Dissertation > Excellent graduate degree dissertation topics show

ICA-based blind signal separation technology and its application

Author: FangXingJie
Tutor: LvShuPing
School: Harbin Engineering University
Course: Pattern Recognition and Intelligent Systems
Keywords: Independent Component Analysis Performance Adaptive Algorithm Learning step Similarity measure
CLC: TN911.7
Type: Master's thesis
Year: 2011
Downloads: 107
Quote: 0
Read: Download Dissertation

Abstract

Blind source separation refers to the source signal, the transmission channel characteristic is unknown, only the source signal from the observed signal and the a priori knowledge (such as probability density function) of the source signal estimation process of the various components. Blind Source Separation rise in recent years as an efficient signal processing method, the antenna array signal processing, medical signal processing, image processing and ship vibration testing has broad prospects for development and other fields, has gradually become the signal processing community and the neural the scientific community of common interest network hotspot. In blind signal processing instantaneous linear mixed model as the keynote papers main work: 1. First describes the major blind source separation problem solutions - independent component analysis (Independent Component Analysis, ICA). A signal based on a mathematical model, detailing the ICA's basic assumptions and uncertainties outlined ICA preparatory measure theory knowledge and independence standards, summed up the commonly used algorithm performance. (2) discuss the blind source separation algorithm batch and adaptive algorithms. Based on negative entropy Fast ICA algorithm has been made on a number of data samples are processed, fast convergence, but tracking performance is poor; while the adaptive algorithm is a single-sample data processing, and real time. In the channel constant, the former separation performance was stronger than the latter. The channel is unstable, the only adaptive algorithm, and its constant presence learning step convergence speed and steady-state accuracy can not take into account the shortcomings, learning step optimization problem is the core thesis adaptive algorithm to make the convergence and stability to achieve the best results. 3 For the main factors affecting online algorithms - the activation function and learning step, first introduced kurtosis online estimate, making the activation function with parameters to meet a combination of different types of signal sources, and analyzes the natural gradient variable step size gradient algorithm characteristics - with strong real-time tracking resistance, but learning step to assist in the selection variable as the basis, and the degree of separation between the output signal regardless of the channel that is not constant, the initial step size is too large case, the algorithm failed to achieve satisfactory results the separation accuracy. The gradient variable step size, based on the thesis improved variable step size adaptive algorithm, neural network defines a representation of the degree of separation between the output of the similarity measure, according to the similarity measure reflected the state adaptive signal separation step long, and the establishment of learning step and similarity measure nonlinear relationship between the amount of change. Comparison with the previous algorithm, this algorithm not only has fast convergence and small steady state error characteristics, and is suitable for the channel is not constant in time-varying environments. 4 based on independent component analysis for blind source separation algorithm in the application of vibration signals Ship preliminary discussion. In-depth analysis of the line spectrum ship noise modeling, modeling, and continuous spectrum modulation envelope modeling. To establish a mathematical model ship radiated noise and ship noise simulation algorithm source separation is feasible.

Related Dissertations

CLC: > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Communication theory > Signal processing
© 2012 www.DissertationTopic.Net Mobile