Multi-channel AFE signal acquisition method and system based on adaptive adjustment

Abstract

The invention discloses a multi-channel AFE signal acquisition method and system based on adaptive adjustment, and the method comprises the steps: synchronously acquiring multi-channel analog signals through a multi-channel parallel AFE module, and configuring an independent programmable gain amplifier and an anti-aliasing filter for each channel; extracting time-frequency domain characteristic parameters of each channel signal in real time; dynamically generating an optimal adjustment strategy based on the time-frequency domain characteristic parameters, and adaptively adjusting the gain coefficient, the filtering bandwidth and the sampling rate parameter of each channel through a closed-loop feedback mechanism; and abnormal signal detection is realized by adopting an inter-channel cross validation algorithm, and automatic isolation of a fault channel and intelligent switching of a redundant channel are triggered. Through dynamic parameter intelligent adjustment, multi-channel cooperative fault tolerance and energy efficiency optimization architecture, the problems of signal distortion, response lag and high power consumption in a traditional multi-channel signal acquisition system are solved, and the signal fidelity and the system reliability are remarkably improved.

Description

Technical Field

[0001] This application relates to the technical field of signal acquisition, and particularly to a multi-channel AFE signal acquisition method and system based on adaptive adjustment. Background Art

[0002] In the scenario of a lithium battery management system (BMS), a multi-channel analog front end (AFE) needs to collect heterogeneous signals such as voltage, current, and temperature in real time to support battery health state assessment or mechanical fault warning. However, the contradiction between the dynamic signal characteristics under complex working conditions (such as high-frequency ripple during lithium battery fast charging and transient shock during equipment vibration) and the multi-channel collaboration requirements is becoming increasingly prominent. The traditional multi-channel AFE system adopts a fixed parameter configuration and an independent channel processing mode, resulting in serious limitations in signal fidelity and system efficiency.

[0003] Taking the monitoring of a lithium battery pack as an example, the voltage difference between individual cells in a battery cluster may cause the coexistence of weak signals and high-amplitude noise. The fixed gain strategy of the traditional AFE easily saturates some channels or deteriorates the signal-to-noise ratio, masking the early lithium plating characteristics. More severely, when the system encounters sudden interference (such as a transient pulse caused by a sudden change in battery contact impedance), the existing technology is difficult to complete noise classification and parameter adaptive adjustment within a millisecond-level window. It often sacrifices bandwidth or sampling rate to avoid signal overflow, but loses the key frequency domain information for fault prediction. Such problems are particularly prominent in resource-constrained edge devices (such as embedded BMS). Although the conservative configuration of high sampling rate and wide bandwidth can cover most scenarios, it leads to a sharp increase in computational load and power consumption, ultimately restricting the practicality of the system in long-term monitoring. When there are multiple aging single cells (low-frequency capacity attenuation) and local micro-shorts (high-frequency impedance mutation) in a lithium battery pack simultaneously, the static parameter strategy cannot achieve differential configuration of gain-bandwidth between channels, resulting in increased aliasing of effective signals and noise. Summary of the Invention

[0004] To solve the above problems, an embodiment of the present invention provides a multi-channel AFE signal acquisition method based on adaptive adjustment. The method includes:

[0005] Synchronously collect multiple analog signals through a multi-channel parallel AFE module, and configure an independent programmable gain amplifier and an anti-aliasing filter for each channel;

[0006] Extract the time-frequency domain characteristic parameters of each channel signal in real time;

[0007] Dynamically generate an optimal adjustment strategy based on the time-frequency domain characteristic parameters, and adaptively adjust the gain coefficient, filter bandwidth, and sampling rate parameters of each channel through a closed-loop feedback mechanism;

[0008] The cross-channel cross-validation algorithm is adopted to implement abnormal signal detection, trigger the automatic isolation of faulty channels and the intelligent switching of redundant channels; the optimized multi-channel signals are subjected to time-domain alignment and phase compensation.

[0009] Furthermore, the method for the dynamic adjustment strategy includes:

[0010] Construct a signal quality evaluation model based on a convolutional neural network, and input a time-frequency domain feature matrix composed of time-frequency domain feature parameters to generate the quality scores of each channel; the model architecture of the signal quality evaluation model includes an input layer, a feature extraction branch, a feature fusion layer, and an output layer;

[0011] Establish a dynamic parameter adjustment rule base, including a noise scenario classification table and an optimal gain-bandwidth-sampling rate combination matrix under different noise environments;

[0012] Adopt a reinforcement learning algorithm to optimize the adjustment strategy online, and dynamically update the rule base parameters according to the historical adjustment effect.

