Challenge 9: Device Classification Under Hardware Impairments

Abstract

This challenge introduces a controlled RF fingerprint dataset designed to evaluate device classification under hardware-induced signal distortions. The dataset is generated from OFDM transmissions affected by residual hardware impairments, Carrier Frequency Offset (CFO), and Symbol Timing Offset (STO).

Multiple feature families are extracted in the frequency domain to capture device-specific signal characteristics. The framework supports supervised classification and enables systematic evaluation of robustness under varying impairment conditions.

The objective is to assess how well machine learning-based classification methods can distinguish devices when the observed signals are distorted by realistic hardware imperfections, reflecting practical deployment conditions in emerging 6G systems.

Dataset

The dataset is available at this link: https://doi.org/10.5281/zenodo.19481618

The dataset consists of OFDM signal realizations generated from two transmitting devices operating under distinct hardware impairment conditions. Each device is characterized by a unique combination of residual hardware impairments (Residual Hardware Impairments (RHI)), CFO, and STO.

Signals are generated using Binary Phase Shift Keying (BPSK)-modulated OFDM waveforms over Rayleigh fading channels with additive white Gaussian noise. The receiver captures the impaired signals, which are transformed into the frequency domain prior to feature extraction.

Residual hardware impairments refer to non-idealities such as amplifier nonlinearities, I/Q imbalance, and phase noise that remain after calibration and act as device-specific fingerprints.

The impairment parameters are configured as follows:

• Residual hardware impairment coefficients: κ₁ = 1, κ₂ = 1.3
• Carrier frequency offsets: 1.25 kHz and 2.5 kHz
• Symbol timing offsets: 5 and 10 samples
• Noise variance: σ² = 10⁻⁵

Each transmitting device has 1000 OFDM realizations, resulting in a balanced dataset.

For each received signal, multiple feature families are extracted, including:

• Power Spectral Density (PSD) statistics
• Spectral flatness and bandwidth
• Statistical descriptors of I/Q components
• L-moments
• Autocorrelation-based metrics
• Amplitude and phase statistics

Challenge Description

The objective of this challenge is to design a machine learning-based classification model capable of identifying the transmitting device based on RF fingerprints extracted from impaired signals.

The classification task is performed under realistic hardware impairment conditions, where distortions caused by RHI, CFO, and STO affect signal characteristics. These impairments introduce variability in the feature space, making the classification problem more challenging and representative of practical 6G deployment scenarios.

Participants are expected to:

• Develop robust classification models
• Analyze feature effectiveness under impairment conditions
• Ensure generalization across different distortion scenarios

This challenge establishes a controlled and reproducible benchmark for evaluating RF fingerprinting techniques in physical layer security and lightweight device authentication for 6G systems.

Input and Output

Input:

• Feature vectors extracted from received OFDM signals (frequency-domain features such as PSD, statistical descriptors, etc.)

Output:

• Predicted device label (e.g., Device 1 or Device 2) for each input sample

The task is a supervised classification problem where each input feature vector must be mapped to the correct transmitting device. The authors shall provide well-commented code and an explanation of the proposed solution.

Evaluation Metric

Performance is evaluated using the following metrics:

Accuracy:

Macro-averaged F1-score:

Accuracy measures the overall proportion of correctly classified samples, while the macro-averaged F1-score evaluates classification performance across all classes equally, regardless of class imbalance.

These metrics jointly assess classification effectiveness and robustness under hardware impairment conditions.

To participate: submit your solution using the submission form.