Signal Modulation Recognition Based On The Improved Self-organizing Feature Map Neural Network

The radio signal modulation and demodulation and modulation recognition technologyhas been a major research topic as an important part of the radio communications field. Theautomatic modulation recognition of the radio signal is that identifying modulation schemesof the signal by the analysis and processing with the absence of a priori knowledge or a littlepriori knowledge and the presence of noise and interference in the channel，which provides abasis for the further processing of the signal. However, the radio communication environmentis getting complex, and signal modulation type and communication institution type increasediversely. This makes the existing recognition theory and methods difficult to effectivelyidentify radio communication signals.Radio monitors observe and analyze the signal spectrum in order to determine themodulation type, which requires very professional skills for radio monitors. The process ofmonitoring and recognition is that radio monitors match the signal spectrum pattern with theprior experience and identify them by utilizing their professional knowledge.Neural network not only has a strong pattern recognition capabilities which can betterhandle the complex non-linear problems, but also has better stability and mistake tolerance, soit has been widely used in the modulation recognition. This paper studies the modulationrecognition algorithm based on the self-organizing competitive neural network and we givesome corresponding improvement in order to improve the network training speed andrecognition accuracy. The main contents are as follows:Firstly, in order to eliminate the impact of environmental noise on the original signalspectrum in modulation identification process, the noise signal is filtered by using fuzzyC-means algorithm (FCM) in this paper. The modulation recognition feature parameters basedon the spectral information are extracted for AM, VSB, FM, FSK modulation signal from theradio monitoring.Secondly, based on a study of the structure and learning algorithm of SOM neuralnetwork, an optimal learning algorithm of SOM network is proposed with FCM clusteringalgorithm. The simulation test has shown that the improved network model can improve thetraining speed and recognition rate.