Noise suppression method, device, equipment and storage medium for balanced photodetector

Abstract

The present application provides a noise suppression method, device, equipment and storage medium for a balanced photoelectric detector. The method includes: obtaining a first sampling sequence of a balanced photoelectric detector; performing digital filtering processing on the first sampling sequence to obtain a second sampling sequence; performing autocorrelation analysis on the second sampling sequence and constructing an autoregressive model to obtain a first prediction sequence; constructing an autoregressive moving average model according to the first prediction sequence, the second sampling sequence and the autoregressive model, and obtaining a second prediction sequence corresponding to the autoregressive moving average model in a preset time period; obtaining a third sampling sequence corresponding to the preset time period, performing noise evaluation according to the second prediction sequence and the third sampling sequence, and obtaining a preliminary noise sequence; obtaining a real-time sampling sequence of the balanced photoelectric detector, and weightedly fusing the real-time sampling sequence with the preliminary noise sequence in the autoregressive moving average model to obtain a photoelectric differential signal.

Description

Technical Field

[0001] The present application relates to the technical field of balanced photoelectric detectors, and in particular to a noise suppression method, device, equipment and storage medium for a balanced photoelectric detector. Background Art

[0002] At present, in the field of high-speed photodetection, balanced photodetector (BPD) has become an important type of detector because its differential detector structure can effectively suppress common-mode noise. However, with the increasing requirements for high-speed response characteristics in application scenarios, how to optimize the noise suppression algorithm while maintaining the high-speed response characteristics of balanced photodetectors has become an urgent problem to be solved. Electronic noise suppression technology has always occupied an important position in photodetectors, especially in high-speed signal processing, the introduction of noise will seriously affect the signal quality and system performance. Traditional noise suppression methods mostly rely on hardware-level improvements, such as the use of high-performance PIN photodiodes and low-noise amplifiers. However, these methods seem to be powerless when faced with complex environmental noise and signal processing requirements. Summary of the invention

[0003] The present application provides a noise suppression method, device, equipment and storage medium for a balanced photodetector, which are used to reduce data noise of the balanced photodetector and improve the accuracy of the balanced photodetector.

[0004] In a first aspect, an embodiment of the present application provides a noise suppression method for a balanced photodetector, the method comprising:

[0005] Acquire a first sampling sequence of a balanced photodetector;

[0006] Performing digital filtering on the first sampling sequence to obtain a second sampling sequence;

[0007] Performing autocorrelation analysis on the second sampling sequence and constructing an autoregressive model to obtain a first prediction sequence;

[0008] According to the first prediction sequence, the second sampling sequence and the autoregressive model, an autoregressive moving average model is constructed, and a second prediction sequence corresponding to the autoregressive moving average model in a preset time period is obtained;

[0009] Acquire a third sampling sequence corresponding to the preset time period, perform noise evaluation according to the second prediction sequence and the third sampling sequence, and obtain a preliminary noise sequence;

[0010] A real-time sampling sequence of the balanced photoelectric detector is obtained, and in the autoregressive moving average model, the real-time sampling sequence is weightedly fused with the preliminary noise sequence to obtain a photoelectric differential signal.

[0011] In a second aspect, an embodiment of the present application provides a noise suppression device for a balanced photodetector, the device comprising:

[0012] A data sampling module, used for acquiring a first sampling sequence of a balanced photodetector;

[0013] A data filtering module, used for performing digital filtering on the first sampling sequence to obtain a second sampling sequence;

[0014] A model building module, used for performing autocorrelation analysis on the second sampling sequence and building an autoregressive model to obtain a first prediction sequence;

[0015] A model optimization model, used to construct an autoregressive moving average model according to the first prediction sequence, the second sampling sequence and the autoregressive model, and obtain a second prediction sequence corresponding to the autoregressive moving average model in a preset time period;

[0016] A noise evaluation module, used for acquiring a third sampling sequence corresponding to the preset time period, performing noise evaluation according to the second prediction sequence and the third sampling sequence, and obtaining a preliminary noise sequence;

[0017] The data processing module is used to obtain the real-time sampling sequence of the balanced photoelectric detector, and in the autoregressive moving average model, weightedly fuse the real-time sampling sequence with the preliminary noise sequence to obtain a photoelectric differential signal.

[0018] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor;

[0019] The memory is used to store computer programs;

[0020] The processor is used to execute the computer program and implement the noise suppression method of the balanced photodetector as described in any one of the embodiments of the present application when executing the computer program.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements a noise suppression method for a balanced photodetector as described in any one of the embodiments of the present application.

