Satellite communication CDMA spread spectrum signal blind detection method and system based on deep learning

Abstract

The invention relates to a satellite communication CDMA spread spectrum signal blind detection method and system based on deep learning. The method comprises the following steps: preprocessing data; building a deep learning model; extracting features; blind detection algorithm processing; and evaluating and optimizing the performance of the algorithm model. According to the method, effective blind detection of the CDMA spread spectrum signal is realized by aiming at the CDMA spread spectrum signal in satellite communication and utilizing the strong feature extraction and mode recognition capability of the deep learning model, the blind detection accuracy and robustness of the CDMA spread spectrum signal in satellite communication in a complex electromagnetic environment are improved, the performance and stability of a satellite communication system are improved, and the method is suitable for popularization and application. According to the method and the device, the dependence on external information is reduced, powerful support is provided for reliable operation of satellite communication, and the problem of accurately detecting the satellite communication CDMA spread spectrum signal in an actual communication scene in which a user spread spectrum code is unknown or difficult to obtain is solved.

Description

Technical Field

[0001] The present invention belongs to the field of satellite communication technology, and in particular to a satellite communication CDMA spread spectrum signal blind detection method and system based on deep learning. Background Art

[0002] Satellite communication is an important means of modern long-distance communication. CDMA (Code Division Multiple Access) spread spectrum technology has been widely used in satellite communication systems due to its strong anti-interference ability and high spectrum utilization. However, in satellite communication systems, CDMA spread spectrum signals face multiple challenges such as multiple access interference and channel distortion, making signal detection particularly complicated.

[0003] Traditional CDMA spread spectrum signal detection methods rely too much on prior information and synchronization signals, but in satellite communications, the acquisition of this information is often limited by multiple factors such as propagation distance, channel conditions, and interference sources, resulting in reduced detection performance. Especially in complex electromagnetic environments, traditional detection methods often cannot accurately identify CDMA spread spectrum signals, thus affecting the reliability and stability of the communication system.

[0004] To overcome this difficulty, researchers began to explore blind detection technology. Blind detection technology can extract and identify signal features by processing and analyzing the received signal without prior information and synchronization signals. However, the application of blind detection technology in CDMA spread spectrum signals still faces many challenges, such as how to effectively extract signal features and how to accurately identify the spread spectrum codes of different users.

[0005] Therefore, developing an efficient and accurate blind detection technology to realize blind detection of satellite communication CDMA spread spectrum signals is of great significance to improving the performance and stability of satellite communication systems. Summary of the invention

[0006] The purpose of the present invention is to provide a blind detection method and system for satellite communication CDMA spread spectrum signals based on deep learning. For CDMA (code division multiple access) spread spectrum signals in satellite communications, the powerful feature extraction and pattern recognition capabilities of deep learning models are utilized to achieve effective blind detection of CDMA spread spectrum signals, improve the blind detection accuracy and robustness of satellite communication CDMA spread spectrum signals in complex electromagnetic environments, improve the performance and stability of satellite communication systems, reduce dependence on external information, and provide strong support for the reliable operation of satellite communications, so as to solve the problem of accurate detection of satellite communication CDMA spread spectrum signals in actual communication scenarios where user spread spectrum codes are unknown or difficult to obtain.

[0007] The present invention provides a satellite communication CDMA spread spectrum signal blind detection method based on deep learning, comprising the following steps:

[0008] Step 1: preprocess the received satellite signals to provide high-quality training data;

[0009] Step 2: Build a deep learning model based on a multi-scale fusion convolutional neural network with feature extraction and sequence processing capabilities to capture subtle differences and dynamic changes in satellite communication CDMA spread spectrum signals;

[0010] Step 3: Use the deep learning model to extract features from the preprocessed signal and extract a distinguishing feature vector from the signal;

