Complexity reduction in designing globally optimal scalar quantiser

A method to reduce the computational complexity in the design of a dynamic-programming-based globally optimal quantiser is proposed. With the new method, the computational complexity involved in designing a q-level quantiser with N training data samples is reduced from O(0.5N/sup 3/) to O((7+q)N/sup 2/) arithmetic and comparison operations. The effect of this reduction is to make the training of large data sets a practical possibility.