[0013] Furthermore, the extraction of time-frequency domain features adopts hybrid signal processing technology; the time-domain features include peak-to-peak value, form factor, and zero-crossing rate; the frequency-domain features are obtained by combining sliding window FFT with wavelet packet decomposition; digital down-conversion technology is used to achieve the feature extraction of bandpass signals.

[0014] Furthermore, the method includes:

[0015] Calculate the mutual information entropy and phase consistency index between adjacent channel signals;

[0016] When the mutual information entropy between the channel signal and the adjacent channel signal is less than the set mutual information entropy threshold and the phase consistency index is greater than the phase locking value threshold, it is determined as an abnormal channel;

[0017] Activate the preset redundant channel to replace the abnormal channel, and keep the total number of acquisition channels constant;

[0018] Mutual information entropy The calculation method of the calculation includes:

[0019]

[0020] In the formula, N is the total number of groups of the channel signal group, p(x i ,y i ) is the joint probability distribution, x i and y i are the channel signal and the adjacent channel signal of the i-th group, p(x i ) and p(y i ) are the marginal probability distributions;

[0021] The calculation method of the phase consistency index PLV includes:

[0022]

[0023] Wherein, φ1(t) and φ2(t) are the instantaneous phases of two signals at time point t, and j is the imaginary unit.

[0024] Furthermore, it also includes a multi - working - mode intelligent switching mechanism, which includes a mode switching mechanism, an environment monitoring mode, a high - precision mode, and a low - power mode;

[0025] Environment monitoring mode: Prioritize wide dynamic range and adopt segmented non - linear gain control;

[0026] High - precision mode: Enable oversampling technology and digital post - filtering processing;

[0027] Low - power mode: Dynamically turn off the bias circuits and high - speed ADC modules of unactivated channels;

[0028] The mode switching mechanism realizes continuous signal phase through a digital isolator and synchronous clock re - configuration.

[0029] Furthermore, phase compensation adopts an adaptive delay - locked technology to generate a phase error signal based on the reference channel signal and finely adjusts the sampling clock of each channel through a digital delay line;

[0030] The method for generating the phase error signal includes:

[0031]

[0032] Wherein, M is the window length; R g,ref (k) is the similarity measure between channel g and the reference channel at time - shift point k; w g (n) is the delay control weight of channel g at the nth iteration; μ is the convergence factor; Δφ g (n) is the phase error signal; x′ g (n) is the signal of channel g after current delay compensation; m is the number of the current signal.

[0033] Furthermore, it also includes a dynamic calibration subsystem:

[0034] Periodically inject a standard test signal for channel characteristic calibration;

[0035] Establish a compensation look - up table for the frequency response and non - linear error of each channel;

[0036] Adopt a temperature sensor to monitor the ambient temperature change and trigger temperature drift compensation.

[0037] Furthermore, the optimization measures implemented in the FPGA include:

[0038] Implement asynchronous sampling rate conversion using a multi-clock domain design;

[0039] Configure a dedicated hardcore IP for parallel digital filtering processing;

[0040] Dynamically update the signal processing algorithm through partial reconfiguration technology.

[0041] A multi-channel AFE signal acquisition system based on adaptive adjustment, the system includes:

[0042] Channel acquisition module, the channel acquisition module synchronously acquires multiple analog signals through a multi-channel parallel AFE module, and each channel is configured with an independent programmable gain amplifier and an anti-aliasing filter;

[0043] Feature extraction module, the feature extraction module extracts the time-frequency domain feature parameters of each channel signal in real time;

[0044] Adaptive adjustment module, the adaptive adjustment module dynamically generates an optimal adjustment strategy based on the time-frequency domain feature parameters, and adaptively adjusts the gain coefficient, filter bandwidth and sampling rate parameters of each channel through a closed-loop feedback mechanism;

[0045] Abnormal signal processing module, the abnormal signal processing module uses an inter-channel cross-validation algorithm to implement abnormal signal detection, triggering automatic isolation of faulty channels and intelligent switching of redundant channels;

[0046] Phase compensation module, the phase compensation module performs time-domain alignment and phase compensation on the optimized multi-channel signals.

