A method and system for separating unknown interference communication signals

By employing multi-path parallel diffusion-based blind separation and matching score recognition, along with optimal branch output, the accuracy and robustness issues of separating unknown interference signals are resolved, achieving efficient signal separation in complex electromagnetic environments.

CN122204602BActive Publication Date: 2026-11-10CHINA ELECTRONICS TECH GRP NO 7 RES INST
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202610293578.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-03-11
Publication Date
2026-11-10
Estimated Expiration
2046-03-11

AI Technical Summary

Technical Problem

In complex electromagnetic environments, existing technologies struggle to effectively separate unknown and time-varying interference signals, leading to increased demodulation error rates and communication link interruptions. Traditional methods suffer from decreased identification accuracy and separation performance under conditions of high signal-to-interference ratios and the coexistence of multiple interference sources.

Method used

A multi-path parallel diffusion-type blind separation method is adopted. The initial estimate of the target signal is generated through blind synchronization and remodulation. The signal is separated using a diffusion model, and the optimal branch is identified by matching score, so as to achieve accurate output of the interference signal.

Benefits of technology

Without requiring prior information, it significantly improves the accuracy and robustness of signal separation, ensuring the real-time performance and separation effectiveness of the algorithm.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122204602B_ABST
    Figure CN122204602B_ABST
Patent Text Reader

Abstract

This application belongs to the field of signal processing and proposes a method and system for separating unknown interference communication signals. The method includes: receiving a mixed signal containing unknown interference; performing blind synchronization and remodulation on the mixed signal based on the modulation type of the target signal to generate an initial estimate of the target signal; inputting the blind-synchronized mixed signal and the initial estimate of the target signal into a diffusion model containing multiple modulation type branches; the diffusion model, based on the blind-synchronized mixed signal and the initial estimate of the target signal, obtaining the target signal estimate and interference signal estimate for each branch through multi-path parallel diffusion-based blind separation; calculating the matching score between the interference signal estimate of each branch and the modulation type corresponding to that branch; determining the optimal branch based on the matching score; and using the target signal estimate and interference signal estimate corresponding to the optimal branch as the separation result. Compared with existing technologies, this invention significantly improves the accuracy of unknown interference signal separation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of signal processing technology, and proposes a signal separation method and system for unknown interference communication signals. Background Technology

[0002] In complex electromagnetic environments such as heterogeneous network converged communication, cognitive radio, and anti-jamming communication, the useful signals received by wireless communication receivers are often inevitably affected by other time-frequency overlapping communication signals. These interference signals may originate from other communication systems in the same frequency band, malicious jamming devices, or non-cooperative communication nodes, and their modulation schemes, symbol rates, power, and other parameters are usually dynamically changing and unknown. Due to the presence of interference signals, the demodulation bit error rate at the receiver increases significantly, and in severe cases, it can even lead to communication link interruption and complete demodulation failure. Traditional methods based on matched filtering or coherent demodulation experience a sharp decline in performance under interference environments, making it difficult to meet the requirements of high-reliability communication.

[0003] Currently, the existing technology typically employs the Blind Source Separation (BSS) method, which separates and removes interfering signals from the received mixed signals without requiring prior channel information.

[0004] However, in cognitive radio or military communication countermeasures scenarios, the modulation patterns and power levels of jamming signals are usually unknown and time-varying, which makes it difficult to directly apply existing BSS methods. Under conditions of high signal-to-interference ratio, complex jamming signal types, or coexistence of multiple jamming sources, the identification accuracy and separation performance often decrease significantly, making it difficult to meet the needs of practical applications. Summary of the Invention

[0005] To overcome the shortcomings of the prior art in terms of reduced interference signal separation performance, this invention proposes a method and system for separating unknown interference communication signals.

