IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics

Abstract

The invention discloses an IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics. The method comprises the following steps: carrying out N-point fast Fourier transform on a received IM signal to obtain N subcarriers, and taking the N subcarriers as sample labels according to activation conditions of marked subcarriers; wavelet decomposition is carried out on the preprocessed subcarrier R (k), statistical characteristics are extracted from high-frequency detail components and low-frequency approximate components obtained through decomposition and serve as sample characteristics of the subcarrier R (k), the sample characteristics are input into a machine learning algorithm and dichotomy is carried out, and according to the activation condition of the subcarrier in a sample characteristic area and comparison with a sample label, the subcarrier R (k) is identified. And training hyper-parameters of a machine learning algorithm. And for IM signals to be detected, after wavelet decomposition is carried out, statistical characteristics of wavelet coefficients are extracted, the statistical characteristics are input into the trained machine learning algorithm, the activation condition of subcarriers is obtained, and detection is completed. According to the method, the threshold value is determined by using the optimization capacity of the model, and the characteristics of the signal can be learned more comprehensively.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and relates to the detection of index modulation signals, and in particular to an IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics. Background Art

[0002] Index Modulation is an emerging modulation technology that transmits data by using the index information of the signal. Common index information carriers are subcarriers, antennas, etc. It can be used to improve the spectrum efficiency and energy efficiency of wireless communication systems.

[0003] Existing signal detection methods based on subcarrier energy, such as maximum likelihood detection, are simple and effective under high signal-to-noise ratios. They can quickly search the energy of each subcarrier and determine whether the subcarrier modulates information. However, they are easily affected by noise. Under low signal-to-noise ratios, the signal energy and noise energy are close. The energy-based detection method has low signal discrimination and a reduced detection probability.

[0004] Wavelet transform uses the method of time-frequency analysis to process signals, and can decompose each subcarrier into a low-frequency approximate component and a high-frequency detail component. Since the wavelet decomposition coefficient is related to the time-frequency characteristics of the signal, it can reflect the characteristics of the modulated signal itself. Therefore, compared with the traditional detection method based on subcarrier energy, the signal detection based on wavelet transform has better performance under low signal-to-noise ratio. However, the traditional IM signal detection method based on wavelet transform directly uses the mean of all wavelet decomposition coefficients of the subcarrier as the threshold for judging the activation state of the subcarrier. This threshold determination method has no theoretical support and cannot well reflect the information of the signal itself, resulting in poor detection effect. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention proposes an IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics. The IM signal is decomposed based on time-frequency analysis wavelet transform, and the statistical characteristic values ​​of the obtained wavelet decomposition coefficients are used as sample features of the IM signal. Whether the subcarrier is activated is used as a classification variable for model training to achieve high-resolution IM signal detection.

[0006] The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics has the following specific steps:

[0007] S1: The received IM signal is subjected to N-point fast Fourier transform to obtain N subcarriers, and necessary preprocessing is performed, including denoising and normalization, to improve the effect of subsequent processing. The activation of the marked subcarriers is used as the sample label for model training.

[0008] S2: Determine the wavelet system and the number of wavelet decomposition layers J, and perform J-layer wavelet decomposition on the preprocessed subcarrier R(k). The larger the value of J, the higher the accuracy of signal analysis:

[0009]

[0010] Where k = 0, 1, 2, ..., N-1, j represents the scaling variable, n represents the translation variable, φ jn (k) represent wavelet function and scaling function respectively.

[0011] Preferably, the db5 wavelet system is selected to perform wavelet decomposition on the subcarriers.

[0012] S3: J groups of high-frequency detail components d obtained by wavelet transformation jn and low frequency approximate component a jn Statistical features are extracted as sample features of subcarrier R(k), which are input into the machine learning algorithm for binary classification. The activation of subcarriers is distinguished according to the sample features and compared with the sample labels to train the hyperparameters of the machine learning algorithm.

[0013] Preferably, the statistical characteristics include mean, variance, standard deviation, median, autocorrelation coefficient, monotonic increasing property, feature cross-correlation, kurtosis, skewness and extreme value.

[0014] Preferably, the machine learning algorithm is support vector machine (SVM), decision tree (Decision Trees), random forest (Random Forest), gradient boosting tree (Gradient Boosting Trees) or extreme gradient boosting tree (Extreme Gradient Boosting, XGBoost).

[0015] S4: For the IM signal to be detected, wavelet decomposition is performed to extract the statistical characteristics of the wavelet coefficients, which are input into the machine learning algorithm trained in S3 to obtain the activation status of the subcarriers and complete the detection.

