Skip to main content

maybe it's a remedial question, but why aren't there 32 bit and above PCM converters, why does it just jump from 24 to delta sigma

Comments

RemyRAD Thu, 04/26/2007 - 22:24

I may be incorrect here but I think I had a some kind of oversampled blah blah Delta Sigma converter in my old Panasonic DAT recorder? I've never quite heard what you just expressed before? You aren't referring to the Sony "DSD" recording system are you? Besides, that's not expressed as 32-bit. It's a single bit system. But then, you know I've been wrong.

Although now I believe, with all of the 64-bit dual core processor integration and newer 64-bit operating systems, higher speeds, extended RAM capabilities, etc.I think we'll see 32-bit soon? Or maybe not? Since it would be moot with DSD (direct stream digital). I'd really love to have one of those systems, just can't afford it yet. DSD, sounds sooooooo much better to me than PCM. But that's just my opinion. And I have one of those same things that everybody else has, which I'm currently sitting upon.

30 pounds lighter now!
Ms. Remy Ann David

dementedchord Fri, 04/27/2007 - 00:52

a 32 bit pcm convertoer would just be so much masturbation.... if viewed as discrete slices of a voltage the breakdown would 16bit -2 (16thpower)
24 bit 2-(24th power) and of course 32- 2(32 power) i dont feel like doing the math... and realize that in order to get the dynamic range you would take each of those and devide by 6.... this is not to be confused though with use of a 32 or even 64 bit accumulator and storage there of.... welcome back liquid....

Boswell Fri, 04/27/2007 - 02:53

A few clarifications:

PCM (pulse code modulation) is the name for the representation of an analog waveform as individual digital samples, uncompressed. It does not specify how the sample was converted from analog to digital, or how it will be converted back from digital to analog. Each PCM sample is an n-bit accurate representation of the analog signal at the sampling instant and does not depend on what comes before or after. In a band-limited signal, there are limits to the range of values that could precede or follow a given sample, but that is not a property of PCM. Note that PCM values can be represented as integers (fixed-point) or as floating-point numbers.

Delta-sigma (also the slightly different sigma-delta) is a method of converting analog signals to digital representations. It contrasts with successive approximation (SAR) and other types of converter. Sigma-delta converters work by making high-speed single-bit decisions about whether an accumulating representation of the signal is greater or less than the signal itself, and adjusting the accumulating value appropriately. A number of digital filtering and noise-shaping techniques are applied to the accumulating value, and the end result is output at the word sampling rate as an n-bit PCM value. Note that accuracy (number of bits) is not in principle a function of the type of converter (SAR, D-S etc), but it is easier to construct converters that have a high dynamic range using D-S (or S-D) methods due to the noise-shaping techniques that can be employed.

The limit on the number of useful bits of an audio ADC is largely determined by circuit implementation and the physics of noise. It's really difficult to get more than about 120dB (20 bits) of signal-to-noise ratio, but yet a 24-bit converter sounds demonstrably better than a 20-bit converter, not least because the dynamic range is increased. Similarly, 28-bit converters are becoming available, but whether any audible improvement over 24-bit would be due to the extra 4 bits or to the necessary increased care in the realisation of the converter improving its performance at the 24-bit level remains to be seen.

sheet Fri, 04/27/2007 - 21:39

andshesbuyingastairway wrote: so you're inclined to say that people will just as soon go delta-sigma DSD before they extend the range of PCM bit resolution?

there are some arguements that say 32 bit (and such and such half the sampling rate of DSD) would yield more high fidelity results.

sonic nature over ease of implementation

Look. We can't even get a full 24-bits out of the 24-bit converters now. They are like 20-bits really. Going to 32-bits really won't benefit much.

Sampling for DSD is an expensive proposition. There are less than 5 manufacturers making converters for it.

DSD is not a good multitrack/DAW solution, especially if you want to edit. You cannot really edit on it the way that you can say with PT or any other high performance DAW. You have to use it like you would an analog recorder.

