Statistical signal analysis for systems with mutually related input functions

Abstract

Modern digital signal processing systems require substantial hardware and software flexibility at a sufficiently low cost. The cost of configuring the hardware and developing the software should be low, and it should be easy to make revisions as requirements such as the number of channels and data rates change. Also important is efficiency for both numerically intensive operations, such as FFT and filter operations, and logical and symbolic manipulation intensive operations such as decision making and display generation. This mixture of computationally intensive operations is ideally suited to concurrent parallel processing. The BBN Butterfly multiprocessor provides a general purpose parallel processor which may be configured for a wide range of applications. It contains up to 256 simultaneously operating asychronous processor nodes, each containing a general purpose microprocessor, local memory, and optional floating-point accelerator, optional high-speed auxiliary processors, and a unique expandable inter-processor communication system. The tight processor coupling and global memory sharing provided by the communication system allows the necessary coordination of the operations in the independent processor nodes, and provides a software environment which dramatically facilitates the ease of programming.