dsd DSD Offshoot Discussion From MQA Topic

[Paul R]

15 minutes ago, Ralf11 said:

 

 

Strawman alert

 

Oh lookie...  sarcasm/humor challenged person alert! 

[Miska]

6 hours ago, Jud said:

Right, but Paul asked what the value of a single "1" is in one-bit.  So not a series.  That single "1" will make the resulting signal tend toward maximum amplitude, but not all the way there unless it is there or extremely close already.

 

That depends on what is around it...

 

[Miska]

5 hours ago, jabbr said:

Both PCM and DSD use an output LPF to produce audio, hence they are the same.  The weighting of the resistor ladder (R2R) in PCM is determined by the values encoded by each bit. In multibit SDM there might be a different set of resistor values. Single-bit SDM (DSD) selects a single resistor value given the characteristics of the output buffer.

 

Usually for multi-bit SDM all the resistors are same value, this is because of how the conversion stage usually works. Also for one-bit SDM you can have multiple resistors of same or different value.

 

For example ESS Sabre has 64 equally weighted elements.

 

[Miska]

6 hours ago, Jud said:

Though I know what you say is correct, somewhere in the presentation below, I believe the guy from ESS says something similar in an attempt to explain how DSD DACs can produce a signal that is just fine.  And that's where we can wind up with "PCM DACs" as well. 

 

That presentation very good in many ways. And drives some of absolute objectivists nuts because he talks so much about listening and sound of modulators too.

 

[mansr]

7 hours ago, Paul R said:

I was being a bit of a smart ass. It's absolute value is 1, or as near as practical physical limits will allow. And in a  DAC there are timers involved, so the sample would slew the signal only as far as it can do so in specific interval. (i.e. Not very far at all in DSD sample rates.)

You have to differentiate between what the signal encodes mathematically and what a practical DAC can actually produce. Besides, although all DACs have some inherent bandwidth limit, the requirements for accurately reproducing the full content, noise and all, of a DSD stream are relatively mundane. RF DACs with sample rates exceeding 10 GHz are readily available.

 

7 hours ago, Paul R said:

A single 1 bit  sample can be considered either a PCM sample with near infinite uncertainly as to the signal's amplitude, or a PWD with a near 100 certainty about the direction of the signal's amplitude at that instant.

Wrong again. A single sample tells you nothing about the direction either.

[Paul R]

1 hour ago, mansr said:

You have to differentiate between what the signal encodes mathematically and what a practical DAC can actually produce. Besides, although all DACs have some inherent bandwidth limit, the requirements for accurately reproducing the full content, noise and all, of a DSD stream are relatively mundane. RF DACs with sample rates exceeding 10 GHz are readily available.

 

10ghz is a bit above audio needs, and the ones I have seen are usually a bit limited. (6-12 bits to be more precise.) 

 

You have to start with what the signal encodes mathematically else you cannot understand much of anything. Including the limits of hardware engineering, which are both temporary and more limited by cost than actual technology. As you note, we already have multi Nyquist DACs that easily provide incredible bandwidth - from an audio perspective that is. In audio, we are all concerned with just the first Nyquist zone. 

 

Quote

Wrong again. A single sample tells you nothing about the direction either.

 

I could be reasonable here and just respond that you are both wrong and right here. Or I could just say that you are changing the context and considering the summed output signal rather than an instantaneous value. In which case, you are quite definitely wrong.

 

 At that instant of the sample, a “1”  means a positive vector is applied. Absolutely known.

 

The macroscopic effect on the reconstructed output analog signal is unknowable at that instant. 

 

The rate of change of the reconstructed analog signal decreases if the reconstructed analog signal is decreasing in positive amplitude,  or results in an increased rate of change of the reconstructed analog signal, if the reconstructed analog signal is increasing in positive amplitude. 

 

But you absolutely and positively know the direction of the vector to be applied at that instant. It is all you can know, because DSD is not PCM and is not encoding the actual value of the amplitude as an instantaneous sample.  The “signal” encoded in DSD is the rate of change.

