A neuro-adaptive TCP-like protocol with communication channels adaptation under cost constraints

In this paper we propose a new approach to control congestion in the Internet via regulating the packet round trip for each application, without violating a maximum cost per packet value, prescribed by the user. For this purpose, we construct an on-line model for the packet round trip time, depending on source rate and congestion measure. The proposed algorithm is comprised of a two level scheme. At the high level the packet round trip is estimated, while at the low, a neural net rate controller is designed to achieve the estimated values. To improve the application sending time, we introduce the notion of communication channels, whose number is updated via an appropriately designed algorithm, parallelizing the transmission process. Simulation studies illustrate the approach.