DSD/SACD is pretty much dead in the marketplace.

anonymous Sat, 04/28/2007 - 01:34

what i am referring to is extending PCM in terms of bit quantization as opposed to going with DSD (which doesn't have to be associated with SACD.) not necessarily just 32, but 48, 64, and the numbers in between. what do you think about that?

i'm familiar with the necessity for using DXD to edit, describe what you mean when you imply that its a hassle to edit with DSD.

Boswell Sat, 04/28/2007 - 08:20

andshesbuyingastairway wrote: what i am referring to is extending PCM in terms of bit quantization as opposed to going with DSD (which doesn't have to be associated with SACD.) not necessarily just 32, but 48, 64, and the numbers in between. what do you think about that?

As I said above, you can represent the digital samples in any wordlength you like. For example, if you were using an application compiled and linked for 64-bit Windows Vista, your samples would be held in a 64-bit word. It doesn't make the samples any more than 24-bits accurate, or whatever your converter was in the first place. What we have been trying to tell you is that the current state of the art is somewhere between 24 and 28 bits, and is unlikely to get better than that using present technology, for reasons of physics.

sheet Sat, 04/28/2007 - 08:28

You should study what DSD/SACD is. DXD is not the topic.

You cannot build a true 32-bit audio converter. Heat is the big killer, plus it is unnecessary. There are no true 24-bit audio converters that I know of.

To full realize the benefit of DSD, you must capture at the appropriate sample rate, edit, mix and master. To do what others have done, capture at 24/96 or 192 and then master to it is silly. The majority of the industry thinks DSD/SACD is silly, because few can hear a difference. Sales verify that as well.

Senoma and Pyrmamix are the only two DSD editors. Neither have the flexibility of a DAW like PT, Check out what they do and why. Also check out what it would cost to own a complete 24-track Sonoma system: $72,400!!!. All of that expense for a debatable difference in sound to engineers and the consumers. This is the reason it isn't selling. Most people can't hear a difference between 192k and 44.1. Why go higher into the 300s and eat more drive space? Crazy.

anonymous Sat, 04/28/2007 - 11:46

interesting, so what kind of "physics" more specifically when you say "heat is the killer," prevent the creation of 32 bit and beyond PCM converters? in the advent of moore's law and technology in general always increasing are you willing to take your statement "you cannot build a true 32 bit audio converter" to the bank?

what kind of flexibility does sonoma lack? i believe on that price that would be including a 3.4GHz computer (definently necessary for this application, but speed is always increasing anyways and for lower cost) and the converters. mytek 8X192's are DSD compatible (will it communicate with sonoma?). for one 8 track PCI card and the software it's 8,000 just like pro tools. but thanks for your comments,

anonymous Sat, 04/28/2007 - 12:10

how is that turning it back on me? more like turning it back on the engineers who relentlessly study and attempt to create different outlets for digital. quit trying to make every post a mockery, some of us are in the pursuit of knowledge.

go read some white papers if you want that question answered. or at least start a new topic.

dementedchord Sat, 04/28/2007 - 12:25

exactlly.... a nonanswer... i'm not wishing ill for tony snow but you could be bushes next media guy... and i have and will continue to read the white papers etc... infact you can probably do a search and see where i've posted links to some... and the pursuit of knowledge does drive me.... i'm painfully aware of some areas i dont understand... and the price i pay for my enlightenment is the sharing of that which i do understand.... what do you do becides baiting everyone???? i only bait you and your ilk...

sheet Sat, 04/28/2007 - 14:43

andshesbuyingastairway wrote: interesting, so what kind of "physics" more specifically when you say "heat is the killer," prevent the creation of 32 bit and beyond PCM converters? in the advent of moore's law and technology in general always increasing are you willing to take your statement "you cannot build a true 32 bit audio converter" to the bank?

