Pseudo excitation parallel system identification algorithm for fixed point digital signed processor
Chen Ke (Graduate student of electrical engineering)
MetadataShow full item record
A common application of the adaptive process is system identification or system modeling. In many real time implementations, because of the poor conditioned input excitation signal and the limited-precision effect as using fixed point digital signal processor, the performance of the real time adaptive on-line modeling is not only affected by the time-variant unknown system but also is undesirably changed by the input excitation signal itself. A Pseudo Excitation Parallel System Identification (PEPSI) algorithm is developed and studied in this thesis. This method utilizes an additional broad band signal and a parallel adaptive filter structure to generate a very well conditioned pseudo excitation signal for the main modeling adaptive filter without interfering with the normal application. Therefore, together with the anti-corruption control technique which is also developed in the thesis, PEPSI process can use any kind of real application signal such as speech signal in acoustic echo canceller application to achieve much better on-line modeling results. Analysis and simulations have been made to estimate the performance of PEPSI process. Results show that the time-variant system can be well modeled with or without using previous off-line modeling results.