Identification and Classification of Automotive Radar Interference using Data-driven Methods
| Title | Identification and Classification of Automotive Radar Interference using Data-driven Methods |
|---|---|
| Summary | This is a collaboration with Radar Reticence. The project investigates how different detection methods handle radar interference, a crucial factor for ensuring accurate perception in automated driving. |
| Keywords | Automotive radar, ADAS, interference, anomaly detection,Property "Keywords" has a restricted application area and cannot be used as annotation property by a user. |
| TimeFrame | Spring 2026 |
| References | Radar interference: https://www.analog.com/en/resources/analog-dialogue/articles/automotive-radar-sensors-and-congested-radio-spectrum-an-urban-electronic-warfare.html
Equipment: https://www.ti.com/tool/AWR2944EVM, and https://www.ti.com/tool/DCA1000EVM, |
| Prerequisites | Programming for data analysis, experience with Matlab/Python (with ML-libraries), signal processing foundations |
| Author | |
| Supervisor | Emil Nilsson, Sławomir Nowaczyk, Elena Haller |
| Level | Master |
| Status | Open |
Background
Transportation systems are currently becoming automated for improved efficiency and safety. The automation requires various types of sensors, and for vehicle automation, cameras and radars are common. The cameras and radars are integrated together with computing devices into Advanced Driving and Assisting Systems (ADAS). Drivers become increasingly dependent on ADAS, and eventually the vehicle will require no human operator at all.
Transportation systems are becoming increasingly automated to improve efficiency and safety. These systems rely on integrated sensor suites — cameras and radars — feeding Advanced Driving and Assistance Systems (ADAS). Radar sensors are particularly valuable because they function well in adverse weather and low-light conditions. Unlike passive cameras, radars are active sensors: they transmit radio-frequency signals and receive weak echoes reflected by the environment. Because received echoes are weak, radars are vulnerable to other transmitters operating in the same frequency band; interference can mask echoes or create false detections. As the number of radar-equipped vehicles (and radars per vehicle) grows, interference will become more frequent and varied.
A better empirical understanding of how automotive radar responds to different interference types is required. Radar hardware is complex and hard to model in full fidelity, therefore controlled experiments and data-driven analysis are preferred approaches for a master thesis.
Project Goal
Develop and evaluate a data-driven pipeline that (1) detects the presence of interference in automotive radar data and (2) classifies the interference into meaningful types (e.g., FMCW, CW, GMSK, pulsed, wideband noise). The project will produce an experimental dataset, baseline algorithms, and an evaluation of accuracy and robustness in controled environment.
Research Questions
1. Which radar data representations (ADC/IQ, range profiles, range-Doppler maps, range-Doppler-angle cubes) provide the “best” information for interference detection and classification?
2. How well can an autoencoder (can be smth else) trained only on “good” data detect interference through reconstruction-error–based anomaly scores, and what are suitable thresholds or statistical criteria to separate “clean” from interfered data?
3. Can residual-based features (e.g., reconstruction error, spectral characteristics of the residual, anomaly scores over time) support accurate supervised classification of interference types?
4. How do different interference types (FMCW overlap, CW tone, pulsed bursts, wideband noise) look like in autoencoder residuals, and which features are most important for classification?
