For typical adaptive systems, the computation of the adaptive weights requires 50 to 150 GFLOPS of computational throughput.
This type of high-throughput adaptive processing calls for special purpose hardware to achieve high efficiency that fully exploits the parallel nature of the adaptive weight computation process.
The system of claim 1 wherein during array operation, all information passed between said computation cells is in the form of packets, and wherein said computation cells comprise means for recognizing and responding to a plurality of different types of packets.4.
The system of claim 3 wherein said matrix arithmetic operations include operations on Cholesky factors, said rotation stages include means for storing Cholesky factor data and means for performing circular/hyperbolic rotations, and said packet types include Cholesky update packets and Cholesky downdate packets, and wherein said array cells comprise means responsive to said Cholesky update packet for zeroing out a leading data element and updating said stored Cholesky factor data, and means responsive to said Cholesky downdate factor to cause said cells to perform circular/hyperbolic rotations on data and downdate said stored Cholesky data.8.
The system of claim 3 wherein said packet types include a constraint vector packet, said constraint vector packet type including constraint vector data used by said array in performance of said matrix arithmetic operations.9.
This essay will highlight contributions of numerical linear algebra to optimization, as well as some optimization problems encountered within linear algebra that contribute to a symbiotic relationship.
A linear systolic array of computation cells, each cell having several vector rotation stages.
The computation of the adaptive weight vectors is one of the most difficult of signal processing algorithms used to implement large scale adaptive processing systems.