Error-decoupling-based PM-BPSK signal demodulation method, device and electronic equipment

Abstract

This application provides a PM-BPSK signal demodulation method, apparatus, and electronic device based on error decoupling. The method includes: acquiring the PM-BPSK signal to be demodulated; performing time-frequency analysis on the PM-BPSK signal using continuous wavelet transform, extracting the wavelet coefficient sequence of the subcarrier neighborhood, and performing circular mapping on its instantaneous phase to obtain the subcarrier phase time-frequency characteristics composed of cosine and sine components; performing decarrier and phase differential processing on the PM-BPSK signal, compensating for the dominant frequency offset based on circular mean, and then orthogonally projecting it with the subcarrier reference signal to obtain phase difference domain phase-locked features; extracting the main channel data from the subcarrier phase time-frequency characteristics and extracting the main channel data from the phase difference domain phase-locked features, and splicing them according to the channel dimension to form a dual-channel fusion feature; inputting the dual-channel fusion feature into a signal demodulation model and outputting a demodulated bit sequence. This application achieves highly robust and high-precision demodulation of PM-BPSK signals, providing a novel technical solution for non-cooperative communication.

Description

Technical Field

[0001] This invention relates to the field of digital signal processing technology, and more specifically to a PM-BPSK signal demodulation method, apparatus, and electronic device based on error decoupling. Background Technology

[0002] PM (Phase Modulation)-BPSK (Binary Phase Shift Keying) is a composite modulation scheme that includes a subcarrier phase modulation structure. Its decision information not only depends on the baseband symbol sequence, but is also closely coupled with the evolution law of the subcarrier phase. Therefore, it has important research value and engineering significance in scenarios such as robust transmission and anti-interference communication in complex electromagnetic environments.

[0003] However, in actual reception, the signal is inevitably affected by non-ideal factors such as additive noise, carrier frequency offset and timing jitter, which leads to cumulative phase drift and symbol boundary offset of the subcarrier component, and in turn causes decision threshold mismatch and bit error rate increase.

[0004] Traditional coherent demodulation methods typically rely on carrier synchronization and timing synchronization loops for parameter estimation and compensation, but their performance is highly sensitive to loop parameters, initial state, and signal-to-noise ratio (SNR) conditions. Especially in low SNR or large frequency offset scenarios, it is often difficult to balance convergence speed and steady-state accuracy, and their robustness under complex and non-ideal conditions is limited. Therefore, researching a highly robust PM-BPSK demodulation method oriented towards frequency offset and timing error disturbances has significant theoretical and practical value.

[0005] In recent years, deep learning has provided new approaches to physical layer demodulation, with related methods gradually shifting from traditional modular processing to end-to-end joint modeling. These methods directly learn the mapping relationship between received signals and information bits through deep networks, aiming to reduce reliance on explicit synchronization, channel estimation, and manually designed rules, and alleviate performance degradation caused by model mismatch and error propagation in traditional serial receiving links. For example, Zheng S and Chen S et al. proposed an end-to-end intelligent receiver framework that unifies the modeling of synchronization, equalization, and demodulation functions, achieving direct recovery from received signals to information bits and verifying its feasibility for blind reception under various modulation and coding schemes. Chen Liujingyan proposed a Transformer-based end-to-end demodulation framework, extracting global dependency information through task-oriented embedding and coding structures, and enhancing local detail representation capabilities by combining deep separable convolutions, thereby improving demodulation robustness under complex and non-ideal conditions. Ye H and Liang L et al. constructed an end-to-end transmit-receive joint optimization framework, which uses CGAN to learn the conditional distribution of unknown channels as a usable channel proxy, thereby realizing gradient backpropagation and joint training at the transmitter in the absence of an explicit channel model.

[0006] While end-to-end joint modeling can reduce the reliance on explicit preprocessing and manual design in traditional receiver links, directly performing decision learning based on the received signal may still face problems such as training instability, convergence difficulties, and insufficient generalization ability in scenarios with complex non-ideal factors, limited sample size, or the need to explicitly preserve modulation structure information. To address this, another type of research tends to first perform explicit representation transformation or feature construction under signal mechanism constraints, and then combine it with a deep model to complete the decision, aiming to achieve a balance between feature interpretability and model robustness. For example, Lijun W and Minghong Z et al. proposed a demodulation method based on generative adversarial networks (GANs), which reconstructs the received signal into an image tensor, allowing information such as phase texture, symbol rate, and carrier frequency texture to be presented in a unified representation. By learning the mapping from the modulated signal image to the demodulated image through GANs, they achieved performance superior to traditional coherent demodulation and CNN demodulation in Gaussian channels. For the short burst SOQPSK demodulation task, Wei Fengyuan, after normalizing, downconverting and synchronizing the signal, rearranged the one-dimensional complex baseband signal into a two-dimensional modulated image according to a fixed structure, and constructed BRCL-Net composed of residual convolution and bidirectional LSTM, which effectively learned the mapping relationship from the modulated image to the demodulated information image, thereby improving the demodulation robustness in short burst scenarios.

[0007] Meanwhile, with the widespread application of self-attention mechanisms in sequence modeling, some studies have further utilized Transformers to characterize long-range dependencies across symbols, achieving global fusion of contextual information and demonstrating certain advantages under complex channel conditions. However, for PM-BPSK signals with subcarrier phase evolution characteristics, if the original phase or instantaneous phase is directly used as a feature, frequency offset will introduce overall phase rotation and slow drift, causing the feature statistical distribution to change over time; while timing jitter will disrupt symbol alignment, making the local decision information exhibit obvious time-varying and unstable characteristics, thereby increasing the difficulty of network training and weakening the model's cross-scenario generalization ability.

