Extended modulation classification scheme based on approximate entropy

Abstract

This paper focuses on the modulation classification problem based on approximate entropy(ApEn). The original ApEn-based classification procedure is extended to a more comprehensive one. In the new procedure, preprocessing method of complex squaring is proposed to cancel out the negative effect introduced by noise. Simulation results demonstrate the effectiveness of our method to alleviate the demanding for extreme high signal-to-noise ratio(SNR). Specifically,almost 20dB gain can be obtained to achieve a reasonable high classification accuracy(≥ 90%) under the special occasion being studied. Furthermore,hybrid-c value scheme is proposed to ensure relatively good performance over the whole SNR range. In this scheme,two different c values are adopted simultaneously to form four-dimensional ApEn feature vectors for classification step,which combines each c value's good performance over the specific SNR range.