広帯域音響信号の量子化への大振幅ディザの適用

抄録

A quantization noise is inevitable in digitizing analog signals. If the input signal passes quickly through many quantization steps, the quantization noise has little correlation to input signals. But if the input signal level is very low and the quantization steps are few or a signal varies very slowly, the quantization noise has a sort of distortion with strong correlation to the input signal. By adding a random"dither"with a uniform probability density function equal to the quantization step Δ to a signal to be quantized and by subtracting the identical dither from the quantizer output, it is possible to make the quantization noise without correlation to the input signal and with a uniform distribution and the power Δ^2/12. however, hardware equipments such as AD or DA converter have the conversion erros different from the ideal quantizing characteristics. Usually these errors do not possess a uniform distribution. In quantizing a signal with these converters, adding and subtracting a dither of much larger amplitude will lead to average the errors and increase the accuracy. This paper includes the optimization of a dither for quantization of wide range audio signals and a brief explanation of the theory and the experiment to increase the accuracy of AD or DA converter by using a large amplitude dither.