Deep learning based channel equalization method for wireless UV MIMO scattering turbulence channel

Aiming at the problems of traditional equalization methods for ultraviolet (UV) multiple-input multiple-output (MIMO) channels in turbulent environments, such as their strong dependence on a priori knowledge of the channel and the low accuracy in coping with the modeling of complex nonlinear channels, this paper proposes a deep-learning-based equalization method for wireless UV-scattering MIMO channels. The method transforms the MIMO signal into a two-dimensional time series, takes the bidirectional long short-term memory (BiLSTM) with bidirectional sequence feature extraction capability as the core, and supplements it with deep neural network for nonlinear modeling to construct a deep learning network model suitable for UV MIMO channel equalization, so as to realize the accurate recovery of the original MIMO signal. Simulation results show that the scheme exhibits stronger BER and MSE performance compared with the least mean square(LMS) algorithm, recursive least squares(LMS) algorithm, and the equalization scheme based on multilayer long and short-term memory(multiLSTM). At SNR of 9 dB, the scheme reduces the BER by about 67.9%, compared with the equalization scheme based on multi-LSTM, and has stable equalization effects in turbulence environments with different intensities.