Spectrum Allocation for Multiuser Terahertz Communication Systems: A Machine Learning Approach

Abstract

In this paper, we propose a novel spectrum allocation design, leveraging machine learning, for multiuser communication systems operating at the terahertz (THz) band. In this design, we propose to (i) change the bandwidth of sub-bands and (ii) underutilize edge spectra of transmission windows (TWs) where the molecular absorption (MA) coefficient is very high. Different from existing studies, our design is not limited to the scenario where the MA coefficient in the spectrum designated for allocation can be accurately modeled by simply using a piecewise exponential function. We establish a constrained optimization problem and introduce an unsupervised learning approach for its solution. Through offline training, we learn a deep neural network (DNN) using a loss function inspired by the Lagrangian of the established problem. The trained DNN is then employed to derive solutions when multiuser distance parameters are given. Based on numerical analysis, we show that when the MA coefficient in the spectrum designated for allocation exhibits highly non-linear variations, our proposed approach can achieve a higher data rate than that of existing approaches which only attain approximate solutions.

Keywords

• Terahertz band communication
• spectrum allocation
• machine learning
• unsupervised learning
• non-convex optimization
• offline training