Issue No 7: November 2003
Research at The Hong
Kong Polytechnic University may help improve the clarity of voice reception
on mobile phones, with less background noise.
Principal Investigator Prof Kok-lay Teo said the total elimination of background noise is not possible. But it can be reduced to a level where it becomes acceptable.
Frequencies for an ordinary telephone for the human voice communication, for example, are in a range from 300 hertz to 3,000 hertz. The range captures most human speech from a sound signal.
The sound signal picked up by a microphone is disturbed by noise and unwanted speech from others. Therefore, it is important to distinguish and enhance only the signal of interest.
Said Prof Teo: What we are trying to do is use mathematics to develop state-of-the-art methods to optimise a given performance measure of filters without scarifying the physical requirements and specifications. The optimising methods devised in the research are generic, he added, and have applications across a broad front, including image filtering in computers.
Objectives given by the research teams collaborators include reducing energy consumption, reducing implementation cost, as well as minimising noise interference. For image filters, it also includes optimisation to enhance the sharpness of pictures.
Said Prof Teo: There are still many unsolved problems in the area of optimum filter design. We have solved a number of them efficiently and brought forward knowledge which will help the industry generally.
The purpose of signal processing is to extract the signal from noise. A family of fast computational algorithms based on a Novel Information Criterion (NIC) has been obtained. It has not been used in this way before, said Prof Teo.
After analysing specification requirements outlined by the collaborators, the research involved finding mathematical methods, and subsequently the computational algorithms, for solving the problems.
Dr Cedric Yiu who was involved in the project said: There are many flexibilities for designing a filter. Our aim was to develop efficient methods for finding the best design of the filter such that a given performance measure is optimised without scarifying the practical requirements and specifications.
We needed to strike a balance. For example, if the noise is suppressed too much, this may have an undesirable impact on the signal of interest; ie, the speech.
We were not trying to reduce noise to zero, in other words, infinity measured in decibels (dB). In many practical applications, it is sufficient to reduce it by 20 to 30 dB.
The actual amount may be set as one of the application criteria. For some applications, a reduction with 40 dB may be required.