A data structure, circuit, and method for serial transmission of neural action potentials

By using a full-channel bitmap trigger marker and dual shift register hardware architecture, low power consumption, low addressing overhead, and high scalability transmission of neural action potential signals are achieved, solving the problems of high transmission power consumption and poor scalability of large-scale neural electrode arrays. This architecture is suitable for data transmission of multi-channel neural electrode arrays.

CN123068338APending Publication Date: 2026-11-10SHANGHAI JUZHI TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611386536.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In large-scale neural electrode array scenarios, traditional neural action potential signal acquisition systems have high transmission power consumption, large addressing overhead, cannot be compatible with hierarchical application requirements, and have poor array scalability.

Method used

It adopts a full-channel bitmap trigger marker and dual shift register hardware architecture to achieve automatic channel correspondence under dynamic data rate and support intra-frame hierarchical transmission. Through the serial transmission structure of full-channel AP trigger marker segment and AP waveform data segment, combined with data shift register chain and marker shift register chain, it simplifies addressing logic and reduces power consumption.

Benefits of technology

It significantly reduces the transmission power consumption of multi-channel AP signals, has extremely low addressing overhead, simple circuit structure, strong array scalability, and data rate that dynamically adapts to AP density, making it suitable for large-scale neural network acquisition scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123068338A_ABST
    Figure CN123068338A_ABST
Patent Text Reader

Abstract

This invention discloses a data structure, circuit, and method for serial transmission of neural action potential signals, belonging to the field of neural electrical signal acquisition and integrated circuit transmission. Each frame of the data structure sequentially includes a frame header check bit, a full-channel AP trigger flag segment, and an AP waveform data segment. The number of bits in the flag segment is equal to the number of channels, with each bit corresponding to the AP detection status of one channel. The waveform data is arranged in channel order, and the receiving end directly completes channel matching through the flag bit position. The circuit adopts a dual independent shift register chain architecture, configured with a bypass switch to automatically skip invalid channels, separately registering the flag signal and waveform data, and driving time-division output by an independent clock. The method supports intra-frame hierarchical transmission, and the waveform data segment can be truncated only when the AP is detected or not. This invention significantly reduces the transmission addressing overhead of multi-channel AP signals, with the data rate reduced to as low as 0.05% of the original sampling rate. The circuit structure is simple, the array is highly scalable, and it is suitable for large-scale neural acquisition systems with thousands of channels.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of neural electrical signal acquisition and integrated circuit data transmission technology, and in particular to a serial transmission data structure suitable for multi-channel neural action potential signals, the corresponding acquisition and transmission circuit, and the transmission method. Background Technology

[0002] Action potentials (APs) are the core representation signals of neuronal electrical activity, and multi-channel neural electrode arrays are the core hardware of implantable and desktop neural signal acquisition systems. Traditional AP signal acquisition systems adopt a parallel acquisition-serial output architecture with a fixed sampling rate: each channel continuously samples neural signals at a sampling rate of 10k~20kSPS and a resolution of 10~12 bits, and the multi-channel sampling results are uniformly serially output through a parallel-to-serial conversion circuit composed of shift registers.

[0003] When the number of channels is small (≤32 channels), the data transmission power consumption of this architecture is within an acceptable range; however, when the number of channels is expanded to more than 256 channels, especially in the case of large-scale neural electrode arrays with thousands of channels, the raw sampling data rate will reach tens of Mbps, and the transmission power consumption can account for more than 50% of the total power consumption of the chip, which seriously restricts the battery life and integration of implantable neural interface devices.

[0004] To reduce the amount of data transmitted, existing technologies typically extract and compress features from the AP signal before transmission. Taking advantage of the time sparsity and refractory period of the AP signal, compressed waveform data is output only when a valid AP signal is detected. However, this approach has three inherent drawbacks: Channel addressing overhead is high, and data matching logic is complex. Because the AP signals of different channels are generated at uncorrelated times, the data output frequencies of different channels are inconsistent, and the compressed data is a non-fixed-length stream, the receiving end cannot directly match the channel number through data position. Existing mainstream solutions, such as the Address Event Representation (AER) protocol, use a method of "packaging channel address + waveform data per event" to solve the channel correspondence problem. However, each AP event requires repeated transmission of address information, and the address overhead increases with the number of channels. Furthermore, this architecture requires complex asynchronous arbitration and FIFO buffering circuits, resulting in relatively high hardware overhead and power consumption.

