Task-oriented Source Coding for Real-Time Applications: Invited Paper

The emerging real-time applications in the Industrial Internet of Things (IIoT) puts more and more strict requirements on extremely low latency communications. In these real-time applications, the duration of codeword transmission cannot be ignored and may cause significant performance loss. To overcome this, task-oriented source coding which integrates the design of source coding and real-time applications may greatly reduce the latency-induced performance loss. In contrast to the traditional source coding that only focuses on the minimization of average codeword length, task-oriented source coding aims at minimizing the cumulative cost in real-time tasks by jointly optimizing source coding and real-time decision making. In this paper, we propose an optimization framework for task-oriented source coding in extremely low latency communications, in which the codeword lengths of all realizations are optimized for minimization of latency-induced cost. To demonstrate the potential of our proposed framework, two specific scenarios of real-time monitoring and real-time control are induced, respectively. In these scenarios, the closed expression of the optimal codeword lengths is derived. Based on this, the performance analysis on minimization of latency-induced cost is further presented. Finally, the simulation results show that the source codebook derived under our proposed framework achieves a significant performance gain in real-time tasks, compared with traditional Huffman coding.