A Cognitive Networking Technique for LTP Segmentation

The time required to reliably deliver a data block with the Licklider Transmission Protocol (LTP) depends on the frame losses occurring during the transmission. LTP establishes overlay links that involve one or more transmission sections at the underlay. Because the state of these sections is time-dependent, LTP works without specific knowledge of the underlay. This property creates challenges to the block segmentation as it involves a tradeoff between the overall header overhead and the higher loss rates of large segments. In this paper, a practical segment loss mitigation method is proposed that benefits block delivery times by deciding segment lengths prior to each block transmission. This goal is achieved with a cognitive networking approach to the problem that leverages the parallel processing and storage capabilities of neuromorphic computing, which is prospectively adequate for onboard systems of known constraints in size and electrical power. The key advantage of this online method is that it does not need complete information about the channel properties, models, protocols, nor state. Instead, the method learns autonomously the best segment length for the current conditions by mapping the delivery performance of prior blocks to the synapse strengths of a spiking neural network, which is then used to generate new decisions for the next segments. Simulation results provide an indication of the performance benefits.