Low-channel EEG amplification circuit and brain-computer interface device for behind-the-ear EEG signals

Abstract

The present application relates to the technical field of wireless communication networks, and provides a low-channel EEG amplification circuit and a brain-computer interface device for behind-the-ear EEG signals, comprising: a dry electrode sensing module for acquiring EEG signals; a pre-active amplifier module comprising an operational amplifier circuit with high input impedance; an anti-aliasing filter module composed of a multi-order passive filter network and an active filter circuit in cascade, connected to the output end of the pre-active amplifier module, for filtering out high-frequency interference signals; a right leg drive feedback module comprising a programmable impedance adjustment circuit and a current limiting protection circuit, forming a closed-loop feedback loop with the pre-active amplifier module, and the output end of the right leg drive feedback module forming a current loop with the human body through a protective resistor; an analog-to-digital conversion module using an integrated low-power analog front-end chip; a wireless transmission module; and a power management module providing independent power supply to the analog front-end chip, the operational amplifier circuit, and the wireless transmission module.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication networks, and in particular relates to a low-channel EEG amplification circuit and a brain-computer interface device for behind-the-ear EEG signals. Background Art

[0002] The human body's EEG signal is a very weak electrophysiological signal at the microvolt level and is accompanied by a lot of noise. To improve the signal-to-noise ratio of the collected EEG signal, the current common practice in medical or scientific EEG equipment is to use a multi-channel EEG amplifier and collector covering the entire brain to expand the collection area, and to use wet electrodes such as conductive paste to enhance the contact between the electrode and the skin and reduce the contact impedance.

[0003] Although multi-channel acquisition can improve the signal-to-noise ratio of the collected signal to a certain extent, it also brings about a sharp increase in equipment costs and increases the size of the equipment, which is not conducive to miniaturization and integration, and is difficult to promote and apply on wearable devices.

[0004] Although the whole-brain EEG signal acquisition device using wet electrodes can ensure good contact between the electrodes and the scalp, reduce contact impedance, obtain high-quality, high-fidelity EEG signals, effectively reduce signal attenuation and the mixing of external interference, and can clearly capture the characteristics of different frequency bands, it has cumbersome preparation work, cannot be used for long periods of time, and has high professional requirements for operators, making it unsuitable for daily EEG signal monitoring and acquisition. Therefore, this type of multi-channel EEG sensor device using wet electrodes is not suitable for consumer wearable electronic products. Summary of the Invention

[0005] The present application provides a low-channel EEG amplification circuit and brain-computer interface device for behind-the-ear EEG signals, aiming to solve the existing whole-brain EEG signal acquisition device using wet electrodes. Although it can ensure good contact between the electrodes and the scalp, reduce contact impedance, obtain high-quality, high-fidelity EEG signals, effectively reduce signal attenuation and the mixing of external interference, and can clearly capture the characteristics of different frequency bands, it has cumbersome preparation work, cannot be used for a long time, and has high professional requirements for operators, resulting in it being unable to be used for daily EEG signal monitoring and acquisition. Therefore, this type of multi-channel EEG sensor device using wet electrodes is not suitable for consumer wearable electronic products.

[0006] In a first aspect, an embodiment of the present application provides a low-channel EEG amplification circuit for behind-the-ear EEG signals, comprising:

[0007] The dry electrode sensing module consists of two differential electrodes and is placed in the hairless area behind the ear to collect EEG signals.

[0008] A preamplifier module includes a high-input impedance operational amplifier circuit, which is directly coupled to the dry electrode sensor module to form a differential signal amplification structure with a common-mode rejection ratio of at least 120 dB; the dry electrode sensor module and the preamplifier module are electrically connected via an ESD protection circuit;

[0009] An anti-aliasing filter module, composed of a multi-order passive filter network and an active filter circuit in cascade, is connected to the output end of the pre-active amplifier module to filter out high-frequency interference signals; the output impedance of the anti-aliasing filter module is less than 500Ω;

[0010] The right leg drive feedback module includes a programmable impedance adjustment circuit and a current limiting protection circuit, and forms a closed-loop feedback loop with the preamplifier module. The output end of the right leg drive feedback module forms a current loop with the human body through a protective resistor;

[0011] The analog-to-digital conversion module uses an integrated low-power analog front-end chip, including a dual-channel differential input interface, a programmable gain amplifier, and a 24-bit high-precision ADC converter;

[0012] The wireless transmission module integrates a low-power Bluetooth communication chip and is connected to the analog-to-digital conversion module through a digital interface;

[0013] The power management module includes a multi-stage low-noise LDO power supply circuit, which provides independent power supply to the analog front-end chip, operational amplifier circuit and wireless transmission module respectively.

[0014] In some embodiments, the dual-channel differential input interface is respectively connected to the two output ends of the anti-aliasing filtering module; wherein, the negative input end of the dual-channel differential input interface is commonly connected as a common reference node, and the positive input end receives the EEG signals from the two differential electrodes respectively; the analog front-end chip has a built-in programmable gain amplifier and a 24-bit high-precision ADC converter, and is connected to the main control chip through a digital interface to transmit the converted digital signal.

[0015] Exemplarily, the analog front-end chip integrates a continuous disconnection detection circuit and a test signal generator, and the corresponding common mode rejection ratio is not less than 120dB and the input reference noise is less than 5μVpp.

[0016] In some embodiments, the operational amplifier circuit of the pre-active amplifier module adopts a non-inverting input amplifier topology, and the corresponding input end is directly coupled to the dry electrode sensor module through a resistor network to form an input impedance greater than 10GΩ.

[0017] In some embodiments, the passive filtering network of the anti-aliasing filtering module is composed of at least three resistor-capacitor devices to form a multi-order low-pass filter, and the corresponding cutoff frequency is set to 50Hz-150Hz to match the EEG signal frequency band; the active filtering circuit is composed of an operational amplifier and resistor-capacitor devices to form a second-order or higher low-pass filtering topology.

[0018] In some embodiments, the programmable impedance adjustment circuit of the right leg drive feedback module includes a parallel resistor and capacitor network, which dynamically controls the feedback bandwidth by adjusting the RC time constant; the current limiting protection circuit is connected in series between the right leg drive output terminal and the human body reference electrode, and forms a bidirectional current limiting structure with the ESD protection diode.

[0019] In some embodiments, the low-noise LDO power supply circuit of the power management module independently supplies power to the analog power pin and the digital power pin of the analog front-end chip, and the power ripple suppression ratio of each corresponding LDO chip is higher than 80dB and an enable control pin is integrated to achieve time-sharing power supply.

[0020] In some embodiments, the Bluetooth communication chip of the wireless transmission module integrates a DC-DC voltage converter, its RF transmission peak current is lower than 10mA and its receiving sensitivity is better than -90dBm, and signal transmission and reception are achieved through an onboard inverted F-type antenna and an impedance matching circuit.

[0021] In some embodiments, the EEG amplification circuit is module-integrated through a four-layer PCB board; wherein, the corresponding analog signal lines and digital signal lines adopt a layered isolation layout, and the dry electrode contact area adopts an immersion gold process to reduce contact impedance.

[0022] In a second aspect, the present application provides a brain-computer interface device, including a low-channel EEG amplification circuit for behind-the-ear EEG signals provided by any embodiment of the present application.

