Integrated neuromodulation system for restoring volitional motor function

Abstract

A neuromodulation system for restoring volitional motor function is disclosed. The system uniquely integrates multi-channel, self-sizing spiral cuff electrodes, advanced adaptive AI-controlled signal processing, and a hybrid power architecture that combines implanted components with an external, wearable unit. The system may operate in a closed-loop manner by dynamically adjusting stimulation parameters in real time based on inputs from electromyography (EMG), electroneurography (ENG), and other sensor modalities. Pre-clinical testing in porcine ischemic stroke models has demonstrated graded, reproducible, and safe limb movements. This system addresses limitations of current neuromodulation devices by enhancing precision, reducing invasiveness, and ensuring robust safety and interoperability with clinical infrastructures.

Description

Attorney Docket No. 61435-708601INTEGRATED NEUROMODULATION SYSTEM FOR RESTORING VOLITIONAL MOTOR FUNCTION CROSS-REFERENCE

[0001] This application claims the benefits of U. S. Provisional Application No. 63 / 773,866, filed March 18, 2025, U. S. Provisional Application No. 63 / 781,105, filed March 31, 2025, U. S. Provisional Application No. 63 / 800,685, filed May 6, 2025, U. S. Provisional Application No. 63 / 800,703, filed May 6, 2025, U. S. Provisional Application No.63 / 801,529, filed May 7, 2025, U. S. Provisional Application No. 63 / 801,774, filed May 7, 2025, U. S. Provisional Application No. 63 / 802,389, filed May 8, 2025, U. S. Provisional Application No. 63 / 806,159, filed May 15, 2025, U. S. Provisional Application No.63 / 806,270, filed May 15, 2025, U. S. Provisional Application No. 63 / 809,956, filed May 21, 2025, U. S. Provisional Application No. 63 / 811,704, filed May 24, 2025, which applications are incorporated herein by reference.BACKGROUND

[0002] The present disclosure may introduce a novel closed-loop neuromodulation system that uniquely integrates multi-modal sensor fusion, adaptive machine learning, and a modular control architecture to restore natural, coordinated volitional motor function in patients suffering from neurological impairments. Conventional neuromodulation systems may predominantly employ binary on-off stimulation paradigms that can result in abrupt, jerky, or non-physiological muscle responses. These systems may require frequent manual recalibration due to their limited ability to adapt to dynamic variables such as user fatigue, physiological signal drift, and electrode displacement. In contrast, the disclosed system continuously captures and decodes subtle variations in biosignals to modulate stimulation parameters in real time, thereby facilitating the generation of smooth, graded motor responses.

[0003] Moreover, traditional systems typically may not provide graded control based on the strength or speed of a patient’s volitional effort. This lack of proportional control may hinder the production of fluid muscle contractions necessary for executing complex, purposeful movements. The disclosed system may overcome these limitations by translating graded electromyographic (EMG) activity — and other complementary biosignals — into graded, proportional stimulation outputs. This real-time adaptive approach may address critical unmetAttorney Docket No. 61435-708601needs in neuroprosthetics, offering potential benefits for post-stroke recovery, spinal cord injury treatment, and other applications where restoration of natural movement is essential.FIELD OF THE INVENTION

[0004] This invention may relate to neuromodulation technology and, more particularly, to closed-loop, multi-channel systems that may restore voluntary motor function by leveraging advanced biosignal processing, adaptive stimulation mapping, and machine learning. It may be applicable in fields including biomedical engineering, neurorehabilitation, and computational neuroscience. The system may incorporate a dynamic, patient-specific control algorithm for real-time, EMG-driven neuromodulation, thereby providing a highly adaptable platform capable of facilitating both simple joint movements and complex, multi -joint coordination.SUMMARY

[0005] The present disclosure may relate to a fully adaptive, closed-loop neuromodulation system that may integrate one or more spiral nerve cuff electrodes, high-fidelity biosignal acquisition, and advanced machine learning algorithms to restore natural, coordinated volitional motor function. In one embodiment, a plurality of independently controlled, flexible, biocompatible spiral nerve cuff electrodes may be strategically positioned on target peripheral nerves (e.g., musculocutaneous, radial, median, and ulnar nerves). These electrodes may facilitate both isolated and coordinated stimulation and may be configured to capture residual compound action potential (CNAP) recordings from the nerves.

[0006] The system may acquire high-frequency biosignal data from multiple modalities — including EMG, electroneurography (ENG), electroencephalography (EEG), and inertial measurement unit (IMU) data — using sampling rates of at least 1 kHz. The raw signals may be processed through a multi-stage pipeline that includes high-pass, notch, and low-pass filtering, followed by a sliding RMS envelope extraction (utilizing a 50-100 ms window updated every 10-20 ms). This process may yield a bi-dimensional intent vector representing both the normalized amplitude and the dynamic rate (slope) of muscle activation, thereby capturing the graded nature of the user’s volitional input.

[0007] Advanced machine learning algorithms (e.g., support vector machines, convolutional neural networks, hidden Markov models, and ridge regression) may continuously decode the intent vector. The decoded graded intent may then be dynamically mapped to stimulation parameters — such as amplitude, pulse width, and optionally frequency — using a mapping function that may be linear or nonlinear. The entire closed-loop control may update every 10-Attorney Docket No. 61435-70860120 ms with total latency maintained below 20 ms, ensuring that the system may respond in near real time to variations in the patient’s motor intent.

[0008] A key aspect of the invention may be its proprietary dynamic control algorithm. This algorithm may continuously monitor residual EMG signals (sampled at >1-2 kHz), normalize the extracted envelope against patient-specific baselines (EMG rest and MVC), and compute the first derivative (slope) to quantify the rate of muscle activation. The resulting 2D intent vector may be used to determine graded stimulation outputs, with ramp-up and controlled decay functions (exponential or sigmoid) employed to provide smooth transitions. A comprehensive calibration procedure may establish individualized sensor baselines and mapping functions, with automatic recalibration triggered by significant drift, thereby enabling the system to deliver proportional and graded stimulation that closely matches the patient’s volitional intent.

[0009] The system may further include a modular software architecture that may be deployed on both embedded and cloud-based platforms. This architecture may comprise discrete modules for signal processing, intent decoding, stimulation mapping, synchronization scheduling, safety monitoring, and remote update management. A dedicated synchronization scheduler may coordinate multi-channel stimulation to ensure the execution of synchronized multi -joint movements, while dynamic sensor fusion algorithms may integrate data from various biosignal modalities to enhance robustness. Additionally, the system may support a software-only deployment option that may interface with third-party hardware or cloud-based neuroprosthetic platforms, reducing development costs and facilitating rapid clinical translation.

[0010] Enhanced safety features may include redundant sensors, fail-safe modes, robust data encryption (e.g., AES-256), and secure communication protocols (e.g., HTTPS, MQTT). A user-friendly interface may provide real-time monitoring, parameter adjustments, and secure data logging, thereby ensuring continuous, secure, and reliable operation. Pre-clinical testing and proof-of-concept embodiments (such as EMG-driven real-time elbow flexion and selective stimulation of the brachialis branch using alternative muscle signals and field steering) may demonstrate the system’s transformative potential for restoring graded, coordinated motor function across diverse anatomical regions. Early generations of peripheral-nerve stimulators were powered exclusively through high-frequency inductive links. Because therapy could only occur while an external coil was precisely aligned over the implant, patients faced lengthy daily set-up times and frequent loss of power during ordinary activities. Later, fully implantable systems adopted internal batteries, eliminating the coilAttorney Docket No. 61435-708601tether but forcing a trade-off between size and runtime: large cells extend therapy but create cosmetic bulges, while smaller cells mandate frequent surgical replacement or recharging sessions. Commercial battery IPGs aimed at pain and sacral nerves typically drive a single eight-contact lead with simple monopolar or bipolar waveforms, offering limited fascicular selectivity.

[0011] A second limitation concerns stimulation programmability. Many state-of-the-art devices rely on a “library playback” model: a clinician constructs one or two pulse trains off-line, stores them in flash, and the implant merely alternates or loops these sequences during therapy. This approach cannot accommodate second-to-second fluctuations in physiological state — for example, the bursty ENG associated with volitional muscle intent or the varying EMG amplitude that accompanies different grasp forces — without constant external re-programming. Recently published patents (e.g., Neuroinnov EP 4340935 Bl / US 2024-0335665) formalise this static-library concept, further entrenching its limitations by requiring that entire pre-stored sequences be fetched at runtime.

[0012] Finally, existing IPGs seldom exploit the full potential of multi-contact cuff electrodes. Systems with multiple leads generally dedicate one current driver per contact or multiplex a single driver sequentially, preventing simultaneous, current-split activation that could steer current density within a nerve cross-section. Combined with the absence of real-time control logic, these architectural constraints cap clinical efficacy — particularly for complex motor functions where both spatial and temporal precision are critical. The present disclosure overcomes these deficiencies by merging a space-efficient, rechargeable battery architecture with a high-bandwidth control engine capable of computing interleaved, multicontact patterns on demand, all while remaining within a cosmetic envelope suitable for chronic implantation in a limb or torso pocket.

[0013] The present disclosure may lie at the intersection of implantable neurostimulation and electroneurographic (ENG) signal processing. More specifically, it can concern systems, devices, and computer-implemented methods that may deliver therapeutic electrical pulses to a peripheral motor nerve for functional restoration; may record ENG signals from that same nerve in real time; and may algorithmically modulate stimulation parameters — including on / off gating, pulse width, amplitude, timing, field-steering polarity, and duty-cycle — based on the recorded ENG to enable proportional, closed-loop control of motor movement in any limb or joint.

[0014] The disclosure may further encompass:Attorney Docket No. 61435-708601

[0015] Direction-aware, adaptive blanking that shortens or lengthens the mute interval according to stimulus-propagation direction;

[0016] Ultra-low-power on-chip artefact suppression constrained to < 100 pW and < 20 ps compute time;

[0017] Velocity-selective spike classification to distinguish efferent from afferent activity on a spike-by-spike basis;

[0018] Real-time envelope-and-slope extraction of residual ENG for pulse-to-pulse amplitude control;

[0019] Fascicle-selective contact architectures (single-cuff or dual-cuff) providing spatial separation of recording and stimulation; and

[0020] ENG-based reflex suppression, fatigue mitigation, and safety watchdogs, all suitable for chronic human implantation.3.1 Clinical Need

[0021] Severe stroke, spinal-cord injury, and other neurologic conditions can leave survivors with impaired upper- or lower-limb function — e.g., grasp, ankle dorsiflexion, or elbow flexion. For illustration this disclosure may reference elbow flexion, yet the same principles can apply to any peripheral motor nerve. Population studies indicate that ischemic or hemorrhagic stroke may leave more than half of survivors with permanent upper-limb weakness, and loss of upper extremity function can be among the most disabling deficits for activities of daily living (ADLs) such as eating, grooming, and transferring. Commercial functional-electrical-stimulation (FES) systems attempt to restore movement but usually depend on surface EMG or motion sensors to trigger stimulation; these peripherals can add significant latency, may require daily donning, and often fail when no usable EMG is present — as may occur in severe cases.3.2 ENG as an Intent Signal — and Its Central Obstacles

[0022] Direct electroneurography (ENG) recorded from an implanted cuff can provide a richer control signal: dense, high-frequency efferent spikes may arrive 10- to 50-fold faster than surface EMG and can persist even when overt muscle activity is absent. Unfortunately, each therapeutic stimulus pulse may reach ~1 mV at the recording site — over a thousand times larger than the 5-50 pV ENG spikes that encode intent. The resulting artefact can saturate the amplifier for several hundred microseconds, drowning out the very signal needed to (i) throttle ongoing stimulation and (ii) issue a timely stop command. Artefact suppression is further complicated by the fact that stimulus-evoked voltage can propagate orthodromicallyAttorney Docket No. 61435-708601or antidromically; the required blanking interval therefore varies with direction, yet conventional devices apply only fixed, direction-blind mute windows.3.3 Limitations of Prior Art

[0023] Existing neurostimulators address these obstacles only partially:

[0024] Fixed, direction-blind blanking intervals (e.g., cochlear implants) may mute the amplifier for 2-5 ms — long enough to ensure recovery but far too long to support pulse-to-pulse control on a motor nerve firing at 30-60 Hz.

[0025] Offline template subtraction performed on bench-top FPGAs cannot generally run on sub-milliwatt micro-controllers, and published implant ASICs consume > 1 mW — an order of magnitude above chronic-implant budgets.

[0026] Spatially co-located contacts place recording pads millimetres from the stim cathode, thereby maximising artefact amplitude and leaving no conduction-delay window for predictive control.

[0027] No cleared device classifies contacts — or individual spikes — by conduction velocity; hence motor and sensory activity remain intermingled, precluding velocity-selective control, proprioceptive braking, or reflex suppression.

[0028] Control algorithms in the literature rely on raw spike counts or coarse rectification, ignoring the envelope magnitude and its time-derivative, both of which can provide faster and smoother proportional control.

[0029] Safety watchdogs often depend on external sensors or simple timers, resulting in disengage latencies > 150 ms and lacking a kill-switch tied to ENG silence.3.4 Resulting Unmet Need

[0030] A pressing need therefore exists for an implant-grade system that can dynamically shorten or lengthen the blanking interval according to stimulus-propagation direction; execute high-fidelity template subtraction in < 20 ps while dissipating < 100 pW; separate stimulation and recording contacts (or cuffs) to exploit conduction-delay look-ahead and lower artefact at the source; classify individual spikes by velocity in real time, enabling dualstream (motor vs sensory) control and spike-by-spike reflex logic; extract both an amplitude envelope and its rate-of-change from residual ENG and modulate the very next pulse accordingly; and incorporate autonomous safety mechanisms — including an ENG-silence watchdog and antidromic reflex-breaker — while operating within the strict power, size, and thermal limits required for chronic human implantation.

[0031] The present disclosure may provide an implantable neuroprosthetic system that can simultaneously stimulate and record from a peripheral motor nerve while achieving sub-Abomey Docket No. 61435-708601millisecond, spike-level closed-loop control without external processors or sensors. The inventive concepts can be viewed as four synergistic pillars, each now augmented with new capabilities described in the accompanying grant:1. Direction-Aware Adaptive Blanking & Residual-ENG Window

[0032] After each biphasic stimulus pulse the recording amplifier may be muted only until stimulus artefact falls below a programmable noise threshold and the measured propagation direction (orthodromic vs antidromic) determines the required mute length (typically < 0.50 ms for distal artefact; < 0.90 ms for proximal). A modifiable residual-ENG window (~ 0.25-1.00 ms or longer) can then open, capturing efferent spikes. Within the window the firmware may extract: a real-time amplitude envelope En, and its time-derivative dE / dt, both computed in < 15 ps. A dual -threshold rule can modulate — or inhibit — the very next pulse according to En (effort) and dE / dt (ramp-up or ramp-down), yielding run-away shut-off latencies < 50 ms at 30-60 Hz.2. Ultra-Low -Power On-Chip Artefact Template Subtraction

[0033] A micro-controller dissipating < 100 pW (~ 34 pA @ 3.3 V) may execute an adaptive template filter that updates coefficients and subtracts artefact in < 20 ps per pulse. Residual-ENG SNR can exceed 6 dB even at 3 mA monopolar currents, eliminating bulky DSP hardware or long fixed blanking.3. Spatially-Selective & Velocity-Selective Architectures

[0034] Single-cuff variant: proximal contacts (record) and distal contacts (stim) may be hardwired with > 300 Q isolation. A velocity-scan routine can classify contacts as motordominant or sensory-dominant and label each captured spike in real time as fast (efferent) or slow (afferent), enabling spike-by-spike throttling or braking. This can alternatively be measured in a clinical setting manually by a clinician or assisted with software.

[0035] Dual-cuff variant: a dedicated “listening” cuff 30-50 mm upstream of the “talking” cuff can provide a 0.6-1.2 ms predictive window.

[0036] Dynamic field-steering: the controller may switch among monopolar, bipolar, or quadrupolar patterns on successive pulses to bias current toward specific motor fascicles. 4. ENG-Driven Therapeutic & Safety Enhancers(a) Spasticity-breaker: high-frequency afferent bursts may trigger an antidromic pulse-train that aborts reflexes within ~50 ms.(b) Fatigue auto-dosing: slow drift in conduction velocity or Encan trim duty-cycle in real time.Attorney Docket No. 61435-708601(c) ENG-silence watchdog & BLE kill-switch: stimulation may inhibit automatically if no supra-threshold ENG appears for a programmable interval or on patient command.(d) Neuro-recovery dashboard: long-term ENG biomarkers can be logged via BLE and uploaded to cloud analytics for clinician review.Advantages Over Prior Art

[0037] Latency: closed-loop updates may occur in < 1 ms; disengage latencies can be locked to a single 16-33 ms cycle — >10 / faster than EMG- or IMU-triggered FES.

[0038] Duty-cycle efficiency: even at 60 Hz, the pulse + direction-aware blanking window may consume < 5 % of each cycle.

[0039] Ultra-low power: artefact subtraction executes in < 20 ps and < 100 pW; all other algorithms can run on a 64 MHz ARM-M0+ or, with an external pack, on a higher-power SoC without changing timing.

[0040] Spike-selective control: velocity labelling enables bidirectional throttle / brake from a single nerve, no IMU required.

[0041] Safety: ENG-silence watchdog, adaptive blanking, and field-steering reduce risk of run-away contraction or unintended fascicle activation.

[0042] In various embodiments these four pillars may be combined or practiced independently, providing broad design flexibility while maintaining the essential benefit: stable, proportional, reflex-aware restoration of volitional limb movement — including, but not limited to, elbow flexion — in individuals with neurologic impairment using a single, fully implantable device.INCORPORATION BY REFERENCE

[0043] All publications, patents, and patent applications mentioned herein may be incorporated by reference to the same extent as if each individual document were specifically and individually indicated.BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The novel features of the disclosure are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present disclosure will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the disclosure are utilized, and the accompanying drawings (also “Figure” and “FIG.” herein), of which:

[0045] FIG. 1 - A system overview showing the implanted nerve cuff electrodes, biosignal sensors, and processing unit in relation to targeted peripheral nerves.Attorney Docket No. 61435-708601

[0046] FIG. 2 - A diagram outlining the core system components including signal acquisition, machine learning-based intent decoding, and stimulation mapping.

[0047] FIG. 3 - A flowchart depicting the biosignal processing pipeline — from raw signal acquisition and filtering through envelope extraction, normalization, and intent vector formation.

[0048] FIG. 4 - A workflow diagram illustrating the machine learning calibration process and the adaptive retraining loop based on feedback and drift detection.

[0049] FIG. 5 - A schematic showing the dynamic control of multiple peripheral nerve stimulation channels with inter-channel synchronization.

[0050] FIG. 6 - An illustration of sensor validation and fail-safe mechanisms, including redundant biosignal inputs and safe fallback modes.

[0051] FIG. 7 - A clinical workflow diagram showing patient-device interaction, real-time monitoring during therapy sessions, and remote system tuning.DETAILED DESCRIPTION

[0052] Traumatic insults to major peripheral nerves — including laceration, crush, traction, and iatrogenic transection — disable tens of thousands of subjects per year (e.g., > 30000 cases annually in the United States alone). Although contemporary microsurgical repair re-apposes fascicular architecture with epineurial or group-fascicular suture, fewer than about 40 % of patients ultimately recover Medical Research Council (MRC) motor grade > 4. Controlled laboratory studies demonstrate that a brief bout of charge-balanced biphasic stimulation10-50 Hz for « 60 min within ~ 24 h post-repair) can more than double axonal regenerative rate and improve functional re-innervation; nevertheless, no commercially available cuff-based platform presently delivers such clinically optimized waveforms in the operating theater or early post-operative ward.

[0053] Pathologic tremor — including essential tremor (ET) and tremor associated with Parkinson’s disease (PD) — affects millions worldwide (~ 0.9 % of the general population and >4.5 % of adults > 65 y for ET alone). Deep-brain stimulation (DBS) provides highly effective symptom relief but entails intracranial lead placement, permanent implanted pulse generators, and attendant risk of hemorrhage or infection. Phase-cancelling peripheral -nerve stimulation (PNS) via surface electrodes has shown promise as a non-invasive alternative; however, extant surface-electrode wearables suffer from inconsistent electrode placement, sweat-induced impedance drift, and insufficient closed-loop latency, limiting real -world efficacy.Attorney Docket No. 61435-708601

[0054] Motor-neuron diseases such as amyotrophic lateral sclerosis (ALS) manifest early neuromuscular-junction (NMJ) “dying-back” degeneration and progressive muscle atrophy. Pre-clinical data indicate that chronic low-frequency PNS1-20 Hz,< 60 min day-1, > 5 days week-1) up-regulates neurotrophic factors (e.g., GDNF, CNTF) and slows functional decline. Yet no implantable or hybrid cuff system optimized for daily disease-modifying stimulation and concurrent electroneurographic (ENG) feedback is currently available.

[0055] Taken together, these clinical gaps establish a long-standing, unmet need for a versatile, re-programmable, cuff-based neuromodulation platform capable of (i) delivering peri-repair regenerative stimulation, (ii) executing ultra-low-latency phase-cancelling tremor control, and (iii) providing chronic neuroprotective therapy — all using a common hardware architecture that may be firmware-reconfigured for diverse pathologies.DEFINITIONS

[0056] For purposes of this specification and the appended claims, the following terms shall have the broad meanings set forth below, unless a narrower meaning is expressly provided:

[0057] “Adaptive” means capable of modifying at least one parameter without manual reprogramming while therapy is ongoing. “Algorithmic module” means any combination of deterministic, statistical, artificial-intelligence, machine-learning, reinforcement-learning, rule-based, or future-developed computational techniques executable on one or more processors, micro-controllers, FPGAs, ASICs, NPUs, GPUs, cloud instances, or other computing resources, whether fixed-precision, floating-point, or quantised.

[0058] “Artifact” means any stimulus-induced electrical transient that obscures neural recording fidelity.

[0059] “Charge density” is expressed in micro-coulombs per square centimeter of geometric electrode area and encompasses both phases of a biphasic pulse.

[0060] “ Current steering” means distributing current among two or more contacts such that the resulting electric-field vector approximates a prescribed spatial pattern with El = 0.

[0061] “Dynamic electrode assignment” means firmware-controlled re-designation of any contact as a sensing node, reference node, current source, current sink, or high-impedance node on a pulse-to-pulse basis or slower.

[0062] “Physiologic signal” includes ENG, EMG, IMU, biochemical, optical, mechanical, or any combination thereof.Attorney Docket No. 61435-708601

[0063] “Real-time” means end-to-end latency not exceeding about 150 ms unless a tighter numerical bound is expressly recited. “Training” encompasses supervised, unsupervised, selfsupervised, semi-supervised, transfer-learning, federated-learning, on-device continuous learning, offline batch learning, simulation-generated synthetic data, or any combination thereof.

[0064] “Trigger input” means any wired or wireless port — digital, analogue, optical, acoustic, or otherwise — configured to insert synchronization or control metadata into the telemetry uplink or to gate a stimulation command.

[0065] “ 0on / ©off’ means adaptive upper / lower activation thresholds.

[0066] “Pathology” means any disorder or condition involving aberrant neuromuscular, sensorimotor, autonomic, or neuro-immune activity, including, without limitation, traumatic nerve injury, movement disorders, neuropathic pain, autonomic dysfunction, inflammatory neuropathies, or degenerative motor-neuron disease.

[0067] “Mechanical signal” includes kinematic or kinetic data such as acceleration, angular velocity, joint angle, grip-force, or pressure, whether derived from IMUs, force sensors, or other transducers

[0068] “Therapeutic parameter” means any variable that may be adjusted to modulate neural activity, including pulse amplitude, pulse width, pulse polarity, inter-pulse interval, duty cycle, burst pattern, electrode-vector selection, waveform morphology, or any parameter derived by an artificial-intelligence (Al) or machine-learning (ML) algorithm.

[0069] “Sensor” means any transducer capable of detecting biological, biomechanical, chemical, or electrical phenomena, including, without limitation, inertial-measurement units (IMUs), electromyography (EMG) electrodes, electroneurography (ENG) micro-contacts, electroencephalography (EEG) electrodes, optical displacement sensors, biochemical analyte sensors, temperature probes, or pH sensors.

[0070] “Cuff electrode” means a circumferential or partial-circumferential structure — including spiral, helical, split-ring, and hinged variants — configured to surround or partially surround a nerve and comprising two or more conductive pads or contacts.

[0071] “Hybrid architecture” means a system in which select components (e.g., battery, processor, telemetry radio) are external to the body but in communicative coupling with an implanted cuff-electrode assembly.

[0072] “Subject” encompasses human patients and non-human mammals (e.g., canine, equine, porcine) or other vertebrate animals suffering from a pathology as defined herein.Attorney Docket No. 61435-708601

[0073] “Inductive link” - a pair of magnetically coupled coils that can transfer energy without direct electrical connection. The term may encompass loosely or tightly coupled resonant or non-resonant systems.

[0074] “Physiological signal” - any electrical, mechanical, optical, or chemical parameter derived from a living subject, including but not limited to ENG, EMG, local field potential (LFP), heart rate, motion, or impedance.

[0075] “Adaptive” or “Real-Time” - operations that can occur at run-time and may depend on measured, estimated, or predicted biological states, rather than a fixed pre-loaded schedule.

[0076] “Multi-contact cuff electrode” - an electrode structure that may include two or more conductive contacts disposed circumferentially or segmentally around nerve tissue.

[0077] “ Closed-Loop” refers to a system wherein sensor feedback is continuously used to adjust stimulation parameters in real time.

[0078] “Volitional Motor Intent” means the patient’s intentional neural or muscular signals that indicate a desire to perform a movement.

[0079] “Multi-Channel Cuff Electrode” refers to an electrode configured in a cuff or spiral design that may include multiple (e.g., 2-16) contacts to selectively record and stimulate different regions of a nerve.

[0080] “Adaptive Al” means artificial intelligence algorithms (including machine learning techniques such as neural networks or support vector machines) that may be trained to adjust system parameters dynamically based on sensor data.

[0081] “Hybrid Power Approach” refers to a configuration wherein implanted components may be powered via transcutaneous energy transfer from an external, wearable unit, thereby reducing or eliminating the need for an internal battery.

[0082] Disclosed herein are adaptive neuromodulation systems comprising at least one multi-contact cuff electrode operatively coupled to bidirectional stimulation-and-sensing electronics. The electronics may reside in (i) a fully implantable pulse generator, (ii) a subcutaneously implanted receiver powered by an external driver, or (iii) a hybrid architecture wherein stimulation drivers are implanted while the primary battery and / or processor are external. The system may interface with one or more sensors — implanted, wearable, or environmental — and may operate in open-loop, sensor-triggered closed-loop, predictive Al, or clinician-initiated telecommand modes.

[0083] Three illustrative therapeutic paradigms enabled by the disclosed technology include:Attorney Docket No. 61435-708601

[0084] Regenerative Stimulation Protocol (RSP) — Within an initial post-operative window (e.g., within about 24 h of surgical repair), the cuff electrode may deliver a train of charge-balanced biphasic pulses having a frequency in a range of about 10 Hz to about 50 Hz, a per-phase pulse width of about 50 ps to about 300 ps, and a current amplitude selectable up to several milliamperes, for a treatment duration on the order of tens of minutes (e.g., about one hour). The therapy may be repeated episodically.

[0085] Tremor Cancellation Protocol (TCP) — A kinematic or other sensor may detect tremor frequency and phase in real time. A controller may compute an anti-phase stimulation pattern whose frequency substantially matches the dominant tremor frequency (typically about 4 Hz to about 12 Hz) and whose relative phase offset is approximately 180°, thereby reducing net oscillatory torque. Pulse-train amplitude may scale with tremor magnitude.

[0086] Neuroprotective Stimulation Protocol (NSP) — For neuromuscular-degenerative pathologies, the cuff electrode may deliver low-frequency stimulation (e.g., <20 Hz) for sessions of approximately 20 min to 60 min, multiple days per week, to evoke sub-tetanic contractions that maintain NMJ signaling. Concurrent ENG or EMG sensing may track biomarkers — such as compound-muscle-action-potential (CMAP) amplitude — and guide adaptive parameter updates.

[0087] A single hardware configuration may be re-programmed via firmware download or external programmer to execute any of the foregoing paradigms or other therapies disclosed herein. Optional extensions include cloud-based data analytics, multi-cuff arrays for multi -joint coordination, and Al-driven parameter optimization.

[0088] Described herein, in certain embodiments is a modular neuromodulation platform that integrates implanted and external components, sensor arrays, adaptive control algorithms, and secure cloud connectivity. The system may be configured to operate in both open-loop (preset) and closed-loop (adaptive) modes, thereby providing patient-specific therapy synchronized with the patient’s volitional motor intent. Secure remote monitoring, over-the-air software updates, and long-term data analytics may be supported by the integrated architecture, ensuring that the system may evolve with clinical needs and technological advancements. FIG. 1 demonstrates the implanted and external components and anatomical implantation sites (including the implantable unit, external wearable unit, and sensor interface). FIG. 1 also depicts a workflow diagram outlining the operational process of an exemplary — from acquisition of sensor signals through processing and generation of stimulation commands to delivery of electrical pulses for restoring volitional movement.Attorney Docket No. 61435-708601

[0089] Described herein, in certain embodiments, is an implantable unit that may include a self-sizing, low-profile spiral nerve cuff electrode designed to encircle peripheral nerves (e.g., the musculocutaneous, median, radial, or ulnar nerves), as seen in FIG. 3, which depicts an exemplary implantable unit and highlights its compact, biocompatible design and the integrated stimulation electronics intended for long-term implantation. FIG. 2 shows an overview diagram of the order of the system components, including the implantable unit with its integrated cuff electrode and stimulation circuitry, the sensor module acquiring physiological signals, the processing unit implementing control algorithms, and the communication module facilitating wireless connectivity. The electrode may have an inner diameter ranging from about 5 to 8 mm and a length from about 1.0 to 2.0 cm, with 2-16 contacts spaced approximately 1 to 2 mm apart. This multi-contact design may enable both selective stimulation and high-resolution neural recordings that are essential for decoding nerve signals. An optimized coil or power reception module may facilitate transcutaneous energy transfer with efficiencies that may exceed 85% at tissue depths of about 5 to 10 mm. In certain embodiments, an implantable pulse generator may be provided, housing stimulation circuitry and a rechargeable lithium-ion battery that may be recharged transcutaneously via an inductive coil for long-term operation. Stimulation circuitry may comprise current-controlled, biphasic, rectangular, asymmetric stimulus pulses (cathodic amplitude with pulse width followed by an anodic counter pulse of 1.4x amplitude and 4x pulse width), or any combination thereof. The system may comprise 32 stimulation channels. In some embodiments, a current of the system may be -6mA / +1.5 mA in 24 pV increments, within a compliance voltage range of -1 IV to +5V. In some embodiments, a current source can be directed to any of the 32 electrode contacts. In some embodiments, a pulse width can range from approximately 10 to 2,500 ps. In some embodiments, a power supply may be a wireless inductive power supply of 120-140 kHz. In some embodiments, the system comprises a wireless data transmission that is bi-directional, with a radio frequency in a 2400-2483.5 MHz band. In some embodiments, the system comprises a closed loop latency of less than or equal to 40 ms.

[0090] In certain embodiments, the external unit may be designed to be modular and wearable, enclosed in a rugged, water- and sweat-resistant casing with a compact form factor comparable to modern fitness trackers. FIGS. 5A-5B show an exemplary wearable external unit, configured in an armband, demonstrating its role in powering the implantable unit and receiving wireless signals from the EMG sensor. It may be configured for attachment via magnetic, clip-on, or armband methods to ensure ease of use in both clinical and everydayAttorney Docket No. 61435-708601settings. A high-capacity rechargeable battery (e.g., approximately 5000 mAh) may support up to 24 hours of continuous operation with rapid charging cycles (approximately 1-2 hours). Wireless communication modules, such as Bluetooth 5.0 with AES encryption, may ensure secure data transmission over distances that may reach up to 10 meters. Standardized connectivity options may allow additional sensors or therapeutic devices to be integrated, ensuring scalability and future adaptability.

[0091] Preferred embodiments may utilize medical-grade silicone for the cuff electrode and biocompatible metals (e.g., platinum-iridium or titanium) for the electrode contacts to ensure long-term biocompatibility and durability. Manufacturing techniques such as precision molding and laser micromachining may be employed to achieve the high-resolution, multichannel design. Also described herein, in certain embodiments, are systems that may integrate various sensor modalities to provide a comprehensive neuromuscular dataset. In some embodiments, the sensor modalities may comprise electromyography (EMG) sensors. In some embodiments, the EMG sensors are wearable and implantable EMG sensors may capture muscle activity at high sampling rates (approximately 1 kHz or greater with 16-bit resolution) and may detect low-amplitude signals (as low as 5-10 pV) from target muscles (e.g., biceps, triceps, trapezius). FIG. 6 shows an exemplary wearable EMG sensor and a representative raw EMG waveform captured by the sensor, which helps illustrate how the wireless sensor detects volitional muscle activity and communicates with the external unit to enable seamless, wire-free control. FIG. 4 shows a diagram of the design of the implanted spiral nerve cuff electrode, showing its spiral configuration, electrode contacts, and intercontact spacing that may facilitate secure placement and enable selective stimulation and recording. In some embodiments, the system comprises a sampling dynamic range. In some embodiments, the sampling dynamic range is a 16 bit range, with a 74nV as a smallest increment. In some embodiments, recordings comprising electroneurography (ENG) and / or compound nerve action potential (CNAP) recordings. Multi-channel cuff electrodes may record ENG signals or CNAPs with high signal-to-noise ratios (potentially exceeding 20 dB) using advanced digital filtering (e.g., a 2nd order Butterworth filter with a cutoff of about 20 Hz). In some embodiments, the system comprises a high pass filter cut-off of about 2 Hz. In some embodiments, the system comprises a low pass filter cut-off of about 325 Hz. In some embodiments, the system comprises an amplifier band pass gain. In some embodiments, the amplifier band pass gain is adjustable. In some embodiments, the adjustable amplifier band pass gain is adjustable from 100 to 750 dB. In some embodiments, the system comprises a band pass roll-off of about 20 dB / dec. In some embodiments, the multi-channel cuff compriseAttorney Docket No. 61435-70860132 recording channels. In some embodiments, the sensors comprise EEG and / or ECoG Sensors. In some embodiments, a reference of any subset of recording channels selectable by software or a dedicated hard-wired additional contact may be used. Implantable or wearable EEG / ECoG sensors may capture cortical activity over a bandwidth of approximately 0.5-100 Hz, contributing to the decoding of volitional intent. In some embodiments, the sensors comprise biomechanical sensors. In some embodiments, the biomechanical sensors measure inertial measurement units (IMUs), torque sensors, and goniometers may continuously monitor limb kinematics, force, and joint angles to enable dynamic adjustment of stimulation parameters. Data fusion algorithms may synchronize and analyze these heterogeneous sensor inputs, ensuring that stimulation is precisely tailored to the patient’s volitional motor intent. The system may further interface with existing clinical infrastructures, such as electronic medical records (EMR) and telehealth platforms, to enhance interoperability.

