Vector Quantization System Based on Scalar SOM/AND-OR Hybrid Network

This paper proposes new vector quantization algorithm, and its hardware configuration is discussed. The system is based on the combination of self-organizing map (SOM) and AND-OR network. The SOM is used to preprocess the input vectors and the AND-OR network is used to identify the categories to which the input vectors belong. The proposed system uses a simplified SOM that handles a scalar input instead of the vector input. N-dimensional input vector is split into N scalars, each of which is processed by the scalar SOM (SSOM). The input vector space is divided into multiple blocks, and the blocks belonging to a same class are selected as a single group. The class is identified by the group to which the input vector belongs. The AND-OR network is trained to do the selection by supervised Hebbian learning rule. The proposed system is designed so that it can be easily implemented with digital hardware. VHDL simulations are conducted to verify the feasibility of the system. The results show that the proposed system successfully classifies data set with very complicated distribution. The circuit size of the system is smaller than that of the conventional SOM, while its performance is estimated as 6.25 million vectors processed per second.