An Untrained DNN Denoiser for Uplink Channel Estimation in Multicell Massive MIMO System

• Published in: 2022 IEEE 33rd Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) pp. 1 - 6
• Main Authors: Bansal, Yatharth, Sah, Abhay Kumar
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12-09-2022
• Subjects: 5G mobile communication; 5G networks; Channel estimation; deep image prior; Massive MIMO; OFDM; pilot contamination; Receiving antennas; Robustness; untrained DNN; Uplink
• ISSN: 2166-9589
• DOI: 10.1109/PIMRC54779.2022.9978046

Massive multiple-input and multiple-output (MIMO) systems are an integral part of 5G cellular networks. They employ a large antenna array at the base station (BS) to serve several users concurrently, which is expected to grow for the 5G and beyond networks. The performance of such systems relies on the quality of channel estimates, and therefore, an accurate channel state information (CSI) is critical for the deployments. However, with the increasing number of users, and antennas at BS, the pilot contamination increases, thus, inducing an extra noise to the received signal. This extra noise leads to an inaccurate and inefficient channel estimate. In this work, we propose a channel estimation technique based upon a specially designed untrained deep neural network (DNN), namely, deep image prior (DIP). We leverage the performance of the untrained DNN model for denoising the received signal and obtaining a superior channel estimation performance. We use normalized mean square error (NMSE) as a metric to corroborate the robustness of the proposed channel estimator against pilot contamination compared to the existing channel estimators.