EEG pre-amplification circuit and brain-computer interface equipment

Abstract

The invention relates to the technical field of emergency protection circuits, and provides an EEG pre-amplification circuit and brain-computer interface equipment. The buffers are connected in parallel, and the corresponding input ends of the buffers are connected with the electrode sensors; the buffer is a four-channel unity gain buffer; the input end of the electrostatic protection diode array is connected with the positive electrode and the negative electrode of a preset power supply, and the input end is further connected between the corresponding input ends of the buffers and the electrode sensor; the first filtering capacitor is connected with the first voltage stabilizing capacitor in parallel, the second filtering capacitor is connected with the second voltage stabilizing capacitor in parallel, the input ends of the first filtering capacitor and the first voltage stabilizing capacitor are connected with the positive electrode of a preset power supply, and the input ends of the first filtering capacitor and the first voltage stabilizing capacitor are connected with a preset grounding end; the input ends of the first filtering capacitor and the first voltage stabilizing capacitor are connected with the negative electrode of the preset power supply, and the input ends are connected with the preset grounding end, so that the power supply noise corresponding to the preset power supply is minimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of emergency protection circuits, and particularly relates to an EEG preamplifier circuit and a brain-computer interface device. Background Art

[0002] EEG (electroencephalogram) is a weak bioelectrophysiological signal, characterized by low amplitude, low frequency band, and susceptibility to interference. Most of the existing EEG acquisition analog front-end circuit designs adopt a single-ended signal link, that is, the electrode and the amplifier are directly connected by a lead wire, and the collected EEG signal is amplified by the amplifier and then subjected to analog-to-digital conversion.

[0003] In the existing EEG acquisition analog front-end circuit design, due to the problem that the impedance of the brain-computer amplifier itself is not high enough, the amplitude of the EEG signal obtained from the human scalp is low and is easily covered by noise, which is not conducive to extracting high-quality EEG signals. In addition, since the EEG signal obtained from the human scalp is very weak itself and is connected to the EEG amplifier after passing through a relatively long wire, the signal is extremely vulnerable to various external factors during the transmission process, thus greatly reducing the quality of the EEG signal. Especially for the wearable EEG signal acquisition usage scenario, dry electrode sensors that are usually easy to wear and have a long service life are used. Due to the high contact impedance of the dry electrodes and the relatively complex and frequent motion artifacts and noise in the daily wearing scenario, it is impossible to collect high-quality EEG signals. In short, the existing EEG signal acquisition analog front-end circuit design has problems such as low signal-to-noise ratio and poor anti-interference ability. In addition, the noise and power consumption problems introduced in the prior art to suppress signal attenuation also need to be solved urgently. Summary of the Invention

[0004] The present application provides an EEG preamplifier circuit and a brain-computer interface device, aiming to solve the problems in the existing EEG acquisition analog front-end circuit design, that is, due to the problem that the impedance of the brain-computer amplifier itself is not high enough, the amplitude of the EEG signal obtained from the human scalp is low and is easily covered by noise, which is not conducive to extracting high-quality EEG signals. In addition, since the EEG signal obtained from the human scalp is very weak itself and is connected to the EEG amplifier after passing through a relatively long wire, the signal is extremely vulnerable to various external factors during the transmission process, thus greatly reducing the quality of the EEG signal.

[0005] In a first aspect, an embodiment of the present application provides an EEG preamplifier circuit, including:

[0006] An electrode sensor for collecting EEG signals;

[0007] Multiple buffers connected in parallel, with the input ends of the multiple buffers corresponding to the electrode sensor, for buffering or preliminarily amplifying the electroencephalogram (EEG) signals collected by the electrode sensor and transmitting the EEG signals to a preset instrumentation amplifier or signal processing circuit; the buffer is a four-channel unity-gain buffer;

[0008] An electrostatic protection diode array, which includes a first input end, a second input end, and a third input end. The first input end and the second input end are respectively connected to the positive and negative poles of a preset power supply, and the third input end is connected between the input ends corresponding to the multiple buffers and the electrode sensor for protecting the buffers and the preset power supply; the preset power supply is used to supply power to the buffers;

[0009] A first filter capacitor, a second filter capacitor, a first voltage-stabilizing capacitor, and a second voltage-stabilizing capacitor. The first filter capacitor and the first voltage-stabilizing capacitor are connected in parallel, and the second filter capacitor and the second voltage-stabilizing capacitor are connected in parallel. The input ends of the first filter capacitor and the first voltage-stabilizing capacitor are connected to the positive pole of the preset power supply and the preset grounding end, and the input ends of the first filter capacitor and the first voltage-stabilizing capacitor are connected to the negative pole of the preset power supply and the preset grounding end, for minimizing the power supply noise corresponding to the preset power supply.

[0010] This EEG preamplification circuit is designed around EEG signal acquisition, preprocessing, and protection. The core modules include: Electrode sensor: As the signal input end, it directly contacts the human scalp to collect weak EEG signals (in the microvolt level), which is the signal source of the entire circuit. Parallel buffer array: Adopts a parallel design with multiple (specifically four-channel unity-gain buffers), and the input ends are connected to the electrode sensor. The unity-gain buffer has the characteristics of high input impedance and low output impedance. After parallel connection, the input impedance is further increased (close to infinity), reducing the attenuation of EEG signals; at the same time, the buffer buffers or preliminarily amplifies the signal (the unity gain means the amplification factor is 1, which does not change the signal amplitude but enhances the driving ability), facilitating subsequent long-distance transmission or access to the instrumentation amplifier. Electrostatic protection diode array: It contains three input ends. The first and second input ends are respectively connected to the positive and negative poles of the preset power supply, and the third input end is connected in series between the buffer and the electrode sensor. Its function is to limit the overvoltage generated by electrostatic discharge (ESD) within the power supply voltage range through the clamping effect of the diode, protecting the buffer from electrostatic impact and preventing device damage. Power supply filtering and voltage stabilization module: Consists of two groups of capacitors connected in parallel. The first filter capacitor (high-frequency filtering) and the first voltage-stabilizing capacitor (low-frequency energy storage, such as electrolytic capacitor) are connected in parallel and then connected to the power supply positive pole and the grounding end, and the second group is connected to the power supply negative pole and the grounding end in the same way. Through the cooperation of high- and low-frequency capacitors, high-frequency noise and ripples in the power supply are filtered, the power supply voltage is stabilized, and the interference of power supply noise on the front-end signal is reduced.

[0011] The EEG signal collected by the electrode sensor first enters the buffer parallel network through the third input terminal of the electrostatic protection diode array, reduces signal attenuation and enhances driving capability through the high-impedance buffer, and is then output to the instrument amplifier or subsequent processing circuit.

[0012] The preset power supply is used to power the buffer, and its positive and negative poles are grounded through a filter capacitor network to ensure the purity of the power supply; the electrostatic protection diode array is connected across the positive and negative poles of the power supply and the signal input terminal to form a bidirectional overvoltage protection.

[0013] The provided circuit has the following beneficial effects:

[0014] 1. Solve the high impedance matching problem and improve the signal acquisition quality: Parallel buffers improve input impedance: A single buffer already has high input impedance. After parallel connection, the equivalent input impedance is further improved (close to the parallel value of the input impedance of each buffer), which consumes almost no signal current of the electrode sensor, avoids signal attenuation caused by insufficient amplifier impedance, and ensures that weak EEG signals (microvolt level) enter the subsequent circuit with minimal loss, reducing the risk of noise coverage. Unit gain buffering avoids signal distortion: The buffer only buffers the signal (does not amplify the amplitude), maintains the original characteristics of the EEG signal, and reduces the output impedance at the same time, enhancing the signal driving ability, which is convenient for resisting signal attenuation caused by wire impedance during long-distance transmission.

[0015] 2. Enhance anti-interference capability and reduce noise impact: Power supply filtering reduces noise introduction: The positive and negative power supplies are grounded through high-frequency filter capacitors (such as ceramic capacitors) and low-frequency voltage stabilizing capacitors (such as electrolytic capacitors) respectively, forming a "full-band filtering of high and low-frequency noise" mechanism, suppressing power supply fluctuations and external electromagnetic interference from entering the front-end circuit through power supply coupling, and cutting off the noise source from the power supply end. Electrostatic protection circuit reliability: The electrostatic protection diode array forms a "voltage clamping barrier" between the electrode sensor (directly in contact with the human body and susceptible to electrostatic shock) and the buffer. When static electricity occurs, the diode conducts and guides the overvoltage to the positive and negative poles of the power supply, avoiding the buffer input end from being subjected to high-voltage shock, protecting the core components from damage, and improving the circuit durability.

