System for real-time decomposition of electrophysiological signals and classification of cardiac arrhythmias using adaptive neural signal processing

Abstract

A system for real-time decomposition of electrophysiological signals and arrhythmia classification using adaptive neural signal processing, comprising: a plurality of electrodes configured to acquire electrophysiological signals from a subject; an analog input stage consisting of an instrumentation amplifier, an impedance matching circuit, and an anti-aliasing filter for processing the acquired electrophysiological signals; an analog-to-digital converter unit operationally coupled to the analog input stage and configured to digitize the processed electrophysiological signals at a programmable sampling rate; a processing unit consisting of a microcontroller and a digital signal processor operationally coupled to each other via a communication bus; and a neural processing unit operationally coupled to the processing unit.a storage unit that stores executable instructions and trained neural parameters; wherein the processing unit and the neural processing unit are configured to jointly perform an adaptive decomposition of the digitized electrophysiological signals into a variety of intrinsic signal components based on dynamically updated baseline representations, extract temporal and morphological features from the intrinsic signal components, and classify the extracted features in real time into one or more arrhythmia categories.

Description

Application area of ​​the invention: The present invention relates generally to biomedical signal processing systems and intelligent diagnostic devices, but in particular to a system and an associated machine structure configured for real-time decomposition of electrophysiological signals and automated classification of cardiac arrhythmias using adaptive neural signal processing techniques, embedded hardware architectures and dynamically trainable computational models. Background of the invention: Electrophysiological signals, particularly electrocardiographic signals, are frequently used to monitor cardiac activity and diagnose cardiac arrhythmias. Conventional arrhythmia detection systems rely on static filtering techniques, fixed feature extraction pipelines, and rule-based or flat machine learning classifiers. However, these are inherently limited in their ability to process non-stationary signal characteristics, motion artifacts, baseline deviations, and interindividual variability. Existing signal decomposition techniques, such as Fourier transform-based methods or wavelet transforms, offer only limited adaptability to dynamically changing signal morphologies and often fail to isolate clinically relevant components in real-time scenarios.Furthermore, traditional arrhythmia classification systems are limited by latency, computational overhead, and a lack of context adaptation, reducing their effectiveness in continuous monitoring environments such as wearable or ambulatory cardiac monitoring systems. These limitations necessitate the development of a system capable of adaptive signal decomposition in real time and intelligent classification, employing neural processing architectures that can dynamically adapt to signal variations and patient-specific electrophysiological signatures. The monitoring of electrophysiological signals, particularly electrocardiography (ECG), has long served as a fundamental diagnostic tool for assessing cardiac function and detecting cardiac arrhythmias. Conventional systems for analyzing such signals have traditionally relied on deterministic signal processing chains with fixed analog pre-filtering, digitization, and subsequent digital processing using linear transformations and threshold-based detection mechanisms. Early approaches primarily utilized time-domain analysis and heuristic rules to identify characteristic waveform components such as P waves, QRS complexes, and T waves. While these methods yielded acceptable results under controlled clinical conditions, they exhibit significant limitations when applied in real-world environments characterized by noise, motion artifacts, baseline drift, and interindividual variability.The inherently non-stationary and non-linear nature of electrophysiological signals further exacerbates the limitations of such traditional approaches, as fixed filters and static feature extraction methods cannot capture transient signal dynamics and morphological variations. To address some of these limitations, frequency-domain and time-frequency analysis techniques such as the Fourier transform, the short-time Fourier transform, and the wavelet transform were introduced. These methods allow for improved analysis of signal components across different frequency bands and are frequently used for noise reduction and feature extraction. However, Fourier-based approaches require a stationary signal within the analysis window and have low temporal resolution, making them unsuitable for capturing abrupt changes in heart rhythm. While wavelet-based methods improve time-frequency localization, they require careful selection of the parent wavelets and decomposition planes, which are typically predefined and not adaptable to individual signal characteristics.Therefore, these methods often cannot be applied to different patient populations and different physiological conditions, leading to suboptimal results in continuous monitoring scenarios. To address the nonlinearity and non-stationarity of