[0005] It cannot meet the needs of hierarchical applications. In large-scale neural network acquisition scenarios with thousands of channels, some applications only need to determine whether each channel generates an AP signal (i.e., only trigger information), without needing to reconstruct the AP waveform details. Existing compressed transmission protocols must transmit the complete data packet regardless of whether the waveform is needed, and cannot truncate the transmission as needed, thus limiting the potential for further power consumption optimization.

[0006] Poor array scalability. Traditional shift register architectures only support fixed-length data output and cannot adapt to compressed AP signals with dynamic data volumes; while AER-type protocol circuit architectures are complex, and two-dimensional array expansion is difficult, making it hard to adapt to the acquisition requirements of large-scale planar electrode arrays. Summary of the Invention

[0007] (a) Purpose of the invention To address the aforementioned deficiencies in the prior art, the present invention aims to provide a data structure, circuit, and method for serial transmission of neural action potentials. Through a full-channel bitmap trigger marker and dual shift register hardware architecture, it achieves automatic channel correspondence under dynamic data rates, while supporting intra-frame hierarchical transmission, significantly reducing the transmission power consumption of multi-channel AP signals. Furthermore, the circuit structure is simple and the array scalability is strong.

[0008] (II) Technical Solution To achieve the above objectives, the present invention adopts the following technical solution: A serial transmission data structure for neural action potentials (APs) is disclosed. Each frame of data includes, in the transmission order, a frame header check bit, a full-channel AP trigger flag segment, and an AP waveform data segment. The frame header check bit is a fixed level of 1 bit, used to identify the start of a frame of data and provide a frame synchronization reference for the receiving end. The number of bits in the full-channel AP trigger flag segment is equal to the total number of sampling channels N. Each bit of binary data corresponds to one sampling channel and is used to identify whether the corresponding sampling channel detects a neural action potential (AP) signal in the current sampling period. The AP waveform data segment contains several sets of compressed AP waveform data. The number of AP waveform data sets is equal to the number of valid bits in the full-channel AP trigger marker segment that indicate "AP signal detected". Each set of AP waveform data is arranged sequentially according to the corresponding channel number. The receiving end directly completes the correspondence between AP data and channel by the position of the marker bits. The data structure supports intra-frame hierarchical transmission: when it is only necessary to determine whether an AP signal is present, a frame of data only transmits the frame header check bit and the full-channel AP trigger marker segment, truncating the AP waveform data segment to complete the simplified mode transmission.

[0009] Furthermore, in the full-channel AP trigger flag segment, the high-level bit indicates that the corresponding sampling channel has detected an AP signal, and the low-level bit indicates that the corresponding sampling channel has not detected an AP signal.

[0010] Furthermore, for sampling channels arranged in an M-row N-column array, data is transmitted in rows: first, the frame header check bit and the full-channel AP trigger flag segment of each row are transmitted row by row. After the flag segments of all rows are transmitted, the AP waveform data segment corresponding to each row is transmitted row by row. The full-channel AP trigger flag segment of each row contains N bits of data, corresponding to the N sampling channels of that row.

[0011] This invention also protects a neural action potential acquisition and transmission circuit for implementing the above-mentioned serial transmission data structure, including N parallel sampling channels, a data shift register chain, a marker shift register chain, and a timing control module; each sampling channel is used to acquire neural action potential signals, generate compressed AP waveform data from the detected AP signals, and synchronously output a 1-bit AP detection marker signal; the data shift register chain is composed of N data registers cascaded in series, each data register corresponding to the compressed AP waveform data output by one sampling channel, and is driven serially by a first clock signal; the marker shift register chain is composed of N marker registers cascaded in series, each marker register corresponding to the AP detection marker signal output by one sampling channel, and is driven serially by a second clock signal; the timing control module is used to control the output timing of the two shift register chains: first, drive the marker shift register chain to output the full-channel AP trigger marker segment, and then, according to the number of valid bits in the marker segment, drive the data shift register chain to output the corresponding number of AP waveform data groups.

[0012] Furthermore, the circuit also includes a multiplexer; for an array sampling channel with M rows and N columns, each row is independently configured with a set of data shift register chains and a flag shift register chain; the input of the multiplexer is connected to the output of each row of shift register chains, and the output serves as the serial data output of the circuit, used to select and output row by row under the control of the timing control module.

[0013] Furthermore, the timing control module is also used to drive the flag shift register chain to complete the flag segment output in the working mode of only detecting the presence or absence of AP, without starting the transmission of the data shift register chain, and directly ending the current frame.

