Deep transfer learning-based performance prediction of URLLC in independent and not necessarily identically distributed Interference Networks

In the rapidly evolving sixth-generation (6G) networks, achieving ultra-high reliability for short packets poses a crucial challenge for network designers as classical Shanon capacity bounds become obsolete. This paper addresses this imperative task and presents a performance analysis of ultra-reliable lowlatency communication (URLLC) systems in clustered networks with short packet communication (SPC) in the finite blocklength regime. In clustered networks, where multiple interferers are present, multiple ground users are grouped into a cluster and communicate with a central node called a cluster head (CH). The system adopts a CH as a wireless relay to accomplish URLLC between a base station and ground users. In this context, we first derive a closed-form expression for the overall block error rate of the system. The analysis is based on a theoretical model that considers the packet size, blocklength, and maximum achievable rate. Moreover, the paper proposes a transfer learning approach using fine-tuned target models with domain adaptation for real-time prediction of the system’s performance. The prediction is necessary to accurately assess the performance of URLLC in statistically independent and not necessarily identically distributed (Non-IID) interference networks. Deep transfer learning is utilized to develop a model that can generalize interference scenario and provide reliable predictions. The transfer learning approach involves using a source model and fine-tuning it with domain-specific data to improve the prediction accuracy. Overall, the paper provides insights into the performance of URLLC systems with SPC and proposes an approach to improve the real-time prediction’s accuracy. The proposed approach has the potential to facilitate the deployment of URLLC systems in practical applications where real-time performance prediction is crucial.

Mots clés

Ultra-reliable low latency communication (URLLC); transfer learning (TL); block error rate (BLER); convolutional neural network (CNN); performance analysis; non-IID interference network; machine learning-based performance prediction