[0016] 3. Adapt to long-distance signal transmission and ensure signal integrity: The low output impedance characteristics of the buffer enable the EEG signal to have a stronger driving ability after buffering. Even if it is transmitted to the subsequent processing circuit through a long wire, it can reduce the signal attenuation and distortion caused by wire resistance and distributed capacitance, solve the problem of "long-distance transmission is susceptible to external interference", and ensure the signal quality received at the back end.

[0017] 4. Compact structure design and strong compatibility: By paralleling four-channel unity-gain buffers, it can flexibly adapt to the multi-channel EEG acquisition requirements. Meanwhile, the modular design (independent modules for buffers, protection, and filtering) facilitates circuit integration and debugging, reduces the design complexity, and is suitable for the miniaturized and portable applications of brain-computer interface devices.

[0018] In summary, through the three core designs of high-input-impedance buffering, power supply noise suppression, and electrostatic protection, this circuit specifically solves the signal quality problems caused by insufficient impedance, noise interference, and transmission loss in the traditional EEG acquisition front-end, provides high-quality original EEG signals for subsequent signal amplification and processing, and significantly improves the acquisition accuracy and reliability of brain-computer interface devices.

[0019] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Brief Description of the Drawings

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0021] Figure 1 is the circuit schematic diagram of the EEG preamplifier circuit provided by an embodiment of this application;

[0022] Figure 2 is the principle schematic diagram of the EEG preamplifier circuit provided by an embodiment of this application;

[0023] Figure 3 is the structural schematic block diagram of the brain-computer interface device provided by an embodiment of this application;

[0024] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Detailed Description of the Embodiments

[0025] The following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the drawings in the embodiments of this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

[0026] The flowcharts shown in the accompanying drawings are merely illustrative examples, and do not necessarily include all the content and operations / steps, nor are they necessarily executed in the order described. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may change according to the actual situation.

[0027] 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 identical or similar items with basically the same functions and roles. Those skilled in the art can understand that the terms "first", "second", etc. do not limit the quantity and execution order, and the terms "first", "second", etc. do not necessarily mean different.

[0028] It should be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0029] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] The following will describe in detail some embodiments of the present application with reference to the accompanying drawings. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0031] EEG (electroencephalogram) is a weak bioelectrophysiological signal, characterized by low amplitude, low frequency band and susceptibility to interference. Most of the existing EEG acquisition analog front-end circuit designs use a single-ended signal link, that is, the electrode and the amplifier are directly connected by a lead wire, and the collected EEG signal is amplified by the amplifier and then subjected to analog-to-digital conversion.

[0032] In the existing EEG acquisition analog front-end circuit design, due to the problem that the impedance of the brain-computer amplifier itself is not high enough, the amplitude of the EEG signal obtained from the human scalp is low and is easily covered by noise, which is not conducive to extracting high-quality EEG signals. In addition, since the EEG signal obtained from the human scalp is very weak in itself and is connected to the EEG amplifier after a relatively long line, the signal is extremely vulnerable to various external factors during the transmission process, thus greatly reducing the quality of the EEG signal. Especially for the wearable EEG signal acquisition use scenario, dry electrode sensors that are easy to wear and have a long service life are usually used. Due to the high contact impedance of the dry electrodes and the presence of complex and frequent motion artifacts and noise in the daily wearing scenario, it is impossible to collect high-quality EEG signals. In short, the existing EEG signal acquisition analog front-end circuit design has problems such as low signal-to-noise ratio and poor anti-interference ability. In addition, the problems of noise and power consumption introduced in the existing technology to suppress signal attenuation also need to be solved urgently.

[0033] To solve the above problems, please refer to Figures 1 to 2 , this application provides an EEG preamplification circuit, including: an electrode sensor 10 ( Figure 1 Only 1 electrode sensor is shown, Figure 2 in the case of showing multiple), the electrode sensor is used to collect EEG signals; a plurality of buffers 20 connected in parallel (one buffer corresponds to one electrode sensor), the input ends of the plurality of buffers are connected to the electrode sensor, and are used to buffer or preliminarily amplify the EEG signals collected by the electrode sensor and transmit the EEG signals to a preset instrumentation amplifier or signal processing circuit; the buffer is a four-channel unity-gain buffer; an electrostatic protection diode array 30, the electrostatic protection diode array includes a first input end, a second input end and a third input end, the first input end and the second input end are respectively connected to the positive and negative electrodes of a preset power supply, and the third input end is connected between the input ends corresponding to the plurality of buffers and the electrode sensor, and is used to protect the buffer and the preset power supply; the preset power supply is used to supply power to the buffer; a first filter capacitor 40, a second filter capacitor 50, a first voltage-stabilizing capacitor 60 and a second voltage-stabilizing capacitor 70 (constitute a voltage-stabilizing / filtering circuit in Figure 2 ), the first filter capacitor and the first voltage-stabilizing capacitor are connected in parallel, the second filter capacitor and the second voltage-stabilizing capacitor are connected in parallel, the input ends of the first filter capacitor and the first voltage-stabilizing capacitor are connected to the positive electrode of the preset power supply, the input ends are connected to the preset grounding end, the input ends of the first filter capacitor and the first voltage-stabilizing capacitor are connected to the negative electrode of the preset power supply, and the input ends are connected to the preset grounding end, and are used to minimize the power supply noise corresponding to the preset power supply.

[0034] Specifically, the EEG preamplifier circuit proposed in this application addresses the deficiencies of the existing single-ended signal link. It improves the signal quality through three core technologies: high-input impedance buffering, electrostatic protection, and power supply noise suppression. The core components and their functions are as follows:

[0035] 1. Electrode sensor: Function: Directly contact the human scalp to collect weak EEG signals (amplitude about 10 - 200 μV, frequency 0.5 - 100 Hz). Adapted scenarios: Especially for dry electrode sensors (such as metal electrodes, micro-needle electrodes), it solves the problem of signal attenuation caused by their high contact impedance (up to 10 - 100 kΩ).

[0036] 2. Parallel buffer:

[0037] Structure: Four-channel unity-gain buffers are connected in parallel (such as operational amplifiers like LMV324, AD8554 configured in voltage follower mode). The input terminals of each buffer are commonly connected to the electrode sensor, and the output terminals are aggregated to the subsequent instrumentation amplifier.

[0038] By paralleling the input impedance of a single buffer (such as up to 10^12 Ω for FET-type operational amplifiers), the equivalent input impedance is close to the single-channel value (because paralleling does not reduce the high-impedance characteristic), reducing the loading effect on the electrode signal and avoiding signal amplitude attenuation. By isolating the high-impedance signal of the electrode from the subsequent circuit, it reduces the influence of the subsequent noise on the front end, and at the same time enhances the signal driving ability and reduces signal distortion in long-line transmission. Due to the low noise figure of the unity-gain buffer itself, the equivalent input noise current is reduced after paralleling (the root mean square of the noise currents is superimposed), improving the signal-to-noise ratio.

[0039] 3. Electrostatic protection diode array:

[0040] Structure: It includes three input terminals. The first and second input terminals are respectively connected to the positive and negative poles of a preset power supply (such as ±3V or +5V), and the third input terminal is connected in series between the electrode sensor and the input terminal of the buffer.

[0041] When an electrostatic pulse appears at the electrode end (such as an ESD event, with a voltage up to several thousand volts), the diode array clamps the overvoltage to within ±0.7V of the power supply voltage through forward conduction, avoiding breakdown of the input stage of the buffer. The forward-biased diode conducts to the positive pole of the power supply, and the reverse-biased diode conducts to the negative pole of the power supply, forming an electrostatic discharge path to protect the precision buffer circuit.

[0042] 4. Power supply filtering and voltage regulation module:

[0043] Structure: The first filter capacitor is connected in parallel with the first voltage regulation capacitor. The first filter capacitor is a high-frequency ceramic capacitor (such as 100 nF) to filter out high-frequency noise of the power supply. The first voltage regulation capacitor is a low-frequency electrolytic capacitor (such as 10 μF) to suppress power supply ripple.

