AI-Assisted Channel Modeling

Through laborious human visual inspection, we are able to “stitch” together hundreds of images of the channel coupled with an environment map, reconstructing an electromagnetic “view” of the channel environment. We believe that visual inspection can be replaced by artificial intelligence (AI), bridging the gap between data collection and data processing. To that end, we have mounted a 360° camera and a 360° Lidar, on the rover robot. They can reconstruct a 3D model of the environment so that we can use AI to classify salient objects in the environment that act as electromagnetic scatterers, reflecting radiation emitted from the transmitter into the receiver.

Classifying objects in the (a) indoor and (b) outdoor environments is shown below: image (left) and classified image (right). Each object classified is labeled with a confidence level up to 100%.

 

 

Armed with the speed and autonomous data collection, we are uniquely slated to pioneer this new AI space. Recently, we developed a beamtracking algorithm that was tested directly on the measurement, adding a greater level of confidence to the results  versus an often-simplified channel model.

More details about our new beamtracking algorithm can be found in the reference below:

[1] S. Blandino, J. Senic, C. Gentile, D. Caudill, J. Chuang, A. Kayani, “Markov Multi-Beamtracking in SU-MIMO with 60 GHz Mobile Channel Measurements,” Submitted to IEEE Open Journal on Vehicular Technology, 2021.

The new measurement campaign was focused on human sensing at mmWave. Preliminary results could be found here.