PyJama: Differentiable Jamming and Anti-Jamming with NVIDIA Sionna

TL;DR

This work addresses the lack of a framework for studying ML-enabled jamming and anti-jamming in realistic wireless systems by introducing PyJama, a differentiable library that extends NVIDIA Sionna to simulate jamming scenarios in MIMO-OFDM with mobility and coding. PyJama enables end-to-end learning of jamming strategies via SGD by optimizing a trainable power-allocation vector under a constraint such as $\|\bm{\rho}\|_1/N_s \le \rho_{\max}$, and demonstrates that learned jammers outperform uniform barrage across both uncoded and coded settings while exposing weaknesses in POS-based mitigation. The paper showcases extensive simulations, including time-domain and frequency-domain jamming, mobility effects, and CP-related interference phenomena, to reveal nontrivial attack strategies and their implications for anti-jamming design. Overall, PyJama provides a practical, differentiable platform for researching jamming threats and designing robust mitigation techniques in realistic wireless systems, highlighting the need for more sophisticated anti-jamming strategies beyond simple spatial filtering. The work thus enables realistic, differentiable exploration of jamming and anti-jamming research with direct implications for safeguarding future wireless networks.

Abstract

Despite extensive research on jamming attacks on wireless communication systems, the potential of machine learning for amplifying the threat of such attacks, or our ability to mitigate them, remains largely untapped. A key obstacle to such research has been the absence of a suitable framework. To resolve this obstacle, we release PyJama, a fully-differentiable open-source library that adds jamming and anti-jamming functionality to NVIDIA Sionna. We demonstrate the utility of PyJama (i) for realistic MIMO simulations by showing examples that involve forward error correction, OFDM waveforms in time and frequency, realistic channel models, and mobility; and (ii) for learning to jam. Specifically, we use stochastic gradient descent to optimize jamming power allocation over an OFDM resource grid. The learned strategies are non-trivial, intelligible, and effective.

Figures (9)

• Figure 1: PyJama simulation pipeline (channel estimation not shown). In Sionna, the frequency-domain and time-domain jammers are implemented as (generalized) frequency-domain and time-domain interference classes, respectively.
• Figure 2: Foreground: Illustration of an OFDM resource grid that uses the PilotPatternWithSilence. Background: Corresponding illustration of the different types of frequency domain jammers (barrage jammer, pilot jammer, data jammer, sparse jammer). The jammer transmit symbols can be drawn from a uniform distribution, a Gaussian distribution, or a QAM constellation.
• Figure 3: Left: Uncoded bit error rates (BERs) vs. SNR for basic MIMO jamming and anti-jamming. Right: Corresponding block error rates (BLERs).
• Figure 4: Left: POS mitigation effectiveness for different levels of jammer mobility. Right: POS mitigation effectiveness for different levels of UE mobility.
• Figure 5: Left: Error rates of CP-compliant and CP-violating single-antenna jammers under one-dimensional POS mitigation. Right: Normalized histogram of the fraction of receive interference per spatial dimension (in descending order); the jammer simulated in the frequency domain is not shown.