[0011] Step 4, blindly detect the CDMA spread spectrum signal based on the trained deep learning model, including: using a multi-scale fusion convolutional neural network to classify the satellite signal, and identifying different spread spectrum codes by comparing the feature vector with the feature space of the known spread spectrum code, and making a decision on the signal based on the classification result and the preset threshold rule; if the confidence of the classification result is higher than the threshold, it is considered that the same type of spread spectrum code of the corresponding satellite communication CDMA spread spectrum signal is detected, and the separated independent signal is reconstructed according to the estimated spread spectrum code to obtain the original user signal;

[0012] Step 5: Quantitatively analyze the performance of the blind detection method and deep learning model on the test set, evaluate the accuracy and efficiency of the blind detection method and model, and provide feedback on the evaluation results to optimize and improve related steps.

[0013] Furthermore, the step 1 comprises:

[0014] 1) Blind source separation: separate the independent signals of each user from the received multi-user CDMA signal;

[0015] 2) Demodulation: Identify the modulation type, carrier type, carrier frequency and modulation rate based on the signal samples and sampling rate, and restore the modulated signal to the original signal;

[0016] 3) De-noising and filtering: Filter the received satellite communication CDMA spread spectrum signal to remove high-frequency noise and interference in the signal to improve signal quality, and use noise suppression algorithm to perform adaptive filtering to further reduce the noise level in the signal;

[0017] 4) Decoding: Identify the encoding type and restore the encoded signal to the original signal.

[0018] Furthermore, the method for building the deep learning model in step 2 is as follows:

[0019] ①Choose model architecture:

[0020] A multi-scale fusion convolutional neural network is selected as the core architecture of the deep learning model to efficiently process and identify multi-scale features in satellite communication CDMA spread spectrum signals. Multiple convolution paths are set in parallel through the multi-scale fusion convolutional neural network, and each path is configured with convolution kernels of different sizes to fully capture the complex structure of the signal in the time and frequency domains.

[0021] ②Define model parameters:

[0022] To ensure the optimal performance of the multi-scale fusion convolutional neural network, the following key parameters are set:

[0023] Convolution layer parameters: Assign different numbers of convolution kernels to each convolution path and set different sizes to capture different scale features in the signal; at the same time, set appropriate step size and padding for the convolution layer according to signal characteristics and computing resources;

[0024] Pooling layer parameters: Select maximum pooling to further reduce feature dimensions and reduce the amount of calculation; at the same time, set the appropriate window size and step size according to the output feature map size of the convolution layer;

[0025] Feature fusion parameters: Convolution fusion is used to effectively integrate features extracted from different paths. Fusion parameters are adjusted to ensure that the fused feature dimensions match the subsequent full connection.

[0026] Fully connected parameters: According to the classification requirements, set the appropriate number of nodes for the fully connected layer; at the same time, select the ReLU activation parameter to enhance the nonlinear expression ability of the model;

[0027] Optimizer and loss function: Select the Adam optimizer to speed up the model training process and improve the convergence speed. At the same time, according to the task type, select the mean square error loss function to balance the difference between the model output and the true value.

[0028] ③Build the model:

[0029] According to the model parameters defined in step ②, follow the steps below to build the model:

[0030] Set the input layer according to the dimension and format of the signal data;

[0031] According to the defined parameters, multiple convolutional layers and pooling layers are built at a time to extract multi-scale features in the signal; the features extracted by different paths are fused to form a more comprehensive and robust feature representation;

[0032] Add a fully connected layer to map the vector to the output category and set the corresponding activation function;

[0033] Set the model output layer according to the selected loss function and optimizer.