[0008] Therefore, how to suppress global phase rotation caused by frequency offset at the feature level, enhance the stability of the representation, and take into account both local discriminative structure extraction and global context modeling at the model level are the key problems that need to be solved to achieve robust demodulation of PM-BPSK signals. Summary of the Invention

[0009] In view of this, this application proposes a PM-BPSK signal demodulation method, apparatus, and electronic device based on error decoupling. Specifically, this application is implemented through the following technical solution: According to a first aspect of the embodiments of this specification, a PM-BPSK signal demodulation method based on error decoupling is provided, comprising: Step S1: Obtain the PM-BPSK signal to be demodulated; Step S2: Perform time-frequency analysis on the PM-BPSK signal using continuous wavelet transform, extract wavelet coefficient sequences in the subcarrier neighborhood, and perform circular mapping on the instantaneous phase of the wavelet coefficient sequences to obtain the subcarrier phase time-frequency characteristics composed of cosine and sine components. The subcarrier phase time-frequency characteristics are used to resist timing errors. Step S3: After decarrier processing of the PM-BPSK signal, construct the complex conjugate product between adjacent sampling points to obtain a complex phase increment sequence. Estimate the dominant rotation of the complex phase increment sequence based on the circular mean and compensate for it. Orthogonally project the real part of the compensated signal with the local subcarrier reference signal to obtain the phase difference domain phase-locked feature. The phase difference domain phase-locked feature is used to resist carrier frequency offset. Step S4: Extract the main channel data from the subcarrier phase time-frequency features, extract the main channel data from the phase difference domain phase-locked features, and stitch the two main channel data together according to the channel dimension to form a dual-channel fusion feature; Step S5: Input the dual-channel fusion feature into the signal demodulation model and output the demodulated bit sequence.

[0010] According to a second aspect of the embodiments of this specification, a PM-BPSK signal demodulation device based on error decoupling is provided, comprising: The signal acquisition unit is used to acquire the PM-BPSK signal to be demodulated; The first feature processing unit is used to perform time-frequency analysis on the PM-BPSK signal using continuous wavelet transform, extract wavelet coefficient sequences in the subcarrier neighborhood, and perform circular mapping on the instantaneous phase of the wavelet coefficient sequences to obtain subcarrier phase time-frequency features composed of cosine and sine components. The subcarrier phase time-frequency features are used to resist timing errors. The second feature processing unit is used to perform carrier removal processing on the PM-BPSK signal, construct the complex conjugate product between adjacent sampling points to obtain a complex phase increment sequence, estimate the dominant rotation amount of the complex phase increment sequence based on the circular mean and perform compensation, and orthogonally project the real part of the compensated signal with the local subcarrier reference signal to obtain the phase difference domain phase-locked feature, which is used to resist carrier frequency offset. The feature fusion unit is used to extract main channel data from the subcarrier phase time-frequency features, extract main channel data from the phase difference domain phase-locked features, and splice the two main channel data according to the channel dimension to form a dual-channel fused feature. The signal demodulation unit is used to input the dual-channel fusion features into the signal demodulation model and output the demodulated bit sequence.

[0011] According to a third aspect of the embodiments of this specification, an electronic device is provided, comprising: a processor; and a computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in the first aspect.

[0012] The embodiments of this application have at least the following technical effects: First, it significantly improves the demodulation robustness of PM-BPSK signals under complex and non-ideal conditions. This application proposes an end-to-end demodulation method that combines multi-feature characterization with a signal demodulation model to address the common composite disturbances such as additive noise, carrier frequency offset, and timing jitter in non-cooperative shortwave signals received in real-world environments. This application can achieve lower bit error rates in scenarios with single frequency offset, single timing error, and a combination of both, with particularly significant performance advantages in the low to medium signal-to-noise ratio (-6~4dB) range. It effectively solves the problem of drastic performance degradation caused by decision threshold mismatch and error accumulation propagation in non-ideal channels of traditional methods. Second, this application proposes a complementary multi-feature representation method, which enhances the adaptability of the input representation to different error mechanisms. This application designs subcarrier phase time-frequency features for timing errors and phase difference domain phase-locked features for frequency offset disturbances, and fuses them into a unified dual-channel time series feature through a main channel selection strategy. This fused feature not only retains the local time-frequency structure related to modulation decision, but also effectively suppresses global phase rotation caused by frequency offset and symbol boundary shift caused by timing jitter. Visualization and ablation experiments show that its ability to preserve decision information and statistical stationarity under composite error conditions are superior to existing schemes such as the original received sequence and structured image representation, providing a more discriminative and robust input for subsequent networks. Third, it reduces the dependence of traditional receiving links on explicit synchronization and manual rules, and simplifies the non-cooperative receiving process. The embodiments of this application do not require the construction of independent carrier synchronization, timing synchronization loops or channel estimation modules. They directly extract dual-channel fusion features from the received signal and combine them with the signal demodulation model to achieve bit-level decision-making. This reduces model mismatch and error propagation problems in serial processing. Compared with existing end-to-end methods, the embodiments of this application are more stable in training and have stronger cross-scenario generalization ability in scenarios where the sample size is limited and the modulation structure needs to be explicitly preserved. This is for shortwave composite modulation signals. In summary, the embodiments of this application effectively overcome the performance bottleneck of existing technologies under non-ideal conditions through mechanism-driven multi-feature representation and deep learning-based network architecture, achieving highly robust and high-precision demodulation of PM-BPSK signals, and providing a new technical solution for intelligent processing of the physical layer of non-cooperative communication. Attached Figure Description

