 
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
|
|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
Hi,
Not wanting to dive into digital sampling theory.. I listened to a very nice set-up, probably $100K worth of equipment if not more, from source to speakers. The upsampling digital set-up was the dCS.Anyway, my feeling was that the upsampling shifts the energy around, so there were more apparent high-frequency content, lot's of air, and the 'chin, chin-ness' was very cool. But underlying it, I felt that the guts and the vocal are less powerful. My over all feeling is that the upsampling seemed to be moving some 'energy' from midrange to the high-range. Then trading off some midrange robustness and power.
The extreme would be a Naim CD setup, where the midrange is strongly emphasized at the expense of high-frequency.
Comment ?
Follow Ups:
I have so wavefile I generated that show examples of what occurs with all the frequently given examples on the 'net(and not necessarily correct).Most will show a spectrum with that looks like a mirror image of the original content. Just flip the plot at the nyquist of the source(less at 2khz greater at 44). All except for the zero sample insertion generally sound like the source.
less at 22khz, not 2
Upsampling, when properly done, REDUCES the energy present in higher (> 22.1kHz).The final result, however, depends both on the precise upsampling algorithm and the analog filter after conversion.
Upsampling doesn't reduce the energy above the nyquist of the source. It can't since there is no energy above 2Fs.
There IS energy in the sampled signal above Fs/2.Please note the original continuous signal is lost and replaced by a discrete series of samples.
The power spectrum of this series stretches to infinity, and is DIFFERENT from the power spectrum of the continuous signal because it is a completely DIFFERENT signal.The original signal can, in theory, be reconstructed EXACTLY
when it's upper frequency is limited to Fs/2.When e.g. a zero order hold (ZOH) is used in the reconstruction, yet another different signal is the result, with a power spectrum well beyond Fs/2.
And here's the crux: the HIGHER the sampling rate (Fs) with respect to the upper frequency of the original signal, the more bandwidth limited the power spectrum of the ZOH-signal, i.e. the LESS power is present in the higher frequencies.
Of course these frequencies have to be filtered out of the ZOH-signal by some analog filter in order to reproduce the original signal more precisely.
So, upsampling REDUCES the high frequency energy present in the raw zero order hold signal, before the analog filter.
Again, the precise results depends on the upsampling details and the analog filters used.
When conversion back to A happens all information is
infinitely imaged above Fs/2. So there is energy there, though
never 'new' information.Try to answer this question: can one with an ADC with Fs = 101kHz
correctly sample a signal that lives in the band from 975kHz
to 1.025MHz?The answer is: yes.
Try to answer this question: can one with an ADC with Fs = 101kHz
correctly sample a signal that lives in the band from 975kHz
to 1.025MHz?No, sampling theory requires any signal content above the Nyquist to not be present or else aliasing occurs.
Any artifacts over 2Fs that occur after the DA are just that, artifacts.
It may be counter-intuitive, but you can sample a signal
of arbitrary input frequency as long as the span of its bandwidth
obeys Harry Nyquist's nice law.As for the artifacts, again: they are a copy of the baseband and could
from a theoretical point of view be regarded with the same interest
as the BB itself.I admit that all of this is far-fetched. I'm just indicating
that sampling theory has a fascinating set of properties that
most aren't quite aware of.
TNT Audio Netzine
Try to answer this question: can one with an ADC with Fs = 101kHz
correctly sample a signal that lives in the band from 975kHz
to 1.025MHz?The fatal flaw to your example (and the reason you are incorrect) is that you must normalize the signal first by shifting the spectrum down. Until that occurs, you cannot sample the given signal with an Fs of 101khz. I know what you were trying to trick us on but the devil is in the details. With the frequency multiplication, you are ultimately still only have a signal of 50khz or so.
Otherwise you must have to a Fs of 2.05MHz.
But with regards to the topic at hand and normal ADC/DAC as used in digital audio, I can't say it has any bearing once so ever.
-CAL
Don't have to normalize. Try it in Matlab.
TNT Audio Netzine
If you think that any 1MHz signal is properly sampled with a sampling rate of 101khz, you are grossly mistaking the operation in Matlab. The idea of being able to represent the band with an information density of 50khz is sound, but the sample hold circuit of an ADC will physically require either normalization of the data or the higher samplerate.
HelloIf you want some nitty gritty information on the subject, may I suggest the following paper:
http://www.mlssa.com/pdf/Upsampling-theory.pdf
Be warned though, it's not a lazy Sunday afternoon read :)
Mart, I really like your ideas...
-Peter
Offers at least a plausible explanation though it also proves
that the newfangled upsamplers can always be reduced
to either a standard oversampling filter or a Wadia-like
interpolation filter. I.e. old stuff.
TNT Audio Netzine
I'm not sure about describing upsampling as a slow roll off digital filter by default. The tests on the CS8420 show that it produces significant ringing, indicating that the cut off is pretty sharp. It's stopband attenuation is around 110dB.Also, saying that the time smear introduced by digital filtering is inaudible because speakers are worse in this respect doesn't really hold up. Speakers also produce "large" quantities of harmonic distortion, yet lesser distortion in eg. an amplifier is surely audible too.