[0034] Furthermore, the feature vector extracted in step 3 includes:

[0035] Power spectral density: used to obtain the frequency distribution information of the signal, reflecting the energy distribution of the signal in the frequency domain, as one of the important features of subsequent blind detection processing;

[0036] Autocorrelation function: reflects the correlation of the signal in the time domain and is used to extract periodic features in the signal;

[0037] Cross-correlation function: used to calculate the correlation between different user signals for multi-user CDMA signals, providing useful information for subsequent blind detection processing;

[0038] 01 conversion probability: used to calculate and analyze the probability distribution changes of 01 conversion under different signal states, integrate with other features, and adjust the detection parameters appropriately to adapt to different signal environments;

[0039] High-order statistics: used to describe the nonlinear characteristics of signals and enhance the robustness of blind detection methods.

[0040] Furthermore, the step 4 also includes:

[0041] The iterative optimization algorithm is used to iteratively update the spreading code until the convergence condition is met.

[0042] Furthermore, the step 5 comprises:

[0043] Select bit error rate as the evaluation metric, use the preprocessed data to construct a test set, and ensure that the test set and the training set have the same distribution and characteristics;

[0044] Run the blind detection method in step 4 on the test set and use the evaluation index to compare with the original data. Recover the received satellite spread spectrum signal based on the detected spread spectrum code and check the data bit error rate.

[0045] Based on the evaluation results, new satellite signal data is continuously collected for updated training to improve the generalization ability of the deep learning model.

[0046] The present invention also provides a satellite communication CDMA spread spectrum signal blind detection system based on deep learning, comprising a blind detection module, which executes the satellite communication CDMA spread spectrum signal blind detection method based on deep learning.

[0047] The present invention also provides a non-transitory computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the method for blind detection of satellite communication CDMA spread spectrum signals based on deep learning is implemented.

[0048] The present invention also provides an electronic device, comprising:

[0049] A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the deep learning-based satellite communication CDMA spread spectrum signal blind detection method by executing the computer instructions.

[0050] Through the above scheme, through the satellite communication CDMA spread spectrum signal blind detection method and system based on deep learning, it is possible to achieve effective blind detection of CDMA spread spectrum signals in an actual communication environment where the user spread spectrum code is unknown or difficult to obtain, and through powerful algorithm iteration and optimization capabilities, it is possible to continuously update the training model, improve the performance and stability of the satellite communication system, reduce dependence on external information, and provide strong support for the reliable operation of satellite communications.

[0051] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a flow chart of a satellite communication CDMA spread spectrum signal blind detection method based on deep learning of the present invention;

[0053] Figure 2 The satellite communication CDMA spread spectrum signal generation principle diagram of the present invention;

[0054] Figure 3 This is a working diagram of signal preprocessing of the present invention;

[0055] Figure 4 Build a flow chart for the deep learning model of the present invention;

[0056] Figure 5 This is a schematic diagram of relevant statistics in one embodiment of the present invention;

[0057] Figure 6 It is a 01 conversion probability statistical diagram in one embodiment of the present invention;

[0058] Figure 7 The figure is a schematic diagram of an electronic device according to the present invention. DETAILED DESCRIPTION

[0059] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0060] Ginseng Figure 1As shown, this embodiment provides a blind detection method for satellite communication CDMA spread spectrum signals based on deep learning. In the case of unknown user spread spectrum codes, the method performs preprocessing, feature extraction and blind detection processing on the received satellite signals to achieve effective detection of CDMA signals. Specifically, the method includes the following steps:

[0061] 1. Data preprocessing:

[0062] Data preprocessing is a necessary step before deep learning is built. According to the principle of satellite signal generation through CDMA spread spectrum ( Figure 2 ), at the receiving end, what is received is a mixed signal of multiple user signals and noise. The preprocessed data directly affects the training effect of the deep learning model and the accuracy of feature extraction. It is necessary to preprocess the data according to relevant principles, convert and clean the original data accordingly, and provide high-quality training data for the model.

[0063] Ginseng Figure 3 As shown in the figure, the data preprocessing process is as follows:

[0064] ① Blind Source Separation: Blind Source Separation (BSS) technology can separate each independent source signal from a mixed signal without knowing the specific parameters of the source signal and the transmission channel. In this embodiment, BSS technology is used to separate the independent signal of each user from the received multi-user CDMA signal.