[0013] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Some specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings in an exemplary and non-limiting manner. The same reference numerals in the drawings indicate the same or similar parts or components. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings: Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of the PM-BPSK signal demodulation method based on error decoupling in this application. Figure 2 This is a schematic diagram of an end-to-end demodulation process based on CRTNet, as illustrated in an exemplary embodiment of this application. Figure 3 This is a schematic diagram of a residual convolution block structure shown in an exemplary embodiment of this application; Figure 4This is a demodulation performance diagram under a frequency offset of 200Hz (0.01Rb) as illustrated in an exemplary embodiment of this application; Figure 5 This is a demodulation performance diagram under a timing error of 0.02T, as shown in an exemplary embodiment of this application. Figure 6 This is a demodulation performance diagram illustrating a frequency offset of 0.01Rb and a timing error of 0.02T under an exemplary embodiment of this application; Figure 7 This is a structural block diagram of an electronic device illustrated in an exemplary embodiment of this application; Figure 8 This is a block diagram of an error-decoupling-based PM-BPSK signal demodulation device illustrated in an exemplary embodiment of this application. Detailed Implementation

[0014] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0015] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0016] In non-cooperative or high-dynamic communication environments, the actual reception of PM-BPSK signals is inevitably affected by a combination of non-ideal factors such as additive noise, carrier frequency offset, and timing jitter. Traditional coherent demodulation methods rely on carrier synchronization and timing synchronization loops, which are prone to lock-out in low signal-to-noise ratio or large frequency offset scenarios, exhibit poor robustness against interference, and struggle to balance convergence speed and steady-state accuracy. On the other hand, existing end-to-end deep learning intelligent demodulation methods face problems such as unstable training, difficulty in convergence, and weak cross-scenario generalization ability when directly processing complex signal systems. In particular, frequency offset introduces global phase rotation, and timing jitter disrupts symbol alignment, leading to significant time-varying and unstable features, thus limiting the robustness of the model.

[0017] Therefore, the main technical objective of this application is to overcome the fundamental defects of traditional coherent demodulation under complex and non-ideal conditions, and at the same time solve the technical bottleneck of insufficient feature time-varying and model robustness caused by frequency offset and timing errors in existing deep learning methods, thereby improving the robustness and accuracy of PM-BPSK signal demodulation.

[0018] To achieve the above objectives, this application proposes a PM-BPSK signal demodulation method based on error decoupling. This method focuses on PM-BPSK signals and is particularly suitable for PM-BPSK signal reception and processing in non-cooperative or high-dynamic communication environments. Error decoupling here refers to analyzing the different impact mechanisms of carrier frequency offset and timing jitter—two major non-ideal factors simultaneously present in the actual reception process—on signal representation, and designing independent and complementary feature extraction paths accordingly. Through this decoupling design, the originally coupled composite errors are separated, each receiving targeted suppression and compensation, thereby providing more stable and discriminative input features for subsequent deep network demodulation.

[0019] Based on the above technical objectives, the overall process of this technical solution includes three core steps: feature construction, network demodulation, and performance evaluation.

[0020] In terms of feature construction, the technical solution of this application designs two complementary types of features to address the different impact mechanisms of timing errors and frequency offsets on signal representation. Timing errors mainly lead to local timing mismatches, i.e., symbol boundary shifts. To address this, the technical solution of this application employs continuous wavelet transform to perform multi-scale time-frequency analysis on the signal and extracts phase correlation features in the subcarrier neighborhood to enhance adaptability to local time offsets, thereby constructing a representation that is insensitive to timing jitter. Frequency offsets cause phase accumulation rotation, resulting in an approximately linear phase shift over time. Therefore, the technical solution of this application introduces a phase difference transform and circular mean offset strategy to suppress the average phase increment caused by frequency offsets, obtaining a more stable and robust phase increment feature.

[0021] Through the feature construction driven by the two mechanisms described above, the technical solution of this application successfully decouples frequency offset and timing error, that is, uses different feature channels to respond to different error sources, and the two do not interfere with each other but complement each other. On this basis, the two types of features are fused to construct a unified multi-feature input representation.

[0022] Secondly, in terms of network structure, the technical solution of this application is designed based on the end-to-end demodulation model of CRTNet (CNN-Residual-Transformer Network). This model uses a convolutional neural network (CNN) to extract local temporal discrimination patterns and short-range structural information, while combining a Transformer encoder to realize cross-symbol global dependency modeling and context fusion, and enhances information transmission and error correction capabilities through a residual error correction mechanism, and finally outputs bit-level decision results.

[0023] Finally, the technical solution of this application systematically evaluates the bit error rate performance of the proposed method under different signal-to-noise ratios, single errors, and composite errors, and verifies the effectiveness of the proposed feature construction and network structure through comparative experiments and ablation analysis.

[0024] The embodiments described in this specification will now be described in detail.

[0025] This application provides a PM-BPSK signal demodulation method based on error decoupling. Figure 1 This is a schematic flowchart illustrating an exemplary embodiment of the PM-BPSK signal demodulation method based on error decoupling, as shown in the following example. Figure 1 As shown, the PM-BPSK signal demodulation method includes the following steps: Step S1: Obtain the PM-BPSK signal to be demodulated.