[0023] This low-channel EEG amplification circuit for post-auricular EEG signals uses two differential electrodes placed in the hairless area behind the ear to directly acquire EEG signals. This dry electrode design eliminates the need for auxiliary materials such as conductive adhesive, eliminating the tedious preparation required for wet electrodes. The preamplifier module, comprising a high-input impedance operational amplifier circuit, directly couples with the dry electrodes to form a differential signal amplification structure with a common-mode rejection ratio (CMRR) ≥120dB, effectively suppressing common-mode interference (such as ambient noise and power frequency interference). The dry electrodes and amplifier module are connected via an ESD protection circuit, enhancing the circuit's anti-static capability and protecting sensitive front-end components. The anti-aliasing filter module, consisting of a cascaded multi-stage passive filter network and active filter circuit, filters out high-frequency interference signals (such as electromagnetic noise and high-frequency harmonics), preventing aliasing distortion during signal sampling. The output impedance is designed to be less than 500Ω, matching the input requirements of the subsequent analog-to-digital conversion module and ensuring stable signal transmission. The right leg drive feedback module integrates a programmable impedance adjustment circuit and current limiting protection circuit, forming a closed-loop feedback loop with the preamplifier module to actively suppress common-mode noise (such as surface potential drift) from the human body. The output terminal forms a current loop with the human body through a protective resistor, reducing common-mode interference while ensuring user safety and preventing excessive current injection. The analog-to-digital conversion module utilizes an integrated low-power analog front-end chip. It features a dual-channel differential input interface that matches the front-end differential signal output; a programmable gain amplifier (PGA) that supports dynamic signal gain adjustment to accommodate EEG signals of varying amplitudes; and a 24-bit high-precision ADC converter that ensures high-resolution digitization of EEG signals while preserving weak signal characteristics. The wireless transmission module integrates a low-power Bluetooth communication chip and connects to the analog-to-digital conversion module via a digital interface, enabling wireless transmission of EEG signals and supporting long-lasting wearable device applications. The power management module utilizes a multi-stage low-noise LDO (low-dropout linear regulator) power supply circuit to provide independent, isolated power supplies for the analog front-end chip, operational amplifier, and wireless module, reducing power supply noise coupling between modules and improving overall system noise performance.

[0024] The provided circuit has the following beneficial effects:

[0025] 1. Solve the pain point of wet electrodes and adapt to daily wear:

[0026] Dry electrode design: No conductive gel is required, avoiding tedious electrode preparation, supporting quick donning and long-term use, reducing the need for professional operation, and is suitable for consumer wearable devices (such as smart headphones, headsets, etc.).

[0027] Acquisition in the hairless area behind the ear: Avoiding interference from scalp hair, ensuring stable electrode contact, reducing contact impedance problems caused by hair, and improving the convenience of signal acquisition.

[0028] 2. High anti-interference capability and signal quality:

[0029] High CMRR differential amplification (≥120dB) and multi-stage filtering (anti-aliasing filtering + right leg drive feedback): effectively suppress environmental noise, power frequency interference and human common-mode noise, ensuring high-fidelity amplification of EEG signals and clearly capturing characteristics of different frequency bands (such as α, β, and γ waves).

[0030] 24-bit high-precision ADC: preserves details of weak EEG signals, avoids quantization noise, and meets the high-precision requirements of physiological signal analysis.

[0031] 3. Low power consumption and integrated design:

[0032] Low-power Bluetooth and multi-level LDO power supply: Reduce system energy consumption, adapt to battery-powered wearable devices, and support long-term monitoring; independent power supply design reduces noise crosstalk between modules and improves stability.

[0033] Low-channel count and simplified structure: Different from traditional multi-channel whole-brain acquisition equipment, the low-channel design focuses on the specific area behind the ear, taking into account performance and portability, reducing hardware cost and complexity.

[0034] 4. Security and reliability:

[0035] ESD protection and current limiting protection: Improve circuit anti-static capabilities and human safety protection, preventing high-voltage static electricity or abnormal current from damaging users and equipment.

[0036] Programmable impedance adjustment: adapts to the physiological impedance differences of different individuals and ensures signal acquisition stability.

[0037] 5. Application scenario expansion:

[0038] By overcoming the limitations of wet electrodes, this circuit is suitable for daily EEG signal monitoring (such as fatigue detection, attention analysis, sleep monitoring, etc.), and promotes the implementation of brain-computer interface technology in the consumer electronics field, such as smart wearables and health management equipment.

[0039] In summary, this circuit, through dry electrode design, high anti-interference architecture, and low-power integrated solution, simplifies operation and improves signal quality while meeting the portability and safety requirements of wearable devices, filling the technical gap in consumer-grade EEG acquisition equipment.

[0040] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0042] Figure 1 This is a schematic block diagram of the structure of a low-channel EEG amplification circuit for behind-the-ear EEG signals provided by an embodiment of the present application;

[0043] Figure 2 This is a first circuit diagram of a low-channel EEG amplification circuit for behind-the-ear EEG signals provided by an embodiment of the present application;

[0044] Figure 3 This is a second circuit diagram of a low-channel EEG amplification circuit for behind-the-ear EEG signals provided in one embodiment of the present application;

[0045] Figure 4 This is a third circuit diagram of a low-channel EEG amplification circuit for behind-the-ear EEG signals provided in one embodiment of the present application;

[0046] Figure 5 This is a fourth circuit diagram of a low-channel EEG amplification circuit for behind-the-ear EEG signals provided in one embodiment of the present application;

[0047] Figure 6 This is a schematic block diagram of the structure of a brain-computer interface device provided in one embodiment of the present application;

[0048] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0050] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0051] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish between identical or similar items having substantially the same functions and effects. Those skilled in the art will understand that terms such as "first" and "second" do not limit the quantity or order of execution, and that terms such as "first" and "second" do not necessarily define differences.

[0052] It should be understood that the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0053] It will also be understood that the term "and / or" as used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0054] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features therein may be combined with each other.

[0055] The human body's EEG signal is a very weak electrophysiological signal at the microvolt level and is accompanied by a lot of noise. To improve the signal-to-noise ratio of the collected EEG signal, the current common practice in medical or scientific EEG equipment is to use a multi-channel EEG amplifier and collector covering the entire brain to expand the collection area, and to use wet electrodes such as conductive paste to enhance the contact between the electrode and the skin and reduce the contact impedance.

[0056] Although multi-channel acquisition can improve the signal-to-noise ratio of the collected signal to a certain extent, it also brings about a sharp increase in equipment costs and increases the size of the equipment, which is not conducive to miniaturization and integration, and is difficult to promote and apply on wearable devices.

[0057] Although the whole-brain EEG signal acquisition device using wet electrodes can ensure good contact between the electrodes and the scalp, reduce contact impedance, obtain high-quality, high-fidelity EEG signals, effectively reduce signal attenuation and the mixing of external interference, and can clearly capture the characteristics of different frequency bands, it has cumbersome preparation work, cannot be used for long periods of time, and has high professional requirements for operators, making it unsuitable for daily EEG signal monitoring and acquisition. Therefore, this type of multi-channel EEG sensor device using wet electrodes is not suitable for consumer wearable electronic products.

[0058] To resolve the above issues, please refer to Figure 1The present application provides a low-channel EEG amplification circuit for behind-the-ear EEG signals, comprising: a dry electrode sensing module 10, consisting of two differential electrodes, configured in a hairless area behind the human ear for EEG signal acquisition; a pre-active amplifier module 20, comprising an operational amplifier circuit with high input impedance, the operational amplifier circuit being directly coupled to the dry electrode sensing module to form a differential signal amplification structure with a common-mode rejection ratio of at least 120 dB; the dry electrode sensing module and the pre-active amplifier module being electrically connected via an ESD protection circuit; an anti-aliasing filter module 30, consisting of a multi-order passive filter network and an active filter circuit in cascade, connected to the output end of the pre-active amplifier module to filter out high-frequency interference signals; the anti-aliasing filter The output impedance of the module is lower than 500Ω; the right leg drive feedback module 40 includes a programmable impedance adjustment circuit and a current limiting protection circuit, which forms a closed-loop feedback loop with the front active amplifier module. The output end of the right leg drive feedback module forms a current loop with the human body through a protective resistor; the analog-to-digital conversion module 50 adopts an integrated low-power analog front-end chip, which includes a dual-channel differential input interface, a programmable gain amplifier and a 24-bit high-precision ADC converter; the wireless transmission module 60 integrates a low-power Bluetooth communication chip and is connected to the analog-to-digital conversion module through a digital interface; the power management module 70 includes a multi-stage low-noise LDO power supply circuit, which provides independent power supply to the analog front-end chip, operational amplifier circuit and wireless transmission module respectively.