[0047] The technical effects and advantages of a multi-channel AFE signal acquisition method and system based on adaptive adjustment provided by the present invention:

[0048] Through dynamic parameter intelligent adjustment, multi-channel collaborative fault tolerance, and energy efficiency optimization architecture, the present invention solves the problems of signal distortion, response lag, and high power consumption in traditional multi-channel signal acquisition systems, and significantly improves the signal fidelity and system reliability in complex noise scenarios. The present invention integrates signal time-frequency domain feature extraction and reinforcement learning strategies, constructs a dynamic parameter rule base, and real-time optimizes the gain, bandwidth, and sampling rate combinations of each channel, so that in the coexistence scenario of strong interference (such as lithium plating monitoring of lithium batteries), it can automatically suppress high-frequency noise and lock the characteristic frequency band, significantly improve the fidelity of weak signals, and avoid signal saturation or SNR degradation caused by traditional fixed parameters; based on the joint criterion of cross-channel mutual information entropy and phase consistency, combined with adaptive delay locking technology, it realizes inter-channel anomaly detection and time-domain synchronization compensation, accurately distinguishes real fault signals from transient interference, quickly isolates the failed channel and switches the redundant link, and ensures the continuous and reliable operation of the system under sudden interference; adopts an environment-aware driven multi-mode dynamic switching strategy (high-precision / balanced / low-power mode), combined with the FPGA multi-clock domain parallel processing architecture, can dynamically allocate hardware resources according to scene requirements, while maintaining high sampling accuracy, significantly reducing the system power consumption and extending the battery life of industrial monitoring equipment. Brief Description of the Drawings

[0049] Figure 1 It is a flowchart of a multi-channel AFE signal acquisition method based on adaptive adjustment in Embodiment 1;

[0050] Figure 2 It is a schematic diagram of the model architecture of the signal quality evaluation model in Embodiment 1;

[0051] Figure 3 It is a schematic connection diagram of a multi-channel AFE signal acquisition system based on adaptive adjustment in Embodiment 2. Detailed Embodiments

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] Embodiment 1:

[0054] Please refer to Figure 1 As shown, the multi-channel AFE signal acquisition method based on adaptive adjustment in this embodiment includes:

[0055] Synchronously collect multiple analog signals through a multi-channel parallel AFE module, and configure an independent programmable gain amplifier and an anti-aliasing filter for each channel;

[0056] Extract the time-frequency domain characteristic parameters of each channel in real time;

[0057] Dynamically generate an optimal adjustment strategy based on the time-frequency domain characteristic parameters, and adaptively adjust the gain coefficient, filter bandwidth, and sampling rate parameters of each channel through a closed-loop feedback mechanism;

[0058] Adopt an inter-channel cross-validation algorithm to detect abnormal signals, trigger the automatic isolation of faulty channels and the intelligent switching of redundant channels; Align the optimized multi-channel signals in the time domain and compensate for the phase.

[0059] The method for dynamically adjusting the strategy includes:

[0060] Construct a signal quality evaluation model based on a convolutional neural network, and input the time-frequency domain characteristic matrix composed of time-frequency domain characteristic parameters to generate the quality score of each channel; The time-frequency domain characteristic parameters include dynamic range, noise floor, and main frequency component distribution;

[0061] Such as Figure 2 As shown, the model architecture of the signal quality evaluation model includes an input layer, a feature extraction branch, a feature fusion layer, and an output layer.

[0062] Input layer:

[0063] Receive a two-dimensional matrix composed of time-frequency domain characteristic parameters. The two-dimensional matrix, namely the time-frequency domain characteristic matrix, includes:

[0064] Row dimension: The number of signal channels (for example, a 16-channel system corresponds to 16 rows);

[0065] Column dimension: Characteristic parameters (preset characteristics such as dynamic range, noise floor, and main frequency component distribution).

[0066] The feature extraction branch includes a time-domain feature branch and a frequency-domain feature branch;

[0067] Time-domain feature branch:

[0068] Use a 1D convolutional layer to extract the inter-channel correlation features, including a convolutional kernel and an activation function;

[0069] The convolutional kernel is 3 groups of learnable filters with different scales (narrow, medium, and wide time windows);

[0070] The activation function is a PReLU non-linear unit to prevent gradient disappearance;

[0071] Frequency-domain feature branch:

[0072] Use a 2D convolutional layer to process the spectral features, including:

[0073] Multi-scale spectrogram processing (main frequency band, harmonic region, and noise region), using frequency-domain attention mechanism to strengthen the features of key frequency bands.

[0074] The feature fusion layer fuses the time-domain and frequency-domain features through cross-channel connections, and uses a gating mechanism to dynamically allocate feature weights.

