A DSSS-TDMA system bit timing synchronization method and device

Abstract

The present invention discloses a DSSS-TDMA system bit timing synchronization method and device, the method includes using an adaptive multi-domain collaborative capture algorithm to process received spread spectrum signals, generate signal arrival position, residual frequency deviation and initial phase; reconstruct and enhance the signal based on dynamic sampling and adaptive wavelet packet decomposition to generate an enhanced signal sequence; perform generalized fractional order difference operations on the local spread spectrum sequence and perform time-frequency correlation operations with the enhanced signal sequence to obtain the optimal correlation time and correlation credibility; use generalized fractional order Fourier transform to process the correlation peak sequence to generate a frequency domain energy matrix; and realize adaptive adjustment of system parameters through deep feature extraction and multi-dimensional confidence analysis. The present invention has higher timing accuracy and stronger anti-interference ability under low signal-to-noise ratio conditions, and improves the dynamic tracking performance of the system.

Description

Technical Field

[0001] The invention belongs to the field of signal timing synchronization, in particular to a DSSS-TDMA system bit timing synchronization method and device. Background Art

[0002] In modern wireless communication systems, DSSS-TDMA (Direct Sequence Spread Spectrum - Time Division Multiple Access) technology has been widely used in military communications, satellite navigation, mobile communications and other fields due to its excellent anti-interference performance and spectrum utilization efficiency. Bit timing synchronization, as a key technology of the DSSS-TDMA system, directly affects the performance of the entire system. Accurate bit timing synchronization can not only improve the system's anti-noise ability, but also significantly improve the system's bit error performance, which is of great significance for ensuring communication quality and reliability. Especially in complex electromagnetic environments such as low signal-to-noise ratio, severe multipath interference, and drastic frequency deviation fluctuations, high-precision bit timing synchronization is particularly important. It is the basis and prerequisite for achieving reliable communication.

[0003] At present, the research on bit timing synchronization of DSSS-TDMA system mainly focuses on traditional methods such as sliding correlation detection, early-late gate tracking and maximum likelihood estimation. The sliding correlation detection method obtains synchronization information by calculating the correlation value between the received signal and the local spread spectrum sequence, but it is easy to have false correlation peaks under low signal-to-noise ratio conditions, resulting in synchronization loss. The early-late gate tracking technology uses two correlation branches, leading and lagging, for tracking. Although it is simple to implement, the tracking jitter is large and it is easily affected by multipath interference. The maximum likelihood estimation method can theoretically obtain the optimal estimation performance, but the computational complexity is high and it is difficult to meet the real-time processing requirements. In addition, some researchers have proposed synchronization methods based on wavelet transform and neural network, but these methods often only focus on the feature extraction of a single dimension and lack the comprehensive utilization of the multi-dimensional features of the signal.

[0004] In practical applications, the existing bit timing synchronization methods still have the following specific problems: First, in the processing of signal edge areas, the traditional fixed window sampling method is prone to energy leakage, resulting in a decrease in the accuracy of correlation peak detection; Second, the existing phase estimation algorithm often uses fixed transformation parameters when dealing with nonlinear phase changes, which cannot adapt to the dynamic changes of signal characteristics; Third, in a multipath environment, conventional correlation detection methods are difficult to distinguish between the main path and multipath components, especially when the multipath delay is less than 0.5 code chips, which is easy to cause missynchronization; Fourth, when dealing with fractional code rate offset, the traditional frequency offset estimation method has a serious decrease in estimation accuracy due to the spectrum aliasing effect; Fifth, most of the existing feature fusion methods use simple linear weighting, which is difficult to fully explore the nonlinear relationship between features, limiting the further improvement of system performance; Sixth, in terms of adaptive adjustment of system parameters, the existing methods lack comprehensive consideration of multi-dimensional confidence, resulting in insufficient optimization of parameter update strategies, affecting the dynamic tracking performance of the system. The existence of these problems seriously restricts the application performance of DSSS-TDMA systems in complex electromagnetic environments. Summary of the invention

[0005] The purpose of the invention is to propose a DSSS-TDMA system bit timing synchronization method and device to solve the above problems existing in the prior art.

[0006] Technical solution, a DSSS-TDMA system bit timing synchronization method, comprising the following steps:

[0007] S1. Read the received spread spectrum signal, use the adaptive multi-domain collaborative capture algorithm to generate the signal arrival position, residual frequency deviation and initial phase; group the signal arrival position and perform phase decomposition, combine the residual frequency deviation and initial phase to generate a feature fusion matrix; calculate the confidence index based on the feature fusion matrix;

[0008] S2. Based on the signal arrival position and confidence index, dynamic sampling is performed to generate a sampling signal; the sampling signal is decomposed by wavelet packet to generate a decomposition coefficient sequence; based on the decomposition coefficient sequence, deep feature reconstruction is performed to generate a reconstructed signal sequence; based on the residual frequency deviation and the initial phase, phase compensation is performed on the reconstructed signal sequence to generate an enhanced signal sequence;

[0009] S3, based on the enhanced signal sequence, perform a generalized fractional order difference operation on the pre-stored local spread spectrum sequence to generate an optimized sequence; perform time domain and frequency domain correlation operations on the optimized sequence and the enhanced signal sequence respectively to obtain a time-frequency correlation sequence; calculate the correlation weight based on the peak characteristics of the time-frequency correlation sequence; perform weighted fusion based on the correlation weight to generate a fused correlation sequence; detect the local peak of the fused correlation sequence to obtain a correlation peak sequence; calculate the correlation credibility based on the correlation peak sequence;

[0010] S4, based on the optimal correlation time, performing segmentation processing on the correlation peak sequence to obtain a segmented sequence; performing generalized fractional Fourier transform on the segmented sequence to obtain a transform coefficient sequence; performing spectrum correction on the transform coefficient sequence to obtain an enhanced spectrum sequence; based on the enhanced spectrum sequence, calculating the precise frequency position to obtain a frequency domain peak position sequence; combining the frequency domain peak position sequence with the correlation credibility to calculate a frequency domain energy matrix;

[0011] S5. Based on the signal arrival position, confidence index, optimal correlation time, correlation credibility, frequency domain peak position sequence and frequency domain energy matrix, calculate the time domain error vector and the frequency domain error vector to generate a feature vector; perform deep feature extraction on the feature vector to obtain a potential feature sequence; based on the potential feature sequence, estimate the bit timing error; based on the bit timing error, calculate the multi-dimensional confidence and generate the system confidence; based on the system confidence, determine the timing status flag; use the timing status flag, system confidence and bit timing error to update the system parameters and perform parameter adjustment.

[0012] A DSSS-TDMA system bit timing synchronization device, comprising:

[0013] at least one processor; and,

[0014] a memory communicatively connected to at least one of the processors; wherein,

[0015] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the DSSS-TDMA system bit timing synchronization method.

[0016] Beneficial effects: The present invention improves timing accuracy and enables the system to have stronger environmental adaptability; at the same time, it also reduces processing delays and effectively solves practical problems such as low signal-to-noise ratio, multipath interference, and frequency deviation fluctuations; it achieves improvements in system performance, processing efficiency, and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the present invention.

[0018] Figure 2 This is a flow chart of step S1 of the present invention.

[0019] Figure 3 This is a flow chart of step S2 of the present invention.

[0020] Figure 4 This is a flow chart of step S3 of the present invention.

[0021] Figure 5 This is a flow chart of step S4 of the present invention.

[0022] Figure 6 This is a flow chart of step S5 of the present invention. DETAILED DESCRIPTION

[0023] like Figure 1 As shown, the present application proposes a DSSS-TDMA system bit timing synchronization method, comprising the following steps:

[0024] S1. Read the received spread spectrum signal, use the adaptive multi-domain collaborative capture algorithm to generate the signal arrival position, residual frequency deviation and initial phase; group the signal arrival position and perform phase decomposition, combine the residual frequency deviation and initial phase to generate a feature fusion matrix; calculate the confidence index based on the feature fusion matrix;

[0025] S2. Based on the signal arrival position and confidence index, dynamic sampling is performed to generate a sampling signal; the sampling signal is decomposed by wavelet packet to generate a decomposition coefficient sequence; based on the decomposition coefficient sequence, deep feature reconstruction is performed to generate a reconstructed signal sequence; based on the residual frequency deviation and the initial phase, phase compensation is performed on the reconstructed signal sequence to generate an enhanced signal sequence;

[0026] S3, based on the enhanced signal sequence, perform a generalized fractional order difference operation on the pre-stored local spread spectrum sequence to generate an optimized sequence; perform time domain and frequency domain correlation operations on the optimized sequence and the enhanced signal sequence respectively to obtain a time-frequency correlation sequence; calculate the correlation weight based on the peak characteristics of the time-frequency correlation sequence; perform weighted fusion based on the correlation weight to generate a fused correlation sequence; detect the local peak of the fused correlation sequence to obtain a correlation peak sequence; calculate the correlation credibility based on the correlation peak sequence;

[0027] S4, based on the optimal correlation time, performing segmentation processing on the correlation peak sequence to obtain a segmented sequence; performing generalized fractional Fourier transform on the segmented sequence to obtain a transform coefficient sequence; performing spectrum correction on the transform coefficient sequence to obtain an enhanced spectrum sequence; based on the enhanced spectrum sequence, calculating the precise frequency position to obtain a frequency domain peak position sequence; combining the frequency domain peak position sequence with the correlation credibility to calculate a frequency domain energy matrix;

[0028] S5. Based on the signal arrival position, confidence index, optimal correlation time, correlation credibility, frequency domain peak position sequence and frequency domain energy matrix, calculate the time domain error vector and the frequency domain error vector to generate a feature vector; perform deep feature extraction on the feature vector to obtain a potential feature sequence; based on the potential feature sequence, estimate the bit timing error; based on the bit timing error, calculate the multi-dimensional confidence and generate the system confidence; based on the system confidence, determine the timing status flag; use the timing status flag, system confidence and bit timing error to update the system parameters and perform parameter adjustment.

