Radial basis function neural network-based algorithm unfolding for energy-aware resource allocation in wireless networks

• Published in: Wireless networks Vol. 30; no. 8; pp. 7041 - 7058
• Main Authors: Prasanna, B. T., Ramya, D., Shelke, Nilesh, Fernandes, J. Bennilo, Galety, Mohammad Gouse, Ashok, M.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.11.2024; Springer Nature B.V
• Subjects: Algorithms; Approximation; Artificial neural networks; Communications Engineering; Computer Communication Networks; Convexity; Design optimization; Effectiveness; Electrical Engineering; Energy management; Engineering; IT in Business; Maximum power; Network topologies; Networks; Neural networks; Radial basis function; Resource allocation; Wireless communication systems; Wireless networks; Deep learning; Radial basis function neural network (RBFNN); Deep unfolding of the successive concave approximation (DUSCA); Wireless networks; Energy-aware resource allocation
• ISSN: 1022-0038; 1572-8196
• DOI: 10.1007/s11276-023-03540-0

Significant advances in high-bandwidth applications and rising power consumption have highlighted the need for energy-efficient design solutions in backbone optical networks. Task allocation is an important application scenario for sensor networks in which a central entity allocates resources to a collection of geographically scattered sensor nodes to achieve an overall goal. Effective resource allocation is essential for optimizing wireless networks' performance and energy efficacy. We provide a novel technique for energy-aware resource allocation in wireless networks that integrates the radial basis function neural network (RBFNN) algorithm with the deep unfolding of the successive concave approximation (DUSCA). The RBFNN is approximated as a function to understand the relationship between resource allocation decisions and network performance measures. Using its capacity to characterize complex nonlinear mappings, the RBFNN offers a flexible framework for depicting the intricate interdependence in wireless network settings. The DUSCA framework is learned using progressive training and stochastic gradient descent. The unsupervised loss is precisely designed to illustrate the objective's monotonic property under maximum power limitations. Extensive numerical findings show that it may be applied to a wide range of network topologies with different sizes, densities, and channel dispersion. In addition, we present the DUSC method, which is frequently employed in resource allocation to tackle non-convex optimization problems. By expressing the SCA's repetitive phases as a deep neural network and optimizing resource allocation concurrently, we can improve the convergence and quality of its solutions. Our proposed RBFNN–DUSCA architecture, which combines RBFNN with the profound unfolding of the SCA algorithm, provides a viable method for addressing resource allocation challenges in wireless networks, paving the way for more energy-efficient and high-performance wireless communication systems.