[0026] Step S2: The PM-BPSK signal is subjected to time-frequency analysis using continuous wavelet transform. Wavelet coefficient sequences are extracted in the subcarrier neighborhood, and the instantaneous phase of the wavelet coefficient sequences is circularly mapped to obtain the subcarrier phase time-frequency characteristics composed of cosine and sine components. The subcarrier phase time-frequency characteristics are used to resist timing errors.

[0027] Step S3: After decarrier processing of the PM-BPSK signal, construct the complex conjugate product between adjacent sampling points to obtain a complex phase increment sequence. Estimate the dominant rotation of the complex phase increment sequence based on the circular mean and compensate for it. Orthogonally project the real part of the compensated signal with the local subcarrier reference signal to obtain the phase difference domain phase-locked feature. The phase difference domain phase-locked feature is used to resist carrier frequency offset.

[0028] Step S4: Extract the main channel data from the subcarrier phase time-frequency features, extract the main channel data from the phase difference domain phase-locked features, and stitch the two main channel data together according to the channel dimension to form a dual-channel fusion feature.

[0029] Wherein, the main channel data of the subcarrier phase time-frequency characteristics is the cosine component sequence after circular mapping, and the main channel data of the phase difference domain phase-locked characteristics is the in-phase component sequence after orthogonal projection.

[0030] Step S5: Input the dual-channel fusion feature into the signal demodulation model and output the demodulated bit sequence.

[0031] This embodiment can receive PM-BPSK demodulated signals in non-cooperative or high-dynamic communication environments. After down-conversion and high-speed analog-to-digital converter (ADC) sampling by the radio frequency (RF) front-end, the signal has timing errors caused by multipath effects and transmit / receive clock mismatch. To describe its impact, the actual sampling time is expressed as... ,in , Let be the sampling interval. Then the received sequence containing timing error can be written as: (1) In formula (1), It is complex Gaussian white noise.

[0032] In some embodiments, step S2, extracting the wavelet coefficient sequence in the subcarrier neighborhood, includes: A continuous wavelet transform is performed on the PM-BPSK signal to generate a two-dimensional time-frequency complex matrix; the frequency point closest to the subcarrier frequency is selected on the frequency axis, and the complex sequence on the time axis corresponding to the frequency point is extracted as the wavelet coefficient sequence.

[0033] In some embodiments, step S2 performs circular mapping on the instantaneous phase of the wavelet coefficient sequence, including: Extract the instantaneous phase value of each element in the wavelet coefficient sequence. ; calculate separately and and with the stated The main channel data, which serves as the time-frequency characteristic of the subcarrier phase, is described as follows: The slave channel data serves as the time-frequency characteristic of the subcarrier phase.

[0034] Specifically, this step uses continuous wavelet transform technology to process the received signal. Perform multi-dimensional joint time-frequency analysis to generate a two-dimensional time-frequency complex matrix, denoted as [equation missing]. Then, select the frequency closest to the subcarrier frequency on the frequency axis. frequency And extract the one-dimensional wavelet frequency sequence corresponding to that frequency point: (2) In formula (1), It is a wavelet coefficient sequence.

[0035] Furthermore, the adjacent frequency information contained in the frequency sequence is extracted. In order to eliminate the numerical discontinuity caused by the phase jump, circular mapping is performed on the synchronization phase, that is, the cosine increment and sine increment are calculated respectively. Based on the cosine increment and sine increment, the subcarrier phase time-frequency feature oriented to timing error is constructed. The subcarrier phase time-frequency feature highlights the modulation-related phase information through local time-frequency focusing in the subcarrier neighborhood.

[0036] (3) In formula (3), This is the instantaneous phase value. The main channel data represents the phase-frequency characteristics of the subcarrier. This is the slave channel data representing the phase-frequency characteristics of the subcarrier.

[0037] In some embodiments, step S3, which estimates and compensates for the dominant rotation of the complex phase increment sequence based on circular mean, includes: Calculate the complex summation of the complex phase increment sequence and use the argument of the summation as an estimate of the dominant rotation amount; perform phase-locked compensation on each phase increment based on the estimate of the dominant rotation amount to obtain the compensated signal.

[0038] In some embodiments, step S3 involves orthogonally projecting the compensated real part of the signal onto the local subcarrier reference signal, including: A cosine reference sequence and a sine reference sequence are constructed using the subcarrier frequency. The real part of the compensated signal is orthogonally projected onto the cosine reference sequence and the sine reference sequence, respectively, to obtain an in-phase component sequence and a quadrature component sequence. The in-phase component sequence serves as the main channel data of the phase difference domain phase-locked feature, and the quadrature component sequence serves as the slave channel data of the phase difference domain phase-locked feature.

[0039] The main effect of carrier frequency offset is that the phase of the received signal undergoes an approximately linear cumulative rotation over time. Considering carrier frequency offset... With initial phase The received signal can be represented as: (4) For carrier frequency caused by Doppler shift or local oscillator frequency offset The original complex baseband digital sequence is first subjected to carrier removal processing: (5) In formula (5), This is the signal sequence after carrier removal.

[0040] In this embodiment, step S2 generates a complex phase increment sequence by constructing the complex conjugate product of the current sampling point and the previous sampling point to process the linearly accumulated absolute phase: (6) In formula (6), It is a complex phase increment sequence.