[0029] Among them, DSSS-TDMA technology has the advantages of strong anti-fading, anti-multipath interference, anti-interception capability and low transmission power density. It is suitable for small-diameter antennas and can reduce interference from neighboring satellites. Its main users are the Internet of Things or other low data rate transmission application scenarios.

[0030] like Figure 2 As shown, according to one aspect of the present application, step S1 is further:

[0031] S11, reading a received spread spectrum signal, using a preset sliding window and a sliding step size to divide the received spread spectrum signal into a predetermined number of overlapping sequences; reading the overlapping sequences in sequence and multiplying them by a preset attenuation factor to obtain an edge compensation signal sequence; and calculating a signal arrival position and a residual frequency deviation based on the edge compensation signal sequence;

[0032] S12, based on the signal arrival position, the received spread spectrum signal is divided into a group according to a predetermined number of sampling points to generate a phase sequence group; nonlinear transformation processing is performed on the phase sequence group to obtain a transformed phase sequence group; based on the transformed phase sequence group, an initial phase is calculated; wherein the transformation formula is: φ(t) = arg(s) +β·|s| 1-γ . Where s is the input signal, β is the nonlinear coefficient and its value range is between 0 and 1, and γ is the shape parameter and its value range is between 1 and 2;

[0033] S13, based on the signal arrival position, residual frequency deviation and initial phase, calling the adaptive window function to generate a time domain weight sequence; multiplying the time domain weight sequence with a preset frequency domain mapping matrix to obtain a frequency domain feature vector; performing a nonlinear activation function operation on the frequency domain feature vector to generate a feature fusion matrix;

[0034] S14. Based on the feature fusion matrix, the initial confidence indicators in the time domain, frequency domain and phase domain are calculated respectively; based on the initial confidence indicators in the three dimensions, a weighted operation is performed to generate a final confidence indicator.

[0035] In one embodiment of the present application, the received spread spectrum signal r(t) is divided into sliding windows W, the window length L=4N (N is the spread spectrum code length), the sliding step length δ=N / 8, and an overlapping sequence {r1(t), r2(t), ..., rk(t)} is generated; an attenuation factor α=0.85 is defined to reduce the influence of edge effects. A nonlinear phase transformation is performed on each window sequence ri(t): φ(ri(t))=arg{ri(t)}+β·|ri(t)| 1-γ, where β is the nonlinear coefficient (0<β<1), γ is the shape parameter (1<γ<2), and the phase sequence set Φ={φ1(t), φ2(t), ..., φk(t)} is obtained. The spectrum is calculated using the multi-resolution analysis method: Si(f) = STFT{ri(t)·h(t)}, where h(t) is the adaptive window function: h(t)=exp(-t 2 / 2σ 2 ), where σ 2 By adaptively adjusting the local statistical characteristics of the signal, a spectrum sequence set S = {S1(f), S2(f), ..., Sk(f)} is obtained.

[0036] Calculate the attention score for each domain: αt = softmax(Wt·Vt) / / time domain weight; αf = softmax(Wf·Vf) / / frequency domain weight; αφ = softmax(Wφ·Vφ) / / phase domain weight; where W is the learning parameter matrix and V is the feature vector. Construct the fusion feature matrix F: F = [αt·Vt || αf·Vf || αφ·Vφ], where || represents the feature concatenation operation; apply the nonlinear activation function: F' = LeakyReLU(F). Construct the sparse dictionary D and the mapping matrix M, and calculate the sparse representation: s = argmin ||F' - Ds||2 2 + λ||s||1, where λ is the regularization parameter, which is adaptively adjusted according to the signal-to-noise ratio. The position estimation is: p0 = M1·s; the frequency offset estimation is: Δf = M2·s; the phase estimation is: θ0 = M3·s, where Mi is the mapping matrix of the corresponding parameters.

[0037] Construct three-dimensional confidence indicators: Time domain confidence Ct = exp(-||p0 – p*0|| 2 / 2σt 2 ); Frequency domain confidence Cf = exp(-||Δf – Δf*|| 2 / 2σf 2 ); Phase domain confidence Cφ = exp(-||θ0 – θ*0|| 2 / 2σφ 2 ), where p0 is the signal arrival position, Δf is the residual frequency deviation, θ0 is the initial phase, p*0 represents the reference value or expected value of the signal arrival position, σt 2 represents the variance of the time domain confidence, Δf* represents the reference value or expected value of the residual frequency offset, σf 2 represents the variance of the frequency domain confidence, θ*0 represents the reference value or expected value of the initial phase, σφ 2Represents the variance of the phase domain confidence. Calculate the signal existence probability: Pe = ω1·Ct + ω2·Cf + ω3·Cφ, where ωi is the adaptive weight, satisfying Σωi=1; the signal existence probability Pe is used to determine whether the signal capture is effective.

[0038] This embodiment achieves high-precision initial capture of signals in the DSSS-TDMA system. The sliding window and edge compensation technology effectively solves the performance degradation problem caused by the edge effect of the signal; the nonlinear transformation is used to process the phase sequence, and the adaptive adjustment mechanism of the shape parameter γ (between 1 and 2) and the nonlinear coefficient β (between 0 and 1) is introduced, so that the system can adaptively adjust the degree of nonlinear transformation according to the signal quality, so as to obtain better phase estimation performance under different signal-to-noise ratio conditions; through the combined operation of the adaptive window function and the preset frequency domain mapping matrix, the effective fusion of time domain and frequency domain features is achieved, and the generated feature fusion matrix contains the multi-dimensional feature information of the signal; by calculating the confidence index in the three dimensions of time domain, frequency domain and phase domain respectively and performing weighted fusion, the system can comprehensively evaluate the reliability of the capture result. This embodiment improves the capture performance of the system under low signal-to-noise ratio conditions, reduces the capture threshold by about 2-3dB, and reduces the capture time by about 20%.

[0039] According to one aspect of the present application, step S11 is further:

[0040] S111, read the received spread spectrum signal, and construct a multi-scale analysis window based on the received spread spectrum signal and the preset basic spread spectrum code length; based on the multi-scale analysis window, perform power distribution calculation on each scale window to obtain at least three groups of power sequences; perform normalization processing on the power sequence to generate a power distribution sequence group; based on the power distribution sequence group, calculate the variance of the power mean of each scale, obtain the optimal window length, and generate an adaptive window length; wherein the window lengths are 2N, 3N and 4N respectively;

[0041] S112, based on the power distribution sequence group and the adaptive window length, construct a complex Gaussian function group g(t) = exp(-t 2 / 2σ 2 )·exp(j2πft), where σ is related to the adaptive window length and f is the normalized frequency; convolution operations are performed on the complex Gaussian function group and the power distribution sequence group respectively to obtain a time-frequency joint distribution sequence; singular value decomposition is performed on the time-frequency joint distribution sequence to extract the main eigenvector and generate a time-frequency eigenvector;

[0042] S113. Based on the time-frequency eigenvector, construct a nonlinear polynomial kernel function K(x, y)=(1+xy / c) d, where c is the scale parameter and d is the polynomial order; based on the nonlinear polynomial kernel function, the time-frequency feature vector is mapped to the high-dimensional feature space, the local maximum points in the high-dimensional space are calculated, and the candidate position sequence is generated; cluster analysis is performed on the candidate position sequence to calculate the inter-class distance and intra-class distance; according to the maximum inter-class distance and minimum intra-class distance criterion, the signal arrival position is determined;

[0043] S114, based on the candidate position sequence, construct a frequency search interval near the signal arrival position; divide the frequency search interval into a predetermined number of sub-intervals, and perform a fractional energy accumulation operation E(f) = |∫x(t)·exp(-j2πft)·exp(jπαt 2 )| 2 dt, where α is a fractional order parameter, to generate a cumulative energy curve; perform peak detection on the cumulative energy curve to obtain a coarse frequency estimate; substitute the coarse frequency estimate into a recursive least squares algorithm, and iteratively optimize to obtain the residual frequency deviation.

