Frequency spectrum state prediction method based on mixed deep learning model

Abstract

The invention belongs to the technical field of frequency spectrum prediction, and particularly relates to a frequency spectrum state prediction method based on a hybrid deep learning model, the hybrid deep learning model is fused with a long short-term memory (LSTM) network and a multi-layer perceptron (MLP), and through three-dimensional frequency spectrum data sensing, self-adaptive dual-threshold energy detection and hybrid model prediction, the frequency spectrum state is predicted. And the accuracy of idle channel spectrum prediction is further improved. According to the method, the secondary user (SU) in the cognitive radio system (CRS) can quickly select the channel with the highest idle probability for access, the repeated sensing frequency is reduced, the total sensing time is reduced by 30%, and the effective data transmission time is improved by 30%. Compared with the prior art, the method provided by the invention is higher in frequency spectrum state prediction precision in a low signal-to-noise ratio (SNR) scene, the throughput of the system is remarkably improved, and the energy consumption of the system is lower.

Description

Technical Field

[0001] This invention belongs to the field of spectrum prediction technology, specifically relating to a spectrum state prediction method based on a hybrid deep learning model. Background Technology

[0002] With the rapid growth in demand for next-generation wireless applications such as large-scale mobile internet and the Internet of Things (IoT), efficient spectrum management is crucial for addressing limited spectrum resources. Cognitive Radio Systems (CRS) are considered one of the most promising solutions to the spectrum shortage problem. They acquire unused frequency bands and idle channels from primary users (PUs) through spectrum sensing and prediction, and dynamically allocate them to other users based on fundamental factors such as location, time, and frequency band. However, if a channel is actually occupied but predicted to be idle (missed detection), a user accessing that channel will interfere with the PU. Conversely, if a channel is actually idle but predicted to be occupied (false alarm), the user will lose the opportunity to access that channel. Therefore, higher spectrum prediction accuracy leads to higher system throughput and spectrum utilization.

[0003] Traditional fixed-threshold energy detection (ED) suffers from increased false alarm and false negative probabilities due to noise uncertainty under low SNR conditions. On one hand, a single LSTM model struggles to capture the nonlinear characteristics of spectral data, and MLPs alone cannot handle long-term temporal dependencies, leading to increased prediction errors in multi-channel switching scenarios. On the other hand, most existing hybrid models do not incorporate adaptive detection thresholds or consider spatiotemporal correlations, resulting in insufficient generalization capabilities in complex electromagnetic environments. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a spectrum state prediction method based on a hybrid deep learning model, which solves the problems of high computational complexity and low prediction accuracy in long-term spectrum prediction in existing technologies.

[0005] The spectrum state prediction method based on a hybrid deep learning model of the present invention includes the following steps: S1, Spectrum Sensing Phase: Secondary users (SU) collect three-dimensional spectrum data from primary users (PU) in the time, frequency, and spatial domains; S2, Spectrum Detection Stage: An adaptive dynamic adjustment dual-threshold energy detection method is adopted, including: S2.1 Clean, format and normalize the received spectrum data, and perform Fast Fourier Transform (FFT). S2.2 Calculate the weighting coefficients of the dual thresholds dynamically based on the signal-to-noise ratio (SNR) to adjust the sensitivity of the detection threshold; S2.3. Based on historical channel state information, determine the current channel occupancy status. The expression for the channel state detection result is as follows: ; in, This indicates that the primary user does not currently exist and the channel is idle. This indicates that the primary user is currently present, and the channel is in use. For secondary users The power of the primary user signal received at any given time, For the main users Transmitted signal power at any given time For secondary users The power of the noise signal received at all times, Channel gain; S3. Spectrum Prediction Stage: A hybrid deep learning model is used for spectrum state prediction. The hybrid deep learning model includes a spectrum data sequence input layer, an LSTM-MLP model hybrid layer, and a binary prediction result output layer. The input layer data includes historical spectrum state training datasets and historical spectrum state test datasets. The LSTM-MLP model hybrid layer consists of Long Short-Term Memory (LSTM) units and Multilayer Perceptron (MLP) units. LSTM processes time series data and captures long-term dependencies. MLP performs feature extraction and classification to improve prediction accuracy. The binary prediction result output layer outputs and saves the spectrum state prediction result binary file.

