The S-ICDF Dataset: Sionna-Simulated Dynamic Interference Characterization and Direction Finding
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: We're looking at the paper: "The S-ICDF Dataset".
Dev: Here's the gist of the paper. Robust interference monitoring (i.e., detection, classification, characterization, and direction finding) is therefore essential to identify and localize anomalous signals.
Rosa: To address this gap, we create and publish S-ICDF, a large-scale indoor interference dataset generated with Sionna, a GPU-accelerated simulation library for physical-layer wireless communications.
Dev: Dataset Creation and Scope The S-ICDF dataset covers 102 interference configurations, including diverse antenna array patterns, bandwidths, and simulation settings such as noise level and reflection depth.
Rosa: The simulation environment models an indoor industrial halllike scenario using Sionna’s ray-tracing engine to capture dense multipath conditions typical of indoor industrial settings.
Paper summary: Rosa: So we've seen how they built this S-ICDF Dataset, and now we're going to talk more about the core thesis of "The S-ICDF Dataset: Sionna-Simulated Dynamic Interference Characterization and Direction Finding". The paper is essentially arguing that to tackle jamming and spoofing threats, you absolutely need a massive dataset that captures the real variability of these signals >
Dev: Right. They are pointing out the difficulty in collecting this data because intentional jamming isn't something you can just do legally, and when you try to attribute interference to a specific source, it gets really complicated by how the signal propagates, the hardware involved, and all those environmental factors >
Rosa: The core claim of this paper is that machine learning promises better performance in these complex environments if we have large datasets that actually represent that variability >
Dev: And they're saying collecting that data in the real world is really tough. This dataset, S-ICDF, solves that problem by simulating realistic scenarios using Sionna, which is a GPU-accelerated simulation library for physical layer wireless communications >
Rosa: So what this means for us right now is that we don't have enough realistic data to train strong ML models for interference monitoring because the real world is too messy and hard to access >
Dev: This dataset fills that gap by providing a controlled, simulated environment where they can systematically test how different signal parameters influence the IQ signal structure and the performance of localization algorithms >
Rosa: They're not just throwing data at a black box. They are building a pipeline that shows exactly how changing one setting affects everything from the raw received signal to the final output of a direction finding algorithm >
Dev: It’s about creating a controlled testbed to understand the relationship between those physical parameters and the resulting performance metrics, which is crucial for developing better detection and localization systems >
Rosa: So in short, they claim that this large-scale indoor interference dataset is the necessary ingredient to move machine learning from theory into practical application for identifying and localizing anomalous signals >
Conclusion: Dev: Wrapping up this discussion on "The S-ICDF Dataset: Sionna-Simulated Dynamic Interference Characterization and Direction Finding". The authors are Christian Wielenberg, Lucas Heublein, Jonathan Ott, Alexander Mattick, Nisha L. Raichur, Jonas Pirkl, Lukas Schelenz, Tobias Feigl from the Fraunhofer Institute for Integrated Circuits >
Rosa: Their work is about providing a way to get past the data scarcity problem in interference monitoring by simulating realistic industrial environments using Sionna to create this S-ICDF Dataset >
Dev: The implication here is that we can start training more reliable machine learning models for detecting and localizing anomalous signals because we have this structured data set where the simulation parameters are fully defined >
Rosa: It means that when you look at future systems designed for robust interference monitoring, they should be tested against scenarios similar to what's captured in this dataset >
Dev: So, in simple terms, they're saying that having a large dataset generated under these specific controlled simulation conditions is the key ingredient to making ML models useful for finding and pinpointing those weird radio signals >
Christian Wielenberg, Lucas Heublein, Jonathan Ott, Alexander Mattick, Nisha L. Raichur, Jonas Pirkl, Lukas Schelenz, Tobias Feigl, George Yammine, Christopher Mutschler, Felix Ott
Fraunhofer Institute for Integrated Circuits IIS
eess.SP, cs.AI
Submitted: 2026-07-03
Updated: 2026-07-03
Comments: 6 pages, 9 figures
Journal ref: IEEE Intl. Conf. on Indoor Positioning and Indoor Navigation (IPIN), Rome, Italy, October 2026
DOI: 10.1109/IPIN72673.2026.11772556
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 82/100
The gist: Robust interference monitoring (i.e., detection, classification, characterization, and direction finding) is therefore essential to identify and localize anomalous signals.
Key concepts
- S-ICDF Dataset
- This is a large collection of simulated indoor interference scenarios created using Sionna, a GPU-accelerated simulation library. It includes 102 different configurations of antennas, bandwidths, and noise levels to test how signal parameters affect performance.