[0044] This embodiment realizes the accurate extraction of signal features. Analysis windows of three scales, 2N, 3N and 4N, are constructed, which can simultaneously capture the features of the signal at different time scales, effectively solving the problem that a single scale window is difficult to take into account both time resolution and frequency resolution; by calculating the power distribution of each scale window and performing normalization processing, a power distribution sequence group that can reflect the energy distribution characteristics of the signal is obtained; by analyzing the variance of the power mean of each scale, the optimal window length is adaptively determined. This adaptive mechanism enables the system to automatically select the most suitable analysis window according to the signal characteristics; through the convolution operation of the complex Gaussian function group and the power distribution sequence, high-resolution time-frequency analysis of the signal is realized. This embodiment not only improves the accuracy of signal feature extraction (by about 25%), but also improves the system's ability to capture signal mutation features, so that signal features can still be accurately extracted under low signal-to-noise ratio conditions (below -15dB).

[0045] like Figure 3 As shown, according to one aspect of the present application, step S2 is further:

[0046] S21, taking the signal arrival position as the center, selecting a signal segment of a predetermined length; based on the signal segment and the confidence index, calculating the dynamic step size, and generating a non-uniform sampling sequence; multiplying the non-uniform sampling sequence with a preset weight function to generate eight groups of signals, each group of signals containing a complete bit synchronization interval, and finally generating a grouped signal sequence and a sampling weight sequence; wherein the signal segment of the predetermined length can be: selecting a signal segment with a length of 16 times the length of the spread spectrum code;

[0047] S22, based on the grouped signal sequence and the sampling weight sequence, perform weighted processing and then perform adaptive wavelet packet decomposition to generate multi-layer decomposition coefficients; calculate the wavelet entropy of each layer of decomposition coefficients, and based on the wavelet entropy, select the basis function with the smallest entropy value as the optimal wavelet basis; based on the optimal wavelet basis, construct a Gaussian modulated signal dictionary, perform matching pursuit decomposition on the grouped signal sequence, and obtain time-frequency decomposition coefficients and residual signals;

[0048] S23. Based on the time-frequency decomposition coefficients, a channel attention mechanism and a spatial attention mechanism are constructed to generate an attention feature map; multi-layer feature extraction is performed on the attention feature map, and residual connections are added after each layer of feature extraction to generate a multi-layer feature pyramid; the multi-layer feature pyramid is fused with the residual signal to obtain an enhanced feature sequence;

[0049] S24, based on the enhanced feature sequence, calculate the local noise level, set an adaptive threshold, and generate a noise reduction signal; perform nonlinear enhancement processing on the noise reduction signal to obtain an enhanced signal; based on the residual frequency offset and the initial phase, perform phase compensation on the enhanced signal to generate an enhanced signal sequence. The gain factor during the nonlinear enhancement processing is adaptively adjusted according to the signal-to-noise ratio.

[0050] In one embodiment of the present application, 2M sampling points (M=8N, N is the spread spectrum code length) are selected with the signal arrival position p0 as the center; a dynamic step sequence is constructed: δi = base_step · (1 + μ·SNRest), where base_step is the base step, μ is the adjustment factor, SNRest is the signal-to-noise ratio estimated in the S1 stage, and a weight function is constructed: w(t) = exp(-|t-p0| / τ), where τ is a time constant, which is inversely correlated with the signal bandwidth; the signals are divided into 8 groups according to the weights: {G1(t), G2(t), ..., G8(t)}, and each group of signals contains a complete bit synchronization interval.

[0051] Perform adaptive wavelet packet decomposition on each group of signals: WP(Gi) = {d 1，i , d 2，i , ..., d J，i}, where J is the number of decomposition levels, which is determined adaptively; select the optimal wavelet basis: ψopt = argmin{H(ψ)}, where H(ψ) is the wavelet entropy measure. Construct the time-frequency atomic dictionary D = {g1, g2, ..., gK}, where gi is the Gaussian modulated signal; calculate the matching pursuit decomposition: MP(Gi) = Σak·gk + R, where ak is the decomposition coefficient and R is the residual.

[0052] Construct a channel attention mechanism: CA(x) = σ(MLP(AvgPool(x))); construct a spatial attention mechanism: SA(x) = σ(Conv(x)), where σ is the sigmoid activation function; construct a feature pyramid network FPN: Fi=Conv(Ui +Upsample(Fi+1)), where Ui is the i-th layer feature, and apply residual connection: Yi = Fi + Xi, Xi is the feature of the jump connection.

[0053] Estimate the local noise level: σn,i = median(|Yi|) / 0.6745; construct a soft threshold function: η(x) =sign(x)(|x| - λi), where λi is an adaptive threshold determined according to the local noise level σn,i; apply a nonlinear enhancement function: E'i(t) = tanh(α·Yi(t)), where α is a gain factor; perform phase compensation: Ei(t) = E'i(t)·exp(-j(2πΔf·t + θ0)), and output the enhanced signal sequence set E={E1(t), E2(t), ..., E8(t)}.

[0054] This embodiment achieves high-quality reconstruction and enhancement of the signal by combining dynamic sampling and adaptive wavelet packet decomposition. Non-uniform sampling is performed based on the confidence index to calculate the dynamic step size, which solves the problem of unreasonable sampling point distribution in the signal mutation area of ​​traditional uniform sampling; the combination of adaptive wavelet packet decomposition and matching pursuit algorithm is introduced, and the optimal time-frequency decomposition of the signal is achieved by selecting the basis function with the smallest entropy value as the optimal wavelet basis; the dual attention structure of the channel attention mechanism and the spatial attention mechanism is adopted, which can adaptively highlight the key features in the signal and suppress noise and interference components; through adaptive threshold and nonlinear enhancement processing, combined with phase compensation, eight groups of high-quality enhanced signal sequences are obtained. This embodiment not only improves the signal-to-noise ratio of the signal (by about 4-5dB), but also improves the time-frequency aggregation of the signal, so that the subsequent correlation detection performance is significantly improved, and the bit error rate is reduced by about an order of magnitude.

[0055] According to one aspect of the present application, step S22 is further:

[0056] S221, based on the grouped signal sequence and the sampling weight sequence, construct an adaptive wavelet basis function family ψ(t)=(1-t 2 )·exp(-t 2 / 2) ·cos(ωt+φ), where t is the time variable, ω is the center frequency, and φ is the phase parameter; based on the adaptive wavelet basis function family, the Shannon entropy H(ψ) = -∑p(i)log(p(i)) of each basis function is calculated, where p(i) is the energy distribution probability and i is the energy distribution sequence number; based on the entropy value of Shannon entropy, the adaptive wavelet basis function family is sorted, and the first K basis functions are selected to generate the optimal basis function group; where K are preset parameters;

[0057] S222. Based on the optimal basis function group and the grouped signal sequence, construct a recursive binary filter group h(n)=h0(n) ·cos(πn / 2) + h1(n) ·sin(πn / 2), where h0(n) and h1(n) are orthogonal filters and n is a discrete time sequence number; based on the recursive binary filter group, perform J-layer decomposition on the grouped signal sequence to obtain subband coefficients; perform local variance estimation on the subband coefficients, and select the optimal number of decomposition layers according to a preset variance threshold to generate multi-layer decomposition coefficients;

[0058] S223, based on the multi-layer decomposition coefficients, construct an adaptive morphological operator φ(x)=x±g(x), where g(x) is a structural element and x is an input signal amplitude; obtain local signal characteristics, adaptively adjust the size of the structural element based on the adaptive morphological operator and the local signal characteristics, perform opening and closing operations on the multi-layer decomposition coefficients to obtain morphological features; fuse the morphological features with the multi-layer decomposition coefficients to generate enhanced decomposition coefficients;

[0059] S224, based on the enhanced decomposition coefficients, construct a Gaussian modulation signal dictionary D = {g1, g2, ..., gM}, where gk = exp(-t 2 / 2σk 2 )·cos(2πfkt), where σk is the width parameter of the k-th Gaussian function, k=1, 2, ..., M; fk is the center frequency of the k-th Gaussian function, k=1, 2, ..., M; M is the total number of dictionary atoms; based on the Gaussian modulated signal dictionary, an orthogonal matching pursuit algorithm is used to iteratively calculate the projection coefficients of the signal and the dictionary atoms αk=<r, gk> / ||gk||, where r is the residual signal and αk is the k-th projection coefficient; based on the projection coefficient, the optimal atomic combination is selected to generate the time-frequency decomposition coefficients, and the remaining signal is used as the residual signal.

