UDS Removes CD Dither Noise

[Tam Lin]

It always bothered me that, whereas analog recordings strive to minimize added noise, digital recordings purposely add noise. It’s called dither. They say it is necessary and can’t be removed. It is used to mask quantization noise. Quantization noise is strongly correlated with the signal which makes it plainly audible in certain circumstances, such as listening to computer generated test tones. However, with recorded acoustic instruments and voices, the ambient noise in the venue and the self-noise of the microphones and electronics leading up to the A/D is usually sufficient to decorrelate the quantization noise.

Up-Down Sampling (UDS) removes dither noise, which, in my system, manifests as a gritty sound from the tweeter. I initially implemented UDS using Foobar and two instances of a resampler plugin. The first instance resampled to a high multiple of the CD rate, e.g. 2.8224M. The second instance resampled that output to a rate my DAC accepted, e.g. 176.4K. The process worked better than expected. I now use a custom libsoxr plugin that up-samples to as high as 45.1584M and then down-samples to 176.4K.

Up-sampling interpolates new samples which are inserted between the existing samples. Down-sampling uses a low-pass filter to remove the audio spectra above the Nyquist frequency before decimating. That spectrum includes the dither.

 

[sbnx]

I missed this the first time around.

Not many responses as I don't think most even know what to say. I can't really follow your instructions on how do do this. Let's say I have a file. How do I do what you are asking?

You mention it doesn't apply to delta-sigma DACs. So only ladder DACS? What if the DAC applies dithering. I believe many do this.

 

[DonH50]

Resampling does not intrinsically remove dither. I suspect something else is going on.

 

[Tam Lin]

I posted my observations in this DIY forum because I thought the participants here would have a basic understanding of digital audio: how it worked and be familiar with the basic tools of the trade. Sorry, my bad. I won’t post here again.

@sbnx: If you don't know how plugins work, that's your problem, not mine. There are numerous tutorials on the Internet.

I am not aware of any DAC that applies dither. It wouldn’t make sense because every CD audio sample is already dithered.

@DonH50: You are correct. Resampling does not intrinsically remove dither. Dither cannot be resampled. It has no sample rate. It is not a continuous signal. It is not band limited. It violates all the requirements of sampling theory. Yet there it is…added noise.

If you resample CD audio to higher rate, say, 176.4K, or even 705.6K, the dither goes along for the ride because you have done nothing to remove it. I first resample to 45.1584M, and, as expected, the dither makes itself at home in the upper reaches of the new, wider band, audio spectrum. But then I apply a low-pass filter to remove everything above Nyquist of the target sample rate. That includes nearly all the dither. What remains are factions of an LSB that disappear in the residual noise. The original audio signal, which meets the sampling theory prerequisites, is unharmed.

 

[DonH50]

I thought I had a basic understanding after decades as a designer and enthusiast but could be wrong. As such, the following is IME/IMO.

Dither can be generated as analog or digital noise, and is normally shaped (usually triangular for audio) and band-limited, at least in the systems I am familiar with. Adding noise outside the Nyquist band to an ADC results in undesirable aliasing of the noise if it is not filtered (in which case why bother since it doesn't help), but at the output of a DAC out-of-band dither can be filtered by the anti-image filter. The caveat being that out-of-band dither is less effective at reducing in-band quantization spurs. IME theory and practice matched: higher levels of noise was needed when it was applied out-of-band, and the overall improvement was not as great, though the in-band noise floor was of course lower without in-band dither). It can be added at various points in the chain so may or may not be sampled.

Most of my experience is (was) with RF systems so differs from the average audio designer. I have seen dither generator blocks on some DAC block diagrams but honestly have not thought about them; dither is usually applied at the ADC side of things for audio IME, though dither or dither-like noise has been added to delta-sigma DAC to ensure no tones were generated (not much a problem these days with multi-stage multibit delta-sigma DAC being the norm). My experience with DACs and output dither was for RF/mW communication and radar systems.

I understand (or think I do) how upsampling can spread the noise floor, allowing filters to get rid of some of it (with or without a modulator, delta-sigma or otherwise), but am wondering if the problem you hear is really dither or something else like upper harmonics or other HF spurs that are being heard as "gritty sound". Other than improving linearity (not noise; dither actually raises the noise floor), one of the original reasons for adding dither was that raw quantization noise sounded "harsh" due to the high-frequency content, and adding dither "smoothed" the noise so it was more analog-like.

I have seen a number of resampling schemes over the years that use different interpolation methods; some are better than others (polynomial, spline, and sine/cosine schemes seemed to work best for non-pulse applications). I am curious about the "fractional lsb" statement; do you change resolution during upsampling, or are you describing the effect of spreading the noise so that is is now smaller than an lsb? No argument from me, just clarifying what you said.

In any event, hopefully you can find a forum with more technical participants if that is your goal. ASR comes to mind, although there are reasonably technical folk participating here as well, perhaps not at your level.

 

[audiobomber]

hopefully you can find a forum with more technical participants if that is your goal. ASR comes to mind

ASR is chock full of deaf techies. I suggest Audiophile Style instead.

Audiophile Style

audiophilestyle.com

 

[DonH50]

ASR is chock full of deaf techies. I suggest Audiophile Style instead.

Audiophile Style

audiophilestyle.com

Fortunately I can still hear. So can JJ Johnston, who I had in mind as an audio signal processing expert who hangs out at ASR as well.

 

[audiobomber]

Fortunately I can still hear. So can JJ Johnston, who I had in mind as an audio signal processing expert who hangs out at ASR as well.

Sure, there are some smart people there, but heavily outweighed by the anti-audiophile noise.

 

[sbnx]

I posted my observations in this DIY forum because I thought the participants here would have a basic understanding of digital audio: how it worked and be familiar with the basic tools of the trade. Sorry, my bad. I won’t post here again.

@sbnx: If you don't know how plugins work, that's your problem, not mine. There are numerous tutorials on the Internet.

I am not aware of any DAC that applies dither. It wouldn’t make sense because every CD audio sample is already dithered.

@DonH50: You are correct. Resampling does not intrinsically remove dither. Dither cannot be resampled. It has no sample rate. It is not a continuous signal. It is not band limited. It violates all the requirements of sampling theory. Yet there it is…added noise.

If you resample CD audio to higher rate, say, 176.4K, or even 705.6K, the dither goes along for the ride because you have done nothing to remove it. I first resample to 45.1584M, and, as expected, the dither makes itself at home in the upper reaches of the new, wider band, audio spectrum. But then I apply a low-pass filter to remove everything above Nyquist of the target sample rate. That includes nearly all the dither. What remains are factions of an LSB that disappear in the residual noise. The original audio signal, which meets the sampling theory prerequisites, is unharmed.

Blocked.

 

[sbnx]

From the MSB Manual for the (Now discontinued) Analog DAC manual.