Graphic Constellations and DBN Based Automatic Modulation Classification

Abstract

In this paper, we propose a low-complexity graphic constellation projection (GCP) algorithm for automatic modulation classification (AMC), where the recovered symbols are projected into artificial graphic constellations. Unlike the existing feature- based (FB) algorithms, we convert the AMC problem into an image recognition problem. Subsequently, the deep belief network (DBN) is adopted to learn the underlying features in these constellations and recognize their corresponding modulation schemes. Simulation results demonstrate that the proposed GCP-DBN based AMC system achieves better performance than other schemes. Specifically, the classification accuracy is beyond 95% at 0 dB, which is very close to the average likelihood ratio test upper bound (ALRT-Upper Bound).