[0006] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows: A method for separating unknown interference communication signals, the method comprising: S1. Receive a mixed signal containing unknown interference, perform blind synchronization and remodulation on the mixed signal based on the target signal modulation type, and generate an initial estimate of the target signal; S2. Input the initial estimate of the mixed signal after blind synchronization and the target signal into a diffusion model containing multiple modulation type branches. The diffusion model obtains the target signal estimate and interference signal estimate of each branch through multi-path parallel diffusion-type blind separation based on the initial estimate of the mixed signal after blind synchronization and the target signal. S3. Based on the interference signal estimation of each branch, calculate its matching score with the modulation type corresponding to that branch; S4. Determine the optimal branch based on the matching score, and use the target signal estimate and interference signal estimate corresponding to the optimal branch as the separation result.

[0007] As a preferred embodiment, the step of generating the initial estimate of the target signal includes: using the mixed signal after blind synchronization as a coarse estimate, performing deconstellation mapping to recover the bit sequence, and remodulating the bit sequence according to the modulation type parameter of the target signal to obtain the initial estimate of the target signal.

[0008] As a preferred embodiment, the diffusion model includes a target signal diffusion model and an interference signal diffusion model; the target signal diffusion model is trained based on pre-constructed target signal training samples; the interference signal diffusion model includes multiple interference signal diffusion models corresponding to the various modulation types.

[0009] Furthermore, the diffusion model, based on the initial estimation of the mixed signal and the target signal after blind synchronization, obtains the target signal estimate and interference signal estimate for each branch through multi-path parallel diffusion-type blind separation, including: For each branch, diffusion separation is performed iteratively for a preset number of iterations. When the iteration is completed, the target signal estimate and interference signal estimate of that branch are obtained. The diffusion separation includes the following steps: Randomly sample the first and second time steps; Based on the first time step and the predetermined noise scheduling strategy, the target signal estimate for the current round is denoised to obtain the first denoised signal; Based on the second time step and the predetermined noise scheduling strategy, the interference signal estimate for the current round is denoised to obtain a second denoised signal; Based on the target signal diffusion model, noise in the first noisy signal is predicted and removed, and the target signal estimate is updated. Based on the interference signal diffusion model corresponding to this branch modulation type, the noise in the second noisy signal is predicted and removed, and the interference signal estimate is updated. The expressions for the target signal estimation and interference signal estimation are as follows:

[0010]

[0011]

[0012]

[0013] in, and They are the firstk The branch in the i The target signal estimation and interference signal estimation obtained in the round iteration are as follows: and To utilize the target signal diffusion model respectively and interference signal diffusion model For noisy samples and The prediction results for noise in the data; and These are the target signal estimation and interference signal estimation at the 1st and 2nd respectively. i Score estimation under sub-noise sampling; The synchronized mixed signal; ~ For the first k The branch in the Random noise independently sampled from a standard normal distribution during each iteration; and They are respectively with time step The corresponding signal attenuation coefficient; For the first k The mixing coefficient of each branch, For the first The learning rate of the round; and The first k The branch in the The noise-added results of target signal estimation and interference signal estimation in the round of iteration.

[0014] As a preferred embodiment, during the training process, the diffusion model performs forward noise addition and reverse denoising according to the standard diffusion mechanism, and is trained and optimized using a complex IQ signal dataset of the corresponding modulation type; the training objective of the diffusion model is to minimize the noise level at any time step. Below, the predicted noise and the actual injected noise are compared. The mean squared error between them is given by the following loss function:

[0015] in, It is the first step t The corresponding signal attenuation coefficient.

[0016] As a preferred embodiment, the noise addition is based on random noise that follows a standard normal distribution, and the expressions for the first and second noise-added signals are as follows:

[0017]

[0018] in, and They are the first The branch in the The target signal estimate and interference signal estimate are obtained from rounds of iteration.

[0019] As a preferred embodiment, the steps for calculating the matching score include: Randomly sample multiple time steps; The interference signal is estimated to be noisy multiple times according to different time steps to obtain multiple noisy samples; Noise prediction is performed on the noisy sample based on the interference signal diffusion model corresponding to the branch modulation type. Based on the noise prediction error, the matching score is calculated, and the formula for calculating the matching score is as follows: Calculation branch Assuming the modulation type matching score:

[0020]

[0021] in, To pass the interference signal diffusion model Noise sampling The noise prediction results and matching scores Indicates interference signal estimation With modulation type The degree of matching; For random sampling points The corresponding signal attenuation coefficient; ~ For the first k The first branch l Random noise from sub-independent sampling L This represents the number of time steps.

[0022] Finally, the branch with the highest matching score is selected as the optimal branch.

[0023]

[0024] As a preferred embodiment, the method further includes: calculating the signal-to-interference ratio and the mixing coefficient after obtaining the separation result, the calculation formula of which is as follows:

[0025]

[0026] in, The signal-to-interference ratio of the k-th branch. Let be the mixing coefficient of the k-th branch. For target signal estimation, For interference signal estimation.

[0027] This invention also proposes a signal separation system for unknown interference communication signals, and the application of the aforementioned signal separation method for unknown interference communication signals includes: The preprocessing module is used to receive the mixed signal and perform preprocessing to obtain an initial estimate of the mixed signal after blind synchronization and the target signal. The multi-modulation diffusion separation module receives the initial estimate of the mixed signal and the target signal after blind synchronization, and performs multi-parallel diffusion-type blind separation based on the diffusion model to obtain the target signal estimate and interference signal estimate for each branch. The parameter estimation module receives interference signal estimates based on each branch, calculates the matching score between the estimated signal and the corresponding candidate modulation type, selects the branch with the highest matching score as the optimal branch, identifies the modulation type of the interference signal, and estimates the mixing coefficients. The output module estimates and outputs the final target signal based on the target signal corresponding to the optimal branch.

[0028] The present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a signal separation method for unknown interference communication signals as described in the present invention.

[0029] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a signal separation method for unknown interference communication signals as described in the present invention.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention introduces a diffusion model containing multiple candidate modulation types to perform multi-path parallel diffusion-based blind separation of mixed signals. It also utilizes the matching score between the interference signal estimation and the candidate modulation type to achieve automatic identification of the interference type and accurate output of the corresponding target signal. This application can effectively address the impact of unknown modulation type interference on target signal extraction without any prior information, significantly improving the accuracy and robustness of signal separation in complex electromagnetic environments. At the same time, the multi-path parallel processing architecture ensures the real-time performance of the algorithm. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the implementation of a signal separation method for unknown interference communication signals proposed in Example 1. Figure 2 This is a flowchart of the mixed signal preprocessing process in Example 1.

[0032] Figure 3Here is an architecture diagram of a signal separation system for unknown interference communication signals proposed in Example 2; Figure 4 The modulation recognition accuracy of interference signals with different modulation types under the condition that the target signal is BPSK in Example 3; Figure 5 The modulation recognition accuracy of interference signals with different modulation types under the condition that the target signal is QPSK in Example 3; Figure 6 This is a curve showing the mean square error of various target signals as a function of SIR under random modulation type signal interference conditions in Example 3; Figure 7 The curves show the bit error rate of various target signals as a function of SIR under random modulation type signal interference conditions in Example 3. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0034] Example 1 This embodiment proposes a method for separating unknown interference communication signals, and the implementation flowchart is as follows: Figure 1 As shown.

[0035] A method for separating unknown interference communication signals, the method comprising: S1. Receive a mixed signal containing unknown interference, perform blind synchronization and remodulation on the mixed signal based on the target signal modulation type, and generate an initial estimate of the target signal; S2. Input the initial estimate of the blind-synchronized mixed signal and the target signal into a diffusion model containing multiple modulation type branches. The diffusion model obtains the target signal estimate and interference signal estimate of each branch through multi-path parallel diffusion-type blind separation based on the initial estimate of the blind-synchronized mixed signal and the target signal. S3. Based on the interference signal estimation of each branch, calculate its matching score with the modulation type corresponding to that branch; S4. Determine the optimal branch based on the matching score, and use the target signal estimate and interference signal estimate corresponding to the optimal branch as the separation result.