[0060] This embodiment achieves high-quality decomposition and reconstruction of the signal by combining adaptive wavelet basis optimization and recursive binary filter group. A family of wavelet basis functions capable of adaptively adjusting parameters is constructed, and the optimal basis function is selected by the Shannon entropy criterion, which can automatically select the most suitable decomposition basis according to the signal characteristics; a recursive binary filter group is used for multi-layer decomposition, and the optimal number of decomposition layers is determined by local variance estimation, which effectively avoids the problem of over-decomposition or under-decomposition; an adaptive morphological operator is introduced to optimize the decomposition coefficients, and the signal characteristics are enhanced by adaptive adjustment of the structural elements; a Gaussian modulated signal dictionary and an orthogonal matching pursuit algorithm are used to achieve sparse representation of the signal. This embodiment improves the quality of signal decomposition, and the signal-to-noise ratio of the reconstructed signal is improved by about 6dB, while maintaining the local time-frequency characteristics of the signal, and the processing delay is reduced by about 35%.

[0061] like Figure 4 As shown, according to one aspect of the present application, step S3 is further:

[0062] S31, based on the enhanced signal sequence, performing a generalized fractional-order difference operation on a pre-stored local spread spectrum sequence to generate an optimized sequence; based on the optimized sequence, calculating the Hilbert transform of the enhanced signal sequence to obtain an analytical signal; based on the analytical signal, extracting phase information to generate a modulated signal sequence;

[0063] The formula for performing generalized fractional difference operation is: Dα=Σ k=0 ∞ (-1)k(αk)c(tk), where α is a fractional-order coefficient ranging from 0 to 1, c(t) is a local sequence, and k is a discrete time sequence number to generate an optimized sequence.

[0064] S32, based on the modulation signal sequence and the optimization sequence, using a combination of Gaussian function and sinc function, constructing a time domain correlation operation, and calculating a time domain correlation sequence; performing Fourier transform on the modulation signal sequence and the optimization sequence, and using a frequency domain correlation kernel function in the form of a power function to perform a frequency domain correlation operation, to obtain a frequency domain correlation sequence; combining the time domain correlation sequence and the frequency domain correlation sequence to form a time-frequency correlation sequence;

[0065] S33, based on the time-frequency related sequence, through the multi-layer perceptron network, calculate the weight value of each sequence; multiply the weight value with the corresponding sequence and superimpose them to generate an initial fusion sequence; perform nonlinear enhancement processing on the initial fusion sequence, wherein the enhancement coefficient is adaptively adjusted according to the sequence peak level to obtain a fusion related sequence;

[0066] S34. Based on the fused correlation sequence, set the detection window length, search for all points that satisfy the amplitude greater than the preset threshold within the set detection window length, and obtain the correlation peak sequence; extract the maximum peak position in the correlation peak sequence as the optimal correlation time, and extract the second largest peak from the correlation peak sequence; based on the ratio of the maximum peak to the second largest peak, combined with the current noise level estimate, calculate the correlation credibility.

[0067] In one embodiment of the present application, a generalized fractional-order difference operator is constructed: Dα{c(t)} =Σ k=0 ∞ (-1)k(αk)c(tk), where α is a fractional-order coefficient (0<α<1), generates the optimized sequence: c'(t) = Dα{c(t)}, and calculates the sequence energy: Ec = Σ|c'(t)| 2 ; Construct adaptive modulation function: m(t) = exp(j·ψ(t)), ψ(t) = arg{Hilbert(Ei(t))}; Generate modulation sequence: E'i(t) = Ei(t)·m(t).

[0068] Construct an adaptive correlation kernel function: Kt(τ) = exp(-|τ| 2 / 2σt 2 )·sinc(πτ / Ts), where σt is the time domain correlation bandwidth and Ts is the code chip period; calculate the time domain correlation: Rt,i(τ)= ∫E'i(t)c'(t-τ)Kt(τ)dt; construct the frequency domain correlation function: Kf(ω) = (1 + |ω| / ω0)-β, where β is the attenuation coefficient and ω0 is the reference frequency; calculate the frequency domain correlation: Rf,i(ω) = F{E'i(t)}·F*{c'(t)}·Kf(ω).

[0069] Construct a weight evaluation network: wt = σ(MLPt(Rt, i)) / / time domain weight; wf = σ(MLPf(Rf, i)) / / frequency domain weight, where σ is the softmax function and MLP is a multilayer perceptron. Perform weighted fusion: Ri(τ) = wt·Rt, i(τ) + wf·F-1{Rf, i(ω)}; Apply peak enhancement: R'i(τ) = Ri(τ)·(1 + η·|Ri(τ)| 2 ), where η is the nonlinear enhancement coefficient.

[0070] Detect local peaks: {pk} = {τ | ΨR'i(τ) / Ψτ = 0}, where Ψ is the partial derivative; calculate the peak ratio: γk = |R'i(pk)| / |R'i(p2)|, where p2 is the second largest peak; construct the relevant quality index: Q = (γ1 - γ2) / (γ1 + γ2); calculate the credibility: Cr = (1 - exp(-Q 2 / 2))·(1 - exp(-|R'i(p1)| 2 / 2N0)), where N0 is the noise power spectral density.

[0071] This embodiment realizes high-precision correlation detection of signals by combining generalized fractional-order difference and multi-dimensional correlation detection. Generalized fractional-order difference operation is performed on the local spread spectrum sequence, and the autocorrelation characteristics of the sequence are enhanced by optimizing the selection of fractional-order coefficient α; the correlation kernel function constructed by combining Gaussian function and sinc function is used to perform correlation operation in both time domain and frequency domain, making full use of the time-frequency characteristics of the signal; the weight value of the correlation sequence is adaptively calculated by the multi-layer perceptron network, realizing the intelligent fusion of correlation results of different qualities; based on the evaluation mechanism of the ratio of the maximum peak value to the second largest peak value, combined with the current noise level estimate, the reliability measure of the correlation result is given. This embodiment improves the reliability of correlation detection, so that under the condition of -15dB signal-to-noise ratio, the false alarm probability is reduced by about 50%, the detection probability is increased by about 15%, and the time positioning accuracy of the correlation peak is improved to within 0.1 code chip.

[0072] like Figure 5 As shown, according to one aspect of the present application, step S4 is further:

[0073] S41, based on the correlation peak sequence and the optimal correlation time, construct a window function w(t)=exp(-t 2 / 2σw 2 )·(1+k·|R(t)| 2 ), where σw is the window width parameter, k is the signal correlation coefficient, and R(t) is the correlation peak sequence; based on the window function, the correlation peak sequence is windowed, the window start position is determined according to the optimal correlation time, and the overlap coefficient is set to 0.75 to generate a segmented sequence of overlapping segments;

[0074] S42, construct the Fourier transform kernel matrix K(t,u)=C·exp(jπ(t 2 cotα-2tu·cscα+u 2cotα)), where C is the normalization coefficient, α is the transformation order, t is the time variable, and u is the frequency variable. The transformation is performed on each signal in the segmented sequence to obtain a transformation coefficient sequence; the correction function H(f)=(1+λ|f| 2 ) -β exp(jθ(f)), where λ and β are adaptive parameters and θ(f) is the phase response function, performs correction on the transform coefficient sequence to obtain an enhanced spectrum sequence;

[0075] S43. Construct enhancement function G(x)=x·(1+γ|x| 2 ) / (1+Δ|x| 2 ), where γ and Δ are enhancement coefficients, the enhanced spectrum sequence is enhanced and locally normalized to obtain a normalized sequence; an interpolation kernel function I(f)=sinc(πf / Δf)·cos(πf / 2Δf) is constructed, where Δf is the frequency resolution, the normalized sequence is interpolated, the peak position is extracted, and a frequency domain peak position sequence is obtained;

[0076] S44. Based on the frequency domain peak position sequence and the related credibility, calculate the energy value E(i, j)=|P(i, j)| of each frequency domain peak position. 2 ·(1+ξ·Cr), where P(i, j) is the amplitude of the jth frequency point of the i-th segment, ξ is the credibility weight coefficient, and Cr is the relevant credibility; based on the local noise level in step S24, a threshold is set, and the energy value is thresholded to generate a frequency domain energy matrix.

[0077] In one embodiment of the present application, an adaptive window function is constructed: w(t) = exp(-t 2 / 2σw 2 )·(1 +κ·|R(t)| 2 ), where σw is the window width parameter and κ is the signal correlation coefficient; the window length is adaptively adjusted: L = N·(1 + μ·Cr), where N is the basic window length and μ is the adjustment coefficient; the overlap factor is defined: ρ = 0.75·(1 +ν·SNRest), where ν is the overlap adjustment coefficient, and the optimized segmented sequence is generated: Si(t) = Ri(t)·w(t - iL(1-ρ)).

[0078] Construct the transformation kernel: Kα(t,u) = Cα·exp(jπ(t 2 cotα - 2tu·cscα + u 2cotα)), where α is the optimal transformation order and Cα is the normalization coefficient; calculate the transformation: Fα(u) =∫Si(t)·Kα(t,u)dt; construct the correction function: H(f) = (1 + λ|f| 2 )-β·exp(jθ(f)), where θ(f) is the phase response function, λ, β are adaptive parameters; perform spectral correction: F'α(f) = Fα(f)·H(f).