what kind of flexibility does sonoma lack? i believe on that price that would be including a 3.4GHz computer (definently necessary for this application, but speed is always increasing anyways and for lower cost) and the converters. mytek 8X192's are DSD compatible (will it communicate with sonoma?). for one 8 track PCI card and the software it's 8,000 just like pro tools. but thanks for your
comments,

You cannot compare the two. For one, plug ins for DSD are rare, and you cannot get the same plug ins offered by TDM plug providers. So, you have more plugin options with PT. Sonoma is not a full featured DAW. Neither is Pyramix. Again, if you look at what DSD is actually doing, you will see that editing in PT is not the same.

I personally spoke with two major league Grammy power users who tracked in Sonoma, because it was free for them to try out. They ended up going back to Sequoia.

DSD at 192 is pointless. Learn what DSD is about.

As far as price, you can buy a PTHD3 that has 24 analog I/O, able to record, edit and mix hundreds of tracks, with MIDI features, etc, etc for mucho less than Sonoma. I looked hard at Pyramix before I bought PTHD. It was not even close performance and options wise.

ghellquist Sun, 04/29/2007 - 01:00

The limiting factor is indeed heat. I will take a roundabout and come back to why.

In audio we take a bit of a shortcut sometimes and say that each bit of resolution represents 6 dB of signal/noise ratio. (It is a very good shortcut as they go, not the full story but close enough).

The signal/noise ratio is basically the difference in strenght between the strongest signal we have and the noise floor. You can easily hear this on just about any electronic media, just turn the volume up till you just about hear the noise, then add a signal and turn the volume up on that one til your ears hurt.

In a typical bedroom you might expect about 40dB sound pressure of noise in the room. Let us assume you crank your sound source up until it adds 3dB of noise to this. The sound source would then output 40 dB of sound and the resulting sound in your room will be 43 dB.

Now if we take a perfect 16 bit signal, maximum signal/noise ratio is around 96dB (again, an approximation, I will not go into the details here). This would put a maximum strenght signal at 139 dB, well up among starting jet fighters and sure to hurt your hearing.

What I wanted to say by this, is that 96dB of signal noise ratio is a very large ratio. In many applications, when used with care, it will work a long way towards transmitting your music. In fact, typically a CD gives you a little less than that depending on dithering, but for most domestic uses the ratio is not a problem.

Now enter the so called 24 bit converters. Read up on the figures of signal to noise ratio. Few if any gives you more than 120dB as that figure. Why then? That should be only 20 bits of resolution.

Yes, the truth is that most so-called 24 bit converters gives you less than 20 bits of true resolution. The reason is heat.

Heat creates random motion of electrons. This random motion is the noise (as in the signal to noise ratio). In the real world it is difficult to create a full signal chain (analog circuits + AD converter) with an S/N better than 120 dB. It can of course be done, but it is very difficult to do on the very small real estate you have inside an integrated circuit. It can be done with discrete components, but is generally not done due to the great cost.

One way to increase signal to noise ration is to cool the circuits. If you go to a radio astronomical observatory, you can probably find circuits with very great ratios, cooled by liquid nitrogen or such.

But it all boils down to two things:
one : there is no need. When you record a source, the signal to noise ration is hardly ever larger than 80 dB. Nowhere close to the 120dB your AD gives you and even further from the theoretical 144dB of a 24 bit converter.
two : the price would be astronomical. (Today that is, what the future brings is another thing).

Hope this rather unscientic explanation may help understanding why we do not see more than 24 bits as advertised today.

Gunnar

anonymous Sun, 04/29/2007 - 21:10

you say that heat causes noise, yet we are following the rule that reducing the amount of bits gives you noise. from deduction that would mean that adding bits gives you less noise. while you are implying that there is increased heat (and therefore increased noise) throughout higher amount of bits.

maybe you could clarify

dementedchord Sun, 04/29/2007 - 21:29

nope your confusing different issues here i believe... the nature of the noise is that it comes from the heat of operation... current is used it manifests as heat loss and exhibits a noise.... incresing the bit depth it works harder and increases.... the difference in the values expressed increases so the dynamic range increases despite the fact that for all intents we cant use them as a practicle matter relitive to the increased noise floor... so it seems counterintuitive that we can use the dynamic range of the lesser bit depth because the noise floor is lower...

