Probabilistic Shaping in MIMO: Going Beyond 1.53dB AWGN Gain With the Non-Linear Demapper
Kirill Ivanov, Wei Yang, Jing Jiang
TL;DR
This work shows that probabilistic amplitude shaping (PS), when combined with a non-linear MIMO demapper such as sphere decoding, can exceed the AWGN-shaped gain bound of $1.53$ dB in multi-antenna channels. By integrating MB-distributed amplitudes via CCDM and incorporating a priori symbol probabilities into the sphere-decoding metric, the authors demonstrate over $2$ dB gains, rising to about $3.5$ dB under moderate antenna correlation for a $4\times4$ MIMO with $256$-QAM. PS also reduces the average number of nodes visited in the decoding tree at low correlation, indicating potential complexity savings. The results highlight the crucial role of non-linear demappers in exploiting shaped interference and motivate further theoretical and practical exploration of shaping in multi-antenna systems, including potential relevance for future networks like 6G.
Abstract
Constellation shaping is a well-established method to improve upon a regular quadrature amplitude modulation (QAM). It is known that the gain achieved by any shaping method for an additive white Gaussian noise (AWGN) channel is upper-bounded by 1.53dB. However, the situation becomes less clear in the multiple-input and multiple-output (MIMO) setting. In this paper, we study the application of probabilistic shaping for MIMO channels. We utilize an efficient near-optimal demapper based on sphere decoding (SD) and demonstrate that it is possible to achieve more than 2dB gains, breaking the AWGN limit. It becomes possible because both signal and interference are shaped and the non-linear methods can capture this property and leverage on it to improve the demodulation performance.