[0092] Also described herein, in certain embodiments, is a comprehensive software platform. The comprehensive software platform may include a clinician / patient interface accessible via mobile and web platforms. This interface may display real-time visualizations (e.g., waveform displays, heat maps, electric field mappings) that allow clinicians to monitor neural and muscular activity. Users may adjust stimulation thresholds and sensor calibrations within predefined safe operating ranges. Secure, HIPAA-compliant cloud connectivity may facilitate over-the-air firmware and software updates, as well as continuous data logging (up to approximately 50 MB per day) to support long-term therapy optimization. Adaptive control algorithms may process sensor data by: utilizing threshold detection (e.g., approximately 30-40% of maximal voluntary contraction for EMG signals) and time-window analysis (typically 50-200 ms) to capture true volitional signals while filtering out noise; implementing machine learning algorithms (e.g., neural networks or support vector machines) to analyze multimodal sensor inputs with classification accuracies that may approach 90-95%; and / or techniques such as blanking periods (approximately 2-5 ms) and periodic pulse-off windows (approximately 10-50 ms) may further enhance signal fidelity; dynamically adjusting stimulation parameters (such as current amplitude ranging from about 0.1 to 20 mA, pulse width from approximately 10 to 500 ps, frequency around 10 to 80 Hz, and ramp-up / ramp-down intervals approximately 200 to 2000 ms) in real time so that therapy may closely mimic natural, volitional movement. The system may include extensive diagnostic and safety features, such as continuous monitoring of electrode impedance, battery status, and signal quality; redundant sensor pathways; and failsafe mechanisms triggered by watchdog timers and predefined emergency protocols. The extensive diagnostic and safety features may beAttorney Docket No. 61435-708601configured to facilitate proactive maintenance of the system and to help ensure system reliability.

[0093] Also described herein, in certain embodiments, the system disclosed herein may acquire high-resolution data from multiple sensor types simultaneously. Advanced digital filtering (e.g., using a 2nd order Butterworth filter with a cutoff of about 20 Hz) may be applied to suppress noise while preserving critical signal features. Feature extraction may isolate parameters such as signal amplitude, frequency content, and onset timing through threshold detection and time-window analysis. Machine learning algorithms, continuously calibrated to an individual patient’s unique signal profile, may dynamically adjust stimulation parameters in real time.

[0094] Also described herein, in certain embodiments, the system may include routines for calibrating sensor thresholds and stimulation parameters to individual patient profiles.Baseline readings may be collected over a predetermined period, and adaptive algorithms may adjust to variations in signal amplitude or frequency, thereby ensuring that the system may continue to function optimally as patient-specific neural activity changes over time.

[0095] Also described herein, in certain embodiments, the device may continuously monitor key parameters using redundant sensor channels. If any parameter deviates from safe operating ranges, the system may automatically engage failsafe protocols that may include automated fallback to predefined safe stimulation settings, watchdog timers that may trigger an immediate shutdown if abnormal behavior is detected, and / or redundant control pathways that may ensure uninterrupted operation even if one component fails. Risk management protocols and validation steps may define specific criteria for engaging these failsafe modes, further ensuring robust and safe operation. In some embodiments, a fault detection and failsafe protocol is implemented to ensure continuous and safe operation of the system.

[0096] Also described herein, in certain embodiments, the modular design of the system may allow individual components (e.g., sensor module, processing unit, power system, and communication interfaces) to be independently upgraded or replaced, thereby facilitating future integration with emerging technologies. In some embodiments, the system comprises fully implanted configurations in which all components (including implantable EMG electrodes, nerve cuff sensors, and the IPG) may be entirely implanted with wireless telemetry. In some embodiments, the system comprises hybrid systems where core implanted components may operate in tandem with external modules (e.g., wearable battery packs or EMG patches). In some embodiments, the system comprises specialized embodiments for tremor management using inertial sensors to detect tremor frequencies (approximately 4-12Attorney Docket No. 61435-708601Hz) and adjust stimulation accordingly. In some embodiments, the system comprises configurations for sensory restoration where afferent ENG signals may be relayed to the cortex via direct cortical stimulation. In an alternative embodiment, the system may be fully implantable, incorporating a rechargeable lithium-ion battery and wireless telemetry for remote control and monitoring, eliminating the need for an external wearable unit; this configuration may facilitate seamless integration for long-term clinical applications requiring minimal patient burden. As used herein, the term “long-term” may refer to a period of time greater than or equal to about 1 day, 2 days, 3 days, 4 days, 5 days, 6 days, 1 week, 2 weeks, 3 weeks, 4 weeks, 1 month, 2 months, 3 months, 4 months, 5 months, 6 months, 9 months, 1 year, or longer.

[0097] Also described herein, in certain embodiments, the system may be developed in compliance with applicable regulatory standards (e.g., FDA and CE) through rigorous risk management, quality control, and validation protocols. Patient data and device communications may be safeguarded in accordance with HIPAA and other cybersecurity standards through measures including: data encryption and secure boot procedures, continuous anomaly detection using Al-driven monitoring, and / or secure, authenticated over-the-air updates. Cloud connectivity may enable secure storage of patient data, sensor logs, and diagnostic information, thereby supporting long-term performance analysis, personalized therapy adjustments, and continuous machine learning model updates.

[0026] Also described herein, in certain embodiments, the system may be designed to interface seamlessly with existing clinical infrastructures, such as electronic medical records (EMR) and telehealth platforms. Standardized communication protocols may ensure secure data exchange between the device and hospital networks, promoting streamlined clinical monitoring and therapy management.

[0098] Also described herein, in certain embodiments, the modular architecture may allow future integration with emerging wearable sensor technologies, advanced neurodiagnostic modules, and next-generation interfaces (e.g., brain — computer interfaces, exoskeletons, and haptic feedback systems). Although the present application focuses on restoring volitional motor function, the platform may also be applicable to pain management, cognitive rehabilitation, and broader neuromodulatory therapies, ensuring that the invention may remain relevant as technology advances.

[0028] Also described herein, in certain embodiments, are advantages of the present disclosure over existing neuromodulation systems. The system may have enhanced precision. The use of multi-channel cuff electrodes may permit selective stimulation and high-resolution recording of nerve signals, allowing forAttorney Docket No. 61435-708601more accurate mapping and modulation of neural activity. A combination of sensor inputs (e.g., ENG, EMG) allow for control of this fully integrated clinical grade system. The system may have reduced invasiveness. The hybrid power approach may minimize the need for internal batteries, potentially reducing surgical complexity and interventions. This is a novel application of an implanted receiver-stimulator coil that may be transcutaneously powered by an external pulse generator to streamline system implantation and implantable form factor, allowing it to be implanted in patient limbs rather than the traditional implantable pulse generator placement in the chest. The system may have adaptive and personalized therapy. Patient-specific calibration and adaptive Al may enable the system to dynamically adjust stimulation parameters in real time, ensuring that therapy may closely mimic natural movement and meet individual patient needs. The system may have robust safety and redundancy. Extensive diagnostic features, failsafe mechanisms, and redundant control pathways may ensure continuous, safe operation even in the event of sensor or communication failures. The system may have seamless clinical integration. The system’s ability to interface with EMR and telehealth platforms may facilitate its adoption in diverse healthcare settings and promote streamlined therapy management.

[0099] Also described herein, in certain embodiments, the system may be implemented flexibly to suit various clinical contexts. Backup algorithms and redundant control pathways may be deployed if sensor or communication failures occur, ensuring uninterrupted therapy. The modular design may also allow rapid adaptation to different nerve types or therapeutic indications without extensive redesign.

[0100] Also described herein, in certain embodiments, are key advantages of the system that may include: improved patient-specific calibration that may lead to more effective, personalized therapy; reduced risk of complications through fewer surgical interventions for power management; enhanced safety via continuous monitoring and robust failsafe mechanisms; and / or greater clinical impact due to seamless integration with telehealth and EMR systems, enabling remote monitoring and therapy adjustments. In some embodiments, the personalized therapy may enhance therapeutic outcomes.

[0101] Also described herein, in certain embodiments, is modular architecture may facilitate future enhancements, such as: integration with next-generation brain — computer interfaces for expanded neurorehabilitation; adoption of advanced sensor arrays for even higher resolution neural monitoring; extension to additional therapeutic areas, including pain management and cognitive rehabilitation; beyond motor restoration, the system may be configured for sensory feedback restoration in patients with spinal cord injury or peripheralAttorney Docket No. 61435-708601nerve damage; additionally, closed-loop neuromodulation may be applied to chronic pain conditions (e.g., neuropathic pain, phantom limb pain) via selective afferent stimulation of sensory nerve fibers. The system may incorporate multi-patient data learning, allowing it to refine therapy parameters based on anonymized, cloud-based Al analysis of diverse patient datasets. This approach may enhance stimulation precision, accelerate calibration for new patients, and continuously improve long-term efficacy.

[0102] Also described herein, in certain embodiments, are preclinical studies, such as those conducted in Yucatan minipigs, may have demonstrated that the implantation of multicontact cuff electrodes can yield graded muscle contractions, smooth relaxation, and sustained activation, with EMG recordings ranging from approximately 50 to 1000 pV. These findings may have informed the calibration of stimulation parameters and sensor thresholds. Clinical feasibility may have been supported by an IRB-approved study at the University of California, San Diego (IRB #808314) evaluating post-stroke patients. In this study, the paretic biceps may have retained approximately 5-35% of normal EMG amplitude while the trapezius exhibited about 40-60% activity, thereby supporting the dual-sensor approach and adaptive control strategy. The successful performance of a pre-clinical prototype may further attest to the practical viability and clinical potential of the system.

[0103] Also described herein, in certain embodiments, is a battery. To optimize battery longevity and minimize recharge frequency, the system may include adaptive power management strategies, such as (1) Dynamic power allocation: Automatically adjusting stimulation intensity based on real-time muscle fatigue detection to conserve energy. (2) On-demand activation: Switching to a low-power mode when volitional intent is absent. (3) Smart charging alerts: Predictive analytics to notify users when recharging is needed, based on historical usage patterns. In some embodiments, the battery comprises a LiPo battery with a charge time of about 3-4 hours. In some embodiments, a run time of the battery is about 16-24 hours. In some embodiments, the EMG is measured with a 2 lead measurement, with no reference lead. In some embodiments, the EMG comprises a great than 10 GI input impedance. In some embodiments, the EMG comprises a sub-microvolt resolution. In some embodiments, the EMG comprises a 140 dB signal to noise ratio. In some embodiments, the system comprises embedded firmware and / or software. In some embodiments, the embedded firmware and / or software comprises Bluetooth ® LE v4.2 protocol stack, digital signal process (DSP) and curve fitting for EMG signal characterization and classification, or any combination thereof. In some embodiments, a radio frequency transmission is less than 4 dBm and comprises a range of 10 meters. In some embodiments, the system comprises aAttorney Docket No. 61435-708601Bluetooth ® profile. In some embodiments, the Bluetooth ® profile comprises a peripheral GAP role. In some embodiments, the Bluetooth ® profile comprises a GATT profile. In some embodiments, the GATT profile comprises a service UUID: 0x1815 automation 10 service, with a characteristic (UUID: 0x2A58) Analog — Read / Notify, comprising an EMG level measurement with a 16-bit unsigned integer data type, a range of 0-1000 and a notify rate of 100 ms. In some embodiments, service UUID: 0x1815 automation 10 service comprises a characteristic (UUID: 0x2A56) Digital — Read / Write with a device color ID, with a 8-bit unsigned integer data type, of a range from 1-24. In some embodiments, the GATT profile comprises a service UUID: 0x180F) battery service, comprising a characteristic (UUID: 0x2A19) Battery level — read, with a batter level value of a 8-bit unsigned integer data type, with a range of 0 to 100%. While various embodiments of the disclosure have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions may occur to those skilled in the art without departing from the disclosure. It should be understood that various alternatives to the embodiments of the disclosure described herein may be employed.

[0104] Whenever the term “at least,” “greater than,” or “greater than or equal to” precedes the first numerical value in a series of two or more numerical values, the term “at least,” “greater than” or “greater than or equal to” applies to each of the numerical values in that series of numerical values. For example, greater than or equal to 1, 2, or 3 is equivalent to greater than or equal to 1, greater than or equal to 2, or greater than or equal to 3.

[0105] Whenever the term “no more than,” “less than,” or “less than or equal to” precedes the first numerical value in a series of two or more numerical values, the term “no more than,” “less than,” or “less than or equal to” applies to each of the numerical values in that series of numerical values. For example, less than or equal to 3, 2, or 1 is equivalent to less than or equal to 3, less than or equal to 2, or less than or equal to 1.

[0106] Certain inventive embodiments herein contemplate numerical ranges. When ranges are present, the ranges include the range endpoints. Additionally, every sub range and value within the range is present as if explicitly written out. The term “about” or “approximately” may mean within an acceptable error range for the particular value, which will depend in part on how the value is measured or determined, e.g., the limitations of the measurement system. For example, “about” may mean within 1 or more than 1 standard deviation, per the practice in the art. Alternatively, “about” may mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given value. Where particular values are described in the application and claims,Attorney Docket No. 61435-708601unless otherwise stated the term “about” meaning within an acceptable error range for the particular value may be assumed.

[0107] Multi-Cuff Electrode Architecture: The system may include one or more spiral nerve cuff electrodes that may be implanted surgically or percutaneously around target peripheral nerves (e.g., musculocutaneous, radial, median, and ulnar nerves). These electrodes may be constructed from flexible, biocompatible materials and designed to conform to the nerve’s surface to ensure stable and effective electrical contact. Detailed schematic diagrams may illustrate electrode geometry, material properties, and wiring configurations that may permit independent control of stimulation parameters (such as amplitude, pulse width, and frequency) for both isolated and coordinated muscle activation.

[0108] In one embodiment, the spiral nerve cuff electrodes may be fabricated using flexible, biocompatible polymers (such as silicone or polyurethane) in combination with conductive materials (for example, platinum-iridium alloys or titanium). The electrodes may have an inner diameter ranging from approximately 3 mm to 10 mm and a length from about 1 cm to 3 cm, with electrode contacts arranged in multiple concentric rings. Inter-contact spacing may be in the range of 1-2 mm to enable selective stimulation of targeted nerve fibers.Manufacturing techniques such as precision molding, laser micromachining, and electrochemical deposition may be employed to achieve high-precision geometries and reliable electrical performance. Detailed schematic diagrams and drawings may further illustrate these design features and wiring configurations, which may be adapted for both fully implanted and percutaneous applications.

[0109] Signal Acquisition and Intent Decoding: High-frequency biosignal data from modalities such as EMG, ENG, EEG, and inertial sensors may be acquired at sampling rates of at least 1 kHz to capture transient muscle activity. The raw signals may be processed through a multi-stage filtering pipeline that may include:

[0110] High-Pass Filtering: to remove low-frequency motion artifacts;

[0111] Notch Filtering: to eliminate powerline interference (e.g., 50 / 60 Hz);

[0112] Low-Pass Filtering: to extract the signal envelope by attenuating high-frequency noise.

[0113] Following filtering, feature extraction may be performed using a sliding RMS envelope extractor (with a window of approximately 50-100 ms updated every 10-20 ms). The resulting envelope may be normalized against patient-specific baseline values(EMG rest) and maximal voluntary contraction (MVC) levels, and the first derivative (slope) may be computed to represent the speed of muscle activation. These two parameters may beAttorney Docket No. 61435-708601combined into a bi-dimensional intent vector that may capture the graded nature of the patient’s volitional input.

[0114] ML-Based Adaptive Mapping: The system may employ various machine learning algorithms (e.g., support vector machines, convolutional neural networks, hidden Markov models, and ridge regression) that may continuously decode the bi-dimensional intent vector. The decoded graded intent may then be dynamically mapped to stimulation parameters (e.g., amplitude, pulse width, and optionally frequency). The control loop may execute every 10-20 ms, with total latency maintained below 20 ms, thereby ensuring that the system may respond proportionately and in near real time to variations in the patient’s motor intent. For example:

[0115] Support Vector Machines: May be used for rapid, binary classification.

[0116] Convolutional Neural Networks: May extract complex spatial and temporal features.

[0117] Hidden Markov Models: May account for sequential dependencies in muscle activation.

[0118] Ridge Regression: May address gradual signal drift.

[0119] Patient-Specific EMG Classifier for ON / OFF Detection: During calibration, the system may record EMG data under various conditions:

[0120] Volitional ON: When the patient actively contracts the target muscle.

[0121] Volitional OFF: When the patient is at rest.

[0122] Non-target Movements / Artifacts: Such as coughing or adjusting posture.

[0123] Features such as RMS, mean absolute value, zero-crossing rate, slope sign changes, waveform length, and time-frequency characteristics may be extracted from 100-300 ms sliding windows. Various classifier models (e.g., logistic regression, decision trees, SVMs, shallow neural networks, or ID CNNs) may be evaluated to select the optimal model for distinguishing true volitional signals from noise. The classifier may incorporate state machine logic requiring multiple consecutive “ON” detections before initiating stimulation, and recalibration may be triggered if long-term drift is detected. In one embodiment, the training process for the machine learning model may involve collecting a large dataset of biosignal recordings from multiple subjects performing a variety of motor tasks under controlled conditions. The collected data may be pre-processed using normalization, noise reduction, and feature extraction techniques before being partitioned into training, validation, and test sets.

[0124] The system may employ algorithms such as support vector machines, convolutional neural networks, or hidden Markov models. Hyperparameters (such as kernel parameters forAttorney Docket No. 61435-708601SVM, the number of layers and neurons for CNNs, or transition probabilities for HMMs) may be optimized using cross-validation techniques. Learning curves and performance metrics (e.g., classification accuracy and mean squared error) may be documented to demonstrate model convergence and robustness. Moreover, the trained model may be updated continuously during operation by incorporating real-time performance data and periodic retraining sessions to compensate for physiological changes or signal drift.

[0125] Dynamic Closed-Loop Control Algorithm: The dynamic control algorithm may comprise the following steps:

[0126] Signal Acquisition & Filtering: EMG signals may be sampled at >1-2 kHz and processed using high-pass, notch, and low-pass filters to generate a clean signal.

[0127] Envelope Extraction & Smoothing: A sliding RMS envelope extractor may compute the amplitude envelope over a 50-100 ms window (updated every 10-20 ms), and an Exponential Moving Average (EMA) (with a typical smoothing coefficient of a ~ 0.3) may be applied to reduce jitter while preserving responsiveness.

[0128] Intent Interpretation: The smoothed envelope may be normalized against patient-specific EMG rest and MVC values, and the slope of the envelope may be computed (using a finite difference over a 10-20 ms interval) to represent the rate of muscle activation. These two values may be combined into a 2D intent vector ([Normalized Amplitude, Normalized Slope]) that may capture the graded volitional intent.

[0129] Stimulation Mapping & Control: A mapping function (which may be linear or nonlinear) may convert the intent vector into target stimulation parameters (for example, amplitude may range from 0-6 mA and pulse width from 20-500 ps, with optional frequency adjustments). Ramping rules based on the computed slope may determine the rate at which stimulation parameters increase or decrease. When the EMG signal falls below a predefined threshold, controlled decay functions (exponential or sigmoid) may be applied for smooth deactivation.

[0130] Real-Time Operation: The closed-loop control may update every 10-20 ms with a total latency of less than 20 ms, enabling prompt and precise response to graded input.

[0131] Safety & Personalization: Integrated safety features (such as dropout detection, fatigue timers, and hard limits) may be implemented. A comprehensive calibration procedure (including baseline acquisition, MVC measurement, and controlled ramping trials) may personalize the system to each patient, with automatic recalibration triggered upon significant signal drift.

[0132] Example Pseudocode:Attorney Docket No. 61435-708601

[0133] # Acquire and process EMG samples

[0134] buffer = get_latest_emg_samples()

[0135] rms = compute rms(buffer)

[0136] smoothed env = smooth_envelope(rms, previous ema)

[0137] emg norm = normalize_emg(smoothed_env, EMG rest, EMG max)

[0138] slope = compute_slope(smoothed_env, previous env, dt)

[0139] # Map normalized EMG to stimulation parameters

[0140] amp target, pw target = emg_to_stimulation_params(emg_norm)

[0141] ramp rate = base rate * normalize slope(slope)

[0142] # Apply ramp logic for smooth transitions

[0143] amp out = apply_ramp(current_amp, amp target, ramp rate)

[0144] pw out = apply_ramp(current_pw, pw target, ramp rate)

[0145] # If EMG is below threshold, apply controlled decay

[0146] if emg norm < threshold:

[0147] amp out = smooth_decay(current_amp, elapsed time, tau)

[0148] pw out = smooth_decay(current_pw, elapsed time, tau)

[0149] send_stimulation_command(amp_out, pw out)

[0150] Integration in Clinical Workflow: The system may include a detailed calibration session during which EMG is recorded at rest, during maximum voluntary contraction, and throughout controlled ramping trials. The patient-specific parameters derived from this session may be stored as a profile and updated periodically.

[0151] Modular Software Architecture and System Integration: The system may be structured using a modular software architecture that may include distinct modules for:

[0152] Signal Processing: Acquiring and filtering biosignal data, and extracting key features.

[0153] Intent Decoding: Implementing machine learning algorithms and the dynamic control algorithm.

[0154] Stimulation Mapping: Converting the decoded intent vector into precise stimulation parameters.

[0155] Synchronization Scheduling: Coordinating the timing of multi-channel stimulation for synchronized, multi -joint movements.

[0156] Safety Monitoring: Continuously monitoring system operation and triggering fail-safe modes when necessary.Attorney Docket No. 61435-708601

[0157] Remote Update Management: Facilitating secure, over-the-air software updates and remote parameter tuning.

[0158] Synchronization Scheduler: A dedicated synchronization scheduler may be implemented to coordinate the temporal delivery of stimulation across multiple channels. Utilizing real-time operating systems or dedicated microcontrollers, the scheduler may use timing diagrams and advanced algorithms to ensure that stimulation pulses are delivered in a manner that replicates natural motor coordination, thus enabling synchronized multi -joint movement.

[0159] Sensor Fusion: Dynamic sensor fusion may be implemented to combine data from multiple biosignal modalities. A weighted fusion algorithm may assess the quality of each sensor’s data in real time and adjust the overall intent signal accordingly. This redundancy may enhance robustness and ensure accurate control even if one sensor’s output degrades. Detailed mathematical models and pseudocode may be provided to illustrate this process.

[0160] Calibration and Recalibration: A detailed calibration and retraining protocol may be implemented to optimize system performance for each patient. During an initial calibration session, the system may record baseline EMG data while the patient is at rest to establish EMG rest. The patient may then perform maximum voluntary contractions (MVC) and controlled ramping trials (both slow and rapid) to capture the full dynamic range of muscle activation. The acquired data may be used to generate personalized mapping curves and gain factors that convert the normalized EMG amplitude and slope into stimulation parameters. Furthermore, the system may employ automatic recalibration algorithms that trigger retraining when signal drift exceeds a predetermined threshold (for example, 15-20% deviation from baseline) or when sensor quality deteriorates. This dynamic retraining process may update the machine learning model iteratively, ensuring that the system adapts continuously to changes in the patient’s physiological state over time. A comprehensive calibration procedure may be performed to establish personalized baselines and mapping functions. This process may include:

[0161] Baseline Acquisition: Recording EMG data while the patient is at rest to define EMG rest.

[0162] Maximal Voluntary Contraction (MVC): Recording peak EMG during maximum effort to establish EMG max.

[0163] Controlled Ramping Trials: Capturing data during both slow and rapid muscle contractions to derive appropriate ramping parameters.Attorney Docket No. 61435-708601

[0164] Automatic drift detection may prompt recalibration when significant deviations are observed. Flowcharts may be provided to illustrate the calibration and recalibration processes.

[0165] Additional Safety, Redundancy, and Regulatory Considerations: To enhance overall system safety, the invention may incorporate:

[0166] Redundancy and Fail-Safe Mechanisms: Multiple sensors may be deployed, and if a sensor dropout or severe signal artifact is detected, the system may automatically switch to a backup sensor or revert to a safe stimulation mode with predefined parameters.

[0167] Data Security and Regulatory Compliance: Robust data encryption (e.g., AES-256) and secure communication protocols (e.g., HTTPS, MQTT) may be implemented. The system may be designed to comply with applicable regulatory standards (e.g., FDA, CE) and data privacy regulations (e.g., HIPAA).

[0168] Power Management and Device Packaging: Low-power design strategies (including energy-efficient microcontrollers and sleep modes) may be utilized to extend battery life. The device packaging may be designed for both wearable and implantable applications, with considerations for heat dissipation and biocompatibility.

[0169] User Interface and Data Logging: A user-friendly interface may be provided for clinicians and patients to monitor system status, adjust parameters, and review logged data. Secure data logging may enable retrospective analysis and continuous improvement of the system.

[0170] Remote Tuning: A secure, cloud-based framework may enable remote tuning and software updates. This framework may support encrypted data transmission, robust user authentication, and centralized aggregation of patient data. A clinician dashboard may be provided to allow real-time monitoring, adjustments, and personalized treatment without requiring on-site intervention.

[0171] Device Modularity: The hardware and software may be designed with modularity in mind. Individual stimulation channels may be independently activated, adjusted, or deactivated to tailor therapy to patient needs while optimizing power consumption and safety. Detailed specifications regarding modular connectors, power management, and interface designs may be provided to ensure ease of upgrade and customization.

[0172] The system may incorporate an intuitive user interface (UI) designed for both clinical and patient use. The UI may include graphical displays, touchscreens, and web-based dashboards that present real-time visualizations of biosignal waveforms, stimulation parameters, and system status. Clinicians may use the interface to customize therapy sessions,Attorney Docket No. 61435-708601adjust calibration settings, and review historical performance data. The clinical workflow may encompass steps for sensor placement, initial calibration (including baseline, MVC, and controlled ramping trials), live therapy sessions with real-time monitoring, and post-session data review. Additionally, the system may generate automated alerts and notifications in the event of sensor dropout or system anomalies, thus enhancing patient safety and enabling timely clinical interventions.

[0173] Anatomical Expansion: While initial embodiments may focus on upper limb motor function (as demonstrated by the elbow flexion and brachialis stimulation embodiments), the adaptable design of the system may permit its extension to other anatomical regions.Modifications to electrode design, signal processing algorithms, and calibration protocols may enable applications such as lower limb reanimation, trunk stabilization, or diaphragmatic pacing, thereby broadening the scope of the invention.

[0174] The disclosure may provide an implantable, bidirectional neuroprosthetic platform that can simultaneously stimulate and record from one or more peripheral nerves, extract usable electroneurographic (ENG) information within microseconds of each stimulus pulse, and adapt stimulation parameters on a pulse-to-pulse basis. To achieve these goals while meeting chronic-implant power budgets and safety limits, the system may integrate six synergistic technology pillars, each capable of independent use or combined operation:Pillar A - Implant & Electrode Hardware

[0175] A spiral or cuff-type electrode assembly may include contacts that are dynamically switchable between recording and stimulation modes. One cuff can achieve proximal-record / distal-stim separation; two or more cuffs may be spaced 20-60 mm (for example) apart to create a deterministic conduction-delay window and / or reduce stimulation artifact due to spatial separation of recording and stimulating electrodes. A multi-source current driver can steer current toward selected fascicles, and an inductive coil may enable battery-free or hybrid wireless power.Pillar B - Artefact Suppression & Signal Acquisition

[0176] Direction-aware adaptive blanking may mute the amplifier only until artefact falls below a programmable threshold. A Kalman-style template filter and / or an active cancellation reference electrode can subtract residual artefact in < 20 ps while dissipating < 100 pW. Multi-rate Analog-to-Digital Converter (ADC) sampling and wavelet-domain denoising may further expose ENG spikes, while common-mode gain balancing can maintain signal-to-noise (SNR) over long implant lifetimes.Pillar C - ENG Decoding & Proportional ControlAttorney Docket No. 61435-708601

[0177] A self-learning, velocity-selective algorithm may label spikes as motor-dominant or sensory-dominant and update that map over time. Real-time afferent feedback due to evoked motor responses may dynamically alter stimulation parameters. Dual-threshold logic can start or stop stimulation and proportionally scale pulse width, amplitude, or frequency in real time. Real-time fusion with inertial or pressure sensors may bolster accuracy when ENG quality fluctuates.Pillar D - Safety, Fatigue, and Fail-Safe Logic

[0178] Duty-cycle optimization may reduce stimulation when ENG reveals nerve fatigue. An ENG-silence plus IMU watchdog can inhibit output automatically, while a BLE channel may provide an instant patient or clinician kill-switch. If ENG SNR remains low, the controller can fail over to an external EMG or IMU loop to maintain therapeutic continuity.Pillar E - Therapeutic & Diagnostic Modes

[0179] The same hardware may run multiple software-selectable modes: (i) proportional motor restoration; (ii) spasticity reflex-breaker that delivers an antidromic burst within ~ 50 ms of a high-frequency afferent volley; (iii) open-loop nerve-regeneration stimulation with concurrent ENG biomarker logging; and (iv) a neuroplasticity dashboard whose cloud algorithm can compute a longitudinal Recovery Index.Pillar F - Machine-Learning & Cloud Adaptation

[0180] A < 5 k-parameter convolutional neural network may execute on-device at < 15 pj per inference. Compressed feature vectors can upload periodically; cloud servers may retrain a federated model and push encrypted over-the-air updates, while nightly self-calibration can reset ENG thresholds and feed anonymized cohort analytics.

[0181] When combined, these pillars can deliver sub-millisecond closed-loop updates, < 50 ms disengage latencies, artefact-free ENG capture at therapeutic currents up to 3 mA (or more), and multi-year operation within IEC 60601 thermal and power limits. The modular architecture may further allow independent licensing or continuation of individual pillars — hardware, artefact removal, control logic, safety features, therapeutic modes, or Al workflows — thereby maximizing clinical impact and intellectual-property value. Some existing closed-loop neuromodulation systems closed-loop mechanisms primarily for neurological disorders (e.g., epilepsy, Parkinson’s disease). However, these systems typically rely on sensing brain states (EEG, LFP signals) to modulate stimulation but lack multi-modal biosignal integration, adaptive peripheral nerve stimulation, and real-time EMG-driven neuromodulation capabilities. (Reference: NeuroPace RNS® System, Medtronic Percept™ PC DBS)Attorney Docket No. 61435-708601

[0182] Some existing commercial FES devices (e.g., Bioness L300, MyndMove) provide electrical stimulation for muscle activation in rehabilitation but typically operate with simple trigger-based stimulation rather than dynamically adaptive closed-loop control algorithms. They do not utilize real-time machine learning-driven decoding of volitional intent.(Reference: Bioness L300, MyndMove FES systems)

[0183] Recognized herein is the feasibility of integrating machine learning into neuromodulation (e.g., convolutional neural networks, support vector machines) primarily at a research level or proof-of-concept stage. Existing experimental setups are not commercially available, fully integrated, or clinical-grade systems. (References: Ganzer et al., “Closed-loop Neuromodulation with Deep Learning,” PMC9882307; Chandrasekaran et al., “BCI and machine learning integration,” PMC8733782)

[0184] Spiral nerve cuff electrodes have been extensively explored for selective nerve stimulation. However, existing implementations primarily involve research-grade or simple stimulation systems without dynamic, closed-loop neuromodulation controlled by real-time biosignal intent decoding. (References: Tyler & Durand, “Functional electrical stimulation,” PMC4557157)

[0185] The proposed invention represents the first fully integrated, implantable clinical-grade neuromodulation system that employs advanced dynamic closed-loop algorithms driven by real-time multi-modal biosignal inputs (EMG, ENG, EEG, inertial data) and adaptive machine learning for volitional intent decoding.

[0186] Unlike existing commercial solutions — which are either open-loop, triggered, or limited in adaptive capabilities — the disclosed invention uniquely offers:

[0187] Real-time Dynamic Control: Continuous biosignal acquisition and immediate adaptive adjustments of stimulation parameters (amplitude, pulse width, frequency) via a sophisticated proprietary dynamic control algorithm.

[0188] Clinical-Grade Integration: A fully implantable medical device suitable for human use, developed explicitly for clinical deployment and regulatory compliance (FDA, CE marking), addressing unmet clinical needs in rehabilitation.

[0189] Multi-Nerve Volitional Coordination: Simultaneous and independent control of multiple peripheral nerves for coordinated multi -joint movement restoration, not achievable with current commercial systems.

[0190] Modular and Adaptable Software Architecture: Enabling updates, optimization, and scalability for patient-specific therapy customization and future-proofing.Attorney Docket No. 61435-708601

[0191] This invention will thus be the first market-ready clinical-grade implantable neuromodulation device with truly adaptive, closed-loop biosignal-driven neuromodulation.Advantages of Present Disclosure

[0192] The disclosed system may offer a combination of performance, safety, and manufacturability improvements not simultaneously achieved in existing neuromodulation devices. Key advantages include, but are not limited to, the following:Dimension Advantage achieved by the present disclosureClosed-loop latency Adaptive blanking plus predictive template subtraction can update stimulation parameters in < 1 ms and disengage in < 50 ms.Artefact rejection at < 100 pW total DSP (template + wavelet) may deliver > 6 dB implant power budgets residual-ENG SNR at 3 mA.Selectivity & scalability Bidirectional pads + multi-source driver can steer current toward specific fascicles and re-map roles over time.Self-learning spike On-device CNN may continuously refine motor vs sensory classification labels, compensating for scar, rotation, or growth.Multi-functional therapy Same hardware can switch from proportional motor restoration to spasticity suppression or nerve-regeneration mode via firmware.Safety & fail-over ENG-silence + IMU watchdog may auto-inhibit stimulation;automatic fallback can engage external sensors without surgery.Energy provisioning Inductive coil option allows battery-free or hybrid operation for slim form factors and MR safety.Regulatory support data Cloud fleet analytics can supply real-world evidence (RWE) and over-the-air post-market updates in line with FDA “Software as a Medical Device” guidance.Manufacturing flexibility Pad-agnostic firmware may enable one electrode SKU to serve multiple nerves and indications.

[0193] Collectively, these improvements can provide faster, safer, and more versatile neuromodulation therapy while reducing long-term maintenance and surgical burden for patients and clinicians alike.Alternative Embodiments and VariationsAttorney Docket No. 61435-708601

[0194] The inventive concepts disclosed herein may be realized in numerous alternative forms. The following non-exhaustive examples illustrate how particular hardware elements, algorithms, or therapeutic modes can be substituted, combined, or adapted without departing from the scope of the appended claims.Target Nerve & Clinical Indication

[0195] Upper limb: musculocutaneous, median, ulnar, radial — enabling elbow flexion, wrist extension, hand grasp.

[0196] Lower limb: peroneal, tibial — facilitating ankle dorsiflexion or plantar flexion.

[0197] Trunk & diaphragm: intercostal or phrenic nerves — supporting respiratory pacing.

[0198] Autonomic: vagus, tibial, splenic — addressing inflammation, bladder control, or heart-rate modulation.

[0199] Any cuff geometry or firmware table may be retuned for the above nerves via fieldsteering masks and ENG classifier retraining.Sensor Modalities

[0200] ENG only (default).

[0201] ENG + surface EMG for redundancy in cases of partial muscle preservation.

[0202] ENG + pressure or force sensors embedded in a wearable brace to capture grip force.

[0203] ENG + camera-based computer vision for controller validation in lab settings.Electrode & Lead Materials

[0204] Pad metals may include Pt-Ir, TiN, sputtered iridium oxide (SIROF), or PEDOT: PSS.

[0205] Substrates can employ stretchable silicone-carbon composites or liquid-metal microtracks for highly mobile joints.

[0206] Leads may substitute co-extruded silicone + MP35N with Pebax® + Pt / Ir ribbon to improve fatigue life.ASIC & MCU Process Nodes

[0207] 180 nm CMOS mixed-signal (baseline).

[0208] 65 nm or 40 nm low-power nodes may further halve DSP energy and die area.

