Quasi Neural Network assisted Channel Estimation for MIMO-OFDM

This paper presents a quasi-neural network (QuasiNN) based parametric channel estimation for multiple inputs and multiple outputs with orthogonal frequency division multiplexing (MIMO-OFDM) systems. This approach jointly considers time delay, multipath gain, direction-of-arrival (DOA), direction-of-departure (DOD), and the number of multipaths. We construct a Quasi-NN that accurately models the signal transmission process via a parametrized channel. By leveraging the layered structure of Quasi-NN, similar to that of artificial neural networks (ANNs), we can directly apply the backpropagation (BP) algorithm. To address parameter initialization in scenarios with clustered multipath distributions, we propose a low-complexity fast Fourier transform (FFT)-based initialization method. Furthermore, we propose a hypothesis testing based method for estimating number of multipaths during the optimization of the Quasi-NN. Simulation results verify that the proposed Quasi-NN assisted parametric channel estimation closely approaches the Cramér Rao Bound (CRB).