An algorithm-based fault tolerant method termed the fault tolerant least-mean-squares (FTLMS) algorithm is extended from 1-D to 2-D. The algorithm enables the LMS to correct for a particular type of weight update failure called a stuck-at fault. The key to the algorithm is an overcomplete set of vectors used to transform the input. In developing the 2-D version, the properties of performance-enhancing transforms are investigated, leading to a transform structure that gives the algorithm a more desirable convergence rate. Various transforms are evaluated according to their reduction in eigenvalue ratio (EVR) for stationary inputs.
展开▼