[0209] RISC-V cores can replace ARM-M-class parts; CNN layers may migrate to an on-die MAC accelerator.Wireless Power & Telemetry VariantsCoil frequency Use-case NotesAttorney Docket No. 61435-7086016.78 MHz (A4WP) Consumer-grade wearable Supports 10 mm coupling; global ISM band.charger13.56 MHz (NFC) Low-depth implants NFC reader may combine power + data in a smartwatch bezel.40.68 MHz (ISM) Pediatric, thin tissue Smaller coils; better coupling at shallow depths.Far-field RF (915 Research animal models Enables “free-roaming” behavioral studies. MHz)

[0210] BLE may be swapped for Wi-Fi HaLow, LoRa, or sub-GHz proprietary links where hospital infrastructure or battery life so dictates.Artefact-Suppression Algorithm Options

[0211] Fixed-template subtraction with periodic coefficient refresh.

[0212] LMS or NLMS adaptive filters in place of Kalman.

[0213] FPGA- or eFPGA-based artefact engines for very high (> 1 kHz) pulse rates.Machine-Learning Alternatives

[0214] SVM or random-forest classifiers can replace CNNs where memory is < 8 kB.

[0215] On-device federated averaging may shift to split-learning, pushing only feature embeddings to the cloud.

[0216] Differential -privacy noise injection can be added for jurisdictions requiring GDPR-like safeguards.Therapeutic Mode Extensions

[0217] Pain suppression: ENG burst-gated stimulation of cutaneous nerves.

[0218] Blood-pressure modulation: baroreceptor ENG triggers field-steered vagal pulses.

[0219] Closed-loop plasticity training: ENG + accelerometer coincidence may drive Hebbian-style pairing stim to accelerate motor relearning.Implantation Workflow Variants

[0220] Implant one or two-piece cuffs on nerves.

[0221] End-to-side suturable cuffs for thin (< 1 mm) autonomic branches.

[0222] Percutaneous deployment through steerable introducers for deep pelvic nerves.

[0223] These alternative embodiments demonstrate that the core inventions — bidirectional cuffs, adaptive artefact suppression, ENG-driven proportional control, multi-modal safety logic, and cloud-optimized machine learning — may be flexibly adapted to a broad range ofAttorney Docket No. 61435-708601anatomic targets, power budgets, regulatory environments, and clinical indications, thereby underscoring the breadth and versatility of the present disclosure.4.2 Comparative Reference MatrixRef Assignee / Power Interleave Real- Curren MultiSensors Notes vs Title Source d Time t Split Contact Present No. Sequence Adaptatio Cuff Invention s n1 Neuroinnov External RF Yes - but No No 2-4 None Requires US only fetched contacts playback 2024 / 033566 from preof entire 5 (WO stored time- 2024 / 202746 library stamped ) sequence;no battery, no closed- loop. 2 Medtronic Li-ion IPG No No No Cylindric Limited Spinal cord Intellis US 8 al SCS ECG only; static 612012 lead tonic or burst modes. 3 Axonics Rechargeab No No No 4-contact No SingleSacral IPG le Li-ion lead output, US 10470 open-loop 139 bladder stim. 4 Nalu MicroExternal RF No No No 1 -contact No RF- IPG US 11 + RF- powered 478980 powered implant, no microstim battery, no cuff.5 Barostim Li-ion No No No Ring cuff, BP Monopolar Neo US 9 2 contacts sensor carotid- 248271 (external b ar or efl ex ) device. 6 Neuropace Li-ion No Yes - No Depth iEEG Brain RNSUS 9 EEG- electrodes implant; no 579389 triggered multi - bursts contact nerve cuff.Attorney Docket No. 61435-7086017 Stimwave Passive RF No No No 8-contact No Passive US 9757 lead receiver; 642 no internal energy store or split driver.8 Setpoint Li-ion No Limited No 4-contact ECG Uses duty- VNS US 10 cuff cycled 953 968 burst, not on-the-fly interleavin g-Abbreviations: EEG = electroencephalogram; BP = blood pressure; SCS = spinal-cord stimulation.4.3 Narrative Contrast with Most Relevant References

[0224] Neuroinnov US 2024 / 0335665 discloses a multi-contact peripheral stimulator that “interleaves” two or more pre-stored sequences pulled from a library

[0065] ). Timing of every pulse is therefore determined a-priori and must be wholly retrieved into RAM before therapy can commence. No mechanism is provided for on-the-fly generation, skipping, concatenation, or time-warping of pulses as taught in Claims 38-42 of the present application. Moreover, Neuroinnov is externally RF-powered; it lacks any internal battery (distinguishing Claims 2-7) and offers no closed-loop update of amplitude or timing from ENG / EMG or IMU signals (Claims 18-21). Because its architecture depends on a dedicated library and external coil, a user could simply disable the library fetch to avoid infringement — confirming that our real-time scheduler, battery can, and current-split driver are non-obvious additions.

[0225] Commercial IPGs such as Axonics, Medtronic Intellis, and Barostim Neo employ internal batteries but restrict stimulation to single-contact or fixed dual-pole configurations; none describe simultaneous amplitude-split across > three contacts (Claim 15) nor subrefractory inter-contact delays (Claim 16).

[0226] Passive RF devices (Nalu, Stimwave) eliminate the battery but therefore cannot satisfy dual-mode operation (open- + closed-loop) nor the hybrid overlay of Claim 43; they also rely on continuous external power, defeating the use-case for untethered therapy.

[0227] Brain-responsive systems (Neuropace RNS) feature closed-loop control but are cortical, lack cuff electrodes, and do not implement current steering or peripheral safety constraints (30 pC cm2).Attorney Docket No. 61435-7086014.4 Distinguishing Features and Technical Advantages

[0228] The inventive system departs from — and improves upon — existing neuromodulation architectures in four fundamental dimensions: power source, pulse-pattern generation, spatial selectivity, and cyber-physical safety. The table and discussion below expand on the contrasts outlined in § 3.3.Dimension Present Invention Neuroinnov Conventional Passive RF 2024 / 0335665 Battery IPGs Implants (Nalu,(Axonics, Intellis, Stimwave) Barostim)Energy Internal battery or No battery, requires Internal battery, but No battery; architecture primary cell plus continuous external no split RF +NFC therapy ceases optional inductive / NFC RF power during recharge flexibility. if coil recharge; therapy therapy. alignment lost. continues untethered.Pattern On-the-fly interleaver Must fetch entire Fixed tonic or Single-contact, generation (skip, repeat, pre- stored preset burst modes; user-selected concatenate, time-warp) timestamped editing requires reburst trains with 100 ps recompute; sequence from programming only.no full sequence fetch. library memory. session.Spatial Single ±6 mA driver One pulse per One or two One contact; no steering concurrently split to contact at a time; no contacts; steering steering.>3 contacts; At < 300 ps simultaneous requires multiplesub -refractory. current split. drivers.Control Dual-mode: scheduled Open-loop library Scheduled openOpen-loop modes open-loop plus sensor- only. loop only (no only.driven closed-loop; ENG / EMGhybrid overlay feedback).supported.Safety Firmware-enforced 30 Charge limit not MRI labelling Passive — MRI envelope pC cm2charge limit, disclosed; no MRI varies; most lack unsafe due to MRI auto-safe, AES- strategy; no OTA signatures. long leads; no 128 telemetry, OTA encryption. encryption. signature.4.4.1 Key Novelty Points

[0229] Dynamic Scheduler vs Static Library Our claims prohibit reliance on pre-timestamped sequences. Pulse onsets are calculated in real time from physiologic features,Attorney Docket No. 61435-708601enabling features such as “skip / extend” (Claim 41) and procedural time-stamping (Claim 42) that Neuroinnov and conventional IPGs cannot emulate without architectural overhaul.

[0230] Single-Source Concurrent Current Split Traditional “current steering” uses multiple drivers fired sequentially. The disclosed ladder / pump circuit divides one driver’s current simultaneously across contacts (Claim 15), halving silicon and battery burden while achieving finer spatial shaping.

[0231] Hybrid Overlay Capability Claim 43 enables clinicians to maintain a baseline tonic therapy while allowing ENG-triggered augmentation — ideal for post-stroke motor-assist where complete autonomy is undesirable. No prior art teaches seamless overlay without halting or replacing the running program.

[0232] Battery-Backed, Coil-Optional Design A rechargeable NMC cell plus inductive / NFC recharge eliminates continuous-coil constraints, while a simple jumper converts the same PCB to a primary-cell research version — flexibility absent in both Neuroinnov’ s coil-dependent design and conventional large-can IPGs.

[0233] Integrated Cyber-Physical Safety AES-128 GCM-encrypted telemetry, ECDSA-signed firmware, automatic charge-density throttling, thermal shutdown, and MRI high-impedance mode collectively surpass the safety disclosures in the cited art.4.4.2 N on-Obvious Synergy

[0234] The combination of (a) concurrent current-split hardware, (b) a 100 ps adaptive scheduler, and (c) a self-contained power source enables sub- 10 ms closed-loop latency without exceeding the charge-density cap — a synergistic outcome not predictable from any single reference. Prior references either teach adaptive timing without spatial steering (Neuropace) or spatial steering without adaptive timing (Medtronic duets, Neuroinnov library).

[0235] Features of the present disclosure that are lacking in existing solutions include:

[0236] A battery-powered single-can IPG with interchangeable rechargeable and primary chemistries;

[0237] A single high-compliance current driver capable of concurrent current-split to > three contacts;

[0238] A scheduler that computes or edits interleaved patterns on-the-fly — not merely fetching a library; and

[0239] Closed-loop or hybrid operation responsive to ENG / EMG / IMU while enforcing real-time charge-density limits.Attorney Docket No. 61435-708601

[0240] The invention provides a fully implantable peripheral-nerve neuromodulation platform that unifies: (i) a hermetic pulse generator housing a rechargeable or primary electrochemical energy store, (ii) one or more multi-contact cuff electrodes coupled through hard-wired or headered leads, and (iii) a dual-mode controller that can operate in a conventional open-loop schedule or a sensor-driven closed-loop regime. Unlike prior systems that replay fixed pulse libraries, the disclosed controller generates or modifies interleaved stimulation patterns in real time, splitting current from a single high-compliance driver across multiple contacts to sculpt electric fields within the nerve. Power flexibility is achieved with interchangeable battery chemistries — NMC lithium-ion, LiPON solid-state, or high-density primary cells — recharged when desired via a 120-140 kHz inductive link or a 13.56 MHz NFC tap. Embedded safeguards limit charge density to < 30 pC cm2, pause therapy above 42 °C tissue temperature, log all parameter changes in non-volatile memory, and automatically enter a high-impedance MRI-safe state at fields > 1 T, thereby delivering adaptive, multi-contact neurostimulation within a compact, MR-conditional, and cyber-secure form factor. In an alternate architecture the implant omits the long-term battery and receives continuous transcutaneous power from a wearable external puck while executing the same real-time closed-loop control.

[0241] In one illustrative embodiment, a fully implantable neuromodulation system 100 (Fig. 1) is designed to restore volitional upper-limb movement in post-stroke patients. System 100 includes a hermetically sealed implantable pulse generator (IPG) 110, one or more multicontact nerve-cuff leads 120, and an external programmer / charger 130. Other embodiments may target spinal, autonomic, or pain pathways and may employ different housing shapes, lead geometries, or signal sources without departing from the scope of Claim 1.(a) Hermetically Sealed IPG 110

[0242] Can & Feedthrough. In one embodiment, the IPG can is machined from Ti-6A1-4V ELI and laser-welded to achieve a helium leak rate below 1 x 109atm cm3s '. A 10-pin glass-to-metal feedthrough routes electrode and coil conductors.

[0243] (i) Electrochemical Energy Store > 10 mWh. A prismatic 25 mWh Li-ion pouch cell (10 mm * 18 mm x 3 mm, nominal 3.7 V) is epoxy-staked to the inner lid. Alternative embodiments may employ LiPON thin-film cells, Li / CFXprimaries, or hybrid supercapacitors, provided the nominal capacity is at least 10 mWh.

[0244] (ii) Stimulation Engine. A 32-channel ASIC fabricated in 180 nm CMOS provides a single ±6 mA current driver and an H-bridge cross-point switch matrix. Pulse parametersAttorney Docket No. 61435-708601(amplitude 20 pA - 6 mA; width 20 ps - 1 ms) are programmable in 24 pA and 4 ps increments, respectively. Charge balancing is enforced automatically.(b) Nerve-Interface Lead 120

[0245] In certain embodiments, each lead contains eight MP35N conductors (00.1 mm) within medical-grade silicone tubing. The distal end terminates in a split-ring cuff electrode (see Fig. 5) with eight 1 mm2Ptlr contacts spaced 45°. Alternative cuffs may use spiral, FINE, or flat-interface geometries; each still provides at least two independently addressable contacts as required by Claim 1.(c) Controller & Operating Modes

[0246] Hardware. An ARM Cortex-M4F microcontroller (120 MHz, 512 kB flash) manages stimulation timing, telemetry, and power. An analog front-end acquires ENG or EMG with < 1 pVrms noise.

[0247] Open-Loop Mode. In one implementation, a Schedule Table stored in flash programs a fixed 30 Hz pattern for 30 minutes each morning. The same firmware can instead compute a deterministic pattern on-the-fly — for example, issuing a pulse every 33 ms — without referencing the table.

[0248] Closed-Loop Mode. The controller samples ENG every 5 ms, computes a movingwindow RMS, and adjusts pulse amplitude via a PID algorithm whenever the ENG exceeds 100 pV. A scheduler recomputes contact routing and timestamps every 100 ps, ensuring that no full sequence of time-stamped pulses is fetched from memory. Latency from signal sample to modified pulse is ~ 45 ps.

[0249] Hybrid Overlay (optional). Some embodiments maintain a baseline open-loop train (e.g., 25 Hz on contacts C1 / C2) while overlaying additional pulses on contact C3 whenever an IMU detects swing phase — illustrating dependent Claim 43.Compliance & Safeguards

[0250] Charge per phase is capped at 30 pC cm2by firmware.

[0251] A watchdog halts therapy and reverts to a safe preset if battery SoC < 10 %.

[0252] Reed-switch and RF -filter place outputs in high-impedance during MRI (> 1 T).

[0253] Note: The foregoing embodiments are provided by way of example only. A person of ordinary skill in the art will recognize that dimensions, component choices, and control algorithms can be varied — e.g., different battery chemistries, alternate microcontroller families, or optical telemetry — without departing from the invention as defined by Claim 1.

[0254] Illustrative Embodiments - Battery / Power Claims 2 - 7Embodiment 2 - Rechargeable Li-Ion Power SourceAttorney Docket No. 61435-708601

[0255] In one embodiment, the electrochemical energy store is a prismatic lithium-ion pouch cell (10 mm * 18 mm x 3 mm) having a nominal voltage of 3.7 V and a 25 mWh capacity. The cell employs a LiCoCh cathode and a graphite anode. In alternative embodiments, the cathode layer may instead be:

[0256] Ni-Mn-Co (NMC 811) for higher energy (4.2 V max),

[0257] Ni-Co-Al (NCA) for fast charge,

[0258] LiFePCh (LFP) for extended cycle life at 3.6 V, ora silicon-graphite blended anode paired with any of the foregoing cathodes to boost capacity by ~ 15 %.

[0259] Each cell includes a bq29707 protector IC and a resettable PTC fuse that disconnect at 4.25 V or 80 °C, satisfying ISO 14708 safety requirements.Embodiment 3 - Primary (Non-Rechargeable) Cell Option

[0260] In another embodiment, the IPG is powered by a Li / CFXspiral-wound cell (0 14.5 mm x 25 mm, 400 mWh) that provides >10 years shelflife with < 1 % yr1self-discharge. Other suitable primaries include:

[0261] Li / CFx-MnOz hybrid bobbin cells for higher pulse current, Li / SOCL cells for -55 °C to +125 °C operation, and zinc-air buttons for low-cost, single-use research implants.

[0262] When a primary cell is selected, the recharge circuitry described below is omitted or permanently disabled by blowing a zero-ohm link.Embodiment 4 - 120 - 140 kHz Inductive Recharge

[0263] In certain embodiments, the IPG incorporates a secondary Litz-wire coil (20 turns on a 12 mm ferrite core, L ~ 11 pH) tuned to 131 kHz with a 150 nF capacitor (see Fig. 4). An external class-E driver transmits up to 250 mW. A synchronous MOSFET rectifier and CC / CV buck charger limit charge rate to < 2 C. A 10 kQ NTC bonded to the can wall pauses charging if skin temperature rises above 45 °C, ensuring compliance with IEC 60601 hot-surface limits.Embodiment 5 - 13.56 MHz NFC Recharge & Data Link

[0264] In an alternative embodiment, the same flex-PCB antenna (4-turn, 18 mm 18 mm, 2 pH) supports NFC class-2 power harvesting at 13.56 MHz (ISO / IEC 15693). The PN5180 front-end IC supplies up to 300 mW average for slow-rate top-ups while concurrently exchanging encrypted commands with a smartphone-based programmer via load modulation.Embodiment 6 - Thin-Film LiPON Solid-State Cell

[0265] In yet another embodiment, the energy store is a LiPON thin-film battery built on a 50 pm alumina substrate: Ti|LiCoO2 (1.2 pm)|LiPON (1.0 pm)|Li metal (1.0 pm)|Ti. TheAttorney Docket No. 61435-708601stack is 0.5 mm thick and delivers > 600 mWh cm-3, providing 10 mWh in a 28 mm x 12 mm footprint. Solid electrolyte eliminates leakage and permits 10 C pulse-charge without thermal runaway.Embodiment 7 - Package Envelope

[0266] Regardless of power chemistry, the IPG housing may be a capsule 45 mm x 18 mm x 8 mm, for a total volume of 24.5 cm3and a maximum thickness of 8 mm — both within the < 25 cm3 / < 9 mm limits of Claim 7. Finite-element analysis verifies structural integrity at 200 N compressive load; helium leak testing confirms hermeticity.

[0267] External-Power-Puck Embodiment (supporting Claims 52-56). In an alternative architecture the implantable pulse generator omits a long-term battery and instead receives continuous energy from an external “power-puck” module worn on the skin surface (e.g., adhered by hydrogel or held by alignment magnets). The puck contains a 300 mWh NMC lithium-ion cell, a class-E driver that excites a primary coil at 130 kHz, and a Bluetooth Low Energy transceiver. A 20-turn Litz-wire secondary coil (11 pH) inside the implant rectifies the coupled field and feeds a low-ESR 47 pF tantalum buffer capacitor that delivers instantaneous stimulation current and bridges link drop-outs of up to 200 ms (Claim 55). Bidirectional data are exchanged over BLE 5.0 (coded PHY, 125 kbit s⁻¹), spectrally isolated from the 130 kHz power carrier. System firmware (§ 6.5-6.6) executes unchanged: the real-time scheduler continues to compute or modify interleaved patterns every 100 ps, and all safety features — charge-density limiting, MRI auto-safe, AES- 128 encryption — remain active. Bench testing with 10 mm tissue phantoms demonstrated > 90 mW delivered power at ±15 mm lateral and ±30 ° angular mis-alignment providing > 3 margin over the worst-case 30 mW therapy budget. The external battery supports at least 8 hours of continuous stimulation and can be recharged via USB-C or a Qi™ pad in < 1 hour.

[0268] The disclosed system satisfies a wide range of unmet clinical needs across neuromotor rehabilitation, chronic pain, and autonomic regulation:

[0269] Stroke and traumatic brain injury (TBI). Multi-contact cuffs placed on radial, ulnar, or musculocutaneous nerves can evoke graded wrist and elbow extension; closed-loop ENG or EMG feedback allows the device to supplement — rather than override — volitional effort, making it suitable for activity-based neuro-rehabilitation and long-term functional electrical stimulation (FES).

[0270] Spinal-cord injury (SCI). When implanted in the lower limb, the system provides ankle dorsiflexion and knee extension to combat foot-drop and improve standing balance. AAttorney Docket No. 61435-708601single IPG can drive multiple cuffs, reducing the number of implanted cans relative to traditional multi-driver approaches.

[0271] Essential tremor and Parkinsonian tremor. ENG- or IMU-triggered high-frequency bursts delivered to peripheral nerves modulate afferent feedback loops and attenuate tremor amplitude within < 10 ms, enabling real-time symptom cancellation.

[0272] Chronic neuropathic pain. Closed-loop amplitude modulation based on autonomic or EMG metrics reduces habituation compared with static spinal or dorsal-root ganglion stimulators.

[0273] Autonomic disorders. Multi-contact cuffs on the vagus or carotid sinus nerves allow current-steering to preferentially activate cardio-inhibitory or anti-inflammatory fibers, while real-time baroreflex sensing tunes pulse parameters.Advantages of the single-can, battery-powered architecture

[0274] No daily coil alignment — therapy continues during sleep, bathing, and vigorous activity.

[0275] Cosmetic acceptability — the < 25 cm3, < 9 mm package fits comfortably in a limb or torso pocket, avoiding bulky external receivers.

[0276] Surgical efficiency — hard-wired or headered leads eliminate secondary receiver modules; a single incision suffices for most implants.

[0277] Scalability— a lone ±6 mA driver, combined with current-splitting and dynamic interleaving, stimulates up to 32 contacts, lowering component count and power draw compared with multi-driver ASICs.

[0278] Regulatory alignment — the MR-conditional, cryptographically secured design meets emerging standards for cyber-security (FDA 21CFR Part 882) and MRI safety (ASTM F2503), facilitating global commercialization.

[0279] Collectively, these attributes position the invention for rapid translation into stroke rehab clinics, pain management centers, movement-disorder programs, and neuroengineering research laboratories.

[0280] Structural and functional equivalents to the features described or illustrated may be substituted and fall within the spirit and scope of the claims. No element, step, or limitation is intended to be dedicated to the public regardless of whether it is expressly recited in the claims. Headings are provided solely for convenience and shall not limit the scope of the appended claims, which alone define the legal metes and bounds of the invention.4.2 Unmet Needs Addressed by the Present InventionArtefact-free simultaneous ENG sensing and stimulation.Attorney Docket No. 61435-708601

[0281] Existing cuffs either multiplex the same pad — necessitating long blank intervals — or forego ENG entirely. A platform that dynamically steers a single, clock-coherent pulse to contacts that are never in the recording set can shorten blank-gate windows to submillisecond ranges, revealing inter-pulse neural dynamics.Stimulus-energy— tuned blank-gate.

[0282] Fixed blank intervals waste recording bandwidth at low amplitudes and may still clip artefact at high amplitudes. A programmable or algorithmically adaptive blank-gate proportional to pulse charge solves both extremes and remains agnostic to future waveform shapes.Reconfigurability without re-operation.

[0283] Neuro-rehabilitation often evolves from single-muscle assist to multi -joint synergy. A contact-selectable switch matrix, plus hub or connectorized leads, lets surgeons implant once and clinicians remap pads, add cuffs, or introduce new sensor modalities — EMG, IMU, EEG — through software updates rather than additional surgery.Machine-learning and parametric control paths.

[0284] Prior systems hard-code linear mappings or amplitude tables. As patient performance stabilises, ML classifiers can improve intent specificity or artefact rejection. Embedding weight-verified neural-net inference and gated firmware-over-the-air updates fulfils this need while preserving safety.Scalable hardware for multi-cuff or sensor-fusion expansion.

[0285] Future indications — splenic anti-inflammatory loops, hepatic metabolic control — may require several nerves and heterogeneous sensors. The disclosed platform scales via (i) a switch matrix indifferent to pad count, (ii) hub-style headers for additional cuffs, and (iii) telemetry bandwidth sized for continuous multi-sensor streams.

[0286] The present invention therefore fills a critical gap by delivering a single, modular, software-defined peripheral-nerve interface that meets the artifact, adaptability, scalability, and algorithmic-flexibility shortcomings of the prior art.5 Summary of the Invention

[0287] The present disclosure provides an implantable, peripheral-nerve interface platform that unifies stimulation, sensing, and real-time control within a single, reconfigurable architecture. Key aspects include — without limitation — the following:

[0288] Selectable-contact topology. Every electrode contact of an extra-neural cuff may be assigned, under software control and without physical rewiring, either to a stimulation driver channel or to a high-impedance recording channel. Assignment may be changed pulse-by-Attorney Docket No. 61435-708601pulse, thereby permitting rapid spatial steering, tri-polar configurations, or conversion of a previously redundant pad into an active recording node.

[0289] Single, clock-coherent pulse sequence. Therapeutic energy is delivered as one charge-balanced pulse train indexed to a common time base. Individual pulses of that single sequence are routed — sequentially or concurrently — to any selected contact or subset of contacts, which avoids the “two interleaved sequences” constraint of prior art while enabling per-pulse field shaping.

[0290] Programmable and adaptive blank-gate. The recording front end is automatically blanked for a duration that is (i) clinician-programmable or (ii) computed as a function of contemporaneous pulse amplitude, pulse width, or other stimulation parameter — optionally via a regression table or a machine-learning regressor — thereby rejecting stimulus artefact without obscuring near-field neural activity.

[0291] Closed-loop control engine. A running-window metric (e.g., RMS, MAV, wavelet energy) is extracted from the recorded signal; the metric is compared to an adaptive threshold that may be derived by a constant-false-alarm-rate (CFAR) algorithm or by o-based statistics. Programmable dwell (“sustain”) intervals govern on / off hysteresis. At least one stimulation parameter — pulse amplitude, width, frequency, contact set, or any combination thereof — is updated in real time according to a mapping that may be linear, logarithmic, piece-wise, LUT -based, or machine-learned.

[0292] Optional open-loop mode. The same hardware can operate as a fixed-parameter therapy generator when neural sensing is undesirable or unavailable; ENG may still be passively logged for outcome tracking.

[0293] Multi-contact corroboration and auto-recalibration. Trigger validity may require contemporaneous threshold crossings on two or more recording pads, and baseline statistics may be re-estimated automatically whenever stimulation is inactive or when long-term drift is detected — thereby suppressing false positives while maintaining sensitivity over chronic implantation.

[0294] Dual-cuff & proximal-record / distal-stim embodiments. A first cuff may deliver stimulation while a second, more proximal cuff records intent or sensory feedback, or a single cuff may dedicate proximal pads to recording and distal pads to stimulation.

[0295] Anodal-blocking pad option. One or more guard electrodes may be energised anodically to prevent backward propagation of the stimulus artefact toward the recording region, improving artefact rejection by > 20 dB.Attorney Docket No. 61435-708601

[0296] Flexible power delivery. The pulse generator may be supplied by an integral rechargeable battery or by a passive inductive link; brown-out detection triggers an automatic safe-state or an open-loop fall-back profile.

[0297] Lead-system options. Electrode leads may be hard-wired, terminated in a single micro-connector, or fanned out through a multi-port hub that permits intra-operative or chronic attachment of additional cuffs.

[0298] Sensor-fusion expansion. Electrophysiological or biomechanical signals — EMG, EEG, mechanomyography, inertial motion, heart-rate, or other biometrics — may be combined with ENG features to enhance intent detection, artefact rejection, or adaptive control.

[0299] Rehabilitation-progress metrics (RPM). Recorded ENG snippets are post-processed to compute longitudinal markers (e.g., CAP amplitude, firing-rate histogram, conductionvelocity shift) that are telemetered to a secure cloud dashboard, thereby providing objective evidence of neural recovery independent of the control loop.

[0300] Collectively, these elements furnish a single, modular platform that can be tailored — via software alone — to acute conditioning, chronic motor restoration, pain blockade, tremor suppression, autonomic regulation, or research mapping, all while remaining within the scope of the appended claims.6. Brief Description of the Drawings

[0301] FIG 1 A - System block diagram (pulse generator, AFE, MCU / FPGA, tele-power).

[0302] FIG IB - Timing diagram: blank-gate, ENG window, pulse routing.

[0303] FIG 2 - Cuff layouts: (i) single spiral; (ii) proximal-record / distal-stim; (iii) dual-cuff

[0304] FIG 3 - Lead variants: (i) hard-wired; (ii) in-line molded connector; (iii) molded Y-hub.

[0305] FIG 4 - Adaptive blank-gate curve vs. pulse amplitude.

[0306] FIG 5 - Algorithm flow: ENG — window metric — adaptive threshold — dwell timers — stim map.

[0307] FIG 6 - Multi-contact voting logic schematic.

[0308] FIG 7 - Auto-recalibration state diagram.

[0309] FIG 8 - ML inference block (CNN / RNN) for parameter update.

[0310] FIG 9 - ENG progress-metric generation & cloud dashboard.7. Detailed Description

[0311] The following description is provided to comply with the written-description and enablement requirements of 35 U. S. C. §112(a) and to support, by way of example and not limitation, the scope of the claims appended hereto. Unless expressly defined otherwise, allAttorney Docket No. 61435-708601technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art.7.0 Purpose and Technical Advantages of the Present Filing

[0312] This application covers a next-generation peripheral-nerve interface that: (i) drives a single, clock-coherent train of charge-balanced pulses while (ii) simultaneously recording true electroneurographic signals on separate contacts, using a switch-matrix and an adaptive blank-gate to remove artefact in sub-millisecond windows. The platform lets any cuff contact be reassigned — on the fly and without surgery — to stimulation, sensing or guard roles; it then modulates amplitude, width, contact set or duty cycle in real time from the recorded ENG or from fused EMG / IMU inputs. Because all steering, blanking and control rules live in firmware, the same implanted hardware can scale from simple open-loop therapy to multi-sensor, machine-learning closed-loop control, creating a broad moat against prior art that either (a) interleaves two separate pulse trains, (b) reuses the same pad for both drive and sense, or (c) lacks software-defined contact routing and adaptive artefact suppression.

[0313] The present application is submitted to secure proprietary rights in a software-defined peripheral-nerve interface that, unlike the prior art, delivers therapeutic stimulation and acquires bona-fide electroneurographic (ENG) data without requiring interleaved or alternating pulse trains. The claimed system accomplishes this through the coordinated operation of

[0314] A time-coherent, single-sequence pulse generator capable of routing each individual, charge-balanced pulse to any subset of electrode contacts on a pulse-by-pulse basis;

[0315] A CMOS switch matrix that assigns every cuff contact — under firmware control and subject to checksum-verified safety interlock — to one of four instantaneous roles: stimulation, high-impedance recording, anodic guard, or open circuit;

[0316] An adaptive blank-gate module that blanks the recording front-end for a programmable or stimulus-energy-scaled interval that decays in sub-millisecond time, thereby suppressing stimulus artefact while exposing inter-pulse neural activity; and

[0317] Closed-loop control logic — parametric or machine-learned — that updates, in real time, at least one stimulation parameter (current amplitude, phase width, pulse frequency, duty cycle or contact set) as a deterministic or algorithmic function of the contemporaneously recorded ENG metric, optionally fused with auxiliary physiological signals (EMG, EEG, inertial, biometric).Attorney Docket No. 61435-708601

[0318] Because every element required for field shaping, artefact rejection and feedback control is implemented in re-programmable firmware, the same implanted hardware can be re-configured — post-implant and without surgical revision — to operate in:

[0319] 1. Simultaneous stimulation-and-record, closed-loop mode for motor restoration, pain block or autonomic regulation;

[0320] Therapy-only open-loop mode when neural sensing is unavailable, undesirable or contraindicated; and

[0321] Hybrid, software-upgradable modes incorporating multi-contact voting, auto-recalibration of thresholds, adaptive charge-density ceiling, inductive-only power, hot-plug cuff expansion and secure machine-learning inference.

[0322] This architecture provides at least the following technical advantages over the closest prior art:

[0323] No shared-pad artefact penalty - sensing never occurs on the same physical contact used for the immediately preceding pulse, permitting blank-gate windows < 1 ms versus > 5 ms in systems that time-multiplex a single pad.

[0324] No requirement for two interleaved therapy sequences - therapeutic energy is delivered as a single pulse train referenced to one master clock, eliminating duty-cycle loss and scheduling complexity inherent to dual-train approaches.

[0325] Software-selectable contact assignment - any stimulation or recording site may be reassigned by authenticated firmware update, enabling long-term therapy evolution (e.g., expanding from single-muscle assist to multi -joint synergy) with zero hardware changes.

[0326] Energy-scaled blank-gate - the duration of recording blanking scales monotonically with pulse charge, preserving bandwidth at low amplitudes while preventing residual artefact at high amplitudes, independent of waveform shape.

[0327] Future-proof algorithmic layer - control mappings may be linear, lookup-table, or neural-network based; encrypted, checksum -verified firmware-over-the-air (FOTA) permits safe deployment of improved classifiers without violating primary safety rails.

[0328] Accordingly, the claimed subject matter establishes a broad and defensible moat around non-interleaved, selectable-contact stimulation with concurrent ENG sensing, adaptive artefact suppression and multi-sensor closed-loop modulation — capabilities not taught or suggested by Shon et al. (artefact-limited shared-pad), Neurinnov EP 4340935 (dual-train interleaving) or other referenced disclosures.7.1 Definitions

[0329] As used in this specification and the appended claims:Attorney Docket No. 61435-708601

[0330] “Contact” means an electrically conductive region disposed on, in, or within an extraneural substrate and configured to exchange charge with adjacent neural tissue. A contact may comprise platinum-iridium, gold, stainless steel, conductive polymer, conductive carbon, or any other biocompatible conductor and may optionally include a surface modification such as PEDOT: PSS, iridium oxide, titanium nitride, or diamond-like carbon.

[0331] “Pulse sequence” means an ordered series of charge-balanced, monophasic or biphasic current or voltage waveforms that share a common time base or clock. The sequence may be delivered to the same contact or to different contacts on a pulse-by-pulse basis and may include pulses that are delivered concurrently to more than one contact.

[0332] “Blank-gate” means an interval during which an amplifier, analog-to-digital converter, or other sensing circuit is disabled, isolated, or otherwise prevented from acquiring neural signals in order to suppress or avoid stimulus artefact. The blank-gate interval may be fixed, clinician-programmable, or automatically computed as a function of one or more stimulation parameters.

[0333] “Metric” means any numerical value or vector derived from one or more samples of a neural or auxiliary signal, including but not limited to root-mean-square (RMS) amplitude, mean absolute value, short-time Fourier magnitude, wavelet coefficient, auto-regressive prediction error, or machine-learned feature.

[0334] “Dwell interval” (also “sustain interval”) means a programmable time period that must elapse with the metric above (ON dwell) or below (OFF dwell) a threshold before stimulation is enabled or disabled, respectively.

[0335] “Auto-recalibration” means automatic adjustment, without clinician intervention, of at least one control parameter — e.g., threshold, blank-gate duration, gain, or calibration constant — based on statistics computed from ongoing recorded data.

[0336] “Multi-contact voting” means a decision process in which a trigger condition is required to be detected contemporaneously on two or more contacts, within a correlation window of < 2 milliseconds, before stimulation is enabled.

[0337] “Electrophysiological signal” collectively encompasses electroneurographic (ENG), electromyographic (EMG), electroencephalographic (EEG), mechanomyographic (MMG), inertial, kinematic, and biometric signals.7.2 Hardware Embodiments7.2.1 Receiver Electronics (Implant Core)

[0338] In a preferred but non-limiting embodiment the implant core is realised as a single monolithic application-specific integrated circuit (“ASIC”). The term ASIC is employedAttorney Docket No. 61435-708601herein to encompass any custom or semi-custom silicon die, including a system-on-chip (“SoC”), a multi-chip module, or a flip-chip assembly bonded on a ceramic interposer.Functional partitioning may therefore be rearranged — e.g., the microcontroller and telemetry radio may reside on a companion die — without departing from the spirit and scope of the invention.

[0339] Stimulation driver. The ASIC includes at least one programmable bi-phasic current source-sink pair capable of sourcing or sinking 0.05 mA to at least 10 mA per phase with a compliance voltage that is preferably > 100 V, thereby accommodating high-impedance extraneural contacts in a safety margin exceeding IEC 60601-2-10. Output resolution is not less than 8 bits and may be achieved by a segmented binary-weighted DAC, a current-mode sigma-delta modulator, or any other digitally controlled current mirror. Pulse width is programmable from 10 ps to 1 ms in < 5 ps increments, suitable for cathodic-first or anodic-first symmetric or asymmetric charge-balanced waveforms.

[0340] Recording front-end. A fully differential low-noise amplifier (“LNA”) exhibits an input-referred noise density < 30 nV / ^Hz, equivalent to < 2 pVRMS integrated over a 300 Hz-5 kHz pass-band, and an input impedance > 1 G 2 || 5 pF to ensure negligible loading of microampere ENG signals. Programmable gain stages (×20–×100) enable optimization for either ENG (< 100 pVpp) or EMG / EEG (> 500 pVpp). Input protection clamps and ESD diodes satisfy ISO 14708-3 section 25.

[0341] Switching matrix. A 16 x 16 CMOS cross-point array (or, in reduced-pad embodiments, an 8×n array) routes any stimulation channel or recording channel to any cuff contact. Each cross-point incorporates a low-leakage transmission gate with < 50 Q on-resistance and < 1 pA off leakage at 37 °C. A configuration register protected by a 16-bit cyclic-redundancy checksum (“CRC-16”) is verified on every power-cycle; an error disables stimulation until a valid map is re-loaded, thereby implementing the safety interlock recited in the claims.

[0342] Data conversion. Neural signals are digitised by an oversampling 12-bit successive-approximation (“SAR”) ADC operating at > 50 kS / s per channel; higher-rate (> 200 kS / s) S-A or pipelined converters may be substituted to support artefact-tolerant oversampling or multi-band ENG-EMG acquisition.

[0343] Digital processor. Control logic is executed by an on-die ARM® Cortex-M-class microcontroller (or equivalent RISC core) clocked at 8 MHz-120 MHz, equipped with 128 kB-2 MB of flash and 64 kB-512 kB of SRAM. DSP extensions (MAC, barrel shifter) accelerate real-time envelope, CFAR, FFT and CNN layers. Alternative embodiments mayAttorney Docket No. 61435-708601replace or augment the MCU with an FPGA fabric, hardware finite-state machine, or a neural -network inference engine.

[0344] Telemetry & power. The ASIC integrates a full-duplex load-shift-keying (“LSK”) modem for inductive power carriers in the range 100-150 kHz and a 2.4 GHz or MICS-band GFSK transceiver for bidirectional data, firmware updates and cloud connectivity. For battery-powered variants, an on-chip charger manages a Li-ion or Li-polymer cell (10 pAh-200 mAh). For passive-receiver variants, a synchronous rectifier and low-dropout regulator furnish a regulated 1.8 V-3.6 V rail to the digital core and a 5 V-12 V rail to the current drivers.

[0345] Packaging. The die (or die stack) is flip-chip bonded to a Ti-Al-Nb or alumina substrate that incorporates a hermetic 16-line feed-through brazed or laser-welded to the implant housing. Feed-through pins are routed to 90-10 platinum -iridium weld pads for the lead wires. Alternate embodiments may employ glass-to-metal-seal, laser-ceramic, or polymer-feed-through technologies.

[0346] Those skilled in the art will recognise that any of the foregoing functional blocks may be implemented in discrete components, folded into additional digital or analog cores, or migrated off-chip without departing from the scope of the invention as claimed.7.2.2 Cuff Fabrication

[0347] In an illustrative embodiment the extraneural interface is realised as a spiral-wrap cuff that completely encircles the target nerve without penetrating the epineurium.Alternative geometries — e.g., split-ring, flat-interface (FINE), semi-cylindrical “half-cuff,” or helical-overlap designs — may be substituted to accommodate nerve diameters ranging from 200 μm to 6 mm, all such variations falling within the term “cuff electrode” as used herein.

[0348] Substrate. The body of the cuff is moulded or die-cut from implanted-grade silicone elastomer having a Shore-A hardness between 40 and 70. A thickness of 100-300 μm affords a bend radius < 1 mm while providing sufficient tear strength (> 20 kPa) for chronic implantation. Radiopaque tungsten or barium-sulfate markers may be co-moulded for fluoroscopic localisation.

[0349] Conductive tracks and contacts. In one embodiment 25 μm platinum-iridium (90-10 or 80-20) foil is laser-patterned into circumferential bands that serve as contacts and longitudinal bridges that serve as lead traces. Trace width is selectable between 100 iim and 500 iim to meet IEC 60601-1 creep-distance requirements. Where higher compliance voltage is desired, foil thickness may be increased to 50 pm or multiple foils may be laminated. OtherAttorney Docket No. 61435-708601biocompatible conductors — e.g., Pt, Pt-black, IrOx, TiN, TiC, conductive diamond-like carbon, or sputtered Au — may be employed interchangeably.

[0350] Surface modification. To reduce impedance and polarisation, the active surface of each contact may be electrochemically coated with poly(3,4-ethylenedioxythiophene)-poly(styrenesulfonate) (PEDOT: PSS) to an areal charge density of 0.5-2 mC cm’2, thereby achieving an impedance of < 10 k 2 at 1 kHz for a 1 mm2pad. Alternate coatings such as iridium oxide (AIROF), titanium nitride, or platinum black may be substituted, and mixed-layer stacks (e.g., IrOx / PEDOT bilayers) are expressly contemplated.

[0351] Contact patterning and spacing. The cuff may incorporate 2-16 circumferential contacts arranged in one or multiple longitudinal rows. A preferred two-row embodiment allocates distal-row contacts for stimulation and proximal-row contacts for recording; in a single-row embodiment individual contacts are dynamically reassigned by the switching matrix. Inter-contact gaps of 200-800 iim minimise shunt capacitance while permitting multi-contact voting as claimed.

[0352] Lead wires and strain relief. Individual conductor wires are 7 / 44 AWG MP35N, Pt-Ir, or annealed Pt ribbon, helically wound to a pitch of 4-6 turns cm’1to withstand > 106flex cycles at 30° bend. Wires are bonded to the foil using laser microwelding, resistance welding, or a gold-stud / thermo-compression process and then over-moulded in silicone. A silicone-rubber “wing” or grommet may be integrally moulded at the proximal cuff edge to serve as a strain-relief anchor and suture tie-down.

[0353] Optional anodal block pad. Where anodal blocking is desired, an additional circumferential band, identical in material to the stimulation contacts, is positioned 0.5-3 mm proximal to the most proximal recording contact. The pad is connected to a dedicated driver line and may be driven with a DC-offset or tonic anodic pulse to attenuate stimulus propagation toward the recording region.

[0354] Closure mechanism. For spiral -wrap cuffs, a self-biasing hook-and-loop tongue or interlocking dovetail may provide mechanical closure; for split-ring cuffs, silicone-embedded sutures or an over-centre latch may be employed. All closures are configured to maintain contact-to-nerve radial pressure below 10 kPa to preserve intraneural perfusion.

[0355] Sterilisation compatibility. Materials and assemblies described herein are compatible with ethylene oxide, low-temperature hydrogen-peroxide plasma, or gamma (25 kGy) sterilisation processes.Attorney Docket No. 61435-708601

[0356] All dimensional values are exemplary and may be scaled or interchanged to accommodate different nerves, patient anatomies, or manufacturing tolerances without departing from the scope of the appended claims.7.2.3 Anodal-Blocking Pad

[0357] To further mitigate stimulus artefact that might otherwise couple into the recording channel, the cuff optionally incorporates at least one anodal-blocking electrode (“block pad”) fabricated from the same conductive foil or thin-film stack used for the primary contacts. Unless specified otherwise, the terms block pad and anodal pad are used interchangeably.

[0358] Geometric placement. In a single-row cuff the block pad is disposed longitudinally between the distal-most stimulation contact and the most proximal recording contact at an inter-pad spacing of 0.5 - 3 mm centre-to-centre. In a dual-row cuff the block pad may reside on either row, preferably the row that lies closer to the epineurium, or may be implemented as a circumferential “guard ring” encircling the entire recording group. Multiple block pads may be placed in series to form a distributed ladder for extended shielding.

[0359] Electrical drive modes. The block pad is connected to a dedicated output of the stimulation driver and may be energised in any of the following, non-limiting modes:(i) Dynamic return mode: During each cathodic stimulation phase the block pad is driven anodically at substantially the same current magnitude and phase width as the cathodic phase (minus interconnect parasitics), thereby creating a local return path that terminates radial current flow and prevents propagation toward the recording contacts.(ii) Tonic DC offset mode: The pad is biased with a constant anodic offset of 5 - 100 pA whenever stimulation is active, independent of per-pulse timing, to establish a steady electricfield barrier.(iii) Adaptive impedance-matched mode: The driver samples the instantaneous artefact on the recording pad and adjusts the block-pad current in closed loop to minimise residual artefact energy; suitable control algorithms include least-mean-squares (“LMS”) or model-predictive estimation.

[0360] Charge-balance compliance. For modes (i) and (iii) the control circuitry ensures that the time-integral of net charge delivered by the block pad over any stimulation second does not exceed 1 pC to remain within chronically safe electro-chemical limits. A blockingcapacitor option (0.1 - 2.2 pF) may be inserted in series as further assurance of zero net DC.

[0361] Artefact suppression performance. Bench measurements in phosphate-buffered saline (PBS) demonstrate that positioning the block pad 1.5 mm proximal to the firstAttorney Docket No. 61435-708601recording pad and energising it in dynamic return mode attenuates residual artefact at the recording amplifier input by > 20 dB, equating to a reduction from ca. 1 mV_pk to < 100 pV_pk for a 2 mA, 300 ps stimulation pulse.

[0362] Material, surface, and size. The block pad uses the same Pt-Ir (or alternative conductor) construction as the primary contacts to avoid galvanic mismatch. Surface area equals or exceeds the aggregate area of all stimulation contacts engaged in the active pulse, thereby keeping charge density per phase below 0.1 mC cm2.

[0363] Programmability. Block-pad mode, current amplitude, and bias profile are clinician-selectable parameters accessible via the external programmer UI and may be altered intra-operatively or post-implant under authenticated session control.

[0364] Alternative embodiments may omit the block pad altogether, may use multiple pads arranged circumferentially or longitudinally, or may employ a resistive or capacitive shunt integrated into the cuff substrate; all such variations are considered equivalents for purposes of the appended claims.7.2.4 Lead Variations

[0365] The invention contemplates several interchangeable architectures for electrically coupling the implant core to one or more cuff electrodes. Each architecture is described with exemplary dimensions and materials that may be varied by ±50 % without departing from the scope of the claims.Permanently-moulded lead (monolithic lead-over-device configuration).

[0366] A bundle of MP35N or Pt-Ir conductors — typically 4 to 24 strands of 7 / 44 AWG — is helically wound to a pitch of 4-6 turns cm1and over-moulded in medical-grade, platinum-cured silicone elastomer to an outer diameter of 2.0-3.5 mm. The over-mould extends continuously from the distal cuff strain-relief wing to the proximal implant header so that no intermediate junctions or connector interfaces are present. Accelerated bend testing (ISO 14708-1 Annex G) demonstrates a fatigue life of > 1,000,000 cycles at 30° flex and a tensile pull strength of > 40 N prior to conductor breakage. A radiopaque marker stripe may be coextruded to facilitate fluoroscopic tunnelling. Because the lead is integral, ingress risk is limited to the hermetic feed-through.Integral nano-connector (single-port disconnectable lead).

[0367] Where explant-ability or modularity is desired, the receiver header incorporates a two-row nano-connector having 16, 18, or 20 pins on < 0.635 mm pitch (e g., Omnetics Nano 360® or Samtec NovaRay®). The connector shell is recessed in a silicone or PEEK over-mould and features a positive-locking latch or titanium set-screw that is surgeon-Attorney Docket No. 61435-708601operable with a standard 0.9 mm hex driver. Electrical contacts are gold-plated berylliumcopper or titanium-nitride coated to ensure < 20 mQ contact resistance and MRI conditionality up to 3 T (ASTM F2503). A silicone boot, moulded over both plug and socket after mate, forms an IPx8 barrier and furnishes strain relief with > 35 N pull-out force. The connector permits intra-operative substitution of cuff variants and simplifies lead routing through tortuous anatomy.Multi-port Y-hub (dual-cuff expansion).

[0368] In an alternate embodiment a silicone-encapsulated Y-hub is placed inline between the receiver lead and two downstream cuff leads. The hub houses two or more female microconnectors (e.g., 8-pin nano latch), each keyed to prevent polarity reversal. Unused ports are sealed with pre-attached silicone plugs that are factory-bonded and detachable only under surgical visualisation; leakage tests per ISO 10993-18 show < 0.5 pA at 10 V for 72 h saline soak. Internal wiring fans-out the 16 feed-through pins into two logically independent bundles, enabling (a) a dedicated stimulation cuff and (b) a dedicated recording cuff, or any other pairing claimed herein. Overall hub diameter is < 6 mm to permit subcutaneous tunnel and to satisfy cosmetic profile requirements for the upper arm.Common features and alternatives.All lead variants incorporate:

[0369] Redundant conductor paths for critical grounds and reference electrodes.

[0370] Retro-reflective laser-engraved serial codes for traceability.

[0371] Optional antimicrobial silicone additives (< 0.5 % Ag-glaze) at the skin exit site for percutaneous prototypes.

[0372] Any of the above lead architectures may be substituted, mixed, or re-dimensioned to accommodate alternate cuff counts, implant locations, or mechanical specifications without departing from the invention as defined by the appended claims.7.2.5 Power-Supply Architectures

[0373] The inventive system may obtain operating energy either from an internal electrochemical cell or from a passive inductive link; both modalities are fully supported by the pulse-generator and telemetry circuits described in §7.2.1. Parameters recited herein are exemplary and may be re-scaled without departing from the scope of the claims.(a) Rechargeable IPG (active implant with on-board battery).

[0374] A hermetically sealed implantable pulse generator (“IPG”) incorporates a lithium-ion or lithium-polymer secondary cell having a nominal capacity selectable from 10 pA-h to 200 mA h The cell is managed by an integrated charger that harvests energy from aAttorney Docket No. 61435-708601transcutaneous resonant inductive link centred at 150 kHz ± 10 kHz. Primary (external) and secondary (implant) coils are wound on solenoid or planar ferrite formers and resonate with respective tuning capacitors to achieve a loaded Q-factor of 20-50. Efficiency at a skin-to-coil separation < 10 mm exceeds 65 % under typical alignment.

[0375] Charging current is limited to C / 10 or 10 mA, whichever is smaller, and cell voltage is clamped to 4.15 V ± 50 mV to satisfy IEC 62133-2. In the illustrative operating regime — 20 Hz stimulus frequency, 2 mA cathodic-first biphasic pulses, 300 ps per phase — the battery supports > 24 h of uninterrupted therapy before requiring recharge. A coulomb counter provides run-time estimation accuracy better than 5 % of full scale. During charging the neurostimulation output may operate, reduce duty cycle, or be inhibited, all selectable via the external programmer.(b) Passive RF-powered receiver (battery-free implant).

[0376] In a second embodiment the implant is entirely battery-less. A ferrite-cored multiturn coil — outer diameter 18 mm, inductance ~ 28 pH — resonates at 120-140 kHz. An on-chip synchronous rectifier and low-dropout regulator furnish a 5 V rail capable of > 20 mW continuous output, which is sufficient to drive a 4-mA, 100-V biphasic pulse pair at 20 Hz with < 50 % duty-cycle margin.

[0377] Down-link data (external-to-implant) are amplitude-shift keyed onto the power carrier at 2-10 kb s'1; up-link data (implant-to-external) are conveyed by load-shift keying (LSK) using a two-level modulation depth of 15 %. Higher-rate telemetry, firmware updates, and streaming diagnostics are transported on an auxiliary 2.45 GHz GFSK link operating at < 100 pW ERP and compliant with ETSI EN 300328 duty-cycle limits.

[0378] The passive receiver automatically enters a quiescent state (< 10 pW) when the external field is removed; volatile state variables are shadowed in ferroelectric RAM to survive power-down, thereby enabling “plug-and-play” home use with a garment-mounted or bedside transmitter.(c) Fail-safe and hybrid considerations.

[0379] Both power architectures share a common brownout monitor; if rail voltage falls below 2.7 V the microcontroller disengages current drivers and re-arms only when the rail stabilises above a programmable hysteresis threshold. A hybrid embodiment employing a small (e.g., 100 pA-h) buffer cell may be used to ride through brief mis-alignments of the external coil, but the presence or absence of such a buffer does not limit the claims.Attorney Docket No. 61435-708601

[0380] All energy-storage and power-transfer components are selected or dimensioned to maintain surface temperature rise below 2 °C under worst-case charging and stimulation duty cycles, in accordance with ASTM F2182.7.3 Pulse-Routing and Switching Matrix

[0381] In order to realise the “single-sequence, multi-contact” stimulation paradigm recited in the independent claims, the implant core incorporates a digitally addressable routing network that can present any stimulus pulse — or set of pulses — to any desired subset of electrode contacts on a pulse-by-pulse basis while simultaneously providing high-impedance isolation for the remaining contacts assigned to recording. Two illustrative hardware realisations are described; other topologies (e.g., time-domain multiplexers, MEMS relays, or micro-liquid metal switches) may be substituted without departing from the claimed invention.(a) Cross-point array (full flexibility embodiment).

[0382] A preferred implementation employs a 16 x 16 CMOS bilateral-switch matrix fabricated in a 180-nm or smaller mixed-signal process. Each cross-point comprises complementary transmission gates with on-resistance < 50 at 5 V gate over-drive and off leakage < 1 pA at 37 °C. The array thus exposes 16 “column” nodes (electrode contacts) and 16 “row” nodes that may be tied, under firmware control, to (i) the stimulation driver outputs, (ii) the recording pre-amplifier inputs, (iii) a guard potential, or (iv) a high-impedance state. Switching transients are suppressed by break-before-make logic that enforces a minimum 200 ns interlock between row and column actuation.(b) Time-division distribution example.

[0383] Suppose a sequence of n equally spaced pulses {Pl... PII} repeats at fundamental rate fEll = 20 Hz. A round-robin scheduler maps Pl — Cl, P2 — C2,..., Pl 1 — Ci, thereby sweeping the stimulus focus around the nerve circumference once every 1 / fEll seconds. This achieves spatial steering while maintaining a single clock-coherent pulse sequence and therefore lies outside the scope of patents that require two interleaved trains.(c) Concurrent delivery example.

[0384] At time tO a single pulse is routed concurrently to contacts Cl and C2 by closing both cross-points (RowSTIM, Coll) and (RowSTIM, Col2). Because the same current source is reflected through equal-value ballast resistors, per-contact charge density is halved while the total recruited fibre volume increases. Concurrent delivery is particularly advantageous for large-diameter nerves or when uniformly distributed activation is clinically desirable.(d) Reduced-pad multiplexer (space-saving embodiment).Attorney Docket No. 61435-708601

[0385] For smaller cuff variants an 8 x n analog multiplexer (where n = 6-12) routes a fixed number of stimulation channels to a larger number of electrode contacts on a pulse-scheduled basis. The multiplexer may be implemented with high-voltage SOI FETs to withstand compliance voltages up to 120 V. Although not offering full any-to-any connectivity, this topology still enables (i) dynamic reassignment of which contacts are “stim” vs. “record,” and (ii) time-division steering across multiple contacts — all that is required to satisfy the independent claims.(e) Configuration integrity and safety interlock.

[0386] A pin-map table — a 32- to 64-byte structure stored in non-volatile flash or FRAM — defines the active routing for each contact. Upon every power-up or brown-out recovery the microcontroller computes a CRC-16 or CRC-32 over the pin-map; if the computed checksum differs from the stored checksum, or if the map fails semantic rules (e.g., a contact marked simultaneously as “stim” and “record”), the firmware inhibits the stimulation driver and raises a wake-up flag over telemetry. This checksum gate constitutes the “safety interlock checksum” element recited in the dependent claims and prevents latent or single-event-upset memory faults from creating unsafe pin states.(f) Electrical isolation during recording.

[0387] While a contact is assigned to the recording front end the associated switch devices connecting that contact to any stimulation node are forced open, guaranteeing a minimum off-isolation > 60 dB at 5 kHz. Conversely, when a contact is assigned to stimulation the corresponding recording switches are opened to prevent amplifier saturation. Isolation timing is synchronised with the blank-gate controller to ensure that switch commutation never coincides with the sampling aperture of the ADC.(g) Scalability.

[0388] The matrix may be expanded to 32 x 32 or folded to 4 x n by standard layout techniques. Because routing decisions are executed by firmware prior to each pulse, the architecture supports future upgrades — such as tri-polar waveforms, tripolar-sequential steering, or ML-optimised pulse selection — without silicon re-spin.

[0389] Thus the pulse-routing and switching matrix delivers the requisite contact- to-contact flexibility, artefact isolation, and safety validation necessary to practise the invention in its broadest scope while providing several manufacturable embodiments for differing size, cost, and channel-count constraints.7.4 Blank-Gate ModuleAttorney Docket No. 61435-708601

[0390] The term “blank-gate” denotes an interval that commences at a reference point within, or immediately following, a stimulation pulse and during which the recording front end is placed in a high-impedance or otherwise non-acquisitive state so as to suppress stimulus artefact. The invention provides a parameterised blank-gate generator that may operate in fixed, linear, piece-wise, or algorithmically adaptive modes; each mode is fully disclosed herein so as to satisfy 35 U. S. C. §112 while reserving maximal design latitude. (a) Parametric definition.

[0391] For any given pulse ] having peak current amplitude Ipeak,j (mA) and, optionally, cathodic-phase width Wj (ps), the blank-gate duration Atblank,j is computed according to \Deltat_{\text{blank},j}\;=\; \mathcal{F}\!\left(I_{\text{peak},j},\, W_{j}\right) where F(-) is one of:1. Linear model2. Lookup-table modelTable Indexed by with bilinear interpolation*

[0393] 3. Piece-wise modelIf A ~~~ ~~~

[0394] 4. Machine-learning model

[0395] A regression network that outputs Atblank given a feature vector that may include Ipeak, W, electrode impedance, temperature, or long-term charge-accumulation history.

[0396] The linear form (1) is preferred for its computational simplicity and is used herein for illustrative numeric bounds.(b) Programmable coefficients and bounds.

[0397] Coefficients k (ps mA'1) and c (ps) are stored in non-volatile memory and are accessible in a clinician GUI subject to dual-credential authentication (engineer + physician). Independent safety limits are enforced in firmware:

[0398] Attorney Docket No. 61435-708601

[0399] Factory defaults are Atmin = 0.5 ms and Atmax = 5.0 ms; the user may tighten but not exceed these limits without developer firmware.(c) Implementation timing.

[0400] Start marker. The blank-gate timer is armed by the rising edge of the cathodic phase (or, in an anodic-first waveform, the anodic phase) and expires after Atblank.

[0401] Record-enable. Upon expiry, the recording channel bias network is re-connected via fully differential, make-before-break switches to prevent inrush transients.

[0402] Switch-matrix interlock. If the next pulse is scheduled on different contacts < Atblank after the present pulse, the switch matrix obeys break-before-make timing with an additional guard of 50 ps to allow LNA settling.

[0403] Physiologic hand-off. Immediately upon expiry of At blank the recording front end re-biases and the ENG processing chain (see § 7.5) resumes acquisition. Stimulation remains ON only while the post-blank endogenous ENG metric exceeds the adaptive threshold; when the metric falls below threshold for the OFF-dwell interval, stimulation is disabled. Because the threshold is updated continuously (a-based or CFAR) it remains high enough to reject transient artefact spikes yet low enough to detect genuine motor-intent activity, thus ensuring that the stimulator is driven solely by physiologic signal, not by residual artefact or noise. (d) Per-contact and class-based blanking.

[0404] In advanced embodiments Atblank may be computed per stimulation contact group so that contacts with higher path impedance receive proportionally longer blanking, while contacts connected to a blocking anode may require shorter blanking. Alternatively, contacts may be grouped into “classes” (e.g., large-diameter vs. small-diameter nerves) each with its own pair of k and c coefficients.(e) Auto-adaptation and fail-safe.

[0405] During normal closed-loop operation the control software monitors residual artefact amplitude on one or more sacrificial “test contacts.” If artefact exceeds a programmable ceiling (e.g., 100 pV_rms) the system extends k by 10 % per violation up to Atmax.Conversely, if artefact remains below a floor (e.g., 20 pV rms) for > 1 min, k is decremented toward its nominal setting in 5 % steps. Should Atblank ever be commanded below Atmin by corrupted memory or telemetry, the stimulus driver is inhibited and a Class III fault is telemetered.(f) Clinician interface.Attorney Docket No. 61435-708601

[0406] A graphical widget displays the present k, c, Atblank, Atmin, and Atmax. The clinician may toggle fixed versus adaptive mode; in adaptive mode maximum slew rate is limited to 10 ps min'1to guard against oscillatory behaviour.

[0407] The foregoing architecture supplies a blank-gate mechanism that (i) is parameterisable and clinician-tunable, (ii) scales with stimulus energy to suppress artefact without unnecessarily masking neural data, and (iii) incorporates auto-recalibration safeguards, thereby satisfying the breadth and specific embodiments recited in the accompanying claims.7.5 Signal-Processing Pipeline

[0408] The control circuitry executes a real-time pipeline that converts raw electroneurographic (“ENG”) samples acquired from the recording contact(s) into one or more decision variables that govern stimulation. The pipeline is modular: each stage may be substituted, bypassed, or supplemented by firmware update without hardware modification, thereby furnishing the scalability contemplated by the independent claims.(a) Pre-conditioning and decimation.

[0409] Raw samples are (i) high-pass filtered at < 300 Hz to remove drift, (ii) low-pass filtered at > 5 kHz (or < 3 kHz when artefact is low) to satisfy Nyquist, and (iii) decimated to a frame rate fD = 1 kHz for envelope processing. Alternate embodiments may use multi-rate filter banks, wavelet packets, or adaptive bandwidth reduction.(b) Window-metric computation.Default implementation (20 ms RMS).

[0410] For each decimated sample window W of length N = 20 samples (20 ms at 1 kHz) the root-mean-square isAlternative implementations.

[0411] Mean-absolute value (MAV). A 5 ms window of absolute values.

[0412] Spectral-bin magnitude. Magnitude of a single short-time-Fourier (“STFT”) bin centred at 1.5-3 kHz.

[0413] Wavelet or CNN feature. Any scalar or vector produced by a machine-learning feature extractor.Attorney Docket No. 61435-708601

[0414] The particular metric is selected from a drop-down menu in the clinician GUI and may be changed in vivo.(c) Thresholding schemes.7.5.1 o-based global threshold.

[0415] The metric’s long-term mean p and standard deviation o are estimated using an exponential moving average with a programmable time constant (typ. 2-10 s). A triggerwhere k is clinician-programmable in 0.1 increments over the range 2 < k < 5.7.5.2 Smallest-of CFAR (“SO-CFAR”) local threshold.

[0416] A sliding reference window of 25 samples on each side of the cell-under-test (“CUT”) produces leading and lagging sums SL and SR. The threshold is the firmware permits dynamic switching between global o-based and SO-CFAR modes, or an external host may blend the two by logical OR / AND to create hybrid detectors.(d) Dwell-interval logic.

[0417] Two independent programmable timers implement hysteresis:

[0418] ON dwell interval AtON programmable 1-100 ms in 1 ms steps. Stimulation is enabled when the trigger condition persists continuously for AtON.

[0419] OFF dwell interval AtOFF programmable 0-50 ms in 1 ms steps.

[0420] Stimulation is disabled when the trigger condition remains false for AtOFF.

[0421] Both timers are clocked by the same decimated sample rate fD and therefore introduce deterministic, bounded latency compliant with IEC 60601-1-10. The ratio AtON / AtOFF is clinician-adjustable to trade sensitivity for specificity.(e) Adaptive and machine-learning variants.The pipeline may be extended by:

[0422] Auto-adjusted k-factor. If the observed false-trigger rate exceeds a programmable ceiling the firmware increments k; if ENG amplitude drifts downward, k is decremented toward its nominal set-point.Attorney Docket No. 61435-708601

[0423] Neural network discriminator. A classifier (e.g., 1-D CNN with 3 convolutional layers, 128 parameters) can replace or augment the dwell logic, outputting a probabilistic trigger flag.

[0424] Multi-sensor fusion. Feature vectors combining ENG metrics with EMG, IMU, or EEG features are concatenated and supplied to a decision tree or DNN that outputs both trigger and stimulation magnitude.(f) Data integrity and telemetry.

[0425] All intermediate metrics (p, c, k, KSO, AtON, AtOFF) are stored in redundant register files with parity protection and are mirrored to the external programmer at < 1 Hz for transparency and audit.

[0426] This flexible, parameterised signal-processing chain supports every control strategy recited in the claims — from simple fixed-threshold triggering to fully adaptive or machine-learned modulation — while offering ample hooks for future firmware expansion without silicon change.7.6 Control-Function Library

[0427] Once a trigger condition has been met, at least one stimulation parameter — preferably pulse amplitude A — is updated in real time according to a control function F(M, 0), where M is the window metric defined in § 7.5 and 0 represents one or more coefficient sets, breakpoints, or network weights. A library of interchangeable control functions is described below to ensure full enablement of the broad “modify at least one stimulation parameter in real time” limitation of the independent claims.(a) Linear gain model.

[0428] a is a slope coefficient (0 < a < 25 mA V1) and P an offset (0-3 mA). Both coefficients are stored in non-volatile memory (NVM) as 12-bit fixed-point numbers and may be modified via the clinician programmer subject to hard safety rails (total charge < 0.35 pC phase'1cm'2).(b) Logarithmic gain model.A ™ -y ln(Af ) 4- A

[0429] Used when the dynamic range of M spans > 20 dB.Implementation utilises the natural-log LUT in the MCU DSP library; overflow is clamped at the maximum safe current.(c) Piece-wise or LUT mapping.Attorney Docket No. 61435-708601

[0430] An up to 16-segment piece-wise-linear curve is stored as successive (M, A) breakpoints in NVM. Linear interpolation produces sub-segment values. This affords arbitrary monotonic or non-monotonic profiles, e.g., dead-zone, saturating, or sigmoid shapes. Breakpoints may be installed at factory or uploaded during a programming session and are checksummed along with the pin-map.(d) Machine-learning controller.

[0431] On-device CNN. A 1-D convolutional neural network of the form Conv(8,3) — ReLU - Conv(4,3) - ReLU - FC(32) - ReLU - FC(1) outputs A directly. Total footprint ~ 2 k parameters, enabling inference in < 500 ps on a 64-MHz Cortex-M4 using CMSIS-NN.

[0432] External RNN. For higher-complexity regimes an LSTM network executes on a wrist-mounted or cloud processor; predicted A values are telemetered to the implant with < 30 ms round-trip latency. Secure firmware-over-air (“FOTA”) allows model-weight updates signed with 256-bit ECDSA.(e) Multi-sensor fusion.X EMGRMS, cJMU, heart. rate, J

[0433] A feature vector isassembled every control cycle. A gradient-boosted decision tree (128 nodes) or neural net (e.g., fully-connected 32-16-1) maps X to output A and, optionally, to ancillary parameters such as pulse width or duty cycle. Sensor weights can be dynamically re-weighted; e.g., during gait the IMU channel may receive higher gain than during seated tasks.(f) Safety and fall-back logic.

[0434] Regardless of the active F( ), a watchdog compares the requested A to a hard upper safety ceiling stored in one-time-programmable (OTP) memory. If A exceeds the ceiling or is NaN / inf, the driver reverts to a clinician-defined default amplitude and logs a Class-II fault.(g) Auto-recalibration hook.

[0435] When auto-recalibration (see § 7.7) updates the threshold or blank-gate, the control function receives the new baseline via an interrupt and re-centres M accordingly, preventing drift-induced excessive stimulation.

[0436] Through the foregoing library the system may execute deterministic linear control, non-linear heuristic control, or adaptive machine-learning control, individually or in hybrid cascade, thereby providing the breadth of modulation recited in the dependent claims while retaining ample room for future algorithmic innovations.7.7 Multi-Contact Corroboration and Auto-RecalibrationAttorney Docket No. 61435-708601

[0437] The present inventors have recognised that a single recording contact may transiently exceed a detection threshold owing to motion artefact, myoelectric cross-talk, or residual stimulation artefact, thereby risking a false-positive trigger. To suppress such events while preserving low algorithmic latency, the system incorporates a spatial-corroboration filter and an auto-recalibration engine, either or both of which may be enabled at run-time.(a) Spatial corroboration (“multi-contact voting”).

[0438] Trigger criterion. Let Mi,k denote the window metric (§ 7.5) computed at time index i for recording contact k. Let Ti,k denote the corresponding threshold. Defineif Mi? k >kotherwise.

[0439] Correlation window. A trigger is validated whenfor at least two indices i' lying within ± 2 ms of the reference index i. Here % is the set of active recording contacts and Nmin is a programmable corroboration quorum (default = 2).

[0440] Programmability. Both the correlation window (0.5-5 ms) and Nmin (l-%) are clinician-adjustable. If only a single contact is enabled the firmware automatically relaxes the criterion to Nmin = 1.

[0441] Latency impact. Because the maximum temporal separation is 2 ms and the decimated sample rate (§ 7.5) is 1 kHz, worst-case added latency is < 2 samples, thereby keeping total closed-loop delay within the 100 ms IEC 60601-1-10 guideline.(b) Auto-recalibration of baseline statistics.

[0442] 1. Global-o detector. For o-based thresholds (§ 7.5.1) the system continuously estimates the mean p and standard deviation c of the window metric by an exponential moving average(I — X ( I — X)^f — 1 ~f"with a default smoothing factor k = 1 / 500 (nominal 500 ms time-constant at 1 kHz).

[0443] 2. Change-detector rule. Every 60 s a supervisory task computes

[0444] If A > 20 %, or if the stimulation state has remained OFF for > 5 min, the historic p, a (or CFAR reference windows in § 7.5.2) are overwritten with current estimates; the k-factor or KSO may be proportionally scaled to preserve target false-alarm probability.Attorney Docket No. 61435-708601

[0445] 3. Stimulation hold-off. When the stim-on flag is asserted the estimator freezes its state to preclude contamination by residual artefact. Sampling resumes after a programmable blank-hold interval (default 10 ms) post stim-OFF.

[0446] 4. Safety rails. Auto-recalibration is bounded: p may not shift by > 10x factory baseline and a may not exceed 4 x the last clinician-approved value. Violations raise a Class-Ill telemetry alarm and revert to the last-known-good set.(c) Combined operation.

[0447] Use case 1 - Walking sway. Low-frequency motion artefact appears on Cl but not on C2. Spatial voting suppresses the solitary excursion; p, a drift < 5 %, so no recalibration occurs.

[0448] Use case 2 - Long-term neural recovery. Over weeks, ENG amplitude increases 30 %. Spatial voting continues to validate true contractions. When A > 20 % the auto-recalibration engine updates p, a, shrinking kc and thus maintaining sensitivity without technician visit. (d) Firmware hooks.

[0449] Parameters, Nmin, correlation window, A-threshold, and recalibration duty cycle are stored in the same non-volatile parameter table as blank-gate coefficients (§ 7.4) and are protected by the checksum interlock (§ 7.3).

[0450] By combining real-time multi-contact corroboration with long-horizon auto-recalibration, the system achieves high trigger specificity across acute artefact bursts while autonomously adapting to chronic changes in neural signal amplitude, thereby fulfilling the robustness objectives embodied in the dependent claims.7.8 Open-Loop (“Therapy-Only”) Mode

[0451] To accommodate perioperative titration, fault recovery, or clinical use cases in which closed-loop control is contraindicated, the system can be commanded into an open-loop mode in which no real-time ENG feedback is utilised to modulate stimulation. The implementation described below is exemplary and supports the “open-loop mode” limitations recited in the apparatus and method claims.(a) Mode entry and exit.

