Multi-feature extraction signal modulation identification method based on miniaturized radio frequency instrument

Abstract

The invention discloses a multi-feature extraction signal modulation identification method based on a miniaturized radio frequency instrument, and belongs to the technical field of information. The invention provides an electromagnetic signal multi-feature fusion feature quantity construction and modulation identification method and scheme which can be deployed on a resource-limited miniaturized portable radio frequency instrument. Comprising the following steps: extracting multi-feature parameters; constructing characteristic quantity parameters; designing a classifier through a modulation recognition algorithm; and through modulation identification test and analysis, identification of a signal modulation mode is completed. The method has the advantages of low algorithm complexity, rich feature representation, small operation resource demand and simple engineering implementation; the method has the advantages of low calculation complexity, short operation time, high recognition speed, high recognition accuracy and the like, can help a user to quickly determine the modulation type of a signal, and improves the new environment exploration efficiency and spectrum monitoring and detection.

Description

Technical Field

[0001] The invention belongs to the field of information technology, and in particular relates to a signal modulation recognition method based on multi-feature extraction of miniaturized radio frequency instruments. Background Art

[0002] As electronic information systems continue to expand their applications in the air, space, land and sea, electromagnetic signal measurement and analysis will develop from known and certain signals to unknown and uncertain signals, which requires not only rapid signal search, but also real-time signal tracking and identification. Electromagnetic signal modulation recognition technology is widely used in cognitive radio, software-defined radio and other communication fields. It can better guide subsequent demodulation signals and improve the flexibility of digital signal processing modules, thereby achieving accurate acquisition of electromagnetic signal-carried information and reconstruction and reproduction of unknown electromagnetic signals.

[0003] In the existing technical solutions, the modulation recognition method based on likelihood ratio is to calculate the likelihood value of the signal sequence, and then compare it with the set threshold to determine the recognition result. It can be divided into average likelihood ratio test, generalized likelihood ratio test and mixed likelihood ratio test. The modulation recognition method based on feature extraction mainly relies on professional and empirical knowledge. It manually extracts features from the signal, establishes a feature extraction model, and realizes signal recognition based on feature differences. It is a process of mapping the original signal data to modulation features. Common signal features include instantaneous features, statistical features, and transform domain features. The modulation recognition technology based on machine learning and deep learning simultaneously completes the functions of feature extraction and classifier by constructing a neural network model, realizing an end-to-end process from the original signal data directly to the modulation mode recognition result. Common machine learning classifiers include decision trees, support vector machines, shallow neural networks, etc. Deep neural networks include convolutional neural networks, long short-term memory networks, capsule networks, etc.

[0004] The existing technical solutions have the following disadvantages:

[0005] The modulation recognition method based on likelihood ratio is suitable for ideal channel conditions. However, factors such as high computational complexity, the need for complete prior information, and poor robustness to environmental interference limit the practical application of this method.

[0006] The data feature representation capability of the modulation recognition method based on feature extraction is relatively simple.

[0007] Modulation recognition technology based on machine learning and deep learning requires a large amount of actual sample data to train the learning model. The algorithm is highly complex and requires high computing resources. It is difficult to deploy on resource-constrained instruments and has weak engineering applications.

[0008] The signal received by the receiving antenna can be divided into an intermediate frequency signal, a baseband signal after down-conversion, and a symbol sequence output by a matched filter. After receiving the RF signal, the miniaturized RF instrument obtains two baseband signals, the in-phase component (I) and the orthogonal component (Q), through down-conversion processing, which contain the effects of frequency offset, noise interference, I / Q imbalance, shaping filter, etc. Figure 1 As shown, the IQ data of QPSK collected at different times is constantly changing, the angle of the point in the constellation diagram changes linearly with time, the constellation diagram begins to rotate and the rotation speed becomes faster as the frequency offset increases. The existing technical solutions are usually applicable to relatively ideal signal data, but are not very applicable to the above signals. Summary of the invention

[0009] In view of the above technical problems existing in the prior art, the present invention proposes a signal modulation recognition method based on multi-feature extraction of miniaturized radio frequency instruments, which has a reasonable design, overcomes the shortcomings of the prior art, and has good effects.

[0010] In order to achieve the above object, the present invention adopts the following technical solution:

[0011] A signal modulation recognition method based on multi-feature extraction of miniaturized radio frequency instruments comprises the following steps:

[0012] Step 1: Multi-feature parameter extraction;

[0013] Step 2: feature parameter construction;

[0014] Step 3: Design a classifier;

[0015] Design a classifier through modulation recognition algorithm;

[0016] Step 4: Complete the identification of signal modulation mode;

[0017] Through modulation identification test and analysis, the signal modulation method can be identified.

