Partially observed cross-layer optimization for vehicular communications

Vehicular Ad-hoc Networks (VANETs) that are characterized by frequently changed topology, worse signal noise ratio, and non-negligible Doppler effect introduce new non-trivial challenges to Quality-of-Service (QoS) provisioning. The methodology of cross-layer optimization aims to jointly optimize the working behaviors over different layers to achieve a better network performance, eg, throughput and transmission latency. This paper presents a novel cross-layer optimization method based on Partially Observed Markov Games (POMG) to improve optimization decision against the inaccurate observed context caused by high-speed movement, sensor errors, and other unavoidable reasons. POMG extends Markov Decision Process (MDP) and Partially Observed Markov Decision Process (POMDP) to dynamically adjust the concerned actions (eg, transmission range, contention window, and bit rate) according to the observed traffic density and thus can improve optimization performance at several aspects, eg,, throughput, channel utilization, delay, and total number of neighbor nodes. The extensive simulations show that POMG could harvest a high entire gain compared with the traditional fixed policy and the other cross-layer optimization schemes.