CN109274626B

CN109274626B - Modulation identification method based on constellation diagram orthogonal scanning characteristics

Info

Publication number: CN109274626B
Application number: CN201811389534.2A
Authority: CN
Inventors: 王卫东; 甘露; 廖红舒
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2018-11-21
Filing date: 2018-11-21
Publication date: 2020-11-13
Anticipated expiration: 2038-11-21
Also published as: CN109274626A

Abstract

The invention belongs to the technical field of communication, and particularly relates to a modulation identification method based on orthogonal scanning characteristics of a constellation diagram. The invention utilizes orthogonal scanning to convert the time complexity of the signal characteristic extraction stage from O (n) compared with the method of converting the constellation diagram into the color density spectrum image by adopting the density statistical window²) O (n) is reduced, and because the extracted signal features are one-dimensional, a simpler neural network classifier can be selected, so that the occupied resources are less, the calculated amount is less, the recognition speed is higher, and the recognition accuracy is also improved.

Description

Modulation identification method based on constellation diagram orthogonal scanning characteristics

Technical Field

The invention belongs to the technical field of communication, and particularly relates to a modulation identification method based on orthogonal scanning characteristics of a constellation diagram.

Background

Since the first article for researching automatic modulation and identification of communication signals was published in 1969 by four scholars such as c.s.weaver in technical reports of stanford university, the automatic modulation and identification technology of communication signals has been a research hotspot in the communication field, has wide application in the fields of electronic reconnaissance and countermeasure, spectrum monitoring and management, and the like, and has important significance for communication intellectualization. Existing modulation identification techniques are mainly divided into two main categories: a maximum likelihood method based on hypothesis testing and a pattern recognition method based on feature extraction.

The maximum likelihood method based on hypothesis test utilizes probability model derivation to find out the most reasonable parameter estimation quantity when the probability density of an observed value reaches the maximum, and the result is optimal from the point of Bayesian estimation. Compared with a maximum likelihood method based on hypothesis test, the pattern recognition method based on feature extraction is stable and has stronger practicability. At present, the signal characteristics for modulation identification mainly include instantaneous amplitude, frequency and phase isochronal characteristics, constellation diagram geometric characteristics, time-frequency distribution characteristics, high-order statistic characteristics, cyclostationary characteristics, and the like.

The constellation diagram of the digital modulation signal can intuitively reflect the modulation type of the signal, but the traditional identification method based on the geometrical characteristics of the constellation diagram, such as fuzzy C-mean clustering and other methods for counting constellation diagram symbol clusters, has higher requirements on the signal-to-noise ratio. In order to solve the problem, researchers propose that a density statistical window can be adopted to convert a common constellation diagram into a colorful density spectrum view, namely a statistical image about the distribution density of symbol sampling points on the constellation diagram, and a neural network is utilized to identify the image, so that the applicability of the method based on the geometrical characteristics of the constellation diagram under the condition of low signal-to-noise ratio is greatly enhanced. However, the time complexity of the method in the signal feature extraction stage is O (n)²) In order to achieve higher image recognition accuracy, deep convolutional neural networks such as GoogLeNet and AlexNet are adopted, the number of parameters of the models is in the order of millions and millions, resources are occupied, the calculation amount is large, and therefore the recognition speed is slow.

Disclosure of Invention

The present invention aims to solve the above problems, and provides a modulation identification method based on orthogonal scanning characteristics of a constellation diagram, which is essentially to perform density statistics on the distribution of symbol sampling points on the constellation diagram.

The technical scheme adopted by the invention is as follows:

a modulation identification method based on constellation diagram quadrature scanning characteristics is mainly used for identifying a digital modulation mode (PSK/QAM), and is characterized by comprising the following steps:

s1: preparing signal data

Assuming that there are v types of signal modulation schemes to be identified, and the current class label T represents each modulation scheme, the signal set to be identified can be represented as T ═ { T ═ 1, 2, …, v }. Generating simulation signals by MATLAB and then simulating realityOr directly collecting the actual signal through a signal receiver. Finally obtaining baseband signals, and acquiring m code element symbols each time to obtain a symbol sequence

Sequence element (x)_i，y_i) Is a symbol s_iCoordinates on the constellation diagram.

S2: performing orthogonal scanning on a constellation diagram

Setting the width w and the scanning length n of the density statistical window, and assuming that scanning is started from the upper left corner of the constellation diagram, the scanning starting position is

The scanning process is shown in fig. 1.

Scanning the X-axis direction of the constellation diagram to obtain a density statistical vector with the length of n

Scanning in the Y-axis direction

The specific scanning process is as follows:

x＝x₀，y＝y₀

fork＝1，2，…，ndo

x′＝x+w/2

y′＝y-w/2

d_x(k)＝count_x(S，x，x′)

d_y(k)＝count_y(S，y，y′)

x＝x′

y＝y′

end for

wherein, the function count_x(S, x, x ') is used for searching symbols which satisfy the condition that the abscissa is more than or equal to x and less than x' in the symbol sequence S, and returning the number of symbol sampling points which satisfy the condition; similarly, the function count_y(S, y, y') is a reference for finding the satisfied ordinate in the symbol sequence SSymbols less than or equal to y and greater than y'.

