Robust WMMSE-Based Precoder With Practice-Oriented Design for Massive MU-MIMO

The challenge induced by imperfect channel state information (CSI) at the transmitter promotes the research of robust precoder design, among which stochastic weighted minimum mean square error (SWMMSE) has gained wide attention due to its excellent performance. However, its considerable computationa...

Full description

Saved in:
Bibliographic Details
Published in:IEEE wireless communications letters Vol. 13; no. 7; pp. 1858 - 1862
Main Authors: Wang, Kexuan, Liu, An
Format: Journal Article
Language:English
Published: Piscataway IEEE 01.07.2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Algorithms
Approximation
Approximation algorithms
Complexity
Complexity theory
Convergence
deep unfolding
Iterative algorithms
linear precoder
Mean square error methods
Mean square errors
Neural networks
Precoding
Robust WMMSE
Robustness
ISSN:2162-2337, 2162-2345
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The challenge induced by imperfect channel state information (CSI) at the transmitter promotes the research of robust precoder design, among which stochastic weighted minimum mean square error (SWMMSE) has gained wide attention due to its excellent performance. However, its considerable computational complexity and extremely slow convergence rate make it impractical to be applied in real massive multi-user multiple-input (MU-MIMO) systems. To combat these drawbacks, in this letter, we first propose a robust weighted minimum mean square error (WMMSE)-based precoder with a practice-oriented design, namely PO-WMMSE. On one hand, it can be approximately seen as a deterministic equivalent of the classical SWMMSE, and thus yield a satisfying performance. On the other hand, due to the low complexity and closed-form approximation of the expectation terms in PO-WMMSE, it can quickly converge with linear complexity (in contrast, the expectation terms are approximated using sample average in SWMMSE). Then, the proposed algorithm is unfolded into a layer-wise neural network, namely PO-WMMSE Net, in which several trainable matrices are induced to compensate for the approximate loss and further accelerate convergence. Finally, numerical comparisons under imperfect CSI with existing algorithms demonstrate the significant advantages of the developed PO-WMMSE Net.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2162-2337
2162-2345
DOI:10.1109/LWC.2024.3393484