[0014] Finally, this invention protects a method for serial transmission of neural action potentials, comprising the following steps: S1. Parallel acquisition of neural signals from N sampling channels, generation of compressed AP waveform data for channels that detect AP signals, and simultaneous generation of N full-channel AP trigger flags, with each flag corresponding to whether an AP signal is detected in a channel; S2. Serial output of 1 fixed level bit as frame header check bit to complete frame synchronization; S3. Serially output the N-bit full-channel AP trigger flag; S4. Based on the number of valid bits in the AP trigger flag of the full channel, the compressed AP waveform data of the corresponding channel is output serially. The output order is consistent with the channel number order. The receiving end directly completes the correspondence between AP data and channel by the position of the flag bits. When working in the mode of only detecting the presence or absence of AP, the current frame transmission ends after step S3 is completed, and step S4 is not executed.

[0015] Furthermore, for the sampling channels distributed in an M-row N-column array, steps S2-S3 are executed row by row to complete the transmission of the marker segments for all rows, and then step S4 is executed row by row to complete the transmission of the AP waveform data for all rows.

[0016] (III) Beneficial Effects Compared with the prior art, the present invention has the following significant technical advancements: The addressing overhead is extremely low, and the channel matching logic is extremely simple. This invention uses a full-channel bitmap-style tagging segment to replace the event-specific address field. Only N bits are used to cover all channel states at once, allowing the receiver to complete channel pairing based solely on bit positions, without needing to parse the address field of each data packet. Taking 1024 channels as an example, only 1024 bits are needed per frame to complete full-channel addressing, far lower than the total overhead of multiple sets of address fields corresponding to multiple events in the AER protocol. The more channels there are, the more significant the overhead savings become.

[0017] Intra-frame hierarchical transmission allows for dynamic power consumption adaptation. This invention inherently supports both full-waveform transmission and trigger-only transmission modes within the same frame structure: waveform data segments can be directly truncated when only AP detection is required, reducing the data rate to below 3% of the full-waveform mode. Taking a 1024-channel, 16kSPS sampling rate, 12-bit sampling precision, and 32-bit compressed waveform as an example, the trigger-only mode data rate is only 0.103Mbps, far lower than the 3.482Mbps of the compressed full-waveform mode, allowing for dynamic switching of power consumption levels according to application requirements.

[0018] The circuit structure is simple and the array is highly scalable. This invention is implemented using pure digital standard cells. The dual shift register chain architecture does not require complex circuits such as asynchronous arbitration and FIFO buffering, resulting in low layout overhead and strong process compatibility. For two-dimensional array electrodes, expansion can be completed simply by selecting row by row using a multiplexer. The architecture is highly scalable and adaptable to large-scale neural acquisition scenarios with thousands of channels.

[0019] The data rate dynamically adapts to the AP density. The total output data is positively correlated with the number of detected AP signals. When there is no AP signal, only the marker segment needs to be transmitted, making full use of the temporal sparsity of AP signals. The average data rate is much lower than that of fixed-length transmission schemes. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the original sampling and transmission circuit structure of a traditional multi-channel AP signal; Figure 2 A schematic diagram of a traditional compressed AP signal transmission circuit with an additional channel address; Figure 3 This is a schematic diagram of the dual-shift register type AP signal acquisition and transmission circuit described in this invention; Figure 4This is a schematic diagram of the serial data frame structure under the three AP triggering states of the present invention; Figure 5 This is a schematic diagram of the array-type M-row N-column AP signal acquisition and transmission circuit of the present invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0022] Example 1: Single-group 4-channel neural action potential acquisition and transmission circuit This embodiment is a basic single-group multi-channel acquisition and transmission circuit, refer to the attached diagram. Figure 3 This is used to explain the core working principle of the present invention.

[0023] This embodiment uses four sampling channels as an example to illustrate the principle. In practical applications, the number of channels can be flexibly expanded according to requirements. The entire circuit is implemented using standard digital CMOS technology and requires no special components.

[0024] 1. Circuit structure details The circuit in this embodiment mainly consists of four parts: a sampling channel group, a data shift register chain, a flag shift register chain 300, and a timing control module.

