Jammer Mitigation Meets Machine Learning

Jammer mitigation is not exactly magic, but it sure feels like it. (Image made with Midjourney.)

Short Description

As wireless communication becomes an indispensable pillar of modern infrastructure, jamming attacks turn into a crucial security threat. The widespread adoption of safety-critical technologies such as autonomous driving or industry 4.0 will therefore depend (among many other things) on our ability to mitigate jammers.

The IIP group has been working on jammer mitigation for a few years now, and has already proposed several novel methods for jammer mitation, such as Joint Jammer Mitigation and Data Detection (JMD) [1] or Jammer Mitigation via Subspace Hiding (MASH) [2]. However, these methods are essentially "classical" signal processing methods: They either pose and solve an optimization problem with first-order methods, or they perform a series of linear transformations based on signal theory.

The goal of this project would be to carry jammer mitigation to the next level by harvesting the power of machine learning. Machine learning has transformed many fields in recent times, and wireless communications is no exception. A prime example is Sionna, a TensorFlow library that has recently been developed by NVIDIA. Sionna allows the simulation of fully differentiable (and hence trainable) wireless communiation systems.

In this project, the student would use Sionna to make existing jammer mitigation methods trainable, and to come up with completely new trainable receive architectures. On the flip side, adversarial training cold be utilized to answer such questions as: What is the worst jammer?

[1] G. Marti and C. Studer, "Joint Jammer Mitigation and Data Detection for Smart, Distributed, and Multi-Antenna Jammers," to be presented at IEEE ICC 2023

[2] G. Marti and C. Studer, "Universal MIMO Jammer Mitigation via Secret Temporal Subspace Embeddings," arXiv:2305.01260

Status: Available

Looking for a Semester/Master student

Contact: Gian Marti

Prerequisites

Basics understanding of digital communication (such as acquired in the lectures Kommunikationssysteme, Communication and Detection Theory, Wireless Communications, or similar)

Familiarity with Python and MATLAB

Prior experience with TensorFlow is advantageous but not required

Character

60% Programming

40% Algorithm Development

Professor

Christoph Studer

Links

↑ top