Research On Key Techniques Of Radio Resource Allocation In Relay-enhanced Cellular Networks

It is accepted that the speech signal processing has high requirement to the quality of voice signal. However, with the system becoming more complex than before, it is very difficult to obtain high-quality speech signal. Besides, the clarity and intelligibility of speech signal have seriously degraded due to the severe noise interferes with the voice of pick-up, which would severely restrict the development and application of speech processing technologies. Therefore, it is very meaningful to study speech enhancement that could effectively separate the target and interference signals.In the voice signal processing, Speech enhancement is one of importance portion, which could directly affect the performance of the subsequent applications. It is well-known that the microphone array speech enhancement has critical role in designing the speech enhancement method. And its applications are more extensive; since it could effectively and accurately separate the target signal and interference signal by using the spatial filter in terms of array signal processing and radar. Based on its efficiency and accuracy in designing the speech enhancement method, in our thesis, both efficient speech enhancement methods are proposed by considering the characteristics of speech signal. And as shown in our experimental results, both methods can improve the effect of the microphone array speech enhancement. The thesis structure is organized as followsFirstly, the research significance and development of array speech enhancement are introduced, and then the characteristics of speech signal are also briefly analyzed in different environments.Secondly, we have introduced the variety of speech enhancement algorithms in detail, and the advantages and disadvantages of these algorithms are also been analyzed. Meanwhile, based on the adaptive filtering mode, a new variable step least mean square algorithm is also illuminated, which could reduce the effect of input noise to step factor.Finally, based on study and analysis of the existing algorithms, in our thesis, two speech enhancement methods are proposed: the frequency decomposition generalized side lobe offset method and the generalized beside the offset rear spectral subtraction method. The former has fixed beamforming of main lobe width, and the relationships between the signal frequencies are also analyzed. And then the digital filter group signal is decomposed into different frequency bands. And further processed through the multiple sampling rate signal processing technology. On account of the characteristics on the samples of each sub-band signals for different sub band signals, the matched processing method are freely and flexibly choose, which greatly reduce the calculation load. The latter one has used spectral subtraction as a rear-mounted filter and adopted the combination of noise estimation and voice detection method to estimate noise. With the second speech enhancement mode, the mode has promoted the enhancement effect and obtained the higher output SNR.