The signal characteristic obtained by performing orthogonal scanning on the constellation diagram is d_xy＝(d_x，d_y)。

S3: training neural network classifier

Repeating the steps S1 and S2, and obtaining a large number of signal characteristic samples d of each modulation mode in the preset signal set_xyThen, a training dataset D { (D) is constructed_xyT), and simultaneously building a neural network shown in figure 2, and optimizing the neural network by using an Adam algorithm.

S4: identifying communication signals

Extracting signal feature d from digital communication signal of unknown modulation mode by using signal feature extraction method mentioned in step S2_xyThen, the neural network classifier trained in step S3 is input to obtain a recognition result t.

The modulation identification method based on the orthogonal scanning characteristics of the constellation diagram utilizes the advantages of orthogonal scanning, and compared with the method of converting the constellation diagram into the color density spectrum image by adopting the density statistical window, the modulation identification method based on the orthogonal scanning characteristics of the constellation diagram utilizes the sliding times of the density statistical window from n²Reduced to n, so that the time complexity of the signal feature extraction stage is from O (n)²) Reduced to O (n); because the signal characteristics obtained by the method are one-dimensional, a simpler neural network classifier can be selected for identifying the three-dimensional color image characteristics, the occupied resources are less, the calculated amount is less, and the identification speed is higher; in addition, the density statistical characteristics obtained by performing orthogonal scanning on the constellation diagram are equivalent to performing secondary statistics on the density spectrum of the constellation diagram in the X-axis direction and the Y-axis direction, and the formed signal characteristics are more stable and effective, so that the final identification accuracy is improved, and experiments prove that the density statistical characteristics are also obtained.

Drawings

Fig. 1 is a schematic diagram of a process for performing quadrature scanning on a signal constellation diagram according to the present invention;

FIG. 2 is a schematic diagram of a neural network classifier selected for use in the present invention;

FIG. 3 is a graph of signal to noise ratio variation with average recognition accuracy for a signal in an embodiment of the present invention;

fig. 4 shows the signal recognition result when the signal-to-noise ratio is 5dB in the embodiment of the present invention.

Detailed Description

The invention is described in detail below with reference to the drawings and examples so that those skilled in the art can better understand the invention.

Examples

The purpose of this embodiment is to identify digital communication signals of different modulation modes and verify the identification accuracy. The data in this embodiment is derived from an actual wireless communication signal, the signal transmission rate is 2M Baud, and the data includes five digital modulation modes { BPSK, QPSK,8PSK,16QAM,64QAM }. The signal receiver is responsible for receiving signals, and the signal-to-noise ratio of the obtained baseband signals is about 35dB through a series of signal preprocessing processes. Each sampling obtains 8192 code element symbols, namely m is 8192, a code element sequence S is formed, the width w of a density statistical window is set to be 0.15, the scanning length n is set to be 256, orthogonal scanning is carried out on a constellation diagram depicted by S, and a signal characteristic d is obtained_xyThis operation is repeated 2 ten thousand times for each modulation scheme signal to obtain a total of 10 ten thousand signal feature samples, and then a data set D { (D) is constructed_xyT) }, where t is 1, 2, …, 5, which corresponds to the five modulation schemes described above. The neural network classifier is then trained with data set D. In order to test the performance of the method provided by the invention under a low signal-to-noise ratio, Gaussian white noise with different intensities is directly added to a baseband signal, so that the signal-to-noise ratio is changed between 0 and 9dB, and under each signal-to-noise ratio, the average identification accuracy is obtained by adopting cross validation, and the result is shown in fig. 3, wherein the specific identification condition when the signal-to-noise ratio is 4dB is shown in fig. 4.

Claims

1. A modulation identification method based on orthogonal scanning characteristics of a constellation diagram is characterized by comprising the following steps:

s1, acquiring signal data:

assuming that there are v types of signal modulation schemes to be identified, and each modulation scheme is represented by a category label T, representing a signal set to be identified as T ═ { T | T ═ 1, 2, …, v };

if m code elements are collected each time, the symbol sequence is obtained by collection

Sequence element (x)_i，y_i) Is a symbol s_iCoordinates on a constellation diagram;

s2, performing orthogonal scanning on the constellation diagram:

setting the width w and the scanning length n of the density statistical window, and starting scanning from the upper left corner of the constellation diagram, namely setting the scanning starting position as

Scanning in the Y-axis direction

The specific scanning process is as follows:

wherein, the function count_x(S, x, x ') is used for searching symbols which satisfy the condition that the abscissa is more than or equal to x and less than x' in the symbol sequence S, and returning the number of symbol sampling points which satisfy the condition; function count_y(S, y, y ') is used for searching symbols satisfying that the ordinate is less than or equal to y and is greater than y' in the symbol sequence S;

the signal characteristic obtained by performing orthogonal scanning on the constellation diagram is d_xy＝(d_x，d_y)；

S3, training a neural network classifier:

repeating the steps S1, S2 for the preset signal setFor each modulation mode in the modulation scheme, a large number of signal characteristic samples d are obtained_xyThen, a training dataset D { (D) is constructed_xyT), simultaneously building a neural network, and training the neural network according to the data set;

s4, identifying communication signals:

signal feature d is extracted from the digital communication signal of unknown modulation scheme by the signal feature extraction method in step S2_xyThen, the neural network classifier trained in step S3 is input to obtain a recognition result t.