[0006] The dual thresholds in S2.2 include and Its expression is: ; in, and Here, T represents the weighting coefficient, and T represents the base threshold. and The calculation formula is: ; in, The nonlinear mapping function is based on the signal-to-noise ratio (SNR). When the SNR is low, the signal is easily overwhelmed by noise. To avoid false alarms (PUs), the false alarm probability needs to be reduced, i.e., the signal density needs to be decreased. To make the low threshold Higher weighting and more conservative dual thresholds reduce noise-induced false triggering; at high SNR, signal energy dominates, so to avoid missed detections of PUs, the detection probability needs to be increased, i.e., increased... , making the high threshold Higher weighting, more sensitive dual thresholds, and faster signal capture; T is calculated according to the Neyman-Pearson theorem, as shown in the following formula: ; in, It is the noise variance. It is the probability of a false alarm.

[0007] The determination method in S2.3 is as follows: S2.3.1. The average energy is obtained by accumulating and averaging the energy spectra of a specific time slot. ,in, ; This represents the number of sample points. S2.3.2 Determine if it is the initial time. If so, since the channel state is unknown, calculate the average value of the threshold. If yes, proceed to S2.3.3; otherwise, proceed to S2.3.4. S2.3.3, each value With average threshold Compare, if This indicates that the current channel state is occupied, denoted as . Otherwise, the current channel state is idle, denoted as... The result is written to the channel state binary file, and then the loop returns to S2.1 to continue. S2.3.4, If the channel state at the previous moment... If yes, proceed to S2.3.5; otherwise, proceed to S2.3.6. S2.3.5, if ,but ,otherwise The result is written to the channel state binary file, and then S2.1 is returned. S2.3.6, if ,but ,otherwise The results are written to the channel state binary file, and then S2.1 is returned, looping until the spectrum sensing task ends.

[0008] The spectrum data in the S3 dataset is collected in a unified standard data format to capture the three-dimensional electromagnetic space environment in the time domain, frequency domain, and spatial domain.

[0009] This invention also provides a spectrum state prediction system based on a hybrid deep learning model, comprising: Spectrum sensing unit: used to acquire three-dimensional spectrum data; Spectrum detection unit: used to perform adaptive dual-threshold energy detection and generate channel state observations; Prediction model unit: contains an LSTM-MLP hybrid model for predicting the future state of the channel.

[0010] The beneficial effects of this invention are: Compared with existing technologies, this invention combines spectrum sensing and spectrum prediction for wireless channel state estimation, simultaneously considering the correlation of three-dimensional data in the time, frequency, and spatial domains, thus improving adaptability to multipath fading environments. It dynamically adjusts the detection threshold weights based on SNR to address noise uncertainty issues under low SNR conditions. Furthermore, it uses spectrum sensing results to apply to a deep learning model to predict future time-slot channel states. By improving the prediction accuracy of idle channels, it reduces the total spectrum sensing time, ensuring that the SU can utilize more time slots for data transmission and avoiding interference with the PU. Compared with traditional spectrum prediction techniques, this invention demonstrates superior prediction accuracy. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments are briefly introduced below.

[0012] Figure 1 This is a schematic diagram of a spectrum state prediction method based on a hybrid deep learning model provided in Example 1.

[0013] Figure 2 This is a schematic diagram of adaptive dynamic modulation dual threshold provided in Example 1.

[0014] Figure 3 This is a schematic diagram of the adaptive dynamic adjustment dual-threshold spectrum detection method based on energy detection provided in Example 1.

[0015] Figure 4 This is a schematic diagram of the spectrum sensing and prediction transmission process of the improved cognitive radio system provided in Example 1.

[0016] Figure 5 This is a schematic diagram of the time-frequency-space three-dimensional spectrum occupancy status provided in Example 1. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. In addition, the technical features involved in the various embodiments described below can be combined with each other as long as they do not conflict with each other.

[0018] Example 1

[0019] See Figure 1 As shown, the spectral state prediction method based on a hybrid deep learning model includes the following steps: S1. Spectrum Sensing Phase: Secondary users (SU) collect three-dimensional spectrum data from primary users (PU) in the time, frequency, and spatial domains; specifically including but not limited to received signal power, noise power, and signal-to-noise ratio (SNR).

