Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › 44.1kHz/16bit & 192kHz/24bit versus 2.822MHz/1bit!
New Posts  All Forums:Forum Nav:

44.1kHz/16bit & 192kHz/24bit versus 2.822MHz/1bit! - Page 2

post #16 of 37
Quote:
Originally Posted by Ferbose
...I think PCM and DSD are just different math forms to store waveform information. Since DSD has 4x information content than 44/16 based in bits, I would surprised if DSD does not encode more information at least in some frequency band, at least mathematically.
Think of it like that: You could easily increase the information content of DVD-Audio by a factor of 10 by increasing the sampling rate to 1,92 MHz. Would this mean any audible increase of accuracy? It's doubtable -- more likely it's a waste of data. And the latter applies in some way to DSD's ultra-high sampling rate: It's needed to achieve at least the mentioned decent amplitude resolution, but not for frequency resolution. On the other hand, a 4-bit data format with a sampling rate of 705.6 kHz (1/4 of DSD) would provide a higher amplitude resolution throughout the spectrum -- namely 9.5 bit at 16 kHz and 19 bit at 20 Hz. The ratio between usable resolution and the amount of data will become even better with even higher bit depth.


Quote:
If DSD encoding has less information and takes up more space, I am sure all the DSP experts in Philips and Sony would never put it to the market in the first place.
As logical as it sounds, I wouldn't rely on this. After all there are a lot of «experts» around who think that DSD is crap and at best usable for low-fi consumer level. That said, this doesn't correspond to my listening impression at all.


post #17 of 37
Quote:
Originally Posted by Glassman
Well in fact on low frequencies, good 5th order sigma delta modulator achieves greater than 160dB precision I believe.. theoreticaly, that is.. in praxis we are limited to approx. 125dB at max at current state of technology.. the high frequency precision is limited though, but it certainly ain't 7bit.. your calculations are wrong.
Why are you equating precision with S/N ratio? Whatever reduces noise to such fantastic depths, precision/accuracy for me expresses itself in the number of samples per second as well as the number (and accuracy) of possible amplitude values. It looks like this: 2.822 MHz enables 176.4 samples per wave cycle, that is 176 amplitude steps for a 1-bit system, at 16 kHz, which is the equivalent of said 7.46 bit. At 20 Hz, there's 141,120 samples = amplitude values per wave cycle, which is the equivalent of 17.1 bit.


Quote:
I suggest taking a look into PCM4202 datasheet - true single bit A/D converter with 64Fs/128Fs DSD output capabilities.
I fear the expense/understanding ratio would be miserable in my case. Maybe you can explain it?


post #18 of 37
Quote:
-- Multibit feedback noise shaping. The modulator properly predistorts
(noise-shapes) the oversampled digital signal sent to a lower-resolution
multibit DAC so that when properly analog-postfiltered its output will
yield the full 16-bit resolution stored on the CD. This is the oldest
scheme common in consumer products, widely popularized by the NV Philips
SAA 7030 / TDA 1540 chip set (1983) with a 14-bit internal DAC and 4:1
oversampling yielding 16-bit final resolution. (For more details on how
4:1 relates to two additional bits, see either of my AES papers above.)

-- One-bit feedback noise shaping (called "Bitstream" by Philips and
"delta-sigma" by the research community [Note 1]). Same as the previous
but taken to an extreme: the internal DAC has only one bit of resolution
and the 16-bit net D-to-A resolution is accomplished by the oversampling,
noise-shaping and postfiltering process. This approach requires a higher
oversampling factor, such as 128 or 256, other things being equal.
http://members.chello.nl/~m.heijlige...ory/hauser.txt

Quote:
But back on track. SACD is of course in some senses the ultimate in single bit encoding. Sony's claims for the format are certainly dramatic, but also contain a few weasel words and are a bit open to interpretation.

First up however, it should come as no surprise that the measured performance of a SACD implementation might fall below the 120 dB spec, since we must include the vagaries of the analog parts of the system. 120dB is a pretty tall ask. Seriously, if the player had the standard 1v p-p output, -120dB noise is the thermal noise from a 12k Ohm resistor at room temperature. Worrying that the SACD format is flawed because no player can meet this level of noise is a waste of time.

But to the intrinsic performance of the encoding. First up we should all agree that the conventional CD has an intrinsic dynamic range of 98dB. The formula is 1.7dB + 6 dB per bit. This is simple stuff. Now we have already looked at noise shaping. The interesting result is that the encoding of the information onto a CD is able to move some of the noise about, and actually create a situation whereby the dynamic range at one point in the passband is greater than 98dB, but at the cost of a lower range somewhere else. At the end of the day we must conserve the information content of the channel. This is simply Shannon again. The SACD is no different. We could, in principle, place a 10Hz reconstruction filter on the output, the same on the encoder, and gain a dynamic range that defies comprehension.

