An efficient soft demapper for APSK signals using extreme learning machine

In this paper we introduce an efficient soft demapping method for amplitude phase shift keying (APSK) constellations using extreme learning machine (ELM). The ELM algorithm is used as a classification framework to perform correct symbol detection in additive white Gaussian noise channels. Despite its spectral efficiency, the demapping process of APSK-like high-order modulation schemes is a computationally complex task. The proposed algorithm can alternatively be used as a neural soft demapper and easily be adapted in conventional receivers. The proposed algorithm has been tested with uncoded data and also applied to coded data that combines iterative decoding process to achieve error-free transmission. In this mean, the study contains both uncoded and coded modulation scenarios and employs two types of networks parameters, namely fixed-type and mixed-type parameters, which has been obtained from fixed-type and mixed-type SNR data sets. The validity of the proposed method has been verified by comparing its symbol- and bit-error rate performance with the well-known max-log-MAP algorithm.