Deep Learning-Based Soft Iterative-Detection of Channel-Coded Compressed Sensing-Aided Multi-Dimensional Index Modulation

The concept of Index Modulation (IM) has been actively researched as a benefit of its flexible trade-off between performance, achievable rate, energy efficiency, hardware cost and complexity. In order to fully exploit its the degrees of freedom, the concept of Multi-dimensional IM (MIM) has been developed in literature, where Compressed Sensing (CS) is often utilized to exploit the sparsity of the multi-dimensional transmitted signals. In this paper, we propose Deep Learning (DL) based detection for CS-aided MIM CS-MIM, where both Hard-Decision (HD) as well as Soft-Decision (SD) detection combined with iterative decoding are conceived. More explicitly, firstly, we propose learning aided hard and soft detection for CS-MIM. Secondly, two novel neural network aided methods are proposed for Iterative Soft Detection (ISD), where iterations are carried out between the CS-MIM detector and a channel decoder. The proposed learning-aided methods are capable of eliminating the overhead and complexity of Channel Estimation (CE), which results in an improved transmission rate. Explicitly, we develop an advanced DL architecture for blind-detection-aided MIM for the first time in the open literature, where the HD and SD CS algorithms are implemented by learning without the need for CE. Our simulation results demonstrate that the proposed learning techniques conceived for SD CS-MIM combined with iterative detection can achieve near-capacity performance at a reduced complexity compared to the conventional model-based SD relying on Channel State Information (CSI) acquisition.