Dither with a triangular probability distribution function gives theoretically zero noise modulation, but does so at the cost of adding LSB2/6 noise power each time it is applied. Most DSP applications run at the original sampling frequency fS, so the dither noise has to be distributed in the spectrum up to fS/2. In these applications the dither is essentially an audio range signal. This paper examines several proposals for shaping the spectrum of the dither to minimize its audibility. A recently developed loudness model is used to estimate the detectability of shaped noises. The investigation examines the detection criterion appropriate to such shaped noises and gives some listening test results. In searching for an efficient dither the authors are looking for a balance between the audibility of the spectrally shaped dither and the computational cost of achieving it.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.