[0022] An embodiment of the present application provides a noise suppression method for a balanced photodetector, the method comprising: obtaining a first sampling sequence of the balanced photodetector; performing digital filtering processing on the first sampling sequence to obtain a second sampling sequence; performing autocorrelation analysis on the second sampling sequence and constructing an autoregressive model to obtain a first prediction sequence; constructing an autoregressive moving average model based on the first prediction sequence, the second sampling sequence and the autoregressive model to obtain a second prediction sequence corresponding to the autoregressive moving average model in a preset time period; obtaining a third sampling sequence corresponding to the preset time period, performing noise evaluation based on the second prediction sequence and the third sampling sequence to obtain a preliminary noise sequence; obtaining a real-time sampling sequence of the balanced photodetector, and weightedly fusing the real-time sampling sequence with the preliminary noise sequence in the autoregressive moving average model to obtain a photoelectric differential signal. Through the above method, the first sampling sequence is digitally filtered to reduce the high-frequency noise interference in the initial data, the autoregressive model is constructed through autocorrelation analysis, and the first prediction sequence is generated to further capture the inherent laws and trends in the data and improve the accuracy of the prediction results. Then, an autoregressive moving average model is constructed based on the autoregressive model and the second sampling sequence to generate a second prediction sequence for a preset time period. The model can better adapt to data changes and provide more reliable noise evaluation. The preliminary noise sequence is then separated by comparison. In the autoregressive moving average model, the real-time sampling sequence and the preliminary noise sequence are weightedly fused to obtain the final photoelectric differential signal. This step ensures the optimal combination of real-time data and noise estimation, greatly reduces the impact of noise, and significantly improves the accuracy and reliability of the output signal of the balanced photodetector in a complex noise environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0024] Figure 1 A schematic flow chart of a noise suppression method for a balanced photodetector provided in an embodiment of the present application;

[0025] Figure 2 A schematic block diagram of a noise suppression device for a balanced photodetector provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

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

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

[0030] See also Figure 1 , Figure 1 The embodiment of the present application also provides a schematic flow chart of a noise suppression method for a balanced photodetector. Figure 1 As shown, the specific steps of the balanced photodetector include: S101-S106.

[0031] S101 , obtaining a first sampling sequence of a balanced photodetector.

[0032] Exemplarily, the balanced photodetector includes a plurality of PIN photodiodes and an amplifier, and the amplifier converts the photoelectric signals of the plurality of PIN photodiodes into differential signals. The differential signal of the balanced photodetector is sampled at high speed, and a high-performance analog-to-digital converter (ADC) is used to collect the differential signal at a sampling rate at least twice the bandwidth of the balanced photodetector to obtain a first sampling sequence. This ensures that the first sampling sequence preserves sufficient signal details, laying a foundation for subsequent noise analysis and suppression.

[0033] S102: Perform digital filtering on the first sampling sequence to obtain a second sampling sequence.

[0034] Exemplarily, the first sample sequence obtained by sampling is preprocessed. Digital filtering technology, such as a finite impulse response (FIR) filter, is applied to remove obvious high-frequency noise and DC offset. This step can initially improve signal quality and provide clearer data input for subsequent model analysis.

[0035] S103: Perform autocorrelation analysis on the second sampling sequence and construct an autoregressive model to obtain a first prediction sequence.

[0036] Exemplarily, after completing the data filtering, an autoregressive model (AR model) is established according to the second sampling sequence. The autoregressive model is a time series analysis model that can predict future values ​​through a linear combination of current data points and several previous data points. In the specific implementation process, it is first necessary to determine the model order of the autoregressive model (that is, how many previous data points are used). The AIC criterion (Akaike Information Criterion) or the BIC criterion (Bayesian Information Criterion) is usually used to select the optimal model order. Then, according to the second sampling sequence, the model parameters are estimated using the least squares method, the autoregressive model is established, and the first prediction sequence of the autoregressive model is output.

[0037] S104: construct an autoregressive moving average model according to the first prediction sequence, the second sampling sequence and the autoregressive model, and obtain a second prediction sequence corresponding to the autoregressive moving average model in a preset time period.

[0038] Exemplarily, in order to further improve the noise suppression effect, an autoregressive moving average model (ARMA model) is introduced on the basis of the autoregressive model. The autoregressive moving average model combines the advantages of the autoregressive model (AR model) and the moving average model (MA model), and can more accurately capture the characteristics of time series data. In the specific implementation process, first, the residual parameters of the AR model are analyzed. During the analysis, the prediction results of the AR model (the first prediction sequence) and the second sampling sequence are used to calculate the residuals, and an MA model is established to describe the moving average characteristics of the residual parameters of the AR model. Then the AR model and the MA model are combined, and the combined model parameters are optimized by the maximum likelihood estimation method (MLE), the ARMA model, and the second prediction sequence of the preset time period is output based on the ARMA model.

[0039] S105 , obtaining a third sampling sequence corresponding to a preset time period, and performing noise evaluation according to the second prediction sequence and the third sampling sequence to obtain a preliminary noise sequence.

[0040] Exemplarily, considering the changes in noise characteristics in complex environments, the ARMA model needs to be updated and adaptively adjusted in real time. First, the third sampling sequence within a preset time period is obtained. The purpose of this step is to provide a real signal benchmark for subsequent noise evaluation. Then, the second prediction sequence is compared and analyzed with the third sampling sequence. By comparing the two sequences, the noise components in the signal can be identified. Specifically, the difference method can be used, that is, the third sampling sequence is subtracted from the second prediction sequence point by point to obtain a residual sequence. This residual sequence can be regarded as a preliminary noise sequence because it represents components that are not predicted by the model, and these components are usually mainly noise.

