Ris-Enhanced 6G Networks: Advanced User Location Prediction with Generative Models

The paper explores the utilization of Reconfigurable Intelligent Surfaces (RIS) as a novel approach in a 6G network. RIS is a powerful method that controls the propagation environment by adjusting incident signals and presents a cost-effective and energy-efficient alternative to complex and expensive Multiple Input Multiple Output (MIMO) systems. Our research work presents a Generative Deep Learning Model (GDLM) based prediction framework that includes a Gated Recurrent Unit (GRU), Long Short Term Memory (LSTM), and Transformer networks to predict accurate user locations in real-time which enables RIS to proactively enhance signal delivery. We opted for these these specific models because of their unique capabilities to process sequential data and their varying architectural strengths in capturing long and short-term dependencies. GRU and LSTM are both recurrent neural networks that offer nuanced approaches to temporal data processing. GRU simplifies the model architecture by using fewer gates as compared to LSTM, making it more suitable for certain tasks, while LSTM's design allows for better handling of extended dependencies at the cost of increased complexity. However, Transformers show better performance in parallelizing operations and managing long-range dependencies across sequences, eschewing recurrence entirely in favor of a purely attention-based mechanism. This latter feature is crucial for real-time user location prediction in dynamic 6G environments. In our comparison investigation, Transformers outperform GRU and LSTM in obtaining a noticeably lower Mean Absolute Error (MAE) in user location prediction. This work lays the groundwork for future investigations into sophisticated GDLM structures for improvements in wireless communication, in addition to clarifying the unique benefits of using these models in a 6G scenario.