Energy Efficient Beamforming Training in Terahertz Communication Systems

TL;DR

This work tackles the energy-inefficient overhead of THz pencil-beam training by introducing EETBF, a framework that splits beam training into EETBF-BT and learning-based EETBF-PA. It leverages historical beam data and reinforcement learning to adapt training beams and allocate training power, achieving lower latency and higher energy efficiency than full-search, iterative, and non-power-controlled baselines. The method is validated through simulations across THz bands, showing that an optimal tradeoff between training beam count, power, and interval yields superior EE and effective rate. The results suggest a practical pathway to enable Tbps THz links in 6G-like systems while mitigating the prohibitive training energy and time costs.

Abstract

Terahertz (THz) enables promising Tbps-level wireless transmission thanks to its prospect of ultra-huge spectrum utilization and narrow beamforming in the next sixth-generation (6G) communication system. Compared to millimeter wave (mmWave), THz intrinsically possesses compellingly severer molecular absorption and high pathloss serving confined coverage area. These defects should be well conquered under the employment of ultra-thin 3D beamforming with enormous deployed antennas with high beam gains. However, pencil-beams require substantially high overhead of time and power to train its optimal THz beamforming direction. We propose an energy efficient (EE) oriented THz beamforming (EETBF) scheme by separating the original complex problem into beamforming training (EETBF-BT) acquirement and learning-enabled training power assignment (EETBF-PA). The historical beam data is employed to update next beam selection policy. The performance results have demonstrated that the proposed EETBF outperforms the existing benchmarks leveraging full beam search, iterative search, linear/binary search as well as non-power-control based mechanism in open literature. Our proposed EETBF scheme results in the lowest training latency and power consumption, achieving the highest effective rate and EE performance.

Figures (9)

• Figure 1: Architecture of THz BS deployed with an antenna array having $N_{h}$ horizontal and $N_v$ vertical antenna elements. The corresponding beamwidths of the $i$-th horizontal and the $j$-th vertical beams are $\theta_{h,i}$ and $\theta_{v,j}$, respectively.
• Figure 2: The beam training model and framework. (a) Conventional full-power beam training. (b) Proposed EE-oriented beam training. (c) The designed protocol of THz beamforming training, including beam training overhead and transmission duration. Note that the policy decision is performed before initialiation of beam training.
• Figure 3: Illustration of the beamforming training regarding (a) forward search in \ref{['s1']}, (b) backward search in \ref{['s2']} and (c) bi-directional search in \ref{['s3']}. (d) The final 3D beamforming training set. Note that the beam with dark red indicates the previous optimal beam index $\psi_{\mathcal{A}}$, whereas the beams with bright yellow represent the expanded search area calculated based on \ref{['sss']}.
• Figure 4: The beamwidth and corresponding antenna gain w.r.t. different numbers of THz antennas.
• Figure 5: Instantaneous performance of training power, latency and SNR values versus timesteps during the beamforming training process of the proposed EETBF scheme.