[0041] During the noise evaluation process, statistical methods can be further used to analyze the preliminary noise sequence, such as calculating its mean and variance, to evaluate the intensity and characteristics of the noise. In this way, the distribution of noise in the signal can be better understood, providing a basis for noise suppression and extraction of photoelectric differential signals in subsequent steps. In this way, it can be ensured that during the operation of the balanced photodetector, noise is effectively identified and suppressed, and the signal-to-noise ratio and detection accuracy of the detector can be improved.

[0042] S106 , obtaining a real-time sampling sequence of a balanced photoelectric detector, and performing weighted fusion on the real-time sampling sequence and the preliminary noise sequence in an autoregressive moving average model to obtain a photoelectric differential signal.

[0043] Exemplarily, the real-time sampling sequence represents the photoelectric signal data currently captured by the balanced photodetector. After the real-time sampling sequence is acquired, it is input into the autoregressive moving average model constructed in the previous step.

[0044] The real-time sampling sequence is processed using the autoregressive moving average model. Since the noise has been preliminarily evaluated in the previous step and the preliminary noise sequence has been obtained, the real-time sampling sequence and the preliminary noise sequence need to be weighted fused in this step. The purpose of weighted fusion is to optimize the quality of the signal and further suppress the noise component. Specifically, the weighted average method can be used to linearly combine the real-time sampling sequence and the preliminary noise sequence with certain weights. The choice of weights can be adjusted according to the results of the noise evaluation to reduce the impact of noise on the signal while retaining the effective signal to the greatest extent.

[0045] The result of weighted fusion is the photoelectric differential signal. The photoelectric differential signal is a signal that has been processed after noise suppression, has a higher signal-to-noise ratio, and can more accurately reflect the actual changes in the measured physical quantity. In this way, the balanced photodetector can perform high-precision photoelectric detection in complex environments and improve the stability and reliability of the system.

[0046] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0047] In some embodiments, digital filtering is performed on the first sampling sequence to obtain the second sampling sequence, and the specific steps include: S1021-S1026.

[0048] S1021. Determine an ideal cutoff frequency according to the bandwidth of the balanced photodetector. The ideal cutoff frequency is 1.2 times the bandwidth of the balanced photodetector.

[0049] For example, the transition of a digital filter from a passband to a stopband is not instantaneous, but requires a certain transition band, the width of which is If the cutoff frequency If the cutoff frequency is set too close to the actual bandwidth of the signal, the presence of the transition band may cause some signal frequency components to be attenuated or filtered out, thereby damaging the integrity of the signal. In the embodiment of the present application, the ratio of the sampling frequency to the bandwidth of the balanced photodetector is 2:1. Therefore, setting the cutoff frequency to 1.2 times the bandwidth can provide a sufficient buffer for the transition band, thereby better retaining the useful frequency components of the signal.

[0050] S1022: Generate an FIR filter according to the ideal cutoff frequency and a preset Kaiser window function, and generate a filter impulse response based on the FIR filter, wherein the filter order of the FIR filter is N.

[0051] Exemplarily, a Kaiser window function is used to design an FIR filter, wherein a β parameter of the Kaiser window function is dynamically adjusted according to a target sidelobe attenuation and an ideal cutoff frequency, and the target sidelobe attenuation is set by a user.

[0052] Next, calculate the order N of the FIR filter using the formula N=ceil((As-7.95) / (2.285*2π*Δf)), where As is the target sidelobe attenuation. is the width of the transition zone.

[0053] The filter impulse response of the FIR filter refers to the output signal of the FIR filter when the input signal is a unit pulse (that is, the first sample is 1 and the rest of the samples are 0). The filter impulse response is specifically expressed as: h(n), where n=0, 1, ..., N-1.

[0054] S1023. Perform a discrete convolution operation on the first sampling sequence and the filter impulse response to obtain a filter sequence.

[0055] Exemplarily, discrete convolution is a basic method for applying a filter to a signal. By performing a convolution operation on the first sampling sequence and the filter impulse response, the frequency selection characteristics of the signal can be realized. Through discrete convolution, the filtered signal sequence can be directly obtained, the unnecessary frequency components can be removed, the desired signal characteristics can be retained, and noise suppression can be preliminarily achieved.

[0056] S1024: Use a preset overlap-add method to process the first sampling sequence into segments to obtain multiple segment sequences, each of which has a length of L, and every two adjacent segment sequences overlap by N-1 sampling points.

[0057] Exemplarily, the order of the FIR filter is N, which means that the length of the filter impulse response is N points. In the convolution operation, each point of the input signal will affect the next N points. Therefore, the convolution operation will introduce a delay of length N-1. In order to ensure the continuity of the processed signal between segments, the overlapping length of adjacent segments must be at least N-1. The overlapping length N-1 ensures that the signals between segments can be seamlessly connected, maintaining the integrity and continuity of the signal. At the same time, it avoids unnecessary repeated calculations and ensures the accuracy of the convolution operation.

[0058] S1025, performing fast Fourier transform on each segmented sequence, multiplying it with a preset frequency impulse response, and then performing inverse fast Fourier transform to obtain a segmented filtering result, wherein the preset frequency impulse response is obtained by performing fast Fourier transform on the filtered impulse response.