[0075] The output layer includes a fully connected layer, which maps the fused features to the quality score:

[0076] The output value range is 0 to 1 (0 indicates severe distortion, and 1 indicates ideal quality);

[0077] The quality score of each channel is generated independently, that is, each channel corresponds to a preset dedicated regression neuron.

[0078] The quality score generation method includes:

[0079] Normalize the feature matrix to eliminate the dimensional difference between channels;

[0080] The dual-branch CNN extracts the deep features of the time domain and frequency domain in parallel;

[0081] Feature fusion based on attention weights;

[0082] The quality score (Q value) of each channel is output through regression by the fully connected layer.

[0083] Establish a dynamic parameter adjustment rule library, including a noise scenario classification table and an optimal combination matrix of gain-bandwidth-sampling rate under different noise environments;

[0084] Exemplary:

[0085] Noise scenario classification table:

[0086] Scenario coding Noise type Feature identifier;

[0087] B01 Steady-state low-frequency noise Main frequency < 1kHz, dynamic range fluctuation < 3dB;

[0088] B02 Transient pulse interference Overload rate > 20%, duration < 50ms;

[0089] B03 Wideband random noise Noise floor > threshold and no significant main frequency.

[0090] Optimal combination matrix:

[0091] Scenario coding Gain coefficient Filter bandwidth Sampling rate Applicable conditions;

[0092] B01 ×300 20kHz 192kHz Q > 0.7 and main frequency is stable;

[0093] B02 ×80 5kHz 48kHz The Q value drops by more than 30% for 3 consecutive frames;

[0094] B03 ×150 10kHz 96kHz The noise floor exceeds the limit for 100 ms.

[0095] The matching logic of time-frequency domain feature parameters includes:

[0096] According to the quality score and noise classification result output by the CNN, select the nearest scene encoding;

[0097] Use Euclidean distance to calculate the matching degree between the current time-frequency domain feature and the scene in the rule library:

[0098] When the matching degree is higher than the matching degree threshold, directly call the preset parameters, otherwise trigger the online optimization of the reinforcement learning algorithm.

[0099] Use the reinforcement learning algorithm to optimize the adjustment strategy online and dynamically update the rule library parameters according to the historical adjustment effect.

[0100] The reinforcement learning algorithm framework includes state space definition, action space definition and reward function:

[0101] State space definition:

[0102] The current quality score of each channel;

[0103] Historical parameter adjustment records (the combination of the last 10 gains, bandwidths and sampling rates);

[0104] The change trend of environmental noise characteristics (statistical by sliding window).

[0105] The action space definition includes action type, adjustment granularity and constraint conditions;

[0106] Exemplary:

[0107]

[0108] Reward function:

[0109] Reward value = 0.6×ΔSNR + 0.2×stability coefficient - 0.2×parameter switching cost;

[0110] ΔSNR is the change in signal-to-noise ratio before and after adjustment;

[0111] Stability coefficient: the reciprocal of the standard deviation of the Q value for 5 consecutive frames;

[0112] Switching cost: the weighted sum of the adjustment amplitudes of gain, bandwidth and sampling rate.

[0113] The dynamic update mechanism of the rule library parameters includes:

[0114] When the new scenario matching fails, start reinforcement learning;

[0115] Record the parameter combinations and environmental characteristics that have been successfully adjusted;

[0116] Generate new scenario encodings and parameter matrices through cluster analysis;

[0117] After verification, insert into the rule library and establish the association mapping between the new and old scenarios.

[0118] The extraction of time-frequency domain features uses mixed signal processing technology:

[0119] Time domain features include peak-to-peak value, form factor, and zero-crossing rate;

[0120] The peak-to-peak signal is the absolute difference between the maximum and minimum values within one cycle, which is used to detect voltage mutations during the battery charging and discharging process (such as the instantaneous voltage drop caused by micro-short circuits) and combined with dynamic baseline calibration (to eliminate the open-circuit voltage drift).

[0121] The form factor is the ratio of the effective value of the signal to the absolute average value, which is used to quantify the distortion degree of the charge and discharge waveform (such as the sawtooth wave caused by PWM charging). An abnormal increase in the form factor may indicate electrode aging or blocked electrolyte diffusion.

[0122] The zero-crossing rate is the number of times the signal crosses the zero point per unit time (a dynamic threshold needs to be set to filter out noise), which is used to detect high-frequency oscillations (such as the micro-vibrations caused by dendritic growth inside the battery). A sudden increase in the zero-crossing rate may indicate loose mechanical structure or internal short circuit.

