We investigate the consequences of the rounding or truncation to a fixed word length of the output of a digital signal processor, such as a digital editor, mixer, equalizer, filter, or sampling standards converter, using fixed-point arithmetic. The signal input to the processor is assumed to be in digital form, and have been quantized with proper analog dither applied. Quantization artifacts are thus of a benign nature in the input signal, and our concern is that the arithmetical manipulations to which these digital words are subjected in the digital signal processor should not introduce any quantization distortion or noise modulation. We show that the use of appropriate digital dithering before final rounding or truncation can keep the input signal free of such degradation even when subjected to severe operations such as gain fading. The penalty is a small increase in the broadband noise level.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.