[0025] The sampling channel group comprises four parallel and independent sampling channels. Each sampling channel integrates a preamplifier low-noise amplifier, a bandpass filter, a 12-bit analog-to-digital converter (ADC), an AP detection unit, and a data compression unit. The preamplifier amplifies the microvolt-level AP signal acquired by the electrodes to the volt level. The bandpass filter removes field potential and high-frequency noise. The ADC uses a 16kSPS sampling rate to digitize the filtered signal. The AP detection unit uses a threshold + slope joint judgment method. When the sampled value exceeds a preset threshold and the rising edge slope matches the AP waveform characteristics, it is determined to be a valid AP signal, and a 1-bit high-level detection flag is set simultaneously. The data compression unit extracts four types of feature parameters from the valid AP waveform: peak amplitude, rise time, fall time, and half-peak width. Each type of parameter is encoded as 8 bits, and the final output is 32-bit compressed AP waveform data, retaining all the features required for waveform classification.

[0026] The data shift register chain consists of four 32-bit data registers cascaded in series. Each register corresponds to a sampling channel, and so on. The parallel input of each data register is connected to the compressed data output of the corresponding sampling channel. All registers share the data shift clock CLK, and the data is shifted bit by bit on the rising edge of CLK, and finally output serially from the DATA port.

[0027] The tag shift register chain consists of four 1-bit tag registers cascaded in series. The input of each tag register is connected to the AP detection tag output of the corresponding sampling channel. All registers share the tag shift clock CLK_E. The tag bits are shifted bit by bit on the rising edge of CLK_E and finally output serially from the DATA_E port.

[0028] Timing control module: The core control unit of the circuit, with built-in frame counter, flag bit statistics unit and clock gating unit, is used to generate clock signals and output enable signals for two shift registers and control the output timing of the entire frame of data.

[0029] 2. Complete single-frame transmission timing process In this embodiment, the time window for a single frame of data is 1ms, which matches the refractory period characteristics of neurons, ensuring that an AP signal appears at most once in a single channel within a single frame. Taking the scenario of E<3:0>=4'b1101 (i.e., AP signals are detected in channels 1, 3, and 4, but not in channel 2) as an example, the complete transmission process is divided into three stages: Frame header synchronization stage: The timing control module first outputs a high level for one clock cycle to the output port as a frame header check bit, which is used by the receiving end to identify the frame start position and solve the frame boundary alignment problem in serial transmission.

[0030] Marker segment output stage: After the frame header is output, the timing control module sends four consecutive clock pulses to the CLK_E port. The first clock edge outputs E. <0> (Channel 1 marker, high level 1), output E on the second clock edge. <1> (Channel 2 marker, low level 0), output E on the 3rd clock edge. <2> (Channel 3 marker, high level 1), output E on the 4th clock edge. <3> (Channel 4 marker, high level 1). After 4 clock cycles, all 4-bit full-channel AP trigger marker segments will be output.

[0031] During the waveform data segment output stage: The statistical unit inside the timing control module counts the number of high-level signals in the marker segment; in this scenario, the count value is 3. Then, 3 × 32 = 96 consecutive clock pulses are sent to the CLK port. The data shift register chain sequentially outputs 32-bit compressed AP waveform data for channels 1, 3, and 4, with the output order strictly corresponding to the channel numbers from smallest to largest. Data from channel 2, which has no AP signal, is automatically skipped, thus not occupying transmission bandwidth.

[0032] If no AP signal is detected on any of the channels (E<3:0>=4'b0000), the count value of the statistical unit is 0. After the timing control module finishes outputting the marker segment, it directly turns off the CLK clock and does not output any waveform data. The entire frame contains only 1 frame header + 4 marker bits, with a total length of only 5 bits.

[0033] 3. Data parsing process at the receiving end The receiving end does not require complex address decoding circuitry; it can complete the correspondence between data and channels simply through shift counting. The parsing steps are as follows: The serial input level is continuously monitored. When a continuous high level is detected, it is determined to be the start of the frame header, and the frame reception process is started. Read 4 bits of data continuously after the frame header and store them in the flag register to obtain the complete state of E<3:0>; Traverse the marker bits from least significant bit to most significant bit, record the channel number corresponding to all bits with a value of 1, and form a valid channel queue; The subsequent serial data is read sequentially, and after each 32 bits are read, it is allocated to the next channel in the valid channel queue, until the queue is traversed and the parsing of the current frame ends.

[0034] 4. Trigger-only mode operation process When the application only needs to monitor the neuron firing state and does not need to reconstruct the AP waveform, the circuit can be switched to trigger-only mode via the configuration register. In this mode, after the timing control module outputs the frame header and 4-bit marker segment, it directly terminates the transmission of the current frame without starting the clock of the data shift register chain, and the data length of each frame is fixed at 5 bits. The receiving end can obtain the presence or absence status of AP for all channels after parsing the marker segment, without needing to receive subsequent waveform data, thus significantly reducing transmission power consumption.

