README.md

---
license: mit
language:
- en
task_categories:
- image-classification
- tabular-classification
- tabular-regression
- time-series-forecasting
multilinguality:
- monolingual
tags:
- multimodal
- jamming-detection
- spectrograms
- time-series
- federated-learning
---

## FedJam Dataset

The FedJam dataset is a **multimodal dataset** for jamming detection and classification in wireless networks, combining time–frequency spectrogram images with
cross-layer network KPI time series. Each sample includes aligned **vision and time-series modalities**, allowing joint analysis of physical-layer signal behavior
and network-layer performance. The data are collected from a real over-the-air experimental testbed, under a variety of operating conditions, including multiple
adversarial jamming scenarios as well as normal operation without any jammer present, reflecting realistic wireless environments.

---

## Code

The dataset was used in the following paper “FedJam: A Multi-Modal Federated Learning Framework for Jamming Detection”, which has been accepted for
publication at the IEEE International Conference on Computer Communications (INFOCOM) 2026. 

The implementation and supporting code are publicly available here: https://github.com/panitsasi/fedJam

The research paper can be found here: https://arxiv.org/pdf/2508.09369

---

## Dataset Overview

- **Modality 1 (Vision)**: Spectrogram images  
  - Format: PNG  
  - Resolution: **224 × 224 × 3**  
  - One spectrogram image per sample

- **Modality 2 (Time Series)**: Network KPIs (WiFi)  
  - Format: CSV / structured sequences  
  - Fixed-length normalized multivariate time series
  - 256 measurements per KPI feature, per sample
  - Features:
    - `Time`
    - `Latency`
    - `Jitter`
    - `Packet Loss Count`
    - `Noise`
    - `SNR`

- **Labels**: Benign traffic and multiple jamming attack types

Each spectrogram image and KPI time series correspond to the **same time window**.

---

## Dataset Schema

Each dataset sample contains:

- **`spectrogram`** *(Image)*  
  Spectrogram image of shape **224 × 224 × 3**.

- **`kpis`** *(sequence / array)*  
  Multivariate time series of shape **[256, F]**, where:
  - `256` is the number of temporal measurements
  - `F` is the number of KPI features  
    (`Time`, `Latency`, `Jitter`, `Packet Loss Count`, `Noise`, `SNR`)

- **`label`** *(integer)*  
  Encoded class label:
  - `0`: Benign  
  - `1`: Jamming type A  
  - `2`: Jamming type B  
  - `3`: Jamming type C  
---

## Loading the Dataset

```python
from datasets import load_dataset
import matplotlib.pyplot as plt
import numpy as np

# Load the FedJam dataset
dataset = load_dataset("panitsasi/FedJam")

# Access train / test splits
train_data = dataset["train"]
test_data  = dataset["test"]

# Select one sample
sample = train_data[0]

# Extract modalities
image = sample["image"]          
timeseries = sample["timeseries"]  
label = sample["label"]          

print(sample.keys())
print("Timeseries shape:", len(timeseries), "x", len(timeseries[0]))
print("Label:", label)

ts = np.array(timeseries)
kpi_names = ["Latency", "Jitter", "Packet Loss", "Noise", "SNR"]

# Plot spectrogram and all KPIs
fig, axes = plt.subplots(1, 1 + ts.shape[1], figsize=(18, 3))

# Spectrogram
axes[0].imshow(image)
axes[0].set_title("Spectrogram")
axes[0].axis("off")

# KPI time series
for i in range(ts.shape[1]):
    axes[i + 1].plot(ts[:, i])
    axes[i + 1].set_title(kpi_names[i])
    axes[i + 1].set_xlabel("Time")
    axes[i + 1].set_ylabel("Value")

plt.tight_layout()
plt.show()

```
---

## Citation

If you use this dataset, please cite:
```python
I. Panitsas, I. Ofeidis, and L. Tassiulas,
“FedJam: Multimodal Federated Learning Framework for Jamming Detection”
arXiv:2508.09369 [cs.NI], 2025. doi:10.48550/arXiv.2508.09369.
```

---