[0123] Frequency domain features are obtained through the combination of sliding window FFT and wavelet packet decomposition;

[0124] After segmenting and windowing the time domain signal (such as a Hanning window), it is converted into the frequency domain energy distribution through FFT; then the signal is decomposed layer by layer into sub-signals of different frequency bands through multi-scale wavelet basis functions (such as db6), and the energy entropy of each sub-band is calculated. The energy entropy calculation of the wavelet packet uses entropy value weighting to highlight the contribution of abnormal frequency bands, thereby obtaining frequency domain features.

[0125] Exemplary:

[0126] Early warning of lithium plating:

[0127] Feature combination:

[0128] Time domain feature: The voltage platform at the end of charging abnormally rises (the peak-to-peak value decreases).

[0129] Frequency domain feature: The energy in the 0.5 - 2 Hz frequency band continuously increases (reflecting the blocked diffusion of lithium ions).

[0130] Response strategy:

[0131] Reduce the charging current to trigger the intervention of the balancing circuit.

[0132] Adopt digital down-conversion technology to achieve feature extraction of band-pass signals;

[0133] Shift the high-frequency signal to the baseband through quadrature demodulation (local oscillator mixing), and cooperate with decimation to reduce the data rate. Thus, when facing the sensitive area of contact impedance change, reduce the sampling rate, reduce the amount of data processing, and while retaining the key features, reduce the consumption of edge computing resources.

[0134] The abnormal signal detection adopts the cross-channel correlation analysis method, and the method includes:

[0135] Calculate the mutual information entropy and phase consistency index between adjacent channel signals;

[0136] When the mutual information entropy between the channel signal and the adjacent channel signal is less than the set mutual information entropy threshold and the phase consistency index is greater than the phase locking value threshold, it is determined as an abnormal channel;

[0137] Activate the preset redundant channel to replace the abnormal channel and keep the total number of acquisition channels constant.

[0138] Mutual information entropy The calculation method of the calculation includes:

[0139]

[0140] In the formula, N is the total number of groups of channel signal groups, p(x i , y i ) is the joint probability distribution, x i and y i are the channel signal and the adjacent channel signal of the i-th group, p(x i ) and p(y i ) are the marginal probability distributions, that is, the independent probabilities of the channel signal and the adjacent channel signal.

[0141] The calculation method of the phase consistency index PLV includes:

[0142]

[0143] In the formula, φ1(t) and φ2(t) are the instantaneous phases of two signals at time point t, and j is the imaginary unit.

[0144] Multi-working mode intelligent switching mechanism, the multi-working mode intelligent switching mechanism includes a mode switching mechanism, an environmental monitoring mode, a high-precision mode and a low-power mode;

[0145] Environmental monitoring mode: Give priority to a wide dynamic range and adopt segmented non-linear gain control;

[0146] The environmental monitoring mode is applicable to the transient process of large - current charging and discharging of lithium batteries (such as fast - charge activation and high - load operation). The segmented non - linear gain control includes a primary gain stage and a secondary gain stage;

[0147] Primary gain stage: Based on the baseline tracking of the total battery - pack voltage (2.5V - 4.5V), dynamically adjust the PGA bias voltage to ensure that the ADC input is always in the linear region;

[0148] Secondary gain stage: When it is detected that the voltage across the current - sampling resistor exceeds the preset safety threshold (corresponding to a current above 50A), switch to the logarithmic compression mode to suppress the amplifier saturation caused by current mutation.

[0149] High - precision mode: Enable oversampling technology and digital post - filtering processing;

[0150] The high - precision mode is applicable to the assessment of the state of health (SOH) of lithium batteries and the static measurement of open - circuit voltage (OCV).

[0151] The oversampling technology includes:

[0152] Perform 512 - fold oversampling on the single - cell battery voltage (3.0V - 4.2V) using a Δ - Σ ADC, and cooperate with a Sinc 4 filter to improve the effective resolution to 20 bits;

[0153] Eliminate the reference - voltage error caused by temperature drift through dynamic reference calibration.

[0154] The digital post - filtering processing includes: Cascading an adaptive FIR filter bank to suppress the common - mode interference introduced by PWM dimming;

[0155] An FFT - based spectrum monitoring module to automatically identify and eliminate specific - band noise (such as the 217Hz burst interference of the GSM module).

