An Evaluation of Entropy Measures for Microphone Identification.

Area of Science:

• Signal Processing
• Information Theory
• Device Forensics

Background:

• Microphones possess unique physical characteristics imprinted on audio signals.
• Device identification via microphones is crucial for security applications.
• Existing methods often face a trade-off between accuracy and classification time.

Purpose of the Study:

• To evaluate the efficacy of various entropy measures for microphone classification.
• To identify optimal entropy measures and hyperparameters for accurate device identification.
• To analyze the accuracy-classification time trade-off using entropy-based features.

Main Methods:

• Application and comparison of Shannon Entropy, Permutation Entropy, Dispersion Entropy, Approximate Entropy, Sample Entropy, and Fuzzy Entropy.
• Validation using an experimental dataset of 34 mobile phone microphones and three audio signals.
• Feature selection analysis using filter methods to determine discriminating entropy measures.

Main Results:

• Selected entropy measures demonstrate high identification accuracy, outperforming traditional statistical features.
• Entropy measures show robustness against background noise.
• Optimal entropy measures and hyperparameters were identified, balancing accuracy and efficiency.

Conclusions:

• Entropy measures are highly effective for unique microphone identification.
• These methods offer a promising, efficient, and noise-resilient approach for mobile device security.
• Further analysis confirmed the practical utility of entropy measures in device classification.