Multi-Task Learning-Based Channel Estimation for RIS Assisted Multi-User Communication Systems

TL;DR: This paper proposes a WGAN-GP model to generate realistic fake channel data for MIMO communication systems, leveraging AI to overcome the high cost of real channel data acquisition and improve wireless channel modeling and generation.

TL;DR: A joint cascaded channel estimation scheme is proposed for STAR-RIS systems with hardware imperfections, leveraging multi-task learning to estimate near- and far-field channels with spatial non-stationarity, achieving superior accuracy with reduced training overhead.

TL;DR: In this article , the authors investigated the computation performance of an RIS-aided MEC system, where an access point services multi-MEC devices by utilizing two distributed RISs to operate a partial offloading strategy.

Abstract: Millimeter wave communications are sensitive to complex scattering environments (e.g., to the presence of obstacles), which can be mitigated by reconfigurable intelligent surfaces (RISs). Traditional nearly passive RISs lack the ability to perform signal processing operations, which makes channel estimation in RIS-aided communications more challenging. Hence, in this paper, we focus on a hybrid RIS architecture, which is equipped with a small number of active elements. These active elements can be connected with baseband processing units through radio frequency (RF) chains. We estimate the whole channel, including the channel between the base station (BS) and the RIS, that between the users and the RIS, and that between the BS and the users. The whole channel is acquired at the hybrid RIS through the transmission of several segments of training pilots. In order to decrease the hardware cost, the BS and the hybrid RIS are equipped with RF chains with low-resolution analog-to-digital converters (ADCs). Since the numbers of BS antennas and RIS elements are very large, the estimation of the full-space channels is not straightforward. To tackle this problem, we propose a channel extrapolation scheme based on a joint selection model. Specifically, we select a BS antenna subset and an RIS element subset to be connected to the RF chains and to estimate the partial-space channels related to these subsets. Then, a reference-based variational auto-encoder model is used to implement the extrapolation from the partial-space channels to the full-space channels. Besides, the optimal joint selection pattern is acquired through a selection network to improve the channel extrapolation performance. Moreover, to overcome the quantization error caused by the use of low-resolution ADCs, we propose a two-stage repair scheme for channel estimation. Simulation results are provided to demonstrate the effectiveness of the designed channel extrapolation scheme.

TL;DR: This article seeks to help ML practitioners apply MTL by shedding light on how MTL works and providing guidelines for choosing appropriate auxiliary tasks, particularly in deep neural networks.

TL;DR: In this paper, the authors provide a detailed overview and historical perspective on state-of-the-art solutions, and elaborate on the fundamental differences with other technologies, the most important open research issues to tackle, and the reasons why the use of reconfigurable intelligent surfaces necessitates to rethink the communication-theoretic models currently employed in wireless networks.

TL;DR: In this article, the authors make the observation that the performance of multi-task learning is strongly dependent on the relative weighting between each task's loss, and propose a principled approach to weight multiple loss functions by considering the homoscedastic uncertainty of each task.

TL;DR: New deep learning methods, namely, deep residual shrinkage networks, are developed to improve the feature learning ability from highly noised vibration signals and achieve a high fault diagnosing accuracy.