Loading...

Outlier Censoring Based on Sparse Signal Recovery Algorithms

Bassak, Elaheh | 2022

203 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55506 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Karbasi, Mohammad
  7. Abstract:
  8. In today’s world, knowledge of the statistical behavior of noise can tremendously affect the accuracy of target detection in radar systems. Therefore, radar systems commonly collect a secondary dataset of homogeneous noise and estimate the statistics of the gathered data, prior to attempting target detection. Specifically, in the case of Gaussian noise with a mean of zero, the entire statistical information of the noise is encoded in its covariance matrix. In practice, however, the challenge is that the training samples do not purely contain homogeneous noise. In fact, some samples contain non-homogeneous outlier signals that do not have the same distribution as the noise samples. In this thesis, therefore, we aim at presenting an efficient method for detecting the address of outliers in order to remove the contaminated samples from the training dataset. It should be noted that once the outlier contaminated samples are removed, the size of the training dataset would be reduced but the remaining samples would be homogeneous. Therefore, the covariance matrix of the noise can be estimated accurately via simple methods such as the sample covariance matrix. Next, by employing concepts of compressed sensing we present several methods for detecting the location of the outlier samples. One novel approach is the match filter method for detecting the index of outlier data. But the drawback of this method is its sensitivity to the choice of angular grid as well as its inability to estimate the true number of outliers present in a scene. In order to overcome this drawback, we propose a novel strategy based on sparse signal recovery techniques which follow a l_q-norm constraint, where 0
  9. Keywords:
  10. Training Dataset ; Compressed Sensing Radar ; Outliers ; Block Sparse Signal ; Model Order Selection Tool ; Sample Covariance Matrix ; Block Sparse Learning via Iterative Minimization (BSLIM)

 Digital Object List

 Bookmark

No TOC