Sub-Nyquist tensor completion for massive MIMO channel estimation: Part 2
Semyon Dorokhin, D.V. Shuvalov, Vladimir A. Lyashev · Journal of Radio Electronics · 2025
5G trend towards larger antenna arrays, higher bandwidth and subcarrier number continues towards 6G. As the size of the channel tensor grow, so does the channel estimation overhead. For conventional methods based on regular pilots and interpolation further overhead reduction is prohibited by the Nyquist limit and delay-domain aliasing. In the first part of our work, we demonstrated that slab sampled tensor completion methods are immune to aliasing effect. The minimal number of slabs is limited not by the delay spread of the channel, but by the channel tensor ranks. Despite good theoretical performance of the algorithms proposed in the previous part, implementation of slab sampling in MIMO systems and its impact on the performance of the completion algorithms is an equally important question. In this article we propose Quadratic Permutation Polynomial (QPP)-based sampling and Simultaneous OMP for wideband slabs estimation. We demonstrate that if this approach is used to estimate the entire tensor, the overhead can be reduced twice without spectral efficiency degradation. Moreover, at 0-5 dB SNR this method yields 10-50 % spectral efficiency increase. SOMP can also be used for wideband slabs estimation and further Tucker-based completion with QPP-interleaved frequency-sparse slabs. This approach yields 4 times overhead reduction compared to conventional method and 12.5 % spectral efficiency increase compared to pure SOMP. To avoid frequency-sparse slabs interleaving for practical reasons, one may switch to Tensor Train model and use regular slabs. This, however, yields smaller overhead reduction (only 3.43 times) and 7.5-15 % spectral efficiency degradation compared to QPP-interleaved Tucker completion.