Machine Learning
Signal Segmentation & Classification
What: Analyze the RF spectrum occupancy through signal detection and classification
Why: Regulate and optimize spectrum utilization and sharing
How: Deep learning based wideband signal recognition
Bayesian Learning
What: Learn approximate a posteriori probabilities (APP) of many random variables from observations
Why: Perform APP computations with polynomial complexity
How: Message passing (local computation at nodes and propagation of results along edges)
Combining Bayesian and Deep Learning
How: Unfold and map Bayesian learning graph to equivalent deep neural network; select and modify node computation to include training
Why: Complexity reduction & enhancement of approximation accuracy
Possible Applications to Sensing & Communications
Application 1: Cell-free massive MIMO
What: Joint channel estimation & data detection
Where: RX including distributed systems as cell-free massive MIMO
Application 2: Random multiple access
What: Joint activity detection, channel estimation & data decoding
Where: Random multiple access for, e.g., IoT