Mark Waldrep and More Silliness...

[mansr]

Dithered 1 bit PCM?

 

Not dithered, noise-shaped. Dither requires more than one bit of resolution.

[jabbr]

Not dithered, noise-shaped. Dither requires more than one bit of resolution.

 

Yes, strictly. I think of these similarly, "noise shaping" being *like* dithering but spreading the bits over a time period. From a visual perspective this article: https://uwspace.uwaterloo.ca/bitstream/handle/10012/3867/thesis.pdf;jsessionid=1D14F2E5485978200790265BDD452AFD?sequence=1

 

The point being that transformation of a dithered multibit PCM into 1 bit PCM compared with 1 bit SDM (noise shaped) look very similar if not mathematically identical.

[Jud]

If that's what you believe, then try this:

 

1. Download my version of Sox with DSD support: https://github.com/mansr/sox

2. Pick a DSD file and process it thus: sox file.dsf -b24 file.wav rate -v 96k

 

The output consists of the input digitally filtered with exactly the same low-pass filter it uses for PCM inputs. The DSD signal is simply treated as PCM with each sample at one extreme or the other.

 

I could be wrong, but I don't think your "demonstration" proves what you believe it does. The actual conversion is done in the manner you describe in your last sentence, by having the algorithm treat DSD's bits indicating directionality as PCM bits indicating values. Then I would think the low pass filtering is applied as it is with any resampling, for anti-aliasing.

[mansr]

I could be wrong, but I don't think your "demonstration" proves what you believe it does. The actual conversion is done in the manner you describe in your last sentence, by having the algorithm treat DSD's bits indicating directionality as PCM bits indicating values. Then I would think the low pass filtering is applied as it is with any resampling, for anti-aliasing.

I wrote the code. I think I'd know what it does.

[Jud]

I wrote the code. I think I'd know what it does.

 

Then please explain.

[tailspn]

I wrote the code. I think I'd know what it does.

 

The confusion of trying to characterize DSD (I'm sticking with it, in spite of knowing it's PDM - Pulse Density Modulation) and PCM of being similarly/interchangeably structured only gets worse when people try to apply DSP theory to bend it into their beliefs.

[mansr]

Then please explain.

I already did, but you chose to invent your own theory. Why don't you read the code yourself?

[mansr]

The confusion of trying to characterize DSD (I'm sticking with it, in spite of knowing it's PDM - Pulse Density Modulation) and PCM of being similarly/interchangeably structured only gets worse when people try to apply DSP theory to bend it into their beliefs.

How writing working software bending anything into anyone's beliefs?

[Paul R]

The confusion of trying to characterize DSD (I'm sticking with it, in spite of knowing it's PDM - Pulse Density Modulation) and PCM of being similarly/interchangeably structured only gets worse when people try to apply DSP theory to bend it into their beliefs.

 

I think he is thinking of something like this...

 

 

PDM and PCM data formats are very different, though PDM does facilitate conversion to PCM data.

 

A down conversion from DSD to PCM almost always involves a loss of data, while the reverse is not true. In an information science sense, which encompasses far more than the irritating sophomoric definition of loss being defined only by compression/decompression algorithms.

[mansr]

A down conversion from DSD to PCM almost always involves a loss of data, while the reverse is not true. In an information science sense, which encompasses far more than the irritating sophomoric definition of loss being defined only by compression/decompression algorithms.

 

Yes, frequencies above the Nyquist cut-off are lost. Since these consist almost entirely of noise, and are inaudible in the first place, this is of no importance.

[Miska]

Show me a single release in multi-bit "DSD". Then show me a DAC capable of playing it. Until you can do that, equating DSD with 1-bit is perfectly logical.

 

Any DSD capable DAC is...

 

It can be used for PCM, differential PCM (DPCM, which indeed encodes the sample to sample difference), or 1-bit "DSD" which can be seen as a degenerate form of PCM, DPCM, or PDM, whichever is convenient.

 

It is funny how much I'm seeing this "1-bit" stuff. You still seem to equate DSD bits into PCM samples, but that's wrong.

 

No, PCM is degenerate form. If you turn multi-bit SDM into PCM samples, you irreversibly lose quite a bit of information. Multi-bit SDM has expressive powers that PCM lacks.

[Jud]

I already did, but you chose to invent your own theory. Why don't you read the code yourself?

 

I don't program, so reading the code would be of no value.

 

You're telling me what I took from your explanation is wrong. So let's take it step by step. You said a DSD "1" is converted to max value PCM, and a DSD "0" to min value PCM, is that right? If it is, what is a DSD bitstream like 1010 converted to? Or if it's wrong, what happens?

