Smartphones Identification Through the Built-In Microphones With Convolutional Neural Network

Abstract

The use of mobile phones or smartphones has become so widespread that most people rely on them for many services and applications like sending e-mails, checking the bank account, accessing cloud platforms, health monitoring, buying on-line and many other applications where sharing sensitive data is required. As a consequence, security functions are important in the use of smartphones, especially because most of the applications require the identification and authentication of the device in mobility. This is usually achieved through cryptographic systems but recent research studies have also investigated alternative or complementary authentication mechanisms which can be used to strengthen cryptographic methods with multi-factor authentication. In this paper, we investigate the identification and the authentication of smartphones using the intrinsic physical properties of the mobile phones built-in microphones. The possibility to identify a microphone on the basis of features extracted from audio recordings is well known in literature but it is mostly used in forensics studies and usually relies on human voice recordings. On the contrary this paper proposes a smartphone identification and authentication approach by stimulating the built-in microphone with non-voice sounds at different frequencies. An extensive data set of 32 phones was used to evaluate experimentally the proposed approach. On the basis of the proven performance of deep learning techniques, a new Convolutional Neural Network architecture is proposed both for the identification and the authentication purposes. Its performance, in comparison to other machine learning algorithms, is demonstrated in presence of different types of noises (e.g., Gaussian White noise, Babble noise and Street noise). Satisfactory results have been obtained showing that the exploitation of a fingerprint from the microphone sensor is a good choice to assess smartphone distinctiveness. INDEX TERMS Smartphone identification, authentication, microphone, machine learning, deep learning. IRENE AMERINI (M'17) received the Laurea degree in computer engineering and the Ph.D. degree in computer engineering, multimedia, and telecommunication from the University of Florence, Italy, in 2006 and 2010, respectively.

An Evaluation of Entropy Measures for Microphone Identification

“…Then, the statistical and entropy features are applied only to the complex magnitude component. This result is consistent with the findings in literature on microphone identification [ 11 , 16 , 18 ], where the complex magnitude components of the spectral domain are used. The reason for this behavior is linked to the frequency response of the hardware elements of the microphones (e.g., filters, amplifiers), in which “fingerprints” are more relevant in the complex magnitude component rather than the complex phase component of the spectral representation.…”

“…From one side, this result shows that an entropy based approach is able to obtain in general (e.g., with different sets of entropy features) a very high classification accuracy even with a significant dimensionality reduction. It performs well in comparison with more sophisticated deep learning methods (see Reference [ 18 ]), which require significant computational power. On the other side, robustness to noise is a very desirable feature in practical applications of the identification technique presented in this paper because of the presence of background noise or distance between the generator of the audio stimulus and the microphone.…”

“…The discrete Fourier transform (DFT) was applied to the recordings using a Fast Fourier Transform (FFT) algorithm. As described in the Section 2 the classification usually exploits the frequency response because it captures the discriminating features of the microphones (due to filters and to the physical shape of the microphones) in a better way than the original time domain representation [ 16 , 18 ].…”

“…Thus, the most common approach used in literature is to add noise to the recorded audio samples to simulate the presence of noise during the recording phase. This is the approach used in Reference [ 16 , 18 , 31 ], where a similar argument is proposed. Additive White Gaussian Noise (AWGN) was added to the audio recordings with decreasing values of Signal Noise Ratio (SNR) to simulate the presence of attenuation due to path loss and background noise.…”

“…Sound stimuli applied to microphones for identification and authentication have also been used in Reference [ 18 ], where a Deep Learning approach based on CNN is adopted. Tones at 1 KHz and 2 KHz are used as stimuli of a set of 32 mobile phones.…”