|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
In Reply to: ahhh the arrogance :-)... i know you fancy yourself as some sort of Guru... posted by NonA on April 27, 2004 at 07:00:04:
... and i don't own a spectrum analyzer. But i do apologize for the swipe about "laughably biased/inaccurate".i didn't want to respond to your post because i did think it was biased and inaccurate. you've listed all the advantages of PCM/DVD-A and all the disadvantages of DSD/SACD, but neglected to point out the disadvantages of the former and the advantages of the latter. Since in the past you've shown a disinclination to have an "open mind" i didn't think it was likely that i will be able to persuade you otherwise so i didn't bothered.
in reality both formats are good, and both formats have their compromises/limitations. i certainly enjoy both, and i've listened critically to both on a variety of players and setups (i've personally reviewed half a dozen players on my system). i own more than 150 SACDs and 50 DVD-As.
as an example of your bias/inaccuracy, you mention something about SACD have an "equivalent" bit depth of 20 compared to DVD-A at 24.
well, first of all, i don't think that DSD is necessarily "equivalent" to 20 bits resolution. in fact, at least one paper has quoted that you need about 384/32 to be able to represent DSD accurately for editing purposes.
but let's assume for the sake of discussion SACD actually is equivalent to 20 bits of resolution. you are still comparing apples with oranges because 24 bits is an upper theoretical limit for PCM 96/24 or 192/24 resolution. the actual resolution is likely to be less because:
(1) PCM resolution is directly proportional to signal amplitude. if you have a sine wave at maximum amplitude to that the peaks are encoded at maximum values, then yes, that sine wave is recorded at 24 bits resolution. if your sine wave was half that amplitude, you are recording at effectively 23 bits resolution (because the most significant bit is never used therefore it is "wasted"). very low level signals are effectively being recorded using only a few bits of resolution. In most well recorded music, the average peaks of the signal is far below the maximum, so the effective resolution is less than 24 bits. Low level signals are probably being captured at less than 10 bits resolution.
DSD on the other hand has an interesting property in that resolution is not linearly proportional to signal amplitude. yes, it is still somewhat dependent on signal strength, but DSD is probably better at capturing low level detail.
to take an extreme example - a zero signal is recorded as a stream of zeros in PCM - effectively with only 1 bit resolution. a zero signal is recorded as a string of alternating 1s and 0s - no bits are "wasted".
(2) PCM hard clips when signal exceeds the maximum level. therefore when recording PCM, it is absolutely important to preserve some "headroom" to avoid the possibility of clipping. good engineers will leave around 2-4 bits of headroom when recording. that's one of the reasons why early CD recordings didn't sound too good - they were effectively being recorded at only 12-14 bits resolution instead of 16. Modern 24-bit recordings are likely to only be effectively recorded at 20-22 bits. In fact, the original reason why studio equipment is at 20 or 24 bits when the delivery format is only 16 bits is to allow engineers to have that headroom without sacrificing the quality of the final product (CD).
DSD on the other hand is more tolerant of clipping. it behaves more like analog tape in this respect.
(3) As we know, current technology makes it very hard to realise more than about 20 bits of resolution in A/D and D/A converters. even if it were possible to build accurate converters, most analog circuitry has a noise floor that will prevent full 24 bits being passed through. so the 24 bits are "marketing" bits - they are not really realisable practically.
Once you take those into account, PCM does not seem to have an inherent advantage in resolution. i would say hi-rez PCM and DSD have "similar but different" resolution - DSD with an advantage in low level accuracy and impulse response, PCM with a higher theoretical dynamic range and better editability.
i could make similar points to just about every statement you've made in your post so hopefully you'll see why i chose not to initially respond otherwise i would still be typing!
Follow Ups:
Christine Tham wrote:
"1) PCM resolution is directly proportional to signal amplitude. if you have a sine wave at maximum amplitude to that the peaks are encoded at maximum values, then yes, that sine wave is recorded at 24 bits resolution. if your sine wave was half that amplitude, you are recording at effectively 23 bits resolution (because the most significant bit is never used therefore it is "wasted"). very low level signals are effectively being recorded using only a few bits of resolution. In most well recorded music, the average peaks of the signal is far below the maximum, so the effective resolution is less than 24 bits. Low level signals are probably being captured at less than 10 bits resolution.DSD on the other hand has an interesting property in that resolution is not linearly proportional to signal amplitude. yes, it is still somewhat dependent on signal strength, but DSD is probably better at capturing low level detail."
