Improved Variable RZA-LMS Algorithm Based On Discrete Cosine Transform For Noise Reduction

Oil field downhole operation will be accompanied by a lot of danger, timely and effective communication measures have an important role in the security, but at present, communication quality in oil field work environment cannot meet the requirements. There is a lot of large power equipment in the oil field environment, which is lead to the environment is extremely complex, including compressors, pumps, drilling machines, fans and so on, it can produce high decibels of noise, so as to produce noise pollution when the equipment is running. Strong noise can seriously reduce the quality of communication among workers, and will harm people’s health, even cause serious consequences at last.At present, the important way to eliminate noise is speech enhancement technology, speech enhancement technique is to make possible the pure speech signal extracted from the noisy speech and improve speech intelligibility, and it must rely on high quality speech enhancement algorithms to complete this work. Previous scholars have done a lot of research for voice enhancement technology research, and provided a large number of algorithms, but the effect in the oil field environment is not very good, because of the noise which is high decibels, mixed and the frequency is in the same range as people speaking. Therefore, it is a difficult problem to improve the quality of speech in a specific environment.Characteristic of this paper is focused on noise in oil field to expand the study of speech enhancement algorithms. Firstly, we found the find the appropriate speech enhancement algorithm, the adaptive filtering algorithm, according to the characteristics of the noise. This algorithm does not require a priori knowledge of the input signal, which can avoid the problem of noise models of the oil can not be estimated, noise reduction can be very good to complete the work through the noise and noisy speech signal. We found the contradiction between the convergence and steady-state error of the classical LMS(Least Mean Square) adaptive filtering algorithm through experiments and simulation analysis, and we first proposed an improved adaptive filtering algorithm named variable zero attraction(VZA) LMS algorithm, and the experimental results show that the new algorithm can improve the convergence speed which is better than the other’s.Secondly, the VZA-LMS algorithm is good for identity spares system, we finally proposed an improved VZA-LMS based on Discrete Cosine Transform(DCT). Discrete Cosine Transform can concentrate the signal power, and reduces the autocorrelation matrix of the signal correlation, the signal by DCT group can be sparse representation, so as to achieve Characteristic algorithms for signal requirements. After a large number of simulation experiments, we can see that our algorithm is superior to other similar algorithms through the analysis of 4 kinds of voice quality evaluation indicators and the effect of speech enhancement is obvious.Finally, we developed a new software for noise reduction with VC++2008, Matlab and My SQL database, which can achieve a variety of noise reduction algorithm simulation experiments, and it can record and play audio, view the audio data, and the resulting experimental data can be stored in the database.