[0041] Based on this, the dominant rotation is estimated using the circular mean square method: (7) In formula (7), The estimated value of the dominant rotation amount.

[0042] Accordingly, phase-locked loop compensation is applied to the phase increment, and the real part of the compensated signal is extracted as a stationary linearized observation: (8) In formula (8), The real part of the compensated signal, The signal after compensation.

[0043] Using subcarrier frequency Construct local in-phase and orthogonal digital reference sequences, and... Perform orthogonal projection to generate a dual-channel phase difference domain phase-locked feature sequence: (9) In formula (9), It is an in-phase component sequence. Orthogonal component sequences .

[0044] Thus, the phase difference domain phase-locked feature constructed in this embodiment weakens the global phase rotation caused by frequency offset through the circular mean value compensation of the phase difference domain, and retains the change components related to the decision through orthogonal projection of the subcarrier. Therefore, it is more suitable for extracting stable decision information under the condition of frequency offset dominance.

[0045] In some embodiments, step S3 uses a main channel strategy to extract and fuse features to generate a unified dual-channel fusion feature.

[0046] The subcarrier phase time-frequency characteristics and phase difference domain phase-locked characteristics are designed to address timing errors and frequency offset disturbances, respectively, and are significantly complementary in terms of applicable scenarios.

[0047] This embodiment combines feature visualization results with pre-experimental analysis to extract main channel data carrying more significant decision-related information from the subcarrier phase time-frequency characterization and phase difference domain phase-locked characterization. A main channel selection strategy is then used to construct fused features, i.e., extracting the main channel data of both types of features separately and combining them according to channel dimension to obtain fused dual-channel time series features. (10) In formula (10), It features dual-channel fusion.

[0048] In some embodiments, the signal demodulation model in step S5 adopts a CRTNet network. The CRTNet network includes a convolutional front-end for extracting local temporal patterns, a Transformer encoder for capturing cross-symbol global dependencies, and an output layer for residual fusion of local master decisions and global correction information. The CRTNet network outputs sample-level decision results, which are then symbol-aggregated to obtain the final demodulated bit sequence.

[0049] like Figure 2 and Figure 3 As shown, the core design idea of ​​the CRTNet network in this embodiment is to adopt a hierarchical collaborative structure of local decision extraction, global residual correction, and fused output. Specifically, the CRTNet network first extracts local temporal discrimination patterns, noise suppression features, and transitional structure information near symbol boundaries from the input features through convolutional branches to form a stable main decision result; at the same time, the Transformer branch performs global modeling on the feature sequence extracted by convolution, captures long-range dependencies across symbols, and learns global context correction information that is complementary to the main decision; finally, the outputs of the two branches are additively fused through a learnable residual error correction fusion mechanism to obtain the final sample-level logits sequence.

[0050] The network architecture in this embodiment fully leverages the advantages of CNN in local fine-grained feature extraction, and also utilizes the powerful sequence global modeling capabilities of Transformer to compensate for non-ideal perturbations, such as local misalignment caused by timing jitter and global drift caused by residual phase drift, thereby significantly improving the training stability and demodulation robustness of the network under compound error conditions.

[0051] The end-to-end demodulation process of the CRTNet network is as follows: First, the dual-channel fused feature sequence is input into the CRTNet network to obtain sample-level decision results; then, a sliding window overlapping fusion strategy is used to average and fuse the predicted probabilities of multiple windows to eliminate boundary effects; finally, the signals are grouped according to the number of sampling points per symbol (sps), the average is calculated, and a decision threshold of 0.5 is used to complete the symbol-level bit decision output. This process achieves a seamless mapping from received features to the final bit sequence, providing a complete demodulation link for subsequent experimental verification.

[0052] Next, the performance of the target signal detection scheme of this application will be verified through experiments.

[0053] (1) The dataset used in this embodiment.

[0054] This embodiment uses PM-BPSK complex baseband signals as the research object. Each sample sequence contains 1024 symbols, with a symbol rate of 20 kBaud, a subcarrier frequency of 64 kHz, a main carrier frequency of 1 kHz, a sampling rate of 320 kHz, corresponding to 16 sampling points per symbol, and a single sequence length of 16384. Sample labels include symbol-level bit sequences and oversampled labels aligned with the sampling points. The dataset is divided into training and validation sets.

[0055] To evaluate the demodulation performance of the proposed method under different non-ideal error conditions, three datasets were constructed: one with a single frequency offset, one with a single timing error, and one with both frequency offset and timing error. The common settings for the three datasets were consistent: the signal-to-noise ratio range was -6 to 4 dB, the stride was 2 dB, the number of training samples was 200, and the number of validation samples was 50 for each signal-to-noise ratio.

[0056] Among them, the single frequency offset dataset is set with a frequency offset of 200Hz and an initial phase offset of 0; the single timing error dataset uses zero-mean Gaussian random jitter and interpolation resampling to construct timing errors, with a jitter standard deviation of 0.02T; the composite error dataset introduces both a 200Hz frequency offset and a 0.02T timing jitter to verify the comprehensive robustness of the model under composite error scenarios.

[0057] (2) Experimental environment of this embodiment.

[0058] The experiments were conducted on an NVIDIA GeForce RTX 4090 server, and deep learning training was implemented using the PyTorch framework. The model was trained in 40 epochs with a batch size of 32, a learning rate of 1e-4, and a weight decay coefficient of 1e-4. Performance was evaluated using the bit error rate (BER). BER was calculated and compared using BER–SNR curves under different signal-to-noise ratio (SNR) conditions.