[0059] S1026: Overlap and add all the segmented filtering results to obtain a second sampling sequence.

[0060] Exemplarily, the segmented filtering results are overlapped and added, and the superposition length is N-1, so that the filtering results of each segment can be combined into a complete filtering signal sequence to ensure the continuity of the signal.

[0061] Through the above process, the effectiveness and computational efficiency of the FIR filter are ensured, providing a good foundation for subsequent noise suppression.

[0062] In some embodiments, autocorrelation analysis is performed on the second sampling sequence, and an autoregressive model is constructed to obtain a first prediction sequence. The specific steps include: S1031-S1037.

[0063] S1031. Extract an autocorrelation coefficient sequence from a second sampling sequence using a sliding window, wherein the sliding window is twice of a preset maximum delay order.

[0064] Exemplarily, a sliding window of appropriate size is selected (usually twice the preset maximum delay order). Then, the second sampling data sequence is scanned using the sliding window method, and the autocorrelation function of the data is calculated in each window to obtain an autocorrelation coefficient sequence.

[0065] It should be noted that the preset maximum delay order is a parameter in the autoregressive model analysis, which determines the maximum time delay considered in the autocorrelation function calculation and model construction process. Specifically, this parameter is used to limit the delay range considered when calculating the autocorrelation coefficient and partial autocorrelation coefficient. It is a value pre-set by the user based on experience or actual application requirements to control the complexity and calculation amount of the model.

[0066] S1032. Construct an autocorrelation coefficient matrix according to the autocorrelation coefficient sequence, and decompose the autocorrelation coefficient matrix using a preset recursive algorithm to obtain a partial autocorrelation coefficient sequence.

[0067] Exemplarily, the preset recursive algorithm is the Levinson-Durbin algorithm, which is an effective recursive algorithm that can quickly calculate the parameters of a high-order autoregressive model. The autocorrelation coefficient matrix is ​​recursively decomposed to obtain a partial autocorrelation coefficient sequence.

[0068] S1033. Determine a model order candidate set according to a preset information criterion function and a partial autocorrelation coefficient sequence, wherein the model order candidate set includes a plurality of candidate orders.

[0069] Exemplarily, the preset information criterion function includes: Akaike information criterion (AIC) or Bayesian information criterion (BIC). For each possible model order, the corresponding AIC value or BIC value is calculated using the preset information criterion function and the partial autocorrelation coefficient. Then, a series of candidate model orders with the smallest AIC or BIC values ​​are selected to form a model order candidate set.

[0070] S1034. Determine a segment length according to each candidate order, and perform segment processing on the second sampling sequence according to each segment length to obtain an order segment sequence.

[0071] Exemplarily, the second sampling data sequence is segmented according to the candidate order, and the length of each segment is equal to the candidate order.

[0072] S1035. Use the least square method to estimate the parameters of each order segmented sequence respectively to obtain multiple groups of autoregressive estimation parameters.

[0073] Exemplarily, for each set of autoregressive parameter estimates, its residual sequence is calculated and the prediction error variance is further calculated.

[0074] S1036. Perform residual analysis on multiple groups of autoregressive estimated parameters, calculate the prediction error variance corresponding to each group of autoregressive estimated parameters, and determine the target model order and target autoregressive parameters from the candidate orders and autoregressive estimated parameters according to the minimum value of the prediction error variance.

[0075] Exemplarily, a set of autoregressive parameters and corresponding model orders with the smallest prediction error variance are selected as the optimal model order and autoregressive parameters.

[0076] S1037. Construct an autoregressive model according to the target model order and the target autoregressive parameters, wherein the model coefficients of the autoregressive model are determined by a preset Yule-Walker equation group, and the prediction error of the autoregressive model is calculated by the geometric mean of the forward and backward linear prediction residuals.

[0077] Exemplarily, according to the target model order and the target autoregressive parameter, a Burg entropy maximization autoregressive model is constructed. Then, the Yule-Walker equations are solved to obtain the autoregressive model coefficients. Finally, the prediction error of the model is determined by calculating the geometric mean of the forward and backward linear prediction residuals, thereby completing the construction of the autoregressive model.

[0078] In some embodiments, an autoregressive moving average model is constructed based on the first prediction sequence, the second sampling sequence and the autoregressive model, including: performing a difference calculation on the first prediction sequence and the second sampling sequence to determine the autoregressive residual sequence of the autoregressive model; performing moving average modeling on the autoregressive residual sequence to generate a moving average model; and fusing the autoregressive model and the moving average model to obtain the autoregressive moving average model.

[0079] Exemplarily, by performing a difference calculation on the first prediction sequence and the second sampling sequence, the autoregressive residual sequence of the autoregressive model is determined, thereby capturing the noise and errors that the autoregressive model cannot fully explain, thereby providing input data for the subsequent moving average model. The autoregressive residual sequence can reflect the short-term random fluctuations and unpredictable parts in the original data. Then the autoregressive residual sequence is modeled by moving average. The moving average model further smoothes the data and eliminates random fluctuations by weighted averaging multiple lagged values ​​of the residual sequence. The generated moving average model can effectively reduce the noise component in the data and improve the prediction accuracy of the overall model.