[0044] The second filtering capacitor is connected in parallel with the second voltage stabilizing capacitor: symmetrically connect the negative pole of the power supply and the ground. If it is a single power supply system, the negative pole is grounded to form a power supply loop decoupling. Through the combination of high and low frequency capacitors, the power supply noise is suppressed to the microvolt level, avoiding the interference of power supply fluctuations to the high-sensitivity buffer.

[0045] The corresponding signal link is: the output end of the electrode sensor → the third input end of the electrostatic protection diode array → the parallel buffer (the input ends of 4 channels are connected in parallel) → the aggregation of the buffer output ends → the subsequent instrumentation amplifier (such as INA128, differential amplification to suppress common-mode noise). The non-inverting input ends of the 4 buffers are commonly connected to the electrode signal, the inverting input end is short-circuited with the output end (unity gain mode), and the output ends are connected to the subsequent circuit through the same point.

[0046] The preset power supply uses a low-noise LDO (low dropout regulator, such as TPS76833) to provide ±3.3V or +5V power supply, ensuring that the power supply ripple ≤ 10μV. The diode array can select an integrated ESD protection device (such as BSS138) or a discrete Schottky diode (forward voltage drop ≤ 0.4V, reverse breakdown voltage ≥ 20V). A 1-10kΩ resistor is connected in series at the third input end to limit the current and prevent the diode from overloading.

[0047] By placing the buffer close to the electrode interface, the input trace is shortened, reducing the attenuation of high-frequency signals by the parasitic capacitance (<10pF). The power supply filtering capacitor is adjacent to the power supply pin of the buffer (distance <5mm), and a 0402 package high-frequency capacitor is used to reduce the trace inductance.

[0048] The corresponding key component selection can refer to the following table:

[0049]

[0050] By using a parallel buffer, the equivalent input impedance is increased to more than 10^13Ω, which is much higher than the human scalp-electrode contact impedance (10-100kΩ), avoiding signal voltage division attenuation (attenuation rate <0.1%), and ensuring that the EEG signal completely enters the subsequent amplification. Compared with the traditional single-ended amplifier (input impedance about 10^9Ω), the signal amplitude is increased by 50%-80%, solving the problem of signal weakening caused by the high contact impedance of dry electrodes.

[0051] Through the combination of high and low frequency capacitors, the power supply ripple is suppressed below 5μV, which is better than the traditional single-capacitor filtering (ripple about 50μV), avoiding the coupling of power supply noise to the signal chain. The ESD protection response time <1ns, and it can withstand ±8kV contact discharge, protecting the circuit from human static electricity damage in the wearable scenario (the traditional circuit is easily damaged by static electricity breakdown of the buffer without protection). The high input impedance reduces the modulation effect of the change in contact impedance caused by electrode movement on the signal. Combined with the subsequent differential amplification, the amplitude of the motion artifact is reduced by more than 60%.

[0052] The power consumption of a single four-channel buffer is ≤100 μA (such as AD8554), which is suitable for battery power supply (such as a 3.7V lithium battery with a battery life of up to 24 hours). The integrated four-channel buffer and chip ESD device have a circuit board area of ≤2 cm², meeting the compact requirements of head-mounted and ear-mounted devices.

[0053] In some embodiments, it further includes: a differential-mode cancellation capacitor, which is connected between the input ends of the first filter capacitor and the first voltage-regulating capacitor and the input ends of the second filter capacitor and the second voltage-regulating capacitor; wherein, the capacitance value of the differential-mode cancellation capacitor is the same as that of the first voltage-regulating capacitor or the second voltage-regulating capacitor.

[0054] In the power supply filtering module, the differential-mode cancellation capacitor (C_dm) is connected in parallel between the positive filtering branch (the parallel connection end of the first filter capacitor and the first voltage-regulating capacitor) and the negative filtering branch (the parallel connection end of the second filter capacitor and the second voltage-regulating capacitor) of the preset power supply, that is, it is connected across the positive and negative power supply input ends. The capacitance value of the capacitor is designed to be the same as that of the first voltage-regulating capacitor or the second voltage-regulating capacitor (for example, both are 4.7 μF), and electrolytic capacitors or tantalum capacitors are used, and the withstand voltage value is 20% higher than the power supply voltage (such as selecting a withstand voltage of 3.3V when the power supply is 2.5V).

[0055] The connection method is: power supply positive pole → first filter capacitor and first voltage-regulating capacitor → C_dm → second filter capacitor and second voltage-regulating capacitor → power supply negative pole (or ground), forming a low-impedance path for differential-mode noise. Differential-mode noise refers to high-frequency common-mode interference existing between the positive and negative poles of the power supply (such as switching power supply ripple, radio frequency coupling noise), which is directly bypassed through C_dm to avoid entering the buffer power supply pins.

[0056] Traditional filtering circuits only process common-mode noise (the in-phase noise of the positive and negative power supply terminals to ground), while differential-mode noise will be directly coupled to the buffer power supply terminal, resulting in high-frequency spikes in the output signal. C_dm capacitively shorts differential-mode noise (especially in the 10 kHz - 1 MHz frequency band), increasing the differential-mode rejection ratio by more than 30 dB and reducing the power supply noise spectral density from 50 μV / √Hz to below 5 μV / √Hz.

[0057] Having the same capacitance value as the voltage-regulating capacitor ensures that both high-frequency and low-frequency noises are effectively filtered. For example, the capacitive reactance of a 4.7 μF capacitor to low-frequency differential-mode ripple below 100 Hz is ≤3 Ω, and the capacitive reactance to high-frequency noise of 1 MHz is ≤30 mΩ, forming full-band noise suppression and avoiding gain drift in the buffer due to power supply fluctuations (the drift coefficient drops from 10 μV / ℃ to 1 μV / ℃).

[0058] Improve the stability of the post-stage circuit: A clean power supply input reduces the offset voltage drift of the buffer, and the unity-gain accuracy is improved from 99.5% to 99.9%, ensuring that the EEG signal is not distorted during the buffering process. In particular, the suppression effect on the baseline drift of ultra-low-frequency signals below 0.5 Hz (such as δ waves) is significant.

[0059] In some embodiments, the input impedance of the four-channel unity-gain buffer is 10 TΩ, and the input bias current ≤ 2 pA, so that the buffer can process high-impedance source signals; the high-impedance source signals include at least any one of electrocardiogram signals, electroencephalogram signals, and electromyogram signals.

[0060] By selecting a four-channel unity-gain buffer chip (such as AD8554, with 4 independent buffers integrated inside), the parameters of a single channel are: input impedance 10 TΩ (10^13 Ω, using a FET input stage, gate leakage current ≤ 2 pA), input bias current ≤ 2 pA (2×10^-12 A), unity-gain bandwidth 1 MHz (meeting the 100 Hz bandwidth requirement of EEG signals). Each buffer is configured as a voltage follower (the inverting input terminal is shorted to the output terminal, and the non-inverting input terminal is connected to the electrode signal). The input terminals of the 4 channels are connected in parallel to the electrode sensor, and the output terminals are aggregated to the post-stage instrumentation amplifier through a single point.

[0061] For high-impedance source signals (such as EEG signal source impedance 10 - 100 kΩ, ECG signal source impedance 1 - 10 kΩ, EMG signal source impedance 5 - 50 kΩ), the input impedance of the buffer is much higher than the signal source impedance (10 TΩ vs 100 kΩ, load effect < 0.001%), and the voltage noise generated by the bias current on the signal source impedance ≤ 2 pA × 100 kΩ = 0.2 μV (much lower than the EEG signal amplitude of 10 μV) and can be ignored.

[0062] Solve the problem of signal attenuation caused by the high contact impedance of dry electrodes (such as 100 kΩ). The traditional 10^9 Ω buffer attenuates by about 1% under a 100 kΩ signal source (100 kΩ / (100 kΩ + 10^9 Ω)), while the 10 TΩ buffer attenuates only 0.001%, and the signal integrity is improved by 100 times.

[0063] Avoid generating additional voltage drops on high-impedance signal sources. For example, a 2 pA bias current generates 0.2 μV of noise on a 100 kΩ resistor, while a traditional buffer with a bias current of 1 nA generates 100 μV of noise, which is equivalent to 10% of the EEG signal amplitude. This design reduces such noise by 500 times.

