Reconfigurable intelligent surfaces (RISs) consist of passive elements that can induce controllable phase shifts on the incident signals. This allows the signals to be focused towards the intended user. The access point (AP) tunes the reflection coefficients of the RIS elements as a function of the cascaded channel gain to ensure that the signals combine coherently at the receiver. Hence, accurate channel estimation is crucial to configure the RIS. However, this requires the transmission of sufficiently many pilots. I focus on developing a low-overhead training scheme for RIS-aided systems in this thesis. In doing so, I found that exploiting channel correlation is essential for reducing the training overhead in RIS-aided systems. Consequently, we first proposed a training framework that leverages the spatial correlation of the cascaded channel to reduce the number of pilot symbols required for channel estimation. We proposed a novel low-overhead spatial correlation-aided training (SpCAT) scheme that jointly designed the pilot powers and RIS reflection coefficients during training as a function of the spatial correlation of the cascaded channels. We developed SpCAT for both line-of-sight (LoS) and non-LoS channel models. We then analyzed the uplink achievable rate, which accounted for the training overhead and the impact of noise-induced errors in configuring the RIS and end-to-end channel estimation. This enabled us to determine the optimal number of pilots and the pilot and data power allocation that maximized the achievable rate. Our numerical results showed that SpCAT has a lower mean-squared channel estimation error and a higher achievable rate than several benchmarking schemes, such as reduced subspace-least squares (RS-LS), discrete Fourier transform (DFT) training with and without grouping, and equal power allocation (EPA), across a wide range of signal-to-noise ratios. Notably, the optimal number of pilots was smaller than the number of RIS elements and even the effective rank of the cascaded channel covariance matrix. The above trends persisted even when the RIS elements have one-bit phase resolution.

List of publications:

I. Chakraborty, N. B. Mehta, and S. S. Gollamudi, “Statistics-aware low-overhead training and rate implications for RIS-aided systems,” In Proc. IEEE ICC, May 2026.

I. Chakraborty, N. B. Mehta, and S. S. Gollamudi, “Low-Overhead Training for RIS-Aided Systems that Exploits Spatial Correlation and its Rate Implications,” Submitted in IEEE Trans. Wireless Communication.

I have attended several CNI seminars, which have been highly beneficial in enhancing my understanding across a wide range of topics.