[0065] ② Demodulation: Based on the signal samples and the sampling rate, the modulation type, carrier type, carrier frequency and modulation rate are identified through an algorithm, and the modulated signal is restored to the original signal.

[0066] ③ Denoising and filtering: Filter the received satellite communication CDMA spread spectrum signal to remove high-frequency noise and interference in the signal, improve the signal quality, use the noise suppression algorithm to perform adaptive filtering, further reduce the noise level in the signal, and provide clear signal input for subsequent processing.

[0067] ④Decoding: Decode through algorithms, identify the encoding type, and restore the encoded signal to the original signal.

[0068] 2. Deep learning model construction:

[0069] Building a deep learning model is the basis for implementing feature extraction and blind detection algorithms. It defines the overall architecture and parameter settings of the model. The performance of the model directly affects the effect of feature extraction and the accuracy of the blind detection algorithm.

[0070] Ginseng Figure 4 As shown in the figure, the process of building a deep learning model is as follows:

[0071] ①Choose model architecture:

[0072] In this embodiment, a multi-scale fusion convolutional neural network is selected as the core architecture of the deep learning model, aiming to efficiently process and identify multi-scale features in satellite communication CDMA spread spectrum signals. MSCNN sets multiple convolution paths in parallel, each path is configured with convolution kernels of different sizes, to fully capture the complex structure of the signal in the time and frequency domain.

[0073] ②Define model parameters:

[0074] To ensure the best performance of MSCNN, set the following key parameters:

[0075] Convolution layer parameters: A different number of convolution kernels are assigned to each convolution path, and different sizes are set to capture different scale features in the signal. At the same time, appropriate step sizes and padding methods are set for the convolution layer based on signal characteristics and computing resources.

[0076] Pooling layer parameters: Select maximum pooling to further reduce the feature density and reduce the amount of calculation; at the same time, set the appropriate window size and step size according to the output feature map size of the convolutional layer.

[0077] Feature fusion parameters: Convolution fusion is used to effectively integrate the features extracted from different paths. By adjusting the fusion parameters, the dimension of the fused features is ensured to match the subsequent full connection.

[0078] Fully connected parameters: According to the classification requirements, the appropriate number of nodes is set for the fully connected layer; at the same time, the ReLU activation parameter is selected to enhance the nonlinear expression ability of the model. The mathematical expression of the ReLU activation parameter is:

[0079] F(x) = max(0, x)

[0080] Optimizer and loss function: Select the Adam optimizer to speed up the model training process and improve the convergence speed; at the same time, according to the task type, select the mean square error loss function to balance the difference between the model output and the true value.

[0081] ③Build the model:

[0082] According to the above parameters, follow the steps below to build the model:

[0083] Set the input layer according to the dimension and format of the signal data; build multiple convolutional layers and pooling layers at a time according to the defined parameters to extract multi-scale features in the signal; fuse the features extracted by different paths to form a more comprehensive and robust feature representation; then add a fully connected layer to map the vectors to the output categories and set the corresponding activation function; finally, set the model output layer according to the selected loss function and optimizer.

[0084] 3. Feature extraction:

[0085] Feature extraction is the core function of the deep learning model. By combing the input data layer by layer, the deep learning model is used to extract features from the preprocessed signal, aiming to extract distinguishing feature vectors from the signal. When the model is applied, the trained deep learning model is used to extract features from the new signal, and the extracted features such as power spectrum density are used as model input. These features reflect the information of the signal in the time domain, frequency domain and statistical characteristics, which is helpful for model training. The specific feature extraction parameters are as follows:

[0086] ① Power spectral density: Calculate the power spectral density of the signal to obtain the frequency distribution information of the signal, which reflects the energy distribution of the signal in the frequency domain, and serves as one of the important features of subsequent blind detection processing.