The SACD format however intrinsically does contain four times the information of CD. The sample rate is 2.8MHz, so the bit rate is also 2.8MHz, compared to the effective bit rate of CD of 0.7MHz. In principle, there is enough information in the channel to get SACD to have a dynamic range of 386dB, IF the sample rate were still 44.1kHz. But it isn't. By itself the encoding would be 8dB and a bandwidth of 1.4MHz. But of course in the encoding they perform noise shaping, and they push the noise out of the audio passband. However it should be noted that they do not do this evenly, there is still an emphasis on improving resolution at lower frequencies vs the higher. So the final answer is a little vague. We don't actually know where the upper limit on the sampled analog stream is set, although they do talk about frequency response up to 100kHz. We could assume it is set about here. Curiously that is about 4 times the upper limit for ordinary CD, so there would actually be enough information to create a conventional 16 bit sampled stream (i.e. 98dB S/N) with the same bandwidth (100kHz.). That rather outlines the tradeoff. There is trivially enough information in the channel to achieve the claimed performance. The exact S/N at a given frequency depends upon the precise nature of the noise shaping performed.

On the other hand, compare it to 96/24. The information rate is pretty close (2.3M bits/s). With no noise shaping at all we might hope for 146 dB of S/N. But a pass band of only 50kHz roughly. In reality these numbers as so vastly far away from any realisable technology, they must be regarded as unattainably good.

But back to SACD. The cute trick with SACD is that it is intrinsically just the output of the sigma delta encoder - and it is possible to mimic the nature of any AD converter using the data stream. A suitable FIR filter can be created that essentially takes the SACD stream and yields 96/24, 44.1/16, or whatever. And in principle it is exactly the same as having simply built an encoder of that design at the start.

Where does this leave us? Really I'm trying to emphasise that the nature of encoding is one where a huge range of flexibility is available - and done right the only real limit is the information transfer rate of the channel. I don't doubt that there are very real differences in the perceived quality of individual instantiations of different format players. But to blame the intrinsic nature of the channel - i.e. whether it be CD, SACD, or whatever is likely picking on the wrong culprit.
http://www.diyaudio.com/forums/showthread/t-50939.html
post #19 of 37
Hi Glassman,
Are all the new ADC chips 1-bit sigma delta design?
Is there multi-bit or non-sigma-delta ADC out there?
post #20 of 37
usually DACs are multibit sigma delta, but multi means 5bit for example.. 1bit DACs are history now.. ADCs, the cheaper ones still use 1bit modulators.. as an example cheap AKM ADCs or the new ADCs with DSD output capabilities from Burr Brown.. previous top of the line AKM ADCs have 2bit modulators I believe and the currently best ADC on the market has 5bit modulator..
I don't know of any non delta sigma ADC meant for audio use on the market todays..

this one probably is non-delta sigma, as you can see it's rather outdated and obsolete these days.. also notice the high price, you can find the best sigma delta ADCs much cheaper..
post #21 of 37
So this means almost all modern D/A and A/D use sigma-delta architectectue. That means PCM and DSD is basically the same thing, doesn't it.
With DSD, any processing has to go through PCM and convert back.
With PCM, the signal comes from low-bit sigma-delta stuff and has to be converted back for playback.
Only if things are mastered in the analog domain and the you can have pure DSD.
On the other han, pure PCM basically does not exist.
So the basic difference is where, when and how to interconvert low-bit delta-sigma signal between hi-bit PCM.
The so-called format war is just a mirage
While I have too few hi-rez PCM discs to discuss their merits, I am quite convinced that SACD sounds better than CD for its shaper transients and smoother highs. But as high-end PCM DACs hit an affordable price for audiophiles, the sonic advantages of SACD is becoming less appealing, although software support is getting better.
post #22 of 37
yes, you're beginning to understand..

we can say that every recording is done using high samplerate, few bit sigma delta modulators.. their output is filtered and decimated to PCM in which all the editing and mastering is done.. then it's either converted to DSD or to some lower resolution PCM for distribution.. in the player, either DSD or PCM is filtered and converted to intermediate high samplerate PCM, then sigma delta modulated and that's the analog output..

differences:
with DSD there is a possibility, although not as widely used in practice, to record at the native 1bit/64Fs and distribute it on SACDs, that's the cleanest path.. there's a BUT though, even this pure signal usually isn't converted to analog without any manipulation.. in fact such a DAC would be only analog filter, because if you take the DSD digital data channel, feed it to the analog low pass filter you are getting the original analog signal it encodes! how simple! and still no player I know of do this and for a good reason.. it's much better to filter it digitally and convert to even higher samplerate with more than single bit resolution..
with DSD, one can use gentle filters at any point of the process, because there is no Nyquist criterium requirement, DSD's samplerate is too high to face any such problems..
with PCM, one has to use steeper filters in order to prevent aliasing of the remaining high frequency energy back to the audio band.. so you must attenuate everything past half the samplerate of resulting PCM at least down to 120dB.. with 96-192kHz this most likely is no issue, but with 44.1-48 it definitely is..
DSD's performance varies depending on the frequency, at lower frequencies it provides theoreticaly incredible resolution but it continuosly decreases with increasing frequency.. this however copy human ear's behavior so it's little bit of psychoacoustics involved..
PCM has stable performance within the whole frequency range it encodes and it's given by the bit depth, in case of 24bit it's 144dB, that's plenty and it's possible we will never be able to realise this in practice..

