Cross-layer optimization using two-level dual decomposition in multi-flow ad-hoc networks

Abstract

The conventional protocol architectures based on rigid and inflexible layering principles are not suitable to meet the key challenges posed by ad-hoc networks. To overcome the performance limitations caused by lack of coordination between layers, the communication protocols crossing different layers need to be optimized jointly. In this paper, we explore a novel cross-layer framework for joint optimization of congestion control, routing, contention control, and power control in multi-flow ad-hoc networks. Accordingly, a generalized mathematical model is developed for cross-layer optimization formulation that can seamlessly address these issues with the aid of network utility maximization. Convexity is essential to attain a global optimum solution for the optimization problem in its feasible region. Therefore, the original non-convex optimization problem is convexified through suitable logarithmic transformations. Then, we employ a two-level dual decomposition technique for relaxing the interlayer coupling constraints in the network. In the first level, the problem is vertically decomposed into four disjoint subproblems that are solved independently across transport, network, MAC and physical layers. In the second level, a horizontal decomposition is applied within the physical layer. Each of the four subproblems is jointly correlated through the set of Lagrangian dual variables that control the interlayer coupling. Subsequently, the subgradient projection procedure is used to solve the associated dual problem. For this, a distributed iterative algorithm is proposed to implement the cross-layer approach with static channels and multiple source–destination pairs. Simulation results and analyses depict the global convergence of the convex optimization problem to a unique optimal solution. We also present the convergence process of the dual variables that act as bridges connecting and coordinating the four subproblems. Finally, we compare our proposed cross-layer algorithm with two previous algorithms employing cross-layer deign of two and three layers. Numerical results show that our algorithm offers considerable performance enhancement over the previous work in terms of throughput, persistence probability, and power consumption.