[0080] Finally, the autoregressive model and the moving average model are fused to obtain the autoregressive moving average model. The fused autoregressive moving average model can not only explain the long-term trends and cyclical changes in the data, but also capture short-term random fluctuations through the moving average part, making the model more adaptable to changes in different time scales.

[0081] In some embodiments, the autoregressive model and the moving average model are fused to obtain an autoregressive moving average model, including: linearly combining the first residual equation of the autoregressive model with the second residual equation of the moving average model to obtain an initial fusion model; optimizing the first residual equation of the initial fusion model according to a conditional maximum likelihood estimation method to obtain an intermediate fusion model; obtaining a global optimization fusion model according to an unconditional maximum likelihood estimation method for the first residual equation and the second residual equation of the intermediate fusion model; calculating the AIC value and BIC value of the global optimization fusion model according to a preset information criterion function; and using a stepwise regression method to fine-tune the order of the global optimization fusion model according to the AIC and BIC values ​​to obtain an autoregressive moving average model.

[0082] Exemplarily, the first residual equation of the AR model is linearly combined with the second residual equation of the MA model to obtain the initial fusion model: the residual sequence of the AR model and the residual sequence of the MA model are respectively assigned weight coefficients α and β to construct a linear combination equation. By minimizing the variance of the combined residual, the optimal weight coefficient is solved to form the initial fusion model. According to the conditional maximum likelihood estimation method, the first residual equation of the initial fusion model is optimized to obtain the intermediate fusion model: the parameters of the moving average part are fixed, and the parameters of the autoregressive part are conditionally estimated for the maximum likelihood. The likelihood equation is iteratively solved using the Newton-Raphson method, and the autoregressive parameters are updated to obtain the intermediate fusion model. According to the unconditional maximum likelihood estimation method, the first residual equation and the second residual equation of the intermediate fusion model are obtained to obtain a global optimization fusion model: the parameters of the autoregressive and moving average parts are considered at the same time, and a joint likelihood function is constructed. The BFGS algorithm is used for unconstrained optimization, and the parameters of the two parts are updated at the same time to obtain a global optimization fusion model. According to the preset information criterion function, the AIC value and BIC value of the global optimization fusion model are calculated, specifically: based on the likelihood function value of the global optimization fusion model, the AIC value and BIC value of the model are calculated. According to the AIC and BIC values, the order of the global optimization fusion model is fine-tuned using the stepwise regression method to obtain the autoregressive MA model: starting from the low-order model, the order of the autoregressive and moving average parts is gradually increased. The AIC and BIC values ​​are calculated for each candidate model, and the model with the smallest AIC and BIC values ​​is selected as the final autoregressive MA model. If the AIC and BIC values ​​cannot be reduced by increasing the order for multiple times in a row, the fine-tuning process is stopped.

[0083] In some embodiments, a third sampling sequence corresponding to a preset time period is obtained, and noise evaluation is performed according to the second prediction sequence and the third sampling sequence to obtain a preliminary noise sequence, including: performing time synchronization processing on the second prediction sequence and the third sampling sequence to obtain a synchronous prediction sequence and a synchronous sampling sequence, wherein the time synchronization processing includes resampling the second prediction sequence using an interpolation algorithm to align its sampling points with the third sampling sequence; calculating a residual sequence according to the synchronous prediction sequence and the synchronous sampling sequence, wherein the residual sequence is obtained by subtracting the synchronous prediction sequence from the synchronous sampling sequence; applying a preset wavelet transform algorithm to the residual sequence to obtain multi-scale wavelet coefficients, wherein the preset wavelet transform algorithm uses discrete wavelet transform, and the basis function of the preset wavelet transform algorithm is Daubechies wavelet; according to the multi-scale wavelet coefficients, coefficients are screened using a preset threshold function, and the screened wavelet coefficients are inversely transformed to obtain a preliminary noise sequence, wherein the preset threshold function is a soft threshold function, and the preset threshold of the preset threshold function is adaptively determined according to the standard deviation of the wavelet coefficients and the data length.

[0084] The consistency and comparability of the data are ensured by performing time synchronization processing on the second prediction sequence and the third sampling sequence. Then the residual sequence is calculated, and the discrete wavelet transform is applied for multi-scale analysis, using the Daubechies wavelet as the basis function. Then the wavelet coefficients are screened using the soft threshold function, and the preliminary noise sequence is obtained by inverse wavelet transform. The advantage of this method is that it can effectively separate the signal and noise, and at the same time, the accuracy and adaptability of noise evaluation are improved by setting the adaptive threshold. The whole process not only considers the synchronization problem of the time series, but also utilizes the multi-resolution characteristics of wavelet analysis, so that the noise components at different scales can be more accurately identified and evaluated, providing a reliable foundation for subsequent signal processing and analysis.

[0085] In some embodiments, obtaining a first sampling sequence of a balanced photodetector includes: obtaining a differential signal of the balanced photodetector according to a preset sampling frequency, and using the differential signal of the balanced photodetector as the first sampling sequence, wherein the preset sampling frequency is greater than or equal to twice that of the balanced photodetector.

