Learning to Estimate: A Real-Time Online Learning Framework for MIMO-OFDM Channel Estimation

In this paper, we introduce StructNet-CE, a novel real-time online learning framework for MIMO-OFDM channel estimation, which only utilizes over-the-air (OTA) reference signals (RS) for online channel estimation on a slot basis without assuming the availability of any channel knowledge. To achieve real-time and efficient channel learning, the design of StructNet-CE leverages the structural information inherent in the MIMO-OFDM system: the repetitive structure of modulation constellation and the invariant property of symbol classification to inter-stream interference. The embedded structural information enables StructNet-CE to conduct channel estimation through the underlying symbol detection task and accurately learn MIMO channels through the limited RS with the scattered RS configuration adopted in 5G/5G-Advanced slots. Numerical experiments demonstrate that the channel estimation performance is significantly improved by incorporating the structural knowledge, achieving a mean square error (MSE) reduction ranging from around 44.41% to 95.54% compared to existing methods. Furthermore, StructNet-CE is compatible and readily applicable to current and future wireless networks, demonstrating the effectiveness, importance, and relevance of combining machine learning techniques with domain knowledge for wireless systems.