RNN Learning for Dynamic Selection of Channel Access Scheme in FANETs

Medium access protocols (MAC) are used to detect/avoid collision between the nodes in a flying ad-hoc network (FANET). Numerous MAC protocols are available, including channelization protocols, controlled access protocols, random access protocols., etc. We have implemented a Recurrent Neural Network using Reinforcement Learning (RNNL) to design a universal MAC protocol for FANET architecture that leverages machine learning techniques to develop an intelligent MAC protocol for optimal transmission. We look at the issue of sharing time slots among numerous time-slotted networks that use various MAC protocols. There are multiple nodes, one of which is our node that RNNL supports to decide which protocol to pick next based on the previous history (memory). The nodes available in the network are using protocols such as ALOHA, Carrier Sensing Multiple Access with Collision Avoidance (CSMA/CA), and Reservation base protocols as reference model. From a series of observations and interactions with those nodes in the network, our RNNL node will decide which protocol to use for the ideal transfer of data with the least overall transmission time to reduce total network traffic. We test our node by allowing to act on a node and compare the performance of the node with the other nodes in the network. Average transmission time, average transmission tries, and whether the transmission was successful, these parameters are compared with the conventional algorithms to analyze the efficiency of the proposed protocol.