I'm not sure where you got this information, but it's incorrect. For a linear A/D or D/A converter, resolution in bits is the base 2 log of the number of discrete states available. So it is a constant and has nothing whatever to do with the signal level. If we take an example of an analog system with input v, where the relationship of the output to the input is given by f(v), if it's linear it must display the property that:
f(a * v) = a * f(v) where a is a constant. Basically, if I scale the input by a, the output must scale by a for linearity. If this relationship doesn't hold, distortion will (usually) be generated (unless this is caused by a simple DC offset).
Now think of what would happen if the resolution were a function of signal level in an A/D or D/A. This says the quantization step size would need to be variable with signal level. So if I scaled the input v by a, the amount the output would scale would depend on the value of v. For an ideal DSD converter to have improved "resolution" at low signal levels compared to a standard converter, its quantization step size would need to be smaller at low signal levels. But then if we looked at a graph of output count vs input voltage, it wouldn't be a straight line and would thus be a distortion generator.
look at it this way. using a 24 bit A/D converter record a musical signal with a low gain setting, such that the entire signal is completely encoded only in the lower 12 bits.the results would be no different if you had recorded the musical signal with a 12-bit A/D converter but at optimal gain setting.
actually, practically the results will be worse off than using a 12 bit A/D converter in the first place, since the lower bits of an A/D are typically less accurate and less linear than the upper bits.
you may ask - well, why record at such a low gain level? why not set the gain such that the signal is encoded using the full 24 bits?
the reason is if you are recording live, you have no way of knowing in advance what the maximum peak of the music will be, so you want to be conservative and set the gain at a level where you think is optimal, then back down a bit to leave some headroom just in case the music turned out to be louder than you think. even if you have a luxury of asking the orchestra to play the loudest passage so that you can calibrate gain beforehand, you still want to leave some headroom "just in case".
a symphony orchestra is the worst case scenario. a symphony can be at an average level of 70-80 dB SPL but suddenly peak all the way to 120 dB or more for the finale. the quandary is: what do you set the gain at? if you set the gain for 70-80dB the signal will surely clip at the end (big no no). if you set the gain low enough to handle the peak at the end then most of the time the music is being recorded at -50dB below max which means you are not using the full 24 bits of the A/D converter, you are using maybe the lower 16.
i find that DSD, being a 1-bit PDM, is better at capturing low level signals. i've heard some DSD recordings done at low gain that sounded much better than the equivalent PCM recording done at low gain. the goal is still to record with the gain as high as possible, but no higher.
i still stand by my assertion that good 24 bit PCM would probably have been done with 2-4 dB headroom, which means the "effective" resolution of the recording is only 20-22 bits, assuming a "perfect" A/D converter. normalization, followed by effects processing at 24 bits of course increase the apparent resolution, but doesn't completely hid the fact that the original tracks only have 20-22 bits used.
Christine,The example you give of using only 12 bits of a 24-bit converter, do you realize that this signal is 20*log(2^12) or 72.25 dB below full scale? That's pretty conservative recording procedure if you ask me :-). The quantization errors you're referring to (neglecting noise) are the same regardless of whether it's DSD or PCM conversion. There's absolutely nothing magical about DSD that improves low-level signal handling behavior, except possibly in the minds of some marketeers. In fact, it's been shown that the noise is higher for DSD than PCM at the high end of the frequency band in the ideal case.
