Adding Dither

[oshifis]

I don't know too much about the theory of dither, but I think it is correct to state that it improves low level linearity. Then I know that either the CD record may contain added dither, or the digital filter may add dither during playback.

If the above is correct (please confirm) then would it be possible to generate dither in hardware and add it to the bit samples? I mean building a random signal generator (PRBS) and some logic circuitry into the CD player could do it.

[wakibaki]

Dither is added to randomize patterns that arise in quantization. Production CDs normally have dither added at the mastering stage e.g when the sample rate is reduced from 24 to 16 bits. The rounding-up and -down which takes place is randomised so that an overflow of 0.1 will be rounded up once for every 9 times it is rounded down, the pattern of 1-to-9 roundings being varied randomly.

I don't think it's practical to add dither at playback, it requires knowledge of the signal being processed (downsampled or filtered), and I can't see a good reason for doing it.

w

[abraxalito]

Originally Posted by oshifis

I don't know too much about the theory of dither, but I think it is correct to state that it improves low level linearity.

You could put it like that - in practice there's no linearity (no signal) at all below the LSB in the absence of dither and even above this the quantization noise is correlated with the signal.

Then I know that either the CD record may contain added dither, or the digital filter may add dither during playback.

I'd hope that dither is used whenever an A/D conversion is performed or the word length is reduced. Hence a digital filter based on a sum-of-products structure ideally adds dither before sending 16/18/20/22 or 24 bits out to the DAC.

If the above is correct (please confirm) then would it be possible to generate dither in hardware and add it to the bit samples?

Certainly its possible but would perform no useful function unless its added prior to word-length truncation. To do that requires access to the internals of the digital filter in use.

I mean building a random signal generator (PRBS) and some logic circuitry into the CD player could do it.

Yes, that would be a primitive way of doing it, but suboptimal in the sense that dither is ideally TPF which is usually generated by summing two uncorrelated noise sources. So to do it well, you should have two independent PRBS and take steps in the design to ensure they remain uncorrelated. But as I've already mentioned, there's no benefit to doing this unless inserted at the correct point - immediately prior to truncation.

[badman]

Originally Posted by abraxalito

Certainly its possible but would perform no useful function unless its added prior to word-length truncation. To do that requires access to the internals of the digital filter in use.

Unless you were stuck with poorly mastered material and wished to add low level noise for masking purposes. I'll take a low-level complex noise spectrum over a low level signal-correlated artifact anyday.

Of course, you'd want this switchable.

[wakibaki]

Helluva long way round to go to add some noise.

w

[SoNic_real_one]

Like it was said before addind dither reduces the available bitdepth. That's why it can be done to the original 20-24 bit signal when is converted to 16 bit.

Addig dither to an existing 16 bit signal is possible only if you can live with resulting 14-15 bit of data (or less). And it won't "improve" the quality of the original signal, it will just make those "lesser" bitdepth samples to sound better that the same bitdepth samples obtained without having the dither applied at the conversion/truncation.

[guido]

Use a PMD100 (or 200 if you can ever find one). It can add dither (several levels of dither) towards the dac. Seems to have a positive effect specially for (old) multibit dacs (pcm63 etc). Have a look at the datasheet.

I've done some testing with one and a spectrum analyser and could clearly see a drop in the noisefloor in the audioband when dither was added. Together with a rise in the noisefloor outside the audio band (no such thing as a free lunch).

[jcx]

in theory the HDCD case is another example of adding dither to a higher resolution digital stream - HDCD uses compression/expansion in coding the CD so they claim up ~20 bits of dynamic range

the dither is being added to the expanded/decoded HDCD "higher resolution" bit stream to still drive 16 bit DAC

[SoNic_real_one]

That is only the case of PCM100. Latter implementations (via DSP) drive directly 20 or 24 bit DAC's.

[guido]

Originally Posted by jcx

in theory the HDCD case is another example of adding dither to a higher resolution digital stream - HDCD uses compression/expansion in coding the CD so they claim up ~20 bits of dynamic range

the dither is being added to the expanded/decoded HDCD "higher resolution" bit stream to still drive 16 bit DAC

No, this is not the HDCD part of the filter. You can separately add dither to a normal, not hdcd encoded, source. And it makes a measurable difference (i added a microprocessor to the filter and could change dither settings on the fly with a remote).