Executive Summary
Millimeter-wave (mmWave) networks are essential for high-speed 5G and 6G communications but suffer from extreme sensitivity to urban blockages. Traditional relaying schemes introduce significant latency due to the serial nature of channel estimation and relay selection. This paper proposes a predictive relay selection scheme that leverages the spatial and temporal correlation of signal shadowing. By predicting the optimal relay for the next time slot while simultaneously beamforming in the current one, the system achieves near-ideal performance with drastically reduced overhead.
Introduction: The Latency Bottleneck in mmWave
In dense urban environments, mmWave signals are frequently blocked by buildings and vehicles. To maintain connectivity, networks utilize relays to bypass these obstacles. However, standard relaying protocols require estimating Channel State Information (CSI) for all potential relays before selection and transmission can occur. This "serial bottleneck" consumes valuable bandwidth and introduces latency that can degrade the user experience in real-time applications. The authors address this by parallelizing relay selection and data transmission through predictive modeling.
Proposed Methodology
The proposed system utilizes a two-stage joint beamforming and relay selection scheme tailored for urban environments.
Urban Channel Modeling
The model replaces standard Euclidean distance with Manhattan distance to better reflect signal propagation along street segments. It specifically accounts for shadowing correlation—the phenomenon where signal strength remains relatively consistent across small spatial and temporal intervals.
Predictive Selection and Parallelization
- Predictive Selection (Stage 1): The system uses Kalman Filtering and Gaussian Process Regression (GPR) on past magnitude-only CSI (Received Signal Strength) to predict the best relay for the next time slot.
- Distributed Beamforming (Stage 2): Simultaneously, the system performs optimal Amplify-and-Forward (AF) beamforming using the relay selected in the previous slot.
- Optimization: To handle the stochastic nature of the environment, the authors employ Sample Average Approximation (SAA), replacing complex expectations with empirical means derived from Monte Carlo sampling.
Core Innovations
- Shadowing as a Predictive Asset: Instead of treating shadowing as noise, the scheme exploits its correlation structure for "one-step-ahead" predictions.
- Magnitude-Only CSI: The authors prove that complex CSI (including phase) is unnecessary for relay selection; magnitude-only measurements are sufficient and easier to obtain.
- Decoupled Architecture: Relay clusters can independently determine their successors, reducing the computational burden on central controllers.
Simulation Environment and Results
The study evaluated the scheme using synthetic CSI data grounded in empirical models for 28 GHz and 38 GHz urban channels (typical of New York City and Seoul). The model assumed specific path-loss exponents and shadowing parameters ($\eta^2 = 40$, $\gamma = 15$) typical of urban microcells.
Key results include:
- Near-Ideal SINR: The predictive SAA-based policy achieved SINR levels within 2.3 dB of an "ideal" scheme with perfect instantaneous CSI.
- Significant Gains: The proposed method outperformed randomized selection policies by approximately 7.7 dB.
- Overhead Efficiency: By moving selection to the previous time slot, the "waiting delay" for CSI estimation is virtually eliminated, and the number of required channel estimations scales efficiently even in dense networks.
Conclusion and Future Directions
While the current model demonstrates high efficiency for static infrastructure, future work is needed to adapt the scheme for mobile relays (e.g., vehicle-to-everything communications). Additionally, expanding the model from a 2D Manhattan grid to a 3D environment with varying building heights could further refine the shadowing kernels and predictive accuracy.
Summary updated