[0064] It can be adapted to collect bioelectric signals such as EEG, ECG, and EMG simultaneously without replacing the front-end circuit. For example, when collecting ECG, the load effect of the 10TΩ input impedance on the 5kΩ source impedance of the limb lead can be ignored, and the noise generated by the bias current < 0.01μV, meeting the medical-grade signal quality requirements (ECG diagnosis requires noise ≤ 5μV).

[0065] The parallel connection of 4 channels does not change the unit gain characteristic (the closed-loop gain of each channel = 1), but the equivalent input capacitance is reduced (the input capacitance of a single buffer is 5pF, and it becomes 2.5pF after the parallel connection of 4 channels), reducing the phase delay of the parasitic capacitance of the electrode cable (usually 50 - 100pF) for high-frequency signals (such as β wave at 30Hz), and the phase error is reduced from 15 degrees to less than 5 degrees.

[0066] Exemplarily, the number of the buffers is 4, the supply voltage of the preset power supply is 2.5V, the capacitance values of the first filter capacitor and the second filter capacitor are 0.1μF, and the capacitance values of the first voltage-stabilizing capacitor and the second voltage-stabilizing capacitor are 4.7μF.

[0067] Four independent buffers are adopted (which can be integrated into a single chip such as AD8554, or four single-channel chips such as AD8551 are connected in parallel). The input ends are commonly connected to the electrode sensor, and the output ends are connected in parallel through a 0.1Ω precision resistor (reducing the signal voltage division caused by the output impedance difference).

[0068] It is powered by a single 2.5V power supply (adapted to the output of the LDO after the step-down of the 3.7V lithium battery, such as TPS76025), the ripple ≤ 5μV, and the power supply current ≤ 400μA (100μA for each of the 4 buffers).

[0069] The first / second filter capacitor: a 0.1μF multi-layer ceramic capacitor (X7R dielectric, 1206 package), installed within 1mm of the power supply pin of the buffer, filtering out high-frequency noise above 10MHz.

[0070] The first / second voltage-stabilizing capacitor: a 4.7μF tantalum capacitor (with a withstand voltage of 6.3V), used for storing charge, suppressing the power supply ripple below 100Hz, the ESR ≤ 50mΩ, ensuring that the voltage fluctuation < 10μV during the transient response.

[0071] With a 2.5V power supply, the total power consumption of the four buffers is only 1mW (400μA × 2.5V), which is 80% lower than that of the traditional ±5V dual-power supply scheme (power consumption of 5mW), making it suitable for batteries at the level of Bluetooth headsets (such as a 50mAh battery with a battery life of 50 hours). The 0.1μF + 4.7μF combination is a classic "high-frequency bypass + low-frequency energy storage" configuration. After actual measurement, the power supply noise can be suppressed to 1μV RMS (in the 10Hz - 100kHz frequency band), meeting the stringent requirements of the buffer for power supply noise (noise sensitivity < 5μV RMS). After the four buffers are connected in parallel, the output impedance is reduced to 1 / 4 of that of a single channel (the output impedance of a single channel is 50Ω, and it becomes 12.5Ω after parallel connection). When driving a 100pF cable capacitance, the rise time is shortened from 100ns to 25ns, avoiding signal edge distortion, and especially restoring the high-frequency components of electromyogram signals (high-frequency components of 50 - 100Hz) more accurately.

[0072] In some embodiments, it further includes: a protection resistor, which is connected in series between the input ends corresponding to the plurality of buffers and the electrode sensor, and is used to reduce the electrostatic damage corresponding to the electrode sensor.

[0073] A protection resistor (R_prot) is connected in series between the electrode sensor and the input end of the buffer. The resistance value is 51Ω (1 / 8W surface mount resistor, accuracy 1%), and it is located behind the electrostatic protection diode array (that is, the signal path is: electrode → ESD diode → R_prot → buffer input). The resistor is selected as a non-inductive resistor (such as a metal film resistor) with a parasitic inductance < 5nH to avoid introducing high-frequency noise.

[0074] When an ESD event occurs (such as an 8kV electrostatic discharge), after the ESD diode conducts, R_prot limits the current within a safe range. For example, the conduction voltage of the diode is 0.7V, the power supply is 2.5V, the voltage of the discharge loop = 8kV - 2.5V - 0.7V = 7.8kV, and the 51Ω resistor limits the current to approximately 7.8kV / 51Ω ≈ 150mA (the actual ESD pulse duration < 1μs and the energy is low), while the input withstand voltage of the buffer is usually ≥ ±20mA, and the 150mA peak current will not damage the device in a short time (the traditional resistorless design may have a current of more than 1A, directly burning out the input stage).

[0075] The input capacitance of the buffer (5pF) and the inductance of the long lead wire (assuming 10nH / m, the inductance of a 10cm wire is 1nH) may form an LC oscillation. The 51Ω resistor acts as a damping resistor, reducing the Q value from 10 to 0.5, eliminating high-frequency oscillations above 10MHz, and ensuring the stable operation of the signal chain in a radio frequency interference environment (such as mobile phone signals).

[0076] The 51Ω resistor has a voltage division ratio of 51 / (10^13 + 51) ≈ 5×10^-12 with respect to the buffer input impedance of 10TΩ, and the signal attenuation is negligible (<0.000001%). At the same time, there is no phase impact on the electroencephalogram signal (the highest frequency is 100Hz, and the capacitive reactance corresponding to the 51Ω impedance is 1 / (2π×100×5pF) = 318kΩ, which is much larger than 51Ω).

[0077] Exemplarily, the resistance value of the protection resistor is 51Ω.

[0078] The selection of the 51Ω resistor is based on the following design principles:

[0079] ESD current limiting calculation: According to the IEC 61000-4-2 contact discharge ±8kV standard, assuming that the forward voltage drop of the ESD diode after conduction is 0.7V, the power supply voltage is 2.5V, and the total resistance of the discharge circuit = R_prot + the internal resistance of the diode (about 10Ω) + the line resistance (about 1Ω), and the target is to limit the peak current <200mA (the upper limit of the safety current of the buffer input stage), then R_prot ≥ (8000V - 2.5V - 0.7V) / 0.2A ≈ 40kΩ. However, the actual ESD pulse energy is extremely low (in the nanojoule level), and there is no need for a too large resistor. 51Ω can limit the current while avoiding signal attenuation.

[0080] Noise and impedance matching: 51Ω is a commonly used matching resistor in radio frequency circuits, which can suppress the antenna effect (radio frequency signals received by the electrode as a small antenna), and attenuate the 1GHz radio frequency noise by 30dB (the 51Ω and the buffer input capacitance of 5pF form a low-pass filter, and the cut-off frequency is 1 / (2π×51×5e-12) ≈ 6.3GHz, and the attenuation of the 1GHz noise is -20log(1000 / 6300) = 16dB, and there are slight differences in actual due to PCB layout differences).

[0081] Mass production compatibility: 51Ω is a standard resistance value in the E24 series, with a low procurement cost. An accuracy of 1% can meet the requirements, and there is no need to customize resistors, which is suitable for large-scale production.

[0082] The 51Ω resistor provides sufficient current limiting ability in ESD protection, and has no measurable impact on the transmission of bioelectric signals. Actual measurements show that after adding the 51Ω resistor, the amplitude attenuation of the EEG signal <0.01%, the noise spectrum has no change in the frequency band below 100Hz, and the noise above 100Hz is reduced by 15% (due to the suppression of high-frequency oscillations).

[0083] Enhanced long-term reliability: During the daily use of wearable devices, when human static electricity (such as ±5kV static electricity generated by taking off clothes) passes through the protection resistor, the instantaneous current borne by the input stage of the buffer drops from 500mA without protection to about 100mA (5kV / (51 + 10) ≈ 82mA), which is lower than the device damage threshold (usually 200mA). This increases the circuit life from 100 ESD events to more than 1000 times.

[0084] In some embodiments, the EEG preamplifier circuit is disposed on a preset PCB board. The placement positions and electrical connection points corresponding to the electrode sensor and the buffer are determined according to the corresponding sizes of the electrode sensor and the buffer, so as to minimize the area of the PCB board.

[0085] The electrode sensor interface (such as a connector or a pad) is placed adjacent to the buffer chip, with a spacing < 5mm, to avoid overly long signal lines (ideally, directly bonded without leads).