[0059] The core design of the dry electrode sensing module: uses dual differential electrodes (reference electrode + measurement electrode), which are placed in the hairless area behind the human ear (such as the mastoid area). Technical advantages: Characteristics of the hairless area: The stratum corneum of the skin behind the ear is thin and the hair is sparse, which naturally reduces the electrode-skin contact impedance (no conductive paste is required), solving the problem of cumbersome preparation of wet electrodes and unsuitability for long-term monitoring. Differential acquisition: Directly obtain the EEG differential signal between the two electrodes, suppress common-mode noise (such as power frequency interference and motion artifacts), and cooperate with subsequent circuits to improve the signal-to-noise ratio. Electrode selection: Use dry electrode materials (such as Ag / AgCl coated electrodes, conductive polymer electrodes), with uniform surface impedance, high comfort for long-term wearing, and avoid skin irritation.

[0060] The preamplifier module includes the following core components: a high-input impedance operational amplifier (such as a field-effect transistor input op amp), which constructs a differential amplifier structure (typically an instrumentation amplifier circuit). Key indicators: Common-mode rejection ratio (CMRR) ≥ 120dB: Through high-precision resistor matching and differential circuit design, it effectively suppresses human common-mode noise (such as electromyography and environmental electromagnetic interference). Direct coupling: The electrode is directly connected to the op amp to reduce signal attenuation and adapt to the preamplification requirements of microvolt-level EEG signals (μV level). ESD protection: ESD protection circuits (such as TVS diodes and current-limiting resistors) are integrated at the electrode input to prevent electrostatic discharge from damaging sensitive front-end components and improve equipment reliability.

[0061] The anti-aliasing filter module includes the following: Structural design: A multi-order passive filter network (RC low-pass) is cascaded with an active filter circuit (such as a Sallen-Key filter) to form a high-order low-pass filter. High-frequency noise filtering: The cutoff frequency is set to ≥ 2 times the highest EEG signal frequency (such as 200 Hz, covering the 0.1-100 Hz EEG frequency band) to meet the Nyquist sampling theorem and avoid aliasing distortion in analog-to-digital conversion. Low output impedance: The output impedance is ≤ 500Ω, matching the input requirements of the subsequent analog-to-digital conversion module and reducing signal transmission loss.

[0062] Closed-loop feedback mechanism: Collects the common-mode voltage of the human body, amplifies it inversely, and then feeds it back to the human body (right leg or ground) through a protective resistor (e.g., 1MΩ), forming a negative feedback loop. Programmable impedance adjustment: An integrated digital potentiometer or variable resistor network dynamically adjusts the feedback gain to accommodate individual skin impedance differences. Current limiting protection: A series current-limiting resistor and overvoltage protection diode ensure that the loop current is ≤10μA, complying with safety standards (IEC 60601) and avoiding the risk of electrical stimulation. Core function: Actively suppresses common-mode noise (e.g., 50Hz power frequency interference), improving the system's common-mode rejection capability and further purifying the signal.

[0063] Integrated design: Uses dedicated analog front-end chips (such as TI ADS1292 and Maxim MAX30003), built-in dual-channel differential input, programmable gain amplifier (PGA, gain 1-2000 times) and 24-bit high-precision ADC.

[0064] High-precision sampling: 24-bit ADC resolution (equivalent noise level ≤ 1μV) adapts to the quantization requirements of microvolt-level EEG signals. Low-power mode: Supports sleep / wake-up functions and dynamically adjusts gain with the PGA, balancing accuracy and power consumption (typical power consumption ≤ 100μA).

[0065] Technical solution: Integrate a low-power Bluetooth (BLE 5.0+) chip (such as Nordic nRF52832, DialogDA14580) and communicate with the ADC module through the SPI / I2C interface.

[0066] Low-power transmission: Uses periodic sleep and adaptive connection intervals (e.g., 20ms-1s) to extend battery life (typical operating time ≥ 24 hours). Anti-interference mechanism: Built-in RF filters and protocol-layer error correction coding ensure wireless transmission stability (bit error rate ≤ 10^-6).

[0067] The multi-level power supply architecture includes:

[0068] Analog power supply: An independent low-dropout (LDO) (such as the ADM7150, noise ≤ 10 μVrms) powers the preamplifier and filtering circuits, isolating them from digital noise interference. Digital power supply: A low-voltage LDO (e.g., 3.3V / 1.8V) powers the ADC and Bluetooth module, supporting battery input (e.g., 3.7V lithium battery) and charging management (with an integrated charging chip, the TP4054). Noise isolation: The analog and digital grounds are connected at a single point using a 0Ω resistor or ferrite bead to reduce ground loop noise and ensure low-noise characteristics in the signal chain.

[0069] Electrode placement and wearing include:

[0070] Positioning: Two differential electrodes are fixed to the mastoid region behind the ear (symmetrically), close to hairless skin. A medical-grade silicone electrode cap or wearable headband is used to secure the electrodes for long-term wear and comfort. Electrode Material: Dry electrodes with Ag / AgCl coating are preferred, with a contact impedance of ≤50kΩ (wet electrodes are typically ≤10kΩ, but dry electrodes compensate for impedance differences by leveraging the skin characteristics behind the ear).

[0071] Circuit design and component selection include:

[0072] Preamplification: An AD8221 instrumentation amplifier (10^10Ω input impedance, 120dB CMRR) is used with a ±2.5V power supply and a gain of 100 (first-stage amplification). Anti-aliasing filtering: A first-order RC passive filter (10kΩ + 10nF, 1.6kHz cutoff frequency) is cascaded with a second-order active Sallen-Key filter (OP27 op amp, 200Hz cutoff frequency, -40dB / decade roll-off). Right leg drive: An OPA2333 op amp acquires the common-mode voltage with an inverting gain of -1, which is connected to a reference point on the body (e.g., ankle electrode) via a 1MΩ resistor. Analog-to-digital conversion: An ADS1292 is configured as a 24-bit delta-sigma ADC with a sampling rate of 250Hz (meeting the required EEG signal bandwidth). The PGA gain is dynamically adjustable (default 100). Wireless transmission: The nRF52832 transmits data at 2Mbps, drawing ≤5mA in BLE connected mode and ≤1μA in standby mode. Power configuration: Powered by 3.7V / 200mAh lithium battery, output 3.3V (analog) and 1.8V (digital) through TPS73633 (LDO), battery life ≥12h.

[0073] PCB layout optimization includes: Layered design: The signal chain (analog portion) and digital circuits are partitioned and laid out in layers. Grounded copper foil is laid on the analog signal layer to reduce electromagnetic coupling. Cable processing: Shielded twisted-pair cables are used for electrode leads, and the preamplifier circuit is partially shielded to reduce ambient noise pickup.

[0074] With only two differential acquisition channels (compared to traditional multi-channel ≥8), the device reduces the number of core components, such as op amps and ADCs, resulting in over 70% cost reduction and a 60% reduction in size (compatible with wearable form factors such as headphones and headbands). No conductive paste is required; only skin cleansing is required before wearing, reducing preparation time from 30 minutes with traditional wet electrodes to 2 minutes, making it suitable for everyday wear (such as exercise and sleep monitoring). The hairless area behind the ear prevents hair from interfering with electrode contact, allowing for long-term wear (≥8 hours) without skin allergies or increased impedance, making it suitable for continuous monitoring scenarios. High signal-to-noise ratio: Through 120dB CMRR differential amplification, right leg drive common-mode rejection, and a 24-bit high-precision ADC, the measured signal-to-noise ratio is increased to over 25dB (compared to ≤20dB for traditional single-channel dry electrode devices), enabling clear capture of signals in the α, β, and γ frequency bands (0.1-100Hz). Interference immunity: Combined anti-aliasing filtering and ESD protection effectively suppress high-frequency noise (such as mobile phone RF interference) and electrostatic shock, achieving signal distortion of ≤0.5%. Battery life optimization: Multi-level LDO independent power supply, BLE low-power transmission, the whole device power consumption ≤ 50mW, support Bluetooth 5.0 long-distance transmission (stable connection within 10m range), to meet the portability requirements of wearable devices.

