Source Enumeration for Practical Multi-Sensor System

In array signal processing, a manifold of sensors is employed to produce observations of some measurable quantity. The type of signals measured by the array will depend on the application for which it is designed. For example sensor arrays have been used in such fields as communications, sonar, speech processing, manufacturing processes and image reconstruction. A common assumption is that we use narrowband signals for processing and that the system operates in the far-field.

Estimation of the number of sources impinging on an array of sensors is the critical first step in a subsequent signal parameter estimation in array signal processing. The most classical methods are based on inferring eigenvalue structure of the sample covariance matrix, with the help of information theoretic criteria or hypothesis testing procedures. To implement these methods, the large sample size and the sample distribution (e.g., Gaussianity ) are assumed necessarily.

Herein, we focus on some practical situations, e.g., small sample size and impulsive noise. Firstly, it is not always possible to collect enough large amount of samples, due to the limitation of operation time and system hardware. Secondly, impulsive noise has been considered as a more accurate description for the ambient noise in many communication channels, due to the impulsive nature of man-made electromagnetic interference and a large amount of natural interference.

The limitation of sample size and the presence of impulsive noise often cause large performance degradation of classical methods. Since the bootstrap is a promising tool in the case of small sample size and sample distribution ambiguity, the bootstrap is applied to cope with the difficulties caused by the practical situations, However, the bootstrap is very sensitive to impulsive noise which results in outliers in the observed data. This is because, the bootstrap method has a very low breakdown point, since the simulated bootstrap distribution may be severely affected by bootstrap samples with a higher proportion of outliers than that in the original data set. Herein, we propose the use of robust statistics in order to prevent the expansion of outliers contamination in some bootstrap samples. Efficient integration of robust statistics into the bootstrap scheme is the most challenging issue in this research.

Multi-Scale Modeling and Simulation; Signal Processing

Zhihua Lu

Dipl.-Ing.

Address: | Dolivostraße 15 |

D-64293 Darmstadt | |

Germany | |

Phone: | +49 6151 16 - 24401 or 24402 |

Fax: | +49 6151 16 - 24404 |

Office: | S4|10-210 |

Email: | |