Network utility maximization and price-based distributed algorithms for rate-reliability tradeoff

The current framework of network utility maximization for rate allocation and its price-based algorithms assumes that each link provides a fixed-size transmission 'pipe' and each user's utility is a function of transmission rate only. These assumptions break down in many practical systems, where, by adapting the physical layer channel coding or transmission diversity, different tradeoffs between rate and reliability can be achieved. In network utility maximization problems formulated in this paper, the utility for each user depends on both transmission rate and signal quality, with an intrinsic tradeoff between the two. Each link may also provide a higher (lower) rate on the transmission 'pipes' by allowing a higher (lower) decoding error probability. Despite non-separability and nonconvexity of these optimization problems, we propose new pricebased distributed algorithms and prove their convergence to the globally optimal rate-reliability tradeoff under readily-verifiable sufficient conditions. We first consider networks in which the rate-reliability tradeoff is controlled by adapting channel code rates in each link's physical layer error correction codes, and propose two distributed algorithms based on pricing, which respectively implement the 'integrated' and 'differentiated' policies of dynamic ratereliability adjustment. In contrast to the classical price-based rate control algorithms, in our algorithms each user provides an offered price for its own reliability to the network while the network provides congestion prices to users. The proposed algorithms converge to a tradeoff point between rate and reliability, which we prove to be a globally optimal one for channel codes with sufficiently large coding length and utilities whose curvatures are sufficiently negative. Under these conditions, the proposed algorithms can thus generate the Pareto optimal tradeoff curves between rate and reliability for all the users. The distributed algorithms and convergence proofs are extended for wireless MIMO multi-hop networks, in which diversity and multiplexing gains of each link are controlled to achieve the optimal ratereliability tradeoff.