Abstract

High-mobility wireless environments—such as those experienced in vehicular or aerial networks—pose significant challenges to reliable communication. Rapid time variations and frequency dispersion in these environments lead to severe intersymbol interference and signal distortion. Conventional modulation schemes and channel estimation methods, which are often based on a time-frequency representation, struggle to maintain performance under such conditions. Generalized Frequency Division Multiplexing (GFDM) has emerged as a promising modulation technique for next-generation wireless networks due to its flexibility and efficient spectrum usage. However, its conventional formulation does not adequately address the dynamic nature of high-mobility channels.
This thesis presents a novel approach that redefines the GFDM system model in the delay-Doppler domain. The delay-Doppler domain offers a natural framework for representing channels in high-mobility scenarios, as it captures the sparse structure of the channel more effectively than the conventional time-frequency domain. By transforming the GFDM signal into the delay-Doppler domain, our method exploits the inherent sparsity of the channel, thereby enabling more accurate and efficient channel estimation. Additionally, a superimposed pilot scheme is introduced, whereby pilot symbols are embedded within the data-bearing frame. This strategy eliminates the need for dedicated pilot-only regions, thus significantly enhancing spectral efficiency.
Based on the new system model, we investigate two channel estimation methods. The first approach employs a compressed sensing technique using the Subspace Pursuit (SP) algorithm. This method reconstructs the channel vector from a limited number of measurements, leveraging the sparse nature of the channel. It offers low computational complexity, which is beneficial for real-time implementations. However, the SP algorithm requires prior knowledge of the channel’s sparsity level—a parameter that is often difficult to determine in practice.
To overcome this limitation, the second method adopts Sparse Bayesian Learning (SBL) for channel estimation. SBL integrates prior information about the channel’s sparse structure directly into a Bayesian inference framework, allowing it to both accurately estimate the key channel parameters and identify the positions of the non-zero elements without requiring a priori sparsity knowledge. Simulation results demonstrate that the SBL-based estimator outperforms the SP algorithm, particularly in scenarios where pilot overhead is constrained.
Building on these estimation techniques, the thesis further extends the proposed framework to incorporate reconfigurable intelligent surfaces (RIS). RIS are composed of numerous passive reflecting elements that can dynamically adjust their reflection coefficients. By optimizing these coefficients, the RIS can steer the reflected signals to constructively combine with the direct path, thereby enhancing the overall channel gain and system capacity. Hence, a low-complexity phase optimization strategy is then employed to tune the RIS phase coefficients, maximizing the effective channel gain and improving the achievable rate.
The extensive simulation results presented in this thesis validate the performance of the proposed methods. Our findings indicate that the new GFDM system model in the delay-Doppler domain leads to significant improvements in channel estimation accuracy and robustness in high-mobility scenarios. The superimposed pilot scheme enhances spectral efficiency by embedding pilot symbols within the data frame, and the SBL-based channel estimator demonstrates superior performance over conventional greedy methods such as SP. Moreover, the integration of RIS with optimized phase shifts further increases the achievable rate and overall system capacity compared to systems with random phase configurations or without RIS support.