[0086] See also Figure 2 , Figure 2 The embodiment of the present application also provides a schematic block diagram of a noise suppression device for a balanced photodetector, wherein the noise suppression device 200 for a balanced photodetector is used to perform the aforementioned noise suppression method for a balanced photodetector, wherein the noise suppression device 200 for a balanced photodetector can be configured in a server.

[0087] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0088] like Figure 2 As shown, the noise suppression device 200 of the balanced photodetector includes: a data sampling module 201, a data filtering module 202, a model building module 203, a model optimization module 204, a noise evaluation module 205 and a data processing module 206.

[0089] The data sampling module 201 is used to obtain a first sampling sequence of a balanced photodetector.

[0090] In some embodiments, when the data sampling module 201 is used to implement the first sampling sequence of obtaining a balanced photodetector, it is specifically used to implement: obtaining the differential signal of the balanced photodetector according to a preset sampling frequency, and using the differential signal of the balanced photodetector as the first sampling sequence, wherein the preset sampling frequency is greater than or equal to twice that of the balanced photodetector.

[0091] The data filtering module 202 is used to perform digital filtering on the first sampling sequence to obtain a second sampling sequence.

[0092] In some embodiments, when the data filtering module 202 is used to implement digital filtering processing on the first sampling sequence to obtain the second sampling sequence, it is specifically used to implement: determining an ideal cutoff frequency according to the bandwidth of the balanced photodetector, where the ideal cutoff frequency is 1.2 times the bandwidth of the balanced photodetector; generating an FIR filter according to the ideal cutoff frequency and a preset Kaiser window function, and generating a filter impulse response based on the FIR filter, wherein the filter order of the FIR filter is N; performing discrete convolution operation on the first sampling sequence and the filter impulse response to obtain a filter sequence; using a preset overlap-add method, segmenting the first sampling sequence to obtain multiple segment sequences, each segment sequence has a length of L, and every two adjacent segment sequences overlap N-1 sampling points; performing a fast Fourier transform on each segment sequence, multiplying it with a preset frequency impulse response, and then performing an inverse fast Fourier transform to obtain a segment filtering result, wherein the preset frequency impulse response is obtained by performing a fast Fourier transform on the filter impulse response; and overlapping and adding all the segment filtering results to obtain the second sampling sequence.

[0093] The model building module 203 is used to perform autocorrelation analysis on the second sampling sequence and build an autoregressive model to obtain a first prediction sequence.

[0094] In some embodiments, when the model building module 203 is used to implement autocorrelation analysis on the second sampling sequence and construct an autoregressive model to obtain a first prediction sequence, it is specifically used to implement: using a sliding window to extract an autocorrelation coefficient sequence from the second sampling sequence, wherein the sliding window is twice the preset maximum delay order; constructing an autocorrelation coefficient matrix according to the autocorrelation coefficient sequence, and decomposing the autocorrelation coefficient matrix using a preset recursive algorithm to obtain a partial autocorrelation coefficient sequence; determining a model order candidate set according to a preset information criterion function and the partial autocorrelation coefficient sequence, wherein the model order candidate set includes multiple candidate orders; determining a segment length according to each candidate order, and respectively determining a segment length according to each segment length. The second sampling sequence is segmented to obtain an order segmented sequence; the least squares method is used to estimate the parameters of each order segmented sequence to obtain multiple groups of autoregressive estimation parameters; residual analysis is performed on the multiple groups of autoregressive estimation parameters to calculate the prediction error variance corresponding to each group of autoregressive estimation parameters, and the target model order and target autoregressive parameters are determined from the candidate orders and autoregressive estimation parameters according to the minimum value of the prediction error variance; an autoregressive model is constructed according to the target model order and the target autoregressive parameters, wherein the model coefficients of the autoregressive model are determined by a preset Yule-Walker equation group, and the prediction error of the autoregressive model is calculated by the geometric mean of the forward and backward linear prediction residuals.

[0095] The model optimization model 204 is used to construct an autoregressive moving average model according to the first prediction sequence, the second sampling sequence and the autoregressive model, and obtain a second prediction sequence corresponding to the autoregressive moving average model in a preset time period.

[0096] In some embodiments, when the model optimization model 204 is used to implement the construction of an autoregressive moving average model based on the first prediction sequence, the second sampling sequence and the autoregressive model, it is specifically used to implement: performing difference calculation on the first prediction sequence and the second sampling sequence to determine the autoregressive residual sequence of the autoregressive model; performing moving average modeling on the autoregressive residual sequence to generate a moving average model; and fusing the autoregressive model and the moving average model to obtain an autoregressive moving average model.

[0097] In some embodiments, when the model optimization model 204 is used to realize the fusion of the autoregressive model and the moving average model to obtain the autoregressive moving average model, it is specifically used to realize: linearly combining the first residual equation of the autoregressive model with the second residual equation of the moving average model to obtain an initial fusion model; optimizing the first residual equation of the initial fusion model according to the conditional maximum likelihood estimation method to obtain an intermediate fusion model; obtaining a global optimization fusion model according to the unconditional maximum likelihood estimation method for the first residual equation and the second residual equation of the intermediate fusion model; calculating the AIC value and BIC value of the global optimization fusion model according to a preset information criterion function; and using the stepwise regression method to fine-tune the order of the global optimization fusion model according to the AIC and BIC values ​​to obtain the autoregressive moving average model.