electrophysiological signals, more advanced decomposition methods such as empirical mode decomposition (EMD) and its variants have been proposed. These methods decompose signals into intrinsic mode functions based on local extrema, thus offering a data-driven approach to signal representation. Despite their theoretical advantages, these methods exhibit problems such as mode mixing, noise sensitivity, and high computational complexity, which limits their applicability in real-time systems. Furthermore, the lack of a robust mathematical foundation for certain decomposition methods leads to inconsistent results and thus limits their reliability in clinical decision-making. In the field of classification, traditional arrhythmia detection systems are based on machine learning techniques such as support vector machines, k-nearest neighbors, and decision trees. These work with manually generated features derived from preprocessed signals. While these approaches have achieved moderate success, their performance depends heavily on the quality and relevance of the extracted features. Feature generation in electrophysiological signal analysis is inherently challenging due to the variability of signal morphology between patients and disease patterns. Consequently, these models often fail to capture complex temporal dependencies and subtle patterns indicative of pathological conditions. Furthermore, such classifiers typically operate offline or in batch mode, making them unsuitable for real-time applications requiring continuous monitoring and immediate response. The advent of deep learning has led to more sophisticated approaches for analyzing electrophysiological signals, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs). These models can learn hierarchical feature representations directly from raw signals, thus reducing reliance on manual feature extraction. Convolutional neural networks are used to extract spatial features, while RNNs such as long short-term memory (LSTM) are used to model temporal dependencies. Despite these advances, existing deep learning-based systems face several challenges, including high computational cost, poor interpretability, and limited adaptability to changing signal conditions.Most models are trained offline with large, annotated datasets and used statically, which limits their ability to adapt to patient-specific variations or changing physiological conditions. Another crucial limitation of existing systems lies in their inability to perform efficient real-time processing in resource-constrained environments such as portable devices. Many current implementations rely on cloud-based processing, where raw data or partially processed signals are sent to remote servers for analysis. This approach introduces latency, increases reliance on network connectivity, and raises privacy and data security concerns. Edge computing solutions have been explored to mitigate these issues; however, they often employ simplified models that sacrifice accuracy for computational efficiency. The lack of optimized hardware architectures that support complex neural processing further restricts the use of advanced techniques in real-time applications. Furthermore, existing systems typically use sequential processing pipelines, where signal acquisition, preprocessing, feature extraction, and classification are performed as separate steps. This rigid architecture limits the possibility of joint optimization across multiple steps and reduces the overall efficiency of the system. Moreover, most current solutions lack adaptive mechanisms that enable continuous learning from incoming data streams. The absence of online learning capabilities prevents the system from improving its performance over time and adapting to new or previously unknown arrhythmia patterns. Another disadvantage concerns the handling of noise and artifacts, which are ubiquitous in outpatient monitoring. Motion artifacts, electrode displacement, muscle sounds, and environmental disturbances can significantly distort electrophysiological signals and lead to misdiagnoses. Existing noise reduction methods are often not robust enough and can unintentionally remove clinically relevant signal components along with the noise. Furthermore, most systems lack the ability for context-sensitive processing to distinguish between physiological fluctuations and pathological anomalies, which limits diagnostic accuracy. Interoperability and scalability remain challenges for current electrophysiological monitoring solutions. Many systems are designed for specific hardware configurations and offer little flexibility for integration with different sensor modalities or healthcare platforms. The lack of standardized frameworks for data visualization and processing further complicates the development of scalable solutions for large-scale deployments. Moreover, existing systems often offer limited interpretability of their results, which impacts clinical acceptance and trust, particularly in critical care settings where traceability is essential. Given the aforementioned limitations, there is a need for a comprehensive system that integrates adaptive signal decomposition and intelligent classification within a unified, real-time architecture. Such a system must be able to dynamically adapt to signal variability, efficiently process noise and artifacts, and ensure accurate and interpretable arrhythmia classification. Furthermore, the system should be designed to operate in resource-constrained environments