Heterogeneous Gateway Planning for Semantic Communication in Industrial Networks

Semantic Communication (SemCom) has become a promising approach to transmit the information entropy instead of the raw data, which can increase the transmission efficiency and robustness. SemCom recently attracts more attention on the research of Internet of Things (IoT). Previous work focused more on new SemCom framework for IoT applications or lightweight SemCom models which can be easily deployed at IoT end-devices. However, the adaptation of SemCom in industrial networks still remains to be explored. In this article, we focus on gateway planning for SemCom in industrial networks. Specifically, in order to meet the different requirements of various nodes in industrial networks, we build a heterogeneous architecture of gateway planning for SemCom. Then, the Structure-Similarity-Index-Measure (SSIM) model in SemCom is introduced to illustrate the Signal-to-Noise Ratio (SNR) and compression ratio. Based on this, we build the bridge between the SSIM and the number of gateways to analyze our problem. To apply minimum gateways to cover the heterogeneous nodes and meet the SemCom transmission conditions, we formulate the problem as an NP-hard problem and simplify the problem to find the solution. Greedy-Based Algorithm (GBA) and Heuristic-Deep Q Network (H-DQN) algorithm are designed to solve this problem, and simulations are demonstrated to compare H-DQN, Q-learning, GBA and Branch and Bound (BB). The experimental results show that H-DQN outperforms other algorithms, deploying less gateways and saving over 50% of operation time. Besides, the throughput of our system is 1.3× larger than that of the single coverage system with less gateways.