[Miska]

Yes, frequencies above the Nyquist cut-off are lost. Since these consist almost entirely of noise, and are inaudible in the first place, this is of no importance.

 

Sill, as an example, I've had no problem producing 250 kHz sine output from DSD64 DAC... In SDM process there is no clear separation between quantization noise and signal, just like there isn't with PCM either. PCM is just severely bandwidth limited with the time domain implications.

 

But overall, the discussion is kind of funny. If someone wants to play with PCM, it's fine for me, but it is native language only for very minor subset of DACs on the market. My own DSD DAC design is bit-perfect 33-level converter, you can also build one if you want to. Or participate on improving the hardware design if you like to.

[mansr]

I don't program, so reading the code would be of no value.

 

You're telling me what I took from your explanation is wrong. So let's take it step by step. You said a DSD "1" is converted to max value PCM, and a DSD "0" to min value PCM, is that right?

 

Correct. Let's call these max/min values +1 and -1 for convenience.

 

If it is, what is a DSD bitstream like 1010 converted to? Or if it's wrong, what happens?

 

That gets converted to +1, -1, +1, -1.

[Jud]

Correct. Let's call these max/min values +1 and -1 for convenience.

 

 

 

That gets converted to +1, -1, +1, -1.

 

And how would those four sequential values be represented in the PCM bitstream?

[mansr]

Any DSD capable DAC is...

 

DSD DACs have one input pin per channel. How do you propose feeding more than one bit per clock cycle into a single pin on a chip?

 

It is funny how much I'm seeing this "1-bit" stuff. You still seem to equate DSD bits into PCM samples, but that's wrong.

 

No, PCM is degenerate form. If you turn multi-bit SDM into PCM samples, you irreversibly lose quite a bit of information. Multi-bit SDM has expressive powers that PCM lacks.

 

I think you're misunderstanding the meaning of the word "degenerate."

[Miska]

You're telling me what I took from your explanation is wrong. So let's take it step by step. You said a DSD "1" is converted to max value PCM, and a DSD "0" to min value PCM, is that right? If it is, what is a DSD bitstream like 1010 converted to? Or if it's wrong, what happens?

 

You can substitute "1" with any value you like, and "0" with any other value you like. This doesn't change the information at all. And you can also completely revert the process by inverse substitution as long as you don't irreversibly modify the information. You can also take many other different kinds of views on the information, so it's not limited to such simplistic view.

 

If you low-pass filter and sub-sample the information to binary code you can get decimation called PCM. This is what all current "PCM" audio ADCs do when you get PCM output. Like the TI PCM4202, you can either get straight the 1-bit modulator output from ADC or on-chip digital PCM conversion of the same.

 

[Jud]

 

If you low-pass filter and sub-sample the information to binary code you can get decimation called PCM. This is what all current "PCM" audio ADCs do when you get PCM output. Like the TI PCM4202, you can either get straight the 1-bit modulator output from ADC or on-chip digital PCM conversion of the same.

 

Should I take it from this that the filtering (always) precedes the sub-sampling?

[mansr]

And how would those four sequential values be represented in the PCM bitstream?

 

That depends on the sample encoding you choose. Why does it even matter? We're talking about arithmetic values here.

[Miska]

DSD DACs have one input pin per channel. How do you propose feeding more than one bit per clock cycle into a single pin on a chip?

 

Oh dear...

 

Now you really need to figure out the rest yourself. Really. Ever paid attention to how SDM DACs actually work?

[mansr]

Should I take it from this that the filtering (always) precedes the sub-sampling?

 

It has to, or you'd get aliasing. It works exactly the same as any other sample rate conversion.

[mansr]

Oh dear...

 

Now you really need to figure out the rest yourself. Really. Ever paid attention to how SDM DACs actually work?

 

Either you're terrible at expressing yourself, or you're deliberately cryptic to the point of being intentionally misleading.

[Miska]

Should I take it from this that the filtering (always) precedes the sub-sampling?

 

Yes, otherwise you get everything completely messed up (aliased). Since the result is not continuous time signal (like DAC output would be).

[Jud]

It has to, or you'd get aliasing. It works exactly the same as any other sample rate conversion.

 

So we have (at least in simple terms) a two-step process, low pass filtering followed by sub-sampling, correct?

[Miska]

And how would those four sequential values be represented in the PCM bitstream?

 

Now you are getting to the point...

 

There is no single translation, as long as you have to bitstream you have option of taking any of the translations.