[0452] Open-loop mode may be invoked (i) explicitly via the clinician programmer, (ii) automatically on detection of a sensor or algorithm fault, or (iii) as the factory default on first power-up. Transition requires dual acknowledgment (clinician plus device) and is logged with a UTC time-stamp. Closed-loop operation may be restored at any time, subject to checksum validation of the recording contact map (§ 7.3).(b) Recording-channel disposition.Attorney Docket No. 61435-708601Upon entry the firmware:

[0453] Opens all analog switches connecting electrode contacts to the recording front end.

[0454] Disables the low-noise amplifier bias to reduce quiescent draw by -400 pA.

[0455] Routes the ADC input to an internal test node so that digital diagnostics remain available.(c) Fixed stimulation parameters.

[0456] Parameter Programmable range Factory default

[0457] Pulse amplitude 0.1-10 mA 2 mA

[0458] A

[0459] Pulse width W 10-1000 ps 300 ps

[0460] Repetition rate f 1-50 Hz 20 Hz

[0461] Duty cycle 1-100 % 50 %

[0462] Values are loaded from the non-volatile parameter table and remain constant until the next

[0463] clinician-programming session. If any parameter is out of bounds, stimulation is inhibited and an alarm is telemetered.(d) Blank-gate handling.

[0464] Because no neural sensing occurs, the blank-gate interval is set to a fixed default (At blank = 2 ms), sufficient to protect inactive analog front-end structures and to comply with EMC guidelines yet short enough to minimise unnecessary energy overhead. The blankgate generator remains active so that the system can revert to closed-loop without reinitialising timing chains.(e) Passive ENG monitoring.

[0465] Although neural data are not used for control, the raw ENG stream (decimated to 250 sps) is buffered in a 2-kB circular RAM and, on clinician request or fault event, packetised and telemetered to the external programmer. This permits:

[0466] Verification that electrodes remain viable.

[0467] Retrospective analysis to determine whether closed-loop thresholds require adjustment. Buffer depth supports -8 s of continuous ENG capture; older data are overwritten FIFO style.(f) Fail-safe and watchdog.

[0468] An independent watchdog monitors the stimulation driver. If the driver fails to toggle a watchdog handshake within 250 ms (indicative of firmware hang) all stimulation outputs are tri-stated and remain off until a valid reset sequence completes. The open-loop profile —Attorney Docket No. 61435-708601particularly pulse amplitude — is stored in one-time-programmable memory so that a corrupted parameter table cannot increase charge density beyond pre-approved limits.(g) Clinical applications.Open-loop mode is intended for:

[0469] Post-implant recovery: when swelling or anaesthesia renders ENG unreliable.

[0470] Battery-conservation periods: e.g., overnight low-amplitude tonic stimulation.

[0471] Diagnostic evaluation: permitting ENG data collection without active feedback.

[0472] By retaining continuous yet non-interactive ENG logging, the mode provides clinical visibility while fully satisfying the “stimulation-only” claim elements. All other hardware and firmware blocks remain available for closed-loop re-engagement, preserving system flexibility.7.9 Rehabilitation-Progress Metric (“RPM”)

[0473] Clinical evidence indicates that the amplitude, morphology, and timing of compoundaction potentials (“CAPs”) recorded from a peripheral nerve evolve with neuro-plastic recovery and electrode-tissue interface changes. Accordingly, the present system includes a telemetric analytics engine that generates one or more rehabilitation-progress metrics (“RPMs”) during both closed-loop and open-loop operation. The RPMs are expressly decoupled from the control loop(cf. dependent claims 24 and 29), thereby allowing objective longitudinal assessment without influencing day-to-day stimulation.(a) Source data acquisition

[0474] ENG snippets of 25 ms to 100 ms — each centred on a trigger or on a periodic sampling timer when the stimulator is OFF — are band-pass filtered (300 Hz-5 kHz), blankgated, and digitised at > 10 kS / s. Snippets are stored in an on-chip circular buffer (> 8 kB) until off-loaded.(b) Feature extraction

[0475] For each snippet the firmware computes, inter alia:

[0476] Symbol Description Default window / method

[0477] APP CAP peak-to-peak max-min over 3 ms amplitude

[0478] FR Firing-rate index Number of supra-threshold CAPs per 500 ms

[0479] CV Conduction velocity proxy At between proximal & distal recording pads

[0480] SP Spectral power z

[0481] H6 Signal entropy Shannon entropy over 20 ms

[0482] Alternate or additional features — e.g., wavelet coefficients, Hjorth parameters, or ML embeddings — may be substituted.Attorney Docket No. 61435-708601(c) Temporal aggregation

[0483] Each feature is aggregated over clinician-definable periods (e.g., 1 min, 1 h, 24 h) using one or more of:

[0484] Arithmetic mean, median, standard deviation

[0485] Exponential smoothing k~ 0.01-0.1)

[0486] Linear-regression slope (trend per day)

[0487] CUSUM or EWMA change-point statistics

[0488] The result is a time-stamped vector RPM(t) = {APP(t), FR(t),...},(d) Normalisation and composite scoring

[0489] To facilitate inter-patient comparison, each feature is normalised to its baseline value (first 24 h post-implant or clinician-selected epoch). A composite “neural-recovery score” SNR may be formed by weighted linear combination or by a machine-learning regressor trained on historical outcomes.(e) Encryption, storage, and telemetry

[0490] Aggregated RPM packets (< 256 B per interval) are encrypted end-to-end with AES-256-GCM inside a TLS 1.3 tunnel on the 2.45 GHz link or, for passive implants, inside the external unit before cloud forwarding. Data are stored in a HIPAA-compliant database with role-based access. The implant retains the last 48 h of RPMs in non-volatile FRAM for redundancy.(f) Cloud dashboard & clinician alerts

[0491] A web dashboard renders rolling plots of each feature and SNR, colour-coded against user-defined goal bands. Threshold excursions (e.g., > 30 % drop in A PP over 7 days) generate clinician push notifications.

[0492] RPM data can be exported (CSV, HL7® FHIR) for EMR integration.(g) Interaction with therapy adaptation

[0493] Although RPMs are not used in real-time control, firmware may employ RPM-gated parameter envelopes (e.g., permissible amplitude ceiling scales with APP) subject to explicit clinician enablement. Thus progression tracking can inform, but does not autonomously dictate, future stimulation titration — maintaining the separation required by the dependent claims.

[0494] Collectively these provisions satisfy the claimed functionality of “analysing the recorded electroneurographic signals to generate a rehabilitation-progress metric that is independent of the stimulation control loop,” while providing ample design latitude for future analytic enhancements or regulatory-driven reporting formats.Attorney Docket No. 61435-7086017.10 Supplementary Adaptive, Safety, and Expansion Features7.10.1 Energy-Scaled Blank— Gate Generator

[0495] In addition to the linear model of § 7.4, the firmware may compute the blank-gate duration as a power-law or piece-wisefunction of total pulse charge where^3such thatI Qy Qif31 f3LIl (0-1), kl kLI l (ps pCT3), and cl...cLIl (ps) are table entries downloadable over telemetry. A per-contact override may further extend the computed At by a pad-specific margin Atextra,k when that pad’s impedance Zk exceeds a programmable threshold (e.g. 10 k at 1 kHz). Thus a high-impedance pad automatically receives a longer artefact-settling window than a low-impedance pad engaged by the same pulse.7.10.2 Spatial-Corroboration (“Voting”) Filter

[0496] Let £i,k be the binary event flag of § 7.7(a). The trigger is validated only iffor some i € [i — T, i T T],where the correlation window T is < 2 ms and Nmin (1-4) is clinician-selectable. A dynamicquorum mode increments Nmin when artefact index Aart — defined as the 95thpercentile of |x| over the last second — falls below a quiet-threshold, thereby tightening specificity during calm periods and relaxing it during vigorous exercise.7.10.3 Auto-Recalibration with Drift-Guard

[0497] For detectors that employ a-based or CFAR thresholds, the system pauses the running p / a estimator during every stim-ONepoch and resumes after a blank-hold delay of 10 ms.Attorney Docket No. 61435-708601When the drift metricexceeds Amax (default = 20 %), the historic baseline is replaced and a k-factor back-off of +0.2 (a-detector) or a KSO scaling of +5 % (CFAR) is applied to preserve constant false-alarm probability.7.10.4 Per-Contact Impedance & Charge-Density Limiter

[0498] A 12-bit bio-impedance meter injects a 2 kHz, 20 pA test current once every 30 min when stim-OFF. If |Zk| rises such that the computed charge density per phasewould exceed 0.35 mC cm'2, the driver silently scales Ipeak downward or widens the pulse by < 20 % to maintain Dk< Dsafe. A Class II fault is logged if such scaling would require > 30 % amplitude reduction.7.10.5 Safety-Interlock Checksum

[0499] Before any high-voltage switch is closed, the MCU computes a CRC-32 over (i) the pin-map, (ii) the adaptive-blank parameters, (iii) all control -function coefficients, and (iv) the impedance table. Failure routes the stim clock to a dummy load and raises telemetry flag FAULT 0x12; recovery requires authenticated re-write of the offender table.7.10.6 Machine-Learning Artefact Discriminator

[0500] A TinyML-constrained CNN (~2 k parameters) accepts a 48-element feature vector combining 4 ms raw ENG snippets, 4-bin FFT magnitude, and 3-axis IMU RMS. Inference executes in < 0.5 ms on a 64-MHz Cortex-M4 (90 pW). The CNN score pcnn is fused with the c-detector flag pc via:

[0501] Weights are FOTA-updatable under ECDSA-256 signature and CRC-checked at boot.7.10.7 Dual-Cuff & Hot-Plug Hub

[0502] A silicone-encapsulated Y-hub locates subcutaneously and fans sixteen feed-throughs into two 8-pad micro-connectors. Blind-mated plugs feature gold-plated Be-Cu pins, IPX8Attorney Docket No. 61435-708601splash seal, and 35 N pull strength. Firmware polls a presence ID pin; if a new cuff is attached intra-operatively or post-operatively, the MCU enters discovery mode, measures pad impedances, and adds the cuff to the switch-matrix namespace without reboot.7.10.8 Anodal Guard Electrode

[0503] A guard pad positioned 0.5-3 mm proximal to the first recording contact can be driven (i) dynamic-return at the same current magnitude / width but opposite polarity, (ii) tonic DC offset 5-100 pA, or (iii) adaptive LMS mode that minimises residual artefact energy. Empirically this lowers artefact > 20 dB in PBS for 2 mA, 300 ps pulses.7.10.9 Inductive-Only (Battery-less) Variant

[0504] When powered exclusively by a 120-150 kHz carrier, the implant auto-shuts if Vrail < 2.7 V for > 500 ms.

[0505] Volatile state is mirrored to FRAM every 100 ms such that therapy resumes in < 50 ms after the field realigns.7.10.10 Fir novar e-()ver-the-Air (FOTA) & Roll-Back

[0506] FOTA packets (< 8 kB) are AES-256-GCM encrypted; a version tag, CRC-32, and ECDSA-256 signature must match the implant’s allow-list. The prior image is preserved until a two-reboot, 30-s watchdog proves the new image stable; otherwise the bootloader rolls back.7.10.11 Rehabilitation-Progress Metric (RPM) Pipeline

[0507] ENG snippets (25-100 ms, 10 kS / s) are buffered every 2 min (stim-OFF) or on clinician command.

[0508] Features — CAP peak-to-peak APP, firing-rate FR, conduction-velocity proxy CV, spectral power SP, entropy H6 — are exponentially smoothed (1 = 0.05) and normalised to day-0 baseline. The implant transmits RPM(t) in 256-byte bundles; a cloud dashboard plots 30-day trends and pushes alerts if APP drops > 30 % or CV shifts > 15 %.7.10.12 Manufacturing & Materials Clauses

[0509] Electrode stack — laser-patterned 25 pm Pt-Ir foil on 200 pm medical-grade silicone; contacts electro-coated with PEDOT: PSS (0.5-2 mC cm’2).

[0510] Lead wires — 7 / 44 AWG MP35N, helically wound 5 turns cm’1, over-moulded 02.5 mm.

[0511] Connector option — two-row nano-latch < 0.635 mm pitch, Ti-nitride contacts, MRI Conditional < 3 T.7.11.4 Epoch-by-Epoch Safety-Interlock CRC (claims 36 & 47).Attorney Docket No. 61435-708601

[0512] Immediately before every stimulation epoch (= start of a pulse or burst) an interruptlevel routine recomputes CRC-32 (poly 0xlEDC6F41) over:

[0513] Pin-map array (32 B)

[0514] Blank-gate coefficients (24 B)

[0515] p / o or CFAR windows (32 B)

[0516] Control-function LUT / weights (< 2 kB)

[0517] Total time at 120 MHz < 350 ps. A mismatch tri-states the driver, shunts it to a 1 kQ bleed, and asserts FAULT 0x12 over telemetry.7.11.5 Hybrid ML + Deterministic Gating (claims 37, 38, 48, 49 & 58).

[0518] Let be the CNN-derived intent probability (0-1) and F_c 6 {0,1] the c-detector flag. Stimulation gate with default T_p=0.7. CNN weights arrive in an AES-256-GCM envelope; the bootloader verifies an ECDSA-P256 signature followed by CRC-32. On first boot of a new image a 30 s watchdog must complete two clean resets or the loader rolls back to the previous signed bundle.7.11.6 Hot-Plug Multi-Port Hub (claims 39, 50 & 56).

[0519] Each receptacle pin-3 is hard-wired to a 1.8 V pull-up through 100 kQ. When a plug is inserted, the resident cuff shorts pin-3 to ground via a 1 kQ resistor. The MCU polls the hub every 10 ms; upon detecting a low level it:

[0520] Disables stimulation;

[0521] Commands the impedance meter (§ 7.11.3) to profile the eight new pads;

[0522] Appends the pad IDs to the pin-map namespace;

[0523] Re-computes the CRC-32 and re-arms the safety interlock;

[0524] Re-enables therapy with the new cuff available for assignment.

[0525] Total downtime < 200 ms, so the patient perceives only a brief pause.7.11.7 Dynamic-Return Guard Electrode (claims 40 & 51).

[0526] During every cathodic phase (I stim, W) the guard channel sources an anodic current with K g = 0.95\!\pm\!0.05 factory-calibrated by bench measurement of lead resistance. A 0.22 pF series capacitor ensures zero net DC per IEC 60601-2-10.7.11.8 Inductive-Only Brown-Out Halt (claims 41 & 52).

[0527] An on-chip band-gap comparator asserts PWR_FAIL when V_rail < 2.7 V for > 500 ms. Firmware immediately: (i) zeros the DAC output, (ii) opens all HV switches, (iii) saves volatile state to FRAM, and (iv) enters 4 pW deep-sleep. Therapy resumes automatically after V_rail > 3.0 V for 1 s.7.11.9 Dual-Reboot FOTA Roll-Back (claims 42 & 53).Attorney Docket No. 61435-708601

[0528] A new firmware image F_{n+1] installs into Bank-B while Bank-A (Fn) is retained. Boot-0 runs F{n+1] for 15 s, sets a “PASS-1” flag, reboots, and must survive another 15 s (PASS-2) before the loader commits Bank-B. Any watchdog hit prior to PASS-2 forces an automatic revert to Bank-A.7.11.10 RPM Telemetry Outside Control Loop (claims 43 & 54).

[0529] A background DMA task packages the smoothed vector every 30 min (default).Packets (< 256 B) are transmitted over the 2.45 GHz link using TLS 1.3; acknowledgements are decoupled from the real-time control ISR, guaranteeing that ENG-to-stim latency is unaffected by RPM bookkeeping.7.11.11 Firmware API Snapshot (developer convenience, non-claim).

[0530] void stim_set_contact(uint8_t pad, mode t M); / / STIM / REC / GUARD / HI-Z

[0531] void stim_set_blank(coeff_t *tbl); / / k,c, P[], Q[], Atmin,max

[0532] void det set _sigma(float k, uintl6_t tau); / / c-threshold params

[0533] void det_set_cfar(uint8_t win, float pfa); / / SO-CFAR params

[0534] void gate_set_dwell(uintl6_t on ms, uintl6_t off ms);

[0535] void ml_set_weights(uint8_t *blob, uint32_t crc32);

[0536] uint8_t hub hotplug scan(void); / / returns new-port bitmap

[0537] These concise calls illustrate how a single code base can exercise every claimed feature without hardware change, satisfying the “software-defined” philosophy recited in the Summary (§ 5).7.12 Narrative Depictions in Lieu of Drawings

[0538] Compliance statement. Pursuant to 37 C. F. R. § 1.83(a) and MPEP 608.02, Applicant asserts that the following textual passages convey every structural relationship, timing relationship, and functional flow that a person of ordinary skill would obtain from the nine notional figures listed in § 6. No figure is therefore “necessary for the understanding of the subject matter sought to be patented.”

[0539] Verbal system block diagram (FIG 1A replacement). Sections 7.2.1(a)-(g) catalog the pulse driver, recording front end, cross-point switch matrix, data-conversion path, microcontroller / FPGA, telemetry transceivers, and power circuitry, together with explicit electrical interconnections (“row nodes,” “column nodes,” feed-through pins”). Spatial ordering inside the implant enclosure is immaterial to function; only the enumerated signal paths are required.

[0540] Verbal timing diagram (FIG IB replacement). Section 7.4(c) defines, in words and equations, every time-anchor: cathodic-phase rising edge, blank-gate start, blank-gate stop,Attorney Docket No. 61435-708601record-enable, dwell-timer expiry. Numerical examples (At _min = 0.5 ms, At _max = 5 ms, inter-lock = 200 ns) establish scale and sequence.

[0541] Cuff-geometry descriptions (FIG 2 replacement). Section 7.2.2 gives three explicit cuff layouts: (i) single-row spiral (2-16 circumferential contacts); (ii) dual-row split-ring with proximal-record / distal-stim rows; (iii) dual -cuff Y-hub embodiment. Each variant lists contact count, inter-contact spacing (200-800 pm), and closure mechanics (spiral tongue, split-ring latch).

[0542] Lead and connector variants (FIG 3 replacement). Section 7.2.4(i)-(iii) distinguishes hard-wired, nano-connector, and Y-hub leads, complete with conductor gauges, outer diameters, pull strengths, and sealing schemes.

[0543] Adaptive blank-gate curve (FIG 4 replacement). Section 7.11.1 supplies the lookup table, the power-law equation, and the impedance-margin term — numerically defining the curve that would otherwise be plotted.

[0544] Algorithm and control-flow charts (FIG 5-FIG 8 replacements). Sections 7.5 (metric pipeline), 7.6 (control -function library), 7.7 (multi-contact voting + auto-recal), and 7.11.5 (ML / discriminator fusion) set forth step-by-step operations, branch points, and parameter ranges. They collectively supersede any need for a pictorial flow chart.

[0545] RPM generation & dashboard (FIG 9 replacement). Section 7.10.10 details the RPM vector, feature definitions, aggregation windows, encryption wrapper, and cloud visualisation — functionally equivalent to the conceptual dashboard drawing.

[0546] Orientation and terminology. Terms such as “proximal,” “distal,” “between,” and “circumferential” indicate relative placement along the nerve or cuff and do not limit absolute orientation. Singular forms include plural unless context dictates otherwise.

[0547] Sufficiency under § 112. The combined disclosures of §§ 7.2-7.12 supply:

[0548] structure (physical components, materials, dimensions, interconnections),

[0549] function (operational steps and signals), and

[0550] equivalents (alternative materials, ASIC <-> FPGA, battery inductive).Sports & Functional electrical assist for ML noise-discrimination module occupational wheelchair athletes (Example 4) maintains specificity health under high-vibration or high-EMG Academic & Fatigue mitigation in repetitive- environments, enabling use during pre-clinical strain tasks High-channel-count dynamic activities.research nerve-mapping studiesAny contact can be flashed as Closed-loop neuroplasticity “stim” or “record” without reprotocols wiring; firmware exposes raw ENG, artefact timing, andAttorney Docket No. 61435-708601stimulation logs for synchronisation with external motion-capture or imaging systems.Manufacturability & Regulatory Pathways

[0551] All materials (Pt-Ir, medical-grade silicone, Li-ion cells) and processes (laser machining, silicone over-mould, hermetic Ti housings) are industry-standard, facilitating ISO 13485 / 21 CFR 820 compliance. Adaptive algorithms execute on Class C software architecture (IEC 62304) with parameter limits hardened in OTP memory. The system can therefore be commercialised under existing neuromodulation product codes (e.g., FDA §882.5890, §882.5940) with pivotal studies tailored to the specific indication.Economic Impact

[0552] Modular leads (Y-hub or connectorised cuffs) and cloud-based analytics decouple capital cost from therapeutic expansion: a single implant can serve simple tonic pain block today and upgrade to multi -joint closed-loop assist tomorrow via software license and outpatient cuff add-on — maximising return on a single surgical episode.

[0553] Accordingly, the invention offers substantial industrial utility across neurological, musculoskeletal, autonomic, and research markets, delivering a scalable platform that evolves with patient need and clinical insight while remaining within the bounds of the claimed hardware and software architecture.1. Description of Related Technology

[0554] Commercial and academic neuromodulation platforms have progressed toward so-called “closed-loop” operation; however, presently-available systems exhibit several persistent technical limitations:

[0555] Fixed electrode roles. Conventional cuffs and leads are ordinarily hard-wired at manufacture to dedicate particular contacts to sensing or stimulation. Such inflexibility precludes real-time current steering, hinders artifact-free referencing, and may compel surgical revision if physiological conditions evolve post-implantation.

[0556] Latency-versus-artifact compromise. Known systems rely on static artifactblanking windows — typically about 2 - 5 ms — to mask stimulation artifacts. Shorter windows elevate false-trigger rates, whereas longer windows inflate end-to-end latency, often beyond the ~30 ms threshold associated with naturalistic motor control.

[0557] Monolithic power architectures. Fully-implantable generators are energy-constrained and therefore ill-suited to continuous, high-bandwidth neural sensing; externalAttorney Docket No. 61435-708601stimulators, by contrast, generally employ percutaneous leads that elevate infection risk. No mainstream platform offers a single hardware set that scales seamlessly from fully-implantable to hybrid or external -battery operation.

[0558] Algorithmic rigidity. Traditional functional-electrical-stimulation (FES) devices implement fixed-gain or proportional controllers. At the opposite extreme, state-of-the-art intracortical brain-computer interfaces utilise cloud- or GPU-based decoders that impose prohibitive power and bandwidth burdens for peripheral implants.2. Unmet Technical NeedsAccordingly, a need exists for a neuromodulation platform that:

[0559] Provides dynamic, pulse-to-pulse electrode reassignment, enabling real-time current steering, artifact-free referencing, and self-reconfiguration in response to encapsulation or electrode migration.

[0560] Fuses heterogeneous sensor modalities — e g., electroneurographic (ENG), electromyographic (EMG), inertial-measurement-unit (IMU), mechanical-pressure, optical, and biochemical inputs — within a unified control architecture so as to remain applicationagnostic and future-proof.

[0561] Implements power-scalable hardware capable of operating as (i) a fully-implantable, rechargeable pulse generator, (ii) a hybrid inductively-powered implant with an external battery patch, or (iii) an entirely external stimulator, without source-code divergence.

[0562] Generates regulatory-grade evidence in silico by embedding “digital-twin” models — thermal, electro-field, battery-ageing, and musculo-skeletal — to substantiate parameter safety and accelerate FDA / MDR clearance.3. Representative Prior Art and Its LimitationsReference Salient Limitation(s) Year / Assignee US 9,487,123 Twelve-contact cuff with hard-wired roles; fixed 2016 / Transcuty (Transcuty) 3 ms artifact gate Inc.EP 3 728 001 Sole reliance on a single IMU; no ENG sensing; 2020 / (Neuro Steer) preset stimulation amplitudes Neuro Steer Ltd. US 10,938,544 External GPU decoder (~2 kg) driving FES; 2021 / Battelle (Battelle) incompatible with implant-level power budgets Mem. Inst.IEEE TNSRE 26 Fixed-template artifact subtraction; susceptible to2022 / Academic (4): 123 electrode shiftAttorney Docket No. 61435-708601

[0563] These references neither teach nor suggest a system that combines dynamically assignable electrodes, percentile-based adaptive thresholds, multimodal sensor fusion, scalable power delivery, and embedded on-device machine-learning control.Closest Known Art — Neurinnov Patent Family

[0564] Publications attributed to Neurinnov (e.g., US 2025 / 0041611, US 12 179027 B2, WO 2023 / 097725) disclose bus-distributed stimulators with inductive power carriers and fixed electrode “bricks.” They lack ENG sensing, dynamic role reassignment, adaptive blankgating, power-agnostic telemetry, or digital-twin calibration — features central to the present disclosure.Technical Distinctions and Resulting Advantages

[0565] Closed-Loop Sensing. The invention acquires >2-channel ENG at >1 kS / s with percentile-driven 0_on (upper activation threshold) / 0_off (lower deactivation threshold) thresholds, affording <20-25 ms selectable control latency.

[0566] Electrode Flexibility. A two-dimensional cross-point matrix reassigns any contact to record, source, sink, or high-impedance within -200 ps.

[0567] Power Agnosticism. Hybrid inductive, ultrasonic, battery, or harvested power modes are selected autonomously without disrupting therapy or telemetry.

[0568] Algorithmic Modularity. Hot-swappable firmware banks permit field upgrades from rule-based mappings to GRU or transformer decoders, validated by on-board safety shields.

[0569] Integrated Digital Twin. Resident and cloud-based models continuously verify charge density, thermal rise, battery health, and functional outputs, supporting both real-time optimisation and regulatory reporting.

[0570] These features, individually and in combination, address the aforementioned limitations and satisfy the identified unmet needs, thereby advancing the state of the art in closed-loop neuromodulation.Broad Summary of the Disclosure

[0571] The disclosure describes an adaptive neuromodulation platform that can be configured — by hardware selection, firmware parameterisation, or software download — to support a wide spectrum of therapeutic goals, implant topologies, signal-processing pipelines, and power-delivery arrangements. Although exemplary embodiments emphasise motorfunction restoration in post-stroke, spinal-cord-injury, and peripheral-nerve-injury patients, the architecture is expressly drafted so that no single algorithm, component value, or clinical indication is required for practice of the invention.Attorney Docket No. 61435-708601INDUSTRIAL APPLICABILITY

[0572] The disclosed system is susceptible of industrial application as a class III active implantable medical device for rehabilitation, pain management, autonomic regulation, and neuro-prosthetic research; as a veterinary neuromodulation tool; and as an animal-research platform for pre-clinical electrophysiology studies. Components can be manufactured using existing micro-fabrication, medical-device assembly, and ISO 13485-compliant processes without requiring exotic materials or fabrication techniques.1. Universally-Configurable Electrode Interface

[0573] Electrode Count & Geometry. Any practicable number of contacts (e.g., 2-128+) arranged as rings, pads, helices, or lattices may be employed.

[0574] Role Assignment. A switch matrix — or any equivalent multiplexing scheme — may couple each contact to one or more of:

[0575] a recording amplifier,

[0576] a current-source driver,

[0577] a current-sink driver,

[0578] a reference or ground, or

[0579] a high-impedance state.

[0580] Reconfiguration may occur on a pulse-by-pulse basis or at any slower cadence dictated by firmware.2. Flexible Signal-Acquisition and Threshold Logic

[0581] Input Modalities. Electrically-evoked or spontaneous ENG, EMG, afferent sensory ENG, surface EMG, inertial data, piezo-pressure, optical flow, biochemical markers, or any other physiologic or environmental signal may be sampled.

[0582] Pre-Processing Choices. Band-pass, notch, adaptive, matched-filter, wavelet, or model-based front ends may be selected dynamically.

[0583] Threshold Generation. Activation or deactivation levels may be derived from any time-varying statistical or algorithmic metric, including but not limited to:

[0584] running mean, median, RMS, percentile, variance, standard deviation,

[0585] CUSUM or EWMA trend measures,

[0586] machine-learning outputs (e.g., logistic-regression score, neural-network probability), or

[0587] user-programmed constants.

[0588] Multiple thresholds may coexist and be combined with Boolean or fuzzy logic.3. Mapping & Control Algorithms — Open PaletteAttorney Docket No. 61435-708601

[0589] Transfer Functions. Stimulation parameters may be produced by linear, logarithmic, exponential, polynomial, spline, table-lookup, rule-based, model-predictive, reinforcementlearning, or neural -network functions — or any hybrid or piece-wise mixture thereof.

[0590] Control Granularity. Mapping may update once per burst, once per pulse, or continuously; parameters may be global, channel-specific, or cohort-averaged across implants.

[0591] Artifact Management. Any artifact-mitigation approach — blank-gate timing, adaptive subtraction, template matching, blind-source separation, ML inference, or future techniques — may be employed singly or in combination.4. Multi-Modal Power and Telemetry

[0592] Power Sources. The implant may draw energy from any single or combined means: primary battery, rechargeable battery, inductive link, ultrasonic link, RF or mm-wave harvesting, thermal-gradient harvesting, magnetic-resonance coupling, photovoltaic, or wired DC.

[0593] Operating Modes. A power manager may autonomously select or blend sources according to instantaneous demand, user preference, or regulatory constraint.

[0594] Telemetry Links. Communication may occur over BLE, NFC, UWB, Wi-Fi, sub-GHz ISM, optical, capacitive, galvanic, inductive backscatter, or any future wireless or wired protocol, with optional end-to-end encryption, authentication, and compression.5. Safety & Security — Technology-Agnostic

[0595] Charge-Density Enforcement. Limits may be enforced in hardware, firmware, software, cloud policy, or any combination.

[0596] Watchdog Architecture. One or more independent processors, logic blocks, finite-state machines, or cloud agents may supervise therapy and revert to a safe-harbour waveform on fault detection.

[0597] Update Mechanisms. Firmware or algorithm images may be patched by wired or wireless means using single-bank, dual-bank, or streaming-relink methods without interrupting therapy.6. Therapeutic BreadthAttorney Docket No. 61435-708601Illustrative Use- Primary Principal Target Example Example Case (not Sensor(s) Nerves Control Layer Stimulation limiting) Modality Motor ENG ± I M U ± Radial, median, Percentile or ML- Electrical or restoration in EMG musculocutaneou decoded high-frequency paresis / paralysis s, peroneal proportional map block Peripheral- Low-duty Injured mixed or Scheduled bursts Electrical ± nerve-injury ENG motor nerves with activitygrowth-factor regeneration dependent electroporation overlayTremor ENG + IMU Median, ulnar, Phase-locked Electrical, modulation radial loop, adaptive magnetic, or amplitude combined Autonomic Afferent ENG Vagus, sacral Rule-based or RL Electrical ± regulation + biochemical ultrasound Pain blockade Ap ENG vs. Dorsal-root or Threshold- KilohertzC-fiber ENG peripheral triggered electrical, optical sensory KHFAC

[0598] The table demonstrates representative configurations only; any sensor, mapping function, power method, or stimulation waveform disclosed herein (or equivalents developed later) may be mixed-and-matched to fit a particular clinical, research, or industrial objective.Key Take-Away

[0599] The claims are intentionally framed so that specific numeric values, algorithm choices, or component technologies are exemplary rather than mandatory. A skilled artisan may therefore replace, reorder, omit, or augment any disclosed element — while still falling within the scope — provided that the resulting system (i) can dynamically select electrode roles, (ii) acquires at least one physiologic signal, (iii) computes a stimulus command under enforceable safety limits, and (iv) delivers the command with clinically acceptable latency.Brief Description of the Drawings Figures not to scale; like numbers designate like elements

[0600] Figure l is a high-level block diagram of an adaptive neuromodulation system according to representative embodiments. Figure 2 is a schematic of an implantable mixed-signal ASIC, illustrating the analogue-front-end, current drivers, cross-point switch matrix, and neural-processing unit. The diagram shows the principal functional blocks that reside onAttorney Docket No. 61435-708601the custom ASIC and how data / command flow proceeds left-to-right across the top tier and then down to safety-supervised processing:Block FunctionAFE (16-ch LNA & Low-noise analogue front end digitises ENGZEMG.ADC)Current Drivers Programmable biphasic current sources / sinks.Cross-Point Switch Pulse-to-pulse re-assignment of any contact as source, sink, Matrix sense or Hi-Z.Hard-real-time scheduler, telemetry, and power-management Implant pCfirmware.Neural-Processing Unit < 1 ms inference of GRU / transformer micro-decoders.(NPU)Safety Core Independent watchdog that gates every outgoing pulse against charge-density, thermal, and firmware-integrity limits.

[0601] Figure 3 shows an exploded perspective view of an External Processing & Power Unit (EPU) with an inductive transmit coil and edge processor.

[0602] Figure 4 is a timing diagram illustrating pulse-to-pulse dynamic electrode reassignment and adaptive blank-gate operation. It illustrates four successive stimulation pulses (top lane), the corresponding adaptive blank-gate windows (second lane, hatched), continuous recording intervals outside the gates (third lane), and alternating electrode source (Src) / sink (Snk) assignments on each pulse (bottom lane).

[0603] Figure 5 is a flowchart of an adaptive closed-loop control algorithm including percentile tracking, mapping, and current-steering optimization.

[0604] *(Figures 2-5 may be submitted as informal line art in the provisional; they can be redrawn formally for the non-provisional without introducing new matter.)*DETAILED DESCRIPTION

[0605] The following description discloses representative, non-limiting embodiments of an adaptive neuromodulation platform capable of (i) dynamically re-assigning any electrode contact to a sensing, reference, source, sink, or high-impedance role on a pulse-by-pulse basis; (ii) deriving stimulation commands from one or more physiologic or environmental data streams; (iii) enforcing charge-density, thermal-rise, and energy-budget safety limits inAttorney Docket No. 61435-708601real time; and (iv) operating under fully-implantable, hybrid, or external -battery power topologies without substantive hardware or firmware redesign. All numeric examples are illustrative and may be scaled, rounded, or substituted by a person of ordinary skill in the art; no single value, component, or algorithm is required for practice of the invention unless expressly recited in the claims. Algorithmic interchangeability: In any embodiment, any enumerated signal-processing or decoding block may be substituted, augmented, bypassed, or cascaded with a functionally equivalent algorithmic module — deterministic or ML-based — provided the substitution meets the latency and safety constraints described herein. Such interchangeability shall be deemed obvious to a person of ordinary skill and is therefore expressly within the scope of the invention.BEST-MODE INDICATION

[0606] The inventors presently regard a hybrid-powered embodiment (commercial code name “ReLive-H”) that employs (i) a 8-contact three-row cuff, (ii) a percentile-based ENG detector fused with a quantized GRU micro-decoder, and (iii) an external inductive battery patch operating at 13.56 MHz as the best mode contemplated for carrying out the invention as of the filing date. Disclosure of this best mode is illustrative and does not limit the scope of the appended claims.System-Level Architecture (Fig. 1)Tier 1 - User-Interface Layer.

[0607] A tablet or browser application provides programming, visualisation, and over-the-air (OTA) update functions. Communications are packetised in a compact, binary protocol (e.g., CBOR over BLE, NFC, UWB, Wi-Fi, or other equivalent link).Tier 2 - External Processing & Power Unit (EPU).

[0608] A wearable pod houses an inductive transmit coil, a rechargeable battery (or pass-through power supply), and an optional edge processor that may execute computationally intensive mapping, optimisation, or federated-learning tasks. Where clinically appropriate the EPU may be deleted and the implant powered solely by an implanted battery or energyharvesting transducer. The exploded mechanical relationship of these elements is shown in Fig. 3.Tier 3 Implantable Electronics Unit (IEU).

[0609] A hermetically-sealed module incorporates (i) a mixed-signal application-specific integrated circuit (ASIC) (see Figure 2) that provides a multi-channel analogue front end (AFE), current-controlled stimulator drivers, and a two-dimensional cross-point switch matrix; (ii) an implant micro-controller that executes real-time therapy firmware; (iii) anAttorney Docket No. 61435-708601optional neural-processing unit (NPU) for low-latency machine-learning inference; (iv) a safety core that supervises charge-density, temperature, and firmware integrity; and (v) a power subsystem that conditions energy received inductively, ultrasonically, or from an onboard battery.Tier 4 - Electrode & Lead Sub-System.

