Abstract
This work develops a novel power control framework for energy-efficient power control in wireless networks. The proposed method is a new branch-and-bound procedure based on problem-specific bounds for energy-efficiency maximization that allow for faster convergence. This enables to find the global solution for all of the most common energy-efficient power control problems with a complexity that, although still exponential in the number of variables, is much lower than other available global optimization frameworks. Moreover, the reduced complexity of the proposed framework allows its practical implementation through the use of deep neural networks. Specifically, thanks to its reduced complexity, the proposed method can be used to train an artificial neural network to predict the optimal resource allocation. This is in contrast with other power control methods based on deep learning, which train the neural network based on suboptimal power allocations due to the large complexity that generating large training sets of optimal power allocations would have with available global optimization methods. As a benchmark, we also develop a novel first-order optimal power allocation algorithm. Numerical results show that a neural network can be trained to predict the optimal power allocation policy.
Introduction
As wireless networks evolve toward 5G and 6G, achieving significant gains in energy efficiency (EE)—measured in bits per Joule—is paramount. A primary mechanism for this is power control: optimizing the transmit power of each user to maximize system-wide EE. However, in wireless interference networks, one user's signal acts as noise for others, making EE maximization a non-convex, NP-hard problem. Traditional global optimization tools are often too slow for real-time use, and existing deep learning (DL) approaches typically rely on suboptimal heuristic data for training because generating globally optimal datasets was computationally prohibitive.
The Global Optimization Framework
The core contribution of this research is a highly efficient Branch-and-Bound (BB) algorithm designed to find globally optimal power settings. Unlike generic solvers, this algorithm exploits the specific mathematical structure of the problem—namely, the pseudo-concavity of individual link energy efficiencies. By developing tighter, problem-specific bounds, the algorithm can prune the search space significantly faster than previous methods.
This efficiency allows the BB algorithm to serve as a "teacher" for machine learning models. In numerical tests, the algorithm generated a massive dataset of over 600,000 globally optimal samples in just 8.4 CPU hours, a task that would have taken orders of magnitude longer with standard global optimization frameworks.
Deep Learning for Real-Time Implementation
To bridge the gap between complex global optimization and real-time network requirements, the authors implemented a Deep Neural Network (DNN). By training the DNN on the "perfect" power allocations generated by the BB algorithm, the network learns to approximate the global optimization strategy itself.
The DNN maps channel state information directly to optimal transmit powers in a single forward pass. Key techniques such as data augmentation via permutation invariance (ensuring the network treats users based on channel conditions rather than arbitrary indices) were used to enhance training efficiency and accuracy.
Key Results and Insights
The proposed framework demonstrates that training on globally optimal data significantly raises the performance ceiling of neural networks in wireless control.
- High Accuracy: The trained DNN achieved a median relative error of only 0.00739 compared to the true global optimum.
- Robustness: Although trained on simple Rayleigh fading models, the network generalized well to complex urban environments (Hata-COST231), suggesting the learned logic is robust across different propagation scenarios.
- Real-Time Capability: The DNN provides near-optimal solutions in milliseconds, making it suitable for dynamic environments where channel conditions change rapidly.
Conclusion
This work provides a path forward for practical, globally optimal power control in energy-efficient wireless networks. By combining a mathematically specialized global optimization algorithm with deep learning, the framework achieves the "best of both worlds": the theoretical performance of global optimization with the real-time execution speed required for next-generation wireless systems. Future research will look toward scaling these methods for massive MIMO systems and dynamic, time-varying network topologies.
Summary updated