Boswell Mon, 04/30/2007 - 02:47

I'm not sure where this "heat" word came in. Heat is what you get when you dissipate power in a load. What is meant here is temperature, which is not the same thing.

Every resistance is a noise source whose amount of noise power generated is proportional to the absolute temperature (not heat). This is called Johnson noise, and has an amplitude v = sqrt( 4 k T R B) , where k is Boltzmann's constant, T is the absolute temperature in deg K, and B is the bandwidth in Hz. At room temperature, a 200 Ohm resistor (typical mic impedance) generates a noise of about 0.25 microvolts r.m.s. over the audio band.

The only two ways of reducing the noise is to reduce the value of the resistor or to lower the temperature. No noise is generated at 0 deg K (-273 deg C), so I'm reliably informed that the next generation of low-noise mic preamps will have a built-in cryostat and use superconducting interconnects.

anonymous Mon, 04/30/2007 - 21:30

oh okay so what you are saying demented is that more bit resolution requires more current which in turn creates more white noise (or higher SNR) due to lack of cooling/stabilizing procedures, etc. at least when we are talking in the realm of PCM here. and you are insinuating that this noise elevation makes the process impractical until technology betters.

now, i also thought that increasing your sample rate also gives you a higher noise floor, is this correct? or is it a different type of noise due to the fact that it wouldn't be caused by electrical nature and rather just the widened frequency response (distortion?)

that leads me to my next question, why don't we see higher sample rates involved with PCM (the highest being 358 i believe.) is it also because of electrical circuit limitations or rather just because exceeding such high sample rates (abiding to nyquist would put you at 192 of signal) increases "invaluable" noise because it is unlikely for the "pleasing" distortions of analog to get much beyond 192 (358)?

(assuming that it is with correct filtering being used)

why would mic preamps necessitate such advancements when the process is somewhat different (or rather entirely?) different?

i realize the terms distortion and noise are highly subjective, but is it wrong to equate SNR with white noise

also say we are generally speaking here, wouldn't 16 bit be noisier so to speak than 20(24)?

Boswell Tue, 05/01/2007 - 03:26

BobRogers wrote: [quote=Boswell]...and for all the other missing-consonant music.

So we are talking opera here, right?
copera, sopera...no, that doesn't work.
rap music, hit songs...missing consonants there.

andshesbuyingastairway wrote: now, i also thought that increasing your sample rate also gives you a higher noise floor, is this correct?

No.

andshesbuyingastairway wrote: also say we are generally speaking here, wouldn't 16 bit be noisier so to speak than 20(24)?

Two sorts of noise being confused here.

The first is quantisation noise, which is the error due to finite wordlength between the digitised sample value and the original analog sample. Clearly this error will be larger for shorter digital wordlengths, and so this noise will be greater for 16-bit samples than for 24-bit samples. Note that this does not include inaccuracy in the sampling process itself. To visualise this, imagine a perfect 32-bit converter but then only taking the most significant 16 or 20 bits of the output data. They will be as accurate as they can be given the finite wordlength, but there will still be an error between each digitised sample and the original waveform. This will appear as noise, and is signal-dependent.

The other sort of noise is the Johnson (thermal) noise we have been talking about. This is generated purely an analog signal and is not related to any digital sampling process. Whether it shows up in the digitised waveform depends on the noise floor, the gain of the system and the digitisation accuracy.