[0036] In this embodiment, firstly, blind synchronization and remodulation are performed on the mixed signal to generate an initial estimate that matches the modulation type of the target signal. This initial estimate provides effective prior information for subsequent accurate separation. Then, parallel diffusion-based blind separation is performed in a diffusion model containing multiple modulation type branches. The matching score between the interference signal estimate of each branch and its corresponding modulation type is calculated to evaluate the separation effect. The branch with the highest matching score is selected as the final separation result. This embodiment effectively solves the technical problem of accurately separating the target signal from the mixed signal when the modulation type of the interference signal is completely unknown by introducing the modulation prior of the target signal and a multi-branch parallel blind separation mechanism, significantly improving the accuracy and robustness of signal separation.

[0037] More specifically, in the BSS problem, the observed complex mixed signal (length is...) ) Target signals that are statistically independent and energy normalized and interference signals It is formed by linear superposition. This process can be formalized as the following underdetermined linear mixture model:

[0038] in, The mixing coefficients correspond to the target signal. and interference signals The letter between them .

[0039] In an optional embodiment, the step of generating an initial estimate of the target signal includes: using the blind-synchronized mixed signal as a coarse estimate, performing deconstellation mapping to recover the bit sequence, and remodulating the bit sequence according to the modulation type parameters of the target signal to obtain the initial estimate.

[0040] In this embodiment, by performing deconstellation mapping and remodulation on the mixed signal after blind synchronization, the simulated mixed signal problem is transformed into an estimation problem based on symbolic knowledge, which provides a starting point for the iterative optimization of the subsequent diffusion model and helps to accelerate the convergence speed of the diffusion model.

[0041] like Figure 2 The diagram shown is a flowchart of mixed signal preprocessing.

[0042] More specifically, for mixed signals Preprocessing such as synchronization and remodulation is performed to generate an initial estimate of the target signal. The steps include: based on the target signal Modulation type, for the received mixed signal Blind synchronization processing is performed to obtain the synchronized mixed signal. , for The signal length. The target signal was eliminated. The sampling timing error and frequency offset and phase offset, and The sampling rate is the target signal Integer multiples of the symbol rate. Then directly... As After performing constellation mapping based on the coarse estimate and recovering the bit sequence, the target signal is then used. Modulation type parameters and The sampling rate is remodulated, and the remodulated signal is used as the initial estimate of the target signal. .

[0043] In one optional embodiment, the diffusion model includes a target signal diffusion model and an interference signal diffusion model; the target signal diffusion model is trained based on pre-constructed target signal training samples; the interference signal diffusion model includes multiple interference signal diffusion models corresponding to the various modulation types.

[0044] More specifically, for the initial set of candidate branches as ,in This represents the total number of predefined candidate modulation types. Each branch corresponds to one candidate modulation type for the interference signal. In the first In each branch, a pre-trained target signal diffusion model is used. With branches The interference signal diffusion model corresponding to the assumed modulation type For the synchronized mixed signal Separation is performed to obtain estimation results of the target signal and interference signal under the current modulation type assumption. and .

[0045] Further optionally, the step of obtaining the target signal estimate and interference signal estimate for each branch through multi-path parallel diffusion-type blind separation based on the initial estimate of the mixed signal and the target signal after blind synchronization in the diffusion model includes: For each branch, diffusion separation is performed iteratively for a preset number of iterations. When the iteration is completed, the target signal estimate and interference signal estimate of that branch are obtained. The diffusion separation includes the following steps: Randomly sample the first and second time steps; Based on the first time step and the predetermined noise scheduling strategy, the target signal estimate for the current round is denoised to obtain the first denoised signal; Based on the second time step and the predetermined noise scheduling strategy, the interference signal estimate for the current round is denoised to obtain a second denoised signal; Based on the target signal diffusion model, noise in the first noisy signal is predicted and removed, and the target signal estimate is updated. Based on the interference signal diffusion model corresponding to this branch modulation type, the noise in the second noisy signal is predicted and removed, and the interference signal estimate is updated. The expressions for the target signal estimation and interference signal estimation are as follows:

