Performance evaluation of AI-based CSI feedback schemes compliant with 3GPP standards

Abstract

This paper addresses the limitations of traditional CSI feedback schemes in 5G massive MIMO-OFDM systems, particularly the challenges of high feedback overhead and computational complexity. To overcome these issues, this paper proposes and presents an in-depth evaluation of “M-CsiNet”, an AI-based feedback channel state information compression and reconstruction model adhering to 3GPP standards using the CDL-C MIMO channel model. The innovation of “M-CsiNet” lies in the extension of CsiNet to “M-CsiNet” and conducting an in-depth evaluation. Unlike legacy methods such as Type-II and Enhanced Type-II codebook-based schemes, “M-CsiNet” demonstrates significant improvements. Experimental results demonstrate that “M-CsiNet” achieves up to 10-15 dB SNR gain in link-level block error rate (BLER) and throughput performance while reducing feedback overhead by two orders of magnitude and with reduced complexity. These advantages make “M-CsiNet” a promising solution for practical deployment in capacity-constrained 5G and future wireless systems across both rank-1 and rank-2 transmission scenarios.