Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenRANet: Neuralized Spectrum Access by Joint Subcarrier and Power Allocation with Optimization-based Deep Learning

Siya Chen, Chee Wei Tan, Xiangping Zhai, H. Vincent Poor

TL;DR

The paper tackles the nonconvex problem of minimizing total transmit power in Open RAN under per-user rate constraints, formulated as $\min\sum_{l,m} p_l^m$ subject to $\sum_m f_l^m(\mathsf{SINR}_l^m(\mathbf p^m)) \ge \bar{r}_l$. It develops a two-pronged approach: a low-complexity primal-dual algorithm exploiting the log-convexity of standard interference functions, and OpenRANet, an optimization-based deep learning model that embeds a projection and a convex optimization layer to enforce feasibility and solve the convex subproblem. The convex subproblems are obtained via decoupling, a logarithmic change of variables, and relaxation, enabling differentiable end-to-end training. Numerical results show that OpenRANet closely tracks globally optimal solutions with far less training cost than purely optimization-based baselines and outperforms purely data-driven methods in constraint adherence, demonstrating strong potential for AI-native resource management in Open RAN and satellite-terrestrial networks. The framework lays groundwork for extending to multi-cell deployments and incorporating broader power and performance metrics with adaptive learning strategies.

Abstract

The next-generation radio access network (RAN), known as Open RAN, is poised to feature an AI-native interface for wireless cellular networks, including emerging satellite-terrestrial systems, making deep learning integral to its operation. In this paper, we address the nonconvex optimization challenge of joint subcarrier and power allocation in Open RAN, with the objective of minimizing the total power consumption while ensuring users meet their transmission data rate requirements. We propose OpenRANet, an optimization-based deep learning model that integrates machine-learning techniques with iterative optimization algorithms. We start by transforming the original nonconvex problem into convex subproblems through decoupling, variable transformation, and relaxation techniques. These subproblems are then efficiently solved using iterative methods within the standard interference function framework, enabling the derivation of primal-dual solutions. These solutions integrate seamlessly as a convex optimization layer within OpenRANet, enhancing constraint adherence, solution accuracy, and computational efficiency by combining machine learning with convex analysis, as shown in numerical experiments. OpenRANet also serves as a foundation for designing resource-constrained AI-native wireless optimization strategies for broader scenarios like multi-cell systems, satellite-terrestrial networks, and future Open RAN deployments with complex power consumption requirements.

OpenRANet: Neuralized Spectrum Access by Joint Subcarrier and Power Allocation with Optimization-based Deep Learning

TL;DR

The paper tackles the nonconvex problem of minimizing total transmit power in Open RAN under per-user rate constraints, formulated as subject to . It develops a two-pronged approach: a low-complexity primal-dual algorithm exploiting the log-convexity of standard interference functions, and OpenRANet, an optimization-based deep learning model that embeds a projection and a convex optimization layer to enforce feasibility and solve the convex subproblem. The convex subproblems are obtained via decoupling, a logarithmic change of variables, and relaxation, enabling differentiable end-to-end training. Numerical results show that OpenRANet closely tracks globally optimal solutions with far less training cost than purely optimization-based baselines and outperforms purely data-driven methods in constraint adherence, demonstrating strong potential for AI-native resource management in Open RAN and satellite-terrestrial networks. The framework lays groundwork for extending to multi-cell deployments and incorporating broader power and performance metrics with adaptive learning strategies.

