A Quantization-Aware DL-Based Channel Estimation Algorithm for OFDM Systems

Abstract

This paper proposes a quantization-aware deep learning (DL)-based channel estimation algorithm for orthogonal frequency-division multiplexing (OFDM) systems under varying effective number of bits (ENOB) configurations. The algorithm addresses two key aspects: generalization in both channel conditions and analog-to-digital converter (ADC) resolutions, and mitigation of quantization noise. Generalization is achieved during the training phase by utilizing a dataset that includes multiple channel realizations across different ADC resolutions. Regarding quantization noise mitigation, the algorithm uses medium- to high- resolution data as target labels during offline training to learn the corresponding ENOBs indirect mapping. This approach enables improved channel estimation accuracy and enhances end-to-end system performance in terms of mean square error (MSE) and bit error rate (BER). Test results demonstrate consistent improvements both locally and globally across ENOBs configurations, with validation conducted in an indoor over-the-air (OTA) scenario to confirm real-world applicability. To the best of the authors’ knowledge, this is the first work to address the mitigation of ADC quantization noise in the channel estimation process without relying on medium- to high- resolution data during inference, while ensuring algorithm generalization across multiple channel realizations and varying ENOB configurations.