Date of Award

August 2020

Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Biao Chen

Subject Categories



Radio frequency (RF) communications are an integral part of many situational awareness applications. Sensing data need to be processed in a timely manner, making it imperative to have a robust and reliable RF link for information dissemination. Moreover, there is an increasing need for exploiting RF communication signals directly for sensing, leading to the notion of multi-function RF.

In the first part of this dissertation, we investigate the development of a robust Multiple-Input Multiple-Output (MIMO) communication system suitable for airborne platforms.Three majors challenges in realizing MIMO capacity gain in airborne environment are addressed: 1) antenna blockage due largely to the orientation of the antenna array; 2) the presence of unknown interference inherent to the intended application; 3) the lack of channel state information (CSI) at the transmitter. Built on the Diagonal Bell-Labs Layered Space-Time (D-BLAST) MIMO architecture, the system integrates three key design approaches: spatial spreading to counter antenna blockage; temporal spreading to mitigate signal to interference and noise ratio degradation due to intended or unintended interference; and a simple low rate feedback scheme to enable real time adaptation in the absence of full transmitter CSI. Extensive experiment studies using a fully functioning $4\times 4$ MIMO system validate the developed system.

In the second part, ambient RF signals are exploited to extract situational awareness information directly. Using WiFi signals as an example, we demonstrate that the CSI obtained at the receiver contains rich information about the propagation environment. Two distinct learning systems are developed for occupancy detection using passive WiFi sensing. The first one is based on deep learning where a parallel convolutional neural network (CNN) architecture is designed to extract useful information from both magnitude and phase of the CSI. Pre-processing steps are carefully designed to preserve human motion induced channel variation while insulating against other impairments and post-processing is applied after CNN to infer presence information for instantaneous motion outputs. To alleviate the need of tedious training efforts involved in deep learning based system, a novel learning problem with contaminated sampling is formulated. This leads to a second learning system: a two-stage solution for motion detection using support vector machines (SVM). A one-class SVM model is first evaluated whose training data are from human free environment only. Decontamination of human presence data using the one-class SVM is done prior to motion detection through a two-class support vector classifier. Extensive experiments using commercial off-the-shelf WiFi devices are conducted for both systems. The results demonstrate that the learning based RF sensing provides a viable and promising alternative for occupancy detection as they are much more sensitive to human motion than passive infrared sensors which are widely deployed in commercial and residential buildings.


Open Access

Included in

Engineering Commons