with high computational efficiency and low latency. Integrating adaptive neural signal processing with optimized hardware architectures represents a promising approach to addressing these challenges and enables the development of next-generation electrophysiological monitoring systems with improved performance, reliability, and clinical applicability. Summary of the invention: The present invention describes a system and a device for the real-time decomposition of electrophysiological signals and the classification of cardiac arrhythmias using adaptive neural signal processing. The system integrates a multi-layered signal acquisition module, an adaptive decomposition unit, a neural inference processor, and a classification and decision module in a unified hardware-software architecture. The invention further describes a machine structure with embedded biosignal acquisition circuitry, programmable processing units, and a neural coprocessor for executing adaptive learning procedures. The system utilizes the generation of dynamic basic functions, recurrent neural filtering, and attention-based feature extraction to isolate signal components corresponding to different cardiac events.Subsequently, classification is performed using deep neural architectures trained on time-varying datasets. The device is designed for continuous monitoring, low-latency processing, and real-time diagnostic output, thus enabling early detection of cardiac arrhythmias and improved clinical decision-making. The present invention primarily relates to a system and an associated device for the real-time decomposition of electrophysiological signals and for arrhythmia classification using adaptive neural signal processing. The system is capable of precisely analyzing non-stationary and non-linear cardiac signals under dynamic physiological and environmental conditions. A further objective of the invention is the development of an integrated signal processing architecture that enables the simultaneous acquisition, decomposition, feature extraction, and classification within a unified framework, thereby reducing latency and improving the overall efficiency of the system for continuous monitoring applications. A further objective of the invention is to provide an adaptive signal decomposition mechanism that dynamically learns and updates basic representations of electrophysiological signals to respond to temporal variations. This enables the precise isolation of clinically relevant waveform components such as P waves, QRS complexes, and T waves. Furthermore, the invention aims to integrate a neural signal processing module that can learn hierarchical and temporal features directly from raw or minimally preprocessed signals. This eliminates the dependence on manual feature extraction and improves the robustness of the classification across different patient populations. A further objective of the invention is to provide a real-time arrhythmia classification system that utilizes advanced neural architectures in combination with probabilistic inference models to improve diagnostic accuracy, reliability, and interpretability. The invention further aims to enable continuous online learning and model adaptation, allowing the system to personalize its performance based on patient-specific electrophysiological patterns and changing physiological conditions. A further objective of the invention is to provide a compact and energy-efficient device structure with integrated processing units, neural coprocessors, and optimized hardware interfaces. This enables its use in portable, mobile, or point-of-care monitoring systems without compromising computing power. Furthermore, the invention aims to ensure low-latency processing and immediate diagnostic feedback to enable the early detection of cardiac arrhythmias and timely clinical intervention. A further objective of the invention is to improve robustness against noise and artifacts by integrating adaptive filtering and context-sensitive signal enhancement techniques into the neural processing chain. This minimizes false-positive and false-negative results in arrhythmia detection. The invention further aims to provide secure and reliable communication functions for transmitting processed data, alerts, and diagnostic summaries to external systems, including clinical dashboards and remote monitoring platforms. A further objective of the invention is to support scalability and interoperability through a system design that is compatible with various sensor configurations, data formats, and healthcare infrastructures, thus enabling seamless integration into existing medical systems. Furthermore, the invention aims to improve the interpretability of classification results through confidence assessment and traceable inference mechanisms, thereby increasing clinical confidence and ease of use. A further objective of the invention is to provide a system that can be used in both standalone and distributed computing environments, and in which computing tasks can be selectively executed on the device or offloaded to edge or cloud platforms, while ensuring data security and system performance. Overall, the invention aims to overcome the limitations of existing electrophysiological monitoring systems by providing a comprehensive, adaptive, real-time solution for the precise detection and analysis of cardiac arrhythmias. BRIEF DESCRIPTION OF THE IMAGE These and other features, aspects and advantages of the present invention will be better understood