VECTOR QUANTIZING SYSTEM HAVING SELF TRAINING FUNCTION

摘要

PURPOSE:To avoid complexity of designing a code book and give versatility for non-stationary input signals by making input signals themselves training signals and updating a code book basing on quantized representative vector and the quantized value of residual of vector quantizing. CONSTITUTION:Signals inputted from an input terminal 1 are inputted to a minimum distance vector detector 2 and treated as N-dimensional vector X and compared with representative vector Vi outputted from a code book memory 3. Input vector is inputted to a subtracter through a fixed delay, and output of the subtracter 5 is inputted to a scalar quantizer as residual of vector quantizing corresponding to vector Vi. A multiplier 10 multiplies each component of local decoded vector L and alpha, and a multiplier 11 multiplies each component of representative vector Vj in a code book and (1-alpha), and each product is added by an adder. The result of addition Vi is the renewed value of vector Vj and stored in the code book in place of Vj.