[0020] S2, Spectrum Detection Stage: (e.g., ...) Figure 2 and Figure 3 As shown, the energy detection method employing adaptive dynamic adjustment of dual thresholds includes: S2.1 Clean, format and normalize the received spectrum data, and perform Fast Fourier Transform (FFT). S2.2 Calculate the weighting coefficients of the dual thresholds dynamically based on the signal-to-noise ratio (SNR) to adjust the sensitivity of the detection thresholds; the dual thresholds include... and Its expression is: ; in, and Here, T represents the weighting coefficient, and T represents the base threshold. and The calculation formula is: ; in, The nonlinear mapping function is based on the signal-to-noise ratio (SNR). When the SNR is low, the signal is easily overwhelmed by noise. To avoid false alarms (PUs), the false alarm probability needs to be reduced, i.e., the signal density needs to be decreased. To make the low threshold Higher weighting and more conservative dual thresholds reduce noise-induced false triggering; at high SNR, signal energy dominates, so to avoid missed detections of PUs, the detection probability needs to be increased, i.e., increased... , making the high threshold Higher weighting, more sensitive dual thresholds, and faster signal capture; T is calculated according to the Neyman-Pearson theorem, as shown in the following formula: ; in, It is the noise variance. It is the probability of a false alarm.

[0021] S2.3. Based on historical channel state information, determine the current channel occupancy status. The expression for the channel state detection result is as follows: ; in, This indicates that the primary user does not currently exist and the channel is idle. This indicates that the primary user is currently present, and the channel is in use. For secondary users The power of the primary user signal received at any given time, For the main users Transmitted signal power at any given time For secondary users The power of the noise signal received at all times, The channel gain is used for determining the channel gain. The determination method in S2.3 is as follows: S2.3.1. The average energy is obtained by accumulating and averaging the energy spectra of a specific time slot. ,in, ; This represents the number of sample points. S2.3.2 Determine if it is the initial time. If so, since the channel state is unknown, calculate the average value of the threshold. If yes, proceed to S2.3.3; otherwise, proceed to S2.3.4. S2.3.3, each value With average threshold Compare, if This indicates that the current channel state is occupied, denoted as . Otherwise, the current channel state is idle, denoted as... The result is written to the channel state binary file, and then the loop returns to S2.1 to continue. S2.3.4, If the channel state at the previous moment... If yes, proceed to S2.3.5; otherwise, proceed to S2.3.6. S2.3.5, if ,but ,otherwise The result is written to the channel state binary file, and then S2.1 is returned. S2.3.6, if ,but ,otherwise The results are written to the channel state binary file, and then S2.1 is returned, looping until the spectrum sensing task ends.

[0022] The spectrum data in the S3 dataset is collected in a unified standard data format to capture the three-dimensional electromagnetic space environment in the time domain, frequency domain, and spatial domain.

[0023] S3. Spectrum Prediction Stage: A hybrid deep learning model is used for spectrum state prediction. The hybrid deep learning model includes a spectrum data sequence input layer, an LSTM-MLP model hybrid layer, and a binary prediction result output layer. The input layer data includes historical spectrum state training datasets and historical spectrum state test datasets. The LSTM-MLP model hybrid layer consists of Long Short-Term Memory (LSTM) units and Multilayer Perceptron (MLP) units. LSTM processes time series data and captures long-term dependencies. MLP performs feature extraction and classification to improve prediction accuracy. The binary prediction result output layer outputs and saves the spectrum state prediction result binary file.

[0024] The specific prediction process is as follows: 1. Secondary users (SUs) acquire historical channel state information of primary users (PUs) through spectrum sensing, including signal power, noise power, signal-to-noise ratio, etc., to form three-dimensional spectrum data information; 2. Input this data into a hybrid deep learning model. In the LSTM layer, set an appropriate number of cells based on the time step of the input sequence to process the time series features of the spectral data and extract long-term dependencies. The output of the LSTM serves as the input to the MLP, which uses this data to further extract features and perform classification, improving prediction accuracy. 3. The model outputs spectrum state prediction results, that is, it predicts whether the future time slot channel will be idle or occupied. Based on the prediction results, the SU can select the channel most likely to be idle for rapid access and data transmission, thereby increasing throughput and reducing perceived energy consumption.

[0025] Improved cognitive radio system spectrum sensing and prediction transmission process, such as Figure 3 As shown. This process includes three stages: spectrum sensing, spectrum prediction, and spectrum decision-making. In the spectrum sensing stage, the SU performs spectrum sensing on the PU to acquire three-dimensional spectrum data information, such as... Figure 4 As shown, a [T×F×S] three-dimensional training dataset is constructed by fusing time-domain sequences, frequency-domain channel features, and spatial geographic features. An adaptive dual-threshold spectrum detection method is used to detect channel occupancy status and generate observations. In the spectrum prediction stage, the observations are input into a hybrid deep learning model to predict the channel state and obtain the prediction results. In the spectrum decision stage, based on the output of the spectrum prediction unit and by using detection probabilities, the channel occupancy status is re-evaluated. This iterative update further improves the accuracy of spectrum state prediction, ultimately determining the channel occupancy status to guide the SU (Suite Array) in channel access and data transmission.

