On the enhancement of signals in the presence of noise

Abstract

This thesis is concerned with the speech enhancement system in the presence of noisy data. The real word presents us with digital data which contains not only the useful signal but also the noise. Noisy data is referred to the unwanted information-bearing signal conveying information about either the sources of the noise or the environment. Unwanted noisy data often causes the algorithms to miss out the useful signal in the data so as to limit the ability of systems. Noise reduction is a broad term where the goal is to remove the information unrelated to the phenomenon we want to study. In this thesis, we resort to tackling the noise reduction problems and seeking the optimal speech enhancement solutions in the present of noisy data. In the speech communication system, there are two kinds of noise. The first one is the environment noise. The other is the channel noise especially in the wireless communication case. Environment noise can be recognized as the measurement noise which is the result of the imperfection of measurement instruments. The channel noise can be treated as the dynamical noise in the inherent system. Since speech data are recorded by pre-mounted microphones, one of the main problems when dealing with the speech data is that the microphones will most likely record noisy speech due to the interference of the environment and background noises. In order to extract the pure speech and suppress the noises, there are two popular methods including the beamforming technique and blind source separation technique. In this thesis, we firstly design a novel distributed acoustic beamformer with blockchain protection such that the interference signals and background noise can be suppressed. After that, we try to improve the performance of the blind source separation system via the optimization of sensor placement in the wireless acoustic sensor network. It observes significant advantages of our proposed methods according to the simulation results. In order to deal with the channel noise, we investigate the optimization methods in terms of the M-QAM constellation in additive white noise channel. In digital communication systems, channel coding is an extremely important task in determining the performance of the system. In implementing such systems, one needs to map a sequence of bits to symbols in the constellation. A good mapping can show a significant difference in the ultimate performance of the designed system in terms of the error rates. Using the technique of the assignment problem, we derive an optimal mapping rule for the Honeycomb-structured constellation such that the bit-error rate is significantly reduced. Apart from that, we also investigate the optimal position of symbols in the constellation. In order to speed up the optimization process, we propose a novel calculation method of the bit error rate (BER) and the tailor-made optimization method is also proposed. Simulation results show that our proposed BER calculation method is accurate and an optimal mapping rule can be achieved using our tailor-made optimization method.