DNN-Based Joint I/O Relation Estimation and Detection in Zak-OTFS

Zak transform-based orthogonal time frequency space (Zak-OTFS), a delay-Doppler (DD) domain modulation, is well-suited for general wireless channels, including channels which are highly doubly-dispersive in nature. Accurate inputoutput (I/O) relation estimation and signal detection at the receiver are crucial for reliable communication. In this paper, we propose a deep neural network (DNN) based Zak-OTFS receiver that jointly performs I/O relation estimation and signal detection, considering an exclusive pilot frame. We focus on two commonly considered DD pulse shaping filters, namely, sinc and Gaussian filters, to demonstrate the effectiveness of the DNN approach. A fully-connected DNN is trained to jointly learn the DD channel within a spatial coherence period and optimize the detection performance. Our simulation results demonstrate that the proposed DNN approach outperforms the conventional approach where model-free I/O relation estimation and minimum mean square error detection are performed separately.