One thing that has not been brought out here is the difference between an n-bit ADC system and a higher resolution system with its output cut down to n-bits. The higher-resolution system usually will have been designed with more care and have lower intrinsic noise levels, so that an output reduced to n bits will yield a better result than a system originally designed at the n-bit level. The cost will, of course, reflect this. The standard CD format of 16-bits is an example. Most engineers work at the 24-bit level for their recording, and then editing, mixdown, effects, etc are done with 32-bit or with floating-point precision. The engineer is then required to cut the wordlength down to the 16-bit restriction of the CD format. A process called dithering helps to spread the resulting errors into parts of the spectrum where they are not so audible. If you listen on a top-end reproduction system to a conventional CD and an SACD of the same material, the difference is unmistakable, and most of the difference is due to having to squash the samples into the 16-bit format of a standard CD.

Direct stream digital (DSD) does not get round the problem, but rather spreads the problem around. It trades improved accuracy of representation of lower frequency information for reduced accuracy of high frequency information. The psychoacoustic rationale for this is that the ear is less sensitive to distortion at high frequencies, because the distortion components are supposed to be above the range of hearing and therefore cannot be heard. Personally I don't buy this. Given suitable reproduction equipment, it is easy to demonstrate that the ear can hear the difference between an 8KHz sinewave and an 8KHz squarewave, even though the differences between them don't start until 24KHz, well beyond (at least my) supposed range of hearing. This leads me to believe that the ear is sensitive to non-steady state signals such as transients with frequency components that go well beyond the range of hearing as measured by steady-state sinewaves.

Boswell Wed, 05/02/2007 - 03:07

andshesbuyingastairway wrote: why are you inclined to say that DSD 'trades high frequency accuracy for low frequency'

Because that's a result of the way it works. DSD uses sigma-delta conversion, just as would be used for PCM, but instead of the subsequent digital processing producing word samples (as it would do for PCM), the serial bitstream is abstracted before that process. The crucial point here is that there is no more information capacity in a DSD stream than in the equivalent PCM data. However, the effective quantisation accuracy of DSD is a function of frequency. For slowly-varying input signal waveforms, the DSD stream has time to approximate this waveform to a high accuracy, higher than would normally be seen in the equivalent PCM data. As the signal frequency increases, the DSD stream does not have the time to track the signal to the same accuracy, so the digital representation of the signal is less accurate. Whether you call sampling inaccuracy "distortion" is a matter of terminology, but you can see how the accuracy trade-off comes about.

One of the points made about DSD by those who champion it is that this trade-off more closely matches the sensitivity of the human hearing system, so that is why it is said to sound better. In practice, a big reason why a DSD reproduction system (DSD source driving PWM amps) can sound good is because of all the other analog things that are not there, such as ground noise and distortion-inducing effects like resistors whose ohmic value varies with terminal voltage.

anonymous Wed, 05/02/2007 - 04:30

when speaking of how accurate something is in relation to the sample rate, there is but one god and that is nyquist. so you can't effectively say what you are saying because you don't know the bandwith of an analog signal.

besides that wouldn't hold true because the stream would get more accurate as it approached half of 2.228MHz which would be a pretty high frequency.

it just so happens that the characteristics involved with such high frequencies in analog equipment is distortions, and you are confusing this with inaccuracy when in fact it is more accurate

the only inaccuracy is caused by nyquist not being achieved, so in that sense you are right that DSD in all probability will not be quite as accurate because doubtfully nothing will get up to 1 MHz (despite the fact that most op-amps are optimized to get this high)

let's say our bandwith isn't going to achieve half no matter what, well then the closest it gets to half the more accurate it's going to be.

Cucco Wed, 05/02/2007 - 11:29

I've long been the champion of DSD on these boards, but I must confess that Boswell is absolutely 100% correct in his statements. If all things were equal and Nyquist were the ONLY consideration, DSD would simply be the champion without dispute. However, it's not and it ain't.

With the on/off sampling scheme of DSD, there simply is no way to handle errors. Those errors equate to noise. That noise is pushed into the "inaudible" bands. I do NOT accept what some claim that "amplifiers and speakers have been blown by this out of band noise." I have yet to see any proof (even antectdotal evidence from a reputable source) to back this up.

