Fast Detection Based on Customized Complex Valued Convolutional Neural Network for Generalized Spatial Modulation Systems

In this paper, a customized Auto-Encoder Complex Valued Convolutional Neural Network (AE-CV-CNN) that has been developed in a prior work is applied to Single Symbol Generalized Spatial Modulation (SS-GSM) scheme with new extracted features. The achieved reductions in the computational complexity at the receiver is at least 63.64% for M-PSK schemes compared to the complexity of Maximum Likelihood (ML) detection algorithm. This Fast detection algorithm is based on a proposed Low Complexity ML (LC-ML) detector that affords a complexity reduction of at least 40.91%. With these proposed algorithms, the complexity is reduced as the spatial constellation size increases. Furthermore, in comparison to other sub optimal detection algorithms, the computational complexity in terms of real valued multiplications of the AE-CV-CNN applied to LC-ML is independent of the spatial spectrum efficiency which means that the total spectrum efficiency increases with larger spatial constellation size at no additional complexity.