[0046]

[0047]

[0048]

[0049] in, and They are the first k The branch in the i The target signal estimation and interference signal estimation obtained in the round iteration are as follows: and To utilize the target signal diffusion model respectively and interference signal diffusion model For noisy samples and The prediction results for noise in the data; and These are the target signal estimation and interference signal estimation at the 1st and 2nd respectively. i Score estimation under sub-noise sampling; The synchronized mixed signal; ~ For the first k The branch in the Random noise independently sampled from a standard normal distribution during each iteration; and They are respectively with time step The corresponding signal attenuation coefficient; For the first k The mixing coefficient of each branch, For the first The learning rate of the round; and The first k The branch in the The noise-added results of target signal estimation and interference signal estimation in the round of iteration.

[0050] In this embodiment, a composite diffusion model architecture is constructed, consisting of a general target signal diffusion model and multiple dedicated interference signal diffusion models. The target signal diffusion model focuses on learning the distribution characteristics of the target signal, while each interference signal diffusion model is trained for a specific modulation type, capturing the statistical characteristics and structural information of that type of interference signal. During the separation process, these models are used in parallel and iteratively to denoise both the target and interference signals, allowing the two estimated signals to approximate their true data distributions during the iteration process. This embeds prior knowledge into the model structure and separation process, improving the model's ability to separate and generalize complex mixed signals.

[0051] In one optional embodiment, during training, the diffusion model performs forward noise addition and reverse denoising according to the standard diffusion mechanism, and is trained and optimized using a complex IQ signal dataset of the corresponding modulation type; the training objective of the diffusion model is to minimize the noise level at any time step. Below, the predicted noise and the actual injected noise are compared. The mean squared error between them is given by the following loss function:

[0052] in, It is the first step t The corresponding signal attenuation coefficient.

[0053] In this embodiment, training is performed by minimizing the mean square error between the predicted noise and the actual noise.

[0054] In an optional embodiment, the noise addition is based on random noise that follows a standard normal distribution, and the expressions for the first and second noisy signals are as follows:

[0055]

[0056] in, and They are the first The branch in the The target signal and the estimated interference signal are obtained from rounds of iteration. hour, , .

[0057] In this embodiment, during the separation stage, random noise is added to the current estimated signal and denoised using a model, thereby using the prior knowledge learned by the model to correct the current estimate and make it more consistent with the inherent pattern of the modulation type signal.

[0058] In an optional embodiment, the calculation step of the matching score includes: Randomly sample multiple time steps; The interference signal is estimated to be noisy multiple times according to different time steps to obtain multiple noisy samples; Noise prediction is performed on the noisy sample based on the interference signal diffusion model corresponding to the branch modulation type. Based on the noise prediction error, the matching score is calculated, and the formula for calculating the matching score is as follows: Calculation branch Assuming the modulation type matching score:

[0059]

[0060] in, To pass the interference signal diffusion model Noise sampling The noise prediction results and matching scores Indicates interference signal estimation With modulation type The degree of matching; For random sampling points The corresponding signal attenuation coefficient; ~ For the first k The first branch l Random noise from sub-independent sampling L This represents the number of time steps.

[0061] If interference signal The actual modulation type is If the noise disturbance along the diffusion path can be accurately reproduced by the corresponding model, the noise error should be small and the matching score should be high; otherwise, if modulation type and If there is a mismatch, the model cannot effectively predict noise, the noise error is large, and the matching score is low.

[0062] In this embodiment, if the modulation type of an estimated interference signal happens to match the modulation type of an interference signal diffusion model, then the model can accurately predict the performance of the estimated signal at various noise levels, meaning the prediction error will be small. Conversely, if the two do not match, the prediction error will be large. By combining the prediction errors from multiple different time steps, a robust and quantified matching index can be obtained. Finally, the branch with the highest matching score is selected as the optimal branch.

