Time Delay Estimation Using the Two-Level Three-Layer Structured Network

• Main Author: 陳鈺林
• Other Authors: 陳巽璋
• Format: Others
• Language: zh-TW
• Published: 1993
• Online Access: http://ndltd.ncl.edu.tw/handle/39926055020772258241

碩士 === 國立中山大學 === 電機工程研究所 === 81 === 　　In this thesis, a new time delay estimation (TDE) scheme based on two-level three-layer structured network is proposed. The two-level three-layer structured network can be viewed as a special class of feedforward neural network for solving moving-average (MA) model parameters based on second-order cumulants (SOC) and third-order cumulants (TOC) matching. In consequence, we apply the parameters estimated from this model to the direct delay estimation formula for TDE. Basically, this does not involve complicated matrix operation, and thus will be more efficient than the conventional TOC adaptive TDE method. 　　In general, the presented method can perform well when the signal is i.i.d. non-Gaussian that is the assumption the original two-level three-layer structured network was derived. However, this can be expanded to meet the need when the desired signal is correlated non-Gaussian process by using a modified method. 　　To demonstrate the performance of the method presented in this thesis, a computer simulation is carried out. As observed from the simulation results, assumed that the signals are i.i.d. non-gaussian signals, the results are satisfactory under the stationary and nonstationary time delay environments with varied signal-to-noise ratio. However, when the non-Gaussian signals became more correlated, the results can be further improved by using the modified patterns.