[0087] ②Autocorrelation function: Calculate the autocorrelation function of the signal to reflect the correlation of the signal in the time domain, and use it to extract the periodic features in the signal. The autocorrelation function has a good reflection of the characteristics of the pseudo-random data sequence, and can be used to detect and obtain the m sequence. The full permutation distance statistics can better reflect the implicit periodicity of the original sequence, and the statistical results are less affected by bit errors. It can better determine the block length of the block code and cyclic code, and the frame length of the data frame. The relevant statistical diagram is as follows Figure 5 shown.

[0088] ③ Cross-correlation function: For multi-user CDMA signals, the correlation between different user signals is calculated to provide useful information for subsequent blind detection processing.

[0089] ④01 conversion probability: Statistical 01 conversion probability, by calculating and analyzing the probability distribution changes of 01 conversion under different signal states, and integrating with other features, the detection parameters can be appropriately adjusted to adapt to different signal environments. The 01 conversion probability statistics are shown in the figure below. Figure 6 shown.

[0090] ⑤High-order statistics: Extract high-order statistics of the signal, such as kurtosis and skewness, to describe the nonlinear characteristics of the signal and enhance the robustness of the blind detection algorithm.

[0091] 4. Blind detection algorithm processing:

[0092] According to the trained model, the classification ability of the deep learning model is used to identify the signal. The model uses a multi-scale fusion convolutional neural network (MSCNN) to classify and identify satellite signals. The model is trained with a training data set, and recognizes different spread spectrum codes by comparing the feature vector with the feature space of the known spread spectrum code. According to the classification result, combined with the preset threshold rule, the signal is decided. If the confidence of the classification result is higher than the threshold, it is considered that the same type of spread spectrum code corresponding to the satellite communication CDMA spread spectrum signal is detected. According to the estimated spread spectrum code, the separated independent signal is reconstructed to obtain the original user signal. At the same time, in order to improve the accuracy of spread spectrum code estimation and the quality of signal reconstruction, an iterative optimization algorithm is used to iteratively update the spread spectrum code until the convergence condition is met.

[0093] 5. Algorithm model performance evaluation and optimization:

[0094] The algorithm model performance evaluation evaluates the accuracy and efficiency of the algorithm and model by quantitatively analyzing the performance of the blind detection algorithm and model on the test set; the evaluation results are fed back to the construction of the deep learning model, feature extraction and the implementation of the blind detection algorithm to optimize and improve each link.

[0095] Specifically, we first select quantitative information such as bit error rate and F1 score as evaluation indicators, and then use the preprocessed data to build a test set to ensure that the test set and training set have the same distribution and characteristics; run the blind detection algorithm on the test set, and use the evaluation indicators to compare with the original data, restore the received satellite spread spectrum signal according to the detected spread spectrum code, and verify the quantitative information such as the bit error rate of the data; based on the evaluation results, we continuously collect new satellite signal data for updated training to improve the generalization ability of the deep learning model.

[0096] The present invention has the following technical effects:

[0097] (1) The present invention optimizes the decoding, demodulation, denoising and filtering of the received satellite signals in data preprocessing, and adopts an algorithm to improve the accuracy of the overall data preprocessing and improve the signal quality.

[0098] (2) Aiming at the characteristics of satellite communication CDMA spread spectrum signals, the present invention selects a suitable deep learning model, adopts a multi-scale fusion convolutional neural network (MSCNN), and uses multiple parallel convolution paths of MDCNN, each path is configured with convolution kernels of different sizes to fully capture the multi-scale features in satellite communication CDMA spread spectrum signals. The model has powerful feature extraction and sequence processing capabilities, can capture subtle differences and dynamic changes in signals, effectively solves the limitations of traditional methods in signal feature extraction, and significantly improves the accuracy and robustness of blind detection.