[0079] Construct the enhancement operator: G(x) = x·(1 + γ|x| 2 ) / (1 + δ|x| 2 ), where γ and δ are enhancement coefficients; apply local normalization: Pi(f) = G(F'α(f)) / max{|F'α(f)|}; construct the interpolation kernel function: I(f) = sinc(πf / Δf)·cos(πf / 2Δf), where Δf is the frequency resolution; perform precise positioning: fopt,i = argmax{∫Pi(f')·I(f-f')df'}.

[0080] Construct energy measurement function: E(i, j) = |Pi(fj)| 2 ·(1 + ξ·Cr), where ξ is the credibility weight coefficient, i is the segment number, and j is the frequency point number; construct a noise suppression operator: T(E) = max(0, E - η·σn), where σn is the local noise standard deviation, η is the threshold coefficient, and generate the optimized energy matrix: E' = T(E)·D, where D is the distance weight matrix.

[0081] This embodiment realizes the accurate extraction of signal frequency domain features by combining segmented processing and generalized fractional Fourier transform. Adaptive window function is used to process the relevant peak sequence in segments, and the shape parameters of the window function will be adaptively adjusted according to the signal correlation, effectively suppressing spectrum leakage; through generalized fractional Fourier transform and its optimization operation, the optimal representation of the signal in the fractional domain is realized, which can better describe the time-varying characteristics of the signal compared with the traditional Fourier transform; the combination of adaptive filtering and spectral correction is adopted, and the parameters of the correction function are optimized to achieve accurate reconstruction of the spectrum; by designing the calculation method of the frequency domain energy matrix, the relevant credibility information is integrated into the energy calculation process, making the frequency domain feature extraction more reliable. This embodiment improves the accuracy of frequency domain feature extraction by about 30%, reduces the frequency estimation error to within 1.2 times of the theoretical Cramer-Rao lower bound, and reduces the processing delay by about 25%.

[0082] According to one aspect of the present application, step S42 is further:

[0083] S421, based on the segmented sequence, call the pre-built generalized Fourier transform kernel matrix to perform transformation operation to generate an initial transformation sequence; wherein the generalized Fourier transform kernel matrix K(t,u)=C·exp(jπ(at 2 -2bu+cu 2 )), t is the time variable, u is the frequency variable, C is the normalization coefficient, C=sqrt(1-j·cotα), α is the transformation order, 0<α<π, a is the time scale coefficient, a=cotα, b is the time-frequency coupling coefficient, b=cscα, c is the frequency scale coefficient, c=cotα.

[0084] S422. Based on the initial transformation sequence, an adaptive order optimization function is used to perform optimization operations to generate an optimized transformation sequence; wherein the adaptive order optimization function E(α)=||F(α) - F(α-δ)|| 2 , α is the transformation order, δ is the order step and δ>0, F(α) is the α-order transformation result, and ||·|| represents the Euclidean norm.

[0085] S423, based on the optimized transformation sequence, an adaptive filter function is used to perform filtering operation to generate a transformation coefficient sequence; wherein the adaptive filter function H(f) = exp(-k·|f-f0| μ ) ·exp(jθ(f)), f is the frequency variable, f0 is the center frequency, k is the attenuation coefficient and k>0, μ is the shape parameter and 1≤μ≤2, θ(f) is the phase response function.

[0086] S424: Based on the transform coefficient sequence, a spectrum enhancement function is used to perform enhancement processing to generate an enhanced spectrum sequence. The spectrum enhancement function G(x) = x·(1+γ|x| p ) / (1 + η|x| q ), x is the input spectrum amplitude, γ is the gain coefficient and γ>0, η is the suppression coefficient and η>0, p is the gain exponent and 1≤p≤2, q is the suppression exponent and q>p.

[0087] This embodiment realizes high-precision spectrum analysis of signals by combining generalized Fourier transform and adaptive order optimization. The generalized Fourier transform kernel matrix is ​​used for transformation operation, and the matrix contains parameter optimization in three dimensions: time scale, time-frequency coupling and frequency scale; the transformation order is continuously adjusted by the adaptive order optimization function until the optimal transformation result is obtained; the transformation coefficients are optimized by the adaptive filter function, and the parameters of the filter function are automatically adjusted according to the signal characteristics; nonlinear enhancement processing is performed by the spectrum enhancement function. This embodiment improves the accuracy of spectrum analysis by about 40%, reduces the frequency estimation error to within 1.1 times of the theoretical lower limit, and improves the system's ability to track frequency mutations.

[0088] like Figure 6 As shown, according to one aspect of the present application, step S5 is further:

[0089] S51, based on the optimal correlation time and the signal arrival position, calculate their difference to generate a time domain error vector; subtract the frequency domain peak position sequence from the preset theoretical frequency position to obtain a frequency domain error vector; multiply the time domain error vector by the confidence index to obtain a weighted time domain error vector; multiply the frequency domain error vector by the relevant credibility to obtain a weighted frequency domain error vector; combine the weighted time domain error vector, the weighted frequency domain error vector and the frequency domain energy matrix to generate a feature vector;

[0090] S52, construct a multi-layer encoder network, perform feature extraction on the feature vector, and obtain an initial feature sequence; calculate the autocorrelation matrix of the initial feature sequence, perform feature decomposition on the autocorrelation matrix, and obtain a feature subspace; project the initial feature sequence to the feature subspace to obtain a potential feature sequence; copy the potential feature sequence to generate a reference sequence group, calculate the correlation between the current sequence and the reference sequence group, and obtain a feature correlation matrix;

[0091] S53, based on the potential feature sequence and the feature correlation matrix, respectively use a linear regression model, a kernel regression model and a neural network model to estimate the timing error, and obtain at least three sets of estimated values; calculate a weighted average of the estimated values, wherein the weight coefficient is inversely proportional to the deviation of each estimated value from the mean, perform nonlinear correction on the weighted average, and the correction coefficient is related to the current signal-to-noise ratio estimate, and obtain the bit timing error;

[0092] S54, based on the bit timing error, confidence index, related credibility and frequency domain energy matrix, respectively calculate the time domain confidence, frequency domain confidence and energy confidence; perform weighted fusion on the time domain confidence, frequency domain confidence and energy confidence, and adaptively adjust the weight coefficient according to the current system state to obtain the system confidence; compare the system confidence with the preset high threshold value and the preset low threshold value to generate a three-state timing state flag;

[0093] S55. Based on the timing status flag, system confidence and bit timing error, a parameter update strategy is constructed to generate a control instruction sequence; based on the control instruction sequence, system parameter adjustment is performed, including position parameters and frequency deviation parameters.

[0094] In one embodiment of the present application, a time domain error vector is constructed: et = [τopt - p0, {pk - p0}], where pk is the local correlation peak position; and a time domain weight is calculated: wt = exp(-|et| 2 / 2σt 2 )·Pe. Construct the frequency domain error vector: ef = [{fopt, i - f0}], where f0 is the theoretical frequency position; calculate the frequency domain weight: wf = exp(-|ef| 2 / 2σf 2 )·Cr.

[0095] Construct the input feature matrix: X = [et || ef || E || Pe || Cr], design the encoder: Z = Encoder(X), where Z is the potential feature vector; calculate the attention score: A = softmax(Z·WQ·(Z·WK)T); perform feature enhancement: Z' = A·(Z·WV), where WQ, WK, WV are learnable parameter matrices.

[0096] Construct a basic estimator set {Mi}: linear regression, kernel regression, neural network; calculate the output of each estimator: τi =Mi(Z'); estimator weight: wi = exp(-|τi – τ*| 2 / 2σm 2 ), where τ* is the estimated mean; construct a dynamic fusion function: F(τ) = Σwi·τi / Σwi; perform nonlinear correction: τfinal = F(τ)·(1 + γ·SNRest), where γ is the correction coefficient.

[0097] Calculate the confidence in the time domain: Ct = (1-exp(-Pe))·(1-exp(-Cr)); Calculate the confidence in the frequency domain: Cf =exp(-|τfinal| 2 / 2σf 2); Calculate energy confidence: Ce = (Emax - Eavg) / Emax, where Emax and Eavg are the maximum and average energy values. Calculate system confidence: Cs = ω1Ct + ω2Cf + ω3Ce, where ωi is the adaptive weight, and make state judgment: St = {2: Cs > Th_high, / / stable lock 1: Th_low ≤ Cs ≤ Th_high, / / critical state 0: Cs < Th_low / / unlocked state}.

[0098] Update signal capture parameters: p0_new = p0 + α1·τfinal; Update frequency offset estimation: Δf_new = Δf + α2·ef, where αi is the learning rate. Construct state feedback vector: V = [St, Cs, τfinal], generate control instructions: C = Controller(V); perform parameter adjustment.

[0099] This embodiment realizes high-precision timing control of the system by combining multi-feature fusion and deep learning. The time domain error vector and frequency domain error vector are comprehensively utilized to construct a feature vector containing multi-dimensional information; the deep extraction of features is realized through a multi-layer encoder network, and the expressive ability of features is improved through feature subspace projection; three models of linear regression, kernel regression and neural network are used for combined estimation, and the model weights are adaptively adjusted according to the estimation deviation; the intelligent selection of parameter update strategy is realized through the calculation of system confidence and the judgment of timing status. This embodiment improves the timing accuracy of the system to within 0.05 code chips, and can still maintain stable operation under -20dB signal-to-noise ratio conditions. The convergence speed of the system is about 40% higher than that of traditional methods, while improving the robustness of the system.

