Embedding AI models for GNSS signal spoofing on Unix boards
Porting deep learning models on custom NPU-enabled boards.
Requirements
- B.Sc. in Computer Science, Telecommunications, Electronics or similar
- Knowledge of Python, Unix systems
- Software development skills
- Basic concepts of image processing
- Basic concepts of data science, concerning data analysis, processing and machine learning
Description
Global Navigation Satellite System (GNSS) spoofing has emerged as a critical concern in satellite-based positioning, navigation, and timing (PNT) applications. As our reliance on GNSS for various industries such as transportation, agriculture, telecommunications, and defense continues to grow, the threat of malicious actors spoofing GNSS signals becomes increasingly important. Latest advances in spoofing detection often exploit machine learning or deep learning techniques, which are often applied on high-performance systems able to process large amount of data in a near real-time fashion. Given the recent advances in edge computing and the increasing hardware costs, optimizing such solutions for embedded low-cost devices is rapidly growing in interest.
The purpose of this work is to optimize existing deep learning models (e.g., CNNs such as ResNet) and techniques in order to efficiently run them on a iMX8 board, leveraging the integrated Neural Processing Unit (NPU). The project will comprise several steps, from the analysis of the literature and current state-of-the-art approaches, to the porting of existing models. The internship will include the following activities: (i) data identification and understanding (ii) data analysis and pre-processing (iii) model adaptation and optimization for NPU- enabled devices using existing frameworks (e.g., PyTorch, Tensorflow), (iv) performance evaluation, (v) result analysis and visualization.