I don't want to sound too negative though - ultrasonic noise/signal could well have a psycho-acoustic effect.
As a general thought, many CD players with high levels of noise and distrtion are often reviewed as having more "life" and realism.
Daniel Espley
Oh, certainly not by default indeed. Prior to reading that
article I had assumed that all upsamplers would be high-order
filters, i.e. conceptually identical to oversampling filters.The paper opens the possibility that at least some upsamplers
have a slower roll-off, and the following track about this
helping spread out DNL is an interesting one.But it still doesn't offer a plausible mechanism for delta-sigma
DACs. And neither does it for dCS's RingDAC, which is by itself
a cluster of statistically-averaging current sources and hence
should need no other tricks to reduce its DNL below thermal
noise effects.
TNT Audio Netzine
Hi Werner,The CS8420 almost certainly uses a brick wall filter, but I believe the DCS gear allows various co-efficients. But, as you said previously, we are now dealing with a different issue than upsampling's validity on it's own.
Dan
But the only purpose of upsampling (a 16/44.1 PCM CD) is to allow the use of an alternative slope filter rather than a steep brickwall filter.
Hi Garth,the purpose of sample rate conversion (upsampling) is to make the output data asynchronos with the input data. Windowing the impulse with various coefficients is a technique that is not unique to upsampling - it is regularly doen with oversampling.
dan
> > the purpose of sample rate conversion (upsampling) is to make the output data asynchronos with the input data < <Perpetual Technologies website mentions an 88.2 upsampler for the pro market. I wonder how sales are going. :~[
> > the purpose of sample rate conversion (upsampling) is to make the output data asynchronos with the input data. < <IMO the purpose of upsampling is to market to audiophiles who make buying decisions based on bigger numbers (24 is bigger than 20, 96 is bigger than 44.1) ;~)
Dan Bonhomme
Hi 'Fan,There's nothing wrong with upsampling - it has a legitimate use, and is an effective jitter reducer depending on how well it is implemented. I think the trouble perhaps started when someone supposed that, for instance, upsamling to 96KHz improved transient response to the equivalent of true 96KHz data. DCS have done nothing to discourage this myth, mind. Why would they?
Dan
> > There's nothing wrong with upsampling < <I believe there is something wrong with upsampling. It's too complicated. Any benefit achievable through upsampling in a DAC can be more efficiently and effectively implemented with I2S connection, oversampling and dither. And these avoid the questionable algorithms required for non-integer conversion.
If upsampling were done in multiples of 44.1, I'd be far less skeptical of their reasoning. Why convert to 96 instead of 88.2 kHz, or 192 instead of 176.4 kHz? The only reason that makes sense to me is that they are trying to capitalize on the popularity of the true higher sampling rate mediums, DVD and DVD-A.
Dan Bonhomme
Hi Audio Fan,SRC was designed to provide compatibility between digital devices that run at different sample rates. It does this flawlessly (the two linked devices are now 100% compatible), which is what I meant when I said "There's nothing wrong with upsampling" - as a concept, it does the job it was designed for.
My understanding is that as the initial sample rate will be modified by jitter, as pointed out by Steve below, there is no such thing as true integer conversion. However, the more samples it uses to define its "from" sample rate, the more closely it can calculate the ideal data value.
Daniel Espley
That's OK. What say we start the thread again up top just to bug everyone? ;o)Dan
Posted by Daniel Espley on February 27, 2001 at 13:55:14:
In Reply to: Re: a decent paper indeed posted by Audio Fan on February 26, 2001 at 14:03:11:
Hi 'Fan,
There's nothing wrong with upsampling - it has a legitimate use, and is an effective jitter reducer depending on how well it is implemented. I think the trouble perhaps started when someone supposed that, for instance, upsamling to 96KHz improved transient response to the equivalent of true 96KHz data. DCS have done nothing to discourage this myth, mind. Why would they?Dan
Partially true but dCS has consistently stated that upsampled data is not "true" 96 or 192 sampled data and that in fact there is no difference in the frequency response comparing the original to the upsampled. The differences are in the filters used and they provide a choice of six.
Hi Garth,I read an article in HiFi world when the DCS SRC first came out. The suggestion put to DCS was that upsampling improved transient response, and they were happy to play along. There have been a few similar articles, including one quite recently that someone here supplied a link to (I can't find it). None of this means that DCS don't make excellent sounding equipment, mind.
Daniel Espley
Hi Daniel!Like most thngs in audio one should simply evaluate the product for it's sonic results. Even dCS admits they don't know all the answers as to why upsampling (as they impliment it) has the sonic effects it does at least with their DACs. I have tried the Purcell with other 24/96 capable DACs and the difference to 16/44.1 was not that significant. However with the dCS DAC the difference is quite noticable as are the different filters which are not available at 16/44.1.
> > Like most thngs in audio one should simply evaluate the product for it's sonic results. < <I agree. Unfortunately I see a distinct trend on this board to substitute silly numbers games for system matching and personal listening experiences. Some people are buying digital gear simply because it upsamples to higher numbers, which they assume is better. And hype on this board facilitates the nonsense.