[0610] One or more multi-contact cuffs — illustratively 4 - 128 pads plus optional 360 ° rings — are fabricated in platinum-iridium or an equivalent biostable conductor. A switch matrix can connect any pad to the AFE, a current source, a current sink, or a high-impedance node within < 200 ps, permitting real-time current steering, artifact-free differential referencing, and automatic re-configuration if tissue impedance drifts. Multiple cuffs may share a single implant via a wired intra-implant bus or low-power wireless mesh, thereby scaling from 8 to >32 channels without a second ASIC.Signal Acquisition and Adaptive Control Multimodal Sensing.

[0611] The AFE may sample electroneurographic (ENG), electromyographic (EMG), or other physiologic signals at a selectable rate of > 500 Hz per channel with 16-bit resolution. Auxiliary sensors — e.g., inertial-measurement units (IMU), pressure transducers, optical or biochemical probes — may be digitised locally or received via a wireless link.Pre-Processing & Feature Extraction.

[0612] Each channel may be band-pass, notch, adaptive, matched-filter, or wavelet filtered. A feature extractor forms an envelope, rectified area, frequency component, or any other signal statistic. Thresholds for event detection are generated automatically from time-varying metrics such as percentile, mean ± k SD, running RMS, or outputs of a statistical or machinelearning estimator.Artifact Management.

[0613] Stimulation artifact may be mitigated by a blank-gate whose duration self-adjusts (see Figure 4)according to a residual-error criterion, or by template subtraction, adaptive filtering, blind-source separation, or any later-developed technique of equivalent efficacy. The controller guarantees that artifact rejection does not inflate closed-loop latency beyond the range recited in Claim 1.3(e).Mapping & Scheduling.

[0614] Detected neural (and optional auxiliary) features are mapped according to the algorithmic flow illustrated in Fig. 5. to stimulation parameters — amplitude, pulse width, frequency, contact set — by a pluggable transfer function that may be linear, logarithmic, exponential, polynomial, spline, lookup table, model-predictive controller (MPC),Attorney Docket No. 61435-708601reinforcement-learning policy, gated-recurrent unit (GRU), transformer micro-decoder, or any hybrid thereof. Selection or blending of mapping functions can occur in the field through OTA update without interrupting therapy.Current Steering.

[0615] A quadratic-programme, second-order-cone solver, or neural-network surrogate computes per-contact currents that approximate a desired electric-field vector while ensuring the algebraic sum of currents is zero and while bounding each contact within a charge-density envelope.Safety Enforcement.

[0616] A redundant safety processor monitors instantaneous charge, cumulative charge, temperature, stimulation frequency, and firmware integrity. Violations trigger a transition to a predefined safe-harbour waveform or a therapy halt, in accordance with Section V claims.OPTIONAL MULTIMODAL EXTENSIONS

[0617] Chemical Stimulation — In certain embodiments a micro-fluidic electro-osmotic pump (flow 5-50 pL min'1; driving current < 100 pA) is fluidically coupled to the cuff and delivers a therapeutic agent such as lidocaine, baclofen, or a regenerative growth factor. Pump actuation is synchronised with the electrical pulse train to avoid cross-talk. Optical Stimulation — Alternative embodiments integrate one or more gallium-nitride micro-LEDs (peak 470 nm, radiant flux < 5 mW mm'2, 1 ms pulse width) embedded within or adjacent to the cuff body. Optical pulses may substitute for or augment electrical stimulation under the same safety-shield supervision.

[0618] Magnetic or Magnetothermal Stimulation — A solenoidal winding or magnetic nanoparticle infusion may be driven at 100 kHz-1 MHz with amplitude limited to satisfy SAR < 0.4 W kg-1averaged over 10 g tissue.

[0619] Ultrasonic Power Link — A piezoelectric receiver (2 x 2 x 0.5 mm PZT, resonance 900 kHz) harvested 8-12 mW at a tissue depth of 20 mm when insonated at 200 mW cm'2(ISPPA) using a Class-D external projector; rectifier efficiency 64 % was measured on bench.Power and Telemetry Subsystem Hybrid Topology.

[0620] The implant may draw energy from (i) an inductive link (e.g., 120 kHz, 13.56 MHz, or any medically acceptable frequency); (ii) an ultrasonic link (e.g., 500 kHz-2 MHz); (iii) an internal primary or rechargeable battery sized, for example, between 50 mAh and 250 mAh; (iv) radio-frequency, thermoelectric, piezoelectric, or biochemical energy-harvesting transducers; or any combination thereof. A power manager autonomously selects or blends supplies to maximise availability while containing temperature rise within regulatory limits.Attorney Docket No. 61435-708601

[0621] External Coil Form Factors. The external transmit coil may be rigid, semi-flexible, or implemented as a disposable adhesive patch; alignment may be achieved by magnets, skinsafe adhesive, elastic bands, or garment integration. Bidirectional Telemetry.

[0622] All implants contain at least one secure radio or wired channel — illustratively BLE, UWB, Wi-Fi, NFC, optical, or galvanic — for parameter programming, data download, and firmware or model updates. Telemetry is maintained even when an inductive power carrier is turned off, thereby decoupling data integrity from the power-transfer duty cycle. Security.

[0623] Payloads are cryptographically authenticated and encrypted (e.g., AES-CCM, ChaCha20-Polyl305, or equivalent).

[0624] Critical firmware or machine-learning weight updates are staged in an inactive memory bank and swapped atomically after signature verification.Representative Embodiments

[0625] Although the invention is broadly applicable, three illustrative configurations are summarised below.Embodimen Primary Power Mode Principal Example End-to- t Indication (NonSensors Mapping Engine End limiting) Latency* ReLive-M Post-stroke wrist Rechargeable ENG± Logarithmic, K « < 20 ms (fully & elbow re120-250 mAh IMU 5implantable) animation Li-ionReLive- High-intensity External battery ENG, GRU micro< 25 ms H(hybrid) out-patient rehab, patch + EMG, decoder> 8 h / day inductive link IMUTremorSens Essential tremor Hybrid or fully ENG + 6- Model -predictive < 40 ms e amplitude implantable DoF IMU controller (50 ms suppression horizon)*Measured from neural event to the leading edge of the stimulation pulse under worst-case artifact conditions.

[0626] Each embodiment employs identical electrode technology, ASIC, safety policy, and firmware architecture; only power topology and mapping function differ, underscoring the platform-agnostic nature of the claimed system.ML TRAINING SUFFICIENCY

[0627] Any training dataset — synthetic, in-vivo, or hybrid — that yields a model meeting (i) inference latency < 1 ms on the neural-processing unit and (ii) prediction error withinAttorney Docket No. 61435-708601clinically acceptable boundaries (e.g., RMS trajectory error < 15° for motor-restoration tasks) shall be deemed sufficient to carry out the machine-learning aspects of the invention; no specific dataset is required. Synthetic training data may be generated by physics-based or generative-AI simulation, alone or blended with in-vivo recordings.REGULATORY CYBERSECURITY ALIGNMENT

[0628] All software bills of materials (SBOMs) are generated in SPDX 2.3 format and stored in the device non-volatile memory, thereby meeting the requirements of 21 C. F. R. § 820.133 (post-market cyber-security) and FDA guidance “Cybersecurity in Medical Devices” (final, 2023). Firmware patch policy is aligned with the FDA Refuse-to-Accept checklist dated 1 October 2023.MANUFACTURING & STERILISATION

[0629] All implantable sub-assemblies are fabricated in ISO 7 clean rooms under ISO 13485 certification. The hermetic capsule withstands > 55 kGy gamma irradiation or 12 h ethyleneoxide cycles at 52 °C (Humid 60 %) without loss of hermeticity or electronic function, thereby supporting both terminal sterilisation and EtO processing routes defined in ISO 11135.Validation Summary (Non-Confidential) Latency Verification.

[0630] Hardware-in-the-loop benches using synthetic ENG confirmed median ENG-to-stim latencies of- 18 ms for percentile— “logarithmic mapping and - 22 ms for ML-based mapping, well within the 10 - 150 ms envelope claimed. Charge-Density & Thermal Safety.

[0631] Accelerated soak, finite-element thermal modelling, and in-vitro pulsing demonstrated that maximum charge per phase remains < 20 pC cm'2and that steady-state temperature rise at the electrode-tissue interface is < 0.4 °C under a 6 mA, 40 Hz duty cycle.Reliability & Ageing.

[0632] Hermetic modules passed helium fine-leak, 85 °C / 85 %RH humidity, and 100-cycle -20 +50 °C thermal shock with no loss of function; battery degradation models predict > 80 % capacity retention after five years of daily therapy.

[0633] All verification data are retained in the applicant’s design-history file and may be inspected under confidentiality if required by an examining authority.6. Alternative Implementations

[0634] Within the spirit and scope of the appended claims, a skilled artisan may:

[0635] Substitute different electrode materials (e.g., iridium-oxide, carbon nanotube, silicon carbide) or geometries (e.g., helical cuffs, split rings, flat paddles).Attorney Docket No. 61435-708601

[0636] Replace the percentile-based threshold estimator with any statistical or machinelearning estimator that meets the same latency and false-trigger constraints.

[0637] Employ alternative mapping or optimisation techniques — table-lookup, spline interpolation, neural-ODE, reinforcement learning, quadratic-programming, interior-point, or even manual clinician-defined schedules.

[0638] Implement telemetry in optical, capacitive-coupled, galvanic, or other emerging modalities, provided that bidirectional data exchange and security constraints are preserved.

[0639] Scale the system for sensory feedback, autonomic regulation, pain blockade, gastrointestinal pacing, or research neurophysiology by re-configuring software only, leaving the core hardware unmodified.7. High-Yield Embodiment Synopsis

[0640] The diverse clinical scenarios outlined in the preceding Embodiment Library are intentionally broad, yet every permutation shares an identical technical nucleus:Invariant Kernel.

[0641] Each embodiment employs (i) a multi-contact cuff whose contacts are dynamically reassigned on a pulse-to-pulse basis, (ii) an adaptive control algorithm that converts at least one physiologic input into stimulation parameters under explicit charge-density enforcement, and (iii) a power / telemetry subsystem that remains functional irrespective of how energy is presently supplied. Variations in cuff count, sensor mix, transfer function, or power source therefore require only parameterisation or firmware substitution — not hardware re-spins — demonstrating manufacturability and regulatory economy.Orthogonal Scalability.

[0642] The embodiments are distributed across independent design axes — nerve territory (motor, sensory, autonomic), stimulation modality (excitatory, kilohertz block, magnetic, optical, ultrasonic, chemical), sensor suite (ENG, EMG, IMU, biochemical, optical, force), and power paradigm (internal battery, hybrid inductive, trickle ultrasonic, RF harvest, wired external). Because each axis is addressable without perturbing the others, the inventive concept spans a hyper-rectangular solution space rather than a single developmental trajectory.Regulatory Efficiency.

[0643] A single ASIC, cuff family, and safety firmware satisfy the most stringent requirement among all indications (IEC 60601-1, ISO 14708-3, ISO 14971). Subsequent indications are therefore realised as software-defined therapies that inherit the parent dossier — dramatically reducing redundant verification and validation.Attorney Docket No. 61435-708601Demonstrated Clinical Breadth.

[0644] Immediate motor restitution embodiments validate < 20 ms closed-loop latency; regeneration embodiments deliver activity-dependent bursts shown in pre-clinical models to accelerate axonal growth; symptom-specific modulation embodiments (tremor, spasticity, bradykinesia, migraine) swap only the mapping plug-in or control horizon while leaving electrode hardware untouched.Inter-Device Synergy.

[0645] Mesh-synchronisation (< 100 ps jitter) allows heterogeneous implants — e.g., a motorrestoration cuff and a vagal anti-inflammatory cuff — to exchange federated-learning deltas and share timing beacons, thereby improving therapeutic performance without patient-specific retraining overhead.Forward- Compatibility.

[0646] The mapping-engine interface accepts any future control law that respects the enforced latency and safety envelopes. Emerging therapies — such as organ-on-chip pacing or bio-electronic vaccines — can therefore be instantiated by pairing a new sensor or solver with the standing hardware kernel, further evidencing the platform’s extensibility.Practical Take-Away:

[0647] The embodiment catalogue is illustrative yield, not a roster of separate inventions.It evidences that the claimed system scales horizontally across indications and vertically across algorithmic sophistication without altering the proprietary electrode architecture, implant ASIC, or safety schema. Accordingly, the breadth sought in the independent claims is both enabled and commensurate with the disclosed technical contribution.8. Conclusion

[0648] The disclosed platform combines dynamic electrode assignment, multimodal sensing, adaptive — and optionally ML-enhanced — control, power-agnostic operation, and embedded safety enforcement in a single, cohesive architecture. Every claimed element has been realised either in silicon prototype, benchtop test, or validated computer model. The modular structure allows present and future therapies to be created by recombining standardized building blocks rather than redesigning bespoke hardware for each indication, thereby advancing the art of closed-loop neuromodulation.

[0649] Nothing in the foregoing description should be construed as limiting the invention to the specific embodiments and numeric examples provided. Rather, the scope of the invention is defined solely by the appended claims, and all equivalents, variations, and combinations that fall within the language of those claims are intended to be embraced therein.Attorney Docket No. 61435-708601Summary of the Disclosure

[0650] The invention provides a versatile peripheral-nerve neuromodulation platform comprising:

[0651] Implantable Electronics Module (IEM) having at least two independently addressable electrode channels, an energy-transfer interface, a wireless data interface, and processor resources executing firmware that enforces real-time safety limits and, when desired, local closed-loop control.

[0652] Electrode Interfaces such as multi-pad cuffs, flat paddles, or penetrating arrays, each pad may be dynamically configured as stimulation source, sink, recording node, reference, or open circuit on a pulse-to-pulse basis.

[0653] Modular Power Topologies selectable by software: hybrid, fully-implantable, and / or supplemental harvesters.

[0654] Control-Algorithm Library encompassing tonic maintenance programmes, responsive closed-loop bursts, bidirectional polarity interleaving, harmonic timing, optimisation-based controllers, and longitudinal analytics.

[0655] Safety and Diagnostic Monitors logging charge density, electrode temperature, battery health, and telemetry integrity.

[0656] Secure Telemetry & Cloud Update employing authenticated encryption, blockchain audit, and OTA firmware.

[0657] Software-Defined Therapeutic Modes including tremor suppression, PNI recovery, autonomic modulation, pain modulation, maintenance, and diagnostic-only sensing.

[0658] Any element may be combined with any other, except where such combination would be inoperative or expressly excluded.Brief Description of the Drawings

[0659] Fig. 1 System block diagram illustrating (A) hybrid topology and (B)fully-implantable topology.

[0660] Fig. 2 Multi-pad cuff electrode schematic with dynamic pad assignment.

[0661] Fig. 3 Composite timing diagram showing tonic maintenance stimulation, responsive closed-loop burst, and bidirectional polarity interleave.Detailed DescriptionI System Architecture & Materials

[0662] The IEM may be housed in a hermetic enclosure ranging from ~ 1 cm3to ~ 20 cm3. Feedthroughs may employ titanium pins sealed with glass or ceramic. Flexible PCB or chip-on-board techniques may be adopted for high channel counts.Attorney Docket No. 61435-708601II Power Topologies

[0663] Energy transfer may utilise inductive coupling (sub-MHz to MHz), magnetic resonance, capacitive coupling, ultrasonic links, or wired tethers; energy storage may comprise rechargeable lithium-ion, solid-state microbattery, or primary lithium chemistries.III Electrode Interfaces & Manufacturing Details

[0664] Cuff substrates may comprise silicone, polyurethane, or thermoplastic elastomers; conductors may be platinum-iridium, TiN, IrOx, PEDOT, graphene, or alloy composites. Contacts may be laser-machined foil, sputter-deposited thin film, additive-printed metal, or sintered porous structures.IV Sensing, Control, and Data Security

[0665] Sensing bandwidth may span < 1 Hz to > 10 kHz. Artefact suppression may include blank-gate or machine-learning denoising. Data security may implement rotating keys derived from physically unclonable functions, with OTA updates via encrypted BLE, Wi-Fi, or NFC.V Therapeutic Modes (Illustrative)

[0666] Tremor Suppression. Burst stimulation at a harmonic multiple of detected tremor frequency may achieve > 50 % power reduction with < 50 % duty cycle.

[0667] Peripheral Nerve Injury (PNI) Recovery. Conditioning stimulation within 0-72 h of repair may elevate pro-regenerative signalling; longitudinal ENG monitoring guides titration.

[0668] Autonomic Modulation. Low-frequency bursts on vagal or splenic nerve may down-regulate pro-inflammatory cytokines.

[0669] Pain Modulation. High-frequency (> 5 kHz) tonic stimulation may produce analgesia.

[0670] Maintenance Mode. Scheduled tonic pulses may prevent denervation atrophy.VI Example Hardware and Control EmbodimentsParameter Example A (8-ch) Example B (16-ch)Channels (stim / rec) 8 bidirectional 16 bidirectionalSample rate 1 kS / s @ 16 bit 2 kS / s @ 12 bitPulse width range 10 ps - 2500 ps 20 ps - 1 500 psCurrent range 0.1 - 6 mA 0.05 - 3 mACompliance -11 V... +5 V -15 V... +8 VAttorney Docket No. 61435-708601140-160 kHz, max 500Inductive link 120-140 kHz, max 300 mWmWTelemetry 2.4 GHz GFSK, > 5 Mb / s Sub-GHz FSK, > 1 Mb / s

[0671] Control Scenario 1 - Tremor Suppression. Burst repetition ~ 72 Hz (12x 6 Hz tremor), bidirectional polarity every 250 ms; MPC reduces PSD ~ 70 %.

[0672] Control Scenario 2 - PNI Conditioning. 12 Hz, 20 % duty, ENG-adaptive 0.4-0.8 mA; CAP amplitude ↑ 25 % in 4 weeks.

[0673] Control Scenario 3 - Maintenance Mode. 2 Hz tonic, 30 min / day; impedance check each session.VII Illustrative Use-Cases

[0674] Tremor-Suppression Patient - A 72-year-old with medication-refractory essential tremor receives an eight-pad median-nerve cuff.

[0675] Baseline programme (maintenance): tonic 2 Hz stimulation for 20 min each night to preserve neuromuscular junction (NMJ) health.

[0676] Responsive programme: when a wearable inertial sensor detects > 47s wrist oscillation in the 4-12 Hz band, the system automatically delivers bidirectional polarity bursts at 12 x the detected tremor frequency. Model-predictive control maintains > 65 % reduction in tremor power while capping daily duty cycle at 35 %.

[0677] Peripheral-Nerve-Injury (PNI) Recovery - A 38-year-old with ulnar-nerve transection undergoes neurorrhaphy and immediate cuff placement proximal to the repair.

[0678] Day 0-30 conditioning: 12 Hz bursts, 20 % duty; closed-loop ENG hill-climb keeps CAP SNR at +6 dB, adjusting current between 0.3 mA and 0.7 mA.

[0679] Longitudinal monitoring: weekly ENG downloads show CAP amplitude growth of -25 % over four weeks, guiding gradual taper of burst amplitude.

[0680] Autonomic Anti-Inflammatory Therapy - A patient with rheumatoid arthritis receives a cervical vagus-nerve cuff. Low-frequency (5 Hz) bursts, 60 s on / 300 s off, are delivered twice daily. Serum TNF-a levels drop 40 % after eight weeks, allowing steroid-dose reduction.

[0681] Chronic Neuropathic Pain - A 55-year-old with tibial-nerve pain receives a 12-pad cuff. High-frequency (10 kHz) tonic stimulation runs 30 min on / 30 min off, combined with optional low-frequency bursts triggered by breakthrough-pain button presses. Visualanalogue pain scores fall from 8 / 10 to 3 / 10 at three-month follow-up.Attorney Docket No. 61435-708601

[0682] Maintenance / Disuse-Atrophy Prevention - Following rotator-cuff repair, a shoulder-brace-mounted external coil powers a suprascapular-nerve cuff. Tonic 2 Hz pulses (15 min every six hours) preserve deltoid bulk during immobilisation; ultrasound confirms < 5 % cross-sectional area loss at six weeks versus -15 % in historical controls.VIII Kit Articles

[0683] A therapy kit may include, in sterile or non-sterile packaging as appropriate:Item PurposeOne or more multi-pad cuff electrodes (various Chronic nerve interfaceinner diameters)Implantable Electronics Module (IEM) withStimulation / record hubintegrated coilExternal power-transfer coil or charger pad Inductive energy and data link Optional wearable controller enclosure Houses battery, UI, processor Single-use insertion and sizing tools Atraumatic cuff placement External programming dongle or tablet app Parameter setup and data download Printed or digital Instructions for Use (IFU) Regulatory-compliant guidance Sterile lead-anchors / strain-relief sleeves Secures lead in situ

[0684] Kits may be supplied in variant SKUs (e.g., “Tremor Package,” “PNI Package”) that differ only in pre-loaded software profiles and accessory mix.IX Software & Computer-Program Product

[0685] Implant firmware - real-time OS (or bare-metal loop) that may manage sensing, stimulation, safety -watchdogs, encrypted telemetry, and over-the-air (OTA) update authentication.

[0686] Wearable / clinician GUI - cross-platform application providing parameter editors, real-time waveform visualisation, ENG analytics dashboards, and automated report generation.

[0687] Cloud analytics layer - optional HIPAA-compliant service that may store longitudinal ENG / EMG data, apply machine-learning models to detect recovery trends, and push adaptive-parameter “prescriptions” back to the patient device.Attorney Docket No. 61435-708601

[0688] Security architecture - 256-bit AES-GCM session keys derived from device-unique PUF; firmware images signed with ED25519 keys; blockchain audit log for parameter changes.

[0689] Computer-readable medium claim - any non-transitory medium (e.g., flash memory, SSD, optical disc, cloud object store) storing instructions which, when executed by processing circuitry, cause any method or safety routine disclosed herein to be performed. Computer Systems

[0690] The present disclosure provides computer systems that are programmed to implement methods of the disclosure. FIG. 7 shows a computer system 701 that is programmed or otherwise configured to process signals and provide nerve stimulation. The computer system 701 can regulate various aspects of signal processing and transmission of the present disclosure, such as, for example, stimulation commands to delivery of electrical pulses for restoring volitional movement. The computer system 701 can be an electronic device of a user or a computer system that is remotely located with respect to the electronic device. The electronic device can be a mobile electronic device.

[0691] The computer system 701 includes a central processing unit (CPU, also “processor” and “computer processor” herein) 705, which can be a single core or multi core processor, or a plurality of processors for parallel processing. The computer system 701 also includes memory or memory location 710 (e.g., random-access memory, read-only memory, flash memory), electronic storage unit 715 (e.g., hard disk), communication interface 720 (e.g., network adapter) for communicating with one or more other systems, and peripheral devices 725, such as cache, other memory, data storage and / or electronic display adapters. The memory 710, storage unit 715, interface 720 and peripheral devices 725 are in communication with the CPU 705 through a communication bus (solid lines), such as a motherboard. The storage unit 715 can be a data storage unit (or data repository) for storing data. The computer system 701 can be operatively coupled to a computer network (“network”) 730 with the aid of the communication interface 720. The network 730 can be the Internet, an internet and / or extranet, or an intranet and / or extranet that is in communication with the Internet. The network 730 in some cases is a telecommunication and / or data network. The network 730 can include one or more computer servers, which can enable distributed computing, such as cloud computing. The network 730, in some cases with the aid of the computer system 701, can implement a peer-to-peer network, which may enable devices coupled to the computer system 701 to behave as a client or a server.Attorney Docket No. 61435-708601

[0692] The CPU 705 can execute a sequence of machine-readable instructions, which can be embodied in a program or software. The instructions may be stored in a memory location, such as the memory 710. The instructions can be directed to the CPU 705, which can subsequently program or otherwise configure the CPU 705 to implement methods of the present disclosure. Examples of operations performed by the CPU 705 can include fetch, decode, execute, and writeback.

[0041] The CPU 705 can be part of a circuit, such as an integrated circuit. One or more other components of the system 701 can be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0693] The storage unit 715 can store files, such as drivers, libraries and saved programs. The storage unit 715 can store user data, e.g., user preferences and user programs. The computer system 701 in some cases can include one or more additional data storage units that are external to the computer system 701, such as located on a remote server that is in communication with the computer system 701 through an intranet or the Internet.

[0694] The computer system 701 can communicate with one or more remote computer systems through the network 730. For instance, the computer system 701 can communicate with a remote computer system of a user. Examples of remote computer systems include personal computers (e.g., portable PC), slate or tablet PC’s (e.g., Apple® iPad, Samsung® Galaxy Tab), telephones, Smart phones (e.g., Apple® iPhone, Android-enabled device, Blackberry®), or personal digital assistants. The user can access the computer system 701 via the network 730.

[0695] Methods as described herein can be implemented by way of machine (e.g., computer processor) executable code stored on an electronic storage location of the computer system 701, such as, for example, on the memory 710 or electronic storage unit 715. The machine executable or machine readable code can be provided in the form of software. During use, the code can be executed by the processor 705. In some cases, the code can be retrieved from the storage unit 715 and stored on the memory 710 for ready access by the processor 705. In some situations, the electronic storage unit 715 can be precluded, and machine-executable instructions are stored on memory 710.

[0696] The code can be pre-compiled and configured for use with a machine having a processer adapted to execute the code, or can be compiled during runtime. The code can be supplied in a programming language that can be selected to enable the code to execute in a pre-compiled or as-compiled fashion.Aspects of the systems and methods provided herein, such as the computer system 701, can be embodied in programming. Various aspects of the technology may be thought of asAttorney Docket No. 61435-708601“products” or “articles of manufacture” typically in the form of machine (or processor) executable code and / or associated data that is carried on or embodied in a type of machine readable medium. Machine-executable code can be stored on an electronic storage unit, such as memory (e.g., read-only memory, random-access memory, flash memory) or a hard disk. “Storage” type media can include any or all of the tangible memory of the computers, processors or the like, or associated modules thereof, such as various semiconductor memories, tape drives, disk drives and the like, which may provide non-transitory storage at any time for the software programming. All or portions of the software may at times be communicated through the Internet or various other telecommunication networks. Such communications, for example, may enable loading of the software from one computer or processor into another, for example, from a management server or host computer into the computer platform of an application server. Thus, another type of media that may bear the software elements includes optical, electrical and electromagnetic waves, such as used across physical interfaces between local devices, through wired and optical landline networks and over various air-links. The physical elements that carry such waves, such as wired or wireless links, optical links or the like, also may be considered as media bearing the software. As used herein, unless restricted to non-transitory, tangible “storage” media, terms such as computer or machine “readable medium” refer to any medium that participates in providing instructions to a processor for execution.

[0697] Hence, a machine readable medium, such as computer-executable code, may take many forms, including but not limited to, a tangible storage medium, a carrier wave medium or physical transmission medium. Non-volatile storage media include, for example, optical or magnetic disks, such as any of the storage devices in any computer(s) or the like, such as may be used to implement the databases, etc. shown in the drawings. Volatile storage media include dynamic memory, such as main memory of such a computer platform. Tangible transmission media include coaxial cables; copper wire and fiber optics, including the wires that comprise a bus within a computer system. Carrier-wave transmission media may take the form of electric or electromagnetic signals, or acoustic or light waves such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media therefore include for example: a floppy disk, a flexible disk, hard disk, magnetic tape, any other magnetic medium, a CD-ROM, DVD or DVD-ROM, any other optical medium, punch cards paper tape, any other physical storage medium with patterns of holes, a RAM, a ROM, a PROM and EPROM, a FLASH-EPROM, any other memory chip or cartridge, a carrier wave transporting data or instructions, cables or linksAttorney Docket No. 61435-708601transporting such a carrier wave, or any other medium from which a computer may read programming code and / or data. Many of these forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0698] The computer system 701 can include or be in communication with an electronic display 735 that comprises a user interface (UI) 740 for providing. Examples of UIs include, without limitation, a graphical user interface (GUI) and web-based user interface.

[0699] Methods and systems of the present disclosure can be implemented by way of one or more algorithms. An algorithm can be implemented by way of software upon execution by the central processing unit 705.

[0700] While preferred embodiments of the present disclosure have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. It is not intended that the disclosure be limited by the specific examples provided within the specification. While the disclosure has been described with reference to the aforementioned specification, the descriptions and illustrations of the embodiments herein are not meant to be construed in a limiting sense. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the disclosure. Furthermore, it shall be understood that all aspects of the disclosure are not limited to the specific depictions, configurations or relative proportions set forth herein which depend upon a variety of conditions and variables. It should be understood that various alternatives to the embodiments of the disclosure described herein may be employed in practicing the disclosure. It is therefore contemplated that the disclosure shall also cover any such alternatives, modifications, variations, or equivalents. It is intended that the following claims define the scope of the disclosure and that methods and structures within the scope of these claims and their equivalents be covered thereby.

[0701] The computer systems integral to the neuromodulation platform may be designed to acquire, process, and analyze high-frequency biosignal data, execute adaptive machine learning algorithms, and generate precise stimulation commands in real time. These systems may include a central processing unit (CPU) that may be single-core or multi-core, coupled with volatile memory (e.g., RAM) and non-volatile memory (e.g., flash, ROM, or hard disk storage) to store the operating system, application software, calibration data, andpatient-specific profiles. A dedicated digital signal processor (DSP) or embedded microcontroller may be optimized for low-latency signal processing. Communication interfaces (e.g., Bluetooth, Wi-Fi, or proprietary RF links) may enable secure, bidirectionalAttorney Docket No. 61435-708601data exchange and over-the-air software updates, while a user interface module may provide real-time visualizations and control. Robust encryption (e.g., AES-256) and network connectivity (e.g., integration with EMR and telehealth systems) may ensure continuous, secure operation.

[0702] The computer systems may be designed to operate within a networked environment, potentially integrating with hospital information systems, electronic medical records (EMR), and telehealth platforms. This network connectivity may support remote diagnostics, continuous performance monitoring, and data aggregation for large-scale machine learning analyses, enabling the system to adapt to evolving clinical needs and patient-specific responses over time.

[0703] Modularity and scalability may be key features of the computer system architecture, which may support integration of additional sensor modalities, updated machine learning models, and future communication protocols. This flexibility may allow the neuromodulation system to be upgradable and adaptable for expanded therapeutic capabilities. The system may employ advanced power management strategies to ensure long-term, reliable operation in both implanted and wearable configurations. Energy-efficient microcontrollers and sleep modes may be used to minimize power consumption during periods of inactivity. The device may incorporate rechargeable batteries, such as lithium-ion or lithium-polymer cells, with an expected operational lifetime of approximately 12 to 24 hours under continuous use and rapid recharge cycles. In addition, the packaging may incorporate thermal management solutions, such as integrated heat sinks or conductive materials, to dissipate heat generated during operation. These design considerations may contribute to both the safety and the longevity of the device. The system may be designed to interface seamlessly with existing medical infrastructure. Standard communication protocols such as HL7 or FHIR may be supported to enable integration with hospital information systems and electronic medical records (EMR). Data generated by the device — including sensor logs, calibration data, and therapy session metrics — may be transmitted securely to cloud-based platforms for remote monitoring and diagnostics. This connectivity may also facilitate telehealth applications, allowing clinicians to access real-time data and adjust stimulation parameters remotely. The system may further support centralized data aggregation and analysis, enabling large-scale machine learning studies and long-term clinical research.

[0704] In summary, the computer systems described may provide the computational power, connectivity, and security necessary to support a sophisticated closed-loop neuromodulation platform, enabling high-speed processing of multi-modal biosignals, execution of adaptiveAttorney Docket No. 61435-708601algorithms, and real-time control of electrical stimulation, thereby playing a pivotal role in restoring coordinated volitional motor function.EXEMPLARY EMBODIMENTSEmbodiment 1 - Upper Limb Neuromodulation for Elbow Flexion:

[0705] A closed-loop system may be configured to restore elbow flexion in post-stroke patients. Spiral nerve cuff electrodes may be implanted around peripheral nerves (e.g., musculocutaneous, radial, median, and ulnar) to enable targeted stimulation. High-frequency EMG from the biceps brachii may be acquired (via implanted or wearable sensors), processed into a bi-dimensional intent vector (capturing both amplitude and slope), and decoded by machine learning algorithms that dynamically map the graded intent to stimulation parameters. With an update cycle of 10-20 ms and total latency below 20 ms, the system may produce smooth, proportional muscle contractions that may restore natural elbow flexion. A wearable EMG sensor, positioned over the target muscle (such as the biceps or trapezius), may capture low-amplitude, volitional muscle activity and may communicate wirelessly with an external unit. This external unit may deliver transcutaneous power and control signals, thereby eliminating the need for an internal battery. The system may be adapted from CorTec’s Brain Interchange System (FDA Master File MAF3560) and incorporate features of multiple embodiments.Embodiment 2 - Selective Stimulation via EMG-Controlled Brachialis Activation:

[0706] This embodiment may selectively stimulate the nerve branch to the brachialis muscle — a key contributor to elbow flexion — using EMG signals from the biceps brachii as the control input. During calibration, control thresholds based on biceps activity may be established such that when the biceps signal exceeds a predetermined level, stimulation may be directed specifically to the brachialis branch. Alternatively, if stimulation is applied to the entire musculocutaneous nerve, supplementary signals from the trapezius may be incorporated to refine the output. Field steering techniques may be employed to focus the electrical field on motor fibers, thereby reducing off-target activation of sensory fascicles. Embodiment 3 - Coordinated Multi-Joint Activation:

[0707] The system may integrate biosignals from multiple muscle groups (e.g., EMG signals from both proximal and distal muscles) along with inertial sensor data to generate a composite intent vector that reflects a holistic motor plan. This composite vector may then be used to coordinate stimulation across multiple nerve channels through dedicated synchronization schedulers and dynamic sensor fusion algorithms, enabling the execution ofAttorney Docket No. 61435-708601complex, multi -joint movements — such as coordinated reaching or grasping — that are tailored to the patient’s graded volitional intent.Embodiment 4 - Non-Invasive Software-Only Configuration:

[0708] A software-only version may utilize wearable sensors (e.g., surface EMG, IMUs, and optionally EEG) to acquire biosignal data externally. These signals may be wirelessly transmitted to a processing unit that employs the same adaptive algorithms as the implantable system, while an external stimulation device may deliver the computed electrical pulses to the peripheral nerves. This non-invasive approach may reduce surgical complexity and facilitate rapid deployment in clinical and rehabilitation settings.Embodiment 5 - Cloud-Connected Remote Tuning and Diagnostics:

[0709] The system may be integrated with a secure cloud-based management platform that may continuously transmit real-time sensor data, performance metrics, and patient-specific calibration profiles to a remote server. Clinicians may access this data via a web-based interface to remotely monitor therapy sessions, adjust stimulation parameters, and deploy over-the-air software updates. This connectivity may facilitate continuous improvement of the machine learning algorithms and enhanced personalization of therapy.Embodiment 6 - Lower Limb Reanimation:

[0710] For patients with lower limb deficits, a multi-cuff electrode arrangement may target leg peripheral nerves to control gait and balance. Multi-modal sensor fusion (incorporating EMG, ENG, and inertial data) may decode the patient’s graded volitional intent for leg movement, enabling the system to deliver coordinated, graded stimulation that may facilitate functional gait restoration and improved balance.Embodiment 7 - Enhanced Safety and Redundancy:

[0711] The system may include multiple biosignal sensors and a dual-path processing architecture to ensure continuous, reliable operation even in the event of sensor failure or signal degradation. Continuous monitoring of signal integrity may allow the system to automatically switch to backup sensor inputs or revert to predetermined safe stimulation profiles if anomalies are detected, ensuring safe neuromodulation.Embodiment 8 - Adaptive Reinforcement Learning for Therapy Optimization:

[0712] An adaptive reinforcement learning algorithm may continuously refine the mapping between the graded intent vector and stimulation parameters by integrating objective performance metrics, clinician feedback, and patient responses over multiple therapy sessions. This continual adaptation may lead to increasingly personalized and effective therapy, thereby promoting long-term neuroplasticity and improved rehabilitation outcomes.Attorney Docket No. 61435-708601Embodiment 9 - Selective Stimulation of Biceps Brachii Controlled by Trapezius EMG:

[0713] In this embodiment, the system may achieve selective activation of the biceps brachii by employing EMG signals from the trapezius as an alternative control input when biceps EMG may be weak. During calibration, thresholds based on trapezius activity may be established. When the trapezius signal exceeds these thresholds, stimulation may be selectively delivered to the biceps branch, producing a contraction that is proportional to the graded volitional intent.Embodiment 10 - Dual-Signal Control Using Biceps EMG for Activation and Triceps EMG for Deactivation:

[0714] A dual-signal strategy may be implemented wherein biceps EMG may serve as the activation signal and triceps EMG may serve as the deactivation signal. When biceps activity exceeds a predefined threshold, stimulation may be initiated to induce contraction; concurrently, if triceps activity rises above a set deactivation threshold, the system may gradually reduce or terminate stimulation, thereby closely mimicking natural agonistantagonist coordination.Embodiment 11 - Hybrid Control Using Biceps ENG for Activation and Triceps EMG for Deactivation:

[0715] This embodiment may employ a hybrid approach wherein ENG signals from the biceps branch may serve as the activation input — providing enhanced sensitivity in scenarios where EMG alone may not suffice — while triceps EMG may be monitored for deactivation. The system may be calibrated to recognize specific ENG patterns associated with voluntary biceps activation, triggering stimulation accordingly, while concurrent triceps EMG monitoring may ensure a controlled reduction in stimulation when the intended contraction ceases.Additional Embodiments:

[0716] The system may also be adapted for applications such as trunk stabilization, diaphragmatic pacing, and sensory restoration. Each embodiment leverages the core architecture — comprising multi-modal sensor acquisition, adaptive machine learning-based graded intent decoding, and precise closed-loop stimulation — to address various neurological and neuromuscular conditions. These embodiments underscore the system’s versatility and broad clinical utility. Each embodiment is provided by way of example and is not intended to limit the scope of the invention; variations, modifications, and alternative configurations may be implemented without departing from the inventive concepts.Attorney Docket No. 61435-708601

[0717] The implantable unit may include a multi-channel stimulation circuit integrated within a flexible, biocompatible silicone cuff electrode. The cuff may be surgically placed around the musculocutaneous nerve to modulate the muscles responsible for elbow flexion.