[0075] Consumer Market: Suitable for smart headphones, headbands, wearable headbands, and other devices, it is used for daily health management such as fatigue monitoring, attention analysis, and sleep quality assessment. Medical Assistance: As a portable pre-screening device, it assists in the initial detection of EEG abnormalities such as epilepsy and insomnia, making up for the bulkiness and complex operation of traditional equipment.

[0076] In summary, this circuit breaks through the cost, size, and usability bottlenecks of traditional multi-channel wet electrode equipment while ensuring the quality of EEG signals through differential acquisition of dry electrodes behind the ear + high integration and low power consumption design, providing a feasible solution for EEG applications in consumer wearable devices.

[0077] In some embodiments, the dual-channel differential input interface is respectively connected to the two output ends of the anti-aliasing filtering module; wherein, the negative input end of the dual-channel differential input interface is commonly connected as a common reference node, and the positive input end receives the EEG signals from the two differential electrodes respectively; the analog front-end chip has a built-in programmable gain amplifier and a 24-bit high-precision ADC converter, and is connected to the main control chip through a digital interface to transmit the converted digital signal.

[0078] The two outputs of the anti-aliasing filter module (the positive and negative outputs of the differential signal) are connected to the dual-channel differential input interface of the analog front-end chip. The negative input of the differential input interface is connected to a common reference node (usually the signal ground or a noise-filtered virtual ground), and the positive input is connected to the EEG signal output of two differential electrodes (the measurement electrode and the reference electrode).

[0079] The analog front-end chip features a built-in programmable gain amplifier (PGA) and a 24-bit delta-sigma ADC. It connects to a host control chip (such as an MCU or Bluetooth chip) via an SPI / I2C digital interface, transmitting 24-bit digital signals (e.g., in two's complement format). The PGA gain is configurable via registers (e.g., 1x, 2x, 128x), and the default gain dynamically adjusts based on the front-end amplification factor (for example, with a 100x preamplification, the PGA is set to 1x to avoid saturation). The ADC sampling rate is set between 250Hz and 1000Hz (to meet the Nyquist sampling theorem for EEG signals with a maximum frequency of 100Hz). An internal anti-aliasing filter is integrated to assist with the external filtering module.

[0080] The differential input structure naturally suppresses common-mode noise (such as power frequency interference and electrode contact noise). Combined with the noise isolation design of the common reference node, the overall CMRR is improved to over 120dB, effectively preserving the differential EEG signal (μV level). The PGA dynamically adjusts gain to accommodate individual EEG signal amplitude differences (for example, the signal amplitude may be higher in patients with epilepsy), avoiding ADC saturation or excessive quantization noise. The 24-bit high-precision ADC achieves microvolt resolution (LSB ≈ 0.15μV), fully preserving low-frequency weak signal characteristics such as theta / delta. The integrated analog front-end chip eliminates the need for external discrete components (such as independent PGAs and ADCs), reducing PCB layout complexity and the risk of noise introduction, while enabling high-speed, low-bit-error-rate data transmission via a digital interface.

[0081] Exemplarily, the analog front-end chip integrates a continuous disconnection detection circuit and a test signal generator, and the corresponding common mode rejection ratio is not less than 120dB and the input reference noise is less than 5μVpp.

[0082] The chip integrates a constant current source (e.g., 1μA) and a voltage comparator, which periodically injects a small test current into the electrode loop to monitor changes in the impedance between the electrode and the skin. If an electrode becomes detached or poorly connected, the loop impedance increases sharply (exceeding a threshold, such as 10MΩ). The comparator outputs a disconnection signal to the main control chip, triggering an alarm (such as a flashing LED or Bluetooth notification). A built-in function generator can output a standard test signal (e.g., a 10μV, 10Hz sine wave), which can be injected into the differential input channel through register configuration for factory calibration or real-time self-testing.

[0083] During the self-test, the electrode inputs are disconnected and a test signal is injected to verify that the gain, linearity, and noise performance of the entire signal chain (amplification, filtering, ADC) meet specifications. A chip with a CMRR ≥ 120dB (such as the ADI AD8233) is used, with input-referred noise below 5μVpp (peak-to-peak, covering the 0.1-100Hz frequency range) and integrated high-precision PGA and noise filters. A disconnection detection function monitors electrode contact status in real time to prevent invalid data acquisition due to electrode detachment, making it particularly useful for athletic scenarios (e.g., electrode displacement during running). The test signal generator supports automated pre-shipment calibration (to compensate for component tolerances) and daily user self-tests (such as automatically running a calibration routine at power-up) to ensure long-term signal consistency. With a CMRR ≥ 120dB and input noise < 5μVpp, the chip can capture weak EEG fluctuations as low as 0.1μV (such as delta waves during sleep), meeting scientific research-grade signal quality requirements while reducing the complexity of subsequent signal processing algorithms.

[0084] In some embodiments, the operational amplifier circuit of the pre-active amplifier module adopts a non-inverting input amplifier topology, and the corresponding input end is directly coupled to the dry electrode sensor module through a resistor network to form an input impedance greater than 10GΩ.

[0085] A non-inverting amplifier structure is adopted, the non-inverting input of the operational amplifier is directly connected to the dry electrode, and the inverting input is grounded through a resistor network or forms feedback.

[0086] Input impedance: By selecting a high input impedance op amp (input bias current ≤ 1pA) and a minimalist resistor network (to avoid leakage current introduced by low-value resistors), an input impedance of >10GΩ is achieved.

[0087] The resistor network configuration includes a 10MΩ feedback resistor, a 100kΩ gain resistor, and an initial gain of 101. A small capacitor (1nF) in parallel with an ESD protection resistor (e.g., 1MΩ) suppresses high-frequency noise. Direct DC coupling between the dry electrode and the op amp input eliminates DC-blocking capacitors, preserving DC offset information (e.g., slow cortical potentials) in the EEG signal. The right leg drive circuit also suppresses DC common-mode voltage drift.

[0088] An input impedance exceeding 10 GΩ matches the high contact impedance of dry electrodes (typically 5-50 kΩ), preventing signal attenuation (e.g., traditional low-input impedance op amps can cause over 10% signal loss) and ensuring distortion-free amplification of microvolt-level signals. The direct-coupled design allows acquisition of ultra-low-frequency EEG signals below 0.1 Hz (such as delta waves and slow cortical potentials), which is crucial for scenarios such as sleep staging and anesthesia depth monitoring. DC-blocking capacitors would lose this information. The noise of the non-inverting amplifier topology primarily comes from the op amp itself. By selecting a low-noise op amp and optimizing the PCB layout (copper grounding the input), the preamplifier noise contribution can be kept below 1 μVpp.

[0089] In some embodiments, the passive filtering network of the anti-aliasing filtering module is composed of at least three resistor-capacitor devices to form a multi-order low-pass filter, and the corresponding cutoff frequency is set to 50Hz-150Hz to match the EEG signal frequency band; the active filtering circuit is composed of an operational amplifier and resistor-capacitor devices to form a second-order or higher low-pass filtering topology.

[0090] A third-order RC low-pass filter (e.g., three-stage RC in series) is used. Each stage is composed of a 10kΩ resistor and a 10nF capacitor. The single-stage cutoff frequency is 1.6kHz. After the third stage is cascaded, the cutoff frequency is reduced to ≈500Hz, and the roll-off slope is -60dB / decade.