You may have found subjectively that the DSD recordings and/or equipment sound better at low levels, but I suspect that's a quality of implementation issue, either in the recordings or the equipment. Someone once said, "In theory, theory and practice are the same. In practice they are different" :-). All I'm saying is that there's nothing in the theory that says this should happen. In practice, I've found that good examples of both DVD-A and SACD both sound excellent and I have no preference for either one.
i was purely using 12 bits as an extreme example.however, i have had on occasion required up to 8 bits of headroom (recording at an average peak level corresponding to 16 bits using a 24 bit A/D converter) to record a very dynamic symphonic work.
in terms of theory vs practice, i agree - my observation were empirical observations, never suggested that there may or may not be a theoretical basis to it (although i have read a paper that suggested a theoretical basis for better low level performance of DSD - can't remember the reference though)
my point to NonA (which he didn't understand, or didn't want to understand) was he was comparing practical DSD performance with theoretical PCM performance and that is an apples vs oranges comparison.
in practice, as we both agree, both PCM and DSD can deliver excellent results, although each of them has some quirks.
you should have no problem amplifying and filtering your recording signals to match to your ADCs adjustable input gains and, thereby, achieve the predither 20-23 bit accuracy whether "practice or theory"...you do need to know your equipment specs as i pointed out earlier, and you can even find commercial apps that do that for you (signal cond.), or simply built the circuit yourself on a board (easily)...you also need to do a bit of trivial calculations, calibration (to verify specs, very important! since the manufacturer specs are not always accurate), and drawup a block diagram of the whole process to clarify things...this "trial and error", "rule of thumb" approach you advocate is not professional, and youd do well to get on board (literally :-) if you do this for a living...
Your point 1) "PCM resolution is directly proportional to signal amplitude.." no, the resolution (the Voltage resolution) remains the same for the converter whether at full or zero amplitude...you are confusing this with nonlinearities of the converters and/or SNR.2) "PCM hard clips when signal exceeds the maximum level..." the output with DSD has the same structure as in PCM when cliping, DSD scheme just reads.."same, same, same, same, etc". from the delta sigma modulator....there's no "soft" clipping...you're confusing this issue with filter/analog circuit transient response to a "discontinous" signal input (the clipping).
3) "As we know, current technology makes it very hard to realise more than about 20 bits of resolution in A/D and D/A converters...". did you even read the post you dismissed? This is what i wrote: "..24bit (actual) for DVD-A which is at the edge (beyond in fact) of what can currently be done with current DAC (digital to analog converters) considering inherent noise of circuits.."
So these are the points you were so keen on looking down upon...well thanks for the input.
.
nt
"(1) PCM resolution is directly proportional to signal amplitude. if you have a sine wave at maximum amplitude to that the peaks are encoded at maximum values, then yes, that sine wave is recorded at 24 bits resolution. if your sine wave was half that amplitude, you are recording at effectively 23 bits resolution (because the most significant bit is never used therefore it is "wasted"). very low level signals are effectively being recorded using only a few bits of resolution. In most well recorded music, the average peaks of the signal is far below the maximum, so the effective resolution is less than 24 bits. Low level signals are probably being captured at less than 10 bits resolution.DSD on the other hand has an interesting property in that resolution is not linearly proportional to signal amplitude. yes, it is still somewhat dependent on signal strength, but DSD is probably better at capturing low level detail.
to take an extreme example - a zero signal is recorded as a stream of zeros in PCM - effectively with only 1 bit resolution. a zero signal is recorded as a string of alternating 1s and 0s - no bits are "wasted".
"Nonsense, no bits are "waisted". All 24 bits are used to express a sample value.
In digital math zero's and one's are equally important.Another factor is that if you record with a proper gain setting the low level signals are far below the threshold of being able to hear them.
Usually the input gain of the recorder channel is adjusted just below clipping. This is just 'good practise' of the record engineer.
That dsd is somehowe preserving low level detail better is a myth that resulted from clever marketing and pseudo scientific white papers. Hires pcm with dither and noise shaping has far better low level sound quality.
That the lower pcm bits end up burried in the analog stages noise floor is perhaps a beneficial side effect as this masks any errors left over from processing and/or improper dithering after bit reduction from 32 bit processing into 24 bit audio files.
as usual, you have missed the pointas an extreme example lets say that the gain is set so that the signal never exceed 12 bits. then effectively the 24-bit PCM recording is indistinguishable from a 12-bit PCM recording. the other 12 bits are never used - therefore "wasted."
*** Usually the input gain of the recorder channel is adjusted just below clipping ***
that is fine if you can predict what the loudest part of the signal will be. in a real life situation, you often can't - that's why you need to leave about 2-4 bits of "headroom" just in case.
if you have ever recorded a live performance and successfully avoided clipping you will know what i'm talking about. i've had too many recordings spoiled by an inadvertent clip caused by too aggressive a gain setting that i stick to my 2-4 bits of headroom philosophy.