[0035] Example 2: 32×32 array 1024-channel neural action potential acquisition and transmission circuit This embodiment is a large-scale array expansion scheme, see the corresponding instruction manual. Figure 5 Designed for implantable high-density neural electrode arrays, it has a total of 1024 channels and adopts a 32-row × 32-column two-dimensional array layout, which is perfectly matched with the physical structure of planar electrodes.

[0036] 1. Details of array-type circuit architecture The entire acquisition circuit is divided into 32 independent row acquisition units. The structure of each row acquisition unit is exactly the same as in Embodiment 1, including 32 sampling channels, one 32-bit data shift register chain, and one 32-bit flag shift register chain. The circuit is globally configured with one 2-to-1 multiplexer (MUX) and a global timing control module. The MUX has two sets of inputs: a marker shift output bus and a data shift output bus for all rows. The outputs are combined into a single serial data output port SDO. The global timing control module has a built-in 5-bit row address decoder, which can sequentially select any row among the 32 rows to control the shift clock and output path of the MUX for that row.

[0037] The architecture adopts a fully array modular design, with each row of circuits having the same structure. The layout can be quickly expanded by copying cells without redesigning the core circuit, and it can be adapted to electrode arrays of different sizes.

[0038] 2. Full-frame batch transmission process This embodiment adopts a transmission order of "batch processing of marker segments first, followed by batch processing of data segments," which facilitates the receiving end to first obtain the discharge distribution of the entire array and then parse the waveform data as needed. The complete transmission process of a single frame is as follows: Full array marker segment transmission phase: The global timing control module selects rows sequentially starting from row 0. When row i is selected, a 1-bit high-level frame header is output first, followed by the 32-bit AP trigger flag Ei<31:0> for that row; after completing one row, the module switches to row i+1 and repeats the above operation until all 32 rows of marker segments have been output. After this phase, the receiver can obtain the complete 1024-channel AP trigger states, completing the full array monitoring of neuron firing distribution.

[0039] Full array data segment transmission phase: After all marker segments have been transmitted, the global timing control module selects rows one by one, starting from row 0. When selecting row i, it outputs the corresponding number of 32-bit compressed AP waveform data based on the number of valid bits in the marker segment of that row; rows without AP signals are skipped directly without occupying transmission time. After traversing row by row, the transmission of all data in the current frame is complete.

[0040] The advantage of this transmission sequence is that if the receiving end only needs discharge distribution information, it can directly terminate the reception of the current frame after the marker segment is transmitted and enter the next frame without waiting for waveform data transmission, thus further reducing system-level power consumption.

[0041] 3. Quantitative Performance Comparison Taking the parameters of 1024 channels, 16kSPS sampling rate, 12-bit sampling resolution, and 32-bit compressed data per AP in this embodiment as an example, the comparison of single-frame data volume and total data rate for the three transmission modes is as follows: Original sampling transmission mode: 12 bits of sampled data are output per channel per frame, the total data volume of a single frame is 1024×12=12288 bits, and the total data rate is 12288×16k=196.6Mbps; The compressed waveform mode of this invention, calculated based on a typical AP occurrence rate (an average of 10% of channels triggering APs per frame), has a total data volume of 32×(1+32)+1024×10%×32=4320 bits per frame, with a total data rate of approximately 3.482Mbps, which is only 1.77% of the original mode. The invention features a trigger-only mode: a single frame transmits only the marker segment, with a total data volume of 32×(1+32)=1056 bits and a total data rate of approximately 0.103Mbps, which is only 0.05% of the original mode, reducing transmission power consumption by three orders of magnitude.

[0042] In large-scale implantable devices with thousands of channels, the trigger mode alone can meet the application needs of most neuromodulation and discharge screening, and can reduce the proportion of transmission power consumption to the total power consumption of the chip from 50% to less than 5%, significantly improving the device's battery life.