[0018] Preferably, step 1 includes the following steps:

[0019] Step 1.1: Calculate different high-order cumulants and instantaneous amplitude characteristic parameters;

[0020] M pq represents the p-order q-conjugate mixed moment, y(n) represents the signal IQ data,

[0021]

[0022]

[0023] Among them, the symbol * indicates conjugation; C pq represents the p-order q-conjugate higher-order cumulant;

[0024] C 20 =M 20 (11)

[0025] C 21 =M 21 (12)

[0026]

[0027] C 41 =M 40 -3M 21 M 20 (14)

[0028]

[0029] C 60 =M 60 -15M 40 M 20 +30M 20 3 (16)

[0030]

[0031] Step 1.2: Calculate the instantaneous amplitude A(n) using the IQ data;

[0032]

[0033] Among them, I(n) and Q(n) represent the real part and imaginary part of y(n) respectively.

[0034] Preferably, step 2 includes the following steps:

[0035] After calculating different characteristic parameters, construct a suitable feature quantity F, and complete the signal recognition of different modulation modes by setting an artificial threshold classifier; the feature quantity is constructed based on the ratio method based on the cumulative amount of each order, and the constructed feature quantity parameters are as follows:

[0036] F1=|C 40 / C 42 |;F2=|C 41 / C 42 |;

[0037] in, σ are the mean and standard deviation of A(n);

[0038]

[0039] Beneficial technical effects brought by the present invention:

[0040] The present invention provides a method and scheme for constructing feature quantities and modulating recognition of electromagnetic signals with multi-feature fusion that can be deployed on resource-constrained miniaturized and portable radio frequency instruments. The method has the advantages of low algorithm complexity, rich feature representation, small computing resource requirements, and simple engineering implementation. The method has the advantages of short running time, fast recognition speed, and high recognition accuracy while ensuring low computational complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is the QPSK constellation diagram at different acquisition times;

[0042] Figure 2 Flowchart for multi-feature extraction signal modulation recognition;

[0043] Figure 3 Confusion matrix plot for multi-feature extraction modulation recognition. DETAILED DESCRIPTION

[0044] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0045] Feature extraction is a process of mapping raw signal data into modulation features, which reduces the data dimension, has low algorithm complexity, and has strong feature interpretability. The instantaneous features of changes in parameters such as signal amplitude, frequency, and phase have the ability to classify analog signals and low-order digital signals. The differences between different modulated signals are also reflected in high-order statistical features, and the algorithm that uses high-order cumulant features to identify the modulation type also has the advantages of being insensitive to interference factors such as frequency deviation and phase deviation, having a small demand for sample data volume, and having certain noise resistance. In order to reduce the algorithm complexity and the demand for instrument computing resources, the present invention reduces the preprocessing of signal data, directly uses the original IQ data to perform multi-feature fusion extraction of high-order cumulant features and instantaneous features, constructs feature parameters, and then designs a classifier to complete the identification of signal modulation methods.

[0046] 1. Multi-feature parameter extraction

[0047] Firstly, the characteristic parameters of different high-order cumulants and instantaneous amplitudes are calculated.

[0048] M pq represents the p-order q-conjugate mixed moment, y(n) represents the signal IQ data,

[0049]

[0050] The symbol * indicates conjugation.

[0051] C pq represents the p-order q-conjugate higher-order cumulant,

[0052] C 20 =M20 (11)

[0053] C 21 =M 21 (12)

[0054]

[0055] C 41 =M 40 -3M 21 M 20 (14)

[0056]

[0057] C 60 =M 60 -15M 40 M 20 +30M 20 3 (16)

[0058]

[0059] Using IQ data, calculate the instantaneous amplitude A(n),

[0060]

[0061] Where I(n) and Q(n) represent the real part and imaginary part of y(n) respectively.

[0062] 2. Feature parameter construction

[0063] After calculating different feature parameters, construct a suitable feature quantity F, and complete the signal recognition of different modulation modes by setting an artificial threshold classifier. First, the feature quantity is constructed based on the ratio method based on the cumulative quantity of each order, which can remove the influence of power on the recognition result. The constructed feature quantity parameters are F1=|C 40 / C 42 |,F2=|C 41 / C 42 |, in σ are the mean and standard deviation of A(n),

[0064]

[0065] 3. Modulation recognition algorithm

[0066] The modulation type set identified by the present invention is Ω = {AM, FM, 2ASK, 2FSK, BPSK, QPSK, 8PSK, 16QAM, MSK}. The identification algorithm steps are as follows:

[0067] Structural feature F1 = |C 40 / C 42 |, T(F1) represents the classification threshold. Identify and classify AM, 2AKS, BPSK, QPSK, 16QAM and FM, 2FSK, 8PSK, MSK signals.

