Timing Considerations in Blockage Prediction with IRS Beamforming for Ultra-Reliable Network Connectivity

Ensuring ultra-reliable network connectivity in dynamic environments requires accurate and timely blockage prediction.

Abstract

Ensuring ultra-reliable network connectivity in dynamic environments requires accurate and timely blockage prediction. This study analyzes the timing considerations in blockage prediction with Intelligent Reflecting Surface (IRS) beamforming, comparing adaptive filtering techniques—Least Mean Squares (LMS), Normalized Least Mean Squares (NLMS), and Recursive Least Squares (RLS)—against Long Short-Term Memory (LSTM) networks. Results show that LSTM achieves superior accuracy, ranging from 96.40% to 93.21%, while adaptive filters decline over time. Despite its superior accuracy, LSTM incurs a computational delay of 50 µs over the baseline model, which itself is 80 µs slower than adaptive filters. To enhance network reliability, we integrate Intelligent Reflecting Surface (IRS) beamforming, optimizing signal reflections under Non-Line-of-Sight (NLoS) conditions. For a base station (BS) communicating with a 64×64 uniform planar array (UPA) Intelligent Reflecting Surface (IRS), blockage proactive prediction must anticipate at least 31 ms into the future to accommodate transmission delays, handover, and beam training. These findings highlight LSTM’s potential in enhancing real-time blockage prediction and real-time network adaptability.