Delay-optimal MAC level LTE scheduling using deep reinforcement learning for downlink
Somayajula Venkata, Someshwar Rao
MetadataShow full item record
The topic of this thesis is to introduce a novel method of LTE schedulingand comparing it to previous research on the same topic. The motivationis to introduce deep reinforcement learning techniques to downlinkLTE scheduling in order to find a delay optimal method of schedulingusers in an LTE network. First, the scope of the problem needs to bedefined in order to present a solution for the problem at hand. A singlenetwork cell with a limited number of users and Physical ResourceBlocks is considered, in order to constrain the complexity of the DeepRL algorithm. From thereon, three popular scheduling algorithms arecompared to the the new solution, namely, MaxCI scheduling, ProportionalFair Scheduling and Max Weight Scheduling. It is shown thatthe Deep RL Scheduler prioritises delay minimisation by compromisingthroughput in order to be delay optimal.