[0068] if F1>T(F1)

[0069] output:AM, 2AKS, BPSK, QPSK, 16QAM

[0070] Structural feature F2 = |C 41 / C 42 |, T(F2) represents the classification threshold. Identify and classify AM, 2AKS, BPSK, QPSK, 16QAM signals

[0071] if F2>T(F2)

[0072] Structural feature T1(F6) and T2(F6) represent classification thresholds, which can be used to identify and classify AM, 2AKS, and BPSK signals.

[0073]

[0074] Structural feature T(F5) represents the classification threshold, which can be used to identify and classify QPSK and 16QAM signals.

[0075]

[0076] Structural feature T1(F4) and T2(F4) represent classification thresholds. They can identify and classify FM, 2FSK, 8PSK, and MSK signals.

[0077] if F4>=T1(F4)

[0078] output:2FKS

[0079] else

[0080] Structural feature T1(F3) and T2(F3) represent classification thresholds, which can be used to identify and classify FM, 8PSK, and MSK signals.

[0081]

[0082] The multi-feature extraction signal modulation recognition process is as follows Figure 2 shown.

[0083] 4. Modulation recognition test and analysis

[0084] The above modulation type signal is generated using a vector signal generator, the symbol rate is set to 4MHz, a root raised cosine roll-off filter is used, the roll-off factor is 0.35, and the carrier frequency is 1GHz. The signal modulation recognition algorithm is deployed on an RF instrument platform with a CPU of ARM dual-core 1.2GHz and 2GB of memory to perform real-time signal modulation recognition and non-real-time signal modulation recognition of captured data files. The test setting processes a signal sequence length of 1024 points, and tests 100 signals for each modulation mode. After actual deployment, the average running time of single signal recognition is about 5ms, which is fast and efficient.

[0085] Plot the confusion matrix of the recognition classification to analyze the modulation recognition accuracy. Figure 3 As shown in the figure, the horizontal axis is the predicted label, the vertical axis is the true label, and the corresponding intersection area of ​​the predicted label and the true label is the confusion matrix of the classification. The numbers in each row represent the recall rate of each modulation type, representing the classification and recognition performance of the model for this type of data; the diagonal of the matrix represents the accuracy of each modulation type recognition. The larger the value, the darker the color, the higher the accuracy, and the smaller the value, the lighter the color, and the lower the accuracy. The values ​​in other positions of each row except the diagonal line represent the probability of misidentification as other types. It can be clearly seen that 100% recognition is achieved for AM, 2ASK, 2FSK, BPSK, QPSK, 8PSK, 16QAM, and MSK, and 87% recognition is achieved for FM, with a high overall recognition accuracy.

[0086] Compared with the existing methods, the feature quantity construction, modulation identification method and scheme of multi-feature fusion of electromagnetic signals deployed on resource-constrained miniaturized and portable radio frequency instruments provided by the present invention have low algorithm complexity, small computing resource requirements, short running time, fast identification speed, high identification accuracy and easy engineering implementation, which can help users quickly determine the modulation type of the signal and improve the efficiency of new environment surveys and spectrum monitoring and detection.

[0087] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A signal modulation recognition method based on multi-feature extraction of miniaturized radio frequency instruments, characterized in that: The steps include: Step 1: Multi-feature parameter extraction; Step 2: feature parameter construction; Step 3: Design a classifier; Design a classifier through modulation recognition algorithm; Step 4: Complete the identification of signal modulation mode; Through modulation identification test and analysis, the signal modulation method can be identified.

2. The signal modulation recognition method based on multi-feature extraction of miniaturized radio frequency instruments according to claim 1 is characterized in that: Step 1 includes the following steps: Step 1.1: Calculate different high-order cumulants and instantaneous amplitude characteristic parameters; M pq represents the p-order q-conjugate mixed moment, y(n) represents the signal IQ data, Among them, the symbol * indicates conjugation; C pq represents the p-order q-conjugate higher-order cumulant; C 20 =M 20 (11) C 21 =M 21 (12) C 41 =M 40 -3M 21 M 20 (14) C 60 =M 60 -15M 40 M 20 +30M 20 3 (16) Step 1.2: Calculate the instantaneous amplitude A(n) using the IQ data; Among them, I(n) and Q(n) represent the real part and imaginary part of y(n) respectively.

3. The signal modulation recognition method based on multi-feature extraction of miniaturized radio frequency instruments according to claim 1 is characterized in that: Step 2 includes the following steps: After calculating different characteristic parameters, construct a suitable feature quantity F, and complete the signal recognition of different modulation modes by setting an artificial threshold classifier; the feature quantity is constructed based on the ratio method based on the cumulative amount of each order, and the constructed feature quantity parameters are as follows: in, σ are the mean and standard deviation of A(n);

Images