[0091] The resistors should be high-precision metal film resistors (error ≤ 1%), and the capacitors should be low-ESR ceramic capacitors (such as X7R dielectric, capacitance stability ≤ ± 5%) to reduce the impact of component tolerance on filtering characteristics.

[0092] The active filter circuit uses a second-order Sallen-Key low-pass filter (op amp OP27 with R1=R2=10kΩ, C1=C2=10nF) with a cutoff frequency of ≈1.6kHz. After cascading with the passive filter, it forms a fifth-order filter with an overall cutoff frequency of 150Hz (by adjusting the capacitor to 33nF to make fc=50Hz or 100nF to 150Hz).

[0093] The gain of the active filter is set to 1 (buffer mode), and the output impedance is reduced to <500Ω, matching the low input impedance requirement of the analog front-end chip.

[0094] The 50-150Hz cutoff frequency covers the primary EEG signal frequency band (0.1-100Hz), exceeding the maximum frequency by a factor of two (meeting the Nyquist theorem). This effectively filters high-frequency noise above 200Hz (such as myoelectric interference and switching power supply ripple), preventing aliasing distortion during ADC sampling (e.g., misinterpreting 150Hz noise as a 50Hz signal). The multi-order filter's steep roll-off (above -40dB / decade) significantly attenuates out-of-band noise, for example, attenuating 500Hz noise by >60dB, reducing the noise power entering the ADC to less than 0.1% of the original signal and improving the purity of the subsequent digital signal. The active filter circuit acts as a buffer, converting the high-impedance output of the passive filter to a low impedance (<500Ω), minimizing signal attenuation over long transmission lines (e.g., signal loss ≤1% for a 50cm electrode lead), thus meeting the input requirements of subsequent chips.

[0095] In some embodiments, the programmable impedance adjustment circuit of the right leg drive feedback module includes a parallel resistor and capacitor network, which dynamically controls the feedback bandwidth by adjusting the RC time constant; the current limiting protection circuit is connected in series between the right leg drive output terminal and the human body reference electrode, and forms a bidirectional current limiting structure with the ESD protection diode.

[0096] A parallel RC network (e.g., R = 100kΩ in parallel with C = 10nF) is connected to the feedback loop. A digital potentiometer (e.g., the AD5290) dynamically adjusts the resistance value (0-1MΩ) to change the RC time constant (τ = RC), thereby controlling the feedback bandwidth (e.g., fc = 1 / (2πτ), adjustable from 10Hz to 1kHz). The feedback signal is taken from the common-mode output of the preamplifier module, amplified by an op amp (gain -1), and then injected into the body's reference electrode (e.g., the right leg or forehead ground electrode).

[0097] A 1MΩ protection resistor is connected in series to the feedback loop to limit the maximum current to ≤10μA (human safety threshold, IEC 60601 standard). A bidirectional TVS diode (such as SMBJ33A) is connected in parallel to clamp the voltage to ±33V to prevent overvoltage shocks caused by static electricity or leakage of medical equipment.

[0098] The programmable RC network dynamically adjusts the feedback bandwidth based on the user's skin impedance (e.g., dry skin impedance 100kΩ vs. wet skin 10kΩ). For example, when the impedance is high, the RC time constant is increased (reducing the bandwidth) to reduce baseline drift interference; when the impedance is low, the bandwidth is reduced to enhance the ability to suppress power-frequency noise.

[0099] The 1MΩ resistor and TVS diode provide dual protection, ensuring that even in the event of a device failure, the current flowing into the human body remains below the safety limit. At the same time, the ESD protection diode absorbs ±8kV contact discharge energy (compliant with IEC 61000-4-2 standard), protecting the front-end circuit from electrostatic damage.

[0100] Negative feedback is used to reduce the human body common-mode voltage to less than 1 / 1000 of its original amplitude (theoretically increasing CMRR by 20dB). The system is particularly effective in suppressing 50Hz power frequency interference, with the measured common-mode noise amplitude reduced from 500μV to <50μV.

[0101] In some embodiments, the low-noise LDO power supply circuit of the power management module independently supplies power to the analog power pin and the digital power pin of the analog front-end chip, and the power ripple suppression ratio of each corresponding LDO chip is higher than 80dB and an enable control pin is integrated to achieve time-sharing power supply.

[0102] Three independent low-dropout (LDO) chips are used to power the analog front-end chip's analog power supply (AVDD), digital power supply (DVDD), and preamplifier. For the analog power supply, the ADM7150 (noise level 10μVrms, PSRR 80dB@1kHz) is selected as the LDO. It takes a 3.7V lithium battery as input and outputs 2.5V to the analog front-end and op amp. For the digital power supply, the TPS73633 (noise level 20μVrms, PSRR 75dB@1kHz) is selected as the LDO. It outputs 3.3V to the Bluetooth chip and MCU, and 1.8V to the digital interface circuits. The analog ground (AGND) and digital ground (DGND) are connected at a single point via a 0Ω resistor to prevent ground loop noise coupling. Each LDO's enable pin (EN) is connected to the MCU, supporting time-sharing wake-up. For example, all power supplies are enabled during Bluetooth transmission, while only the disconnection detection circuit remains powered during standby mode, reducing power consumption from 50mW to 5μW.

[0103] Independent analog and digital power supplies, combined with a single-point grounding design, completely isolate high-frequency noise from digital circuits (such as clock jitter from the Bluetooth chip) from the analog signal chain. Measured analog power supply ripple has been reduced from 500μVpp to <50μVpp, preventing ripple from modulating μV-level EEG signals. Enable control enables "active-sleep" mode switching, allowing the device to enter sleep mode when no data is being transmitted (retaining only electrode disconnection detection). Battery life is extended from 12 hours to 72 hours, meeting long-term wearable requirements. The LDO's PSRR of >80dB suppresses power supply noise to less than one ten-thousandth. Even with a ±0.5V fluctuation in the lithium battery voltage, the output voltage remains <50μV, ensuring stable reference voltages for the op amp and ADC, preventing gain drift and quantization errors.

[0104] In some embodiments, the Bluetooth communication chip of the wireless transmission module integrates a DC-DC voltage converter, its RF transmission peak current is lower than 10mA and its receiving sensitivity is better than -90dBm, and signal transmission and reception are achieved through an onboard inverted F-type antenna and an impedance matching circuit.

[0105] This Bluetooth 5.0 chip uses an integrated DC-DC converter (such as the Nordic nRF52840). The DC-DC converter boosts the 3.7V lithium battery to 3.3V, achieving a conversion efficiency of ≥90%, a transmit peak current of ≤8mA (at 0dBm output power), and a receive sensitivity of -96dBm (better than the typical -90dBm). The RF front end integrates a π-type impedance matching network (1nH inductor + 33pF capacitor) to match the 50Ω characteristic impedance of the inverted-F antenna. The antenna is designed to be 15mm × 8mm (compacting with the PCB space of the wearable device) and operates in the 2.402-2.480GHz frequency band.

[0106] BLE custom service (GATT) is used to transmit EEG data. The data packet format is: synchronization header (2 bytes) + channel 1 data (3 bytes, 24-bit ADC) + channel 2 data (3 bytes) + checksum (1 byte). The transmission rate is 200kbps, and the bit error rate is controlled below 10^-7 through CRC check.

[0107] The -96dBm high sensitivity and omnidirectional radiation characteristics of the inverted-F antenna ensure uninterrupted transmission within a 10m range, maintaining a connection even when the user is in motion (e.g., with an arm blocking the view), preventing data loss. DC-DC conversion improves power efficiency, with a transmit current of ≤10mA (compared to 15-20mA for traditional Bluetooth chips). Combined with BLE's periodic broadcasting (e.g., transmission every 20ms), this reduces overall power consumption by 30%, making it compatible with button batteries or micro lithium batteries. The integrated Bluetooth chip and onboard antenna eliminate the need for external RF components, saving PCB space (occupying an area of ​​≤10mm×10mm), making it suitable for size-sensitive wearable devices such as headphones and headbands.