[0086] The power filter capacitors (C1 / C2 / C3 / C4) are closely attached to the power pins of the buffer, adopting the shortest path of "power pin → capacitor → ground" to reduce the loop inductance (target < 1nH). The protection resistor (R_prot) and the ESD diode are adjacent to the electrode interface, forming a straight signal path of "electrode → ESD → R_prot → buffer" to avoid the introduction of parasitic inductance due to bent traces. Taking the center of the buffer chip as the origin, the electrode interface pads are symmetrically distributed around, the signal lines are 4mil wide (impedance controlled at 50Ω), covered with copper for shielding, and the ground layer is complete to avoid the signal loop passing through the power layer.

[0087] Through a compact layout, a 4-channel buffer + filtering + protection circuit can be integrated on a 20mm × 20mm PCB, which is 60% smaller than the traditional discrete design (area 50mm × 50mm), and is suitable for the sandwich design of head-mounted devices (such as the PCB thickness of the headband is 0.8mm). The short signal lines result in a parasitic capacitance < 2pF and a parasitic inductance < 2nH, and the transmission delay of the EEG signal (highest frequency 100Hz, wavelength 3000km, parasitic parameters can be ignored) is < 1ps, with completely no distortion. At the same time, the external interference coupling path is reduced, and the induced voltage of the spatial radiation noise drops from 10μV to below 1μV (induced voltage at a 100MHz noise field strength of 1V / m). The standardized layout reduces the trace crossing, avoids the use of vias (all connections are surface traces), and the welding yield rate is increased from 95% to 99.5%, which is suitable for automated production.

[0088] Exemplarily, the electrical connection point is located at the center point corresponding to the EEG preamplifier circuit, so that the electrode sensor can be directly connected to the buffer without passing through a lead wire.

[0089] The input pins of the buffer chip are centrally arranged at the center point of the PCB. The electrode sensor is directly pressed onto the central pad through spring pins or elastic contacts, eliminating the traditional lead wires (usually 10 - 50 cm in length), forming a zero-length connection of "electrode → buffer input pin". The PCB is designed as circular or square, with the central area being the electrode interface, and buffers, filter capacitors, and protection circuits are distributed around it. The power supply and the output of the subsequent stage signal are led out through edge connectors.

[0090] Traditional lead wires are prone to introducing power frequency interference (50 / 60 Hz), motion artifacts (electrostatic noise generated by cable friction), and distributed parameters (such as the parasitic capacitance of a 10 cm wire is 50 pF and the inductance is 10 nH). After removing the lead wires in this design, the amplitude of the 50 Hz common-mode noise drops from 500 μV to 50 μV (suppressing 90%), and the signal fluctuation during movement drops from ±20 μV to ±2 μV, which is especially suitable for wearable devices (such as a sleep EEG headband, without cable drag interference when the user turns over). When there are no lead wires, the input capacitance of the buffer (5 pF) is directly connected to the electrode, and the high-frequency cut-off frequency is increased from 10 kHz (when there is a 10 cm wire, the wire capacitance is 50 pF + the input capacitance is 5 pF, and the RC cut-off frequency is 1 / (2π×100 kΩ×55 pF) = 29 kHz) to the theoretical limit (only limited by the buffer bandwidth, 1 MHz), ensuring the complete acquisition of β waves (30 Hz) and γ waves (80 Hz). After removing the lead wires, the problem of poor plug contact is avoided (the contact resistance of traditional connectors can fluctuate up to 10 Ω, introducing 1 μV of noise), and the connection reliability is increased from 90% to 99.9%, which is suitable for long-term wear (such as continuous monitoring for more than 7 days).

[0091] In some embodiments, the electrode sensor at least includes a sensing electrode, a reference electrode, and a bias driving electrode; the sensing electrode is used to collect the EEG signal, the reference electrode is used to receive a reference signal, and the sensing electrode and the reference electrode are connected to the corresponding input ends of a plurality of the buffers; the bias driving electrode is used to output a bias driving signal generated by the instrumentation amplifier or the signal processing circuit.

[0092] Sensing electrode: Directly contacts the scalp to collect the EEG signal (such as at lead positions like FP1, FP2, etc.), and is connected to the non-inverting input end of the buffer.

[0093] Reference electrode: Placed at an irrelevant part (such as the earlobe), collects the reference signal (including common-mode noise), and is connected to the non-inverting input end of another group of buffers (or the differential input of the same buffer, a dual-channel buffer is required).

[0094] Bias driving electrode: Driven by the common-mode voltage output by the subsequent instrumentation amplifier (i.e., the "right leg driving" circuit), and is connected to E3 through the inverting output of the buffer (unity-gain inverting amplifier) to form negative feedback and suppress common-mode noise.

[0095] Signal processing chain: The acquisition electrode signal (acquisition electrode + noise) and the reference electrode signal (reference electrode + noise) are input into an instrumentation amplifier after passing through a buffer. The differential amplification factor G = 1 + 2Rf / Rg. At the same time, the bias drive electrode inverts the common-mode voltage output by the instrumentation amplifier and applies it to the human body to reduce the common-mode coupling of the skin-electrode contact impedance.

[0096] The CMRR of traditional single-ended acquisition is only 60 dB (suppressing 50 Hz power frequency noise by 1000 times). Through the reference electrode and differential amplification in this design, the CMRR is increased to 120 dB (suppressing 1,000,000 times), and the 50 Hz noise is reduced from 500 μV to 0.5 μV, meeting the medical-grade EEG acquisition standard (CMRR ≥ 100 dB).

[0097] Dry electrodes have high contact impedance (100 kΩ) and are unstable. The bias drive electrode outputs a signal that is in-phase and opposite to the common-mode noise, canceling the capacitive coupling at the electrode-skin interface, reducing the contact impedance to less than 10 kΩ equivalently. The signal attenuation is reduced from 30% to 5%. At the same time, the baseline drift caused by impedance changes during movement is reduced (the drift amplitude is reduced from ±50 μV to ±5 μV). It supports at least a 3-electrode configuration (single-channel differential acquisition) and can be extended to 16 / 32 leads (each lead has an independent buffer group), adapting to the different requirements of clinical EEG devices (requiring multi-channel synchronous acquisition) and consumer-grade EEG devices (single-channel or a small number of channels), with strong architecture compatibility.

[0098] In some embodiments, the buffer of the present application is a precision, low-power, FET-input four-channel unity-gain buffer, with high input impedance and low-noise characteristics, which is very suitable for EEG signal conditioning and amplification. Its high input impedance (10 TΩ) and low input bias current (maximum 2 pA, 25°C) enable it to process high-impedance source signals, such as biopotential signals like electrocardiogram (ECG), electroencephalogram (EEG), and electromyogram (EMG). These signals usually have high source impedance, and traditional operational amplifiers may introduce large errors due to leakage current, while using the provided buffer can effectively avoid these problems.

[0099] In some embodiments, traditional fixed-parameter filters (such as 50 Hz notch filters) cannot handle time-varying noise (such as frequency shift of myoelectric noise caused by movement and amplitude fluctuation of power frequency noise). Especially, the effective components in the EEG signal in the range of 0.5 - 30 Hz are easily contaminated by non-steady noise. By adding a 24-bit ADC (such as ADS1299, sampling rate 2 kSPS) at the buffer output, the 4-channel signals are synchronously sampled, and the quantization noise ≤ 0.1 μV RMS. An integrated low-power MCU (such as STM32L4) or NPU (such as EdgeTPU) with a computing power ≥ 100 GOPS is supported to enable real-time filtering algorithms.

[0100] By constructing the EEG signal state equation \(x_k = Ax_{k - 1}+w_k\), where the state variables include the amplitudes of δ / θ / α / β waves and noise components, and the observation equation \(z_k = Hx_k + v_k\), the noise covariance matrices \(Q\) and \(R\) are updated in real time. Every 100 ms, the noise statistical characteristics are estimated by the recursive least squares (RLS) method, and the Kalman gain \(K_k\) is dynamically adjusted to achieve the optimal estimation and deduction of power frequency noise (50 ± 2 Hz) and electromyogram noise (variable frequency band at 20 - 200 Hz). The boundary constraints are set by setting the physiologically reasonable range of the EEG signal (such as the amplitude fluctuation of a single lead ≤ 100 μV / ms) to suppress burst pulse noise (such as electrode contact transient interference).