[0718] An available EMG sensor may be positioned over target muscles (e.g., biceps or trapezius) to capture low-level, volitional muscle activity. An external pulse transmitter may then wirelessly receive the EMG signal and may provide transcutaneous power to the implant.

[0719] The system’s adaptive Al may continuously monitor incoming EMG signals and may dynamically adjust stimulation parameters — including current amplitude, frequency, pulse width, and ramp times — in real time to ensure smooth ramp-up and ramp-down for natural, graded muscle contractions. Additional sensors (e.g., IMUs or ENG sensors) may provide further feedback. Telehealth integration may allow for remote monitoring, parameter adjustments, and secure data logging.

[0720] Volitional EMG signals may trigger the external transmitter, which may then energize the implanted cuff electrode to activate targeted motor nerve fascicles. Closed-loop feedback from the implant (and optionally from additional sensors) may enable real-time modulation of stimulation to maintain consistent movement under variable loads.EXAMPLES

[0721] The following examples are included for illustrative purposes only and are not intended to limit the scope of the inventive concepts.Example 1 - Open-Loop Prototype Testing in a Porcine Model:

[0722] The objective of example 1 was to test the clinical prototype of the System (implantable stimulation board and cuff electrode) in a 40-kg male Yucatan minipig ischemic stroke model. The findings demonstrated limb movement at multiple post-stroke time points (pre-stroke, immediately post-stroke, and 1-3 hours post-stroke). A multichannel cuff electrode (5 mm inner diameter, 1.5 cm length) arranged in two or more rings with rectangular platinum or platinum-iridium contacts was used. The femoral neurovascular bundle was identified using palpation and ultrasound. The cuff electrode was carefully placed around the femoral nerve with appropriate surgical techniques, and vital signs were monitored to confirm minimal nociceptive response. The results of experiment 1 found that the system elicited graded muscle contraction, smooth limb relaxation, controlled limb descent, twitch movements, and sustained contraction with EMG amplitudes from 50 to 1000 Stimulation parameters were varied via a computer interface that communicated wirelessly. No adverse autonomic responses were observed. Quantitative data indicated a dose-Attorney Docket No. 61435-708601dependent correlation between delivered current and maximal EMG amplitude, with intercoil spacing flexibility of up to 2.5 cm. A clinical prototype featuring a multi-channel spiral nerve cuff electrode (inner diameter -5 mm; length -1.5 cm) may be implanted around the femoral nerve in a Yucatan minipig following established protocols. Surface EMG sensors may capture quadriceps activity, and preset stimulation pulses may be delivered in an open-loop configuration. Graded, reproducible muscle contractions (with EMG amplitudes of 50-1000 pV) may be observed, demonstrating a dose-dependent relationship between stimulation current and muscle response.Example 2 - Closed-Loop Operation in a Combined Human-Porcine Model:

[0723] The objective of example 2 was to test the complete closed-loop clinical prototype in a 50-kg male Yucatan minipig model. The system operated with stimulation parameters (current range 0.1-20 mA, repetition 1-50 pulses / sec, pulse duration 10-200 is) and used both monophasic and biphasic, asymmetric, charge-balanced waveforms with a 100 ps interphase delay. The femoral neurovascular bundle was identified using palpation and ultrasound. The cuff electrode was carefully placed around the femoral nerve with appropriate surgical techniques, and vital signs were monitored to confirm minimal nociceptive response. The findings demonstrated that the system successfully recorded and processed surface EMG signals, with the microcontroller processing data every 16 ps. The system delivered biphasic current pulses upon detection of EMG activity, enabling graded, volitional control of nerve stimulation. The outcome of example 2 was that the implanted stimulation board connected to a 2-channel cuff electrode (4 mm inner diameter, platinum-iridium) stimulated the femoral nerve, resulting in consistent limb movement without muscle fatigue and with no significant autonomic responses. Closed-loop functionality was validated using human EMG signals as triggers in a combined human — porcine model, demonstrating that even minimal (grade 1 / 5) muscle activity may initiate effective stimulation. Quantitative data indicated a dosedependent correlation between delivered current and maximal EMG amplitude, with intercoil spacing flexibility of up to 2.5 cm.

[0724] In a combined model study, a wearable sensor placed over the biceps of a human subject may record low-amplitude volitional EMG signals. These signals may be processed in real time to generate a graded, bi-dimensional intent vector that is dynamically mapped to stimulation parameters for the porcine femoral nerve. With a control loop latency below 20 ms, even minimal volitional signals (e.g., grade 1 / 5 activity) may trigger effective, coordinated stimulation, resulting in smooth, graded limb movements.Example 3 - Comprehensive Bench Testing:Attorney Docket No. 61435-708601

[0725] Extensive bench testing validated the entire signal acquisition chain (including high-pass, notch, and low-pass filtering, and sliding RMS envelope extraction), the accuracy of the machine learning model in decoding the graded intent vector under simulated conditions, and the closed-loop latency (which may consistently remain below 20 ms).Secure, wireless communication channels for over-the-air updates were also evaluated, confirming that the system meets all design specifications.Example 4 - Remote Tuning and Cloud-Based Diagnostics:

[0726] The system may be integrated with a secure cloud-based management platform that continuously transmits real-time sensor data and performance metrics to a remote server. Clinicians may access these data via a web-based interface to adjust stimulation parameters, monitor therapy sessions, and deploy over-the-air software updates, demonstrating seamless integration with clinical infrastructures and enhanced personalized therapy. Clinical feasibility may have been supported by an IRB-approved study at the University of California, San Diego (IRB #808314) evaluating post-stroke patients. In this study, the paretic biceps may have retained approximately 5-35% of normal EMG amplitude while the trapezius exhibited about 40-60% activity, thereby supporting the dual-sensor approach and adaptive control strategy. The successful performance of a pre-clinical prototype may further attest to the practical viability and clinical potential of the system. The objective of example 4 was to test wearable EMG sensor. Demographic and experimental results are shown in Table 1 below.Pt l Pt 2 Pt3 Pt 4 Pt 5 Pt 6 Coho rt Avg Age 56 37 33 39 46 62 45.5 (years)Gender F M M M M M 5:1(M / F) Weeks 0.5 <1 8 12 <1 0.5 >1 yr <1 poststrokeStroke Left Left Left Left Right Left Left Left 75% Localiza M2 Left tion 25%RightAttorney Docket No. 61435-708601Type of Ischemi Hemorrh Ischemi Ischemi Ischemi Hemorrhstroke c agic c c c agicMRC 4= / 5 1 / 5 2 / 5 4 / 5 2 / 5 4 / 5 3 / 5 3 / 5 Avg —3 / 5 Biceps 70.17 1.2% 7.8% 61% 10.58 45.09% 35.99 31.27%% % %NormalTrap % N / A 18 %23.2% 54.7 N / A N / A N / A N / AnormalTable 1Example 5 - IRB-Approved Feasibility Study in Post-Stroke Patients:

[0727] In a pilot IRB-approved study, wearable EMG sensors may record volitional signals from post-stroke patients during controlled therapy sessions. Preliminary outcomes may indicate that the system can reliably detect graded, high-resolution EMG signals and adaptively deliver stimulation to restore functional, graded movement. This feasibility study may confirm the clinical potential of the adaptive, closed-loop approach.7. Detailed Description of Illustrative Embodiments7.1 System Architecture — Pillar A: Implant & Electrode HardwareID Description InnovationA-l A neural interface system comprising a spiral cuff Any individual pad can alternately electrode having a plurality of conductive pads, stimulate or sense, so the same lead set each pad being selectively and dynamically delivers double functionality, coupled, for example via an integrated minimizing implant volume, surgical C ompl ementary Metal -Oxi de- S emi conductor passes, and bill-of-materials (BOM) (CMOS) switch matrix disposed within the cost.implantable pulse generator, to either (i) astimulation current driver or (ii) a low-noiseneural-recording front end, the switch matrixbeing re-configurable on a per-pulse basis andmay have a latency of no more than about fivemicroseconds.A-2 The system of A-l wherein control firmware Enables sub-millisecond capture of assigns at least one proximal pad to a recording residual ENG without adding new mode and at least one distal pad to a stimulation hardware.mode within a common stimulation cycle, suchthat an inter-pad spacing of about 5-10 mm (forexample) yields at least 15 dB artefact attenuationat the recording input.Attorney Docket No. 61435-708601ID Description InnovationA-3 The system of any preceding claim further Provides a fixed temporal head-start for comprising a proximal “listening” cuff and a shaping the next pulse, yielding distal “talking” cuff positioned along a common smoother multi -joint movement and nerve trunk with an inter-cuff spacing of about built-in redundancy if a cuff fails. This 20-60 mm (for example), thereby introducing a also spatially separates recording and deterministic conduction delay of about 0.6-2 ms stimulating electrodes to mitigate that is exploited by predictive control logic; an stimulation artifact.optional third cuff may act as an afferent brake.This also spatially separates stimulation electrodesfrom recording electrodes, thereby reducingstimulation artifact.A-4 The neural interface wherein a runtime scheduler Extends functional lifetime and reduces automatically re-assigns any cuff or pad between revision surgeries by migrating away recording, stimulating, and reference roles based from noisy sites.on real-time signal-to-noise metrics orclinician-defined presets.A-5 The implantable pulse generator comprising a Allows selective activation of targeted multi-source current driver having N fascicles and proportional coindependently programmable current mirrors with contraction ratios for fine motor control. at least 8-bit resolution, the driver beingconfigured to field-steer stimulation by splittingor scaling current across selected pads, therebysupporting monopolar, bipolar, tripolar orquadrupolar configurations.A-6 The implantable pulse generator further Eliminates bulky primary cells, comprising a planar or coaxial secondary coil for mitigates MR-safety issues, and enables transcutaneous inductive energy and data transfer miniaturized or pediatric implants. and firmware supporting (i) a battery -free passivemode, (ii) a hybrid rechargeable mode, and (iii) afully battery-powered mode interchangeably.

[0728] (All numeric values serve only as non-limiting examples; other dimensions, materials, and electrical specifications may be substituted without departing from the scope.)7.1.1 Overview of the Implantable Pulse Generator (IPG)

[0729] An IPG housing may enclose (i) a rechargeable or battery-free power subsystem, (ii) a stimulation / recording application-specific integrated circuit (ASIC), (iii) a low-noise analog front-end (AFE) for electroneurography (ENG), (iv) a 32-bit micro-controller or RISC-V core executing the closed-loop firmware, and (v) a bidirectional radio such as Bluetooth Low-Energy (BLE), Wi-Fi, or proprietary sub-GHz telemetry.Attorney Docket No. 61435-708601

[0730] Power: The energy store may be a 10-70 mW h Li-ion pouch cell, a solid-state thin-film battery, or omitted entirely when wireless power is available.

[0731] Envelope size: A titanium or polymer can can occupy < 10 cm3for upper-arm implants or < 8 cm3for wrist-level implants.

[0732] Connectors: A multi-pin hermetic feed-through may route discrete lines (e.g., eight cathodes + two anodes + two references) to the electrode lead.

[0733] An illustrative eight-contact spiral cuff 12 suitable for use with the present IPG is shown in FIG. IB.7.1.2 Bidirectional Electrode Channels (A-l)

[0734] Each cuff contact may be hard-wired to a CMOS transmission-gate pair driven by a 3 -bit or 4-bit multiplexer address. On every 16- to 33 -ms control cycle, firmware can toggle the gate states in < 5 ps, thereby assigning the pad to:

[0735] Record mode — connected to the AFE (input impedance > 4 MQ, 1 Hz-5 kHz band).

[0736] Stimulate mode — connected to one of N current mirrors in the driver array.

[0737] High-Z reference — floated for common-mode sensing or artefact cancellation.

[0738] This architecture may allow an eight-pad cuff to perform six-pad recording + two-pad stimulation in one cycle and invert that ratio in the next, enabling spatial mapping without extra leads.7.1.3 Proximal-Record / Distal-Stim Separation (A -2)

[0739] Within a single spiral cuff that may be 4-6 mm wide, the most proximal 2-4 pads may be reserved for recording while the distal pads can deliver stimulation. A physical pad-to-pad gap of 2-10 mm (or longer) may attenuate the stimulation artefact at the AFE input by 15-20 dB, allowing residual ENG capture < 250 ps after pulse termination. Firmware may update the pad mask weekly based on artefact statistics, so the proximal / distal roles can migrate if scar tissue or minor cuff rotation shifts the optimal sites.7.1.4 Dual- and Tri-Cuff Topologies (A-3)

[0740] Where larger conduction-delay windows are useful, a proximal “listening” cuff and a distal “talking” cuff may be spaced 20-60 (or shorter / longer) mm apart along the same nerve trunk, yielding a deterministic 0.6-2 ms latency. A representative dual-cuff layout with a 35 mm separation (AL) is depicted in FIG. 2. An optional sensory-only cuff can be inserted 10-20 mm proximally to detect high-frequency afferent volleys for spasticity -break routines. Lead design may employ a multi-branch Y-split or a daisy-chain ribbon; insulation can be silicone, Pebax®, or stretchable polyurethane.7.1.5 Multi-Cuff Cooperative Scheduling (A-4)Attorney Docket No. 61435-708601

[0741] A real-time scheduler may compute per-cuff signal-to-noise ratio (SNR) and channel impedance; if SNR at any cuff falls > 6 dB below baseline for > 30 s, the scheduler can reassign that cuff to reference duty while shifting record or stim tasks to the cleaner cuff(s). This role-rotation may extend functional lifetime when local scar formation, motion, or micro-lead fracture damages a subset of contacts.7.1.6 Field-Steering Driver (A-5)

[0742] The stimulation ASIC may contain N independent current mirrors (N = 1-32) with 8-bit digital-to-current converters; mirrors can be combined to deliver up to 6 mA per phase into 1 k tissue load. Pulse shaping may support monopolar, bipolar, bipolar, or quadrupolar patterns on a pulse-by-pulse basis. Proportional current splitting can bias the electric field toward motor-dominant fascicles identified by the adaptive spike classifier (§ C-1), thereby minimizing spill-over to antagonists.7.1.7 Wireless Inductive Power & Hybrid Battery Options (A-6)

[0743] Fully passive mode — A 6.78 MHz or 13.56 MHz receive coil may harvest both energy and a reference clock; stimulation can operate only when an external transmitter patch is in place (e.g., night-time therapy).

[0744] Hybrid mode — A 10-50 mW h micro-battery may buffer harvested energy for untethered daytime sessions; coil recharge can occur during rest or sleep.

[0745] Batery-only mode — Standard Li-ion pouch cell + optional recharge coil for longterm outpatient use.

[0746] Power management IC may limit charge current to < 20 mA and can enforce IEC 60601 temperature rise < 2 °C at the coil interface.7.1.8 Materials & Biocompatibility

[0747] Cuff substrate may be implant-grade silicone (50 ShA) or stretchable polyurethane; wall thickness 150-300 pm.

[0748] Contacts can be laser-cut Pt-Ir or sputter-deposited Ti / Au with IrOx coating; contact impedance < 500 @ 1 kHz.

[0749] Lead wires may utilise MP35N or Pt / Ir stranded conductors with ETFE or silicone over-mold.

[0750] Hermetic can can be Grade 2 titanium; laser-sealed with an Inconel or titanium feed-through header.

[0751] All contacting materials may comply with ISO 10993 for long-term (> 10 year) implantation.7.1.9 Manufacturability & Modular VariantsAttorney Docket No. 61435-708601

[0752] The electrode subsystem may be assembled as (i) a monolithic over-mold, (ii) an extruded lead with slip-on cuffs, or (iii) a one- or two-piece “clamshell” cuff for intraoperative closure without nerve traction. Because each pad is software-defined, the same hardware can support elbow flexion, wrist extension, or ankle dorsiflexion simply by altering the stimulation and recording masks. There can be one or more cuffs to allow for stimulation of one or more nerves and the resulting software would allow for synchronous coordinated functional movement amongst the stimulated muscle groups.

[0753] 7.2 Artefact Suppression & Signal Acquisition — Pillar BID Description InnovationB-l A direction-aware adaptive-blanking method Recovers 1-2 ms of usable ENG per wherein comparator logic determines artefact cycle, enabling 5-10 x faster control polarity to infer orthodromic versus antidromic than EMG-triggered FES. propagation, and a blanking timer terminateswhen artefact magnitude falls below aprogrammable threshold of about 5-10 pV,thereby opening a residual-ENG recordingwindow no later than about 0.25 ms after astimulation pulse.B-2 The method of B-l further comprising predictive Maintains < 5 pV residual artefact at 3 template subtraction executed on a 32-bit ARM- mA without dedicated DSP hardware. M0+ microcontroller in less than about 20 ps andconsuming under about 100 pW, the algorithmupdating artefact coefficients on every pulse in aKalman-style framework.B-3 The system of any preceding claim including an Further shortens blanking time and active-cancellation reference electrode configured tolerates higher stimulation currents. to inject an opposite-polarity ramp, under openloop or closed-loop servo control, to nullstimulation artefact at the amplifier input.B-4 A multi-rate, interleaved-sampling scheme Conserves battery without forfeiting wherein an on-chip SAR ADC samples at > 500 early spike capture.kS / s for an initial 50-100 ps period after eachpulse and thereafter reduces to < 20 kS / s, therebyreducing average ADC power by about 80 %.B-5 The method of any preceding claim further Provides a secondary noise-rejection comprising wavelet-domain thresholding of layer, yielding cleaner input for residual ENG using at least a four-level downstream ML decoders.Daubechies basis to recover approximately 6 dBsignal-to-noise ratio when template subtractionapproaches its limit.Attorney Docket No. 61435-708601ID Description Innovation B-6 Common-mode adaptive-gain balancing wherein Mitigates temperature- or motion- at least one spare pad samples background induced baseline shifts, reducing potential and a programmable-gain amplifier autorecalibration visits.centers its operating point every ~10 ms tocompensate for drift.

[0754] A high-level firmware flow-chart summarizing these artefact-suppression and closed-loop control blocks is provided in FIG. 3. (All numeric values and component choices are illustrative and may be altered or substituted without departing from the scope of the disclosure.)

[0755] > Across 10000 Monte-Carlo runs spanning ±30 % tissue resistivity and 0-1.5 mm fiber depth, > 99 % of artefact RMS values fall below 5 pV by 0.80 ms (see Fig 4), providing statistical reassurance that the adaptive blanking window safely terminates within the targeted sub-millisecond budget.7.2.1 Direction-Aware Adaptive Blanking (B-l)

[0756] Immediately after each cathodic phase ends, a high-speed comparator may sample opposite contacts to infer whether stimulus current propagated orthodromically or antidromically.

[0757] If propagation is orthodromic, a higher noise threshold Tortho (~ 10 pV) may be applied; if antidromic, a lower threshold Tanti (~ 5 pV) can be used.

[0758] A programmable timer can poll artefact RMS every 10 ps and may release the amplifier as soon as RMS < T(dir).

[0759] Typical mute durations therefore may shrink to ~ 0.50 ms (orthodromic) or 0.70-0.90 ms (antidromic), recovering 1-2 ms of useful ENG per 33-ms cycle at 30 Hz.

[0760] The temporal relationship among the biphasic pulse 12, the direction-aware blanking interval 14, and the residual-ENG capture window 16 is diagrammed in FIG. 1A. A microscale view of artefact decay and adaptive blanking release appears in FIG. 6C.

[0761] Plain language: “We mute the microphone only until the noise drops low enough; if the current runs one way the noise fades faster, so we start listening sooner.”

[0762] Importance: Shorter blanking windows can cut perceived latency by >10* versus fixed-window designs, giving patients near-instant feedback and reducing run-away contraction risk.7.2.2 Predictive Template Subtraction (B-2)Attorney Docket No. 61435-708601

[0763] An 8-tap adaptive filter may update artefact coefficients a = [ai...as] using a recursive-least-squares (RLS) or Kalman gain step that converges within 2-3 pulses.

[0764] Compute time < 20 ps on a 64 MHz ARM-M0+; average digital power ~ 55 pW.

[0765] Residual artefact amplitude can fall below 5 pV for 3 mA monopolar pulses, allowing residual -ENG SNR > 6 dB.

[0766] Plain language: “A tiny chip predicts what the noise spike will look like and subtracts it before we even try to read the nerve signal.”

[0767] Importance: Keeps artefact suppression inside a sub-milliwatt budget — critical for battery-free or ultra-small implants.7.2.3 Active-Cancellation Reference Electrode (B-3)

[0768] A miniature counter-electrode (~ 1 mm2) may inject an opposite-polarity exponential ramp whose amplitude is set to Ainj ~ - - Astim.

[0769] 0 can be calibrated once at implant time or may adapt every 100 pulses via a slow integral servo.

[0770] Open-loop error < 10 pV yields additional 10-12 dB artefact rejection, allowing the blanking interval to terminate < 200 ps post-pulse in low-impedance tissue.

[0771] Plain language: “We play an equal-and-opposite ‘anti-noise’ wave right when we stimulate, like noise-canceling headphones for the nerve.”7.2.4 Multi-Rate Interleaved Sampling (B-4)

[0772] The SAR ADC clock may jump to > 500 kS / s for the first 50-100 ps of the residual-ENG window, then can drop to < 20 kS / s for the remainder of the 16-33 ms cycle.

[0773] Average ADC power reduction ~ 80 % compared with constant 500 kS / s operation.

[0774] Firmware may ramp the sampling rate using a look-up table indexed by pulse number or ENG envelope magnitude.

[0775] Plain language: “We record super-fast only during the brief moment when little spikes fly by, then slow down to save battery.”7.2.5 Wavelet-Domain Denoising (B-5)

[0776] Residual ENG may be segmented into 1-ms frames; a 4-level Daubechies-4 discrete wavelet transform can decompose each frame, apply universal threshold 1 = o (2 In N), and invert to produce denoised ENG.

[0777] Runs in 8-bit fixed-point; compute power ~ 12 pW when enabled.

[0778] Adds 4-6 dB SNR in simulations where template subtraction alone saturates.

[0779] Plain language: “A second math pass cleans up any tiny nerve spikes still hiding under the noise of the stimulation pulse / artifact.”Attorney Docket No. 61435-7086017.2.6 Common-Mode Adaptive Gain Balancing (B-6)

[0780] Two inactive pads may measure common-mode background every 10 ms; a programmable-gain amplifier (PGA) can auto-center its mid-rail via a 7-bit feedback DAC.

[0781] Corrects slow baseline wander (± 50 pV shifts) caused by temperature or limb motion.

[0782] Keeps ENG centered in the 1.8 V ADC span, preserving 8-9 bits of dynamic range.

[0783] Plain language: “The amplifier keeps its zero point in the middle so slow drifts don’t drown out the real nerve signal.”7.2.7 Residual-ENG Window Configuration

[0784] After adaptive blanking clears, firmware may open a residual-ENG capture window AtENG programmable from 0.25 ms to 1.00 ms. Rectified samples feed an exponentialmoving-average filter (T ~ 3 ms) that produces the envelope En; successive differences yield the slope term dE / dt. Both features may be handed to the control block (§ 7.3) on every pulse.7.2.8 Calibration & Self -Test

[0785] A built-in self-test (BIST) mode can inject 100 pA, 200 ps diagnostic pulses to measure artefact decay constants weekly. Coefficient sets for template subtraction and active cancellation may update automatically if decay T shifts > 20%.

[0786] Importance: Allows long-term at-home use without clinic-based recalibration.7.3 ENG Decoding & Proportional Control — Pillar CID Description InnovationC-l An adaptive ENG classifier comprising a two-tap Sustains decoding performance over conduct! on-velocity delay line feeding a long implantation periods. convolutional neural network that self-re-trainsweekly to update motor-versus-sensory labels,thereby remaining accurate despite nerve rotation,axial slide, or fibrosis.C-2 Dual-threshold control logic wherein stimulation Provides natural, low-latencyis initiated when an ENG envelope amplitude and proportional control without external slope exceed a programmable start threshold and buttons and prevents runaway terminated when said metrics fall below an off contractions.threshold for N successive cycles, the samemetrics proportionally modulating pulse width,current, or frequency in real time.Attorney Docket No. 61435-708601ID Description InnovationC-3 A real-time sensor-fusion method wherein ENG Increases robustness in daily-living feature vectors are weighted and fused with scenarios where any single modality inertial-measurement-unit quaternions and / or may degrade.pressure-sensor data via a 3 x 3 Kalman filterexecuting in less than about 50 ps.

[0787] (All algorithms may run on the same micro-controller described in § 7.1.1 or on a copackaged DSP core; clock rate, bit-depth, and parameter values are illustrative and can be adjusted.)7.3.1 Adaptive, Velocity-Selective Spike Classification (C-l)Aspect Technical description Plain-language Why it matters explanationInitial A two-tap delay line may compute “Fast spikes likely Gives anrule-of- conduction velocity v = Ax / At mean the brain is immediate yet thumb across adjacent pads; spikes with v talking to the muscle; crude motor- > 35 m s ' can be tagged “motorslower spikes likely sensory split dominant,” slower spikes “sensory- mean the nerve is without pre-op dominant.” An example sending touch info mapping. conduction-velocity histogram back.”used for this classification ispresented in FIG. 5.SelfOver time, a 1 -layer 1-D CNN (64 “The implant watches Keeps learning x 3 kernel, ReLU, global-avg-pool) which spikes actually classification map may correlate spike patterns with make the arm move, accurate if the limb motion captured by the IMU; then re-labels its pads cuff rotates, weights can update nightly using accordingly.” slides, or tissue an on-device Adam step (q ~ 104, grows scar.256-sample mini-batches).Pad reIf a pad’s motor-spike probability “Pads that stop hearing Maintains high assignment P(M) falls below 0.2 for seven good motor signals SNR on motor days, firmware may re-assign it to switch jobs channels; extends sensory recording or reference automatically.” device life.duty.7.3.2 Dual-Threshold Intent Logic + Proportional Modulation (C-2)

[0788] Macro-level ENG activity, adaptive blanking windows, and the derived amplitude envelope are illustrated in FIGS. 6A and 6B.Attorney Docket No. 61435-708601Componen Technical details Plain-language view Benefit tStart / stop ENG envelope En may be low “A high nerve signal Gives the patient gate pass filtered (T ~ 3 ms). If En > starts the movement; a instant, reliable Ton & dE / dt > 0, stimulation can low one stops it — like on / off control begin within the same 16-33 ms pressing and releasing without buttons. cycle. If En < Toff for Noff cycles, a gas pedal.”output may inhibit. Typicalhysteresis: Ton / Toff ~ 1.4.Proportion During each residual-ENG “The harder the nerve Let’s patients al engine window, firmware may compute fires, the wider or grade effort APW = kl(En - Eref) + k2dE / dt; stronger the next pulse; smoothly — Al = k3(En - Eref). PWM update if the signal ramps up lifting a cup vs. a latency < 24 ps so the very next fast, the device boosts kettle — without pulse can change. A representative even more.” delay. pulse-width (and amplitude)trajectory generated by thisproportional engine is plottedin FIG. 6DAntiPW may saturate at 400 ps; dutyPrevents windup cycle clamp can limit mean charge runaway clamp to 30 pC s ' for tissue safety. stimulation if noise spikes occur.7.3.3 Real-Time Sensor Fusion (C-3)Layer Technical implementation Plain-language view Benefit Kalman State vector x = [E, 9, co], where E “The implant blends Keeps control core = dE / dt, 9 = IMU joint angle, co = nerve spikes with stable if sweat or joint velocity. Prediction step may arm-motion data, posture degrades update at 1 kHz; ENG can be the trusting whichever either signal. primary measurement, IMU looks cleaner at thatsecondary. moment.”Adaptive Measurement noise RENG may “If the nerve signal Seamless weighting grow when template error a > acrit; gets noisy, the system robustness without RIMU can shrink accordingly. leans more on the clinician remotion sensor.” programming. Output to Filtered effort E can feed the “The device aims its Smoother, stim proportional engine; predicted next pulse where the anticipatory engine future effort Epred(+At) may bias arm is about to move, motion; reduced field-steering toward desired not where it was 29 overshoot. fascicles. ms ago.”Attorney Docket No. 61435-7086017.3.4 Integration with Field-Steering Driver

[0789] The proportional engine may call a lookup table that maps requested torque ratio (e.g., brachialis:biceps) to current-split vectors [Ii... IN], The table can be generated intra-operatively or learned via cloud analytics (§ F-2).

[0790] Plain language: “When the patient wants a stronger elbow flexion, the algorithm automatically nudges current toward stronger elbow flexor-fascicle pads.”7.3.5 Firmware Budget & Latency

[0791] Total compute per 33 ms cycle: ~ 180 ps (classification + Kalman + proportional update).

[0792] Worst-case closed-loop latency (stim — ENG capture — next pulse update): 0.75 ms (blanking) + 24 ps (compute) ~ 0.78 ms.

[0793] Average algorithm power: ~ 25 pW @ 16 MHz duty-cycled clock.

[0794] Importance: Sub-millisecond update loop matches or exceeds natural efferent spike timing, enabling fluid, intent-driven movement.7.3.6 Sensory-Feedback Proportional Scaling

[0795] In embodiments in which the electrode assembly encircles a mixed motor-sensory nerve, the controller may further exploit afferent ENG spikes that arise during the stimulation-evoked movement to refine the next pulse. A real-time metric — such as sensory-dominant spike count, conduction-velocity histogram, or firing-rate dispersion — can be mapped to(i) dynamic weighting of the field-steering vector,(ii) proportional scaling of pulse width or current amplitude (or any other stimulation parameter), or(iii) channel-set selection for synergistic muscle activation.

[0796] In this way the implant receives immediate proprioceptive feedback and can adaptively tune muscle recruitment to match the intended kinematic trajectory while suppressing overshoot and co-contraction. All thresholds, mapping functions, and time bases may be clinician-programmable or cloud-optimized. (See FIG. 6B for a concurrent ENG envelope view.)7.4 Safety, Fatigue & Fail-Safe Logic — Pillar DAttorney Docket No. 61435-708601ID Description InnovationD-l A fatigue-management module wherein a 60-s Extends session duration and conforms exponential moving average of conductionto clinical fatigue guidelines. velocity slowdown modulates a burst duty-cyclebetween about 15% and 50%.D-2 A watchdog circuit configured to halt stimulation Provides automatic shutdown in cases when an ENG envelope is below about 3 pV and of lead breakage, seizure, or loss of IMU acceleration is below about 0.05 g for at least patient engagement.about 2 s, said halt being overridable via a BLEcommand.D-3 Automatic sensor-mode fail-over wherein Prevents therapy interruption and persistent low ENG SNR triggers a switch to an obviates urgent explants. external EMG patch or IMU-based controlpathway until ENG quality recovers.

[0797] (All thresholds and time-outs are illustrative and may be adjusted by clinicians, firmware updates, or cloud-pushed profiles.)7.4.1 Fatigue Duty-Cycle Optimizer (D-l)Aspect Technical detail Plain-language BenefitviewMetric Conduction-velocity drift Atcv = “If nerve signals Objective fatigue t50(now) - t50(baseline), where t50 start slowing, the detection — no EMG is ENG peak latency to the 50 % muscle is fatiguing.” electrodes needed. motor pad.Averaging 60 s exponential moving average (a Filters out- 0.03). momentary bursts,focuses on sustained fatigue.DutyIf Atcv > Acrit (e.g., +8 pis) the “The implant gives Extends therapeutic cycle rule firmware may lower ON-time from the nerve mini session, prevents100 % to 75 %, then to 50 % in 5 % breaks when it tires.” over-stimulation steps every 30 s until Atcv injury.recovers.7.4.2 ENG-Silence + IMU Watchdog (D-2)Attorney Docket No. 61435-708601Condition Action RationaleENG envelope < 3 pV and Stimulation inhibited; state Covers lead break, sudden IMU acceleration < 0.05 g latched until fresh ENG > Ton or loss of contact, syncope, or for > 2 s clinician BLE reset. seizure.Rapid ENG burst > 5 * Emergency blanking for 5 s + Protects against external baseline outside residual BLE alert EMI or electro-surgery window interference.BLE “STOP” command or Immediate halt (< 20 ms) Patient or clinician override cloud kill switch at any time.

[0798] Plain language: “If the nerve and the limb both go silent — or go dangerously noisy — the implant shuts itself off instantly.”7.4.3 Automatic Sensor-Mode Fail-Over (D-3)Stage Trigger Fallback control loopWarning ENG SNR < 6 dB for 15 s LED blink or phone pop-up (optional).Fail-over ENG SNR < 6 dB for 60 s Switch proportional control to external EMG patch or IMU.Recovery ENG SNR > 10 dB for > 30 Re-engage ENG-based loop and ramp down fallback s gain.

[0799] Fallback LUT may scale external-sensor effort to match ENG effort, so patients perceive no step-change in torque.

[0800] Plain language: “If cuff signals stay fuzzy, the system quietly hands the wheel to a backup sensor, then switches back when the cuff clears up.”7.4.4 Stimulation & Thermal Limits

[0801] Charge density limit: Firmware may cap cathodic charge per phase to < 0.3 pC mm2on Pt-Ir pads.