[0026] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not describe all details exhaustively, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification.

Claims

1. A method for predicting spectral states based on a hybrid deep learning model, characterized in that, Includes the following steps: S1, Spectrum Sensing Phase: Secondary users (SU) collect three-dimensional spectrum data from primary users (PU) in the time, frequency, and spatial domains; S2, Spectrum Detection Stage: An adaptive dynamic adjustment dual-threshold energy detection method is adopted, including: S2.1 Clean, format and normalize the received spectrum data, and perform Fast Fourier Transform (FFT). S2.2 Calculate the weighting coefficients of the dual thresholds dynamically based on the signal-to-noise ratio (SNR) to adjust the sensitivity of the detection threshold; S2.

3. Based on historical channel state information, determine the current channel occupancy status. The expression for the channel state detection result is as follows: ； in, This indicates that the primary user does not currently exist and the channel is idle. This indicates that the primary user is currently present, and the channel is in use. For secondary users The power of the primary user signal received at any given time, For the main users Transmitted signal power at any given time For secondary users The power of the noise signal received at all times, Channel gain; S3. Spectrum Prediction Stage: A hybrid deep learning model is used for spectrum state prediction. The hybrid deep learning model includes a spectrum data sequence input layer, an LSTM-MLP model hybrid layer, and a binary prediction result output layer. The input layer data includes historical spectrum state training datasets and historical spectrum state test datasets. The LSTM-MLP model hybrid layer consists of Long Short-Term Memory (LSTM) units and Multilayer Perceptron (MLP) units. LSTM processes time series data and captures long-term dependencies. MLP performs feature extraction and classification to improve prediction accuracy. The binary prediction result output layer outputs and saves the spectrum state prediction result binary file.

2. The spectral state prediction method based on a hybrid deep learning model according to claim 1, characterized in that, The dual thresholds in S2.2 include and Its expression is: ； in, and Here, T represents the weighting coefficient, and T represents the base threshold. and The calculation formula is: ； in, The nonlinear mapping function is based on the signal-to-noise ratio (SNR). When the SNR is low, the signal is easily overwhelmed by noise. To avoid false alarms (PUs), the false alarm probability needs to be reduced, i.e., the signal density needs to be decreased. To make the low threshold Higher weighting and more conservative dual thresholds reduce noise-induced false triggering; at high SNR, signal energy dominates, so to avoid missed detections of PUs, the detection probability needs to be increased, i.e., increased... , making the high threshold Higher weighting, more sensitive dual thresholds, and faster signal capture; T is calculated according to the Neyman-Pearson theorem, as shown in the following formula: ； in, It is the noise variance. It is the probability of a false alarm.

3. The spectral state prediction method based on a hybrid deep learning model according to claim 2, characterized in that, The determination method in S2.3 is as follows: S2.3.

1. The average energy is obtained by accumulating and averaging the energy spectra of a specific time slot. ,in, ； This represents the number of sample points. S2.3.2 Determine if it is the initial time. If so, since the channel state is unknown, calculate the average value of the threshold. If yes, proceed to S2.3.3; otherwise, proceed to S2.3.

4. S2.3.3, each value With average threshold Compare, if This indicates that the current channel state is occupied, denoted as . Otherwise, the current channel state is idle, denoted as... The result is written to the channel state binary file, and then the loop returns to S2.1 to continue. S2.3.4, If the channel state at the previous moment... If yes, proceed to S2.3.5; otherwise, proceed to S2.3.

6. S2.3.5, if ,but ,otherwise The result is written to the channel state binary file, and then S2.1 is returned. S2.3.6, if ,but ,otherwise The results are written to the channel state binary file, and then S2.1 is returned, looping until the spectrum sensing task ends.

4. The spectral state prediction method based on a hybrid deep learning model according to claim 3, characterized in that, The spectrum data in the S3 dataset is collected in a unified standard data format to capture the three-dimensional electromagnetic space environment in the time domain, frequency domain, and spatial domain.

5. A spectrum state prediction system based on a hybrid deep learning model, characterized in that, include: Spectrum sensing unit: used to acquire three-dimensional spectrum data; Spectrum detection unit: used to perform adaptive dual-threshold energy detection and generate channel state observations; Prediction model unit: contains an LSTM-MLP hybrid model for predicting the future state of the channel.

Images