LSTM-Based Time-Frequency Domain Channel Estimation for OTFS Modulation

Orthogonal Time Frequency Space (OTFS) is a promising modulation scheme that works in the delay-Doppler (DD) domain, offering resistance to frequency selective fading and time-varying channels. Thus, OTFS channel estimation assumes great significance for successful transmission. Typically, it requires allocating pilots in the DD domain, which often results in an increase in the peak-to-average power ratio (PAPR) and high complexity to detect the received signal. In response to these issues, we present a solution that estimates the channel in the time-frequency (TF) domain. In addition, although several works in the literature present solutions for OTFS channel estimation, few consider the presence of high-power amplifiers (HPAs) and explain the impact of nonlinear effects on channel estimation and system performance. Starting from channel estimation based on preambles and pilots in the TF domain, we present a solution capable of obtaining reliable channel estimation using a long short-term memory (LSTM) network in highly selective channel conditions, effectively compensating for nonlinearities in signal reception. Our results validate the effectiveness of the proposed solution, highlighting its potential to improve the performance and robustness of OTFS communication systems in real scenarios with nonlinear HPA effects.