[0108] In some embodiments, the EEG amplification circuit is module-integrated through a four-layer PCB board; wherein, the corresponding analog signal lines and digital signal lines adopt a layered isolation layout, and the dry electrode contact area adopts an immersion gold process to reduce contact impedance.

[0109] Four-layer board structure design:

[0110] Layer 1: Analog signal layer, layout of dry electrode contacts, preamplifier circuits, anti-aliasing filter components, signal line width ≥ 0.3mm, avoid sharp angle routing to reduce EMI radiation.

[0111] Layer 2: Ground layer (complete copper foil), analog ground and digital ground are divided into partitions, and single-point grounding is connected through vias.

[0112] Layer 3: Power layer, paving 3.7V lithium battery power supply and LDO output voltage. The power trace width is ≥1mm to reduce impedance.

[0113] Layer 4: Digital signal layer, layout of Bluetooth chip, MCU, ADC digital interface, clock line (such as SPI clock) uses differential routing and grounding.

[0114] The dry electrode contact area (PCB pad) utilizes an immersion gold process (gold plating thickness ≥ 3μm) with a surface roughness of ≤ 1μm. This reduces contact impedance from 500mΩ with conventional tin-spraying to 50mΩ, improving the reliability of the connection between the electrode and the PCB. A continuous ground copper foil is laid beneath the analog signal traces to provide electromagnetic shielding and reduce noise coupling from the digital layer (such as interference from Bluetooth RF noise on the preamplifier circuit).

[0115] The layered isolation layout physically separates analog and digital circuits. In actual electromagnetic compatibility (EMC) testing, the interference of RF radiation noise on EEG signals was reduced from 20μVpp to <2μVpp, meeting CE / FCC certification requirements. The smooth surface and low oxidation properties of the immersion gold process ensure long-term stable contact between the electrodes and the PCB solder joints (even when worn and sweating, the impedance change is ≤10%), avoiding signal baseline drift caused by poor contact (common in conventional tin-spraying processes, which can drift up to 50μV / s). The four-layer board is kept to a thickness of less than 1.6mm by stacking power and ground layers, adapting to the lightweight design of wearable devices. The immersion gold pads offer superior corrosion resistance compared to traditional processes, extending the device's service life (for example, no oxidation failure of solder joints for over three years).

[0116] In some embodiments, a lightweight convolutional neural network (CNN) is linked with a hardware filtering module to achieve real-time classification and targeted suppression of power frequency noise, electromyographic noise (EMG), and motion artifacts.

[0117] Over 1000 hours of EEG data containing various noise types were collected and annotated for three types of noise: 50Hz power frequency (including harmonics), EMG (μV-level high-frequency glitches), and motion artifacts (baseline drift >50μV). Time-frequency domain features (short-time Fourier transform spectra and wavelet energy entropy) were extracted as a training set. A three-layer CNN model (inputting a 128-point time-domain signal and outputting probabilities for the three noise types) was trained using TensorFlow Lite. The model parameters were compressed to under 50KB to fit within the 64KB SRAM of an MCU (such as the STM32H7). The main control chip acquires preprocessed signals (128-point window) at 100Hz and feeds them into a CNN model for real-time noise classification. For power-frequency noise, the hardware notch filter is triggered (using a digital potentiometer to switch RC parameters, creating a notch at 50Hz or 60Hz). For EMG noise, the anti-aliasing filter cutoff frequency is dynamically adjusted to 150Hz (originally 100Hz), attenuating high-frequency components above 200Hz. For motion artifacts, the high-pass filter mode of the right leg drive feedback is activated (increasing the RC time constant to enhance baseline drift suppression). The analog front-end chip's PGA gain, anti-aliasing filter capacitor array (programmable capacitor bank), and right leg drive RC network are all connected to the MCU via an I2C interface. The control commands output by the algorithm reconstruct the hardware parameters within 1ms.

[0118] Improved dynamic noise suppression: For different noise types, power frequency suppression is increased from 40dB with a fixed notch to 65dB with an adaptive notch, EMG high-frequency attenuation is increased from -30dB@200Hz to -50dB, and baseline drift caused by motion artifacts is reduced from ±100μV to ±20μV.

[0119] Enhanced generalization: Through continuous online learning (model parameters are updated every 10 minutes), the system can adapt to the differences in noise characteristics of different users (for example, automatic identification of 50Hz / 60Hz power frequencies in different countries). Traditional fixed filtering solutions can only handle a single noise scenario.

[0120] Computing efficiency optimization: The model's single inference time is less than 50μs (MCU main frequency 200MHz), and the power consumption increment is less than 0.5mW, which is much lower than the 2mW power consumption of traditional digital signal processing (such as FFT+threshold judgment).

[0121] In some embodiments, compressed sensing (CS) theory is used to compress high-frequency redundant information in real time after 24-bit ADC sampling, and sparse representation algorithm is combined to achieve data reduction and feature enhancement.

[0122] Aiming at the sparsity of EEG signals (α / β waves are sparse in the frequency domain), discrete cosine transform (DCT) is used as the sparse basis to design a 128×256-dimensional Gaussian random measurement matrix (the hardware is implemented as a fixed multiplier array).

[0123] Each 256-point raw EEG data (24 bits × 256 = 768 bytes) is compressed to 128 points (384 bytes) through the observation matrix, with a compression ratio of 2:1, while retaining more than 95% of the energy information.

[0124] A simplified version of the orthogonal matching pursuit (OMP) algorithm (iterations limited to 10) is run on the main control chip, and the reconstruction error is controlled within 5μV, with a reconstruction time of <200μs / frame.

[0125] The compressed data is transmitted via Bluetooth, along with the sparse coefficient index (16 bytes). The receiving end (mobile app) uses the complete observation matrix to restore the signal. A dedicated CS calculation unit is integrated into the PCB design: two 16-bit multipliers and an accumulator (embedded in the MCU FPU). The ADC's DMA channel directly imports data into the calculation buffer, eliminating CPU intervention.

[0126] Transmission power consumption is reduced by 40%: The amount of Bluetooth transmission data is reduced from 768 bytes / frame to 400 bytes / frame, and the transmission time is shortened from 1.6ms to 1.0ms. Combined with BLE's energy-saving mode, the power consumption of the entire device is reduced from 15mW to 9mW, and the battery life is extended to more than 10 hours.

[0127] Enhanced anti-aliasing performance: Compressed sensing naturally has undersampling anti-aliasing capabilities. When the sampling rate is reduced from 250Hz to 125Hz (meeting the sparsity conditions of CS theory), the signal reconstruction error is less than 3μVpp, which effectively improves the effective utilization efficiency of the hardware sampling rate.

[0128] Feature enhancement preprocessing: Non-sparse noise (such as white noise) is automatically suppressed during the sparse representation process. The signal-to-noise ratio (SNR) of the preprocessed signal is improved from 20dB to 28dB, providing higher-quality data for subsequent deep learning model input.

[0129] In some embodiments, signal statistical characteristics are combined with dynamic electrode impedance modeling to achieve a multi-dimensional assessment of contact quality (contact impedance, signal stability, and baseline drift). Hardware detection: Impedance measurement of the disconnection detection circuit (once per second, with a resolution of 10kΩ); Signal characterization: Calculation of the standard deviation (reflecting the noise level), kurtosis (reflecting the probability of impulse noise), and baseline drift rate (the difference between the mean values ​​of adjacent windows) of the signal within a 5s window; Temperature compensation: An integrated NTC temperature sensor (accuracy of ±0.5°C) corrects for the effect of temperature on skin impedance (impedance decreases by approximately 2% for every 1°C increase).

