Separation Principles in Wireless Networking

A general wireless networking problem is formulated whereby end-to-end user rates, routes, link capacities, transmit-power, frequency, and power resources are jointly optimized across fading states. Even though the resultant optimization problem is generally nonconvex, it is proved that the gap with its Lagrange dual problem is zero, so long as the underlying fading distribution function is continuous. The major implication is that separating the design of wireless networks in layers and per-fading state subproblems can be optimal. Subgradient descent algorithms are further developed to effect an optimal separation in layers and layer interfaces.