OpenRANet: Neuralized Spectrum Access by Joint Subcarrier and Power Allocation with Optimization-based Deep Learning
This repository addresses a non-convex problem involving joint subcarrier and power control, aiming to minimize total power while satisfying rate requirements. The complexity stems from non-convexity, coupled constraints, and implicit resource uncertainties. The code primarily implements the reweighted primal-dual algorithm for achieving local optimality and the OpenRANet algorithm for approximating global optimal solutions under varying transmission rate constraints.
Experiments are conducted on a 64-bit workstation running Windows 10, equipped with an Intel(R) Core(TM) i7-8700K CPU at 3.70GHz and 32.00GB of RAM, utilizing Python 3.7 and PyTorch 1.11.0.
Toy examples of problem instances, along with code for generating data for training and validating the algorithms, are located in the “data_generation” folder.
The file “iteration_solver.py” implements the reweighted primal-dual algorithm for local optimality, while “OpenRANet.py” contains the implementation for constructing the OpenRANet.