[0059] (3) Performance simulation.

[0060] Figure 4 The results of the bit error rate (BER) comparison of different demodulation methods at various signal-to-noise ratios (SNRs) were obtained under a frequency offset of 200Hz. Overall, the BER of the demodulation methods based on the original received sequence and TDemodulation network, the traditional coherent demodulation method, and the demodulation method based on subcarrier phase time-frequency representation and CRTNet network are generally at a high level, and the decrease with increasing SNR is relatively limited. This indicates that the above methods are insufficient in representing effective decision information when frequency offset disturbances exist, and therefore it is difficult to achieve stable and accurate demodulation.

[0061] In comparison, the demodulation method based on phase difference domain phase-locked representation and CRTNet network shows a more significant decrease in BER as the signal-to-noise ratio increases, demonstrating superior resistance to frequency offset. This indicates that phase difference domain phase-locked representation can, to some extent, weaken the overall phase rotation effect caused by frequency offset, thereby improving the robustness of feature representation and enhancing the network's ability to characterize symbol change information.

[0062] Furthermore, the fused feature constructed using subcarrier phase time-frequency representation and phase difference domain phase-locked representation achieved optimal BER results across the entire signal-to-noise ratio range under the CRTNet network. This phenomenon indicates that the two types of features have strong complementarity under frequency offset conditions: the subcarrier phase time-frequency representation can characterize the time-frequency phase evolution structure of the signal, while the phase difference domain phase-locked representation helps suppress global phase drift caused by frequency offset. The fusion of the two further enhances the ability of the input representation to preserve key information, thereby significantly improving the robust demodulation performance of CRTNet for PM-BPSK signals under frequency offset error conditions.

[0063] Figure 5 The results show the BER comparison of different demodulation methods at various signal-to-noise ratios (SNRs) with a timing error of 0.02T. Overall, the BER of all methods decreases with increasing SNR, but the magnitude of the decrease varies significantly, indicating that different representations have different adaptability to timing errors. Specifically, the traditional coherent demodulation, the demodulation method based on structured representation and BRCLNet network, and the demodulation method based on the original received sequence and TDemodulation network all have relatively high BERs, and their improvement is limited in the medium-to-high SNR range, indicating that these methods are insufficient in preserving effective decision information when timing offsets exist.

[0064] In comparison, both the demodulation method based on phase difference domain phase-locked loop representation and CRTNet network and the demodulation method based on subcarrier phase time-frequency representation and CRTNet network exhibit lower bit error rates, demonstrating better resistance to timing errors.

[0065] Furthermore, the fused features constructed using subcarrier phase time-frequency representation and phase difference domain phase-locked loop representation achieved optimal BER results across the entire signal-to-noise ratio range under the CRTNet network. These results demonstrate that the two types of features exhibit good complementarity under timing error conditions, and their fusion can further enhance the robustness of the input representation, thereby improving the robust demodulation performance of CRTNet for PM-BPSK signals under timing error conditions.

[0066] Figure 6 The results show the BER comparison of different demodulation methods at various signal-to-noise ratios (SNRs) under the conditions of frequency offset of 0.01Rb and timing error of 0.02T. Overall, the performance differences among the methods further increase under the combined effects of frequency offset and timing error, indicating that the combined error conditions place higher demands on the robustness of the demodulation algorithms. Specifically, the traditional coherent demodulation, the demodulation method based on structured representation and BRCLNet network, the demodulation method based on the original received sequence and TDemodulation network, and the demodulation method based on subcarrier phase time-frequency representation and CRTNet network all exhibit relatively high BERs, and their decrease with increasing SNR is limited, indicating that these methods have insufficient ability to represent effective decision information under combined perturbation conditions.

[0067] In contrast, the phase difference domain phase-locked loop representation and CRTNet network exhibit a lower bit error rate, indicating that this feature has certain advantages in suppressing the overall phase drift caused by frequency offset.

[0068] Furthermore, the fused representation constructed by fusing the subcarrier phase time-frequency representation and the phase difference domain phase-locked representation achieves the best BER across the entire signal-to-noise ratio range under the CRTNet network. This result demonstrates that the complementary fusion of the two types of features can more effectively improve the robustness of the input representation, thereby enhancing the robust demodulation performance of CRTNet for PM-BPSK signals under composite error conditions.

[0069] In summary, the embodiments of this application have achieved at least the following beneficial effects: First, it significantly improves the demodulation robustness of PM-BPSK signals under complex and non-ideal conditions.

[0070] This application proposes an end-to-end demodulation method combining multi-feature representation with the CRTNet network to address the common composite disturbances in practical non-cooperative shortwave reception, such as additive noise, carrier frequency offset, and timing jitter. Compared with traditional coherent demodulation and existing end-to-end deep learning methods (T-Demodulator, BRCL-Net, etc.), this application achieves a lower bit error rate in scenarios with a single frequency offset (200Hz), a single timing error (0.02T), and a combination of both. The performance advantage is particularly significant in the low to medium signal-to-noise ratio (-6~4dB) range, effectively solving the problem of drastic performance degradation caused by decision threshold mismatch and error accumulation propagation in non-ideal channels of traditional methods.

[0071] Second, a complementary multi-feature representation method is proposed, which enhances the adaptability of input representation to different error mechanisms.

