Radio signal classification method based on random forest
Through the random forest-based radio signal classification method, the problem of limited computing resources in edge devices is solved, and radio signal classification with high accuracy and low resource requirements is achieved, which is suitable for intelligent radio signal processing.
Patent Information
- Application Number
- CN202411116448.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-08
AI Technical Summary
There is a contradiction between the accuracy and computing resources of the existing radio signal classification method. Although the deep learning method has high accuracy, it requires a large number of computing and storage resources. The accuracy of spectrum analysis is low, making it difficult to meet the edge devices with limited computing resources. need.
A random forest-based radio signal classification method is adopted, and a random forest integrated neural network of radio signals is designed by determining the number of intervals of the histogram, the number and depth of the random forest decision tree, and the random forest integrated neural network is designed to conduct offline training and realize online classification.
It realizes high-accurate radio signal classification on edge devices with limited resources, reduces the computing resource requirements, and is suitable for intelligent radio signal processing.
Smart Images

Figure CN118916795A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of intelligent processing of radio signals, and in particular relates to a radio signal classification method based on random forest. Background Art
[0002] In fields such as radio spectrum monitoring, it is usually necessary to classify the collected radio signals, identify and analyze the content, and then take applications such as abnormal signal location or behavior analysis.
[0003] There are currently a variety of radio signal classification methods, including spectrum analysis and deep learning. Spectrum analysis requires fewer computing resources, but has lower accuracy; deep learning methods usually have higher accuracy due to their powerful learning capabilities, but the determination of machine learning hyperparameters and offline training processes require more computing and storage resources, making them difficult to use on edge devices with limited computing resources.
[0004] In practical applications, a radio signal classification method with high accuracy and less computing resources is needed. The present invention is produced to meet this practical demand. Summary of the invention
[0005] 1. Technical issues to be resolved
[0006] The technical problem to be solved by the present invention is how to provide a radio signal classification method based on random forest to achieve a radio signal classification method with high accuracy and less computing resources required.
[0007] (II) Technical solution
[0008] In order to solve the above technical problems, the present invention proposes a radio signal classification method based on random forest, which comprises the following steps:
[0009] S1. Determine the number of intervals of the histogram and generate a histogram data set of wireless signal samples
[0010] Assume that there are m samples in total, each sample is an N-dimensional vector, and a set of histograms is generated for each sample. The histogram of each sample is an n-dimensional vector, where n is the number of intervals, and the histogram data set is an m×n matrix;
[0011] S2. Determine the number of decision trees in the random forest
[0012] The number of random forest decision trees T is calculated according to the statistical characteristics of the histogram data set of the radio signal samples;
[0013] S3. Determine the depth of random forest
[0014] The number of trees in the random forest is fixed, and the depth of the random forest is determined according to the iterative process;
[0015] S4. Offline training of random forests, and online classification of radio signals based on the trained random forest ensemble learning model.
[0016] (III) Beneficial effects
[0017] The present invention proposes a radio signal classification method based on random forests. The radio signal classification method based on random forests of the present invention adopts the characteristics of radio signals to design a random forest integrated neural network for radio signal classification. The method has a simple structure and requires few resources, and is particularly suitable for edge devices with fewer resources. It can play an important role in the intelligent processing of radio signals. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 The figure is a flow chart of the radio signal classification method based on random forest of the present invention. DETAILED DESCRIPTION
[0019] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.
[0020] The present invention discloses a radio signal classification method based on random forest, and designs a random forest neural network for radio signal classification. The method mainly includes: (1) determining the number of intervals of the histogram and generating a histogram data set of samples. Assume that there are M samples, and each sample is an N-dimensional vector. Generate a set of histograms for each sample, and the histogram of each sample is an n-dimensional vector, where n is the number of intervals. The histogram data set is an M×n matrix. (2) Determine the number of decision trees of the random forest. Calculate the number T of random forest decision trees according to the statistical characteristics of the radio signal samples. (3) Determine the depth of the random forest. (4) Offline train the random forest, and realize online classification of radio signals based on the trained random forest ensemble learning model.
[0021] The present invention proposes a radio signal classification method based on random forest, designs a random forest integrated neural network for radio signal classification, which has high accuracy and requires less computing resources. It can be implemented on edge devices with less resources. The method includes the following steps:
[0022] S1. Determine the number of intervals of the histogram and generate a histogram data set of wireless signal samples.
[0023] Assume there are m samples in total, each sample is an N-dimensional vector. Generate a set of histograms for each sample, and the histogram of each sample is an n-dimensional vector, where n is the number of intervals. The histogram data set is an m×n matrix.
