ReFLEX: Length-Generalizable CSI Denoising for MIMO-OFDM via Relative-Frequency Bias

2026-05-29

AI Summary

This work addresses the limited generalization capability of channel state information (CSI) denoising in MIMO-OFDM systems under variable resource block (RB) allocations. To overcome this challenge, the authors propose ReFLEX, a length-generalizable Transformer model featuring a novel relative frequency positional bias (RFPB) based on subcarrier offsets. This mechanism enables a frequency-domain attention architecture that allows a single model—without retraining—to effectively denoise CSI from arbitrary RB lengths and sparse DM-RS observations. Evaluated under the 3GPP TR 38.901 UMa NLOS channel model, ReFLEX achieves an NMSE of approximately −9.6 dB on unseen RB lengths and reduces the 10% BLER threshold by 2–3 dB in NR PUSCH/UL-SCH simulations, significantly outperforming conventional methods that rely on fixed input dimensions.

Abstract

This letter studies CSI denoising for MIMO--OFDM with variable NR resource block (RB) allocations. ReFLEX is a length-generalizable Transformer whose frequency attention uses a relative-frequency position bias (RFPB) generated from subcarrier offsets. A single checkpoint handles unseen RB lengths and can be applied to sparse DM-RS observations in the tested RB5/RB10 PUSCH setup without retraining. In a 3GPP~TR~38.901 UMa NLOS channel, ReFLEX achieves about $-9.6$~dB NMSE on unseen RB lengths. In NR PUSCH/UL-SCH simulations, ReFLEX denoising followed by time-frequency interpolation reduces the 10\% BLER threshold by about 2--3~dB.

Problem

• CSI denoising
• MIMO-OFDM
• resource block allocation
• length generalization
• sparse DM-RS

Innovation

• length-generalizable
• relative-frequency position bias
• CSI denoising
• MIMO-OFDM
• Transformer