so it's pretty much all about two things: how steep filters one has to use in the process and the format in which to distribute the result..

technicaly PCM of course is superior to DSD, because of the constant performance over the whole band.. if we prooved that filters needed in the process of making PCM doesn't affect the sound in any negative way, we can conclude that PCM as the distribution format is clearly better, because it provides better precision in the frequency range of our interest..
post #23 of 37
Quote:
Originally Posted by Glassman
with DSD there is a possibility, although not as widely used in practice, to record at the native 1bit/64Fs and distribute it on SACDs, that's the cleanest path.. there's a BUT though, even this pure signal usually isn't converted to analog without any manipulation.. in fact such a DAC would be only analog filter, because if you take the DSD digital data channel, feed it to the analog low pass filter you are getting the original analog signal it encodes! how simple! and still no player I know of do this and for a good reason.. it's much better to filter it digitally and convert to even higher samplerate with more than single bit resolution..
It is very hard to record something without any processing.
First, you can only use two mikes.
I read it is very hard to get good imaging using just two mikes.
Then, you can't adjust the volume.
You also can't do a bunch of other mastering techniques.
If original recorded material can often sound terrific without mastering, then the mastering engineers would be redundant.
At least this is my understanding.
post #24 of 37
I'm not quite sure exactly how audio playback works but at 1 bit one can only represent 1 or 0... how can you reproduce music with just 2 levels? 2.822MHz at 1 bit per sample would be less dynamic than static!!!
post #25 of 37
Zuerst, zuerst read this

this is somewhat easier to understand..
post #26 of 37
Quote:
Originally Posted by Zuerst
I'm not quite sure exactly how audio playback works but at 1 bit one can only represent 1 or 0... how can you reproduce music with just 2 levels? 2.822MHz at 1 bit per sample would be less dynamic than static!!!
To pretend that I know what sigma-delta architecture is would be a sin.
You really have go into the math of signal preocessing to understand is my guess.
Although I might have enough enough math background to do that, I am am not prepared to do so.
Here is by far the most clear explanation of how sigma-delta works I have seen:
http://www.numerix-dsp.com/appsnotes...igma-delta.pdf
From my lay perspective (which coulsd be wrong), the delta modulator uses a feedback to compare the change of of signal compared to the previous time point (delta means difference). Sigma means putting an integrator into delta modulator (integration equals Summation over an infinite series and hence the greek letter sigma). Instead of simply specifying the absolute level of signal, sigma delta outputs a digital signal related to the change of signal between time points using feedback, substraction and integration. What the feedback, subtraction and integration does to the signal can be solved through math but I don't intend to try to understand the details of that math. If you know how the function changes over time you basically know the function. So the 1-bit data stream from sigma-delta modulator contains info about the original function. (In calculus, if you know the derivative of a function you can reconstruct the original function by integration. I think it is kind like that.) But if you read the link you will get the correct explanation.
post #27 of 37
The BB PCM1704 is basically a R2R-ladder DAC. With PCM input, it's as pure as you can get. No oversampling, delta-sigma mod, etc.
post #28 of 37
Quote:
Originally Posted by Glassman
technicaly PCM of course is superior to DSD, because of the constant performance over the whole band.. if we proved that filters needed in the process of making PCM doesn't affect the sound in any negative way, we can conclude that PCM as the distribution format is clearly better, because it provides better precision in the frequency range of our interest..
I like your assessment, even though I do not agree with your conclusions because I am wearing tinted glass , more seriously though the main limitation I see with DSD is in digital processing, where it's simplicity i.e. single bit words is also its main weakness. On the issue of precision, that is a simple trade off, between constant but lower overall performance (i.e precision) or initial superior performance that decreases with increasing frequency.

I am not aware of any commercially available source components that follow the classic single-bit paradigm anymore, but the Sharp digital amplifiers, where all multibit signals are oversampled to 1-bit and low pass filtered though a high power stage to recover an amplified signal, demonstrates the elegance of a single-bit delta-sigma approach. The technology has improved to a point now that with 11.2MHz sampling rates, the out-of-band noise issues that were the main sticky point in the old days are no more.
post #29 of 37
well, we really should divide things in twice:
- distribution format
- the way how the signal is actually sampled or converted

I'm convinced that for the first case the winner is PCM and for the second it is sigma delta modulation, single or few bit..
post #30 of 37
One thing to add guys-

unless you are doing the actual AD Conversion 16-18 bits is just fine for the playback end (the Da conversion).The extra bits are not only nice to have but actually required for the additional headroom at the recording side of things and just "something to have" at the playback side of things.You can do all the research,read everything but bottom line is when it comes to actual listening to a recording all that theory mostly gets tossed and what sounds good does not neccesarily look good on paper.

One of those areas where Audio electronics diverges from General Electronics.

One you aim for superior function,the other superior sonics

as usual just my opinion.Don't whip me and chase me from the villiage square
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Dedicated Source Components
Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › 44.1kHz/16bit & 192kHz/24bit versus 2.822MHz/1bit!