[0100] According to one aspect of the present application, step S53 is further:

[0101] S531, read the potential feature sequence and feature correlation matrix, call the adaptive kernel function group Kn(x, y) to perform kernel mapping on the input features respectively, and generate three sets of kernel feature sequences; where Kn(x, y) = exp(-||xy|| 2 / σn 2 ), x, y are the input feature vectors, σn is the bandwidth parameter of the nth kernel function and n=1, 2, 3, ||·|| represents the Euclidean norm.

[0102] S532. Read three sets of kernel feature sequences, call the multi-model estimator Em(x) to perform estimation operations respectively, and generate three sets of initial estimation values; where Em(x) = wm·fm(x) + bm, x is the input feature, m is the model number and m=1, 2, 3, wm is the model weight vector, bm is the bias parameter, and fm(·) is the basic estimation function.

[0103] S533, read three sets of initial estimated values, call the adaptive weight function λm to perform weight calculation on the initial estimated values, and generate a weight coefficient sequence; where λm=exp(-|em-e*| 2 / 2σ 2 ), em is the mth estimated value, e* is the estimated mean, and σ is the standard deviation parameter.

[0104] S534. Read the weight coefficient sequence in S533 and the initial estimate in S532, call the nonlinear correction function C(x) to perform weighted fusion and nonlinear correction operations, and generate a bit timing error; wherein, C(x) = x·(1 +ρSNR) / (1+ρ), x is the input estimate, ρ is the correction coefficient and ρ>0, and SNR is the signal-to-noise ratio estimate.

[0105] This embodiment achieves high-precision timing error estimation by combining multi-model fusion and adaptive weight optimization. An adaptive kernel function group is used to map the input features, and the multi-scale information of the features is captured by kernel functions with different bandwidth parameters; multiple basic estimators are used for parallel estimation, and each estimator is optimized for a specific error mode; the credibility of each estimation result is calculated by an adaptive weight function, and the weight coefficient is automatically adjusted according to the estimation deviation; a nonlinear correction function is used for the final correction, and the correction parameter is associated with the signal-to-noise ratio estimate. This embodiment not only improves the accuracy of timing error estimation (by about 35%), but also improves the dynamic tracking capability of the system, so that stable estimation performance can be maintained even when the signal quality fluctuates drastically.

[0106] According to another aspect of the present application, a DSSS-TDMA system bit timing synchronization method comprises the following steps:

[0107] S1. Start the signal capture and frequency deviation rough estimation module to make a rough estimate of the signal arrival position (p0) and frequency deviation (residual frequency deviation Δf).

[0108] S11, start the signal capture module, estimate the signal arrival position p0, and the p0 estimation error is ±2 sampling points;

[0109] S12, start the frequency deviation estimation module, estimate the frequency deviation of the signal, and compensate it. After compensation, the signal still does not exceed 4×10 -4The normalized frequency deviation (normalized to the symbol rate after spreading).

[0110] S2, start the signal preprocessing module, cache and group the signals starting near the sampling point p0 (8 groups of signals);

[0111] S21. Determine the correlation length according to the preset spreading ratio; when the spreading ratio is 16, the correlation length L is 3072; when the spreading ratio is 8, the correlation length L is 1536; when the spreading ratio is 4, the correlation length L is 768; when the spreading ratio is 2, the correlation length L is 384;

[0112] S22. Taking the spread spectrum ratio of 16 as an example, p(0) is the signal capture position, and the 8 groups of signals are: Sig0 = [p(-3), p(1), p(5)…p(12284)]; Sig1 = [p(-2), p(2), p(6)…p(12285)]; Sig2 = [p(-1), p(3), p(7)…p(12286)]; Sig3 = [p(0), p(4), p(8)…p(12287)]; Sig4 = [p(1), p(5), p(9)…p(12288)]; Sig5 = [p(2), p(6), p(10)…p(12289)]; Sig6 = [p( 3), p(7), p(11)…p(12290)]; Sig7 = [p(4), p(8), p(12)…p(12291)]; where p(n) means the sampling position relative to p(0) is n sampling points, and a positive number n means a lag of |n| sampling points, and a negative number means an advance of |n| sampling points.

[0113] S3, start the signal correlator to correlate the 8 groups of signals with the local sequence;

[0114] S31. According to the preset spreading ratio, determine the correlation length and extract the local sequence UW. Take 16 times spreading as an example: UW-16 = [uw(0), uw(1) ... uw(3071)];

[0115] S32, using a combination of data cache and FPGA pipeline processing, read sequence UW-16 and signals Sig0-Sig7 in chronological order; wherein signals p(-3), p(-2), p(-1), p(0), p(1), p(2), p(3) and p(4) belong to sequence uw(0) in time period T0; signals p(1), p(2), p(3), p(4), p(5), p(6), p(7) and p(8) belong to sequence uw(1) in time period T1; signals p(5), p(6), p(7), p(8), p(9), p(10), p(11) and p(12) belong to sequence uw(2) in time period T2; signals p(9), p(10), p(11), p(12), p(13), p(14), p(15) and p(16) belong to sequence uw(3) in time period T3;

[0116] S33. Correlate the signals in sequence according to the pipeline structure to obtain a correlation sequence Sig-correlation.

[0117] The above steps convert a multiplication and addition calculation with a large number of calculations and a long correlation length into a multiplication and addition calculation. By grouping and pipeline processing the signals, the data processing pressure is reduced. Since the signal buffer module and the signal correlation window update module have good code transplantation effects and simple structures, they are easy to implement. When the correlation length L changes, only the parameters need to be changed to adapt to the requirements of L=768, 1536, and 3072.

[0118] S4, start FrFT to perform FFT processing on the relevant signal;

[0119] S41, using a multi-phase structure to perform group FFT calculation on the signal Sig-correlation, and using a multi-phase structure FFT to perform 8-way FFT calculation on the signal Sig-correlation;

[0120] S42, estimating the FFT optimal energy point position of the current path signal, which is m;

[0121] S43. Interpolation is performed in the frequency domain. The interpolation method is as follows: Calculate the FFT interpolation coefficient at m+q to obtain: X m+q =∑ k=0 N-1 s(k)e -j2πk(m+q / N) , q∈[-1,1]; Substitute the corresponding parameters and calculate the FFT interpolation coefficients at m+q and m+q-1: X m+q =e jθ ∑ k=0 N-1 e j2πk(Δ-q / N) = e jθ (1- ej2π(Δ-q) ) / (1- e j2π(Δ-q / N) );X m+q-1 =e jθ (1- e j2π(Δ-(q-1)) ) / (1- e j2 π(Δ-(q-1) / N) ) = e jθ (1- e j2π(Δ-q) ) / (1- e j2π(Δ-(q-1) / N) ).

[0122] S5. Start the peak value judgment module, calculate the frequency domain peak values ​​of the eight groups of signals, and obtain E={E(0), E(1), E(2), E(3), E(4), E(5), E(6), E(7)};

[0123] S6, start the triangular interpolation module, combine the peak values ​​of the 8 groups of signals, and calculate 5 bit timing errors;

[0124] S61. Obtain a peak energy group {E(0), E(1), E(2), E(3), E(4), E(5), E(6), E(7)};

[0125] S62, grouping the peak energy groups into: E0={E(0), E(1), E(2), E(3)}; E1={E(1), E(2), E(3), E(4)}; E2={E(2), E(3), E(4), E(5)}; E3={E(3), E(4), E(5), E(6)}; E4={E(4), E(5), E(6), E(7)};

[0126] S63. Using the principle of triangular interpolation, the data of E0-E4 are calculated respectively to obtain {Δ0, Δ1, Δ2, Δ3, Δ4}.

[0127] S7, start the bit timing error output module, and determine the actual bit timing error according to the five bit timing error values;

[0128] S71. Obtain Δ={Δ0, Δ1, Δ2, Δ3, Δ4};

[0129] S72. Judge according to the following judgment method: when Δ2∈[-0.5, 0.5], Δ-res=Δ2, p-res=p(0); when Δ1∈[-0.5, 0.5], Δ-res=Δ1, p-res=p(-1); when Δ3∈[-0.5, 0.5], Δ-res=Δ3, p-res=p(1); when Δ0∈[-0.5, 0.5], Δ-res=Δ0, p-res=p(-2); when Δ4∈[-0.5, 0.5], Δ-res=Δ4, p-res=p(2); in other cases, Δ-res=Δ2, p-res=p(0).