[0063]

[0064] In an optional embodiment, the method further includes: calculating the signal-to-interference ratio and the mixing coefficient after obtaining the separation result, the calculation formula of which is as follows:

[0065]

[0066] in, The signal-to-interference ratio of the k-th branch. Let be the mixing coefficient of the k-th branch. For target signal estimation, For interference signal estimation.

[0067] In this embodiment, after obtaining the separated target signal and interference signal, key indicators such as signal-to-interference ratio and mixing coefficient are further calculated to evaluate the separation performance of this method in real time. It can also provide valuable reference data for subsequent applications such as communication demodulation, interference suppression or signal identification, thereby enhancing the practicality and information output capability of this method.

[0068] Example 2 This embodiment proposes a signal separation system for unknown interference communication signals, applying the signal separation method for unknown interference communication signals proposed in Embodiment 1. For example... Figure 3 The diagram shown is an architecture diagram of a signal separation system for unknown interference communication signals according to this embodiment.

[0069] This embodiment proposes a signal separation system for unknown interference communication signals, including: The preprocessing module is used to receive the mixed signal and perform preprocessing to obtain an initial estimate of the mixed signal after blind synchronization and the target signal. The multi-modulation diffusion separation module receives the initial estimate of the mixed signal and the target signal after blind synchronization, and performs multi-parallel diffusion-type blind separation based on the diffusion model to obtain the target signal estimate and interference signal estimate for each branch. The parameter estimation module receives interference signal estimates based on each branch, calculates the matching score between the estimated signal and the corresponding candidate modulation type, selects the branch with the highest matching score as the optimal branch, identifies the modulation type of the interference signal, and estimates the mixing coefficients. The output module estimates and outputs the final target signal based on the target signal corresponding to the optimal branch.

[0070] In an optional embodiment, a dynamic branch selection strategy is adopted to improve the robustness and computational efficiency of the algorithm. This strategy dynamically selects a set of candidate branches with higher matching scores after the parameter estimation module has been executed. ( ), and update the corresponding branch. Then, proceed to the next round of inter-module iteration.

[0071] The characteristic structure of a certain signal modulation type generally shows high similarity only to a limited number of other modulation types, rather than broad similarity. Therefore, most modulation type assumptions, due to significant model mismatch with the real interference signal, will quickly exhibit low matching scores after the initial blind separation. Thus, by dynamically filtering branches after each round of inter-module iterations, a large number of low-quality candidate branches can be quickly and effectively eliminated, significantly reducing the computational complexity of the algorithm.

[0072] Termination condition for inter-module iteration: When the maximum number of inter-module iterations is reached Or when the optimal branch in this round Compared to the optimal branch in the previous round The mean square error between the target signal estimates in the two rounds of module iterations is the same, and the mean square error between the two rounds of module iterations is less than a preset threshold. When the result reaches 1e-5, the blind separation process is determined to have converged, and the algorithm terminates; otherwise, the current result is updated to the previous round's optimal result, and the next round of inter-module iteration continues.

[0073] It is understood that the system in this embodiment corresponds to the method in Embodiment 1 above, and the options in Embodiment 1 above are also applicable to this embodiment, so they will not be described again here.

[0074] Example 3 This embodiment provides a specific simulation of the signal separation method for unknown interference communication signals proposed in Embodiment 1.

[0075] 1) Simulation of Modulation Recognition Accuracy Performance In the performance simulation experiment of modulation recognition accuracy, the modulation type of the target signal was fixed, and the interference signal used five digital modulation methods: BPSK, QPSK, 16-QAM, OFDM-BPSK, and OFDM-QPSK. The signal-to-interference ratio (SIR) was fixed at 10 dB, and the signal-to-noise ratio (SNR) ranged from {0, 5, 10, 15, 20, 25, 30} dB. Each SNR was repeated 100 times, and the modulation recognition accuracy for interference signals of different modulation types was statistically analyzed.

[0076] like Figure 4 As shown, this represents the modulation recognition accuracy of interference signals with different modulation types under the condition that the target signal is BPSK.