[0072] This application's embodiments design subcarrier phase time-frequency features for timing errors and phase difference domain phase-locked features for frequency offset disturbances, respectively. These features are then fused into a unified dual-channel time series feature through a main channel selection strategy. This fused feature not only preserves the local time-frequency structure related to modulation decision but also effectively suppresses global phase rotation caused by frequency offset and symbol boundary shift caused by timing jitter. Visualization and ablation experiments show that, under combined error conditions, its ability to preserve decision information and its statistical stationarity are superior to existing schemes such as the original received sequence and structured image representation, providing a more discriminative and robust input for subsequent networks.

[0073] Third, the CRTNet network was designed to achieve collaborative optimization of local fine-grained modeling and global context fusion.

[0074] The CRTNet proposed in this application employs a convolutional front-end, which, for example, uses residual dilated convolutional blocks to extract local discriminative patterns and noise suppression features within symbols. The Transformer encoder models long-range dependencies across symbols, and a learnable residual error correction fusion mechanism achieves information complementarity and error compensation. This structure maintains training stability while significantly improving adaptability to timing jitter and residual phase drift. Furthermore, ablation experiments demonstrate that the performance of CNN alone or Transformer alone is inferior to the complete CRTNet, indicating that the synergistic effect of local and global modeling and the residual error correction mechanism is key to performance improvement.

[0075] Fourth, it reduces the reliance of traditional receiving links on explicit synchronization and manual rules, and simplifies the non-cooperative receiving process.

[0076] This application's embodiments eliminate the need to construct independent carrier synchronization, timing synchronization loops, or channel estimation modules. Bit-level decision-making is achieved directly from received observations through multi-feature representation and CRTNet, reducing model mismatch and error propagation issues in serial processing. Compared to existing end-to-end methods, this invention offers more stable training and stronger cross-scenario generalization capabilities in scenarios with limited sample size and explicit preservation of modulation structure.

[0077] Figure 7 This is a schematic diagram of an electronic device illustrated in this specification according to an exemplary embodiment. Please refer to... Figure 7 At the hardware level, the device includes a processor 702, an internal bus 704, a network interface 706, memory 708, a hardware acceleration device 710, and non-volatile memory 712, and may also include other hardware required for its functions. One or more embodiments of this application can be implemented in software, for example, the processor 702 reads the corresponding computer program from the non-volatile memory 712 into the memory 708 and then runs it. Of course, in addition to software implementation, one or more embodiments of this application do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the above processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0078] Figure 8 This is a structural block diagram of an error-decoupling-based PM-BPSK signal demodulation device illustrated in an exemplary embodiment of this application. The PM-BPSK signal demodulation device can be applied to, for example... Figure 7 The electronic device shown implements the technical solution of this application. The PM-BPSK signal demodulation device includes: a signal acquisition unit 810, a first feature processing unit 820, a second feature processing unit 830, a feature fusion unit 840, and a signal demodulation unit 850, wherein: The signal acquisition unit 810 is used to acquire the PM-BPSK signal to be demodulated; The first feature processing unit 820 is used to perform time-frequency analysis on the PM-BPSK signal using continuous wavelet transform, extract wavelet coefficient sequences in the subcarrier neighborhood, and perform circular mapping on the instantaneous phase of the wavelet coefficient sequences to obtain subcarrier phase time-frequency features composed of cosine and sine components. The subcarrier phase time-frequency features are used to resist timing errors. The second feature processing unit 830 is used to perform carrier removal processing on the PM-BPSK signal, construct the complex conjugate product between adjacent sampling points to obtain a complex phase increment sequence, estimate the dominant rotation amount of the complex phase increment sequence based on the circular mean and perform compensation, and orthogonally project the real part of the compensated signal with the local subcarrier reference signal to obtain the phase difference domain phase-locked feature, which is used to resist carrier frequency offset. The feature fusion unit 840 is used to extract main channel data from the subcarrier phase time-frequency features, extract main channel data from the phase difference domain phase-locked features, and splice the two main channel data according to the channel dimension to form a dual-channel fusion feature. The signal demodulation unit 850 is used to input the dual-channel fusion features into the signal demodulation model and output the demodulated bit sequence.

[0079] In some embodiments, the first feature processing unit 820 is used to extract the instantaneous phase value of each element in the wavelet coefficient sequence. ;calculate and with the stated The main channel data serves as the time-frequency characteristic of the subcarrier phase.

[0080] In some embodiments, the first feature processing unit 820 is used to perform continuous wavelet transform on the PM-BPSK signal to generate a two-dimensional time-frequency complex matrix; select the frequency point on the frequency axis that is closest to the subcarrier frequency, and extract the complex sequence on the time axis corresponding to the frequency point as the wavelet coefficient sequence.

[0081] In some embodiments, the second feature processing unit 830 is used to calculate the complex sum of the complex phase increment sequence and use the argument of the sum as an estimate of the dominant rotation amount; and to perform phase-locked compensation on each phase increment based on the estimate of the dominant rotation amount to obtain the compensated signal.

[0082] In some embodiments, the second feature processing unit 830 is used to construct a cosine reference sequence and a sine reference sequence using the subcarrier frequency; and to orthogonally project the real part of the compensated signal with the cosine reference sequence and the sine reference sequence to obtain an in-phase component sequence, wherein the in-phase component sequence serves as the main channel data of the phase difference domain phase-locked feature.

[0083] In some embodiments, the signal demodulation model employs a CRTNet network, which includes a convolutional front-end for extracting local temporal patterns, a Transformer encoder for capturing cross-symbol global dependencies, and an output layer for residual fusion of local master decisions and global correction information. The CRTNet network outputs sample-level decision results, which are then symbol-aggregated to obtain the final demodulated bit sequence.

