Date of Award

8-2018

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

Committee Chair

Mark R. Bell

Committee Member 1

James V. Krogmeier

Committee Member 2

David J. Love

Committee Member 3

Michael D. Zoltowski

Abstract

The delay-Doppler resolution of a radar system determines its ability to separate targets that are spatially close together and have similar radial velocities with respect to the radar. For this reason, the design of radar signals and signal processing with good delay-Doppler resolution has long been an ongoing active area of research. Radar systems typically use matched-filters to maximize the received signal-to-noise ratio and maximize target detection performance, in which case the delay-Doppler resolution of the radar waveform is given by the ambiguity function of the transmitted waveform.

Neurobiologists studying echolocation in bats have noted their outstanding resolution capabilities of these animals, and it appears these resolution capabilities are superior to that of the matched filter. While the details of biological echolocation processing are not fully understood, neurobiological experiments indicate that the processing being done is not matched filtering. This suggests that animal echolocation may be suboptimal in terms of object detection performance, but may have superior delay-Doppler resolution when compared to matched-filter processing.

In this work, we study the design of waveforms and signal processing mimicking the characteristics of biological echolocation. In initial work done by Rasool and Bell, a simple nonlinear processing model based on known characteristics of the mustached bat demonstrated significant increase in delay-Doppler resolution while only suffering a modest decrease of approximately 0.5 dB in detection performance. This model mimicked the echolocating bat by processing different segments of the transmitted waveform–an up-chirp and a down-chirp– separately and then combining the results with point-wise operations on the resulting delay-Doppler maps. While this approach showed significant improvement in delay-Doppler resolution over the matched-filter, the pointwise nonlinear operation failed to take into account the neighboring delay-Doppler cell responses. Real biological systems do not ignore the neighboring responses. They use information in a neighborhood surrounding the response in a process called lateral inhibition to sharpen resolution. Lateral inhibition is present in neural processing in both the eye and the ear to sharpen resolution response, and any system which ignores the neighborhood response cannot fully exploit the resolution characteristics present in the received signal. Therefore, we investigate neighborhood nonlinear processing using the inverse filter and blanker filter. We show that using neighborhood nonlinear processing, we can separate two targets that are much more closer together. We also propose that in future work, convolutional neural networks should be investigated as an approach to this problem. Further refinement of inverse filter techniques may also be considered.

Download

Share

COinS