Event-Triggered Communications for Industrial IoT: Channel Coding Rate and Reconstruction Distortion
Yiyu Qiu, Wei Chen · 2024
Event-triggered communication has attracted considerable recent attention because it holds the promise of saving bandwidth and energy in real-time monitoring, networked control, federated learning, etc. However, its performance limits remain unknown in theory. In this paper, we investigate from an information-theoretic perspective the fundamental tradeoff between the transmission power and the real-time reconstruction distortion of event-triggered communications. In particular, we focus on the real-time monitoring of a discrete event system, characterized by a jump or semi-Markov process, through AWGN channels. It is found that an all-zero codeword with the probability of the non-triggered event is forcibly added to the codebook, no matter how other codewords are selected. In this context, we derive the maximum achievable channel coding rates in both finite and infinite blocklength regimes, bridging the transmission power and the instantaneous throughput. By borrowing the idea of information freshness and queuing analysis, we further reveal the real-time reconstruction error of event-triggered communication given its instantaneous rate. It is shown that the event-triggered communication with Lebesgue sampling outperforms conventional periodic communication with Riemann sampling in terms of the power-distortion tradeoff. Numerical results validate our theoretical analysis and demonstrate the substantial power-saving gain given various target mean square errors (MSE).