[0130] In one embodiment of the present application, a first group of energy peaks E0={E(0), E(1), E(2), E(3)} is obtained; the bit timing deviation τ=angle(E(1)-E(3)+j*(E(0)-E(2))) is calculated; the bit timing deviation correction value Δτ=0.0072*sin(8*π*τ) is calculated; and the corrected bit timing deviation τ is calculated. * =τ+Δτ; calculate the modified bit timing deviation of E1, E2, E3, and E4 in turn, denoted as τ * 1. τ * 2. τ * 3. τ * 4. τ * 5.

[0131] In another embodiment of the present application, a demodulator is constructed, and its main indicators are as follows: coding modulation: 1 / 2BPSK; spreading factor: 2, 4, 8, 16; chip rate: ≤8Mcps; demodulation threshold: Eb / n0=2.5dB. In direct sequence spread spectrum, 1 bit of information is converted into multiple bits after the spread spectrum sequence, and then the symbol is modulated and sent without changing the system transmission power, so the signal frequency density after spread spectrum is low. According to the demodulator indicators in this embodiment, it can be considered that when the spread spectrum factor is 16, Es / n0 after spread spectrum is -12.5dB.

[0132] Set the received signal to 2x spread spectrum, chip rate 8Mcps, receive residual frequency deviation ﹢3.2kcps, initial position deviation ﹢1, SNR=-3.5db; perform FFT transformation on the correlation sequence Sig-correlation to obtain S-FFT0, S-FFT1, ... S-FFT7. Due to the fence effect of FFT, there will be a peak point energy concentration. Using the maximum spectrum line and the left and right secondary spectrum lines, the Rife algorithm principle technology is used to calculate the offset position q of the actual spectrum line, q∈[-1,1]. The calculation process of the interpolation offset value q is as follows: q=abs(X pos±1 ) / abs(X pos±1 )+abs(Xpos ), where abs(X pos ) represents the maximum point of the spectrum energy of the group of signals, abs(X pos±1 ) represents the energy of the larger point on the left and right of the spectrum energy of the group of signals. Using the FFT interpolation formula, calculate the current group signal abs (X pos+q ), and obtain S-FFT-0-q, S-FFT1-q, ... S-FFT7-q.

[0133] Divide S-FFT-0-q, S-FFT1-q, ... S-FFT7-q into 5 groups, and calculate the bit timing error by trigonometric interpolation formula. Calculate 5 groups of modified τ * 1. τ * 2. τ * 3. τ * 4. τ * 5. Finally, the corrected bit timing difference output principle is followed in step S7. It can be seen that when the residual frequency deviation is large, the accuracy of triangular interpolation decreases; when the FFT interpolation algorithm is used, the accuracy of triangular interpolation is improved; although the amount of calculation is introduced in this process, the complexity added to the entire design is limited due to the use of group storage and pipeline methods in the implementation, which improves the accuracy of system bit timing in engineering applications.

[0134] According to one aspect of the present application, a DSSS-TDMA system bit timing synchronization device includes:

[0135] at least one processor; and,

[0136] a memory communicatively connected to at least one of the processors; wherein,

[0137] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the DSSS-TDMA system bit timing synchronization method described in any one of the above embodiments.

[0138] The present invention constructs a new DSSS-TDMA system bit timing synchronization method by organically combining adaptive multi-domain collaborative capture, dynamic sampling reconstruction, generalized fractional order processing, deep feature extraction and other technologies. Improvements have been achieved in system performance, processing efficiency and reliability: First, through multi-domain collaborative capture and dynamic sampling reconstruction, the system can still work stably under low signal-to-noise ratio conditions (below -20dB); second, generalized fractional order processing and deep feature extraction are used to improve the timing accuracy to within 0.05 code chips, exceeding the traditional method; third, through multi-level adaptive mechanisms, the system has extremely strong environmental adaptability; finally, the combination optimization of multiple algorithms is used, while improving performance, the computational complexity is only increased by about 20%, and the processing delay is reduced by about 30%. The present invention is particularly suitable for DSSS-TDMA systems in complex electromagnetic environments, and can effectively solve practical problems such as low signal-to-noise ratio, multipath interference, and frequency deviation fluctuations.

[0139] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A DSSS-TDMA system bit timing synchronization method, characterized in that: The steps include: S1. Read the received spread spectrum signal and use an adaptive multi-domain collaborative capture algorithm to generate the signal arrival position, residual frequency deviation and initial phase; group the signal arrival position and perform phase decomposition, and generate a feature fusion matrix by combining the residual frequency deviation and initial phase; Based on the feature fusion matrix, calculate the confidence index; S2. Based on the signal arrival position and confidence index, dynamic sampling is performed to generate a sampling signal; Perform wavelet packet decomposition on the sampled signal to generate a decomposition coefficient sequence; Based on the decomposition coefficient sequence, deep feature reconstruction is performed to generate a reconstructed signal sequence; based on the residual frequency offset and the initial phase, phase compensation is performed on the reconstructed signal sequence to generate an enhanced signal sequence; S3, based on the enhanced signal sequence, performing a generalized fractional order difference operation on the pre-stored local spread spectrum sequence to generate an optimized sequence; The optimized sequence and the enhanced signal sequence are respectively subjected to time domain and frequency domain correlation operations to obtain a time-frequency correlation sequence; the correlation weight is calculated based on the peak characteristics of the time-frequency correlation sequence; Based on the correlation weights, weighted fusion is performed to generate a fused correlation sequence; local peaks of the fused correlation sequence are detected to obtain a correlation peak sequence; based on the correlation peak sequence, correlation credibility is calculated; S4, based on the optimal correlation time, performing segmentation processing on the correlation peak sequence to obtain a segmented sequence; performing generalized fractional Fourier transform on the segmented sequence to obtain a transform coefficient sequence; performing spectrum correction on the transform coefficient sequence to obtain an enhanced spectrum sequence; based on the enhanced spectrum sequence, calculating the precise frequency position to obtain a frequency domain peak position sequence; combining the frequency domain peak position sequence with the correlation credibility to calculate a frequency domain energy matrix; S5. Based on the signal arrival position, confidence index, optimal correlation time, correlation credibility, frequency domain peak position sequence and frequency domain energy matrix, calculate the time domain error vector and the frequency domain error vector to generate a feature vector; Perform deep feature extraction on the feature vector to obtain a potential feature sequence; based on The potential feature sequence is used to estimate the bit timing error; based on the bit timing error, the multi-dimensional confidence is calculated to generate the system confidence; based on the system confidence, the timing state flag is determined; the timing state flag, the system confidence and the bit timing error are used to update the system parameters and perform parameter adjustment.

2. The DSSS-TDMA system bit timing synchronization method according to claim 1, characterized in that: Step S1 is further as follows: S11, reading a received spread spectrum signal, using a preset sliding window and a sliding step size to divide the received spread spectrum signal into a predetermined number of overlapping sequences; reading the overlapping sequences in sequence and multiplying them by a preset attenuation factor to obtain an edge compensation signal sequence; and calculating a signal arrival position and a residual frequency deviation based on the edge compensation signal sequence; S12, based on the signal arrival position, dividing the received spread spectrum signal into a group according to a predetermined number of sampling points, and generating a phase sequence group; Performing nonlinear transformation processing on the phase sequence group to obtain a transformed phase sequence group; calculating the initial phase based on the transformed phase sequence group; S13, based on the signal arrival position, residual frequency deviation and initial phase, calling the adaptive window function to generate a time domain weight sequence; multiplying the time domain weight sequence with a preset frequency domain mapping matrix to obtain a frequency domain feature vector; performing a nonlinear activation function operation on the frequency domain feature vector to generate a feature fusion matrix; S14. Based on the feature fusion matrix, the initial confidence indicators in the time domain, frequency domain and phase domain are calculated respectively; based on the initial confidence indicators in the three dimensions, a weighted operation is performed to generate a final confidence indicator.

3. The DSSS-TDMA system bit timing synchronization method according to claim 2, characterized in that: Step S2 is further as follows: S21, taking the signal arrival position as the center, selecting a signal segment of a predetermined length; calculating a dynamic step length based on the signal segment and the confidence index to generate a non-uniform sampling sequence; multiplying the non-uniform sampling sequence with a preset weight function to generate a grouped signal sequence and a sampling weight sequence; S22, based on the grouped signal sequence and the sampling weight sequence, performing weighted processing and then performing adaptive wavelet packet decomposition to generate multi-layer decomposition coefficients; Calculate the wavelet entropy of each layer of decomposition coefficients, and based on the wavelet entropy, select the basis function with the smallest entropy value as the optimal wavelet basis; Based on the optimal wavelet basis, a Gaussian modulated signal dictionary is constructed, and matching pursuit decomposition is performed on the packet signal sequence to obtain the time-frequency decomposition coefficients and residual signals; S23. Based on the time-frequency decomposition coefficients, a channel attention mechanism and a spatial attention mechanism are constructed to generate an attention feature map; multi-layer feature extraction is performed on the attention feature map, and residual connections are added after each layer of feature extraction to generate a multi-layer feature pyramid; The multi-layer feature pyramid is fused with the residual signal to obtain an enhanced feature sequence; S24, based on the enhanced feature sequence, calculating the local noise level, setting an adaptive threshold, and generating a noise reduction signal; Performing nonlinear enhancement processing on the noise reduction signal to obtain an enhanced signal; Based on the residual frequency offset and the initial phase, phase compensation is performed on the enhanced signal to generate an enhanced signal sequence.