[0099] (3) The present invention utilizes the automatic feature extraction capability of the deep learning model to extract distinguishing features from satellite communication CDMA spread spectrum signals, optimizes specific statistical feature algorithms, such as correlation statistics and 01 transition probability statistics, and improves detection performance by fusing with other features. At the same time, the model continuously optimizes and adjusts the parameters of feature extraction through an adaptive learning mechanism to improve the recognition accuracy of the signal.

[0100] (4) In the absence of prior information and synchronization signals, the present invention proposes a blind detection algorithm based on deep learning. By constructing a signal feature space and utilizing the classification capability of a deep learning model, the algorithm can automatically learn and identify the complex features of the signal to achieve recognition of different spread spectrum codes, which not only simplifies the detection process but also greatly improves the detection efficiency and accuracy.

[0101] (5) The present invention has powerful algorithm iteration and optimization capabilities, and can continuously collect new signal data for model update training. This capability ensures that the algorithm can continue to adapt to the rapid development and changes in satellite communication technology.

[0102] This embodiment also provides a satellite communication CDMA spread spectrum signal blind detection system based on deep learning, including a blind detection module, which executes the satellite communication CDMA spread spectrum signal blind detection method based on deep learning.

[0103] This embodiment also provides a non-transitory computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the method for blind detection of satellite communication CDMA spread spectrum signals based on deep learning is implemented.

[0104] Ginseng Figure 7 As shown, this embodiment also provides an electronic device, including:

[0105] A memory 201 and a processor 202, wherein the memory 201 and the processor 202 are communicatively connected to each other, the memory 201 stores computer instructions, and the processor 202 executes the computer instructions to execute the blind detection method for satellite communication CDMA spread spectrum signals based on deep learning.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A satellite communication CDMA spread spectrum signal blind detection method based on deep learning, characterized in that: The steps include: Step 1: preprocess the received satellite signals to provide high-quality training data; Step 2: Build a deep learning model based on a multi-scale fusion convolutional neural network with feature extraction and sequence processing capabilities to capture subtle differences and dynamic changes in satellite communication CDMA spread spectrum signals; Step 3: Use the deep learning model to extract features from the preprocessed signal and extract a distinguishing feature vector from the signal; Step 4, blindly detect the CDMA spread spectrum signal based on the trained deep learning model, including: using a multi-scale fusion convolutional neural network to classify the satellite signal, and identifying different spread spectrum codes by comparing the feature vector with the feature space of the known spread spectrum code, and making a decision on the signal based on the classification result and the preset threshold rule; if the confidence of the classification result is higher than the threshold, it is considered that the same type of spread spectrum code of the corresponding satellite communication CDMA spread spectrum signal is detected, and the separated independent signal is reconstructed according to the estimated spread spectrum code to obtain the original user signal; Step 5: Quantitatively analyze the performance of the blind detection method and deep learning model on the test set, evaluate the accuracy and efficiency of the blind detection method and model, and provide feedback on the evaluation results to optimize and improve related steps.

2. The satellite communication CDMA spread spectrum signal blind detection method based on deep learning according to claim 1 is characterized in that: The step 1 comprises: 1) Blind source separation: separate the independent signals of each user from the received multi-user CDMA signal; 2) Demodulation: Identify the modulation type, carrier type, carrier frequency and modulation rate based on the signal samples and sampling rate, and restore the modulated signal to the original signal; 3) De-noising and filtering: Filter the received satellite communication CDMA spread spectrum signal to remove high-frequency noise and interference in the signal to improve signal quality, and use noise suppression algorithm to perform adaptive filtering to further reduce the noise level in the signal; 4) Decoding: Identify the encoding type and restore the encoded signal to the original signal.