if the following detailed description is read with reference to the accompanying drawing, in which the same symbols represent the same parts: Fig. 1 shows a block diagram of a system for real-time decomposition of electrophysiological signals and for classifying arrhythmias using adaptive neuronal signal processing. Furthermore, those skilled in the art will recognize that the elements in the drawing are simplified and not necessarily drawn to scale. For example, the flowcharts illustrate the process by highlighting the main steps to facilitate understanding of the present disclosure. With regard to the construction of the device, one or more components may be represented in the drawing by conventional symbols. The drawing may show only those specific details relevant to understanding the embodiments of the present disclosure, so as not to clutter the drawing with details that are already apparent to those skilled in the art from the description contained herein. Detailed description of the invention To facilitate understanding of the principles of the invention, reference is made below to the embodiment shown in the drawing, which is described using specific terms. It is understood, however, that this does not limit the scope of protection of the invention. Rather, modifications and further developments of the depicted system, as well as further applications of the inventive principles shown therein, are conceivable, insofar as they would normally occur to a person skilled in the art in the field of the invention. It will be clear to those skilled in the art that the foregoing general description and the following detailed description are exemplary and explanatory of the invention and are not to be understood as a limitation of it. References to “an aspect”, “another aspect”, or similar phrases in this description mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, phrases such as “in one embodiment”, “in another embodiment”, and similar expressions in this description may, but do not necessarily, all refer to the same embodiment. The terms "includes," "comprehensive," or similar expressions denote non-exclusive inclusion. Thus, a procedure or method containing a list of steps does not only include those steps but may also include further steps not explicitly listed or inherent in the procedure or method. Likewise, the statement "includes..." for one or more devices, subsystems, elements, structures, or components, without further limitations, does not preclude the existence of other devices, subsystems, elements, structures, or components. Unless otherwise defined, all technical and scientific terms used herein have the same meanings generally known to those skilled in the art in the field to which this invention belongs. The systems, methods, and examples described herein serve only for illustration and are not to be understood as limiting. Embodiments of the present disclosure are described in detail below with reference to the attached drawing. Fig. 1 shows a block diagram of a system for real-time decomposition of electrophysiological signals and arrhythmia classification using adaptive neuronal signal processing.System 100 comprises: several electrodes (102) for recording electrophysiological signals from a subject; an analog input stage (104) with an instrumentation amplifier, an impedance matching circuit, and an anti-aliasing filter for processing the recorded electrophysiological signals; an analog-to-digital converter unit (106) connected to the analog input stage, which digitizes the processed electrophysiological signals at a programmable sampling rate; a processing unit (108) with a microcontroller and a digital signal processor connected via a communication bus; a neural processing unit (110) connected to the processing unit; and a memory (112) for storing executable instructions and trained neural parameters.wherein the processing unit and the neural processing unit are configured to jointly perform an adaptive decomposition of the digitized electrophysiological signals into a multitude of intrinsic signal components based on dynamically updated basic representations, extract temporal and morphological features from the intrinsic signal components, and classify the extracted features in real time into one or more arrhythmia categories. In one embodiment, the analog input unit (104) further comprises a multi-stage amplifier circuit configured to enable programmable gain control based on signal amplitude variability, and a common-mode rejection circuit configured to suppress disturbances caused by external electromagnetic sources and impedance mismatches between electrode and skin. In one embodiment, the analog-to-digital converter unit (106) is configured to operate with a variable resolution and sampling frequency, which is determined by the processing unit based on the detected signal quality. This enables adaptive resolution control to improve signal fidelity during transient cardiac events. In one embodiment, the processing unit (108) is configured to perform a recursive filtering procedure using temporal correlation parameters derived from previously acquired signal segments, so that noise components such as motion artifacts, baseline drift and muscle interference are selectively attenuated while clinically relevant waveform features are preserved. In one embodiment, the neural processing unit (110) comprises a variety of interconnected computational layers, including convolutional processing layers configured to