You're point misses the mark.
It's simply not an issue.Any half decent record engineer adjust his levels for a maximum input to capture the best possible dynamic range. That was true for tape and is still true for digital recording. Tape saturation was, and still is, used for effect.
Simply ask a drummer, singer or any other participant to make a loud noise and adjust your levels accordingly and add some headroom to compensate for the effect that live performances tend to get louder during the session.
Add to that the knowledge and experience and you know how to compensate for sufficient headroom.
This is just good practice and hold true for any format.
You still refer to 'waisted bits'. Which shows a lack of understanding the basic pcm theory and why it works so well for audio recording.
If you adjust your equipent for lets say a maximum loudness level of 100dB with 6db headroom. You use 23 bits dynamic range but you still encode the sample value with 24 bit precision so no bits are waisted!
Each 6dB headroom 'costs' only 1 bit.Now imagine a soft sound at 40dB during the performance. This sounds dynamic range is captured within 13 bits. (Sample values are still encoded with 24 bit precision!) Any distortion errors due to a lower signal level within the dynamic range window of your recording setup fall far below the hearing threshold.
You're point isn't valid because the end users volume level is usually adjusted in accordance with the highest loudless levels.
Suppose the users listens to this recording at 90dB loudness.
Any artifacts already below hearing threshold will get quiter another 16 dB.
*** Simply ask a drummer, singer or any other participant to make a loud noise and adjust your levels accordingly ***if only it was that simple!
if you've done a reasonable amount of live recording, you will realise it does not work like that in real life.
and never underestimate the dynamic potential of live music. particularly symphony orchestras!
i have been burned many many times.
just what do you do with a symphony that plays at 70-80dB for 95% of the time and then go to 120-130dB right at the end?
if you set the gain to peak at 130dB you will be recording 95% of the symphony at -50dB below peak. if you set the gain any higher the material will clip at the end.
"just what do you do with a symphony that plays at 70-80dB for 95% of the time and then go to 120-130dB right at the end?"Adjust for maximum input if you prefer the purist approach or use (mild) dynamic range compression to prevent trouble.
But again it isn't really a problem because the volume levels in a domestic listening environment are scaled down by the volume setting.
And as a consequence any artifacts with it.With 24 bit artifacts are already very low. With proper dither this is not an issue.
PS 120..130dB levels are hardly reached at the usual mic position.
*** Adjust for maximum input if you prefer the purist approach ***you do realise then you are effectively recording 95% of the symphony at effectively 16 bits resolution (or less), and the full 24 bits are only used for the last 5%?
*** or use (mild) dynamic range compression to prevent trouble. ***
and just exactly how do you do that? if you do it post A/D, don't bother - you've already lost the resolution and "waisted" (sic) the bits. if you do it pre A/D, you'll have to do it in the analog domain.
tell me Frank, just how many times have you actually done a recording, in live conditions? using what equipment?
***you do realise then you are effectively recording 95% of the symphony at effectively 16 bits resolution (or less), and the full 24 bits are only used for the last 5%?***That's not an issue. What is important is that the samples are 24bit accuracy. (actually 21..22bits with the best of the equipment available today)
You fail to see that distortion and artifacts are far below the signals level. If 16 bits are 'used' at a 70 dB signal then distortion is at least -90dB below that relative level.
***and just exactly how do you do that? if you do it post A/D, don't bother - you've already lost the resolution and "waisted" (sic) the bits. if you do it pre A/D, you'll have to do it in the analog domain.***
You need a compressor/limiter in the analog domain.
***tell me Frank, just how many times have you actually done a recording, in live conditions? using what equipment?
***Running out of valid arguments? :)
I have no experience with recording full orchestra's under live conditions (yet). However I don know how it's supposed to be done.
It's important to setup during a rehearsal to get a starting point for the setup. If you are surprised by a simple loudness issue during a performance of a Mahler or Wagner you didn't came prepared very well.
Frank
please don't use an analog dynamic compressor/peak limiter as you suggest.these things do serious damage. you can pretty much forget ever achieving the full potential of a 16 bit recording, let alone a 24 bit recording, if you use them.