Abstract

The next-generation radio access network (RAN), known as Open RAN, is poised to feature an AI-native interface for wireless cellular networks, including emerging satellite-terrestrial systems, making deep learning integral to its operation. In this paper, we address the nonconvex optimization challenge of joint subcarrier and power allocation in Open RAN, with the objective of minimizing the total power consumption while ensuring users meet their transmission data rate requirements. We propose OpenRANet, an optimization-based deep learning model that integrates machine-learning techniques with iterative optimization algorithms. We start by transforming the original nonconvex problem into convex subproblems through decoupling, variable transformation, and relaxation techniques. These subproblems are then efficiently solved using iterative methods within the standard interference function framework, enabling the derivation of primal-dual solutions. These solutions integrate seamlessly as a convex optimization layer within OpenRANet, enhancing constraint adherence, solution accuracy, and computational efficiency by combining machine learning with convex analysis, as shown in numerical experiments. OpenRANet also serves as a foundation for designing resource-constrained AI-native wireless optimization strategies for broader scenarios like multi-cell systems, satellite-terrestrial networks, and future Open RAN deployments with complex power consumption requirements.
Paper Structure (22 sections, 7 theorems, 53 equations, 8 figures, 2 tables, 2 algorithms)

This paper contains 22 sections, 7 theorems, 53 equations, 8 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. System Model
  3. Reweighted Primal-dual Algorithm for Total Power Minimization
  4. Reformulation of the Total Power Minimization
  5. Reweighted Primal-dual Algorithm for Total Power Minimization
  6. OpenRANet: An Optimization-based Deep Learning Model with Projection and Convex Optimization Layers for Total Power Minimization
  7. The OpenRANet for Total Power Minimization
  8. Training Process
  9. Numerical Examples
  10. Performance of the Primal-dual Algorithm
  11. Performance of the OpenRANet
  12. Performance Comparison
  13. Conclusion
  14. Proof of Lemma \ref{['p_standard']}
  15. Proof of Theorem \ref{['theorem2']}
  16. ...and 7 more sections

Key Result

Theorem 1

If the problem in opt:power_mnos_global is feasible, then the transmission rate constraints in opt:power_mnos_global are tight when opt:power_mnos_global reaches to its optima, which means

Figures (8)

  • Figure 1: The architecture of the system model for a downlink multi-carrier open RAN system with $L$ active users and $M$ subcarriers.
  • Figure 2: The architecture of the OpenRANet for approximating the optimal solution to \ref{['opt:power_mno_alt']}. The channel gain $G$ is dimensionally reduced by trainable convolutional filters. Then, concatenated with the Gaussian noise $\sigma$ and transmission rate $\bar{\mathbf{r}}$, it is fed into the dense layers as the feature vector. The output $\mathbf{h}_k(\bm{W})$ is then input into the projection layer \ref{['eq:project']} such that the transmission rate requirements \ref{['nonconvexity']} are satisfied. Finally, the projected transmission rates are input into a convex optimization model, which can be solved using the fast Algorithm \ref{['alg:decentralizedAlg']}. The loss function considers the deviation of $\bar{\mathbf{r}}$ and $\mathbf{p}^*$, where $C$ is the penalty of violation to \ref{['nonconvexity']}.
  • Figure 3: Power allocation per subcarrier with Algorithm \ref{['alg:decentralizedAlg']} for the 4-users by 2-subcarriers wireless network, under Rayleigh (a), Rician (b), Nakagami (c), and Weibull fading (d) channels, resp. The 95% confidence intervals were plotted for 100 trials over the same random sample.
  • Figure 4: Power allocation per user with Algorithm \ref{['alg:decentralizedAlg']} for the 4-users by 2-subcarriers wireless network, under Rayleigh (a), Rician (b), Nakagami (c), and Weibull fading (d) channels, resp. The 95% confidence intervals were plotted for 100 trials over the same random sample.
  • Figure 5: The mean squared error (a) and R-squared (b) of approximating optimal power allocation for the OpenRANet on the power minimization problem under different rate function constraints.
  • ...and 3 more figures

Theorems & Definitions (15)

  • Theorem 1
  • Definition 1
  • Lemma 1
  • Lemma 2
  • Corollary 1
  • Remark 1
  • Definition 2
  • Lemma 3
  • Corollary 2
  • Remark 2
  • ...and 5 more