Research on Signal Processing Supercomputers
Abstract
Signal processing is an area where the required computational bandwidth in an application can be unbounded. Applications such as radar, sonar and communications already call for signal processing systems capable of delivering billions or tens of billions of operations per second. In developing a new signal processor to meet these requirements, it is essential to understand the underlying computational models. An ad-hoc processor development effort that is unclear on the computational models will likely be wasteful and unable to meet the long-term performance goal. Fortunately, because the control in signal processing is typically data-independent, computational models in this area can be relatively simple. Based on the study performed under this contract, this report describes some important computational models for parallel signal processing, and illustrates how the Warp machine developed by Carnegie Mellon supports these models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1988
- Accession Number
- ADA206911
Entities
People
- H. T. Kung
Organizations
- Carnegie Mellon University