[0101] For the electromyogram noise caused by movement (the frequency band drifts from 20 Hz to 80 Hz), the suppression ratio is increased from 20 dB of the fixed filter to 45 dB, and the signal-to-noise ratio of the α wave (8 - 12 Hz) is increased from 25 dB to 38 dB. The adaptive tracking of the power frequency noise frequency offset (such as the 50 Hz ± 1 Hz change caused by power grid fluctuations) reduces the notch bandwidth from the fixed 2 Hz to 0.5 Hz dynamically, avoiding the attenuation of the adjacent α wave by the traditional notch filter (the original attenuation is 5%, and now it is < 1%). Based on the boundary constraints of the physiological model, pulse interference > 200 μV is effectively filtered out (such as the transient signal of electrode detachment), the missed detection rate is < 0.1%, and the false detection rate is < 0.05%, which is better than the traditional threshold detection method (the missed detection rate is 5%). The estimated value of the noise component is output in real time for the subsequent circuit to dynamically adjust the hardware filtering parameters (such as adjusting the gain of the instrumentation amplifier through a digital potentiometer to form an algorithm-hardware closed loop). The power consumption of the NPU in the filtering mode is < 5 mW, only 1 / 3 of the traditional DSP solution. Combined with the sleep mode (waking up once every 1 ms to update the model), the overall computing power occupancy is < 20%, meeting the 72-hour battery life requirement of a 50 mAh battery.

[0102] In some embodiments, the contact impedance of the dry electrode (usually 10 - 100 kΩ) changes with skin humidity and pressure. When the impedance > 50 kΩ, the signal attenuation > 10%. The traditional manual calibration cannot cope with the dynamic changes. By integrating an excitation signal generator (10 Hz sine wave, amplitude 100 μV, output through DAC) at the input end of the buffer, it is injected into the electrode through a 51 Ω protection resistor. The response voltage of the excitation signal is synchronously collected, and the AD8302 phase detector is used to measure the amplitude ratio and phase difference between the excitation signal and the response signal, and calculate the contact impedance \(Z = V_{response} / I_{excitation}=V_{response} / (100 μV / 51 Ω)\)

[0103] By offline training a random forest model, inputting the impedance amplitude, phase angle, and temperature (integrating a DS18B20 temperature sensor), and outputting the electrode contact quality level (excellent / good / poor). The training set contains more than 1000 different skin state samples, and the accuracy is 95%.

[0104] When the impedance > 80 kΩ and the quality grade is "poor" are detected, a prompt for adjusting the trigger electrode position (through bone conduction headphones or LED lights); dynamically allocate lead weights based on impedance distribution (for example, if the impedance of a certain channel in 4 channels is too high, automatically increase the signal weights of other channels), and use the weighted least squares method to fuse multi-channel signals to compensate for single-channel attenuation.

[0105] The impedance measurement error < 3% (in the range of 10 - 100 kΩ), the phase angle resolution is 0.1 degree, and it can distinguish the change in the stratum corneum thickness (a 10-μm thickness change corresponds to a 0.5-degree phase angle shift). Complete an impedance scan of the entire electrode group (4 channels) every 2 seconds, with a power consumption < 0.5 mW (the duty cycle of the excitation signal is 10%), which is much lower than the traditional constant current source impedance measurement scheme (power consumption 5 mW). When the single-channel impedance increases from 50 kΩ to 100 kΩ, the attenuation of the fused signal decreases from 20% to 3%, and the measurement error of the α-wave amplitude decreases from 15% to 2%. The automatic calibration algorithm enables the system to maintain signal stability in the impedance fluctuations (±30 kΩ change) caused by user movement (such as turning the head), and the standard deviation of the baseline drift decreases from 15 μV to 5 μV. The intelligent prompt based on impedance quality reduces the user's manual calibration frequency from once per hour to once every 8 hours, improving the usage experience of wearable devices, especially suitable for the sleep monitoring scenario (no intervention is required at night).

[0106] In some embodiments, the traditional EEG front end relies on hardware redundancy (such as spare channels), but it cannot locate soft faults in real time (such as buffer bias current drift, increase in capacitor ESR), and the fault undetected rate > 10%. By integrating a current monitoring ADC (INA219, resolution 0.1 μA) at the power supply pin of each buffer, the operating currents of 4 channels of buffers are collected in real time (the normal range is 80 - 120 μA). Configure redundant buffer channels (the total number of channels is 5, with 1 hot standby), and dynamically switch the signal path through an analog switch (ADG732).

[0107] Construct a 1D-CNN model, the input includes 4-channel signal waveforms, power supply current, temperature, impedance data (a total of 10-dimensional features), and the output is the fault type (buffer failure / capacitor aging / resistor drift). The training set contains more than 2000 fault samples, and the F1-score is 98%. When it is detected that the buffer current > 150 μA (overheat fault) or < 50 μA (bias failure), switch to the redundant channel within 10 ms, and record the fault log at the same time; for the increase in capacitor ESR (judged by the change in the power supply ripple spectrum), automatically adjust the noise covariance in the Kalman filter model to compensate for the degradation of the filtering performance.

[0108] The soft fault detection delay is less than 5 ms, which is 10 times faster than traditional threshold detection (with a delay of 50 ms), and can capture the slow drift of the leakage current at the input stage of the buffer (changing by 1 pA per month). The signal interruption introduced by redundant channel switching is less than 1 sampling period (less than 0.5 ms at 2 kSPS), and the continuity of EEG signals is not affected, which is particularly crucial for long-term monitoring (such as 24-hour EEG recording). The mean time between failures (MTBF) of the system has been increased from 500 hours to 5000 hours, and the maintenance cost has been reduced by 70%. When the power supply ripple increases due to capacitor aging (rising from 1 μV to 5 μV), the algorithm compensation only increases the buffer output noise by 0.5 μV, maintaining the signal quality within the medical-grade standard. Through fault trend analysis (such as the buffer current rising month by month), potential failures can be warned 72 hours in advance, which is better than the traditional after-failure maintenance mode and is suitable for implantable or field-deployed devices.

[0109] In some embodiments, the traditional fixed power consumption mode wastes energy during the low-activity period of the signal (such as the sleep stage dominated by δ waves), and the battery life of the wearable device is limited by the front-end power consumption of more than 30 mW. By adopting a configurable power supply buffer (such as AD8551, supporting power supply from 1.8 V to 5.5 V), combined with a DC-DC converter (TPS61040, with an efficiency of 95%), channel-level voltage regulation is achieved. An integrated neuromorphic sensor (such as the analog front end based on the principle of the Prophesee event camera) triggers event pulses for the rate of change of the voltage of the EEG signal (dV / dt > 1 μV / μs) and outputs a sparsified signal. A spiking neural network (SNN) is used to process the event pulse stream in real time to distinguish between high-activity states (dominated by β waves, event rate > 100 Hz) and low-activity states (dominated by δ waves, event rate < 10 Hz), with a classification delay of less than 1 μs and a power consumption of less than 100 nW.

[0110] The dynamic power consumption strategy includes: high-activity state: the buffer is powered at 2.5 V, the ADC sampling rate is 2 kSPS, and the NPU runs at full speed; low-activity state: the buffer voltage is stepped down to 1.8 V (with a 60% reduction in power consumption), the ADC is downsampled to 100 SPS, the NPU enters the sleep state, and only the SNN is retained to maintain monitoring.

[0111] The power consumption in the awake state (high activity) is 12 mW, and in the sleep state (low activity), it drops sharply to 1.5 mW. The overall average power consumption is 4 mW, a 60% reduction compared to the fixed mode (10 mW). The battery life of 50 mAh is extended from 12 hours to 33 hours. The event-driven mechanism of the neuromorphic front-end reduces the data transmission volume by 80% (sampling only when the signal changes significantly), reducing the power consumption of the subsequent wireless transmission (such as Bluetooth) (the transmission power consumption accounts for 40% of the overall, and now it is reduced by 32%). The SNN classification accuracy is 98% (distinguishing δ / β wave states), the distortion introduced by voltage regulation is <0.5% (the gain error of the buffer is <0.1% at 1.8 V), and there is no significant change in the baseline drift of the measured δ wave (0.5 Hz). The lightweight SNN model (parameters <10 KB) runs on an ultra-low-power MCU without external storage, and the response speed is better than that of traditional CNN (the latency is reduced from 10 ms to 1 μs), making it suitable for closed-loop brain-computer interfaces with high real-time requirements (such as epilepsy monitoring and warning).

