Where is dither?

SoundConsulting Question

Mar 8, 2026 00:00

The Data Sheet section on Hardware Accelerators says "The core includes accelerators like division, square root, barrel shifters, Base 2 logarithm, Base 2 exponential, slew, and a pseudorandom number generator." ... "The pseudorandom number generator can efficiently produce white noise, pink noise, and dither."

Looking around, I see white noise and pink noise generators, but there is nothing that I can find for dither.

I seem to recall a thread here mentioning that dither is not implemented. If that's the case, are there plans from Analog Devices to add support for dither as hinted in the data sheet?

Brian

Audio Signal Processors

software

sigmaStudio 4.7

adau1463

SigmaDSP Audio Processors

Report

DaveThib Reply

Mar 10, 2026 00:00

Hello SoundConsulting,

I will inquire with the programming team but I think all you need to do is add some low level white noise and that will dither the bottom bit or two depending on the level of the noise.

But, I may be missing something significant....

Dave T

Report

SoundConsulting Reply

Mar 10, 2026 00:00

Thanks, Dave.

Theoretically, that's all that would be needed.

Two noise sources, combined, would give a TPDF (triangular probability distribution function) dither, and if that were further high-pass filtered then it would have some of the benefits of psychoacoustic masking of the noise.

One challenge would be to precisely set the amplitude of the noise source to exactly 1 LSB of the result (probably 24-bit). For TPDF, the amplitude becomes 2 LSB. If the difference between the two noise sources is used, then that becomes +/-1 LSB, which is ideal.

Other DSP products with building blocks include handy, pre-made TPDF and TPDF High-Pass options. I can probably figure it out, but I would want to check my math to confirm that the final result is correct.

FYI: RPDF (rectangular probability distribution function) dither from a single noise source ends up having a DC bias that needs to be removed, somehow. TPDF, when computed as the difference between two noise sources, has both the benefit of ideal dither as well as having no DC bias to remove.

Report