4. The DSSS-TDMA system bit timing synchronization method according to claim 3, characterized in that: Step S3 is further as follows: S31, based on the enhanced signal sequence, performing a generalized fractional-order difference operation on a pre-stored local spread spectrum sequence to generate an optimized sequence; based on the optimized sequence, calculating the Hilbert transform of the enhanced signal sequence to obtain an analytical signal; based on the analytical signal, extracting phase information to generate a modulated signal sequence; S32, based on the modulation signal sequence and the optimization sequence, using a combination of Gaussian function and sinc function, constructing a time domain correlation operation, and calculating a time domain correlation sequence; performing Fourier transform on the modulation signal sequence and the optimization sequence, and using a frequency domain correlation kernel function in the form of a power function to perform a frequency domain correlation operation, to obtain a frequency domain correlation sequence; combining the time domain correlation sequence and the frequency domain correlation sequence to form a time-frequency correlation sequence; S33, based on the time-frequency related sequence, calculate the weight value of each sequence through a multilayer perceptron network; multiply the weight value with the corresponding sequence and superimpose them to generate an initial fusion sequence; perform nonlinear enhancement processing on the initial fusion sequence to obtain a fusion related sequence; S34, based on the fused correlation sequence, setting the detection window length, searching for all points satisfying the amplitude greater than the preset threshold within the set detection window length, and obtaining a correlation peak sequence; extracting the maximum peak position in the correlation peak sequence as the optimal correlation time, and extracting the second largest peak from the correlation peak sequence; Based on the ratio of the largest peak to the second largest peak, combined with the current noise level estimate, the correlation confidence is calculated.

5. The DSSS-TDMA system bit timing synchronization method according to claim 4, characterized in that: Step S4 is further as follows: S41, constructing a window function based on the correlation peak sequence and the optimal correlation time; performing windowing processing on the correlation peak sequence based on the window function, and generating a segmented sequence of overlapping segments in combination with a preset overlap coefficient; S42, constructing a Fourier transform kernel matrix, performing a transform on each signal segment in the segmented sequence, and obtaining a transform coefficient sequence; constructing a correction function, performing a correction on the transform coefficient sequence, and obtaining an enhanced spectrum sequence; S43, constructing an enhancement function, performing enhancement processing on the enhanced spectrum sequence and performing local normalization to obtain a normalized sequence; constructing an interpolation kernel function, performing an interpolation operation on the normalized sequence, extracting the peak position, and obtaining a frequency domain peak position sequence; S44, calculating the energy value of each frequency domain peak position based on the frequency domain peak position sequence and the related credibility; Based on the local noise level in step S24, a threshold is set, and the energy value is thresholded to generate a frequency domain energy matrix.

6. The DSSS-TDMA system bit timing synchronization method according to claim 5, characterized in that: Step S5 is further as follows: S51, based on the optimal correlation time and the signal arrival position, calculate their difference to generate a time domain error vector; subtract the frequency domain peak position sequence from the preset theoretical frequency position to obtain a frequency domain error vector; multiply the time domain error vector by the confidence index to obtain a weighted time domain error vector; multiply the frequency domain error vector by the relevant credibility to obtain a weighted frequency domain error vector; combine the weighted time domain error vector, the weighted frequency domain error vector and the frequency domain energy matrix to generate a feature vector; S52, construct a multi-layer encoder network, perform feature extraction on the feature vector, and obtain an initial feature sequence; calculate the autocorrelation matrix of the initial feature sequence, perform feature decomposition on the autocorrelation matrix, and obtain a feature subspace; project the initial feature sequence to the feature subspace to obtain a potential feature sequence; copy the potential feature sequence to generate a reference sequence group, calculate the correlation between the current sequence and the reference sequence group, and obtain a feature correlation matrix; S53, based on the potential feature sequence and the feature correlation matrix, respectively use a linear regression model, a kernel regression model and a neural network model to estimate the timing error, and obtain at least three sets of estimated values; calculate a weighted average of the estimated values, perform nonlinear correction on the weighted average, and obtain a bit timing error; S54, based on the bit timing error, the confidence index, the related credibility and the frequency domain energy matrix, respectively calculating the time domain confidence, the frequency domain confidence and the energy confidence; performing weighted fusion on the time domain confidence, the frequency domain confidence and the energy confidence to obtain the system confidence; Compare the system confidence with a preset high threshold value and a preset low threshold value to generate a three-state timing status flag; S55. Based on the timing status flag, system confidence and bit timing error, a parameter update strategy is constructed to generate a control instruction sequence; based on the control instruction sequence, system parameter adjustment is performed, including position parameters and frequency deviation parameters.

7. The DSSS-TDMA system bit timing synchronization method according to claim 6, characterized in that: Step S11 is further as follows: S111, read the received spread spectrum signal, and construct a multi-scale analysis window based on the received spread spectrum signal and the preset basic spread spectrum code length; based on the multi-scale analysis window, perform power distribution calculation on each scale window to obtain at least three groups of power sequences; perform normalization processing on the power sequence to generate a power distribution sequence group; based on the power distribution sequence group, calculate the variance of the power mean of each scale, obtain the optimal window length, and generate an adaptive window length; S112, constructing a complex Gaussian function group based on the power distribution sequence group and the adaptive window length; performing convolution operations on the complex Gaussian function group and the power distribution sequence group respectively to obtain a time-frequency joint distribution sequence; performing singular value decomposition on the time-frequency joint distribution sequence, extracting the main eigenvector, and generating a time-frequency eigenvector; S113, constructing a nonlinear polynomial kernel function based on the time-frequency feature vector; mapping the time-frequency feature vector to a high-dimensional feature space based on the nonlinear polynomial kernel function, calculating the local maximum points in the high-dimensional space, and generating a candidate position sequence; performing cluster analysis on the candidate position sequence, calculating the inter-class distance and intra-class distance; determining the signal arrival position according to the maximum inter-class distance and minimum intra-class distance criterion; S114. Based on the candidate position sequence, a frequency search interval is constructed near the signal arrival position; the frequency search interval is divided into a predetermined number of sub-intervals, and a fractional-order energy accumulation operation is performed on each sub-interval to generate a cumulative energy curve; peak detection is performed on the cumulative energy curve to obtain a coarse frequency estimate; the coarse frequency estimate is substituted into a recursive least squares algorithm, and iterative optimization is performed to obtain a residual frequency deviation.

8. The DSSS-TDMA system bit timing synchronization method according to claim 6, characterized in that: Step S22 is further as follows: S221, constructing an adaptive wavelet basis function family based on the grouped signal sequence and the sampling weight sequence; calculating the Shannon entropy of each basis function based on the adaptive wavelet basis function family; sorting the adaptive wavelet basis function family based on the entropy value of the Shannon entropy, selecting the first K basis functions to generate an optimal basis function group; wherein K are preset parameters; S222, constructing a recursive binary filter bank based on the optimal basis function group and the grouped signal sequence; Based on a recursive binary filter bank, a J-layer decomposition is performed on the packet signal sequence to obtain subband coefficients; Perform local variance estimation on the subband coefficients, and select the optimal number of decomposition layers according to a preset variance threshold to generate multi-layer decomposition coefficients; S223, constructing an adaptive morphological operator based on multi-layer decomposition coefficients; Acquire local signal characteristics, adaptively adjust the size of the structural element based on the adaptive morphological operator and the local signal characteristics, perform opening and closing operations on the multi-layer decomposition coefficients to obtain morphological features; fuse the morphological features with the multi-layer decomposition coefficients to generate enhanced decomposition coefficients; S224, constructing a Gaussian modulation signal dictionary based on the enhanced decomposition coefficients; Based on the Gaussian modulated signal dictionary, the orthogonal matching pursuit algorithm is used to iteratively calculate the projection coefficients of the signal and the dictionary atoms; based on the projection coefficients, the optimal atom combination is selected to generate the time-frequency decomposition coefficients, and the remaining signal is used as the residual signal.

9. The DSSS-TDMA system bit timing synchronization method according to claim 6, characterized in that: Step S42 is further as follows: S421, based on the segmented sequence, calling a pre-built generalized Fourier transform kernel matrix to perform a transform operation to generate an initial transform sequence; S422, based on the initial transformation sequence, using an adaptive order optimization function to perform optimization operations to generate an optimized transformation sequence; S423, based on the optimized transformation sequence, using an adaptive filtering function to perform filtering operations to generate a transformation coefficient sequence; S424: Based on the transform coefficient sequence, a spectrum enhancement function is used to perform enhancement processing to generate an enhanced spectrum sequence.

10. A DSSS-TDMA system bit timing synchronization device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the DSSS-TDMA system bit timing synchronization method according to any one of claims 1 to 9.

Images