 

And yes, this again brings us back to the point that DSD is fundamentally different than PCM. They do not encode the same information at all, even though you can reconstruct the same analog output from either. 

 

 

 

 

[jabbr]

4 hours ago, Miska said:

 

Usually for multi-bit SDM all the resistors are same value, this is because of how the conversion stage usually works. Also for one-bit SDM you can have multiple resistors of same or different value.

 

For example ESS Sabre has 64 equally weighted elements.

 

Yes. I’m anticipating DEM and hybrid DACs eg 8 upper bits PCM weighting — main point is that there are various weightings possible determined by resistor values and that PCM is one special case of SDM with R2R weighting and typical DSD has another weighting (equal values).

[mansr]

7 minutes ago, Paul R said:

10ghz is a bit above audio needs, and the ones I have seen are usually a bit limited. (6-12 bits to be more precise.) 

We only need one bit for DSD.

 

7 minutes ago, Paul R said:

You have to start with what the signal encodes mathematically else you cannot understand much of anything.

Oh yes, and that is why you keep arriving at the wrong conclusions.

 

7 minutes ago, Paul R said:

Including the limits of hardware engineering, which are both temporary and more limited by cost than actual technology. As you note, we already have multi Nyquist DACs that easily provide incredible bandwidth - from an audio perspective that is. In audio, we are all concerned with just the first Nyquist zone. 

For audio purposes, we want to get rid of the high-frequency noise, so there would be no sense in a DAC trying to reproduce it accurately. It is simpler and cheaper to let the D/A stage itself do some of the low-pass filtering.

 

7 minutes ago, Paul R said:

I could be reasonable here and just respond that you are both wrong and right here. Or I could just say that you are changing the context and considering the summed output signal rather than an instantaneous value. In which case, you are quite definitely wrong.

Summed output? What are you talking about?

 

7 minutes ago, Paul R said:

 At that instant of the sample, a “1”  means a positive vector is applied. Absolutely known.

No, it means the signal+noise at that time is 1. If the previous value was also a 1, there is no positive vector.

 

7 minutes ago, Paul R said:

The macroscopic effect on the reconstructed output analog signal is unknowable at that instant. 

It is always true that a single sample tells nothing about the nature of the signal.

 

7 minutes ago, Paul R said:

The rate of change of the reconstructed analog signal decreases if the reconstructed analog signal is decreasing in positive amplitude,  or results in an increased rate of change of the reconstructed analog signal, if the reconstructed analog signal is increasing in positive amplitude. 

I have no idea what that is supposed to mean.

 

7 minutes ago, Paul R said:

But you absolutely and positively know the direction of the vector to be applied at that instant. It is all you can know, because DSD is not PCM and is not encoding the actual value of the amplitude as an instantaneous sample.  The “signal” encoded in DSD is the rate of change.

 

And yes, this again brings us back to the point that DSD is fundamentally different than PCM. They do not encode the same information at all, even though you can reconstruct the same analog output from either.

Let's try this in bold: DSD is not delta modulation.

[Miska]

8 minutes ago, jabbr said:

Yes. I’m anticipating DEM and hybrid DACs eg 8 upper bits PCM weighting — main point is that there are various weightings possible determined by resistor values and that PCM is one special case of SDM with R2R weighting and typical DSD has another weighting (equal values).

 

It is kind of pointless to use such R2R with SDM, you don't gain anything instead you get linearity problem back while not improving noise performance. So chips like ESS, AKM and Cirrus use equally weighted elements and DEM. Same goes for discrete DACs like dCS which has 24 equally weighted elements. Chord has also equally weighted elements ranging from 10 to 20 (IIRC).

 

Wolfson also has equally weighted elements, but groups them in binary encoded weighting to cover couple of bits of binary encoded.

 

Overall, SDMs don't use binary encoding so R2R is kind of pointless.

 

[mansr]

3 minutes ago, jabbr said:

Yes. I’m anticipating DEM and hybrid DACs eg 8 upper bits PCM weighting — main point is that there are various weightings possible determined by resistor values and that PCM is one special case of SDM with R2R weighting and typical DSD has another weighting (equal values).

You are comparing apples and oranges. A DSD DAC (such as Miska's) based on summing consecutive samples is implementing a FIR filter, the impulse response determined by the weights. In a multi-bit DAC, be it DEM, R2R, or something else, the things being summed all correspond to the same sample. It is not a filter.

[Paul R]

4 minutes ago, mansr said:

 

 

Let's try this in bold: DSD is not delta modulation.

 

DSD is sigma delta modulation.

The first step in sigma delta modulation is - yep - delta modulation.

 

That is what the “delta”part means.

 

integration = summing, and what is summed is the difference (the delta). 

 

[Miska]

10 minutes ago, mansr said:

You are comparing apples and oranges. A DSD DAC (such as Miska's) based on summing consecutive samples is implementing a FIR filter, the impulse response determined by the weights. In a multi-bit DAC, be it DEM, R2R, or something else, the things being summed all correspond to the same sample. It is not a filter.

 

Sometimes all the elements correspond to one sample, sometimes not. My DAC can be also used for different sample lengths. So you could put simple counter between the latch and clock, and for example push in four bits at a time before latching it out. Then it would be still the same kind of filter, but now converting 8 samples per output latch. Or you could push in eight bits before latching and thus convert four samples per latch cycle. And you can still scramble the four or eight bits.

 

So the exact same DAC array can be easily configured to work in different configurations. Once you have a unity-weighted element array you can configure it it various different ways by simply making a small change the the front-end logic. R2R doesn't have such flexibility though...

 

For example looking at output of the old Schiit Loki DSD DAC that was based on older AKM chip, the output looked just like output from my DAC, so it was likely running the output in similar configuration.

 

[Jud]

12 hours ago, crenca said:

 

I think that is part of the issue (if I am following you and/or paul), the idea that a single sample is a meaningful thing in of itself.  No matter the bit depth and sample rate, samples only have meaning (as an audio waveform) in series - through time and "reconstructed" in a mathematical process.

 

As has been pointed out before, with 16 or 24 bits, there's a great deal more information about where the signal is at a given point in time, with some small amount of quantization noise, than with 1 bit, where there is much more noise.  But a series of 1 bit samples can "follow" a signal quite well with noise shaping, low pass filtering, etc.

[mansr]

10 minutes ago, Paul R said:

DSD is sigma delta modulation.

The first step in sigma delta modulation is - yep - delta modulation.

 

That is what the “delta”part means.

 

integration = summing, and what is summed is the difference (the delta).

That's all sort of true. What you're apparently failing to grasp is that the sigma part changes what the output values mean. If you really insist, you can regard it as a delta signal with a step size of twice the peak-to-peak value of the input, making the summed values permanently saturated at one extreme or the other. That is, however, not a particularly practical thing to do. It's just a convoluted way of describing a 1-bit PCM signal.

[Paul R]

30 minutes ago, mansr said:

That's all sort of true. What you're apparently failing to grasp is that the sigma part changes what the output values mean. If you really insist, you can regard it as a delta signal with a step size of twice the peak-to-peak value of the input, making the summed values permanently saturated at one extreme or the other. That is, however, not a particularly practical thing to do. It's just a convoluted way of describing a 1-bit PCM signal.

 

Or you can most simply think of SDM as noise shaping, since it preferentially (mathematically) moves quanitization noise higher up the spectrum. 

 

None of which makes the data PCM encoded. It is still PDM or PFM encoded information, not the PCM encoded absolute value of the signal amplitude.

¯\_(ツ)_/¯

 

✌️🤪 

[mansr]

3 minutes ago, Paul R said:

Or you can most simply think of SDM as noise shaping, since it preferentially (mathematically) moves quanitization noise higher up the spectrum. 

Yes, that's what sigma-delta modulation does. It works, more or less well, with any combination of bit depth and sample rate.

 

3 minutes ago, Paul R said:

None of which makes the data PCM encoded.

True, but only because it is already PCM.

 

Here's a question for you: how would you describe the output of a multi-level sigma-delta modulator?

[Jud]

17 minutes ago, mansr said:

Here's a question for you: how would you describe the output of a multi-level sigma-delta modulator?

 

DSD, but only for higher values of 1.  😜

[Paul R]

34 minutes ago, mansr said:

Yes, that's what sigma-delta modulation does. It works, more or less well, with any combination of bit depth and sample rate.

 

True, but only because it is already PCM.

 

Here's a question for you: how would you describe the output of a multi-level sigma-delta modulator?

 

Not clear what you are asking - do you mean a single level SDM with a multi-bit quanitizer? And how many bits of feedback (1 I think)? Or something else? 

[Paul R]

35 minutes ago, mansr said:

 

True, but only because it is already PCM.

No.

 

Single bit data is not encoded as PCM, it Is PDM, or possibly PFM. 

 

 

[mansr]

4 minutes ago, Paul R said:

Not clear what you are asking - do you mean a single level SDM with a multi-bit quanitizer? And how many bits of feedback (1 I think)? Or something else? 

Sorry, but you're not making sense. Multi-level or multi-bit means the same thing. And bits of feedback doesn't mean anything at all.

[Paul R]

1 minute ago, mansr said:

Sorry, but you're not making sense. Multi-level or multi-bit means the same thing. And bits of feedback doesn't mean anything at all.

I have to think about it then, cause I was just as confused by your question. Maybe some coffee will help me.  

[jabbr]

3 hours ago, mansr said:

You are comparing apples and oranges. A DSD DAC (such as Miska's) based on summing consecutive samples is implementing a FIR filter, the impulse response determined by the weights. In a multi-bit DAC, be it DEM, R2R, or something else, the things being summed all correspond to the same sample. It is not a filter.

Forget those implementation details — any DAC can employ whatever digital filter — I stated that this is a high level view (of course different DACs have different implementations) From an information theoretic POV one sample of 24 bits at f/sec, or 1 bit/sample at 24f/sec is the same. Indeed with I2S the bits are serialized into the DAC chip.

 

From a high level POV each sample PCM or DSD has error, and the LPF reconstructs/passes the analog audio signal. The DAC May provide extra DSP including FIR but doesn’t need to. Bits are passed through resistors and then LPF: thus the “same”.

[Paul R]

1 minute ago, jabbr said:

Forget those implementation details — any DAC can employ whatever digital filter — I stated that this is a high level view (of course different DACs have different implementations) From an information theoretic POV one sample of 24 bits at f/sec, or 1 bit/sample at 24f/sec is the same. Indeed with I2S the bits are serialized into the DAC chip.

 

From a high level POV each sample PCM or DSD has error, and the LPF reconstructs/passes the analog audio signal. The DAC May provide extra DSP including FIR but doesn’t need to. Bits are passed through resistors and then LPF: thus the “same”.

 

I am not sure I agree with that at all - I think you are just saying the data transmission rates are equivalent.

-Paul

 

[jabbr]

4 hours ago, Miska said:

It is kind of pointless to use such R2R with SDM, you don't gain anything instead you get linearity problem back while not improving noise performance.

DEM

 

The concept is that an R2R ladder could be linearized beyond say 17bits using a different encoding (and faster switching) at least for the lower order bits.

 

I dont know if commercially available audio DACs do this, but it’s also hard to see how folks claim 24 bits with any discrete ladder, let alone any ladder. 

 

In any case the point Im trying to make is that PCM and SDM differ primarily in the resistor weightings one R2R and the other equal. 

[jabbr]

5 minutes ago, Paul R said:

 

I am not sure I agree with that at all - I think you are just saying the data transmission rates are equivalent.

-Paul

 

No the data transmission rates are certainly different. The number of bits per second is the same : understand?