Intelligent spectrum resource management method based on blind signal separation and intelligent feature extraction

Abstract

The invention discloses a spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction, and belongs to the technical field of satellite communication. The method comprises the following steps: S1, spectrum data acquisition and preprocessing: acquiring spectrum data through high-precision equipment and performing primary processing; s2, blind signal separation: extracting a pure signal from a complex signal by adopting an advanced technology; s3, intelligent feature extraction: capturing spectrum dynamic features and extracting key features by using deep learning technologies such as LSTM and the like; s4, optimizing a spectrum resource allocation strategy, and dynamically adjusting the allocation strategy based on deep reinforcement learning; s5, performing anomaly detection and classification, introducing a deep learning model to quickly identify an abnormal mode, and enhancing the anti-interference capability; and S6, integrating and testing the system, and verifying the effectiveness of the system through simulation and actual measurement. The spectrum utilization rate and the anti-interference capability of the system are remarkably improved, an efficient solution is provided for spectrum management in a complex communication environment, and the method has important application value.

Description

Technical Field

[0001] The present invention relates to the field of satellite communication technology, and in particular to a spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction. Background Art

[0002] Cognitive radio technology is a technology that improves spectrum utilization and communication system performance by intelligently sensing and utilizing radio spectrum resources. By using software-defined radio equipment, this technology enables dynamic access and flexible allocation of spectrum resources, adapting to changing communication needs and environmental conditions.

[0003] However, existing cognitive radio technologies still have some problems:

[0004] Existing spectrum data preprocessing methods are inefficient and difficult to adapt to complex and changing communication environments, resulting in low spectrum utilization.

[0005] In response to the above problems, the present invention proposes a method based on blind signal separation and intelligent feature extraction combined with deep learning methods to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide a spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction, so as to solve the problem of low efficiency of spectrum data preprocessing in existing cognitive radio technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction, the spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction comprising the following steps:

[0008] S1: Spectrum data acquisition and preprocessing

[0009] Acquire spectrum data using high-precision equipment and perform preliminary processing; the preliminary processing includes spectrum calibration, data alignment, Doppler effect correction, interference detection and suppression, and dynamic range compression to ensure data accuracy and consistency;

[0010] The main filtering process in the interference detection and suppression can be expressed as:

[0011]

[0012] Where x[n] is the original signal, h[n] is the impulse response of the filter, and y[n] is the filtered signal;

[0013] The dynamic range compression uses a logarithmic transformation:

[0014] y=log(1+x)

[0015] Where x is the original signal and y is the compressed signal;

[0016] In satellite communications, since the velocity v of the satellite relative to the ground station is usually very small, the Doppler shift Δf is also very small, and accurate calculation is required to avoid communication errors;

[0017] In order to correct the Doppler effect, the receiving end needs to adjust the receiving frequency according to the calculated Doppler frequency shift. The corrected receiving frequency f received It can be expressed as:

[0018] f received =f0+△f

[0019] Alternatively, if the received frequency and Doppler shift are known, the original transmitted frequency f0 can be calculated:

[0020] f0=f received -△f;

[0021] S2: Blind Signal Separation

[0022] Use advanced algorithms to extract pure signals from complex signals without prior knowledge, improving the accuracy of spectrum perception;

[0023] The advanced algorithms are independent component analysis and adaptive filtering;

[0024] The independent component analysis algorithm specifically comprises the following steps: first, the signal is centered; then, correlations between the signals are eliminated through a whitening step; then, a separation matrix is found using a non-Gaussian maximization criterion, wherein the separation matrix can convert the observed signal into a statistically independent source signal; and finally, the original source signal is restored through an inverse transformation.

[0025] Applying the independent component analysis algorithm in blind signal separation requires solving the following optimization problem, namely, finding a set of weights that maximizes the non-Gaussianity of the observed signal:

[0026]

[0027] Where: s i is the i-th column of the source signal matrix S; E represents the expected value;

[0028] S3: Intelligent Feature Extraction

[0029] A long short-term memory network algorithm is used to capture the dynamic characteristics of the spectrum and extract key features, thereby providing high-quality data support for resource allocation;

[0030] The LSTM algorithm effectively solves the gradient vanishing or gradient exploding problems of traditional recurrent neural networks when processing long sequences through its gating mechanism, enabling the network to capture long-term dependencies in spectral data.

[0031] Key features are extracted through the long short-term memory network algorithm. The core of the long short-term memory network algorithm is the cell state update formula, which determines how information flows in the network and is remembered for a long time. The cell state update formula is:

[0032]

[0033] in:

[0034] C t is the cell state at the current time t; Ct-1 is the cell state at the previous time t-1;

[0035] f t It is the activation value of the forget gate, which determines how much of the cell state at the previous moment is retained to the current moment;

[0036] i t is the activation value of the input gate, which determines how much of the current input is written into the cell state;

[0037] Is the candidate cell state, which is calculated by the current input and the hidden state of the previous moment through the activation function of the candidate cell state;

[0038] ⊙ represents element-level multiplication, i.e. Hadamard product;

[0039] S4: Spectrum resource allocation strategy optimization

[0040] The spectrum resource allocation strategy optimization is achieved by applying the deep reinforcement learning algorithm Q-learning. This algorithm simulates the interaction between the intelligent agent and the environment, learns the optimal actions to take under different states, and dynamically adjusts the spectrum allocation strategy to improve spectrum utilization and achieve intelligent spectrum management.

[0041] S5: Anomaly Detection and Classification Algorithms

[0042] By deploying an autoencoder and isolation forest model based on deep learning, the system can quickly identify abnormal patterns in the spectrum. The autoencoder and isolation forest model algorithms enhance the system's resistance to interference signals and ensure efficient use of spectrum resources.

[0043] S6: System integration and testing

[0044] Design a system, which includes the spectrum data acquisition and preprocessing module, blind signal separation module, intelligent feature extraction module, spectrum resource allocation strategy optimization module, and anomaly detection and classification algorithm module. Verify the effectiveness of the system through simulation and actual measurement to ensure the stability and reliability of the system.

[0045] Preferably, in S1, the spectrum data is updated using environmental parameters and operating parameters to ensure accurate reflection of the current state of the spectrum environment.

[0046] Preferably, in S2, the blind signal separation technology uses the independent component analysis to recover statistically independent source signals from the observed mixed signals, thereby improving the accuracy and efficiency of signal separation.

[0047] Preferably, in S2, the adaptive filtering algorithm specifically comprises: first pre-processing the input signal to ensure that it meets the requirements of the filter processing; then initializing the filter coefficients and setting the performance evaluation criteria; then using the adaptive algorithm and the minimum mean square error algorithm to adjust the filter coefficients in real time to adapt to the signal characteristics, so as to minimize the error between the desired signal and the filter output; finally processing the input signal through the filter to obtain an optimized output signal, and restoring the original source signal through inverse transformation;

[0048] The core formula of the minimum mean square error algorithm is the update rule of the filter coefficients, which describes how to adjust the filter coefficients according to the input signal and the expected output to minimize the error. The following is the core formula of the minimum mean square error algorithm:

[0049] Filter output:

[0050]

[0051] Where y[n] is the output of the filter at time n, w[i] is the filter coefficient, x[ni] is the input signal, and N is the length of the filter;

[0052] Error calculation:

[0053] e[n]=d[n]-y[n]

[0054] Where e[n] is the error signal, d[n] is the desired output, i.e. the target signal, and y[n] is the actual output of the filter;

[0055] Coefficient update:

[0056] w[i]=w[i]+μe[n]x[ni], fori=0,1,...,N-1

[0057] Where w[i] is the updated filter coefficient, μ is the step size, i.e. the learning rate, e[n] is the error signal, and x[ni] is the delayed version of the input signal.

[0058] Preferably, in S3, the intelligent feature extraction process is to integrate a wavelet transform algorithm to extract the periodicity, trend and randomness characteristics of the spectrum data, thereby fully capturing the dynamic change characteristics of the spectrum data.

[0059] Preferably, in S4, the spectrum resource allocation strategy optimization includes a feedback mechanism, which can dynamically adjust and optimize subsequent spectrum allocation decisions based on actual communication performance after spectrum allocation, so as to continuously improve spectrum utilization and system performance.

[0060] Preferably, in S5, the autoencoder is an unsupervised learning algorithm that can learn an effective representation of data and identify anomalies by reconstructing errors; and the isolation forest isolates samples by randomly selecting features and segmentation values, thereby identifying anomalies that are not easily isolated.

[0061] Preferably, in S5, the anomaly detection and classification algorithm includes dynamic spectrum access, dynamically adjusting the spectrum allocation strategy based on real-time information of spectrum holes and interference sources; cognitive radio technology uses spectrum sensing and learning mechanisms to intelligently avoid interference sources and use spectrum holes for communication to quickly adapt to changes in the spectrum environment and ensure the stability and reliability of the communication system.

[0062] Preferably, in S5, in the isolation forest algorithm, the calculation formula of the abnormality score of each tree is expressed as follows:

[0063]

[0064] in:

[0065] Score(x) is the anomaly score of sample x;

[0066] avgPathLength(x) is the average path length of sample x in all trees;

[0067] X is the set of all samples.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] (1) In the present invention, by using advanced blind signal separation technology, pure signals can be extracted from complex signals, significantly improving the accuracy of spectrum perception, reducing misjudgments and missed judgments, and improving the performance and reliability of the communication system.

[0070] (2) In the present invention, the dynamic change characteristics of spectrum data can be automatically captured and key features can be extracted through intelligent feature extraction technology, providing high-quality data support for spectrum resource allocation and improving the intelligence level and efficiency of resource allocation.

[0071] (3) In the present invention, through the optimization of spectrum resource allocation strategy based on deep reinforcement learning, the allocation can be dynamically adjusted according to the historical data and real-time status of spectrum usage.

[0072] (4) In the present invention, by designing and implementing a spectrum data preprocessing and resource allocation method based on blind signal separation and intelligent feature extraction, key technologies such as blind signal separation, intelligent feature extraction, and deep learning are effectively integrated into a complete system to achieve accurate analysis and optimal allocation of spectrum resources, providing an efficient solution for spectrum management in complex communication environments, which has important application value and broad development prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 Schematic diagram of the process of the present invention;

[0074] Figure 2 This is a diagram showing the measured results of the anti-interference performance of the present invention;

[0075] Figure 3 It is the algorithm flow chart of the present invention. DETAILED DESCRIPTION

[0076] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0077] This invention's technology is primarily used in spectrum resource management within integrated air-space-ground communication systems. In this area, efficient spectrum utilization and intelligent allocation are crucial for ensuring communication quality and improving system capacity. Using blind signal analysis (ICA) and intelligent feature extraction techniques, it is possible to preprocess spectrum data in complex communication environments, thereby optimizing spectrum resource allocation.

[0078] Figure 1 It is a schematic flow chart of the steps of the present invention. Figure 3 It is the algorithm flow chart of the present invention.

[0079] Example

[0080] The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction specifically includes the following steps:

[0081] S1: Spectrum data acquisition and preprocessing

[0082] Spectrum data from the integrated air-ground-space communication system is collected using ground base stations and satellite communication equipment. The collected data is preprocessed, including filtering, denoising, and normalization, to improve data quality. The specific steps for spectrum data collection and preprocessing are as follows:

[0083] Spectrum data is acquired through high-precision equipment and preliminarily processed; the preliminarily processed data includes spectrum calibration, data alignment, Doppler effect correction, interference detection and suppression, and dynamic range compression to ensure data accuracy and consistency.

[0084] The main filtering process in interference detection and suppression can be expressed as:

[0085]

[0086] Where x[n] is the original signal, h[n] is the impulse response of the filter, and y[n] is the filtered signal.

[0087] Dynamic range compression uses a logarithmic transformation:

[0088] y=log(1+x)

[0089] Where x is the original signal and y is the compressed signal.

[0090] In satellite communications, since the velocity v of the satellite relative to the ground station is usually very small, the Doppler shift Δf is also very small, but it still needs to be calculated accurately to avoid communication errors.

[0091] In order to correct the Doppler effect, the receiving end needs to adjust the receiving frequency according to the calculated Doppler frequency shift. The corrected receiving frequency f received It can be expressed as:

[0092] f received =f0+△f

[0093] Alternatively, if the received frequency and Doppler shift are known, the original transmitted frequency f0 can be calculated:

[0094] f0=f received -△f;

[0095] In S1, the spectrum data is updated using environmental parameters and operating parameters to ensure that the current state of the spectrum environment is accurately reflected.

[0096] S2: Blind Signal Separation

[0097] Advanced technologies are used to extract pure signals from complex signals, eliminating the need for prior knowledge and improving spectrum sensing accuracy. Advanced algorithms such as independent component analysis (ICA) and adaptive filtering automatically identify and isolate useful signals, effectively reducing human intervention. These technologies adapt to changing signal environments and maintain efficient signal extraction performance in various communication scenarios. Incorporating deep learning models further enhances adaptability to new environments and signal processing accuracy.

[0098] The independent component analysis (ICA) algorithm is applied to perform blind signal separation on the preprocessed spectrum data to separate statistically independent source signals. The specific steps of blind signal separation are as follows.

[0099] In S2, blind signal separation technology is implemented through the independent component analysis (ICA) algorithm, a specially designed algorithm used to accurately separate statistically independent source signals from the received mixed signal, thereby improving the accuracy and efficiency of signal separation. The ICA algorithm improves the accuracy and efficiency of signal separation by identifying and utilizing the statistical independence between signal sources, thereby optimizing the performance of signal processing. The independent component analysis algorithm specifically: first, the signal is centered; then the correlation between signals is eliminated through a whitening step; then, the non-Gaussian maximization criterion is used to find a separation matrix that can convert the observed signal into statistically independent source signals; finally, the original source signal is restored through an inverse transform.

[0100] Applying the independent component analysis algorithm in the blind signal separation requires solving the following optimization problem, namely, finding a set of weights that maximizes the non-Gaussianity of the observed signal:

[0101]

[0102] Where: s i is the i-th column of the source signal matrix S; E represents the expected value.

[0103] The adaptive filtering algorithm specifically involves: first, preprocessing the input signal to ensure that it meets the requirements of filter processing; then initializing the filter coefficients and setting performance evaluation criteria, such as minimum mean square error; then using an adaptive algorithm, the least mean square error (LMS) algorithm, to adjust the filter coefficients in real time to adapt to signal characteristics, so as to minimize the error between the desired signal and the filter output; finally, processing the input signal through the filter to obtain an optimized output signal, and restoring the original source signal through inverse transformation.

[0104] The core formula of the LMS algorithm is the update rule of the filter coefficients, which describes how to adjust the filter coefficients based on the input signal and the desired output to minimize the error. The following is the core formula of the LMS algorithm:

[0105] Filter output:

[0106]

[0107] Where y[n] is the output of the filter at time n, w[i] are the filter coefficients, x[ni] is the input signal, and N is the length of the filter.

[0108] Error calculation:

[0109] e[n]=d[n]-y[n]

[0110] Where e[n] is the error signal, d[n] is the desired output (target signal), and y[n] is the actual output of the filter.

[0111] Coefficient update:

[0112] w[i]=w[i]+μe[n]x[ni],fori=0,1,...,N-1

[0113] Where w[i] is the updated filter coefficient, μ is the step size, i.e. the learning rate, e[n] is the error signal, and x[ni] is the delayed version of the input signal.

[0114] This update rule is the core of the LMS algorithm, which optimizes the performance of the filter by continuously adjusting the filter coefficients to reduce the mean square value of the error signal e[n].

[0115] S3: Intelligent Feature Extraction

[0116] The Long Short-Term Memory (LSTM) algorithm is used to capture the dynamic characteristics of the spectrum and extract key features, thereby providing high-quality data support for resource allocation.

[0117] The LSTM algorithm is a special type of recurrent neural network (RNN) that can learn long-term dependencies, making it ideally suited for processing and predicting important features in time series data. The LSTM algorithm is a core component and crucial for addressing the dynamic management and optimization of spectrum resources. The Long Short-Term Memory (LSTM) algorithm effectively addresses the vanishing or exploding gradient problems of traditional RNNs when processing long sequences through its unique gating mechanism, enabling the network to capture long-term dependencies in spectral data. This unique gating mechanism refers to the specialized structure within the LSTM network that allows the network to learn long-term dependencies. By introducing input, forget, and output gates to control the flow of information, the LSTM effectively addresses the vanishing or exploding gradient problems found in traditional RNNs.

[0118] The key features are extracted through the LSTM algorithm. The core of the algorithm is the cell state update formula, which determines how information flows in the network and is remembered for a long time. The cell state update formula is:

[0119]

[0120] in:

[0121] C t is the cell state at the current time t; Ct-1 is the cell state at the previous time t-1;

[0122] f t It is the activation value of the forget gate, which determines how much of the cell state at the previous moment is retained to the current moment;

[0123] i t is the activation value of the input gate, which determines how much of the current input is written into the cell state;

[0124] Is the candidate cell state, which is calculated by the current input and the hidden state of the previous moment through the activation function of the candidate cell state (usually tanh);

[0125] ⊙ represents element-wise multiplication, i.e., Hadamard product.

[0126] In S3, the intelligent feature extraction process integrates the wavelet transform algorithm to extract the periodicity, trend and randomness characteristics of the spectrum data, thereby fully capturing the dynamic change characteristics of the spectrum data.

[0127] S4: Spectrum resource allocation strategy optimization

[0128] According to the extracted feature vectors, intelligent algorithms (such as genetic algorithms, particle swarm optimization, etc.) are used to dynamically allocate spectrum resources to maximize spectrum utilization and communication quality.

[0129] Dynamically adjust allocation strategies based on deep reinforcement learning to improve spectrum utilization and achieve intelligent spectrum management. Spectrum resource allocation strategy optimization is achieved by applying a specific deep reinforcement learning algorithm, Q-Learning. This algorithm simulates the interaction between an intelligent agent and its environment, learning the optimal actions to take under different conditions. This allows for dynamic adjustment of spectrum allocation strategies to improve spectrum utilization and achieve intelligent spectrum management.

[0130] In S4, spectrum resource allocation strategy optimization includes a feedback mechanism that can dynamically adjust and optimize subsequent spectrum allocation decisions based on the actual communication performance after spectrum allocation, so as to continuously improve spectrum utilization and system performance.

[0131] S5: Anomaly Detection and Classification Algorithms

[0132] By deploying the autoencoder and isolation forest model based on deep learning, the system can quickly identify abnormal patterns in the spectrum; the autoencoder and isolation forest model algorithms enhance the system's resistance to interference signals and ensure the efficient use of spectrum resources.

[0133] An autoencoder is an unsupervised learning algorithm that can learn an effective representation of data and identify anomalies through reconstruction errors. Isolation forest isolates samples by randomly selecting features and split values, thereby identifying anomalies that are difficult to isolate.

[0134] In the isolation forest algorithm, the calculation formula for the anomaly score of each tree is expressed as follows:

[0135]

[0136] in:

[0137] Score(x) is the anomaly score of sample x;

[0138] avgPathLength(x) is the average path length of sample x in all trees;

[0139] X is the set of all samples.

[0140] In S5, the anomaly detection and classification algorithm includes dynamic spectrum access (DSA), which dynamically adjusts the spectrum allocation strategy based on real-time information about spectrum holes and interference sources; cognitive radio technology, which uses spectrum sensing and learning mechanisms to intelligently avoid interference sources and utilize spectrum holes for communication to quickly adapt to changes in the spectrum environment and ensure the stability and reliability of the communication system.

[0141] S6: System integration and testing

[0142] Design an efficient, stable and scalable system, which includes the spectrum data acquisition and preprocessing module, blind signal separation module, intelligent feature extraction module, spectrum resource allocation strategy optimization module, and anomaly detection and classification algorithm module. Verify the effectiveness of the system through simulation and actual measurement to ensure the stability and reliability of the system.

[0143] Effects of application examples

[0144] This technology has achieved remarkable results in the integrated air-ground-space communication system:

[0145] (1) Improved spectrum utilization: Through intelligent feature extraction and resource allocation, spectrum utilization has been increased by more than 20%.

[0146] (2) Improved communication quality: In complex communication environments, communication quality has been significantly improved, with the bit error rate reduced by 15%.

[0147] (3) Increased system capacity: Due to the efficient use of spectrum resources, the system capacity has increased by 30%, which can support more users and higher data transmission rates.

[0148] (4) Improved user satisfaction: Due to the improvement in communication quality and the increase in system capacity, the end-user satisfaction has been significantly improved.

[0149] This invention not only proves the effectiveness of spectrum data preprocessing and resource allocation methods based on blind signal separation and intelligent feature extraction in integrated air-ground-space communication systems, but also provides valuable experience for future technology development and applications. Figure 2 (Figure 2 shows the measured results of the anti-interference performance of the present invention).

[0150] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction, characterized by: The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction comprises the following steps: S1: Spectrum data acquisition and preprocessing Acquire spectrum data using high-precision equipment and perform preliminary processing; the preliminary processing includes spectrum calibration, data alignment, Doppler effect correction, interference detection and suppression, and dynamic range compression to ensure data accuracy and consistency; The main filtering process in the interference detection and suppression can be expressed as: Where x[n] is the original signal, h[n] is the impulse response of the filter, and y[n] is the filtered signal; The dynamic range compression uses a logarithmic transformation: y=log(1+x) Where x is the original signal and y is the compressed signal; In satellite communications, since the velocity v of the satellite relative to the ground station is usually very small, the Doppler shift Δf is also very small, and accurate calculation is required to avoid communication errors; In order to correct the Doppler effect, the receiving end needs to adjust the receiving frequency according to the calculated Doppler frequency shift. The corrected receiving frequency f received It can be expressed as: f received =f0+Δf Alternatively, if the received frequency and Doppler shift are known, the original transmitted frequency f0 can be calculated: f0=f received -Δf; S2: Blind Signal Separation Use advanced algorithms to extract pure signals from complex signals without prior knowledge, improving the accuracy of spectrum perception; The advanced algorithms are independent component analysis and adaptive filtering; The independent component analysis algorithm specifically comprises the following steps: first, the signal is centered; then, correlations between the signals are eliminated through a whitening step; then, a separation matrix is found using a non-Gaussian maximization criterion, wherein the separation matrix can convert the observed signal into a statistically independent source signal; and finally, the original source signal is restored through an inverse transformation. Applying the independent component analysis algorithm in blind signal separation requires solving the following optimization problem, namely, finding a set of weights that maximizes the non-Gaussianity of the observed signal: Where: s i is the i-th column of the source signal matrix S; E represents the expected value; S3: Intelligent Feature Extraction A long short-term memory network algorithm is used to capture the dynamic characteristics of the spectrum and extract key features, thereby providing high-quality data support for resource allocation; The LSTM algorithm effectively solves the gradient vanishing or gradient exploding problems of traditional recurrent neural networks when processing long sequences through its gating mechanism, enabling the network to capture long-term dependencies in spectral data. Key features are extracted through the long short-term memory network algorithm. The core of the long short-term memory network algorithm is the cell state update formula, which determines how information flows in the network and is remembered for a long time. The cell state update formula is: in: C t is the cell state at the current time t; Ct-1 is the cell state at the previous time t-1; f t It is the activation value of the forget gate, which determines how much of the cell state at the previous moment is retained to the current moment; i t is the activation value of the input gate, which determines how much of the current input is written into the cell state; Is the candidate cell state, which is calculated by the current input and the hidden state of the previous moment through the activation function of the candidate cell state; ⊙ represents element-level multiplication, i.e. Hadamard product; S4: Spectrum resource allocation strategy optimization The spectrum resource allocation strategy optimization is achieved by applying the deep reinforcement learning algorithm Q-learning. This algorithm simulates the interaction between the intelligent agent and the environment, learns the optimal actions to take under different states, and dynamically adjusts the spectrum allocation strategy to improve spectrum utilization and achieve intelligent spectrum management. S5: Anomaly Detection and Classification Algorithms By deploying an autoencoder and isolation forest model based on deep learning, the system can quickly identify abnormal patterns in the spectrum. The autoencoder and isolation forest model algorithms enhance the system's resistance to interference signals and ensure efficient use of spectrum resources. S6: System integration and testing Design a system, which includes the spectrum data acquisition and preprocessing module, blind signal separation module, intelligent feature extraction module, spectrum resource allocation strategy optimization module, and anomaly detection and classification algorithm module. Verify the effectiveness of the system through simulation and actual measurement to ensure the stability and reliability of the system.

2. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S1, the spectrum data is updated using environmental parameters and operating parameters to ensure that the current state of the spectrum environment is accurately reflected.

3. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S2, the blind signal separation technology uses the independent component analysis to recover statistically independent source signals from the observed mixed signals, thereby improving the accuracy and efficiency of signal separation.

4. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S2, the adaptive filtering algorithm specifically comprises: first pre-processing the input signal to ensure that it meets the requirements of the filter processing; then initializing the filter coefficients and setting the performance evaluation criteria; then using the adaptive algorithm and the minimum mean square error algorithm to adjust the filter coefficients in real time to adapt to the signal characteristics, so as to minimize the error between the desired signal and the filter output; finally, processing the input signal through the filter to obtain an optimized output signal, and restoring the original source signal through inverse transformation; The core formula of the minimum mean square error algorithm is the update rule of the filter coefficients, which describes how to adjust the filter coefficients according to the input signal and the expected output to minimize the error. The following is the core formula of the minimum mean square error algorithm: Filter output: Where y[n] is the output of the filter at time n, w[i] is the filter coefficient, x[ni] is the input signal, and N is the length of the filter; Error calculation: e[n]=d[n]-y[n] Where e[n] is the error signal, d[n] is the desired output, i.e. the target signal, and y[n] is the actual output of the filter; Coefficient update: w[i]=w[i]+μe[n]x[ni], fori=0, 1,..., N-1 Where w[i] is the updated filter coefficient, μ is the step size, i.e. the learning rate, e[n] is the error signal, and x[ni] is the delayed version of the input signal.

5. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In the S3, the intelligent feature extraction process is to integrate the wavelet transform algorithm to extract the periodicity, trend and randomness characteristics of the spectrum data, so as to fully capture the dynamic change characteristics of the spectrum data.

6. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S4, the spectrum resource allocation strategy optimization includes a feedback mechanism, which can dynamically adjust and optimize subsequent spectrum allocation decisions based on actual communication performance after spectrum allocation, so as to continuously improve spectrum utilization and system performance.

7. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S5, the autoencoder is an unsupervised learning algorithm that can learn an effective representation of data and identify anomalies through reconstruction errors; the isolation forest isolates samples by randomly selecting features and segmentation values, thereby identifying anomalies that are not easily isolated.

8. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S5, the anomaly detection and classification algorithm includes dynamic spectrum access, dynamically adjusting the spectrum allocation strategy based on real-time information of spectrum holes and interference sources; cognitive radio technology uses spectrum sensing and learning mechanisms to intelligently avoid interference sources and use spectrum holes for communication to quickly adapt to changes in the spectrum environment and ensure the stability and reliability of the communication system.

9. The spectrum resource intelligent management method based on blind signal separation and intelligent feature extraction according to claim 1, characterized in that: In S5, in the isolation forest algorithm, the calculation formula for the abnormality score of each tree is expressed as follows: in: Score(x) is the anomaly score of sample x; avgPathLength(x) is the average path length of sample x in all trees; X is the set of all samples.

Images