MIMO-Zak-OTFS with Superimposed Spread Pilots
Abhishek Bairwa, Ananthanarayanan Chockalingam
TL;DR
The paper tackles channel estimation for MIMO-Zak-OTFS with superimposed spread pilots by designing spread pilots that separate in the cross-ambiguity domain and employing turbo iterations between channel estimation and data detection. It introduces a cross-ambiguity-domain pilot separation scheme based on rotated and shifted lattices $\Lambda_{q_j,\text{shift}}$ so that end-to-end channel taps $h_{\text{eff},ij}[k,l]$ can be read off from cross-ambiguities $A_{y_i,x_{s,j}}[k,l]$, followed by pilot cancellation and iterative MMSE-LAS-based data detection. Complexity is dominated by an MMSE inversion with $\mathcal{O}(n_{\mathrm{itr}}(n_r MN)^3)$ operations. In simulations for $2\times2$ and $3\times3$ MIMO, using Gaussian-sinc pulse shaping in a Vehicular-A channel, the proposed method with three turbo iterations approaches the performance of perfect CSI, highlighting its potential for high-mobility, wideband scenarios and throughput-preserving pilot design.
Abstract
In this paper, we consider the problem of spread pilot design and effective channel estimation in multiple-input multiple-output Zak-OTFS (MIMO-Zak-OTFS) with superimposed spread pilots, where data and spread pilot signals are superimposed in the same frame. To achieve good estimation performance in a MIMO setting, the spread pilots at different transmit antennas need to be effectively separated at the receiver. Towards this, we propose a spread pilot design that separates the pilot sequences in the cross-ambiguity domain and enables the estimation of the effective channel taps by a simple read-off operation. To further alleviate the effect of pilot-data interference on performance, we carry out turbo iterations between channel estimation and detection. Simulation results for $2\times 2$ and $3\times 3$ MIMO-Zak-OTFS with Gaussian-sinc pulse shaping filter for vehicular-A channel model show that the proposed pilot design and estimation scheme with three turbo iterations can achieve very good estimation/detection performance.