3. The satellite communication CDMA spread spectrum signal blind detection method based on deep learning according to claim 2 is characterized in that: The method for building the deep learning model described in step 2 is as follows: ①Choose model architecture: A multi-scale fusion convolutional neural network is selected as the core architecture of the deep learning model to efficiently process and identify multi-scale features in satellite communication CDMA spread spectrum signals. Multiple convolution paths are set in parallel through the multi-scale fusion convolutional neural network, and each path is configured with convolution kernels of different sizes to fully capture the complex structure of the signal in the time and frequency domains. ②Define model parameters: To ensure the optimal performance of the multi-scale fusion convolutional neural network, the following key parameters are set: Convolution layer parameters: assign different numbers of convolution kernels to each convolution path and set different sizes to capture different scale features in the signal; At the same time, according to the signal characteristics and computing resources, set the appropriate step size and padding method for the convolution layer; Pooling layer parameters: Select maximum pooling to further reduce feature dimensions and reduce the amount of calculation; at the same time, set the appropriate window size and step size according to the output feature map size of the convolution layer; Feature fusion parameters: Convolution fusion is used to effectively integrate features extracted from different paths. Fusion parameters are adjusted to ensure that the fused feature dimensions match the subsequent full connection. Fully connected parameters: According to the classification requirements, set the appropriate number of nodes for the fully connected layer; at the same time, select the ReLU activation parameter to enhance the nonlinear expression ability of the model; Optimizer and loss function: Select the Adam optimizer to speed up the model training process and improve the convergence speed. At the same time, according to the task type, select the mean square error loss function to balance the difference between the model output and the true value. ③Build the model: According to the model parameters defined in step ②, follow the steps below to build the model: Set the input layer according to the dimension and format of the signal data; According to the defined parameters, multiple convolutional layers and pooling layers are built at a time to extract multi-scale features in the signal; The features extracted from different paths are integrated to form a more comprehensive and robust feature representation; Add a fully connected layer to map the vector to the output category and set the corresponding activation function; Set the model output layer according to the selected loss function and optimizer.

4. The satellite communication CDMA spread spectrum signal blind detection method based on deep learning according to claim 3 is characterized in that: The feature vectors extracted in step 3 include: Power spectral density: used to obtain the frequency distribution information of the signal, reflecting the energy distribution of the signal in the frequency domain, as one of the important features of subsequent blind detection processing; Autocorrelation function: reflects the correlation of the signal in the time domain and is used to extract periodic features in the signal; Cross-correlation function: used to calculate the correlation between different user signals for multi-user CDMA signals, providing useful information for subsequent blind detection processing; 01 conversion probability: used to calculate and analyze the probability distribution changes of 01 conversion under different signal states, integrate with other features, and adjust the detection parameters appropriately to adapt to different signal environments; High-order statistics: used to describe the nonlinear characteristics of signals and enhance the robustness of blind detection methods.

5. The satellite communication CDMA spread spectrum signal blind detection method based on deep learning according to claim 4 is characterized in that: The step 4 also includes: The iterative optimization algorithm is used to iteratively update the spreading code until the convergence condition is met.

6. The satellite communication CDMA spread spectrum signal blind detection method based on deep learning according to claim 5 is characterized in that: The step 5 comprises: Select bit error rate as the evaluation metric, use the preprocessed data to construct a test set, and ensure that the test set and the training set have the same distribution and characteristics; Run the blind detection method in step 4 on the test set and use the evaluation index to compare with the original data. Recover the received satellite spread spectrum signal based on the detected spread spectrum code and check the data bit error rate. Based on the evaluation results, new satellite signal data is continuously collected for updated training to improve the generalization ability of the deep learning model.

7. A satellite communication CDMA spread spectrum signal blind detection system based on deep learning, characterized in that: It includes a blind detection module, which executes a satellite communication CDMA spread spectrum signal blind detection method based on deep learning as described in any one of claims 1-6.

8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, which, when executed by a processor, implement a satellite communication CDMA spread spectrum signal blind detection method based on deep learning as described in any one of claims 1 to 6.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the deep learning-based blind detection method for satellite communication CDMA spread spectrum signals as described in any one of claims 1 to 6 by executing the computer instructions.

Images