[0077] like Figure 5 As shown, this represents the modulation recognition accuracy of interference signals with different modulation types under the condition that the target signal is QPSK.

[0078] Depend on Figure 4 and Figure 5It can be seen that the modulation recognition accuracy of interference signals increases significantly with the increase of SNR. When the SNR is low (SNR=5, SNR=10dB), this application can still maintain a high recognition accuracy (≥70%), indicating that it has good noise robustness; when the SNR is high (SNR≥20dB), the recognition accuracy of various interference signals is close to 100%, verifying that this application can effectively identify the modulation type of interference signals.

[0079] 2) Simulation of signal separation accuracy performance In the performance simulation experiment of signal separation accuracy, the target signal was selected with BPSK, QPSK and 16QAM modulation types in sequence, corresponding to 3 sets of simulation experiments. Each set was repeated 200 times. In each repeated experiment, the interference signal was randomly selected from BPSK, QPSK, 16QAM, OFDM-BPSK and OFDM-QPSK. The SNR was fixed at 20dB, and the SIR was taken as {-25, -20, -15, -10, -5} dB.

[0080] like Figure 6 As shown, this is a curve of the mean square error (MSE) of various target signals as a function of SIR under random modulation type signal interference conditions in Example 3; like Figure 7 As shown, the curves of bit error rate (BER) of various target signals as a function of SIR under random modulation type signal interference conditions in Example 3 are shown.

[0081] Depend on Figure 6 and Figure 7 It is evident that this application can achieve stable and accurate signal separation under conditions of target signals and interference signals with different modulation types and different SIRs. When the SIR is low (SIR≤-20dB), the interference signal is strong, and the MSE and BER are relatively high; as the SIR increases, the MSE and BER decrease significantly, fully verifying that the proposed method has the ability to effectively separate and extract the target signal under interference conditions of unknown modulation type signals.

[0082] The terminology used in the accompanying drawings is for illustrative purposes only and should not be construed as limiting the scope of this patent. Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for separating unknown interference communication signals, characterized in that, The method includes: Receive a mixed signal containing unknown interference, perform blind synchronization and remodulation on the mixed signal based on the target signal modulation type, and generate an initial estimate of the target signal; The initial estimate of the target signal and the mixed signal after blind synchronization are input into a diffusion model containing multiple modulation type branches. The diffusion model obtains the target signal estimate and interference signal estimate of each branch through multi-path parallel diffusion-type blind separation based on the initial estimate of the mixed signal and the target signal after blind synchronization. Based on the interference signal estimation of each branch, calculate its matching score with the modulation type corresponding to that branch; The optimal branch is determined based on the matching score, and the target signal estimation and interference signal estimation corresponding to the optimal branch are used as the separation result. The diffusion model includes a target signal diffusion model and an interference signal diffusion model; the target signal diffusion model is trained based on pre-constructed target signal training samples; the interference signal diffusion model includes multiple interference signal diffusion models corresponding to the various modulation types. The diffusion model, based on the initial estimation of the mixed signal and the target signal after blind synchronization, obtains the target signal estimation and interference signal estimation for each branch through multi-path parallel diffusion-based blind separation. The steps include: For each branch, the initial estimate of the mixed signal after blind synchronization and the target signal is used as the starting state of the iteration. The diffusion separation is performed iteratively for a preset number of iterations. When the iteration is completed, the target signal estimate and the interference signal estimate of that branch are obtained. The diffusion separation includes the following steps: Randomly sample the first and second time steps; Based on the first time step and the predetermined noise scheduling strategy, the target signal estimate for the current round is denoised to obtain the first denoised signal; Based on the second time step and the predetermined noise scheduling strategy, the interference signal estimate for the current round is denoised to obtain a second denoised signal; Based on the target signal diffusion model, noise in the first noisy signal is predicted and removed, and the target signal estimate is updated. Based on the interference signal diffusion model corresponding to this branch modulation type, the noise in the second noisy signal is predicted and removed, and the interference signal estimate is updated. The expressions for the target signal estimation and interference signal estimation are as follows: in, and They are the first k The branch in the i The target signal estimation and interference signal estimation obtained in the round iteration are as follows: and To utilize the target signal diffusion model respectively and interference signal diffusion model For noisy samples and The prediction results for noise in the data; and These are the target signal estimation and interference signal estimation at the 1st and 2nd respectively. i Score estimation under sub-noise sampling; The synchronized mixed signal; ~ For the first k The branch in the Random noise independently sampled from a standard normal distribution during each iteration; and They are respectively the first time step t Second time step u The corresponding signal attenuation coefficient; For the first k The mixing coefficient of each branch, For the first The learning rate of the round; and The first k The branch in the The noise-added results of target signal estimation and interference signal estimation in the round of iteration.

