A Research Of Modulation Recognition Based On Efficient Attribute Reduction With Neighborhood Rough Set And Neural Network

Modulation recognition of digital communication signal has always been a research focus for countries, which has been more widely used in both military and civilian communication fields. The purpose of it is to identify the correct modulation type of communication signals and provide the signal information that further signals processing required.After studying many of documents about modulation recognition at home and abroad, also owing to the fact that the theory of Rough Set possesses powerful data mining capabilities and Artificial Neural Network which is well known for its superior fault tolerance, an algorithm that combines efficient attribute reduction with Neighborhood Rough Set and the Back-Propagation Network (BP) is given in this paper, and the main works can be summarized as follows:1. Firstly, the basic principles of common communication signals of digital modulation is analyzed on theories, and the signals’instantaneous amplitude, frequency and phase are simulated by MATLAB in my paper.2. Secondly, the theory of high-order cumulants is introduced in the paper. According to the seven kinds of modulation signals researched in this paper, a group of feature parameters (γmax、σap、σap、σaa、σaf、κ1、κ2) based on the time-frequency characteristics and high-order cumulants of communication signals selected by algorithm. Experimental results showed this group of parameters could identify these modulation signals effectively. 3. The theory of Neighborhood Rough Set and an optimized method used in attribute reduction based on Neighborhood Rough Set is introduced in detail. The method uses the property that positive region increases with the amount of attributes, in order to decrease comparison times and improve computational efficiency. Efficient attribute reduction with Neighborhood Rough Set is applied to the research of modulation signals recognition for the first time, and the experiments show that the algorithm is effective.4. Using Back-Propagation Network as classification instruments, single recognition rate and the average recognition rate of2ASK、2FSK、2PSK、4ASK、4FSK、4PSK、16QAM are simulated, which are collected factually, proved the algorithm used in this paper has better recognition effect compared to other methods.