Low-Complexity Cross-Domain Iterative Detection Based on Local LMMSE Filters for OTFS

Cross-domain iterative signal detection is known as a promising demodulation technique for orthogonal time frequency space (OTFS), where signal estimation is performed in both the time-delay (TD) and delay-Doppler (DD) domains, and extrinsic information according to their soft estimates is iteratively exchanged via unitary transforms between them, thereby reaping the benefits of both domains. However, its high computational nature due to the large dimension of TD-domain channel hinders practical use. Although a dimensionality reduction method according to its sparse structure has been investigated, inversion operations of a matrix with a size corresponding to the number of sub-carriers are still required. To circumvent this computational burden, we propose a novel local filtering strategy that divides the TD-domain channel into smaller overlapping sub-channels of size dependent on the maximum delay and performs linear minimum mean square error (LMMSE) detection on each of them. By combining the posterior estimates of the local filters to account for overlap between sub-channels, the performance degradation due to dimensionality reduction can be suppressed. Simulation results show that the proposed method can achieve asymptotic performance approaching that of the state-of-the-art cross-domain schemes with much less computational effort.