Polarization-Space Modulation With Reconfigurable Intelligent Surfaces, Robust Estimation, and Quaternion Neural Networks

Contemporary communications systems use large arrays in order to exploit the spatial domain requiring multiple radio-frequency (RF) chains leading to prohibitive cost and power consumption. Spatial degrees of freedom are also achieved by utilizing the polarization states of an electromagnetic (EM) wave. To this end, we propose a polarization-state modulation system based on carrier-wave reflections from a reconfigurable intelligent surface (RIS), where a sequence of data bits is mapped to polarization states. We consider a system with an upper millimeter-wave or low-terahertz (THz) RF source so that the channel model is line-of-sight-dominant. The data symbol constellation consists of 16 quaternion-valued symbols, of which two correspond to linear polarizations and the remaining 12 represent elliptical polarizations. Our channel models consist of both additive white Gaussian noise (AWGN) and impulse noise. We utilize the Weiszfeld algorithm and generalized M-estimators (GM-estimators) to handle the impulse noise and robustly decode the polarization-space modulated (PSM) signals. Furthermore, we train and evaluate quaternion neural networks (QNNs) for decoding PSM signals using seven different activation methods. Our numerical experiments indicate that the RIS is capable of directing signal power toward the location of the receiver. The bit error rate performance of our QNNs and robust decoders exceeds that of OFDM-BPSK.