extract spatial features from segmented signal windows, recurrent processing layers configured to model temporal dependencies between sequential signal segments, and attention-based processing layers configured to assign dynamic weighting factors to time-frequency regions corresponding to the components of the heart waveform. In one embodiment, the adaptive decomposition of the electrophysiological signals is performed by iteratively updating a set of basis vectors stored in the memory unit. The basis vectors are modified using feedback signals generated by the neural processing unit based on the reconstruction error between the original signal and a reconstructed signal derived from the intrinsic signal components. In one embodiment, the processing unit (108) is further configured to segment the digitized electrophysiological signals into overlapping time windows and to align the segmented signals based on recognized cardiac cycle landmarks in order to ensure temporal consistency in feature extraction. In one embodiment, the classification of arrhythmia categories is performed using a probabilistic inference procedure implemented in the neural processing unit. The classification outputs are linked to confidence values ​​derived from estimating the a posteriori probability. In one embodiment, the system further comprises a communication unit configured to transmit classified arrhythmia data, alarm signals, and processed electrophysiological data to an external device via a wireless communication protocol with encryption enabled for data security. In one embodiment, the system further comprises an energy management unit with a rechargeable energy storage element, a voltage regulation circuit, and a dynamic power distribution controller configured to adjust the power consumption of the processing unit and the neural processing unit based on the computational load. The described system is implemented as an integrated array of physical electronic hardware components that acquire, condition, convert, process, and transmit electrophysiological signals. The electrodes form physical, conductive interfaces for biosignal acquisition, while the analog input stage is implemented with discrete and / or integrated circuit components such as instrumentation amplifiers, impedance matching networks, filters, and multi-stage amplification circuits, which electrically manipulate the signal amplitude and frequency characteristics. The analog-to-digital converter unit is a hardware converter that transforms continuous analog voltages into discrete digital values ​​using quantization circuits with selectable sampling rates and resolutions.The processing unit and neural processing unit are implemented as semiconductor-based computing devices, including microcontroller circuits, dedicated signal processing cores, and layered neural accelerator circuits. These consist of interconnected logic gates, arithmetic units, and memory registers configured for real-time decomposition, feature extraction, and classification via physical data paths. The memory unit comprises non-volatile memory hardware that stores parameter values ​​and executable configurations electronically. Supporting subsystems, such as the communication unit and power management unit, are also implemented using high-frequency transceivers, encryption-capable hardware modules, voltage regulators, energy storage elements, and power control circuits. The system according to claim 1 operates with a tightly integrated sequence of signal acquisition, adaptive decomposition, feature extraction, and classification processes, implemented in a coordinated hardware architecture comprising a processing unit and a neural processing unit. During operation, the electrophysiological signals acquired via multiple electrodes are first processed by the analog input stage. This process involves amplification, impedance matching, and anti-aliasing to ensure the signals operate within an optimal dynamic range. The processed signals are then digitized by the analog-to-digital converter unit at a sampling rate dynamically selected by the processing unit based on the captured signal variability. This allows for higher sampling resolution during rapid cardiac events and lower resolution during stable rhythm segments, thus optimizing computational efficiency. Following digitization, the processing unit initiates a recursive filtering procedure that utilizes the temporal correlation of successive signal segments to attenuate noise components. This procedure employs a predictive filtering approach, using previously processed signal segments to estimate the expected signal behavior. Deviations from the estimated behavior are selectively suppressed unless they exhibit features consistent with physiological waveform transitions. Baseline deviations are corrected through adaptive trend correction using locally estimated means, while motion artifacts are reduced by identifying abrupt amplitude fluctuations that do not correspond to learned cardiac cycle patterns.The resulting filtered signal is then segmented into overlapping time windows, and each window is aligned using identified reference points corresponding to the cardiac cycle markers. This ensures consistent temporal positioning of the waveform components across successive segments. After preprocessing, the system performs adaptive signal decomposition using an iterative basis learning algorithm jointly implemented by the processing unit and the neural processing unit. First, a