Joint User Association and Resource Allocation for Multi-Cell Networks with Adaptive Semantic Communication

Xingqiu He, Chaoqun You, Tony Q. S. QuekarXiv 2023 · DOI · arXiv

Abstract

Traditional wireless communication focuses on bit-level accuracy, often ignoring the semantic meaning of data. Semantic Communication (SemCom) leverages AI to transmit only task-relevant information. However, the heterogeneity of user devices and dynamic network conditions pose significant challenges for SemCom deployment. This paper proposes an Adaptive Semantic Communication (ASC) framework that jointly optimizes user association, resource allocation, and semantic model complexity to maximize network utility in multi-cell environments.

Introduction

As 5G and 6G networks evolve, the demand for high-efficiency data transmission grows. Traditional methods prioritize the "perfect delivery of bits," but Semantic Communication (SemCom) offers a paradigm shift by using Deep Neural Networks (DNNs) to extract and transmit only the information relevant to a specific task (e.g., identifying an object in a video).

Despite its potential, SemCom faces three major hurdles in practical deployment:

  1. User Heterogeneity: Devices range from high-end smartphones to low-power IoT sensors with vastly different computational capabilities.
  2. Network Dynamics: Signal quality varies, requiring flexible data transmission rates to maintain accuracy.
  3. Resource Competition: In dense "Small-cell" networks, users compete for limited radio spectrum, and their choice of semantic "intensity" affects overall cell performance.

System Architecture

The proposed ASC framework addresses these challenges by allowing devices to dynamically adjust their semantic processing based on available resources and hardware constraints.

The ASC framework involves local feature extraction within the Wireless Device (WD) before transmitting semantic information to a Small-cell Base Station (SBS).

As shown in Figure 1, the process "Inside WD" involves raw data passing through a feature extraction layer (a DNN). The resulting semantic information is then sent via a transmitter to an SBS. The complexity of this extraction and the volume of data sent are the primary variables for optimization.

The Optimization Problem

The authors formulate a joint optimization problem to maximize total network utility, considering computation workloads, communication overhead, and resource constraints. The problem is proven to be strongly NP-hard, meaning it cannot be solved perfectly in real-time for large networks.

Notation used in the mathematical formulation of the joint optimization problem.

To manage this complexity, the authors decompose the problem into a three-stage hierarchical approach:

  1. ASC Scheme Selection: Finding the optimal balance between local computation (CPU cycles) and communication (transmitted bits) for each device. The authors prove a "tightness" property: at the optimal point, a device should ideally use its entire available energy or time budget to maximize accuracy.
  2. Spectrum Allocation: Distributing Resource Blocks (RBs) within each base station using a greedy approach for standard utility functions or dynamic programming for complex scenarios.
  3. User Association: Connecting users to base stations using a kurtosis-based heuristic. This prioritizes users with a "clear favorite" base station to avoid network congestion and mitigate interference.

Key Insights and Results

A critical finding is the Computation-Communication Trade-off: there is an optimal "sweet spot" where a device's local processing reduces airtime without over-draining the battery. Simulations in a 5G environment (10 SBSs, 100 WDs) demonstrate several key advantages:

  • Superior Utility: ASC consistently achieves higher system utility than traditional bit-based methods, especially when wireless resources are highly constrained.
  • Adaptability: Unlike fixed semantic models, ASC scales its performance based on the user's battery and CPU, leading to better results for low-power devices.
  • Efficiency: The three-stage algorithm achieves near-optimal utility with polynomial-time complexity, making it viable for real-time network management.

Conclusion

The ASC framework provides a robust solution for deploying semantic communication in heterogeneous, multi-cell networks. By dynamically adapting to device capabilities and network states, it ensures high task accuracy and efficient resource utilization. Future work will look to extend this model to high-speed mobility scenarios and downlink communications.

Summary updated

BibTeX
@misc{he2023joint,
  author = {Xingqiu He and Chaoqun You and Tony Q. S. Quek},
  title = {{Joint User Association and Resource Allocation for Multi-Cell Networks with Adaptive Semantic Communication}},
  year = {2023},
  eprint = {2312.01049},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2312.01049},
}
Figure