We propose a decentralized stochastic control solution for the broadcastmessage dissemination problem in wireless ad hoc networks with slow fadingchannels. We formulate the control problem as a dynamic robust game which iswell justified by two key observations; first, the shared nature of thewireless medium which inevitably cross-couples the nodes' forwarding decisions,thus binding them together as strategic players; second, the stochasticdynamics associated with the link qualities which renders the transmissioncosts noisy, thus motivating a robust formulation. Given the non stationarityinduced by the fading process, an online solution for the formulated game wouldthen require an adaptive procedure capable of both convergence to and trackingstrategic equilibria as the environment changes. To this end, we deploy thestrategic and non stationary learning algorithm of regret tracking, thetemporally adaptive variant of the celebrated regret matching algorithm, toguarantee the emergence and active tracking of the correlated equilibria in thedynamic robust forwarding game. We also make provision for exploiting thechannel state information, when available, to enhance the convergence speed ofthe learning algorithm by conducting an accurate transmission cost estimation.This cost estimate can basically serve as a model which spares the algorithmfrom extra action exploration, thus rendering the learning process more sampleefficient. Simulation results reveal that our proposed solution excels in termsof both the number of transmissions and load distribution while alsomaintaining near perfect delivery ratio, especially in dense crowdedenvironments.
展开▼