[0156] Low - power mode: Dynamically turn off the bias circuits and high - speed ADC modules of unactivated channels.

[0157] The low - power mode is suitable for device standby sleep and small - current trickle charging.

[0158] The mode - switching mechanism realizes continuous signal phase through a digital isolator and synchronous clock re - configuration, avoiding transient jumps in voltage and current data.

[0159] The multi - working - mode intelligent switching mechanism can be applied to the battery management module (BMS) of consumer electronics products (such as smartphones and laptops). By dynamically adapting to the signal - acquisition requirements under different working conditions, it realizes an optimal balance among accuracy, anti - interference, and power consumption.

[0160] Phase compensation adopts an adaptive delay-locked technique, generates a phase error signal based on the reference channel signal, and finely tunes the sampling clocks of each channel through a digital delay line;

[0161] The reference channel selects the channel with the optimal signal quality (such as a channel with SNR > 60 dB) as the benchmark, injects a standard test signal (such as a 1 kHz sine wave) into all channels at startup; and uses the least mean square algorithm to dynamically optimize the delay compensation amount of each channel.

[0162] The generation method of the phase error signal includes:

[0163]

[0164] In the formula, M is the window length. A short window (such as N = 64) can quickly respond to dynamic changes and is suitable for high-frequency signals, while a long window (such as N = 1024) can improve the anti-noise ability and is suitable for low-frequency signals; k is the number of time shift points, which is the time shift amount of the reference signal relative to the compensated signal; R g,ref (k) is the similarity measure between channel g and the reference channel at time shift point k; w g (n) is the delay control weight of channel g at the nth iteration; μ is the convergence factor, which is used for the adaptive step coefficient to control the update rate of the weight; Δφ g (n) is the phase error signal, that is, the real-time phase deviation signal between channel g and the reference channel; x′ g (n) is the signal of channel g after the current delay compensation; m is the number of the current signal.

[0165] The dynamic calibration subsystem includes:

[0166] Periodically inject a standard test signal for channel characteristic calibration;

[0167] Establish a compensation look-up table for the frequency response and nonlinear error of each channel;

[0168] Use a temperature sensor to monitor the environmental temperature change and trigger temperature drift compensation.

[0169] During the calibration phase, the built-in signal source injects a test signal covering the main working frequency band (such as a sweep from low frequency to high frequency) into each channel, and analyzes its frequency response characteristics and nonlinear error after collecting the response data.

[0170] For example, when a certain channel shows amplitude attenuation at a typical frequency point, the dynamic calibration subsystem automatically generates the corresponding gain compensation coefficient and stores it in the look-up table, so that the coefficient can be applied in real time to correct the signal amplitude in subsequent measurements.

[0171] For the phase deviation, by analyzing the cross-correlation peak offset between the test signal and the reference signal, calculate the phase compensation value to ensure the synchronization accuracy between multiple channels.

[0172] Regarding the temperature drift problem, the dynamic calibration subsystem deploys temperature sensors at key circuit nodes and establishes an association model between temperature changes and channel characteristics.

[0173] When it is detected that the temperature change exceeds the preset temperature threshold, dynamic compensation is triggered: predicting the current temperature drift amount based on historical calibration data and fine-tuning the compensation parameters in the compensation lookup table.

[0174] For example, the gain attenuation trend of a certain channel at high temperatures has been pre-modeled. When the temperature rises to a specific range, the system automatically increases the gain compensation coefficient to offset the influence of temperature drift. This process does not require interrupting normal measurements and only requires millisecond-level calculations to complete parameter updates.

[0175] To balance calibration efficiency and system resource occupancy, a hierarchical trigger strategy is designed. In the normal mode, full-band calibration is performed at a fixed period, while in the case of sudden environmental changes or when errors exceed the limit, fast local calibration is started, and only the parameters of the affected frequency band are updated.

[0176] The dynamic calibration subsystem can significantly reduce the amplitude-phase differences between multiple channels and maintain measurement consistency within a wide temperature range. The test signal injection and data processing are realized through a dedicated hardware acceleration module, ensuring that the calibration process is completed efficiently without interfering with normal signal acquisition. By continuously optimizing the compensation parameters, the dynamic calibration subsystem can adapt to device aging and environmental fluctuations, breaking through the limitations of traditional fixed calibration schemes.

