Efficient bit-level systolic array for the linear discriminant function classifier

The linear discriminant function classifier is a widely used but computationally demanding method in statistical pattern recognition. This paper describes a bit-level systolic array for the linear discriminant function classifier to improve its processing speed. The system includes a new scheme for inner product computation, which not only has 100% efficiency but also gains a speed improvement over a previous method, and yields classification results at an average rate of one per B cycles of the array, where B is the wordlength of the input data. The throughput is higher than those of the related bit level arrays described previously.