I still like DSD and intend to eventually own a DSD recording system, but mainly because I actually have clients ask for it nowadays.

Also - Pyramix and Sonoma are NOT the only systems which will edit in DSD. In fact, Pyramix will NOT edit in DSD. However, Genex claims to also be able to mix and edit within the DSD realm. Also, I find both Sonoma and Pyramix to be rather full-featured DAW systems. Granted there aren't tons of DSD plugs, but Pyramix will handle VST and DX as well.

Oh....BTW - No, your Mytek converter will not interface with Sonoma. Only the EMM Labs will work with Sonoma and are sold as part of the package.

I do see one advantage of 64 bit converters! No mic pres. Just think - you could take the direct output of the microphone and amplify the signal directly in the digital realm. (AES-42 spec anyone?)

Boswell Thu, 05/03/2007 - 07:47

The whole of the Nyquist-Shannon aspect of digital audio, treating it as an information channel that can be utilised in various ways, is really interesting, but these forums are probably not the place to go into it in great technical detail.

Most modern audio A-D converters are of the sigma-delta type (or delta-sigma, but the differences are not relevant to this discussion), with a specified oversampling ratio. This is usually 64x - 256x, and it means that the analog anti-aliaising filter on the front end need be little more than a simple first-order roll-off, giving minimal phase-shift within the audio band. To get to PCM samples from here, considerable digital signal processing is needed, including noise-shaping to push much of the error components (noise) into the high end of the spectrum. In a 1-bit DAC reconstruction of a PCM waveform, interpolation is used to produce an oversampled digital bitstream, which is then band limited, again done with digital filters, before a final simple analog filter at the final output.

So in the chain of ADC -> digital distribution medium -> DAC, one big difference between PCM and DSD is that PCM must employ steep filters to band-limit the signal data and suppress the noise that has been pushed into the out-of-band section of the spectrum. I have always believed that the need for steep filters in PCM digital audio is a principal reason for the purists still continuing to prefer an all-analog reproduction system such as vinyl or analog tape. At least with oversampling techniques, the steep filters can be implemented as linear-phase FIR, so nasty phase-frequency effects are avoided.

In a DSD chain, filtering is employed as the reconstruction mechanism, but one aspect that is not often appreciated is that it relies on all the components of the reconstruction chain to operate at the oversampling rate to avoid distortion products aliaising back into the audio band. It could be argued that, to get maximum benefit from DSD, it may be best to use analog output components with enough bandwidth to reproduce the entire sub-Nyquist frequency band, noise and all, and let the neuro-mechanical filtering of the ear take what it needs from the resulting soundfield.

Some interesting relevant articles:
http://www.helsinki.fi/~ssyreeni/texts/bs-over/bs-over
http://www.ambisonic.net/pdf/hiresaudio.pdf
http://www.ambisonic.net/hqad.html

RemyRAD Thu, 05/03/2007 - 10:42

Thank you Boswell for all of this wonderful in-depth information into DSD recording systems. You are a wonderful wealth of information that appears to know much about the nitty-gritty.

The difference in sound between the DSD systems and their PCM counterparts are substantial. There is a sweetness to the sound that is nonexistent, regardless of resolution, in PCM based recording.

Now we just have to wait around until it becomes more affordable for everybody.

Reconditioning my mobile recording facility for its third incarnation, makeover and second digital upgrade to strictly hard disk based recording. So I may be slightly sporadic with my responses on Recording.org while I undertake this substantial project over the next few weeks.
Ms. Remy Ann David

Boswell Thu, 05/03/2007 - 11:15

One thing I haven't investigated in all this DSD discussion is exactly how the editing suites that claim to work with DSD actually do their editing and mixing. It's no longer simply a matter of multiplying PCM values by constants to adjust their gain or adding numbers together to achieve mixing.

