Passive Radar Aided Communication Channel APS Estimation Using DL:

• Predicted communication channel APS directly from radar channel APS based on DL.
• Work on communication channel eigenvector predictions.

Passive Radar Aided Communication Channel Covariance Estimation Using DL:

• Proposed an Encoder-Decoder architecture for radar channel covariance to communication channel covariance translation for moving vehicles.
• Achieved a similarity value of > 85% beween the azimuth power spectrum (APS) of the predicted covariance and the real APS.

Deep Learning (DL) Based 5G Air-to-Ground Network Design and Optimization:

• Proposed a double-DNN architecture for 5G A2G network behavior approximation as well as network deployment optimization.
• Optimized A2G network parameters such as ISD, antenna tilts, etc. for high user throughput.
• Achieved similar or better performance w.r.t user throughput and SINR comparing with the deployments in the dataset.

Collision-free UAV Navigation with a Monocular Camera Using Deep Reinforcement Learning:

• Proposed a UAV navigation system with a monocular camera using deep reinforcement learning (DRL).
• Reduced 25% of the flight distance and avoided 50% of the unnecessary turns for the UAV.
• Alleviated wrong predictions from the deep networks by combining object detection.

Efficient Drone Mobility Support Based on (Deep) Reinforcement Learning:

• Proposed a Q-learning based handover (HO) scheme for UAVs, striking a balance between the connectivity quality and HO cost.
• Extended the work to DQN based HO scheme, which allows for larger state and action space.
• Realized 80% reduction in the number of HOs while guaranteed reliable connectivity.

TA of Course: Probability and Random Process:

Help organize class, review homework and assignments, hold office hours, etc.