Generative channel estimation for intelligent reflecting surface-aided wireless communication

Abstract

Intelligent reflecting surface (IRS) has emerged as a viable technology to enhance the spectral efficiency of wireless communication systems by intelligently controlling wireless signal propagation. In wireless communication governed by the IRS, the acquisition of channel state information (CSI) is essential for designing the optimal beamforming. However, acquiring the CSI is difficult as the IRS does not have radio frequency chains to transmit/receive signals and the capability to process the signals is also limited. The cascaded channel linking the base station (BS) and a user through the IRS does not necessarily adhere to a specific channel distribution. Conventional and deep learning-based techniques for channel estimation face challenges: the pilot overhead and compromised estimation accuracy due to assumptions of prior channel distribution and noisy signal. To overcome these issues a novel generative cascaded channel estimation (GCCE) model based on a generative adversarial network (GAN) is proposed to estimate the cascaded channel. The GGCE model reduces the reliance on pilot signals, effectively minimizing pilot overhead, by deriving CSI from received signal data. To enhance the estimation accuracy, the channel correlation information is provided as a conditioning factor for the GCCE model. Additionally, a denoising network is integrated into the GCCE framework to effectively remove noise from the received signal. These integrations collectively enhance the estimation accuracy of the GCCE model compared to the initial GAN setup. Experimental results illustrate the superiority of the proposed GCCE model over conventional and deep learning techniques when provided with the same pilot count.