[0084] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0085] Accordingly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the above embodiments.

[0086] Accordingly, embodiments of this application also provide a computer program product configured to perform the methods described in any of the above embodiments.

[0087] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0088] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0089] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0090] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random-access memory (SRAM), dynamic random-access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0091] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0092] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0093] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0094] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0095] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A PM-BPSK signal demodulation method based on error decoupling, characterized in that, Includes the following steps: Step S1: Obtain the PM-BPSK signal to be demodulated; Step S2: Perform time-frequency analysis on the PM-BPSK signal using continuous wavelet transform, extract wavelet coefficient sequences in the subcarrier neighborhood, and perform circular mapping on the instantaneous phase of the wavelet coefficient sequences to obtain the subcarrier phase time-frequency characteristics composed of cosine and sine components. The subcarrier phase time-frequency characteristics are used to resist timing errors. Step S3: After decarrier processing of the PM-BPSK signal, construct the complex conjugate product between adjacent sampling points to obtain a complex phase increment sequence. Estimate the dominant rotation of the complex phase increment sequence based on the circular mean and compensate for it. Orthogonally project the real part of the compensated signal with the local subcarrier reference signal to obtain the phase difference domain phase-locked feature. The phase difference domain phase-locked feature is used to resist carrier frequency offset. Step S4: Extract the main channel data from the subcarrier phase time-frequency features, extract the main channel data from the phase difference domain phase-locked features, and stitch the two main channel data together according to the channel dimension to form a dual-channel fusion feature; Step S5: Input the dual-channel fusion feature into the signal demodulation model and output the demodulated bit sequence.

2. The method according to claim 1, characterized in that, Step S2 involves circular mapping of the instantaneous phase of the wavelet coefficient sequence, including: Extract the instantaneous phase value of each element in the wavelet coefficient sequence. ; calculate and with the stated The main channel data serves as the time-frequency characteristic of the subcarrier phase.

3. The method according to claim 1, characterized in that, Step S2, which extracts the wavelet coefficient sequence in the subcarrier neighborhood, includes: Perform continuous wavelet transform on the PM-BPSK signal to generate a two-dimensional time-frequency complex matrix; Select the frequency point on the frequency axis that is closest to the subcarrier frequency, and extract the complex sequence on the time axis corresponding to the frequency point as the wavelet coefficient sequence.

4. The method according to claim 1, characterized in that, Step S3, which estimates and compensates for the dominant rotation of the complex phase increment sequence based on circular mean, includes: Calculate the complex summation of the complex phase increment sequence, and use the argument of the summation as an estimate of the dominant rotation amount; Based on the estimated value of the dominant rotation, phase-locked compensation is performed on each phase increment to obtain the compensated signal.

5. The method according to claim 1, characterized in that, Step S3 involves orthogonally projecting the compensated real part of the signal onto the local subcarrier reference signal, including: Construct cosine and sine reference sequences using subcarrier frequencies; The real part of the compensated signal is orthogonally projected onto the cosine reference sequence and the sine reference sequence to obtain the in-phase component sequence, which serves as the main channel data of the phase difference domain phase-locked feature.

6. The method according to claim 1, characterized in that, In step S4, the main channel data of the subcarrier phase time-frequency characteristics is the cosine component sequence after circular mapping.

7. The method according to claim 1, characterized in that, In step S4, the main channel data of the phase difference domain phase-locked feature is the in-phase component sequence after orthogonal projection.

8. The method according to any one of claims 1 to 7, characterized in that, In step S5, the signal demodulation model adopts the CRTNet network. The CRTNet network includes a convolutional front-end for extracting local temporal patterns, a Transformer encoder for capturing cross-symbol global dependencies, and an output layer for residual fusion of local master decisions and global correction information. The CRTNet network outputs sample-level decision results, which are then symbol-aggregated to obtain the final demodulated bit sequence.

9. A PM-BPSK signal demodulation device based on error decoupling, characterized in that, The signal acquisition unit is used to acquire the PM-BPSK signal to be demodulated; The first feature processing unit is used to perform time-frequency analysis on the PM-BPSK signal using continuous wavelet transform, extract wavelet coefficient sequences in the subcarrier neighborhood, and perform circular mapping on the instantaneous phase of the wavelet coefficient sequences to obtain subcarrier phase time-frequency features composed of cosine and sine components. The subcarrier phase time-frequency features are used to resist timing errors. The second feature processing unit is used to perform carrier removal processing on the PM-BPSK signal, construct the complex conjugate product between adjacent sampling points to obtain a complex phase increment sequence, estimate the dominant rotation amount of the complex phase increment sequence based on the circular mean and perform compensation, and orthogonally project the real part of the compensated signal with the local subcarrier reference signal to obtain the phase difference domain phase-locked feature, which is used to resist carrier frequency offset. The feature fusion unit is used to extract main channel data from the subcarrier phase time-frequency features, extract main channel data from the phase difference domain phase-locked features, and splice the two main channel data according to the channel dimension to form a dual-channel fused feature. The signal demodulation unit is used to input the dual-channel fusion features into the signal demodulation model and output the demodulated bit sequence.

10. An electronic device, characterized in that, include: processor; as well as A computer-readable storage medium storing computer program instructions that, when executed by the processor, cause the processor to perform the method as described in any one of claims 1 to 8.

Images