Parallel algorithms and computational structures for linear estimation problems

Modern acquisition techniques produce high volume data sets that must be processed not only very quickly, but also reliably. The only realistic means for achieving high throughput is the use of parallelism. In terms of reliability, we need the capability of efficient and timely fault detection, location recovery and repair. In this paper, we consider computational structures for a specific type of signal processing-linear estimation problems-that incorporate both parallelism and fault detection and location features. The first part of the paper is devoted to a discussion of real time fault detection requirements. Then the following section considers fault detection and parallel processing of statistical signals using the concept of stochastic redundancy. Finally, parallel algorithms and reliable network designs are presented for a linear regression problem.