Audio data hiding for acoustic data transmission in reverberant aerial space

Abstract

In this dissertation, audio data hiding methods suitable for acoustic data transmission are studied. Acoustic data transmission implies a technique which communicates data in short-range aerial space between a loudspeaker and a microphone. Audio data hiding method implies a technique that embeds message signals into audio such as music or speech. The audio signal with embedded message is played back by the loudspeaker at a transmitter and the signal is recorded by the microphone at a receiver without any additional communication devices. The data hiding methods for acoustic data transmission require a high level of robustness and data rate than those for other applications.

For one of the conventional methods, the acoustic orthogonal frequency division multiplexing (AOFDM) technique was developed as a reliable communication with reasonable bit rate. The conventional methods including AOFDM, however, are considered deficient in transmission performance or audio quality. To overcome this limitation, the modulated complex lapped transform (MCLT) is introduced in the second chapter of the dissertation. The system using MCLT does not produce blocking artifacts which may degrade the quality of the resulting data-embedded audio signal. Moreover, the interference among adjacent coefficients due to the overlap property is analyzed to take advantage of it for data embedding and extraction.

In the third chapter of the dissertation, a novel audio data hiding method for the acoustic data transmission using MCLT is proposed. In the proposed system, audio signal is transformed by the MCLT and the phases of the coefficients are modified to embed message based on the fact that human auditory perception is more sensitive to the variation in magnitude spectra. In the proposed method, the perceived quality of the data-embedded audio signal can be kept almost similar to that of the original audio while transmitting data at several hundreds of bits per second (bps). The experimental results have shown that the audio quality and transmission performance of proposed system are better than those of the AOFDM based system. Moreover, several techniques have been found to further improve the performance of the proposed acoustic data transmission system which are listed as follows: incorporating a masking threshold (MM), clustering based decoding (CLS), and a spectral magnitude adjustment (SMA).

In the fourth chapter of the dissertation, an audio data hiding technique more suitable for acoustic data transmission in reverberant environments is proposed. In this approach, sophisticated techniques widely deployed in wireless communication is incorporated which can be summarized as follows: First, a proper range of MCLT length to cope with reverberant environments is analyzed based on the wireless communication theory. Second, a channel estimation technique based on the Wiener estimator to compensate the effect of channel is applied in conjunction with a suitable data packet structure. From the experimental result, the MCLT length longer than the reverberation time is found to be robust against the reverberant environments at the cost of the quality of the data-embedded audio. The experimental results have also shown that the proposed method is robust against various forms of attacks such as signal processing, overwriting, and malicious removal methods.

However, it would be the most severe problem to find a proper window length which satisfies both the inaudible distortion and robust data transmission in the reverberant environments. For the phase modification of the audio signal, it would be highly likely to incur a significant quality degradation if the length of time-frequency transform is very long due to the pre-echo phenomena. In the fifth chapter, therefore, segmental SNR adjustment (SSA) technique is proposed to further modify the spectral components for attenuating the pre-echo. In the proposed SSA technique, segmenatal SNR is calculated from short-length MCLT analysis and its minimum value is limited to a desired value. The experimental results have shown that the SSA algorithm with a long MCLT length can attenuate the pre-echo effectively such that it can transmit data more reliably while preserving good audio quality. In addition, a good trade-off between the audio quality and transmission performance can be achieved by adjusting only a single parameter in the SSA algorithm.

If the number of microphones is more than one, the diversity technique which takes advantage of transmitting duplicates through statistically independent channel could be useful to enhance the transmission reliability. In the sixth chapter, the acoustic data transmission technique is extended to take advantage of the multi-microphone scheme based on combining. In the combining-based multichannel method, the synchronization and channel estimation are respectively performed at each received signal and then the received signals are linearly combined so that the SNR is increased. The most noticeable property for combining-based technique is to provide compatibility with the acoustic data transmission system using a single microphone. From the series of the experiments, the proposed multichannel method have been found to be useful to enhance the transmission performance despite of the statistical dependency between the channels.