[0024] S11, calculate the strength of the wireless signal sample signal. Assume that the IQ signal sampling value of the i-th sample is X i =((I i1 , Q i1 ),(I i2 , Q i2 ),...,(I ij , Q ij ),...,(I iN , Q iN )), where I ij is the in-phase signal, Q ij is an orthogonal signal, i = 1, 2, ..., m, j = 1, 2, ..., N,. The intensity information of the calculated sample is:
[0025]
[0026] S12, calculate the number of histogram intervals. The number of intervals n of the sample histogram is related to the sample dimension N. The number of intervals n is determined according to the sample dimension N. The value range of n is
[0027]
[0028] S13, generate a histogram. First, A i The elements in are arranged in ascending order, and then the sample range and group boundaries are calculated respectively. Finally, the frequency of elements in each interval is counted, and the sample histogram is obtained as H i =(H i1 ,H i2 ,...,H in ).
[0029] S2. Determine the number of decision trees in the random forest.
[0030] The number T of random forest decision trees is calculated based on the statistical characteristics of the histogram data set of radio signal samples.
[0031] S21. Calculate the mean and variance of the sample histogram. Assume that the i-th sample histogram is H i =(H i1 ,H i2 ,...,H in ), calculate the mean value μ of the sample i and variance σ i for
[0032]
[0033] S22. Calculate the statistical characteristics of each sample
[0034]
[0035] S23, average value of the statistical characteristics of m samples of wireless signal
[0036]
[0037] The number of random forest decision trees is the integer closest to T.
[0038] S3. Determine the depth of the random forest. The number of trees in the random forest is fixed, and the depth of the random forest is determined according to the iterative process. The depth of the random forest is equal to the minimum value of the depth of the random forest when the accuracy is the highest.
[0039] S4. Offline training of random forests, and online classification of radio signals based on the trained random forest ensemble learning model.
[0040] Embodiment 1:
[0041] The following describes the specific implementation of the present invention, but does not limit the specific implementation method.
[0042] The specific implementation process of the present invention is as follows.
[0043] (1) Determine the number of intervals of the histogram and generate a histogram data set of samples. Assume that there are m samples and each sample is an N-dimensional vector. Generate a set of histograms for each sample. The histogram of each sample is an n-dimensional vector, where n is the number of intervals. The histogram data set is an m×n matrix.
[0044] In specific implementation, assume that the IQ signal of the i-th sample is X i =((I i1 ,Q i1 ),(I i2 ,Q i2 ),...,(I iN ,Q iN )),I ij is the in-phase signal, Q ij is an orthogonal signal. The intensity information of the sample is
[0045] First A i The elements in A are arranged in ascending order i =(A′ i1 ,A′ i2 ,...,A′ iN ),A′ i1 ≤A′ i2 ≤,....,≤A′ iN ;
[0046] Then the sample range and group boundary are calculated as R = A' iN -A′ i1 CB=A'i1 +j×R / n, j=1,2,…,n. It is worth noting that n is the number of intervals in the sample histogram, and its value is related to the sample dimension N. The value range of n is
[0047] Finally, the frequency of elements in each interval is counted, and the sample histogram is obtained as H i =(H i1 ,H i2 ,...,H in ).
[0048] (2) Determine the number of decision trees in the random forest. The number of decision trees in the random forest is calculated based on the statistical characteristics of the radio signal samples.
[0049] In specific implementation, assume that the histogram of the i-th sample is H i =(H i1 ,H i2 ,...,H in ), respectively calculate the mean μ of the histogram i and variance σ i for Then calculate the statistical characteristics of each sample Finally, the average value of the statistical characteristics of the m samples of the wireless signal can be obtained The number of random forest decision trees is the integer closest to T.
[0050] (3) Determine the depth of random forest.
[0051] The number of trees in the random forest is fixed, and the depth of the random forest is determined according to the iterative process. As the depth increases, the accuracy will first increase and then remain unchanged, but the complexity keeps increasing, so the depth of the random forest is the minimum value of the depth of the random forest when the accuracy is the highest.
[0052] (4) Offline training of random forests and online classification of radio signals based on the trained random forest model.
[0053] Embodiment 2:
[0054] A radio signal classification method based on random forest, the method comprising the following steps:
[0055] (1) Determine the number of intervals of the histogram and generate a histogram data set of samples.
[0056] (2) Determine the number of decision trees in the random forest.
[0057] (3) Determine the depth of random forest.
[0058] (4) Offline training of random forests and online classification of radio signals based on the trained random forest ensemble learning model.