[0177] The optimization measures implemented in the FPGA include:

[0178] Adopting a multi-clock domain design to achieve asynchronous sampling rate conversion;

[0179] Exemplarily:

[0180] In the communication baseband processing scenario, the input signal enters the FPGA at a sampling rate of 48 kHz, while the core algorithm needs to operate at 96 kHz. A multi-clock domain architecture is adopted to achieve asynchronous sampling rate conversion: first, the input data is buffered through an asynchronous FIFO, and a circular buffer constructed by a dual-port RAM is used to achieve safe cross-clock domain transmission. In the 96 MHz clock domain, a linear interpolation module is inserted to achieve 2-fold upsampling, and the image components are eliminated by a CIC compensation filter.

[0181] Compared with the traditional synchronous interpolation scheme, it occupies less resources and can avoid error codes caused by metastability.

[0182] Configuring a dedicated hard core IP for parallel digital filtering processing;

[0183] For the multi-channel filtering calculation requirements, the built-in DSP hardcore of the device is called to construct a parallel processing array. For example, in radar echo processing, a dedicated FIR filter core is configured for each receiving channel, and the coefficients are stored in the Block RAM to form a reconfigurable filter bank. When running in parallel with 16 channels, each DSP hardcore only needs 3 clock cycles to complete the multiply-accumulate operation, and the crossbar switch is used to achieve dynamic allocation of data flow.

[0184] In the case of burst data scenarios, through the dynamic clock gating technology, the DSP units of the idle channels automatically enter the low-power mode.

[0185] The signal processing algorithm is dynamically updated through the partial reconfiguration technology;

[0186] To achieve online update of the algorithm, a dynamic loading mechanism based on partial reconfiguration is developed, and the signal processing chain is divided into a static area and a reconfigurable partition;

[0187] The static area maintains interface communication and basic control logic, and the reconfigurable partition carries variable algorithm modules (such as adaptive equalizers, encryption engines, etc.);

[0188] When the system detects a change in the channel conditions, a new bitstream file is transmitted through the PCIe interface, and the erasure and reprogramming of the target partition are completed in a short time.

[0189] In the case of maintaining the normal service data stream, the increase in the instantaneous bit error rate caused by the algorithm switching process is less than that of the traditional overall reconstruction scheme.

[0190] Embodiment 2:

[0191] As Figure 3 shown, based on the same inventive concept as a multi-channel AFE signal acquisition method based on adaptive adjustment in the foregoing embodiment, the present application provides a multi-channel AFE signal acquisition system based on adaptive adjustment. The system and method embodiments in the present application are based on the same inventive concept.

[0192] Among them, the system includes:

[0193] A channel acquisition module, which synchronously acquires multiple analog signals through a multi-channel parallel AFE module, and each channel is configured with an independent programmable gain amplifier and an anti-aliasing filter;

[0194] A feature extraction module, which extracts the time-frequency domain feature parameters of each channel signal in real time;

[0195] An adaptive adjustment module, which dynamically generates an optimal adjustment strategy based on the time-frequency domain feature parameters, and adaptively adjusts the gain coefficient, filter bandwidth, and sampling rate parameters of each channel through a closed-loop feedback mechanism;

[0196] An abnormal signal processing module, which uses an inter-channel cross-validation algorithm to implement abnormal signal detection and trigger the automatic isolation of faulty channels and the intelligent switching of redundant channels;

[0197] A phase compensation module, which performs time-domain alignment and phase compensation on the optimized multi-channel signals.

[0198] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

[0199] The above-mentioned is only the preferred specific implementation mode of the embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application, according to the technical solution and its concept of the present application, makes equivalent substitutions or changes, and should be covered by the protection scope of the present application.

Claims

1. A multi-channel AFE signal acquisition method based on adaptive adjustment, characterized in that It includes: Simultaneously collect multiple analog signals through a multi-channel parallel AFE module, and configure an independent programmable gain amplifier and an anti-aliasing filter for each channel; Extract the time-frequency domain characteristic parameters of each channel signal in real time; Dynamically generate an optimal adjustment strategy based on the time-frequency domain characteristic parameters, and adaptively adjust the gain coefficient, filter bandwidth, and sampling rate parameters of each channel through a closed-loop feedback mechanism; Adopt an inter-channel cross-validation algorithm to implement abnormal signal detection, and trigger the automatic isolation of faulty channels and the intelligent switching of redundant channels; Perform time-domain alignment and phase compensation on the optimized multi-channel signals.

