Efficient Likelihood Function Learning Method for Time-Varying MIMO Systems Using One-Bit ADCs

This study proposes a likelihood function (LF) learning method for time-varying multiple-input multiple-output (MIMO) systems using one-bit analog-to-digital converters. In time-varying channels, the initially-estimated LF becomes inconsistent with the true LF owing to channel variation. To mitigate this inconsistency, the LF can be learned by using input-output samples obtained from the MIMO detection. However, the input-output samples inevitably have some uncertainties due to detection errors. To address this challenge, we consider a Markov decision process (MDP) to minimize the mismatch between the true and learned LFs under sample uncertainty. Subsequently, we propose a computationally efficient LF learning method to solve the MDP. In the proposed method, we first simplify an LF update model in the MDP, which efficiently captures temporal channel variations by considering the expectation of the channel model noise. Based on the LF update model, we obtain the optimal policy in a closed-form solution by considering the most probable state transition. Simulation results show that the proposed learning method significantly improves complexity without sacrificing performance.