[0098] The noise evaluation module 205 is used to obtain a third sampling sequence corresponding to a preset time period, and perform noise evaluation according to the second prediction sequence and the third sampling sequence to obtain a preliminary noise sequence.

[0099] In some embodiments, when the noise assessment module 205 is used to obtain a third sampling sequence corresponding to a preset time period, and to perform noise assessment according to the second prediction sequence and the third sampling sequence to obtain a preliminary noise sequence, it is specifically used to implement: time synchronization processing is performed on the second prediction sequence and the third sampling sequence to obtain a synchronous prediction sequence and a synchronous sampling sequence, wherein the time synchronization processing includes resampling the second prediction sequence using an interpolation algorithm so that its sampling points are aligned with the third sampling sequence; a residual sequence is calculated according to the synchronous prediction sequence and the synchronous sampling sequence, wherein the residual sequence is obtained by subtracting the synchronous prediction sequence from the synchronous sampling sequence; a preset wavelet transform algorithm is applied to the residual sequence to obtain multi-scale wavelet coefficients, wherein the preset wavelet transform algorithm adopts discrete wavelet transform, and the basis function of the preset wavelet transform algorithm is Daubechies wavelet; according to the multi-scale wavelet coefficients, coefficients are screened using a preset threshold function, and the screened wavelet coefficients are inversely transformed to obtain a preliminary noise sequence, wherein the preset threshold function is a soft threshold function, and the preset threshold of the preset threshold function is adaptively determined according to the standard deviation of the wavelet coefficients and the data length.

[0100] The data processing module 206 is used to obtain a real-time sampling sequence of the balanced photoelectric detector, and perform weighted fusion on the real-time sampling sequence and the preliminary noise sequence in an autoregressive moving average model to obtain a photoelectric differential signal.

[0101] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement a noise suppression method for a balanced photodetector such as any one of the embodiments of the present application when executing the computer program.

[0102] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements a noise suppression method for a balanced photodetector as described in any one of the embodiments of the present application.

[0103] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A noise suppression method for a balanced photodetector, characterized in that: The method includes: Acquire a first sampling sequence of a balanced photodetector; Determine an ideal cutoff frequency according to the bandwidth of the balanced photodetector, wherein the ideal cutoff frequency is 1.2 times the bandwidth of the balanced photodetector; generate an FIR filter according to the ideal cutoff frequency and a preset Kaiser window function, and generate a filter impulse response based on the FIR filter, wherein the filter order of the FIR filter is N; perform discrete convolution operation on the first sampling sequence and the filter impulse response to obtain a filter sequence; use a preset overlap-add method to segment the first sampling sequence to obtain a plurality of segment sequences, each of which has a length of L, and each two adjacent segment sequences overlap N-1 sampling points; perform fast Fourier transform on each segment sequence, multiply it with a preset frequency impulse response, and then perform inverse fast Fourier transform to obtain a segment filtering result, wherein the preset frequency impulse response is obtained by performing fast Fourier transform on the filter impulse response; overlap-add all segment filtering results to obtain a second sampling sequence; Performing autocorrelation analysis on the second sampling sequence and constructing an autoregressive model to obtain a first prediction sequence; According to the first prediction sequence, the second sampling sequence and the autoregressive model, an autoregressive moving average model is constructed, and a second prediction sequence corresponding to the autoregressive moving average model in a preset time period is obtained; Acquire a third sampling sequence corresponding to the preset time period, perform noise evaluation according to the second prediction sequence and the third sampling sequence, and obtain a preliminary noise sequence; A real-time sampling sequence of the balanced photoelectric detector is obtained, and in the autoregressive moving average model, the real-time sampling sequence is weightedly fused with the preliminary noise sequence to obtain a photoelectric differential signal.

2. The noise suppression method of a balanced photodetector according to claim 1, characterized in that: The performing autocorrelation analysis on the second sampling sequence and constructing an autoregressive model to obtain a first prediction sequence includes: Extracting an autocorrelation coefficient sequence from the second sampling sequence using a sliding window, wherein the sliding window is twice the preset maximum delay order; Constructing an autocorrelation coefficient matrix according to the autocorrelation coefficient sequence, and decomposing the autocorrelation coefficient matrix using a preset recursive algorithm to obtain a partial autocorrelation coefficient sequence; Determine a model order candidate set according to a preset information criterion function and the partial autocorrelation coefficient sequence, wherein the model order candidate set includes a plurality of candidate orders; Determine a segment length according to each candidate order, and perform segment processing on the second sampling sequence according to each segment length to obtain an order segment sequence; The least square method is used to estimate the parameters of each segmented sequence of the order, and multiple groups of autoregressive estimation parameters are obtained; Performing residual analysis on the multiple groups of the autoregressive estimated parameters, calculating the prediction error variance corresponding to each group of the autoregressive estimated parameters, and determining the target model order and the target autoregressive parameters from the candidate orders and the autoregressive estimated parameters respectively according to the minimum value of the prediction error variance; The autoregressive model is constructed according to the target model order and the target autoregressive parameters, wherein the model coefficients of the autoregressive model are determined by a preset Yule-Walker equation group, and the prediction error of the autoregressive model is calculated by the geometric mean of the forward and backward linear prediction residuals.

