Date of Award

2023

Degree Type

Thesis

Degree Name

Master of Science in Electrical Engineering (MSEE)

Department

Electrical, Computer, and Biomedical Engineering

First Advisor

Ken Q. Yang

Abstract

Radar sensors are an essential component of Advanced Driver Assistance Systems (ADAS). They use radio waves to monitor the surroundings of a vehicle, capturing data that can be utilized to provide assistance to the driver. Major challenges of this procedure are the high complexity and the considerable power consumption that the transfer of radar data to subsequent processing units requires. This work addresses these issues by using deep learning architectures to compress the raw data captured by radar before it is being transmitted. As proceeding step so-called Range-Angle (RA) Heatmaps are reconstructed from the compressed data. These heatmaps present an effective format for the visualization of radar data and serve as input for object detection and identification tasks.

In order to implement the intelligent data compression and heatmap generation, a joint model combining the encoding-unit of a neural autoencoder, and the U-Net architecture is used. The data to train, validate and test the neural network model is obtained in two steps: First raw data samples are collected using a Texas Instruments AWR1243 Frequency Modulated Continuous Wave (FMCW) radar testbed. Thereafter, analytical radar imaging algorithms, including a 3-dimensional Fast Fourier Transform and our proposed detection and clustering framework, are employed to compute RA Heatmaps from the raw data.

The results indicate that our proposed neural network architecture is able to achieve a significant reduction of the initial radar data while still maintaining all relevant features required for object recognition tasks. Excellent accuracy scores exceeding 99 % on input data that has been compressed by a factor of four validate the capabilities of the model and underline its potential for radar data processing in the context of ADAS.

Download

Available for download on Friday, September 05, 2025

Share

COinS