| Summary: | 碩士 === 國立雲林科技大學 === 電機工程系 === 107 === The design of partial updating (PU) technique for the weights in an active noise controller (ANC) is discussed in this thesis. Functional link artificial neural network (FLANN) equipped with filtered-s least mean square (FSLSM) algorithm is chosen as the structure of the ANC. Full update (FU) technique is first compared with some known partial update technique, such as M-max and Sequential. Then we propose an interval adaptive maximum-between-two (IAMBT) algorithm, which was evolved from an earlier version, namely, maximum-between-two (MBT). Comparisons among these algorithms are presented in this thesis.
The effect of noise reduction is highly dependent on the technique we choose when PU techniques are included in an FLANN. The three main PU techniques investigated in this thesis are M-max, Sequential and IAMBT PU. M-max technique only updates the output from secondary channel filter with comparatively larger values. Although it can indeed reduce a lot of unnecessary weight updating, the sorting process before updating unfortunately brings additional computational burden. Sequential technique, on the other hand, partially updates the weights in a prescribed order. So the performance is greatly affected by the order chosen by designer. IAMBT technique proposed in this thesis also chooses the output from secondary channel filter, but determines the size of data window to be compared in each updating step. Whether the weights need to be updated or not will be determined by checking the data values in the window. As the system approaches steady, the size of data window can be increased. This means that computation needs to be carried out in a larger interval. Thus system burden can be reduced.
It can be seen from simulation results that, comparing with FU technique, M-max and Sequential both save about 50% of the computational burden. The reduction of the proposed IAMBT PU technique can be up to 85%. On the basis of operational amount, our proposed IAMBT presents the best performance among all the techniques mentioned.
|
|---|