2. The signal separation method for unknown interference communication signals according to claim 1, characterized in that, The steps for generating the initial estimate of the target signal include: using the blind-synchronized mixed signal as a coarse estimate, performing deconstellation mapping to recover the bit sequence, and remodulating the bit sequence according to the modulation type parameters of the target signal to obtain the initial estimate of the target signal.

3. The signal separation method for unknown interference communication signals according to claim 1, characterized in that, During training, the diffusion model performs forward noise addition and inverse denoising according to the standard diffusion mechanism, and is optimized using a complex IQ signal dataset of the corresponding modulation type. The training objective of the diffusion model is to minimize the noise level at any time step. Below, the predicted noise and the actual injected noise are compared. The mean squared error between them is given by the following loss function: in, It is the first step t The corresponding signal attenuation coefficient.

4. The signal separation method for unknown interference communication signals according to claim 1, characterized in that, The noise addition is based on random noise that follows a standard normal distribution. The expressions for the first and second noise-added signals are as follows: in, and They are the first The branch in the The target signal estimate and interference signal estimate are obtained from rounds of iteration.

5. The signal separation method for unknown interference communication signals according to claim 1, characterized in that, The steps for calculating the matching score include: Randomly sample multiple time steps; The interference signal is estimated to be noisy multiple times according to different time steps to obtain multiple noisy samples; Noise prediction is performed on the noisy sample based on the interference signal diffusion model corresponding to the branch modulation type. Based on the noise prediction error, the matching score is calculated, and the formula for calculating the matching score is as follows: Calculation branch Assuming the modulation type matching score: in, To pass the interference signal diffusion model Noise sampling The noise prediction results and matching scores Indicates interference signal estimation With modulation type The degree of matching; For random sampling points The corresponding signal attenuation coefficient; ~ For the first k The first branch l Random noise from sub-independent sampling L This represents the number of time steps.

6. The signal separation method for unknown interference communication signals according to any one of claims 1 to 5, characterized in that, The method further includes: calculating the signal-to-interference ratio and the mixing coefficient after obtaining the separation results, the calculation formulas are as follows: in, The signal-to-interference ratio of the k-th branch. Let be the mixing coefficient of the k-th branch. For target signal estimation, For interference signal estimation.

7. A signal separation system for unknown interference communication signals, applied to the signal separation method for unknown interference communication signals according to any one of claims 1 to 6, characterized in that, The system includes: The preprocessing module is used to receive the mixed signal and perform preprocessing to obtain an initial estimate of the mixed signal after blind synchronization and the target signal. The multi-modulation diffusion separation module receives the initial estimate of the mixed signal and the target signal after blind synchronization, and performs multi-parallel diffusion-type blind separation based on the diffusion model to obtain the target signal estimate and interference signal estimate for each branch. The parameter estimation module receives interference signal estimates based on each branch, calculates the matching score between the estimated signal and the corresponding candidate modulation type, selects the branch with the highest matching score as the optimal branch, identifies the modulation type of the interference signal, and estimates the mixing coefficients. The output module estimates and outputs the final target signal based on the target signal corresponding to the optimal branch.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the signal separation method for unknown interference communication signals as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Multi-time-frequency overlapped mixed signal blind separation method based on diffusion model

    CN121502317A