- Direction Finding (DF)
- DF is the process of determining the direction from which a signal is arriving. The paper tests classical methods like MUSIC and ESPRIT against machine learning models to see which performs best at locating anomalous signals within the dataset.
- XceptionTime Model
- This is a machine learning model trained to characterize interference by predicting its specific characteristics. It uses cross-entropy loss for classification (identifying the type of interference) and mean squared error for estimating angles like azimuth and elevation.
- OAT Strategy
- One-ata-time parameter variation is the simulation method used to generate data. This means only one variable—like noise level or reflection depth—is changed at a time while keeping all other simulation settings fixed, allowing researchers to isolate the effect of each parameter.
Terminology
Summary
Robust interference monitoring (i. p. 0e., detection, classification, characterization, and direction finding) is therefore essential to identify and localize anomalous signals. p. 0 To address this gap, we create and publish S-ICDF, a large-scale indoor interference dataset generated with Sionna, a GPU-accelerated simulation library for physical-layer wireless communications. p. 0
Dataset Creation and Scope
The S-ICDF dataset covers 102 interference configurations, including diverse antenna array patterns, bandwidths, and simulation settings such as noise level and reflection depth. The simulation environment models an indoor industrial halllike scenario using Sionna’s ray-tracing engine to capture dense multipath conditions typical of indoor industrial settings. p. 2 The dataset is designed to facilitate a detailed assessment of how individual signal and channel parameters influence IQ signal structure and the performance of signal localization algorithms. p. 2
Simulation Methodology
The simulation pipeline systematically quantifies how individual simulation parameters affect both the received signal structure and the performance of direction finding algorithms. A one-ata-time (OAT) parameter-variation strategy is employed, where only a single parameter of interest is varied over its predefined range while all remaining parameters are fixed at their respective default values. p. 2 The receiver is modeled as a uniform antenna array with either a 2×2 planar geometry or an 8×1 linear geometry, and the received signal is sampled over a simulation bandwidth of 100 MHz, resulting in 1,024 complex IQ samples per snapshot. p. 2
Evaluation Metrics and Baselines
The dataset is used to benchmark ML models against classical estimation and direction finding (DF) methods, including MUSIC, ESPRIT, and CAPON. For characterization tasks, an XceptionTime model is trained to predict the interference characteristics using a cross-entropy loss for classification and mean squared error (MSE) loss for regression of azimuth and elevation angles. p. 3 The overall objective function balances the losses associated with characterization, DF, position, and distance estimation. p. 3
Performance Findings
The evaluation shows that XceptionTime achieves 99.89% interference classification accuracy and competitive DF performance. p. 5 ESPRIT provides the most robust classical baseline with a mean error of 0. p. 578◦ for azimuth estimation under default settings. p. 4 The ML-Based DF model achieves a mean error of 1. p. 585◦ in azimuth, 2. p. 333◦ in elevation, and 0. p. 326 m in position. p. 1 The results indicate that the interference modulation has a noticeable impact on DF accuracy, with ESPRIT providing the most robust overall performance. p. 4 The evaluation also shows that below −8 dB SNR, the azimuth error increases strongly for all methods. p. 4
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Robust Interference Monitoring System Development: The improved system can perform detection, classification, characterization, and direction finding of anomalous radio frequency signals by leveraging a large-scale dataset like S-ICDF to train machine learning models.
-
Enhanced ML Model Training: The system can be trained on the 440,821 training samples from the S-ICDF dataset using an XceptionTime model to predict interference characteristics and regress azimuth and elevation angles, achieving 99.89% accuracy in classification.
-
Classical Baseline Performance Benchmarking: The system can compare its performance against classical estimation methods such as MUSIC, ESPRIT, and CAPON to establish a robust baseline for direction finding under various signal parameters like SNR and antenna geometry.
-
Sensitivity Analysis Capability: By systematically varying individual simulation parameters within Sionna, the system can perform a
one-ata-time
parameter variation strategy to attribute performance changes clearly to specific variables like reflection depth or antenna spacing. -
Adaptive Direction Finding Under Motion: The system can be designed for time-series based direction finding by simulating an antenna platform that moves through the environment during data acquisition, enabling the evaluation of DF under realistic time-varying multipath conditions.
Sources
- Sionna: An Open-Source Library for Next-Generation Physical Layer Research
- XceptionTime: A Novel Deep Architecture based on Depthwise Separable Convolutions for Hand Gesture Classification
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization