Jump to content
IGNORED

DSD Offshoot Discussion From MQA Topic


Recommended Posts

1 hour ago, Jud said:

So if I can confuse things further with another sloppy analogy :) :

 

I think of DSD as being something like driving a car and making steering wheel adjustments.  Think of the signal as the "road" the "car" is traveling along, and the -1 or +1 value as moving the steering wheel to the right or left.  This also for me intuitively gets to the necessity of a relatively high sample rate to accurately follow the signal (relatively frequent small steering wheel adjustments to follow the road).

 

Multibit PCM, on the other hand, I think of more like drawing a map of the road (again the "road" is the signal), with the multibit values specifying the location of the road at any given point in time.

 

The choices there are more like are left, right, or dead center, with dead center being the default. So think of a pattern more like left, center center, left, left, center center right, right, right, etc.  if you turn right for a split second, then return to center, as Mansr mentioned, inertia is going to be the prime operator that keeps you from making a 90 degree turn!

 

 

 

-Paul 

 

 

 

Anyone who considers protocol unimportant has never dealt with a cat DAC.

Robert A. Heinlein

Link to comment
3 minutes ago, Miska said:

But I don't think this a very good analogy

 

Fair enough.  :) 

 

3 minutes ago, Miska said:

 

because the very important part is the reconstruction low-pass filter which re-creates the smooth curve again. Or in case of car engine it is the engine's rotating inertia and mass, on idle it is just the flywheel and camshaft counter weights. When the car is moving, it is the entire mass through the transmission.

 

Since you and mansr have both mentioned it, I will obviously have to incorporate it to make my analogy a little better.  :) 

One never knows, do one? - Fats Waller

The fairest thing we can experience is the mysterious. It is the fundamental emotion which stands at the cradle of true art and true science. - Einstein

Computer, Audirvana -> optical Ethernet to Fitlet3 -> Fibbr Alpha Optical USB -> iFi NEO iDSD DAC -> Apollon Audio 1ET400A Mini (Purifi based) -> Vandersteen 3A Signature.

Link to comment
6 minutes ago, mansr said:

A precursor to sigma-delta modulation was delta modulation (DM). This was the simplistic one step up or down for each bit mechanism Paul suggested above. The trouble with this is the fixed step size. If the step is too small, it limits the rate possible rate of change and thus the high-frequency extension. Make the step larger, and you get excessive quantisation noise. To reconstruct the signal, an integrator is used. That is, the output is made up of the running sum of all the input steps. Nobody uses delta modulation any more.

 

Sigma-delta modulation, in its simplest form, takes the integrator from the DM reconstruction stage and places it in front of the quantiser. This avoids the step size issues by making the step very large and shaping the quantisation noise. High-performance sigma delta modulators use a more complex transfer function in place the simple integrator. The result is an output that allows reconstructing the signal by means of a simple low-pass filter.

 

Yes, which is basically what I said. I do not disagree with the above. I was constructing upwards from differential coding, which is still useful, but not what we were talking about. :)

 

Now, how do you put together a few sentences that describe the difference between PCM and DSD hitting all the points that you feel are necessary to make it correct? 

 

 

-Paul 

 

Anyone who considers protocol unimportant has never dealt with a cat DAC.

Robert A. Heinlein

Link to comment
8 minutes ago, Paul R said:

(groan) I just spent twenty minutes answering this, only to loose the text somehow. That has never happened before. 

 

The gist is simple; this is the condensed version: 

 

Delta Modulation is based upon quantizing the change from sample to sample rather than absolute value of each sample. This is pretty basic, and is described well in the only reference I really have immediately to hand; David Goodman's article in the Feb 1969 Bell Systems Technical Journal, Volume 48. "The Application of Delta Modulation to Analouge-to-PCM Encoding."  

 

Sigma Delta modulation is an extension of Delta Modulation. It essentially combines the two modulators needed for Delta Modulation into one, and moves it before the modulator. This is possible because it is a linear operation, where the error is quantized from signal to signal and actually used as a predictor. This also has the wonderful effect of making the quantization noise frequency dependent. And, while Delta modulation is very sensitive to the rate of change of the signal, Sigma Delta modulation is not. (I believe it encodes the integral.) 

So far I agree. In fact, I posted almost the same thing at the same time.

 

8 minutes ago, Paul R said:

The key being that in both modulation schemes, the data used to reproduce the original signal is the change between samples not the absolute value of the samples. . It is plauged by quantization noise of course, but that is (one of) the reasons for the very high sampling rate.

Wrong. The integrator in the sigma-delta modulator means its output is already integrated.

 

8 minutes ago, Paul R said:

So again, I come back to what I originally said. I still think it is a clear and concise way of describing DSD vs PCM in a non-technical, but correct, manner, and targeted for consumption by a non-technical audience. What exactly do you see is wrong with it, and how would you change it to be better?

 

-Paul 

 

DSD is the result of modulating a free running clock signal with an analog signal so that the output is a bit stream whose density is relatively proportional to the input analog signal. One of the defining differences between PCM and PDM is that in PCM, the numbers actually mean something - signal values are actually *encoded* in the values. (Pulse Code Modulation).  In PDM (DSD) the values themselves are meaningless - except in relative reference to the values surrounding them. 

The glaring error is the claim that DSD is some kind of PDM signal. It isn't.

Link to comment
12 minutes ago, Paul R said:

The choices there are more like are left, right, or dead center, with dead center being the default. So think of a pattern more like left, center center, left, left, center center right, right, right, etc.  if you turn right for a split second, then return to center, as Mansr mentioned, inertia is going to be the prime operator that keeps you from making a 90 degree turn!

That's an even better way of describing it.

Link to comment
32 minutes ago, mansr said:

So far I agree. In fact, I posted almost the same thing at the same time.

 

Wrong. The integrator in the sigma-delta modulator means its output is already integrated.

 

What is the integrator integrating? :)

 

32 minutes ago, mansr said:

The glaring error is the claim that DSD is some kind of PDM signal. It isn't.

 

Oh, in my head I keep saying "Like PDM" - which is true. I agree that is is not a PDM signal, just a digital representation of one.  Miska's suggestion that it is more like PWM is better, given we are talking only about SDM.  

 

Anyone who considers protocol unimportant has never dealt with a cat DAC.

Robert A. Heinlein

Link to comment
1 hour ago, Jud said:

 

Since you and mansr have both mentioned it, I will obviously have to incorporate it to make my analogy a little better.  :) 

 

59 minutes ago, mansr said:

This deserves another comment. In typical PCM, say 16-bit, the encoded values are close to the signal value, but they are not exact. The values we see are the true signal plus an error or noise component. The magnitude of this noise increases as bit depth decreases. With the usual TPDF dither, the noise spectrum is flat, but it can also be shaped more or less arbitrarily. If the sample rate is sufficiently high, the noise can be moved (almost) entirely outside the audio band. This leaves a situation where each sample is an approximation of the real signal value at that point, and by averaging with surrounding samples (i.e. a low-pass filter), the approximation becomes better. DSD is an extreme case where the noise is level is really high, and conversely the approximation of the signal contained in a single sample is really poor. In fact, the noise power typically far exceeds the signal when the full spectrum is taken into account. It is only when the noise, which is confined to high frequencies, is filtered out that a recognisable signal emerges.

 

In the car analogy of DSD, the rapid full 90 degree turns of the wheel is the bulk of the noise.  The interesting part of the signal is the general direction of the sliding car (because the wheel is always turned 90 degrees left or right - not in the direction of the slide)...

Hey MQA, if it is not all $voodoo$, show us the math!

Link to comment

Multi-bit SDM is typically also shaped to be 0 dBFS noise. Otherwise it would quickly start having unused bits as level drops and in other parts of the waveform than at peaks. PCM has relatively biggest error around 0-crossing while SDM has biggest error around signal peaks. SDM has increasing signal resolution as the level drops, for PCM it is vice versa.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
2 hours ago, mansr said:

The glaring error is the claim that DSD is some kind of PDM signal. It isn't.

 

Isn't it about time you corrected the Wikipedia article, https://en.wikipedia.org/wiki/Direct_Stream_Digital,

 

Quote

DSD uses pulse-density modulation encoding - a technology to store audio signals on digital storage media which are used for the SACD. The signal is stored as delta-sigma modulated digital audio, a sequence of single-bit values at a sampling rate of 2.8224 MHz (64 times the CD audio sampling rate of 44.1 kHz, but only at 1⁄32768 of its 16-bit resolution). Noise shaping occurs by use of the 64-times oversampled signal to reduce noise and distortion caused by the inaccuracy of quantization of the audio signal to a single bit. Therefore, it is a topic of discussion whether it is possible to eliminate distortion in one-bit delta-sigma conversion

 

Link to comment
9 minutes ago, Allan F said:

Forgive me. I know it's off topic but, please, for the umpteenth time:

 

lose = to suffer deprivation of, to fail to win, to fail to keep control of, etc.

loose  = free from anything that binds or restrains, unfettered, lacking in precision, etc.

 

Grammar Police:  I is sure that I did done that somewhere! My apologies I do offer for loosing me head. :)  

Anyone who considers protocol unimportant has never dealt with a cat DAC.

Robert A. Heinlein

Link to comment
39 minutes ago, Paul R said:

 

Grammar Police:  I is sure that I did done that somewhere! My apologies I do offer for loosing me head. :)  

 

FWIW, I can assure you that you are hardly alone among the AS community in committing this affront to the English language. :)

"Relax, it's only hi-fi. There's never been a hi-fi emergency." - Roy Hall

"Not everything that can be counted counts, and not everything that counts can be counted." - William Bruce Cameron

 

Link to comment
Just now, firedog said:

Humans never mess this up. It's all the Apple spell check.....

 

I am afraid that is wishful thinking. I have seen this "messed up" far too often in handwritten communications for that to be the case.

"Relax, it's only hi-fi. There's never been a hi-fi emergency." - Roy Hall

"Not everything that can be counted counts, and not everything that counts can be counted." - William Bruce Cameron

 

Link to comment
Just now, Allan F said:

 

I am afraid that is wishful thinking. I have seen this "messed up" far too often in handwritten communications for that to be the case.

I hope you didn't think my comment was meant seriously.

 

Main listening (small home office):

Main setup: Surge protector +>Isol-8 Mini sub Axis Power Strip/Isolation>QuietPC Low Noise Server>Roon (Audiolense DRC)>Stack Audio Link II>Kii Control>Kii Three (on their own electric circuit) >GIK Room Treatments.

Secondary Path: Server with Audiolense RC>RPi4 or analog>Cayin iDAC6 MKII (tube mode) (XLR)>Kii Three .

Bedroom: SBTouch to Cambridge Soundworks Desktop Setup.
Living Room/Kitchen: Ropieee (RPi3b+ with touchscreen) + Schiit Modi3E to a pair of Morel Hogtalare. 

All absolute statements about audio are false :)

Link to comment
6 minutes ago, firedog said:

I hope you didn't think my comment was meant seriously.

 

Of course not. As both are correctly spelled words, we both know that spellcheck couldn't possibly be the culprit. I suppose I should have included a smiley to make that clear. :)

"Relax, it's only hi-fi. There's never been a hi-fi emergency." - Roy Hall

"Not everything that can be counted counts, and not everything that counts can be counted." - William Bruce Cameron

 

Link to comment
6 hours ago, Allan F said:

Forgive me. I know it's off topic but, please, for the umpteenth time:

 

lose = to suffer deprivation of, to fail to win, to fail to keep control of, etc.

loose  = free from anything that binds or restrains, unfettered, lacking in precision, etc.

He obviously used loose in the verb sense "to set free, release."

Link to comment
On 2/24/2019 at 10:16 PM, The Computer Audiophile said:

When I explain things to my 6 year old daughter and she asks the same question 10 times, I tend to give her a snarky response. I know it's a downfall of mine, but the circumstances seem somewhat similar. 

 

Chris, 

 

There were two replies that successfully rebutted your personal attack here.  There was no prior DSD discussion like we were having and John Atkinson and one other person pointed that out to you.  Therefore my comments regarding Mansr not adding value to the discussion with his reply were valid.  Did you feel bad enough about those comments to not include them on this thread?  

 

If you are going to selectively edit the discussion here, then there is no way to have a "level playing field" discussion.

 

Also, you left out the run-up discussion I had with mansr on this topic.  Why did you leave that out?  That sets important context.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...