[0059] Furthermore, step (1) of the method includes the following sub-steps:
[0060] (11) Calculate the strength of the wireless signal sample. Assume that the IQ signal of the i-th signal sample is X i =((I i1 ,Q i1 ),(I i2 ,Q i2 ),...,(I iN ,Q iN )),I ij is the in-phase signal, Q ij is an orthogonal signal. The intensity information of the sample is
[0061] (12) Calculate the number of histogram intervals. The number of intervals n of the sample histogram is related to the sample dimension N. The range of n is
[0062] (13) Generate a histogram. First, A i The elements in are arranged in ascending order, and then the sample range and group boundaries are calculated respectively. Finally, the frequency of elements in each interval is counted, and the sample histogram is obtained as H i =(H i1 ,H i2 ,...,H in ).
[0063] Furthermore, step (2) of the method includes the following sub-steps:
[0064] (21) Calculate the mean of the sample histogram. Assume that the i-th sample histogram is H i =(H i1 ,H i2 ...H in ), calculate the mean μ of the sample i and variance σ i They are: and
[0065] (22) Calculate the statistical characteristics of each sample
[0066] (23) The average value of the statistical characteristics of the m samples of the wireless signal The number of random forest decision trees is the integer closest to T.
[0067] The random forest-based radio signal classification method of the present invention uses the characteristics of radio signals to design a random forest integrated neural network for radio signal classification, has a simple structure, requires few resources, and is particularly suitable for edge devices with fewer resources. It can play an important role in the intelligent processing of radio signals.
[0068] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A radio signal classification method based on random forest, It is characterized in that The method comprises the following steps: S1. Determine the number of intervals of the histogram and generate a histogram data set of wireless signal samples Assume that there are m samples in total, each sample is an N-dimensional vector, and a set of histograms is generated for each sample. The histogram of each sample is an n-dimensional vector, where n is the number of intervals, and the histogram data set is an m×n matrix; S2. Determine the number of decision trees in the random forest The number of random forest decision trees T is calculated according to the statistical characteristics of the histogram data set of the radio signal samples; S3. Determine the depth of random forest The number of trees in the random forest is fixed, and the depth of the random forest is determined according to the iterative process; S4. Offline training of random forests, and online classification of radio signals based on the trained random forest ensemble learning model.
2. The radio signal classification method based on random forest as claimed in claim 1, It is characterized in that The S1 specifically includes: S11. Calculate the strength of the wireless signal sample signal Assume that the IQ signal sampling value of the i-th sample is X i =((I i1 , Q i1 ),(I i2 , Q i2 ),...,(I ij , Q ij ),...,(I iN , Q iN )), where I ij is the in-phase signal, Q ij is an orthogonal signal, i = 1, 2, ..., m, j = 1, 2, ..., N; the intensity information of the calculated sample is: S12. Calculate the number of histogram intervals The number of intervals n of the sample histogram is related to the sample dimension N, and the number of intervals n is determined according to the sample dimension N; S13. Generate histogram First, A i The elements in are arranged in ascending order, and then the sample range and group boundaries are calculated respectively. Finally, the frequency of elements in each interval is counted, and the sample histogram is obtained as H i =(H i1 ,H i2 ,...,H in ).
3. The radio signal classification method based on random forest as claimed in claim 2, It is characterized in that In S12, the value range of n is 4. The radio signal classification method based on random forest as claimed in claim 3, It is characterized in that The S13 includes: First A i The elements in A are arranged in ascending order as A' i =(A' i1 ,A' i2 ,...,A' iN ),A' i1 ≤A' i2 ≤,....,≤A' iN ; Then calculate the sample range and group boundaries respectively; Finally, the frequency of elements in each interval is counted, and the sample histogram is obtained as H i =(H i1 ,H i2 ,...,H in ).
5. The radio signal classification method based on random forest as claimed in claim 4, It is characterized in that The sample range is R = A' iN -A' i1 ; Group boundary is CB=A' i1 +j×R / n, j=1,2,…,n.
6. The radio signal classification method based on random forest according to any one of claims 2 to 5, It is characterized in that The S2 specifically includes: S21. Calculate the mean and variance of the sample histogram Assume that the histogram of the i-th sample is H i =(H i1 ,H i2 ,...,H in ), calculate the mean value μ of the sample i and variance σ i ; S22. Calculate the statistical characteristics of each sample S23, average value of the statistical characteristics of m samples of wireless signal The number of random forest decision trees is the integer closest to T.
7. The radio signal classification method based on random forest as claimed in claim 6, It is characterized in that The mean value of the sample is μ i for 8. The radio signal classification method based on random forest as claimed in claim 7, It is characterized in that The sample variance σ i for:
9. The radio signal classification method based on random forest as claimed in claim 8, It is characterized in that In S3, the depth of the random forest is equal to the minimum value of the depth of the random forest when the accuracy is the highest.
10. The radio signal classification method based on random forest as claimed in claim 9, It is characterized in that This method is implemented on edge devices with fewer resources.