[0016] The present invention has the following beneficial effects:

[0017] The wavelet decomposition technology is applied to IM-based signal detection. Compared with the detection method based on signal energy, it can effectively reduce the probability of missed detection of activated subcarriers under low signal-to-noise ratio. At the same time, in the traditional IM signal detection method based on wavelet transform technology, the mean of all wavelet decomposition coefficients is simply used as the threshold for determining whether the subcarrier is activated. This method uses the wavelet decomposition coefficient as the eigenvalue and whether the subcarrier is activated as the classification variable for model training, and uses the model's optimization ability to determine the threshold. It can learn the characteristics of the signal more comprehensively, and the probability of IM signal detection is improved by about 0.017 compared with the traditional signal detection method based on wavelet transform, and by about 0.03 compared with the detection method based on signal energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics;

[0019] Figure 2 is an index modulation design diagram in the embodiment;

[0020] Figure 3 It is a schematic diagram of wavelet decomposition;

[0021] Figure 4 is the subcarrier detection probability in the embodiment;

[0022] Figure 5 is the probability of missed detection of activated subcarriers in the embodiment. ; DETAILED DESCRIPTION

[0023] The present invention will be further explained below with reference to the accompanying drawings;

[0024] like Figure 1 As shown, the improved IM signal detection method based on XGBoost algorithm and wavelet transform specifically includes the following steps:

[0025] Step S1: Index modulation is performed on the speech signal from the public data set LJSpeech, and after being transmitted through the channel, it is used as the received IM signal in this embodiment. The index modulation design method is as follows: Figure 2 As shown, every 4 system subcarriers form a subblock, and the first subcarrier in the subblock is fixedly activated to carry pilot symbols, and the remaining 3 subcarriers are selectively activated according to index information.

[0026] The subcarriers of the IM received signal after fast Fourier transformation are processed by Kalman filtering to effectively filter out high-frequency noise, retain useful low-frequency components, and provide smooth output to improve the effect of subsequent processing.

[0027] Step S2: Due to its excellent orthogonality and compact support, it is particularly suitable for signal processing. Therefore, this embodiment uses the db5 wavelet system to perform a 5-layer wavelet decomposition on the subcarrier signal, such as Figure 3 As shown, we get a set of wavelet decomposition coefficients [d 1n , a 1n , d 2n , a 2n , ..., d 5n , a 5n ]. Use variable y i ={-1 or 1} marks whether each subcarrier is activated.

[0028] Step S3: Extract statistical features from the wavelet decomposition coefficients of subcarrier R(k), including mean, variance, standard deviation, median, autocorrelation coefficient, monotonically increasing property, characteristic cross-correlation, kurtosis, skewness, and extreme value, as subcarrier R i (k) Sample feature X ik ={z1,z2,z3,...,z 10}, variable y i As the sample label, 1 indicates that the subcarrier is activated and recorded as a positive sample, and -1 indicates that the subcarrier is silent and recorded as a negative sample.

[0029] Step S4: Divide the original data into a training set D and a test set T in a ratio of 1:1. Input the sample features of the subcarrier R(k) in the training set D into the machine learning algorithm for binary classification. Distinguish the activation of the subcarriers according to the sample features and compare them with the sample labels to train the hyperparameters of the machine learning algorithm.

[0030] In this embodiment, the extreme gradient boosting tree (ExtremeGradientBoosting, XGBoost) is selected to complete the binary classification task, the logistic function is used as the classification model, the probability model is used for calculation, and the Gini coefficient G is used. ini As the basis of feature importance, a greedy search algorithm is applied to put important features in the first decision-making place. The specific steps are as follows:

[0031] S4.1: The Gini coefficient represents the probability of randomly selecting a sample from a data set and measuring its probability of being incorrectly classified into other categories. The larger the Gini coefficient, the greater the uncertainty of the sample set. First, calculate the Gini coefficient G of the sample set D ini (D), as the uncertainty of the sample set D:

[0032]

[0033] Where |D| represents the number of samples in the dataset D, |C q| is the number of samples belonging to the qth class in the data set D; Q is the total number of sample categories in the data set D. In this embodiment, Q=2, C1 represents the positive sample set, and C2 represents the negative sample set.

[0034] For the 10 statistical features of the subcarriers in the sample set D, calculate each statistical feature z in turn A Gini coefficient G after segmenting the sample set D by eigenvalue a ini (D,A,a):

[0035]

[0036] Gini coefficient ini (D, A, a) indicates that according to the statistical characteristics z A The value of z A <a and z A ≥a The uncertainty of sample set D after dividing sample set D into subsets D1 and D2.

[0037] The difference in the Gini coefficient of the dataset D after segmentation based on feature A at value a is defined as the Gini gain G ain (D,A,a):

[0038] G ain (D,A,a)=G ini (D)-G ini (D,A,a)

[0039] Gini gain G ain The larger the value of (D,A,a), the better the effect of the segmentation scheme using the value a as the segmentation point in reducing the uncertainty of the set.

