Deep Learning for OTFS Transceiver Design

Summary

Deep neural networks (DNNs) have found applications in a wide range of fields. They are being increasingly used for inference tasks like regression and classification. This chapter focuses on introducing and describing deep learning techniques that find application in the design of efficient orthogonal time frequency space (OTFS) transceivers. It presents a short primer on the learning framework and DNNs and introduces popular NN architectures such as fully connected neural networks, recurrent neural networks, and convolutional neural networks. The chapter focuses on which architecture will be appropriate for which communication task. It also presents DNN‐based solutions to various OTFS transceiver tasks such as signal detection in the DD domain, channel estimation and tracking in the delay‐Doppler domain, and IQ imbalance estimation and compensation. IQ imbalance occurs due to imperfect match/balance between in‐phase (I) and quadrature phase (Q) arms in the radio frequency chain because of which the orthogonality between I and Q arms is compromised.