DEEP LEARNING COGNITIVE TDMA SCHEDULING & ANTI-JAMMING COMMUNICATIONS WITH REAL-TIME JAMMER IDENTIFICATION AND MACHINE-LEARNING CONTROLLED RECONFIGURABLE ANTENNA ARRAY BEAMFORMING

To address the performance degradation of fixed scheduling policies used by TDMA-based tactical networks in contested and congested environments, a distributed cognitive TDMA scheduling and anti-jamming protocol is proposed. Technical approach consists of: 1. Developing a deep learning (DL) classifier for jammer/interference identification and automatic detection of network congestion that exchanges information among distributed nodes in an ad-hoc network. 2. Developing a deep reinforcement policy learning algorithm to achieve cognitive anti-jamming communications across channels and timeslots. 3. Developing a distributed cognitive TDMA scheduling algorithm that takes cues from the deep reinforcement learning anti-jamming policy, the DL signal classifier and the congestion/jamming detector. 4. Developing a machine-learning controlled reconfigurable antenna array beamforming technique for interference avoidance that takes cues from the jammer identification algorithm and the congestion/jamming detector. 5. Developing realistic models of jamming, interference, congestion and networks to evaluate performance of the designed machine-learning and scheduling algorithms under various M&S scenarios. 6. Prototyping as FPGA IPs the components of the designed algorithms and protocols most suitable for hardware implementation. 7. Implementing a prototype of the developed cognitive TDMA scheduling and anti-jamming protocol on a COTS SDR to provide a demonstration of the cognitive communications capability in contested and congested environments.