[0112] In some embodiments, traditional fixed leads (such as the international 10-20 system) do not consider individual differences, and when the prefrontal electrodes are in poor contact, it is easy to cause the failure of α wave acquisition, so the lead combination needs to be dynamically optimized. By designing an 8-channel electrode interface (compatible with the original 4-channel buffer, extended by analog switches), it supports the flexible access of up to 8 electrodes. Each electrode interface integrates a pressure sensor (membrane type, resolution 0.1 N) to monitor the contact pressure between the electrode and the scalp (the ideal range is 1-3 N). Based on the user's initial 5-minute resting-state EEG data, the particle swarm optimization (PSO) algorithm is used to search for the optimal 4-lead combination, and the objective function is:

[0113] ;

[0114] where SNR is the signal-to-noise ratio of each lead, Z is the contact impedance, and the constraint condition is that the pressure > 0.5 N. Every hour or when a sharp drop in contact pressure is detected, rapid re-optimization is triggered (completed within 10 seconds), and the historical data is used to warm up the model to reduce the calculation amount.

[0115] For different users, the optimal lead combination can increase the overall SNR by 20-30%. For example, for users with thick hair, electrodes in the temporal lobe area (T3 / T4) are selected to replace the prefrontal area, and the SNR is increased from 20 dB to 28 dB.

[0116] Contact pressure constraint avoids noise caused by loose electrodes (noise increases by 50% when the pressure < 0.5N), and the measured pressure compliance rate is increased from 70% to 95%. It supports different forms of devices such as helmet type and headband type, and compensates for the hardware layout limitations through lead optimization (for example, when the top electrode cannot be installed on the headband, the weight of the occipital region electrode is automatically enhanced). After the user changes the posture (such as from sitting to lying down), the signal recovery time is shortened from 5 minutes to 30 seconds during dynamic reconfiguration, which is suitable for multi-scenario applications (office / sports / sleep). Customized lead schemes are provided for special populations such as newborns and elderly patients to solve the adaptation problem of traditional fixed leads under different skull morphologies. For example, in the EEG monitoring of premature infants, the interference of electrodes in the fontanelle area is reduced through optimization, and the effective signal extraction rate is increased from 60% to 85%.

[0117] Please refer to Figure 3 , this application provides a brain-computer interface device, including the EEG preamplification circuit provided by any embodiment of this application.

[0118] In some embodiments, the brain-computer interface device includes: a front-end acquisition module, including:[[]]END]]

[0119] Electrode array: 8-channel pluggable dry electrodes (silver-plated / silver chloride material), integrated with a micro pressure sensor (accuracy 0.1N) and a temperature sensor (DS18B20), supporting the standard positions of the international 10-20 system (such as Fp1, Fp2, C3, C4, etc.) and custom extensions, and dynamically selecting 4 main acquisition channels + 1 redundant channel through an analog switch (ADG732).

[0120] Preamplification circuit: Each channel has an independent buffer (AD8551, configurable power supply of 1.8 - 5.5V), integrated with a 10Hz excitation signal generator (DAC outputs a 100μV sine wave) for impedance measurement, a 51Ω protection resistor to prevent overcurrent, and the signal is digitized by a 24-bit ADC (ADS1299, 2kSPS synchronous sampling), with quantization noise < 0.1μV RMS.

[0121] Sensor fusion: Synchronously access an accelerometer (ADXL345, detecting motion artifacts) and a gyroscope (ITG-3200, attitude monitoring) for motion noise modeling.

[0122] Heterogeneous computing architecture: A low-power MCU (STM32L4, main frequency 80MHz) is responsible for real-time control and communication, and an NPU (EdgeTPU, 100GOPS) runs intelligent algorithms (Kalman filter, SNN, CNN, etc.). The two perform high-speed data interaction through the AXI bus, and the computing power allocation is dynamically adjustable (for example, noise suppression accounts for 60% of the computing power, and lead optimization accounts for 30%).

[0123] Storage and Communication: 2MB of on-board SRAM caches the raw EEG data. The BLE 5.2 module (nRF52840) enables low-power wireless transmission (rate: 2Mbps, distance: 10m), supports real-time synchronization with mobile phone / PC-side APPs, and also reserves a USB-C interface for high-speed wired data export (5Gbps).

[0124] Power Supply System: A 50mAh lithium battery (supporting wireless charging), paired with a DC-DC converter (TPS61040, efficiency: 95%), realizes channel-level voltage regulation. The neuromorphic front end (Example 12) reduces the buffer voltage to 1.8V in the low-activity state, and together with the sleep mode, the standby power consumption is <10μA.

[0125] Energy Management Module: According to the real-time activity classification results of the SNN (high / low EEG activity states), it dynamically adjusts the ADC sampling rate (2kSPS → 100SPS), the NPU frequency (100GOPS → 10GOPS), and the Bluetooth transmission interval (10ms → 100ms), forming a three-level power consumption optimization strategy.

[0126] Adaptive Noise Suppression: The Kalman filter algorithm updates the state space model every 1ms, real-time estimates and subtracts power frequency noise (50±0.5Hz dynamic notch), electromyogram noise (time-varying frequency band tracking), and at the same time applies physiological boundary constraints (amplitude change rate <100μV / ms) to filter out pulse interference, and outputs 4-channel EEG signals after noise suppression (bandwidth: 0.1 - 100Hz, signal-to-noise ratio improved to 40dB+).

[0127] Electrode Status Management: Scans the impedance of all electrodes every 2 seconds (in the range of 10 - 100kΩ, error <3%), maps the contact quality level through a random forest model, triggers automatic calibration (weight assignment compensation) or user prompts (bone conduction headset voice: "The right frontal electrode needs adjustment"), and at the same time inputs the impedance data into the lead optimization algorithm as a constraint condition.

[0128] Fault Self-Healing Mechanism: The 1D-CNN real-time monitors the working current of the buffer (normal range: 80 - 120μA) and the power supply ripple (<2μV). After detecting an anomaly (such as current >150μA), it switches to the redundant channel within 10ms, and at the same time records the fault log (stored in the on-board Flash, supporting more than 1000 records), and the MTBF is increased to more than 5000 hours.

[0129] Dynamic Lead Configuration: Collects 5 minutes of resting-state data at initial power-on, searches for the optimal 4-lead combination through particle swarm optimization (PSO) (objective function: maximizing the sum of SNR + minimizing the sum of impedance), and triggers rapid reconfiguration (completed within 10 seconds) every hour or when the pressure drops suddenly during operation, adapting to individual skull differences and posture changes.

[0130] Through lightweight interaction design: bone conduction headphones (power consumption < 1 mW) are used for status indication. The LED matrix displays the electrode connection status (green: normal, yellow: calibration required, red: fault), and supports the blind operation mode (switch the working mode through touch buttons: monitoring / training / warning). Multimodal output: In addition to the original EEG data, it outputs the power spectra of α / β / θ / δ waves, attention index (based on the energy ratio of β waves), and fatigue score (δ / α wave ratio) in real time, and transmits them to the mobile APP via Bluetooth, supporting EEG biofeedback training (such as controlling concentration games).

[0131] Headband form: Elastic silicone headband, the electrode position is adjustable (within the range of ±10 mm). It is equipped with a built-in micro air pump (power consumption < 2 mW) to automatically adjust the electrode pressure to the ideal range of 1 - 3 N, suitable for users with a head circumference of 50 - 60 cm, and the single wearing time can exceed 8 hours (the pressure is evenly distributed to avoid local compression).

[0132] Protection level: IP54 dust and waterproof. The electrode contact surface is treated with a nano - coating (reducing protein adsorption, extending the cleaning cycle to 7 days), suitable for daily wear and sports scenarios (such as EEG monitoring during running and cycling).

[0133] The time - varying noise suppression ratio reaches 45 dB (traditional devices < 25 dB). It can accurately track the myoelectric noise drift (20 - 80 Hz frequency band) and power frequency fluctuation (50 ± 1 Hz) caused by movement, and the SNR of α wave is increased from 25 dB to 38 dB, supporting the reliable acquisition of weak EEG signals (such as γ waves < 5 μV), meeting the requirements of high - precision brain - computer interfaces (such as P300 evoked potential detection).