[0043] 4. Process and Expansion Specifications This embodiment is implemented entirely using standard digital logic units, eliminating the need for analog calibration circuits, asynchronous arbitration circuits, or large-capacity FIFOs. It operates normally under mainstream CMOS processes ranging from 180nm to 7nm, demonstrating strong process compatibility. The shift register chain is laid out flat along the row direction of the electrode array, corresponding one-to-one with the electrode pads, resulting in low layout and routing complexity, suitable for mass production. If further expansion of the channel count is required, only the number of rows and columns needs to be increased, along with the corresponding row cells. The global timing control module and MUX structure remain unchanged, exhibiting extremely high scalability and supporting ultra-large-scale neural electrode arrays with tens of thousands of channels.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A serial transmission data structure for neural action potential signals, characterized in that, Each frame of data contains, in the order of transmission, a frame header check bit, a full-channel AP trigger flag segment, and an AP waveform data segment; The frame header check bit is a fixed level of 1 bit, used to identify the start of a frame of data; the number of bits in the full-channel AP trigger flag segment is equal to the total number of sampling channels N, and each bit of binary data corresponds to one sampling channel, used to identify whether the corresponding sampling channel detects a neural action potential (AP) signal in the current sampling period; The AP waveform data segment contains several sets of compressed AP waveform data. The number of AP waveform data sets is equal to the number of valid bits in the full-channel AP trigger marker segment that indicate "AP signal detected". Each set of AP waveform data is arranged sequentially according to the corresponding channel number. The receiving end directly completes the correspondence between AP data and channel by the position of the marker bits. The data structure supports intra-frame hierarchical transmission: when it is only necessary to determine whether an AP signal is present, a frame of data only transmits the frame header check bit and the full-channel AP trigger marker segment, and the AP waveform data segment is truncated.

2. The serial transmission data structure according to claim 1, characterized in that, In the full-channel AP trigger flag segment, a high-level bit indicates that the corresponding sampling channel has detected an AP signal, and a low-level bit indicates that the corresponding sampling channel has not detected an AP signal.

3. The serial transmission data structure according to claim 1, characterized in that, For sampling channels arranged in an M-row N-column array, data is transmitted in rows: first, the frame header check bit and the full-channel AP trigger flag segment of each row are transmitted row by row. After the flag segments of all rows are transmitted, the AP waveform data segments of each row are transmitted row by row. Each row's full-channel AP trigger flag segment contains N bits of data, corresponding to the N sampling channels of that row.

4. A neural action potential signal acquisition and transmission circuit, used to implement the serial transmission data structure of claim 1, characterized in that, It includes N parallel sampling channels, a data shift register chain, a tag shift register chain, and a timing control module; each sampling channel is used to acquire neural action potential signals, generate compressed AP waveform data from the detected AP signals, and synchronously output a 1-bit AP detection tag signal; The data shift register chain consists of N data registers cascaded in series. Each data register stores compressed AP waveform data output from a sampling channel and is driven serially by the first clock signal. The tag shift register chain consists of N tag registers cascaded in series. Each tag register stores an AP detection tag signal output from a sampling channel and is driven serially by a second clock signal. The timing control module is used to control the output timing of the two shift register chains: first, it drives the marker shift register chain to output the full-channel AP trigger marker segment, and then, according to the number of valid bits in the marker segment, it drives the data shift register chain to output the corresponding number of AP waveform data groups.

5. The neural action potential signal acquisition and transmission circuit according to claim 4, characterized in that, It also includes a multiplexer; for an array sampling channel with M rows and N columns, each row is independently configured with a set of data shift register chains and a flag shift register chain; the input of the multiplexer is connected to the output of each row of shift register chains, and the output is used as the serial data output of the circuit, which is used to select and output row by row under the control of the timing control module.

6. The neural action potential signal acquisition and transmission circuit according to claim 4, characterized in that, The timing control module is also used to drive the flag shift register chain to complete the flag segment output in the working mode of only detecting the presence or absence of AP, without starting the transmission of the data shift register chain.

7. A method for serial transmission of neural action potential signals, characterized in that, Includes the following steps: S1. Parallel acquisition of neural signals from N sampling channels, generation of compressed AP waveform data for channels that detect AP signals, and simultaneous generation of N full-channel AP trigger flags, with each flag corresponding to whether an AP signal is detected in a channel; S2. Serial output of 1 fixed level bit as frame header check bit; S3. Serially output the N-bit full-channel AP trigger flag; S4. Based on the number of valid bits in the AP trigger flag of the full channel, the compressed AP waveform data of the corresponding channel is output serially. The output order is consistent with the channel number order. The receiving end directly completes the correspondence between AP data and channel by the position of the flag bits. When working in the mode of only detecting the presence or absence of AP, the current frame transmission ends after step S3 is completed, and step S4 is not executed.

8. The serial transmission method according to claim 7, characterized in that, For a sampling channel with an M-row N-column array distribution, first execute steps S2-S3 row by row to complete the transmission of the marker segments for all rows, and then execute step S4 row by row to complete the transmission of the AP waveform data for all rows.