3. The noise suppression method of a balanced photodetector according to claim 1, characterized in that: The step of constructing an autoregressive moving average model according to the first prediction sequence, the second sampling sequence and the autoregressive model comprises: Performing a difference calculation on the first prediction sequence and the second sampling sequence to determine an autoregressive residual sequence of the autoregressive model; Performing moving average modeling on the autoregressive residual sequence to generate a moving average model; The autoregressive model and the moving average model are fused to obtain the autoregressive moving average model.

4. The noise suppression method of a balanced photodetector according to claim 3, characterized in that: The step of fusing the autoregressive model and the moving average model to obtain the autoregressive moving average model includes: Linearly combining the first residual equation of the autoregressive model and the second residual equation of the moving average model to obtain an initial fusion model; According to the conditional maximum likelihood estimation method, optimizing the first residual equation of the initial fusion model to obtain an intermediate fusion model; According to the unconditional maximum likelihood estimation method, a global optimization fusion model is obtained for the first residual equation and the second residual equation of the intermediate fusion model; According to a preset information criterion function, the AIC value and the BIC value of the global optimization fusion model are calculated; According to the AIC and the BIC values, the order of the global optimization fusion model is fine-tuned using a stepwise regression method to obtain the autoregressive moving average model.

5. The noise suppression method of a balanced photodetector according to claim 1, characterized in that: Acquiring a third sampling sequence corresponding to the preset time period, and performing noise evaluation according to the second prediction sequence and the third sampling sequence to obtain a preliminary noise sequence, including: Performing time synchronization processing on the second prediction sequence and the third sampling sequence to obtain a synchronous prediction sequence and a synchronous sampling sequence, wherein the time synchronization processing includes resampling the second prediction sequence by using an interpolation algorithm so that its sampling points are aligned with those of the third sampling sequence; Calculating a residual sequence according to the synchronous prediction sequence and the synchronous sampling sequence, wherein the residual sequence is obtained by subtracting the synchronous prediction sequence from the synchronous sampling sequence; Applying a preset wavelet transform algorithm to the residual sequence to obtain multi-scale wavelet coefficients, wherein the preset wavelet transform algorithm adopts discrete wavelet transform, and the basis function of the preset wavelet transform algorithm is Daubechies wavelet; According to the multi-scale wavelet coefficients, coefficients are screened using a preset threshold function, and the screened wavelet coefficients are inversely transformed to obtain the preliminary noise sequence, wherein the preset threshold function is a soft threshold function, and the preset threshold of the preset threshold function is adaptively determined based on the standard deviation of the wavelet coefficients and the data length.

6. The noise suppression method of a balanced photodetector according to claim 1, characterized in that: The step of acquiring a first sampling sequence of a balanced photodetector comprises: A differential signal of the balanced photodetector is acquired according to a preset sampling frequency, and the differential signal of the balanced photodetector is used as the first sampling sequence, wherein the preset sampling frequency is greater than or equal to twice that of the balanced photodetector.

7. A noise suppression device for a balanced photodetector, characterized in that: The noise suppression device of the balanced photodetector comprises: A data sampling module, used for acquiring a first sampling sequence of a balanced photodetector; A data filtering module, used to determine an ideal cutoff frequency according to the bandwidth of the balanced photodetector, wherein the ideal cutoff frequency is 1.2 times the bandwidth of the balanced photodetector; generate an FIR filter according to the ideal cutoff frequency and a preset Kaiser window function, and generate a filter impulse response based on the FIR filter, wherein the filter order of the FIR filter is N; perform discrete convolution operation on the first sampling sequence and the filter impulse response to obtain a filter sequence; use a preset overlap-add method to segment the first sampling sequence to obtain multiple segment sequences, each of which has a length of L, and each two adjacent segment sequences overlap N-1 sampling points; perform a fast Fourier transform on each segment sequence, multiply it with a preset frequency impulse response, and then perform an inverse fast Fourier transform to obtain a segment filtering result, wherein the preset frequency impulse response is obtained by performing a fast Fourier transform on the filter impulse response; overlap-add all segment filtering results to obtain a second sampling sequence; A model building module, used for performing autocorrelation analysis on the second sampling sequence and building an autoregressive model to obtain a first prediction sequence; A model optimization model, used to construct an autoregressive moving average model according to the first prediction sequence, the second sampling sequence and the autoregressive model, and obtain a second prediction sequence corresponding to the autoregressive moving average model in a preset time period; A noise evaluation module, used for acquiring a third sampling sequence corresponding to the preset time period, performing noise evaluation according to the second prediction sequence and the third sampling sequence, and obtaining a preliminary noise sequence; The data processing module is used to obtain the real-time sampling sequence of the balanced photoelectric detector, and in the autoregressive moving average model, weightedly fuse the real-time sampling sequence with the preliminary noise sequence to obtain a photoelectric differential signal.

8. An electronic device, characterized in that: The electronic device comprises a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the noise suppression method of the balanced photodetector according to any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the noise suppression method of a balanced photodetector according to any one of claims 1 to 6.

Images