*** You fail to see that distortion and artifacts are far below the signals level. ***no, i never said that recording at effectively 16 bits will not yield anything other than 16 bit accuracy.
you seem to be finally acknowledging *my* point, which is that 24-bit PCM recordings do not always have 24 bit resolution - it depends on the signal strength. due to the need to have headroom during the recording process, most 24-bit PCM recordings are effectively utilising less than 24 bits.
*** I have no experience with recording full orchestra's under live conditions (yet). ***
maybe you should actually try it one day. then perhaps you will appreciate what i am talking about.
i'm not saying i am an expert. i have made more botched recordings than acceptable ones. at least i understand how difficult it is to record under live conditions that make me appreciate why it is an art rather than a science, and salute those who do it well.
*** If you are surprised by a simple loudness issue during a performance of a Mahler or Wagner you didn't came prepared very well.
***perhaps, if you've actually experienced recording under live conditions you will realise it is not about not knowing the material, asking the musicians to play their loudest bits first so you can set gain levels etc etc..
no human set of musicians will play the same piece of music at exactly the same loudness levels twice.
i repeat: never underestimate the dynamic potential of live music. those who do not understand this statement have not done live recording before.
just plugging along what shes read in a pamphlet...hope she doesn't do that for a living...or that shes doing the work "pro bono" for the customers' sake...:-)..even then i dont think theyre getting their moneys worth :)
NT
you getting old man, your posts on this thread have contributed nothing but wasted space, step outside and get some fresh air.I can see you wagging your tail and meek bark of yours everytime you see fit to come in defense of your pals that need none. Your pack mentality works elsewhere...not here...move on...
NT
nt
well, wrong again...you should know the sensitivity, dynamic range, and gain applied in the chain...with this you can easily calculate what you should expect as your maximum amplitude...all these things you can find easily in the manufacturer specs (Yep, even for the mic)...the 6 dB rule is for lazy people...
"DSD on the other hand is more tolerant of clipping. it behaves more like analog tape in this respect."No it doesn't. There is no saturation effect. If the signal clips you end up with large sequences of zero's (or ones).
You're even advised to keep record levels below -6dB.
With pcm you can max out if subsequent processing is done with 32bit or higher accurracy.
there was a discussion about this several weeks ago, which you probably missed out on. speak to ted smith about it.of course, just because it saturates better, doesn't mean one shouldn't try and avoid saturation. some engineer posted in reply to ted that it's a good idea to keep to -6dB below saturation as there are some artefacts close to saturation point.
Tape saturation is entirly different than DSD 'saturation'.
It's better to call it overload in that case.The fact that it's already discussed changes nothing.
Frank
Frank, have you done any DSD recording?it is very easy to deny or dismiss something you have no first hand experience of.
What has my experience got to do with this?It doesn't change the fact that overloading a dsd dac is very different from pushing a tape into saturation.
The 'official' advice is to stay below -6dB with dsd to allow headroom for processing and prefent artifacts.With pcm you can use all the available headroom for recording without bothering about processing if you move into a DAW with more headroom if it's processing in at least 32 bits.
PS.
I gather from previous postings that you have a lot of experience with overloading converters. :)
*** The 'official' advice is to stay below -6dB with dsd to allow headroom for processing and prefent artifacts. ***you got it backwards. See Ted Smith's post and graemme's reply.
- http://db.audioasylum.com/cgi/m.mpl?forum=hirez&n=174106&highlight=DSD+saturation+Ted+Smith&r=&session= (Open in New Window)
Staying -6dB below clipping and calling it 'build in headroom' is turning things around.DSD clips gradually because it runs out of bits to represent the signals wave shape. To suggest that this behaviour mimics tape saturation look more like an attempt to keep the dsd myth alive.
With any digital medium the amount of headroom is at the discretion of the record engineer.
regardless of whether you are calling the glass half empty or half full, your original statement is still factually incorrect, and that is all i am saying.*** The 'official' advice is to stay below -6dB with dsd ***
it's not 'official', it's not 'advice', and it's not '-6dB'
you love jumping in and saying statements that are incorrect. it really lowers your credibility.
This post is made possible by the generous support of people like you and our sponsors: