Reconfigurable intelligent surface (RIS) is a promising technology for future wireless communication systems. Channel estimation (CE) of RIS device is a critical but also challenging issue for its development. The mainstream of existing CE methods is confined to the so-called cascaded channel (CscdChn) estimation scheme, which treats the multiplicative two-hop RIS channels as an effective one and measures it as a whole. This CscdChn training method suffers from severe double-fading attenuation loss, which significantly degrades the CE accuracy. In this paper, we propose a novel RIS-transmitting (RIS-TX) based CE scheme, which has lower pilot overhead than CscdChn scheme and effectively overcomes the double-fading curse via incorporating only one single transmit radio frequency (RF)-chain into RIS. We develop highly efficient gradient descent (GD) and penalty duality decomposition (PDD)-based solutions to resolve the pilot design task for the RIS-TX CE scheme, which is a difficult quartic optimization problem. Our designed pilot signal outperforms the discrete Fourier transform (DFT) sequence, which is reported to be optimal for CscdChn scheme. Besides, both theoretical analysis and numerical results demonstrate that our proposed RIS-TX scheme exhibits distinct performance characteristics as opposed to its CscdChn counterpart and yields superior accuracy when RIS device is not extremely large.

翻译：可重构智能表面（RIS）是未来无线通信系统的一项有前景技术。RIS装置的信道估计（CE）是其发展过程中关键且具有挑战性的问题。现有CE方法的主流局限于所谓的级联信道（CscdChn）估计方案，该方案将乘性的两跳RIS信道视为一个等效信道并进行整体测量。这种CscdChn训练方法遭受严重的双重衰落衰减损失，显著降低了CE精度。本文提出一种基于RIS发射（RIS-TX）的新型CE方案，其导频开销低于CscdChn方案，并通过在RIS中仅纳入单个发射射频（RF）链有效克服了双重衰落难题。针对RIS-TX CE方案的导频设计任务（一个困难的四次优化问题），我们开发了基于梯度下降（GD）和惩罚对偶分解（PDD）的高效解法。我们所设计的导频信号优于离散傅里叶变换（DFT）序列——后者被报道为CscdChn方案的最优序列。此外，理论分析与数值结果均表明，与CscdChn方案相比，所提RIS-TX方案展现出截然不同的性能特征，并在RIS装置并非极端庞大时实现更高的精度。