FSIM-Net: Neural Network-Based Symbol Detector for Filter Shape Index Modulation

Filter shape index modulation (FSIM) is a promising approach that unlike most existing index modulation schemes does not sacrifice resource utilization by deactivating available resources but rather uses pulse shaping filters as indexing parameters. This principle is the key to enhance spectral efficiency and/or energy efficiency. However, the filter bank design relaxes the Nyquist criterion for zero intersymbol interference (ISI) to have more distinguishable filters and the inherent controlled ISI can be managed at the receiver side. This letter presents FSIM-Net, a simple neural network (NN)-based detector that enables joint detection of filter indices and QAM/PSK symbols in the presence of filters’ inherent ISI. FSIM-Net is trained offline in an additive white Gaussian noise (AWGN) channel and tested under various conditions by decoupling of inherent ISI and channel effect handling. The proposed receiver demonstrates superior performance especially in more realistic scenarios while simplifying the FSIM receiver architecture and avoiding potential error propagation compared to the conventional one.