Learning based Delay-Doppler Channel Estimation with Interleaved Pilots in OTFS

Traditionally, channel estimation in orthogonal time frequency space (OTFS) is carried out in the delay-Doppler (DD) domain by placing pilot symbols surrounded by guard bins in the DD grid. This results in reduced spectral efficiency as the guard bins do not carry information. In the absence of guard bins, there is leakage from pilot symbols to data symbols and vice versa. Therefore, in this paper, we consider an interleaved pilot (IP) placement scheme with a lattice-type arrangement (which does not have guard bins) and propose a deep learning architecture using recurrent neural networks (referred to as IPNet) for efficient estimation of DD domain channel state information. The proposed IPNet is trained to overcome the effects of leakage from data symbols and provide channel estimates with good accuracy (e.g., the proposed scheme achieves a normalized mean square error of about 0.01 at a pilot SNR of 25 dB). Our simulation results also show that the proposed IPNet architecture achieves good bit error performance while being spectrally efficient. For example, the proposed scheme uses 12 overhead bins (12 pilot bins and no guard bins) for channel estimation in a considered frame while the embedded pilot scheme uses 25 overhead bins (1 pilot bin and 24 guard bins).