Automatic modulation classification for cognitive radio systems: Results for the symbol and waveform domains

This work presents two approaches to the problem of automatic modulation classification in cognitive radios. The first approach deals with classifying distinct QAM modulations. The proposed solution consists of a SVM classifier that uses as input parameters the distance between the received symbol and its nearest neighbor in each constellation. The results obtained by the proposed classifier is compared to a state-of-art technique. The comparison show a good performance for the 16-QAM, 32-QAM and 64-QAM modulation schemes. In the second approach, the cyclic spectral analysis is adopted as the method for extracting the signals features, and the SVM classifier is used for the pattern recognition stage. The proposed method achieved excellent results for the AM, BPSK, QPSK and BFSK modulation schemes. In both approaches, different scenarios are evaluated which may serve as a starting point for researchers who want to compare results systematically.

I. Introduction

The automatic modulation classification (AMC) problem consists in identifying the modulation scheme of a given communication system with a high probability of success and in a short period of time. Recent overviews about the subject can be found in [1]–[3]. Historically, AMC has been used for decades in military applications. More recently, the interest for ACM has been renewed by the research on the cognitive radio area. Cognitive radios (CG) [4] are intelligent wireless communication systems capable of modifying their transmission parameters such as: operation frequency, modulation scheme, transmitted power, communication protocol and so on, based on the interactions with their environment.