Reliable Beamforming at Terahertz Bands: Are Causal Representations the Way Forward?

Christo Kurisummoottil Thomas, Walid Saad

Abstract

As wireless communication moves toward the Terahertz (THz) band to support the massive bandwidth requirements of the metaverse and extended reality (XR), maintaining reliable links in high-mobility scenarios remains a critical challenge. Traditional beamforming methods often struggle with the "pencil-like" narrow beams required at these frequencies when users move rapidly. This paper presents a semantically aware beamforming framework based on Variational Causal Networks (VCN). By learning the causal factors that explain both multi-modal data generation and wireless channel dynamics, the proposed system achieves significant gains in reliability and information rate over conventional tracking methods.

Introduction

The Terahertz band (300 GHz to 10 THz) is a cornerstone for future 6G networks, offering the ultra-high bandwidth and low latency needed for immersive digital environments. However, THz signals are highly susceptible to atmospheric effects and blockage, necessitating highly directional beamforming. In high-mobility environments (e.g., 90 km/h), keeping these narrow beams aligned is difficult. Traditional rigid mathematical models or data-hungry "black-box" AI often fail to generalize or adapt to real-world dynamics.

Proposed Methodology: Causal-Aware Beamforming

The core innovation is a beamforming solution that moves beyond simple bit-level transmission to a model that understands the "why" and "how" of data generation.

Variational Causal Networks (VCN)

Instead of direct position prediction, the system learns a latent "causal representation" ($z_k$). This representation captures the underlying factors influencing both the multi-modal data (such as video or sensor inputs) and the wireless channel. Using Causal Graphical Models (CGMs), the system becomes "interventional," allowing it to adapt robustly to environmental changes.

System architecture for causality-aware beamforming.

Two-Part Beamforming and Optimization

The beamforming (BF) vector is decomposed into a static component, learned from long-term channel history and data semantics, and a dynamic component adjusted based on instantaneous channel estimates. The authors utilize Generalized Eigen Vectors (GEV) to derive the BF and combiner matrices, balancing signal maximization for the target user with interference mitigation for others. A max-min optimization strategy ensures fairness across the network by maximizing the minimum semantic information rate for all users.

Experimental Results

The framework was evaluated in a multi-user MIMO setup (64 antennas) at 300-450 GHz, simulating high-mobility conditions (90 km/h). The "Causality-Aware BF" was compared against perfect Channel State Information (CSI), naive beamforming, and standard DFT-based beam tracking.

Received semantic information vs. number of users.

Key findings include:

  • Semantic Information Rate: The proposed system achieved a 43% improvement in the received semantic information rate compared to traditional schemes, as shown in the performance curves.
  • Reliability: The scheme demonstrated a 14% improvement in semantic reliability over conventional beam tracking.
  • Mobility Resilience: The causal representation allowed for stable links even at high speeds where traditional methods typically degrade.

Conclusion and Future Work

By leveraging the underlying causal structure of data and channels, this framework offers a more data-efficient and robust alternative to traditional AI in wireless communications. While promising, future research must address the computational complexity of training VCNs for real-time THz applications and account for hardware impairments like phase noise and power amplifier non-linearities.

Summary updated

Abstract

Future wireless services, such as the metaverse require high information rate, reliability, and low latency. Multi-user wireless systems can meet such requirements by utilizing the abundant terahertz bandwidth with a massive number of antennas, creating narrow beamforming solutions. However, existing solutions lack proper modeling of channel dynamics, resulting in inaccurate beamforming solutions in high-mobility scenarios. Herein, a dynamic, semantically aware beamforming solution is proposed for the first time, utilizing novel artificial intelligence algorithms in variational causal inference to compute the time-varying dynamics of the causal representation of multi-modal data and the beamforming. Simulations show that the proposed causality-guided approach for Terahertz (THz) beamforming outperforms classical MIMO beamforming techniques.
BibTeX
@inproceedings{thomas2023reliable,
  author = {Christo Kurisummoottil Thomas and Walid Saad},
  title = {{Reliable Beamforming at Terahertz Bands: Are Causal Representations the Way Forward?}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  doi = {10.1109/icassp49357.2023.10096153},
}
Figure