Communication Signal Automatic Modulation Recognition

The classification of modulation types of communication signals is a problem of typical pattern recognition. It involves many perplexing factors. With rapidly development of communication technology, the system and modulation manner of communication signals becomes more and more complicated and various, and circumstance of signals becomes increasing denseness. It results in that the routine methods and theory of recognition can hardly satisfy practical requirement. So the strict demand has been presented for study on recognition of communication signals. However, the outstanding problem is that when communication signals are transmitted by wireless channel, the variation range of SNR is very large, and it is generally between several dB and several ten dB. The result is that the serious distortion of the same sort feature extracted from the different SNR samples for same type signals is caused. It is equal to increase multiply types of the recognized signals, and the recognition probability of classification is reduced.The main contribution of this dissertation includes four aspects. They are instantaneous parameters extraction, fuzzy feature selection, single classifier design and combined classifier design.The main results are as follows:1. This dissertation mainly studies the instantaneous parameters extraction methods that are built on wavelet ridge, short time Fourier ridge, wavelet transform and adaptive time-frequency analysis respectively. Experimental results illustrate that the proposed four methods have strong anti-noise performance.2. A feature selection method is studied in this dissertation. It is fuzzy feature evaluation method, which adopts fuzzy genetic algorithm to find optimal feature. Experimental result testifies that the proposed method is feasible.3. The designing methods of single classifiers that are based on neural network, fuzzy neural network and fuzzy support vector machine respectively, are studied in detail. Experimental results illustrate that the discussed three single classifiers have good generalization ability.4. Three new methods which are based on fuzzy integral and neural network, fuzzy integral and support vector machine, interclass distance and fuzzy neural network respectively, are used to design combined classifier. Experimental results illustrate that they have high recognition rate with large variation range of SNR.