Comparative Performance Investigation of MIMO-OTFS and MIMO-OFDM using Deep Neural Network Modeling

We consider two popular wireless physical layers (PHYs), namely, multi-input multi-output (MIMO)-orthogonal frequency division multiplexing (OFDM) and MIMO-orthogonal time frequency and space (OTFS) modulation. For them, we develop and implement a deep neural network (DNN) model for fading channel estimation and investigate the mean square error (MSE) performance. While MIMO offers spatial multiplexing to enhance data rates, OFDM provides diversity to combat intersymbol interference. However, OFDM has limitations in high Doppler environments. The OTFS modulation, a newly developed two-dimensional modulation scheme, overcomes OFDM’s limitations. The OTFS modulation technique is helpful in high Doppler fading channels and offers several advantages. In the high Doppler scenario, we find that the OTFS outperforms the OFDM modulation scheme. The added benefit of OTFS over OFDM lies in exploiting channel diversity better in both the time domain and frequency domain when we employ the technique with appropriate equalizers. We show via simulations that DL-enabled MIMO-OTFS DNN exhibits a minimum MSE of value 0.4246, which is less than that of MIMO-OFDM DNN, which is 0.5355 for fading scenarios with high Doppler. Furthermore, we compare our DNN model with the existing linear interpolation technique for channel estimation in MIMO-OTFS at high Doppler.