[0040] S4.2: Generate a decision tree with a depth of 0. The goal of building a decision tree is to minimize the sum of the objective functions of all leaf nodes of the decision tree. According to the importance of statistical features obtained from prior information, the left and right leaf nodes are generated according to the splitting order of variance, standard deviation, autocorrelation coefficient, mean, feature cross-correlation, extreme value, monotonically increasing, kurtosis, skewness, and median. Then, the feature sorting is forward adjusted according to the model detection probability to achieve the purpose of optimizing the detection probability. The best splitting order is mean, variance, standard deviation, autocorrelation coefficient, feature cross-correlation, extreme value, kurtosis, monotonically increasing, median, and skewness.

[0041] Calculate each statistical feature z in turn A The gain coefficient O corresponding to the eigenvalue a as the split point bj gain , the eigenvalue a corresponding to the optimal gain coefficient value is used as the splitting point of the current feature:

[0042]

[0043] Among them, λ is the regularization coefficient; It is the square of the sum of the first-order derivatives of the samples contained in the left leaf node; H is the square of the sum of the first-order derivatives of the samples contained in the right leaf node; L is the sum of the second-order derivatives of the samples contained in the left leaf node; H R is the sum of the second-order derivatives of the samples contained in the right leaf node; is the objective function value of the left leaf node; is the objective function value of the leaf node on the right; is the objective function value when no other features are selected.

[0044] Construct the loss function L oss :

[0045]

[0046] Where l(·) is the logistic function; m is the total number of samples, is the predicted label of the model for the i-th sample.

[0047] S4.3: Use the XGBoost model to model the training set D 100 times, and select Bayesian optimization to train the model hyperparameters, including the tree depth, number of trees, regularization term parameters and learning rate of the XGBoost model, so that the model can achieve better performance on the training set.

[0048] Step S5: Using the model trained in step S4, make discrimination predictions on the samples in the 50,000 test set T under different signal-to-noise ratio conditions, and compare them with the energy-based detection method and the traditional wavelet transform-based detection method. Figure 4 and Figure 5 The detection probability and missed detection probability of activated subcarriers of different methods are shown respectively. When the signal-to-noise ratio is -4dB, the detection probability of this method is 0.989692, which is about 0.017 higher than the traditional wavelet transform-based detection method and about 0.03 higher than the energy-based detection method. The missed detection probability of this method is about 0.8%, which is about 1.2 percentage points higher than the traditional wavelet transform-based detection method and about 2.7 percentage points higher than the energy-based detection method. This method is improved compared with the other two existing methods, which shows that this method is superior to the energy-based detection method and the traditional wavelet transform-based detection method.

[0049] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection claimed in the present invention is defined by the attached claims and their equivalents. The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art of the present invention may make various modifications or supplements to the described specific embodiments or replace them in a similar manner, but will not deviate from the spirit of the present invention or exceed the scope defined by the attached claims.

Claims

1. An IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics, wherein the received IM signal is subjected to an N-point fast Fourier transform to obtain N subcarriers, and the activation status of the marked subcarriers is used as a sample label to detect the subcarrier activation status of the IM signal using a machine learning algorithm, wherein: The specific steps are as follows: S1: denoising and normalizing the subcarrier signal; S2: Determine the wavelet system and the number of wavelet decomposition layers J, and perform wavelet decomposition on the preprocessed subcarrier R(k); S3: J groups of high-frequency detail components d obtained by wavelet decomposition jn and low frequency approximate component a jn Extract statistical features as sample features of subcarrier R(k), input them into the machine learning algorithm for binary classification, distinguish the activation of subcarriers according to sample features, and compare them with sample labels to train the hyperparameters of the machine learning algorithm; S4: For the IM signal to be detected, wavelet decomposition is performed to extract the statistical characteristics of the wavelet coefficients, which are input into the machine learning algorithm trained in S3 to obtain the activation status of the subcarriers and complete the detection.

2. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 1, characterized in that: The subcarrier signal is processed by Kalman filtering to retain the low-frequency components and filter out the high-frequency noise.

3. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 1, characterized in that: The db5 wavelet system is selected to perform wavelet decomposition on the subcarriers.

4. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 3, characterized in that: Set the number of wavelet decomposition levels J=5.

5. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 1, characterized in that: The statistical characteristics include mean, variance, standard deviation, median, autocorrelation coefficient, monotonic increasing property, feature cross-correlation, kurtosis, skewness and extreme value.

6. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 1, characterized in that: The machine learning algorithm is a support vector machine, a decision tree, a random forest, a gradient boosting tree or an extreme gradient boosting tree.

7. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 6, characterized in that: Use the logistic function as the classification model for the extreme gradient boosting tree.

8. The IM signal detection method based on machine learning and wavelet decomposition coefficient characteristics as claimed in claim 6, characterized in that: The Gini coefficient is used as the basis for feature importance, and a greedy search algorithm is applied to put important features in the first place for decision making, to determine the order of features to be split in the extreme gradient boosting tree.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 8.

Images