set of basis vectors stored in the memory unit is used to approximate the input signal as a weighted combination of intrinsic signal components. The neural processing unit evaluates the reconstruction error between the original signal and the signal reconstructed from the current basis representation. Based on this error, the basis vectors are updated through a feedback-driven optimization process, suppressing components that contribute minimally to signal reconstruction and amplifying those that capture important morphological features.This adaptive update process continues iteratively, so that the system converges to a set of basic representations that are optimally tailored to the current electrophysiological signal characteristics. This decomposes the signal into multiple intrinsic components, each corresponding to different physiological or noise-related features. The decomposed signal components are then processed by the neural processing unit through a sequence of computational layers configured for hierarchical feature extraction. Convolutional processing layers operate with segmented signal windows to identify localized spatial patterns corresponding to waveform shapes, including abrupt transitions and gradual rises associated with various cardiac events. These extracted features are then passed to recurrent processing layers, which model temporal dependencies between sequential segments by maintaining a state representation that captures rhythm continuity and temporal variations. This temporal modeling allows the system to distinguish between transient anomalies and sustained arrhythmia patterns.An attention-based processing mechanism is further applied to dynamically assign weights to specific time-frequency ranges within the signal, thereby highlighting features associated with clinically relevant waveform components while suppressing irrelevant or redundant information. The processing unit aggregates the outputs of the neural processing unit to create a multidimensional feature representation that integrates amplitude characteristics, time intervals, frequency distributions, and morphological descriptors of the intrinsic signal components. This feature representation is normalized and structured into a format suitable for classification. Subsequently, the neural processing unit performs arrhythmia classification using a probabilistic inference procedure, assigning the feature representation to predefined arrhythmia categories. The classification process generates both categorical outputs and associated confidence scores based on the match of the feature patterns with the learned representations stored in memory. A key aspect of the system lies in its ability to learn online. The neural processing unit continuously updates its internal parameters based on incoming signal data. This update process is controlled by a sophisticated adaptation mechanism that selectively integrates new information without destabilizing previously learned patterns. The system maintains a balance between stability and adaptability by regulating parameter updates based on confidence thresholds and signal consistency metrics. This allows the system to gradually adjust its performance to the electrophysiological characteristics of the subject, thereby improving classification accuracy over time. The processing unit also monitors signal quality indicators such as signal-to-noise ratio, waveform consistency, and electrode contact stability. Upon detecting a deterioration in signal quality, the system initiates recalibration procedures, including gain adjustment, reconfiguration of filter parameters, and reinitialization of the basis vectors used for decomposition. This ensures the robustness of the method under varying physiological and environmental conditions. Furthermore, the system performs feature fusion across different timescales by integrating information from various temporal resolutions. Short-term signal segments are analyzed to capture rapid events such as premature contractions, while longer-term segments are processed to identify arrhythmias such as atrial fibrillation. The fusion of these features across different timescales enables a comprehensive characterization of cardiac activity, thereby increasing the reliability of arrhythmia detection. The classified results, along with confidence levels and summarized signal processing values, are transmitted via the communication unit to external devices for visualization or further analysis. The entire system is optimized for real-time execution, with parallel processing distributed between the processing unit and the neural processing unit to minimize latency. Through the described sequence of adaptive decomposition, hierarchical feature extraction, probabilistic classification, and continuous learning, the system achieves precise and efficient real-time analysis of electrophysiological signals, thus enabling the reliable detection of cardiac arrhythmias under a wide range of operating conditions. In a preferred embodiment, the invention comprises a device-integrated system with a signal acquisition system for acquiring electrophysiological signals from a subject using multiple electrodes arranged in a configurable topology. The acquisition system includes an analog input stage with instrumentation amplifiers, impedance matching networks, and anti-aliasing filters, followed by an analog-to-digital converter unit for digitizing the acquired signals at a programmable sampling rate. The digitized signals