[0130] A three-level evaluation index (good / average / poor) is established, and the input variables are fuzzified and reasoned through the rule base:

[0131] If "Impedance < 50 kΩ and standard deviation < 10 μV and drift < 5 μV / s" → Good;

[0132] If "Impedance 50-100 kΩ or standard deviation 10-30 μV" → Normal (triggering electrode cleaning prompt);

[0133] If "Impedance > 100kΩ or Kurtosis > 3" → Bad (triggering disconnection warning).

[0134] The evaluation results are displayed through the LED color (green / yellow / red) and the Bluetooth APP pop-up window. In the "normal" state, the test signal injection (test signal generator in Example 2) is automatically started to verify the integrity of the signal chain.

[0135] Traditional disconnection detection can only determine "on / off". This solution implements a quantitative scoring system of 0-100 points (such as good ≥80 points, average 60-80 points, and poor <60 points). Users can intuitively understand the wearing status, which is particularly suitable for signal quality control in scientific research scenarios.

[0136] After fusing multimodal data, misjudgment caused by sweat conductivity in single impedance detection (such as scenarios with low impedance but motion artifacts) is avoided, and the measured false alarm rate is reduced from 5 times / hour in the traditional solution to 1.5 times / hour.

[0137] In the "normal" state, users are advised to wipe the electrodes or adjust the wearing position to reduce the invalid data collection time caused by poor contact from an average of 20% to less than 5%.

[0138] In some embodiments, a baseline model is trained using historical user data, combined with transfer learning to quickly adapt to the individual differences of new users and dynamically subtract DC drift and physiological baseline shift. A cloud server pre-trains a universal baseline model (based on resting-state data from over 100,000 healthy individuals), extracting the baseline mean and variance of delta, theta, alpha, and beta waves as prior knowledge. When the edge device is first used, 5 minutes of resting-state signals from the user are collected. Transfer learning is then used to fine-tune the last fully connected layer (updating only 10% of the parameters) to generate a personalized baseline model (storage capacity <1KB). A sliding window (30s) is used to calculate the mean drift of the current signal. Combined with the baseline mean of the personalized model, an adaptive filter (such as the RLS algorithm) is used to dynamically subtract DC offset and low-frequency drift: V^(t) = V(t) − (μuser + k*drift_rate), where μuser is the personalized baseline mean and k is the drift suppression coefficient (adjustable from 0.1 to 0.5).

[0139] The signal after baseline subtraction is fed back to the PGA gain control of the analog front end to automatically compensate for signal amplitude fluctuations caused by baseline offset (for example, when the baseline rises by 20μV, the PGA gain is dynamically reduced by 10% to avoid ADC saturation).

[0140] Traditional fixed-baseline calibration can have an error rate of ±30μV across different users. This solution, through transfer learning, reduces this error to within ±5μV, significantly improving compatibility with children (who experience faster baseline drift) and the elderly (who experience lower signal amplitude). Combined with dynamic adjustment of the RLS algorithm, the suppression of slow drift below 0.1Hz is improved from the traditional high-pass filter's -10dB to -30dB, preserving valuable slow cortical potential signals (such as the P300 component). The baseline stability of the calibrated signal meets clinical EEG device standards (drift <10μV / min), enabling the device to be used in scenarios requiring a high-precision baseline (such as event-related potential (ERP) testing).

[0141] In some embodiments, the pulse emission mechanism of biological neurons is imitated to construct a low-power spiking neural network (SNN) to achieve real-time detection of the edges of abnormal waveforms such as epileptic spikes and sharp waves.

[0142] The SNN model utilizes a two-layer spiking neuron network: an input layer with 100 LIF neurons (encoding the rate of change of signal amplitude within a 50ms window), and an output layer with three neurons corresponding to "normal," "suspected abnormal," and "urgent abnormal," respectively. Synaptic weights are pre-trained through supervised learning (using the CHB-MIT epilepsy database). Single-neuron computations require only comparators and counters, eliminating the need for floating-point operations.

[0143] When the output layer "urgent abnormal" neuron discharges (exceeding the threshold frequency of 5Hz), the following actions are completed within 100μs:

[0144] The hardware storage module (1MB SRAM) is triggered to save the signal data of the first 5 seconds and the last 3 seconds;

[0145] Send emergency interrupt signal via Bluetooth (takes priority over regular data transmission);

[0146] The control LED flashes at 2Hz (visual alarm).

[0147] SNN runs in the MCU's low-power mode (Cortex-M4F core, 16MHz low-frequency clock), with a single detection power consumption of <100nJ and a continuous operation of 24 hours of power consumption of <1mAh, which is only 1 / 20 of the traditional deep learning model.

[0148] The detection delay is shortened from 10ms of traditional CNN to 1ms (neuronal pulse conduction simulation), ensuring the capture of spike signals lasting only 20ms, and the missed detection rate is reduced from 5% to below 1%.

[0149] On the CHB-MIT dataset, the F1 score of epileptic seizure detection reached 0.92, which is higher than mainstream edge algorithms (such as 0.85 of support vector machines), and does not require cloud data upload, protecting user privacy.

[0150] Even when Bluetooth transmission is interrupted, the local storage module can record 8 seconds of key data (including waveforms before and after the attack), providing a complete chain of evidence for subsequent clinical diagnosis. Traditional solutions can only record the current transmission data.

[0151] In some embodiments, a reinforcement learning (RL) algorithm is used to dynamically adjust the PGA gain and ADC sampling rate to achieve Pareto optimality of signal-to-noise ratio and dynamic range in scenarios with signal amplitude fluctuations.

[0152] State-action space definition:

[0153] Status: current signal peak (0-80% of full scale), PGA gain range (1 / 2 / 4 / 8 / 16 times), ADC sampling rate (250 / 500 / 1000Hz);

[0154] Action: gain adjustment (±1 level), sampling rate adjustment (±1 level);

[0155] Reward function: R = \text{SNR} - 0.1 \times |\text{peak-full-scale 80%}| - 0.05\times \text{sampling rate} , balancing signal-to-noise ratio, dynamic range, and power consumption.

[0156] The initial strategy uses a greedy algorithm (based on signal amplitude histogram statistics). After two hours of operation, it switches to Q-Learning online learning, updating the Q table (stored in the MCU's 512KB flash) every 100ms. When the signal amplitude remains stable within the optimal range for a given gain level (30%-70% of full scale) for 30 consecutive seconds, the system enters "energy-saving mode," freezing the current parameters and reducing the sampling rate to 250Hz. Safety margins are set: When adjusting gain, ensure that the signal peak does not exceed 90% of the ADC's full scale, and adjust the sampling rate to at least twice the signal's highest frequency (to satisfy the Nyquist theorem).

[0157] For signals with large amplitude fluctuations (such as 50μV during waking alpha waves to 500μV during epileptic seizures), gain switching delay is less than 50μs, avoiding the 100ms signal distortion window associated with traditional manual gain switching and increasing effective signal retention from 70% to 95%. For signals in the resting state (stable amplitude), the sampling rate automatically drops to 250Hz, reducing power consumption by 40%. During active states (amplitude fluctuations), the gain is dynamically adjusted to maintain optimal quantization accuracy, improving overall system energy efficiency (EE) by 35% compared to fixed-parameter solutions. By using reward function constraints, the probability of a signal peak exceeding 80% of full scale is reduced from 15% in traditional solutions to less than 3%, virtually eliminating clipping distortion caused by ADC saturation. This approach is particularly suitable for scenarios with dynamic signal fluctuations, such as emotion monitoring.