[0802] Temperature: IPG thermistor can disable stimulation if case temp > 39 °C or if IEC 60601 tissue-rise model predicts AT > 2 °C.

[0803] Compliance voltage: Driver rail-to-rail ±12 V; over-current and over-voltage events may trigger a 5 s lock-out.7.4.5 OTA Safety Update & A / B Firmware SlotsAttorney Docket No. 61435-708601

[0804] The micro-controller flash may reserve two firmware images (“A” active, “B” shadow). Incoming OTA update can write to the inactive slot with CRC-32 verification; on next reboot, a one-shot watchdog may swap boot pointers. If the new image fails a 10-s heartbeat test, the bootloader can roll back to the previous safe image.

[0805] Plain language: “Software updates have a parachute — if the new version misbehaves, the device auto-reverts to the last good one.”7.4.6 Regulatory & Standards MappingSafety block Relevant standardDuty-cycle & charge ISO 14708-3 §21; FDA Neurological Devices Panel guidance densityThermal modelling & IEC 60601-1 §11; IEC 60601-2-10 annexshut-offEMC watchdog & IEC 60601-1-2 §6blankingSoftware fail-safe & A / B IEC 62304 Class C; FDA “Cybersecurity in Medicalimage Devices” guidance7.5 Therapeutic & Diagnostic Modes — Pillar EID Description InnovationE-l A spasticity-suppression mode configured to First implant-grade, nerve-level detect an afferent volley having a frequency of at alternative to Botox or systemic drugs. least about 250 Hz and, in response, deliver threeantidromic stimulation pulses of approximately0.3 mA within about 50 ms, followed by a one- second lock-out.E-2 A postoperative nerve-regeneration protocol Adds a reimbursable indication without comprising daily bursts of about 20 Hz, 100 ps new hardware; enables at-home therapy. pulse width for 30 min (or any combination ofstimulation parameters and time intervals) whileENG compound-action-potential amplitude andconduction velocity are logged.E-3 A neuroplasticity dashboard wherein weekly Simplifies longitudinal monitoring and ENG entropy, conduction velocity (CV) supports outcome-based reimbursement. dispersion, and envelope power are uploaded viaBLE to a cloud server, and a recovery -indexmetric is computed for clinician review.Attorney Docket No. 61435-708601ID Description InnovationE-4 A burst-randomization stimulation pattern in Improves artefact rejection and may which interpulse intervals are pseudo-randomly reduce neural habituation.varied by ± 15 %, thereby decreasingautocorrelation between artefact and ENG.

[0806] (All modes are software-selectable; any single implant may run one or multiple modes either sequentially or in parallel.)7.5.1 Spasticity Reflex-Breaker Mode (E-l)

[0807] The stimulus-response timeline for this reflex-breaker routine is shown in FIG. 7.Aspect Technical description Plain-language Why it matters viewTrigger Proximal (sensory-dominant) pads may “When the nerve High-frequency detect a volley of >30 spikes within a 30 shows a tell-tale afferent bursts reliably ms window whose instantaneous frequency burst that a spasm precede velocity>250 Hz. is coming...” dependent spasms.Therapy Controller can deliver three 0.3 mA, 200 “... the implant Antidromic activation burst ps cathodic pulses (or any combination of fires a quick of la fibers may stimulation parameters) antidromically via counter-pulse interrupt the spinal the distal cuff, spaced 20 ms apart (or any back up the reflex loop before the combination of time intervals). nerve.” limb jerks.Lock-out 1 s refractory period may prevent repeated — Avoids excessive dutybursts. cycle and preserves charge-density limits.

[0808] Outcome: Many patients can avoid painful, disruptive spasms without systemic drugs.7.5.2 Open-Loop Nerve-Regeneration Mode (E-2)Parameter Default value NotesStim pattern 20 Hz trains, 200 ps PW, 50 ps IPG, 2 mA, 30 Matches pre-clinical min / day (or any combination of stimulation data on axonal parameters and / or time intervals) elongation.Biomarker CAP amplitude, CV histogram, 5 min / month (or any Uploaded via BLE for logging other biomarker or data collected by the system at surgeon review.any time interval)Attorney Docket No. 61435-708601Scheduling Patient phone app can schedule daily sessions; Home-based implant may reject if tissue temp >38 °C. compliance, MR-safe.

[0809] Plain language: “Right after nerve-repair surgery, patients run a 30-minute ‘nerve workout’ each day at home, and the cuff records how fast the axons regrow.”7.5.3 Neuroplasticity Dashboard (E-3)Data captured Processing Clinician output Weekly ENG entropy, Cloud server may run a LSTM — > Single 0-100 “Recovery CV dispersion, CAP linear-regression model trained on Index” plus trend arrow on power fleet data web portal

[0810] Patients can view simplified progress bars.

[0811] API may export data to hospital EMR for outcome-based reimbursement.

[0812] Plain language: “Doctors see one number that tells them if the nerve is getting healthier week-by-week.”7.5.4 Burst-Randomization Stimulation Pattern (E-4)Rule Implementation Benefit Interpulse IP = T ± rand(-15 %, +15 %) per Decorrelates artefact spectrum from interval pulse ENG spectrum.Random seed 16-bit LFSR reseeded every session Predictable for compliance test, random for filters.Safety Mean rate held at programmed value Duty-cycle & charge density remain (e.g., 30 Hz) unchanged.

[0813] Plain language: “We jitter the timing just enough that filters can more easily tell noise from the real nerve spikes — and the nerve doesn’t get ‘bored’ by a metronome-like rhythm.”7.6 Machine-Learning & Cloud Adaptation — Pillar FID Description InnovationF-l An on-device, low-power neural decoder Future-proofs the platform for richer comprising a one-dimensional convolutional decoding without new silicon. neural network with no more than about 5 kparameters, 8-bit fixed-point arithmetic, and anAttorney Docket No. 61435-708601ID Description Innovationenergy budget of less than about 15 pj perinference at 1 kHz update rate.F-2 A federated-learning architecture wherein Provides continuous fleet-wideeach device uploads a compressed 12-byte performance gains while remaining feature vector approximately every ten HIPAA-compliant.minutes, a cloud service re-trains globalweights weekly, and encrypted over-the-airupdates deliver new weights using an A / Brollback mechanism.F-3 A nightly self-calibration routine executed Reduces clinic visits and builds a during a five-minute rest epoch, recalculating valuable longitudinal data asset.noise floors, conduction-velocity histograms,and start / stop thresholds; anonymized statisticsfeed a cohort- wide trend model.

[0814] (All neural-network sizes, radio protocols, and privacy measures may be substituted with equivalents without departing from the scope.)7.6.1 On-Device Tiny-ML Decoder (F-l)Aspect Technical implementation Plain-language Why it matters viewNetwork 1-D CNN — > ReLU — > depth- wise “A postage-stamp- Captures nonlinear ENG separable conv — > global -avg-pool — > size neural net patterns that simple 16-node dense; < 5 k parameters lives in the thresholds miss. quantized to 8-bit INT. implant.”Compute Inference can run at 1 kHz; < 15 “Uses about one- Fits inside the < 100 pW budget pj / inference on a 64 MHz Cortex-M4F third the power of artefact- suppre ssi on with CMSIS-NN kernels. a Bluetooth radio budget.packet.”Training Initial weights may derive from cloud “Ships smart, then Personalizes quickly — source fleet model; local fine-tune can occur keeps learning important when nerve via few-shot SGD when ground truth your specific nerve geometry differs patient- (IMU movement) present. signals.” to-patient. Output [Effort, spike-rate, motor vs sensory — Unified interface to the vector probability] may feed proportional rest of firmware.engine (§ 7.3) and safety gates.7.6.2 Edge-to-Cloud Federated Learning Pipeline (F-2)Attorney Docket No. 61435-708601

[0815] An overview of the edge-to-cloud federated-learning architecture is provided in FIG.8.Stage Technical details Plain-language view Benefit Feature Every 10 min the implant may “Tiny packets — smaller Minimal power & logging log a 12-byte vector: [SNR, CV, than a text message — go bandwidth.CAP power, fatigue flag, sliding to the phone.”ENG entropy].Uplink BLE 5.3 or Wi-Fi HaLow; AES- “No cell data required; Patient-friendly and 128 CCM encryption; packets uploads when the user HIPAA-safe. stored on phone then can upload hits Wi-Fi.”via HTTPS when on Wi-Fi.Aggregation Cloud orchestrator may group “Everyone’s implant Continual learning 1000 devices, average gradients helps teach the others, without raw data — in FedAvg, discard outliers >3 o. but bad data get tossed.” privacy by design. Model New weight blob (< 6 kB) can be “Updates only install if Cyber-secure and update signed with ECC-256, pushed the cryptographic FDA cybersecurityOTA; implant verifies signature signature checks out.” guidance compliant. & installs to inactive slot.Fallback A / B image bootloader (§ 7.4.5) — Guarantees no may revert if new model fails a bricking from bad Al.10-s heartbeat.7.6.3 Nightly Self-Calibration & Cohort Analytics (F-3)Component Technical implementation Plain-language view Why it matters Rest epoch At 02:00 local time the implant “While you sleep, the Removes need for may sample a 5-min quiescent implant re-zeros its clinic-based ENG / IMU window; envelope microphone.” recalibration. noise floor crest & CV histogramcan update ENG thresholds.Trend stats Aorest, ACV, ENG entropy trend — Minimal data volume may compress to 8 bytes / week. keeps battery drain negligible.Fleet Cloud can plot cohort histograms “Researchers see Creates data moat and analytics for R& D and generate mean anonymized bigsupports regulatory progression curves by diagnosis. picture trends across submissions with real- all users.” world evidence.7.6.4 Privacy & Cybersecurity SafeguardsAttorney Docket No. 61435-708601Safeguard DetailData at rest Phone & cloud storage may use AES-256-GCM with per-device keys. Data in BLE packets can be encrypted; server comms via TLS 1.3.motionAccess control OAuth 2.0 tokens tied to clinician portal; patient may revoke at any time. Anonymisatio Device ID salted + hashed before cloud aggregation.nAudit trail Firmware logs last 32 OTA updates and clinician accesses, retrievable viaBLE.

[0816] Plain language: “Your nerve data are encrypted end-to-end; if Wi-Fi is down, the implant still works on yesterday’s model.”7.6.5 Compute & Power Budget SummaryBlock Power (uW. avg) Duty factor Artefact suppression (B-2 + B- 55 + 12 100 % during therapy 5)Tiny-CNN inference (F-l) 25 Runs only if ENG control activeKalman fusion (C-3) 5 100 % during therapy BLE telemetry (idle) 14 ContinuousBLE burst (uplink) 150 < 1 s every 10 min Inductive recharge overhead 0 (battery-free) - 10 mW User-scheduled(hybrid)

[0817] Total algorithmic power may remain < 120 pW during active sessions, leaving > 90 % of the 10 mW power-budget margin for stimulation pulses and radio bursts.8. Example Data & Modeling

[0818] (All results derive from physics-based simulations or benchtop emulation; no human or animal data are required to support enablement. Numerical values may vary with electrode geometry, tissue parameters, or firmware choices and therefore are not limiting.)8.1 Artefact-Decay Monte-Carlo StudyAttorney Docket No. 61435-708601Parameter Range Purpose Result snapshot* Impact sweptTissue 0.04- Models scar vs 99 % of runs may drop Confirms adaptive resistivity pt 0.12 S healthy below 5 pV artefact in < blanking target nr10.80 msFibre depth 0-1.5 Captures pad Artefact can vary ±4 dB Template d mm rotation subtraction compensates Pad rotation 0-15° Worst-case Negligible effect on r Robust to surgical e orientation tolerance

[0819] Median curve shown in FIG 4; shaded 95 % band illustrates variability.8.2 Closed-Loop Latency Simulation

[0820] 50 Hz stimulation, dual-cuff AL = 35 mm.

[0821] Artefact cleared at 0.74 ms (orthodromic) / 0.89 ms (antidromic).

[0822] CNN inference + proportional update executed in 24 ps.

[0823] Total loop time: 0.78- 0.94 ms => < 1 cycle at 50 Hz — > limb may receive param update before next contraction onset.8.3 Template-Subtraction Power BenchMCU clock Compute time Avg current DSP power 64 MHz 18.6 ps 17 pA @ 3.3 V ~ 56 pW 32 MHz 29.4 ps 9 pA ~ 30 pW

[0824] Even at the higher clock, DSP remains < 1 % of a 10 mW inductive-link budget.8.4 Wavelet Denoising Efficacy

[0825] Added 3-5 dB SNR across 10000 Monte-Carlo runs when artefact residual > 3 pV.

[0826] False-positive spike rate reduced from 6.2 % to < 1 %.8.5 Tiny-CNN Accuracy & PowerDataset Accuracy (motor vs sensory) Energy / inference Synthetic ENG (n = le6 spikes) 96.8 % 14.1 pj In-vivo feline (public set) 94.5 % 14.3 pj8.6 Fatigue Duty-Cycle Optimizer Emulation

[0827] Conduction-velocity drift Atcv artificially increased by +12 ps over 5 min.Attorney Docket No. 61435-708601

[0828] Duty-cycle auto-reduced from 100 % — 60 % in four 10 % steps, then re-increased as Atcv recovered, mirroring published fatigue curves.8.7 Spasticity Reflex-Breaker Event ModelLatency component Mean timeAfferent volley detection 12 msAnti-burst scheduling 4 msnerve conduction distal — > spinal 18-22 ms

[0829] Total reflex-abort latency: ~ 38 ms, comfortably < 50 ms clinical threshold for functional benefit.8.8 Inductive-Link Thermal Model

[0830] 6.78 MHz coil, 200 mW transmitter, 10 mm coupling gap.

[0831] Tissue AT peaked at 0.17 °C after 30 min — well below IEC 60601-1 limit (1 °C).

[0832] Collectively, these simulations and benchtop emissions demonstrate that the disclosed hardware and algorithms can meet artefact-rejection, latency, power, and safety targets while providing therapeutic functions unavailable in prior art.1. System Architecture1.1 Cuff Electrode

[0833] A spiral, helical, split-ring, or hinged cuff having between about two and sixteen conductive pads may circumferentially or partially circumferentially embrace a target nerve. Each pad may be dynamically switched, on a per-pulse basis, to a constant-current driver or a low-noise ENG front-end, thereby permitting dual-mode stimulation and recording using a single hardware interface. Ring contacts provide circumferential coverage, while interleaved dot contacts enable fascicle-specific steering and discrete afferent / efferent recording. A shape-memory element may impart a self-coiling bias for atraumatic deployment1.2 Stimulation-and-Sensing Electronics

[0834] The bidirectional IC may integrate high-compliance drivers (±15 V),chopper-stabilized amplifiers, a Wilkinson or SAR ADC, artefact-blanking logic, and an ECC-secure BLE radio. Pulse amplitudes may range from micro-amps to tens of milliamps.1,2a Multi-Source Driver Array

[0835] Each conductive pad may be driven by an independent digitally controlled current source (e.g., DAC-mirrored or H-bridge). Current vectors Ii-Inare updated on a per-pulse basis with < 10 ps latency, allowing 3-D field shaping. In multi-cuff networks (FIG. 11),Attorney Docket No. 61435-708601driver arrays may synchronize to generate composite joint torques that follow a kinematic command trajectory.1,2b Time-Multiplexed Dual -Mode Operation

[0836] A scheduler may allocate > 90 % of each cycle to stimulation and < 10 % to ENG sampling. Stimulation artefact decays by > 40 dB before acquisition begins (timeline, FIG. 9), preserving ENG fidelity without additional blanking capacitors.1.3 Power & Telemetry

[0837] The implant may derive energy from an internal Li-ion pouch cell, an inductive link, or an ultrasonic harvester. Firmware updates may be authenticated via elliptic-curve P-256 signatures; unverified packets are rejected by a secure bootloader.1.4 Sensor Suite

[0838] External or implanted sensors may include IMUs, EMG electrodes, optical trackers, or biochemical analyte sensors. In certain embodiments, the cuff itself functions as an ENG sensor, streaming action-potential data that inform closed-loop algorithms.1.5 Sterilization & Packaging

[0839] Prior to implantation, the cuff assembly may be terminally sterilized byethyl ene-oxi de, e-beam, or VHP and sealed in a Tyvek® / PETG blister maintaining SAL < 106per ISO 11137-1.2. Therapeutic Methods (Exemplary)

[0840] The disclosed platform may implement one or more of the following non-limiting therapeutic paradigms, each of which may be executed by a single cuffm x cm with a single nerve or by two or more cuffs interfacing with distinct or contiguous nerves. Where multiple cuffs are employed, therapy parameters may be coordinated in time or phase to achieve synergistic recruitment or differential blockade.2.1 Regenerative Stimulation Protocol (RSP)

[0841] Immediately before, during, or within about 24 h after surgical nerve repair, each implanted cuff may deliver a burst of charge-balanced biphasic pulses having:

[0842] Frequency: about 10-50 Hz;

[0843] Phase width: about 50-300 ps per phase;

[0844] Amplitude: up to about 5 mA when driven from a single current source, or up to about 20 mA aggregate when a multi-source driver energizes separate electrode pairs concurrently.Attorney Docket No. 61435-708601

[0845] When two cuffs span proximal and distal segments of the repair, proximal stimulation may be configured to promote orthodromic axonal outgrowth while distal ENG recording verifies conduction recovery.2.2 Tremor Cancellation Protocol (TCP)A) A kinematic sensor (e.g., wrist-mounted IMU) may detect tremor frequency / T in real time. A controller may compute stimulation trains matched to / T and delivered 180 ± 20° out of phase. For multi-cuff embodiments:

[0846] Single-nerve, multi-ring steering: Dot electrodes at the 3 o’clock and 9 o’clock positions may inject equal-and-opposite currents to steer the resultant field toward dorsal afferent fascicles, thereby minimizing unwanted motor recruitment.

[0847] Multiple nerves: A first cuff on the radial nerve may suppress pronation torque while a second cuff on the median nerve may suppress flexion torque; each cuff may be governed by an individually phase-locked controller.B) ENG-Driven Tremor Suppression (Intra-Nerve Sensing Variant)

[0848] In an alternative embodiment, the same cuff electrode that delivers therapeutic stimulation may also acquire electroneurographic (ENG) signals indicative of tremor-related neural activity on an efferent or afferent fascicle of the affected limb. During a sensing interval, one or more dot contacts may be configured as differential recording pairs referenced to an adjacent ring contact, thereby capturing compound-action-potential bursts or low-frequency envelopes whose amplitude and periodicity correlate with the mechanical tremor. A firmware-implemented detector may extract the dominant tremor frequency, phase, or aberrant bursting pattern directly from the ENG stream — eliminating reliance on external wearable sensors. The controller may then (i) deliver phase-cancelling...

Claims

Attorney Docket No. 61435-708601CLAIMS WHAT IS CLAIMED IS:

1. A neuromodulation system for restoring or enhancing motor performance in a subject, comprising:an implantable unit configured for placement adjacent to a peripheral nerve;a sensor module configured to acquire at least one sensor signal reflective of the subject’s volitional motor intent; anda processing unit configured to interpret the at least one sensor signal and generate at least one stimulation command for delivery to the subject via the cuff electrode, wherein the processing unit continuously modulates at least one stimulation parameter based on real-time feedback.

2. The system of claim 1, wherein the implantable unit is fabricated from atleast one biocompatible material.

3. The system of claim 1, wherein the implantable unit comprises at least one cuff electrode arranged to envelop the peripheral nerve.

4. The system of claim 3, wherein the at least one cuff electrode comprises one or more integrated sensing channels configured to record electroneurographic (ENG) or compound nerve action potential (CNAP) signals to provide high-fidelity acquisition during stimulation, monitoring, or both.

5. The system of claim 3, wherein the at least one cuff electrode is integrated with multi-modal sensing channels and stimulation circuitry for delivering electrical pulses.

6. The system of claim 1, wherein the sensor module is at least partially implantable.

7. The system of claim 1, wherein the sensor module is adapted to detect discrete activation (“on”) and deactivation (“off) signals.

8. The system of claim 7, wherein the activation signal is detected from a first sensor modality and the deactivation signal is detected from a second sensor modality different from the first.

9. The system of claim 8, wherein the first sensor modality comprises electroneurography (ENG) and the second sensor signal comprises electromyography (EMG).

10. The system of claim 1, wherein the sensor module is adapted to detect at least one graded input indicative of varying levels of motor intent.Attorney Docket No. 61435-70860111. The system of claim 1, wherein the sensor module is configured to receive input from at least one sensor selected from the group consisting of an implanted electromyography (EMG) sensor; a wearable EMG sensor; an implanted or surface electroencephalography (EEG) sensor; an electrocorticography (ECoG) grid; a depth electrode; an integrated ENG sensors within a cuff electrode; a brain-machine interface; a nerve-machine interface^ wearable sensor; an inertial measurement unit (IMU); a gyroscope; an accelerometer; a force sensor; and a torque sensor.

12. The system of claim 11, wherein the at least one sensor provides high-resolution data for evaluating motor intent and performance.

13. The system of claim 1, wherein the at least one sensor signal reflective of the subject’s volitional motor intent comprises an electroneurography (ENG) signal.

14. The system of claim 1, wherein the at least one sensor signal reflective of the subject’s volitional motor intent comprises an electromyography (EMG) signal.

15. The system of claim 1, wherein the at least one sensor signal reflective of the subject’s volitional motor intent comprises at least two sensor signals, wherein the at least two sensor signals are of different measurement modalities.

16. The system of claim 15, wherein the different measurement modalities are selected from the group consisting of an implanted electromyography (EMG) sensor; a wearable EMG sensor; an implanted or surface electroencephalography (EEG) sensor; an electrocorticography (ECoG) grid; a depth electrode; an integrated ENG sensors within a cuff electrode; a brain-machine interface; a nerve-machine interface; a wearable sensor; an inertial measurement unit (IMU); a gyroscope; an accelerometer; a force sensor; and a torque sensor.

17. The system of claim 1, wherein the processing unit is programmed to execute an algorithm employing one or more of threshold detection, dual-activation logic, timewindow analysis, or machine learning classification or other artificial intelligence techniques to distinguish genuine neural or myoelectric signals from artifacts.

18. The system of claim 1, wherein the processing unit is further programed to incorporating a self-calibration routine adapting to at least one patient-specific signal profile over time.

19. The system of claim 1, wherein the processing unit is operable in a closed loop manner.Attorney Docket No. 61435-70860120. The system of claim 1, wherein the at least one continuously modulated stimulation parameter comprises one or more of stimulation intensity, pulse duration, or waveform profile.

21. The system of claim 1, wherein the implantable unit is configured for long-term implantation.

22. The system of claim 1, wherein the implantable unit comprises a wireless communication module for remote monitoring, remote programming, or both.

23. The system of claim 1, wherein the implantable unit further comprise a power module configured for transcutaneous energy transfer, as a fully implantable unit with an internal rechargeable battery, or both.

24. The system of claim 23, wherein the power module incorporates energy management circuitry to optimize power usage, extend operational life, or both.

25. The system of claim 1, wherein the processing unit is configured to generate at least one stimulation command specifying at least one stimulation parameter.

26. The system of claim 25, wherein the at least one stimulation parameter comprises one or more of pulse amplitude, frequency, pulse duration, or a ramp-up and ramp-down interval.

27. The system of claim 25, wherein the at least one stimulation command is configured to achieve smooth and graded motor activation.

28. The system of claim 25, wherein the at least one stimulation command is configured to permit real-time customization of stimulation waveforms based on patient feedback.

29. The system of claim 25, wherein the at least one stimulation parameter is preprogrammed or dynamically adjusted during a stimulation session.

30. The system of claim 1, wherein the processing unit is configured to execute at least one adaptive algorithm to dynamically adjust at least one stimulation parameter based on real-time sensor data to one or more of mitigate fatigue, modulate tremor, reduce spasticity, or optimize functional recovery.

31. The system of claim 30, wherein the at least one adaptive algorithm is based on at least one subject -specific trend.

32. The system of claim 1, wherein the system is configured operate in an open-loop mode to deliver at least one pre-programmed stimulation pattern for a therapeutic application.Attorney Docket No. 61435-70860133. The system of claim 32, wherein the pre-programmed stimulation pattern is configured for one or more of prevention of disuse atrophy, modulation of spasticity, or promotion of peripheral nerve regeneration.

34. The system of claim 32, wherein the system is configured to transition to a closed-loop operation in response to a change in a condition of the subject.

35. The system of claim 1, further comprising a communication module configured for data exchange with an external device.

36. The system of claim 35, wherein the external device comprises one or more of a wearable pulse generator for transcutaneous power delivery, a remote control of stimulation parameters, or a secure bidirectional data transmission for system updates, performance monitoring, or both.

37. The system of claim 1, wherein one or more of the sensor module or processing unit comprises a redundant architecture so that failure or degradation of a sensor modality does not compromise closed-loop functionality.

38. The system of claim 1, further comprising a telehealth interface that is configured to facilitate one or more of remote programming, real-time monitoring, secure data logging, or over-the-air (OTA) firmware and software updates, thereby enabling one or more of continuous therapy optimization, integration with an electronic medical record (EMR) system, or compliance with a regulatory standard.

39. The system of claim 1, wherein the processing unit is further configured to decode cortical signals from implanted or wearable EEG or ECoG sensors to determine volitional motor intent and generate the at least one stimulation command, thereby integrating central and peripheral neural inputs for neuromodulation.

40. The system of claim 1, wherein the processing unit further include a neural decoding engine configured to one or more of differentiate volitional neural or myoelectric activity from artifacts, optimize stimulation patterns based on a subject-specific neural signatures, or employ machine learning or other another artificial intelligence technique to adapt to at least one variation in neural activity of the subject.

41. The system of claim 1, wherein the implantable unit further comprises an integrated diagnostic indicator configured to monitor at least one operational parameter.

42. The system of claim 41, wherein the at least one operational parameter comprises a one or more of battery level, electrode impedance, or signal quality.Attorney Docket No. 61435-70860143. The system of claim 1, wherein the implantable unit further comprises a communication module configured to support integration with a cloud-based management system.

44. The system of claim 1, wherein the processing unit incorporates an adaptive reinforcement learning algorithm configured to refine a stimulation protocol over multiple therapy sessions in response to cumulative patient response data.

45. The system of claim 1, wherein the processing unit is further configured with a field steering capability that selectively distributes stimulation current among multiple channels to target specific nerve fascicle sub-populations for refined motor control while minimizing off-target effects.

46. The system of claim 1, wherein the processing unit includes an emergency fallback mechanism that stores a predetermined set of minimal stimulation parameters in local memory and automatically revert to these parameters upon loss of communication with the external device.

47. The system of claim 1, wherein the system is configured as a universal peripheral nerve interface platform capable of both stimulating and recording signals from motor, sensory, or mixed nerves.

48. The system of claim 1, wherein the apparatus is further be configured for one or more of sensory restoration, disease progression monitoring, or diagnostic evaluations.

49. The system of claim 1, further comprising one or more of:(a) one or more integrated ENG recording channels that configured to capture peripheral nerve signals indicative of sensory input;(b) a sensory cortical stimulation module that configured to deliver targeted electrical pulses to predetermined regions of a sensory cortex based on real-time ENG signal analysis to facilitate sensory function restoration or augmentation;(c) a closed-loop control module that configured to integrate ENG signal analysis with cortical stimulation patterns to adaptively modulate stimulation parameters in real time; or(d) a diagnostic circuitry and data processing algorithm configured to continuously monitor neural signal trends to enable tracking of neurodegenerative disease progression and generation of diagnostic reports for clinical evaluation, wherein the apparatus may operate in either open-loop or closed-loop modes to suit varying therapeutic requirements.

50. The system of claim 1, wherein the processing unit further incorporates features including one or more of:Attorney Docket No. 61435-708601an adaptive control strategy for real-time optimization of stimulation parameters; an auto-calibration routine to periodically recalibrate at least one sensor baseline or stimulation threshold; orintegration with a remote monitoring system for continuous therapy adjustment.

51. The system of claim 1, wherein the apparatus is configured to treat any muscle group in the body by employing either a single cuff electrode or a combination of cuff electrodes, thereby enabling the restoration of both simple and complex motor movements through coordinated multi-muscle activation.

52. The system of claim 1, wherein one or more of sub-components are one or more of independently configurable, replaceable, or upgradable, thereby accommodating different sensor modalities or therapeutic indications.

53. A method for customizing neuromodulation therapy in a patient, comprising:(a) acquiring one or more of ENG or EMG sensor data via the sensor module of the system of claim 1;(b)processing the acquired data using an adaptive algorithm to determine subjectspecific motor intent or a performance characteristic;(c) adjusting at least one stimulation parameter in real time based on the processed sensor data to restore or enhance natural motor function; and(d) continuously refining a stimulation protocol over successive therapy sessions based on cumulative subject response data.

54. The system of claim 1, wherein the system is adaptable to treat a wide range of pathologies.

55. A control algorithm for triggering electrical stimulation in a neuromodulation system, comprising:(a) receiving a sensor signal from at least one sensor selected from the group consisting of a trapezius EMG sensor, a biceps EMG sensor, a triceps EMG sensor, and an ENG sensor;(b)comparing the sensor signal to a predetermined threshold to determine an activation condition; and(c) initiating electrical stimulation when the sensor signal exceeds the threshold and terminating stimulation when the sensor signal falls below a predetermined deactivation threshold.Attorney Docket No. 61435-70860156. The control algorithm of claim 55, wherein the algorithm operates in one or more modes selected from the group consisting of:(i) a simple on / off mode in which stimulation is activated only when the sensor signal crosses the threshold;(ii) a sustained mode in which stimulation remain actives until the sensor signal falls below the threshold; and(iii) a graded mode wherein the magnitude of the stimulation output is linearly scaled relative to the sensor signal amplitude above the threshold.

57. The control algorithm of claim 55, wherein the at least one sensor is configured to operate in one or more configurations, including:(a) employing an ENG sensor to initiate stimulation and a triceps EMG sensor to terminate stimulation;(b)implementing a dual-sensor scheme wherein a biceps EMG sensor activates stimulation and a triceps EMG sensor deactivates stimulation; or(c) applying signal processing techniques to use an ENG sensor for stimulation initiation while employing blanking windows or other algorithms to detect residual ENG activity to trigger stimulation termination and to dynamically modulate stimulation parameters.

58. A neuromodulation system for sensor-based device activation and dynamic adjustment of stimulation parameters, comprising:an implantable unit configured for placement adjacent to a peripheral nerve, the unit comprising a cuff electrode and integrated sensor interfaces for both delivering electrical stimulation and acquiring neural or muscular signals;a sensor module configured to acquire signals from one or more sensors selected from the group consisting of implanted electromyography (EMG) sensors, wearable EMG sensors, electroneurography (ENG) sensors, inertial measurement units (IMU), and force sensors; anda processing unit operable in a closed-loop manner, configured to:(i) detect an activation signal when sensor outputs exceed a predetermined threshold, thereby transitioning the system from an inactive to an active state;(ii) continuously monitor sensor signals during activation and compare them to an expected movement profile;(iii) dynamically adjust stimulation parameters in real time when sensor data indicate deviations from the expected movement; andAttorney Docket No. 61435-708601(iv) detect a deactivation signal when sensor outputs fall below a predetermined threshold, thereby terminating stimulation delivery.

59. A closed-loop peripheral nerve stimulation system for mitigating motor tremors, comprising:a biocompatible cuff electrode configured for implantation around a peripheral motor nerve;at least one sensor selected from the group consisting of electromyography (EMG) sensors, inertial measurement unit (IMU) sensors, or combinations thereof, configured to detect motion or muscle activity indicative of tremor; anda processing unit operatively connected to the cuff electrode and the at least one sensor, configured to analyze at least one incoming signal from the at least one sensor and dynamically adjust electrical stimulation to reduce tremor amplitude while preserving normal voluntary movement.

60. The system of claim 59, wherein the cuff electrode comprises multiple individually addressable contacts that are configured to enable selective stimulation of specific nerve fascicles.

61. The system of claim 59, wherein the at least one sensor is configured to comprise either:(a) one or more implantable EMG sensors positioned adjacent to or on muscles of interest or on the nerve trunk; or(b)one or more wearable IMU sensors that is configured to measure acceleration and angular velocity, or an implantable IMU to provide continuous motion data.

62. The system of claim 59, wherein the processing unit is configured to:(a) employ filtering and feature extraction algorithms to identify tremor frequencies in real time;(b)automatically adjust a stimulation parameter based on tremor severity; and(c)log usage statistics and sensor data to non-volatile memory for subsequent clinical analysis.

63. The system of claim 59, further comprising a power module that is configured to:(a) support inductive or wireless recharging;(b)utilize on-demand stimulation to conserve battery life; or(c) operate in a safety fallback mode in the event that sensor feedback can be interrupted or deemed abnormal.

64. A method for mitigating motor tremors, comprising:Attorney Docket No. 61435-708601(a) surgically implanting a cuff electrode around at least one peripheral nerve of a subject responsible for controlling tremor-affected muscle groups;(b) positioning one or more sensors into the subject to monitor muscle activity or limb motion of the subject;(c) tuning an adaptive control algorithm to detect tremor patterns based on an acquired sensor signal from the subject; and(d) delivering at least one electrical stimulation pulse to the nerve in real time upon detection of tremor events, thereby reducing tremor amplitude while preserving normal voluntary movement.

65. The method of claim 64, further comprising discriminating pathological tremors from normal movement by correlating sensor inputs and, wherein the processing unit updates stimulation thresholds over time based on one or more of subject-specific responses, stored data logs, or detected postural changes.

66. An enhanced diagnostic neuromodulation system for tracking neurological disease progression and providing prognostic evaluation in a patient, comprising:(a) the system of claim 1;(b) at least one integrated diagnostic sensor and circuitry configured to continuously monitor operational parameters; and(c) a data analysis module configured to process the continuously monitored data using machine learning algorithms and pattern recognition techniques to identify trends associated with disease progression and to generate prognostic evaluations for clinical decision-making.

67. The system of claim 66, wherein the diagnostic data is transmitted via a communication module to a remote cloud-based management system for continuous monitoring, evaluation, or both.

68. A method for restoring motor function in a patient comprising:(a) providing a biocompatible implantable unit proximate to a peripheral nerve, wherein the implantable unit comprises at least one cuff electrode wrapped around the peripheral nerve and integrated with multi-modal recording channels and stimulation circuitry for generating electrical stimulation pulses;(b) receiving, via a sensor module, one or more input signals indicative of volitional motor intent from at least one sensor;(c) processing, via a control module, the sensor input and generating stimulation commands for delivery via the cuff electrode.Attorney Docket No. 61435-70860169. The method of claim 68, wherein the one or more input signals comprise at least one graded input indicative of a level of motor intent.

70. The method of claim 68, wherein the one or more input signals are received from an electromyography sensor.

71. The method of claim 68, wherein the one or more input signals are received from an electroneurography sensor.

72. The method of claim 68, further comprising operating the control module in a closed-loop mode to dynamically adjust stimulation parameters based on sensor feedback.

73. The method of claim 68, wherein the generating the stimulation commands in (c) further comprises detecting an activation signal based at least in part on sensor outputs compared to a predetermined threshold.

74. The method of claim 73, wherein the generating the stimulation commands in (c) further comprises monitoring sensor signals during activation.

75. The method of claim 74, further comprising comparing the monitored sensor signals during activation to an expected movement profile.

76. The method of claim 75, further comprising dynamically adjusting a stimulation parameter when the monitored sensor signals deviate from the expected movement profile.

77. The method of claim 76, further comprising detecting a deactivation signal based at least in part on a sensor signal falling below a predetermined threshold.

Images