are transmitted to an embedded processing unit comprising a microcontroller, a digital signal processor, and a neural processing unit, interconnected via a high-speed data bus. The system further includes an adaptive signal decomposition engine, implemented as a hybrid computing module that combines data-driven and model-based decomposition techniques. The decomposition engine utilizes an adaptive basic learning mechanism, in which a set of basic functions is iteratively optimized using neural network feedback to represent the input signal as a superposition of intrinsic mode components. The system employs a recurrent neural filter architecture configured to dynamically suppress noise and artifacts by learning temporal dependencies in the signal. The decomposition process also incorporates a time-frequency attention mechanism that selectively amplifies signal components corresponding to clinically relevant cardiac features such as P waves, QRS complexes, and T waves. The neural signal processing module comprises a deep neural network architecture with convolutional layers for extracting spatial features, recurrent layers for temporal modeling, and attentional layers for context-dependent feature weighting. The neural network operates adaptively, with model parameters continuously updated via online learning based on incoming data streams. The system also includes a feature fusion layer that combines the decomposed signal components into a multidimensional feature representation. This representation is then processed by a classification module. The classification module includes a probabilistic inference engine that categorizes the processed signals into predefined arrhythmia categories, including atrial fibrillation, ventricular tachycardia, bradycardia, and ventricular premature beats. Classification is performed using a hybrid model that combines deep neural networks and probabilistic graphical models to improve robustness and interpretability. The system also includes a decision support module that generates alerts, diagnostic summaries, and confidence scores, transmitting them wirelessly to external devices. The device structure of the invention comprises a compact, portable housing that encloses the signal acquisition circuitry, the processing units, the power management module, and the communication interfaces. The housing is ergonomically designed to allow continuous monitoring and includes devices for electrode attachment and user interaction. The power management module comprises a battery, voltage regulation, and energy-efficient processing controls for extended operation. The device further includes a thermal management system for dissipating the heat generated by the high-performance processing components. In an advanced implementation, the system utilizes a distributed processing architecture, where portions of the neural processing are offloaded to an edge or cloud computing platform. The device features a secure communication module with encryption protocols to ensure data integrity and privacy. Furthermore, the system supports firmware updates and model retraining via remote interfaces, enabling continuous improvement in classification accuracy. The invention also provides mechanisms for calibration and personalization, whereby the system adapts to individual patient characteristics by optimizing decomposition parameters and neural model weights based on historical data. The adaptive neural signal processing framework enables the system to process nonlinear, non-stationary electrophysiological signals with high accuracy and low latency. The presented system and device thus offer a comprehensive solution for the real-time decomposition of electrophysiological signals and the classification of arrhythmias. By integrating adaptive neural processing, advanced signal decomposition techniques, and optimized hardware architectures, the limitations of existing technologies are overcome. The present invention relates to the fields of biomedical engineering, medical signal processing, and intelligent diagnostic systems, in particular a system and a device for the real-time processing of electrophysiological signals for applications in cardiac monitoring. The invention specifically relates to adaptive neural signal processing methods implemented in embedded hardware architectures for the decomposition of non-stationary electrophysiological signals and for the automated classification of cardiac arrhythmias. The described system integrates advanced signal acquisition circuits, digital processing units, and neural processing units to enable continuous, low-latency analysis of bioelectrical signals in portable, mobile, and clinical monitoring environments. The drawing and the preceding description illustrate embodiments. Those skilled in the art will recognize that one or more of the described elements can be combined to form a single functional element. Alternatively, certain elements can be divided into several functional elements. Elements of one embodiment can be added to another. For example, the process flows described here can be modified and are not limited to the manner described herein. Furthermore, the actions of a flowchart need not be performed in the sequence shown; nor do all actions necessarily need to be carried out. Actions that do not depend on other actions can be performed in parallel with the other actions. The scope of protection of the embodiments is in no way limited by these specific examples. Numerous variations, whether explicitly stated in the description or not, such as...Differences in structure, dimensions, and materials are possible. The scope of protection of the embodiments is at least as comprehensive as described by the following claims. The advantages, other benefits, and problem solutions have been described above with reference to specific embodiments. However, the advantages, benefits, problem solutions, and any components that can effect or enhance an advantage, benefit, or solution are not to be construed as critical, necessary, or essential features or components of the claims. REFERENCES 100 A system for real-time decomposition of electrophysiological signals and classification of cardiac arrhythmias using adaptive neural signal processing. 