[0158] In some embodiments, as shown in the attached Figure 2 As shown, U1 is a high-performance, two-channel analog front-end. This chip integrates two low-noise PGA amplifiers and a 24-bit, high-resolution ADC, consuming only 350uW per channel. It boasts a common-mode rejection ratio of up to 120dB and a minimum input-referred noise of just 2.5uVpp. It also features a right-leg drive amplifier, continuous disconnection detection, and test signals. Its high integration and high performance make it ideal for use in EEG acquisition equipment. The chip's package size is only 4mm x 4mm, making it ideal for wearable devices. U1 is equipped with two differential input channels to accommodate small behind-the-ear wearables. The N terminals of the two differential inputs are connected together as a common reference input. An anti-aliasing low-pass filter circuit, consisting of R46, R50, R51, C83, C85, and C86, is placed between the N terminal and the two P terminals of the signal input chip to filter out high-frequency signals while retaining the low-frequency signals of interest, improving the signal-to-noise ratio. U1 has a built-in right leg drive circuit. The negative feedback size is set externally through R58 and C67 to adjust the bandwidth. R57 acts as a protective resistor from the right leg drive output to the human body, effectively limiting the current from the right leg drive to the human body to prevent damage to the human body. D10~D13 are ESD protection diodes to prevent ESD static electricity from the human body from damaging the amplifier. U1 is connected to the main control MCU through the SPI interface. The MCU reads the ADC channel data converted by U1 through this SPI interface. As shown in the attached figure Figure 3 U4 and U14 are analog power supply and digital power converter LDO chips for U1 respectively. The device has a power supply ripple rejection ratio of up to 90dB and an ultra-low noise of only 6.5uVrms. The package size is only 1mm X 1mm, which can greatly reduce the size of the whole device and is particularly suitable for wearable behind-the-ear devices. Figure 4 U13 is a highly integrated, low-power Bluetooth BLE5.3 master chip. It has a rich set of peripherals, 521kB flash, 64kB RAM, up to 64MHz operating speed, integrated DCDC converter, effectively reducing power consumption, BLE transmit and receive peak current as low as 5.3mA, and has an excellent -96dBm sensitivity. At the same time, the package size is only 3.2mmX3.0mm, which is very suitable for wearable devices. Figure 5 U2 is a preamplifier and active low-pass filter circuit for the active electrode amplifier, consisting of a high-performance, low-power precision operational amplifier. D2 is an ESD protection diode, protecting the op amp input from ESD damage. The non-inverting input amplifier formed by R10 and R13 provides extremely high input impedance, suitable for EEG signal acquisition in dry electrode applications. The active low-pass filter formed by R3, R4, C3, and C4 achieves better filtering and has a low output impedance, enhancing the signal transmission's immunity to interference.

[0159] Please refer to Figure 6 The present application provides a brain-computer interface device, including a low-channel EEG amplification circuit for behind-the-ear EEG signals provided by any embodiment of the present application.

[0160] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A low-channel EEG amplification circuit for behind-the-ear EEG signals, characterized in that: include: The dry electrode sensing module consists of two differential electrodes and is placed in the hairless area behind the ear to collect EEG signals. A preamplifier module includes a high-input impedance operational amplifier circuit, which is directly coupled to the dry electrode sensor module to form a differential signal amplification structure with a common-mode rejection ratio of at least 120 dB; the dry electrode sensor module and the preamplifier module are electrically connected via an ESD protection circuit; An anti-aliasing filter module, composed of a multi-order passive filter network and an active filter circuit in cascade, is connected to the output end of the pre-active amplifier module to filter out high-frequency interference signals; the output impedance of the anti-aliasing filter module is less than 500Ω; The right leg drive feedback module includes a programmable impedance adjustment circuit and a current limiting protection circuit, and forms a closed-loop feedback loop with the preamplifier module. The output end of the right leg drive feedback module forms a current loop with the human body through a protective resistor; The analog-to-digital conversion module uses an integrated low-power analog front-end chip, including a dual-channel differential input interface, a programmable gain amplifier, and a 24-bit high-precision ADC converter; The wireless transmission module integrates a low-power Bluetooth communication chip and is connected to the analog-to-digital conversion module through a digital interface; A power management module includes a multi-stage low-noise LDO power supply circuit, which provides independent power supply to the analog front-end chip, operational amplifier circuit and wireless transmission module respectively; the dual-channel differential input interface is respectively connected to the two output terminals of the anti-aliasing filter module; wherein the negative input terminal of the dual-channel differential input interface is commonly connected to a common reference node, and the positive input terminal receives the EEG signals from the two differential electrodes respectively; the analog front-end chip has a built-in programmable gain amplifier and a 24-bit high-precision ADC converter, which is connected to the main control chip through a digital interface to transmit the converted digital signal. The analog front-end chip integrates a continuous disconnection detection circuit and a test signal generator, and the corresponding common-mode rejection ratio is not less than 120dB and the input reference noise is less than 5μVpp; The baseline model is trained using historical user data, combined with transfer learning to quickly adapt to the individual differences of new users, and dynamically deduct DC drift and physiological baseline offset; the cloud server pre-trains a general baseline model and extracts the baseline mean and variance of δ, θ, α, or β waves as prior knowledge; when the edge device is used for the first time, 5 minutes of resting-state signals from the user are collected, and the last fully connected layer is fine-tuned through transfer learning to generate a personalized baseline model; a sliding window is used to calculate the mean drift of the current signal, and combined with the baseline mean of the personalized model, the DC offset and low-frequency drift are dynamically deducted through an adaptive filter; the signal after baseline subtraction is fed back to the PGA gain control of the analog front end to automatically compensate for the signal amplitude fluctuation caused by baseline offset.

2. The low-channel EEG amplifying circuit for behind-the-ear EEG signals according to claim 1, characterized in that: The operational amplifier circuit of the pre-active amplifier module adopts a non-inverting input amplifier topology structure, and the corresponding input end is directly coupled to the dry electrode sensor module through a resistor network to form an input impedance greater than 10GΩ.

3. The low-channel EEG amplifying circuit for behind-the-ear EEG signals according to claim 1, characterized in that: The passive filtering network of the anti-aliasing filtering module consists of at least three resistor-capacitor devices forming a multi-order low-pass filter, and the corresponding cutoff frequency is set to 50Hz-150Hz to match the EEG signal frequency band; the active filtering circuit consists of an operational amplifier and resistor-capacitor devices forming a second-order or higher low-pass filtering topology.

4. The low-channel EEG amplifying circuit for behind-the-ear EEG signals according to claim 1, characterized in that: The programmable impedance adjustment circuit of the right leg drive feedback module includes a parallel resistor and capacitor network, which dynamically controls the feedback bandwidth by adjusting the RC time constant; the current limiting protection circuit is connected in series between the right leg drive output terminal and the human body reference electrode, and forms a bidirectional current limiting structure with the ESD protection diode.

5. The low-channel EEG amplifying circuit for behind-the-ear EEG signals according to claim 1, characterized in that: The low-noise LDO power supply circuit of the power management module independently supplies power to the analog power pin and digital power pin of the analog front-end chip, and the power ripple rejection ratio of each corresponding LDO chip is higher than 80dB and an enable control pin is integrated to achieve time-sharing power supply.

6. The low-channel EEG amplifying circuit for behind-the-ear EEG signals according to claim 1, characterized in that: The Bluetooth communication chip of the wireless transmission module integrates a DC-DC voltage converter, its RF transmission peak current is less than 10mA and its receiving sensitivity is better than -90dBm, and signal transmission and reception are achieved through an onboard inverted F-type antenna and an impedance matching circuit.

7. The low-channel EEG amplifying circuit for behind-the-ear EEG signals according to claim 1, characterized in that: The EEG amplification circuit is integrated into a module through a four-layer PCB board; Among them, the corresponding analog signal lines and digital signal lines adopt a layered isolation layout, and the dry electrode contact area adopts the immersion gold process to reduce the contact impedance.

8. A brain-computer interface device, characterized in that: A low-channel EEG amplifying circuit for behind-the-ear EEG signals comprising any one of claims 1-7.

Images