An information-theoretic framework for vision-aided communication over memoryless on-off fading channels

Abstract

We investigate a vision-aided communication system consisting of a transmitter, a receiver, and a vision sensor (such as a camera) co-located with the receiver. In this setup, the vision sensor acquires awareness of the communication environment and shares vision-based state knowledge with the receiver. The receiver then uses both the received signal and this state knowledge to decode the transmitted messages. While the vision sensor may not fully determine the channel impulse response, it can often provide useful environmental information that assists in decoding. To quantify the benefit of such vision-based state knowledge, we adopt an information-theoretic framework and model the system as a state-dependent channel where the physical environment influences the channel state. We apply the framework to an on-off fading channel that captures the link blockage at the millimeter-wave (mmWave) or terahertz (THz) frequencies. For comprehensive analysis, we consider both binary symmetric and Gaussian noise cases under perfect, imperfect, and no vision knowledge. We also investigate sequential and joint processing for decoding and prove that the performance gap between the two strategies is bounded by the sensing accuracy, revealing a simple design rule that when the vision sensing errors are small, sequential processing is near-optimal.