102 Variety of electrodes 104 Analog front-end unit 106 Analog-to-digital converter unit 108 Processing unit 110 Neural processing unit 112 Storage unit

Claims

A system for real-time decomposition of electrophysiological signals and arrhythmia classification using adaptive neural signal processing, comprising: a plurality of electrodes configured to acquire electrophysiological signals from a subject; an analog input stage consisting of an instrumentation amplifier, an impedance matching circuit, and an anti-aliasing filter for processing the acquired electrophysiological signals; an analog-to-digital converter unit operationally coupled to the analog input stage and configured to digitize the processed electrophysiological signals at a programmable sampling rate; a processing unit consisting of a microcontroller and a digital signal processor operationally coupled to each other via a communication bus; and a neural processing unit operationally coupled to the processing unit.a storage unit that stores executable instructions and trained neural parameters; wherein the processing unit and the neural processing unit are configured to jointly perform an adaptive decomposition of the digitized electrophysiological signals into a variety of intrinsic signal components based on dynamically updated baseline representations, extract temporal and morphological features from the intrinsic signal components, and classify the extracted features in real time into one or more arrhythmia categories. System according to claim 1, wherein the analog input stage further comprises a multi-stage amplifier circuit configured to enable programmable gain control based on signal amplitude variability, and a common-mode rejection circuit configured to suppress interference caused by external electromagnetic sources and impedance mismatches between electrode and skin. System according to claim 1, wherein the analog-to-digital converter unit is configured to operate with a variable resolution and sampling frequency determined by the processing unit on the basis of the detected signal quality, thereby enabling adaptive resolution control to improve signal fidelity in transient cardiac events. System according to claim 1, wherein the processing unit is configured to perform a recursive filtering procedure using temporal correlation parameters derived from previously acquired signal segments, such that noise components including motion artifacts, baseline drift and muscle interference are selectively attenuated while clinically relevant waveform features are preserved. System according to claim 1, wherein the neural processing unit comprises a plurality of interconnected computational layers, including convolutional processing layers configured to extract spatial features from segmented signal windows, recurrent processing layers configured to model temporal dependencies between sequential signal segments, and attention-based processing layers configured to assign dynamic weighting factors to time-frequency ranges corresponding to the components of the heart waveform. System according to claim 1, wherein the adaptive decomposition of the electrophysiological signals is performed by iteratively updating a set of basis vectors stored in memory, wherein the basis vectors are modified using feedback signals generated by the neural processing unit on the basis of the reconstruction error between the original signal and a reconstructed signal derived from the intrinsic signal components. System according to claim 1, wherein the processing unit is further configured to segment the digitized electrophysiological signals into overlapping time windows and to align the segmented signals based on recognized cardiac cycle features to ensure temporal consistency in feature extraction. System according to claim 1, wherein the classification of arrhythmia categories is carried out by means of a probabilistic inference procedure implemented in the neuronal processing unit, wherein the classification outputs are linked to confidence values ​​derived from the estimation of the a posteriori probability. System according to claim 1, wherein the system further comprises a communication unit configured to transmit classified arrhythmia data, alarm signals and processed electrophysiological data to an external device via a wireless communication protocol with encryption enabled for data security. System according to claim 1, wherein the system further comprises an energy management unit comprising a rechargeable energy storage element, a voltage regulation circuit and a dynamic power distribution controller configured to adjust the power consumption of the processing unit and the neural processing unit based on the computational load.

Images