A robust neuro-adaptive congestion control scheme with respect to exogenous disturbances and delay

A novel neuro-adaptive congestion controller is presented, capable of achieving certain time-quality characteristics, by regulating the per packet round trip. The controller is implemented at the source and is proven robust against modelling imperfections, exogenous disturbances (UDP traffic) and delays (propagation, buffering). The proposed source control theoretic framework is comprised of three dominant modules: i) an on-line desired round trip estimator, whose main objective is to output reliable piecewise constant predictions of feasible (i.e., achievable) round trip times; ii) an adaptive and saturated transmission rate controller, capable of precise tracking the desired round trip time commands that are downloaded from the first module and iii) a future path congestion level estimator, which provides critical information to the aforementioned two modules.