2. The method according to claim 1, wherein The method of the dynamic adjustment strategy includes: Construct a signal quality evaluation model based on a convolutional neural network, and input a time-frequency domain characteristic matrix composed of time-frequency domain characteristic parameters to generate the quality score of each channel; the model architecture of the signal quality evaluation model includes an input layer, a feature extraction branch, a feature fusion layer, and an output layer; Establish a dynamic parameter adjustment rule library, including a noise scenario classification table and an optimal gain-bandwidth-sampling rate combination matrix under different noise environments; Adopt a reinforcement learning algorithm to online optimize the adjustment strategy, and dynamically update the rule library parameters according to the historical adjustment effect.

3. The method according to claim 1, characterized in that, The extraction of time-frequency domain characteristics adopts a hybrid signal processing technology; the time-domain characteristics include peak-to-peak value, waveform factor, and zero-crossing rate; the frequency-domain characteristics are obtained by combining sliding window FFT with wavelet packet decomposition; digital down-conversion technology is used to realize the feature extraction of band-pass signals.

4. The method according to claim 1, characterized in that The abnormal signal detection adopts a cross-channel correlation analysis method, and the method includes: Calculate the mutual information entropy and phase consistency index between adjacent channel signals; When the mutual information entropy between the channel signal and the adjacent channel signal is less than the set mutual information entropy threshold and the phase consistency index is greater than the phase locking value threshold, it is determined as an abnormal channel; Activate the preset redundant channel to replace the abnormal channel, and keep the total number of acquisition channels constant; Mutual information entropy The calculation methods include: where N is the total number of groups of channel signal groups, p(x i , y i ) is the joint probability distribution, x i and y i are the channel signal and the adjacent channel signal of the i-th group, p(x i ) and p(y i ) are the marginal probability distributions; The calculation method of the phase consistency index PLV includes: In the formula, φ1(t) and φ2(t) are the instantaneous phases of two signals at time point t, and j is the imaginary unit.

5. The method according to claim 1, characterized in that, It also includes a multi-working mode intelligent switching mechanism, and the multi-working mode intelligent switching mechanism includes a mode switching mechanism, an environmental monitoring mode, a high-precision mode, and a low-power mode; Environmental monitoring mode: Prioritize a wide dynamic range and adopt a segmented non-linear gain control; High-precision mode: Enable oversampling technology and digital post-filtering processing; Low-power mode: Dynamically turn off the bias circuit and high-speed ADC module of unactivated channels; The mode switching mechanism realizes signal phase continuity through a digital isolator and synchronous clock reconfiguration; 6. The method according to claim 1, wherein Phase compensation adopts an adaptive delay lock technology, generates a phase error signal based on the reference channel signal, and finely adjusts the sampling clock of each channel through a digital delay line; The generation method of the phase error signal includes: where M is the window length; R g,ref (k) is the similarity measure between channel g and the reference channel at the time shift point k; w g (n) is the delay control weight of channel g at the n-th iteration; μ is the convergence factor; Δφ g (n) is the phase error signal; x′ g (n) is the signal of channel g after the current delay compensation; m is the number of the current signal.

7. The method according to claim 1, wherein It also includes a dynamic calibration subsystem: Periodically inject a standard test signal for channel characteristic calibration; Establish a compensation look-up table for the frequency response and non-linear error of each channel; Adopt a temperature sensor to monitor the environmental temperature change and trigger temperature drift compensation.

8. The method according to claim 1, wherein The optimization measures implemented in the FPGA include: Adopt a multi-clock domain design to realize asynchronous sampling rate conversion; Configure dedicated hard-core IP for parallel digital filtering processing; Dynamically update the signal processing algorithm through partial reconfiguration technology.

9. A multi-channel AFE signal acquisition system based on adaptive adjustment, characterized in that the system Including: Channel acquisition module, which synchronously acquires multiple analog signals through a multi-channel parallel AFE module. Each channel is configured with an independent programmable gain amplifier and an anti-aliasing filter; Feature extraction module, which extracts the time-frequency domain feature parameters of each channel signal in real time; Adaptive adjustment module, which dynamically generates an optimal adjustment strategy based on the time-frequency domain feature parameters, and adaptively adjusts the gain coefficient, filter bandwidth, and sampling rate parameters of each channel through a closed-loop feedback mechanism; Abnormal signal processing module, which uses an inter-channel cross-validation algorithm to implement abnormal signal detection, triggering the automatic isolation of faulty channels and the intelligent switching of redundant channels; Phase compensation module, which performs time-domain alignment and phase compensation on the optimized multi-channel signals.

Images