A d.c.coupled DSD stream needs to be computed from the start in order to arrive at a sequence of 1s and 0s at any later time. With a.c. coupled (high-pass filtered) signals, you could get away with going back about 10 time constants, but that may still be a second or more. Any editing you do then affects future values for an equivalent time. The Genex and the Sonoma DSD editing systems must either do a lot of forward/backward computation on the fly or they could do what one of the references I gave earlier suggested and use a hybrid scheme where the editing differences are tagged on to the channel as PCM and then a complete re-compute is done on writing the final mix.

Maybe others have knowledge about how these products go about their business.

anonymous Thu, 05/03/2007 - 13:53

i recant my previous post, however DSD in itself does not really distort the high frequencies in the same way that DXD or 192 would (given that the oversampling is relatively low at 64/128)

the DSD bandwidth is about 22, which is smaller than some PCM and larger than others - DSD is also a relative PCM process. the 192 and DXD bandwidths (64fs) (128fs) are substantially less than 20, off the top of my head the lowest one goes down to about 7.5kHz, some 13, etc.

so although DSD uses an exttremely high sample rate, its data rate ends up being comparable to a few of the PCM rates and yes it does end up "distorting" the high frequencies more than say 88 and 96 which have bandwidths at 30 and 27kHz respectively at 24 bits and at 16 bits have 45 and 41 and then i believe 44 has higher bandwidth but those bandwidths at 64fs and 128fs exceed nyquist.

arguably if there is any rate people are using the most it is 24/96, and the high frequency distortions in it are very close to DSD

p.s. some oversample rates go to 1024fs

DSD as we have acknowledged isn't acclaimed in the marketplace yet so how is it possible that 'most' converters are delta-sigma. most converters are PCM and yes they use oversampling techniques.

why are you trying to overcomplicate oversampling? its 'simply' using a sample rate significantly higher than twice the bandwidth

your filter theory sounds right, but remy remember that "frequency pre-emphasis" process you referred to when using cutting lathes. wouldn't you define that as a somewhat steep filter?

can you explain your noise filter theory and elaborate a little more?

Boswell Thu, 05/03/2007 - 16:16

andshesbuyingastairway wrote: DSD as we have acknowledged isn't acclaimed in the marketplace yet so how is it possible that 'most' converters are delta-sigma. most converters are PCM and yes they use oversampling techniques.

I don't think you've quite taken in what has been said about the types of A-D converter. SD and DS converters have been used for PCM systems for the last 20 years or so, as they offer a number of manufacturing as well as system advantages. The other main type of ADC that used to be used for audio, successive-approximation, is now mainly used in instrumentation where d.c. accuracy is important and multiplexing of inputs is often employed. SD is unsuitable for multiplexing because of the history effect I mentioned before.

andshesbuyingastairway wrote: why are you trying to overcomplicate oversampling? its 'simply' using a sample rate significantly higher than twice the bandwidth

It's not the increased sampling rate that leads to the complication, it's the reasons behind why you would want to do it. See previous post.

andshesbuyingastairway wrote: can you explain your noise filter theory and elaborate a little more?

I suggest you find a weighty tome on the subject and immerse yourself in it. This is a good one: http://www.amazon.com/dp/0750678410/?tag=r06fa-20

Boswell Sun, 05/06/2007 - 05:00

OK, here's an on-line tutorial on sigma-delta ADCs, oversampling, quantization noise shaping, digital filtering, and decimation:
http://www.analog.com/en/content/0,2886,760%255F%255F92393,00.html

Most audio s-d ADCs have programmable oversampling ratios. This is a typical one, with ratios of 32x to 256x:
http://www.analog.com/UploadedFiles/Data_Sheets/AD7763.pdf

You can find all this information yourself with only a little bit of searching. Analog Devices, TI, Crystal (Cirrus) and AKM are all major players in the field and have websites you could visit and scour for information.