SYSTOLIC MIMD ARCHITECTURES FOR 4-LEVEL SPHERICAL SUBSPACE TRACKING

The 4-level spherical subspace tracker tracks the dominant or subdomin ant subspace associated with an input vector time series. The tracker maintains four distinct eigen-levels which can be used with the Minimu m Description Length (MDL) criterion to make subspace dimension adjust ments on the fly. This is important as the associated subspace based d irection finding algorithms require knowledge of the subspace rank. In systems involving large radar or sonar arrays, the achievable perform ance will often be limited by the real time MIPS rate. Hence, in this paper the 4-level spherical subspace tracker is parallelized and seque nced to execute on systolic MIMD architectures. The systolic MIMD arch itectures are ideally suited for implementation with custom CORDIC VLS I or networks of available parallel numeric processing chips such as i Warps and TMS320C40's, Pseudo C-language cell programs specify the fun ctionality and timing of the architecture using the fine grain MIMD/sy stolic model of computation.