The Terahertz (THz) frequency band is one of the major factors for reaching the ultra-high data rates and sub-millisecond latency that are expected in sixth-generation (6G) wireless networks. On the other hand, there are critical challenges in THz propagation, such as high path loss, molecular absorption, and severe rain attenuation, which are all outdoor factors that greatly reduce the reliability of the link. To tackle this issue, the paper presents a Deep Reinforcement Learning (DRL)-based adaptive beamforming technique that will work in real-time to offset the effects of the atmosphere in Massive MIMO THz systems.
The technique that is put forward uses a Hybrid Beamforming structure and also has a Soft Actor-Critic (SAC) agent that will help choose the most suitable beamforming vectors from the codebook that has been set up. The SAC agent's training is based on a reward function that endorses the instantaneous throughput and at the same time considers the dynamic rain conditions which vary between 0 to 100 mm/hr. The setup is compared with a static beamformer and a Deep Q-Network (DQN) standard during a realistic THz channel model that includes the impact of rain on the signal.The outcomes of the simulation are that the method based on SAC keeps almost total connectivity during any rain conditions while at the same time decreasing the average outage from 0.5 (static beamformer) to almost nothing. Furthermore, it gets the minimum manageable data rate to be increased by as much as 7.5×, which is very close to the performance of the hypothetical codebook-based oracle solution.Finally, the proposed DRL-driven adaptive beamforming technique not only displays but also brings a new energy, thus leading to support for ultra-reliable, high-throughput THz communication under rain-dense atmospheric conditions, thus elevating the physical-layer intelligence necessary for practical 6G deployments.