[0134] The physiological constraint mechanism makes the missed detection rate of pulse interference < 0.1%, superior to the traditional threshold detection method (5%), ensuring the complete capture of transient signals such as epileptic spikes (the missed detection rate is reduced from 20% to 1%).

[0135] Intelligent lead configuration improves the SNR of different users by 20 - 30%. For example, those with thick hair use temporal region electrodes to replace the frontal region electrodes, solving the adaptation blind area of traditional fixed leads. The effective signal extraction rate of special populations such as newborns / elderly patients is increased from 60% to 85%, expanding the clinical application scenarios (such as EEG monitoring of premature infants and early screening of Alzheimer's disease).

[0136] Dynamic pressure regulation and impedance compensation reduce the signal attenuation caused by poor electrode contact from 20% to 3%, and the standard deviation of baseline drift in the movement scenario is reduced from 15 μV to 5 μV, supporting full - scenario EEG monitoring (seamless switching between resting state / movement state / sleep state).

[0137] Dynamic power consumption optimization reduces the average power consumption to 4 mW (traditional devices > 10 mW). The 50 mAh battery supports 33 hours of continuous monitoring (traditional devices < 12 hours). Combined with wireless charging technology, it enables "charging while in use" to meet long-term requirements such as 24-hour sleep monitoring and all-day EEG training.

[0138] The event-driven mechanism of the neuromorphic front-end reduces the data transmission volume by 80%, reduces the Bluetooth power consumption by 32%, and simultaneously alleviates the pressure of back-end data processing (the cloud computing power demand is reduced by 60%), constructing a lightweight edge computing architecture.

[0139] The fault self-healing system improves the MTBF from 500 hours to 5000 hours, and the redundant channel switching interruption < 0.5 ms, avoiding monitoring interruption caused by hardware failures in traditional devices (such as manual restart when the buffer fails), which is suitable for unattended scenarios (such as home health monitoring).

[0140] Automatic calibration and intelligent prompts reduce the user intervention frequency from once per hour to once every 8 hours. The headband-type adaptive pressure adjustment enables "measurement as soon as worn" without the assistance of professionals, significantly reducing the usage threshold and promoting the popularization of consumer-grade brain-computer interfaces.

[0141] It meets medical device standards (such as signal distortion < 0.5%, noise < 1 μV RMS), supports clinical EEG monitoring (such as EEG topographic mapping, seizure warning), and at the same time provides consumer-grade functions such as neurofeedback games and concentration training through the APP, constructing a "medical + health" dual-scenario device. The open API interface supports the access of third-party algorithms (such as custom EEG feature extraction models) and is compatible with mainstream brain-computer interface protocols (BCI2000, LSL), contributing to the construction of the scientific research and industrial ecosystem.

[0142] The 8-channel electrode interface and configurable front-end circuit support the expansion of the number of channels (upgraded to 16 / 32 channels in the future). The heterogeneous computing architecture reserves computing power redundancy (the current computing power occupancy < 60%) to adapt to more complex EEG signal processing algorithms (such as deep learning EEG classification models). The modular design (the electrode / processing unit / power supply can be independently replaced) reduces the maintenance cost. It is expected that the product life cycle will be extended to 5 years, and the total cost of ownership (TCO) is reduced by 40% compared with traditional devices.

[0143] Integrate technologies such as Kalman filter dynamic noise suppression, neuromorphic power consumption optimization, and particle swarm lead configuration into wearable devices to form a "signal acquisition - processing - adaptation" closed-loop, with the technical complexity leading 2 - 3 generations of similar products. Continuously optimize the lead configuration model through the user's EEG data (anonymous data training at the edge). For every additional 1000 users, the individual adaptation accuracy is increased by 1.5%, constructing a positive cycle of "device use - algorithm evolution" to form a differentiated competitive advantage.

[0144] This brain-computer interface device not only breaks through the performance bottleneck of traditional EEG acquisition devices, but also realizes the leap from a "signal acquisition tool" to a "personalized brain function monitoring terminal" through the deep cooperation between intelligent algorithms and hardware, and has significant application value and market prospects in the fields of medical health, human-computer interaction, neuroscience research, etc.

[0145] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims.

Claims

1. An EEG preamplifier circuit, characterized in that: include: An electrode sensor, wherein the electrode sensor is used to collect brain electrical signals; A plurality of buffers connected in parallel, wherein the input ends corresponding to the plurality of buffers are connected to the electrode sensors, and are used to buffer or preliminarily amplify the EEG signals collected by the electrode sensors, and transmit the EEG signals to a preset instrument amplifier or signal processing circuit; The buffer is a four-channel unity gain buffer; An electrostatic protection diode array, the electrostatic protection diode array comprising a first input terminal, a second input terminal and a third input terminal, the first input terminal and the second input terminal are respectively connected to the positive electrode and the negative electrode of a preset power supply, the third input terminal is connected between the input terminals corresponding to the plurality of buffers and the electrode sensor, and is used to protect the buffer and the preset power supply; the preset power supply is used to power the buffer; A first filter capacitor, a second filter capacitor, a first voltage-stabilizing capacitor and a second voltage-stabilizing capacitor, the first filter capacitor and the first voltage-stabilizing capacitor are connected in parallel, the second filter capacitor and the second voltage-stabilizing capacitor are connected in parallel, the input ends of the first filter capacitor and the first voltage-stabilizing capacitor are connected to the positive electrode of a preset power supply, and the input ends are connected to a preset ground end, the input ends of the first filter capacitor and the first voltage-stabilizing capacitor are connected to the negative electrode of the preset power supply, and the input ends are connected to a preset ground end, for minimizing the power supply noise corresponding to the preset power supply.

2. The EEG preamplifier circuit according to claim 1, characterized in that: Also includes: A differential mode elimination capacitor, wherein the differential mode elimination capacitor is connected between the input ends of the first filter capacitor and the first voltage stabilizing capacitor and the input ends of the second filter capacitor and the second voltage stabilizing capacitor; The capacitance of the differential mode elimination capacitor is the same as that of the first voltage stabilizing capacitor or the second voltage stabilizing capacitor.

3. The EEG preamplifier circuit according to claim 1, characterized in that: The input impedance of the four-channel unity gain buffer is 10 TΩ, and the input bias current is ≤2 pA, so that the buffer can process high-impedance source signals; the high-impedance source signal includes at least any one of an electrocardiogram signal, an electroencephalogram signal, and an electromyogram signal.

4. The EEG preamplifier circuit according to claim 3, characterized in that: The number of the buffers is 4, the supply voltage of the preset power supply is 2.5V, the capacitance of the first filter capacitor and the second filter capacitor is 0.1μF, and the capacitance of the first voltage stabilizing capacitor and the second voltage stabilizing capacitor is 4.7μF.

5. The EEG preamplifier circuit according to claim 1, characterized in that: Also includes: A protection resistor is connected in series between the input terminals corresponding to the plurality of buffers and the electrode sensor, and is used to reduce electrostatic damage corresponding to the electrode sensor.

6. The EEG preamplifier circuit according to claim 5, characterized in that: The resistance of the protection resistor is 51Ω.

7. The EEG preamplifier circuit according to claim 1, characterized in that: The EEG preamplifier circuit is arranged on a preset PCB board, and the placement positions and electrical connection points of the electrode sensor and the buffer are determined according to the corresponding sizes of the electrode sensor and the buffer, so as to minimize the area of ​​the PCB board.

8. The EEG preamplifier circuit according to claim 7, characterized in that: The electrical connection point is located at a center point corresponding to the EEG preamplifier circuit, so that the electrode sensor is directly connected to the buffer without passing through a lead wire.

9. The EEG preamplifier circuit according to claim 1, characterized in that: The electrode sensor at least includes a collection electrode, a reference electrode and a bias drive electrode; The collection electrode is used to collect the EEG signal, the reference electrode is used to receive a reference signal, and the collection electrode and the reference electrode are connected to the input ends corresponding to the plurality of buffers; The bias driving electrode is used to output the bias driving signal generated by the instrumentation amplifier or the signal processing circuit.

10. A brain-computer interface device, characterized in that: The EEG preamplifier circuit comprises the EEG preamplifier circuit according to any one of claims 1 to 9.

Images