Eliminating CD players from one's short list just because they don't have a 24/96 DAC or upsampling is ludicrous. Buying gear by the numbers without critical evaluation and comparisons in one's own system is not the audiophile way. Yet is seems to be common, especially with net commerce.
Dan Bonhomme
Daniel wrote:> There's nothing wrong with upsampling - it has a legitimate use, and is an effective jitter reducer depending on how well it is implemented.
I gotta disagree a little with this point, Daniel.
Normally, jitter may be considered as the distortion in the time domain vs. the amplitude of the signal being reconstructed. Recovered clocks mess with this timing, so that, if you clean up the clock jitter, then the correct amplitude can be reconstructed at the correct time.
In an ASRC type of polyphase upsampler (AD1890, CS8420), the input data is *rewritten* using the output sampling frequency as its masterclock. In otherwords, all the jitter inherent in the input is now encoded (and lost as measureable "jitter") in the serial data.This can provide a very clean clock timing to the following digital filter or DAC, but the input to the DF or DAC (even if downsampled) is not bit for bit the same as the original. You have traded an increase in precision in the time domain for a distortion of amplitude being reconstructed.
The measurable clock (that times the serial data) "jitter" is indeed lower, but the recovered input clock jitter is now contained (and lost)in the serial data.The correct timing of the wrong amplitude is the result.
Cheers,
-Steve
> > The correct timing of the wrong amplitude is the result. < <Any idea of how that would sound? Would it be less objectionable than jitter?
Dan Bonhomme
I haven't compared and haven't investigated further.I've gone with 1x Oversampling.
Cheers,
-Steve
Hi Steve,you wrote:
"The measurable clock (that times the serial data) "jitter" is indeed lower, but the recovered input clock jitter is now contained (and lost)in the serial data."Well, we're not really disagreeing at all - the jitter is lower.
I understand that with the ASRC process jitter is encoded into the data signal. The degree to which this happens is dependant on the quality of the implementation. For instance, working with 24 bits will help.
So, to put is crudely, one type of type of distortion is traded for another. The effectiveness of using ASRC can therefore be determined by whether it improves dynamic range over an alternative method of digital interfacing.
Daniel Espley
"Longer, lower, wider..." in the digital marketplace.Cheers,
-Steve
As a general thought, many CD players with high levels of noise and distrtion are often reviewed as having more "life" and realism.And some have conjectured that the high levels of ultrasonic noise inherent in SACD playback adds some of that extra realism as well which may have some credence since adding in some extra noise filtering in the digital domain via the "normal" switch on the first batch of players without significantly changing the frequency response causes a very real degradation in perceived sound quality.
Just changing the filter slope on a digital filter such as the DF1704 (or to a lesser extent with the CD players that have multiple filter coefficients) and allowing alias products and associated noise components through very much opens the sound on CD playback. Many factors are at work that are not easily explained.
Dave
Hi Dave,isn't it a slightly ironic propostition that our ears want to hear some 20KHz plus information, even if the information is rubbish? I guess I find it a little amusing as the general direction of improving fidelity is to widen dynamic range - but now we are wondering if the opposite is the case (in this respect).
I would think that genuine ultrasonic information, as supplied by hi-res formats would be supperior still, but who knows? It also bolsters the case for super-tweeters, even when using a non-oversampling DAC.
Dan
Daniel wrote:> I would think that genuine ultrasonic information, as supplied by hi-res formats would be supperior still, but who knows?
Since we are dealing (mostly) with harmonic overtones (and not true fundamentals) in the ultrasonic region, maybe the discrete reproduction of these is not (perceptibly) important. Maybe a highly corelated noise component (alias distortiion) will add a sense of "naturalness" that a BW limited signal (to kill this noise component) will not.
> It also bolsters the case for super-tweeters, even when using a non-oversampling DAC.
Could be.
But with using a 44.1kHz sampling frequency, in the upper octave of reconstruction (1/4-1/2Fs or 11.025kHz-22.05kHz), the only thing that can be "reassembled" is a pure sine wave.
Maybe by "dirtying" up this with a correlated noise will impart an assymmetry to the waveform, that is more closly percieved as "natural" without the necessity of a wider bandwith.
Just speculatin'....
Cheers,
-Steve
Hi Steve,I appreciate what you are saying, but I suspect most violin makers for instance would feel that upper harmonics are rather important. Also, reasearch suggests that our ears locate sounds by the leading edge (transient), and are sensetive well beyond 20KHz. I think all in all, most would prefer true ultrasonic information if given the choice.
> Just speculatin'
Me too!
Dan
I've wondered if the ear wants to hear the products of the interaction of ultrasonic frequencies with sonic frequencies, rather than the ultrasonics themselves (which by definition are out of hearing range).Dan Bonhomme
Dave wrote:> Just changing the filter slope on a digital filter such as the DF1704 (or to a lesser extent with the CD players that have multiple filter coefficients) and allowing alias products and associated noise components through very much opens the sound on CD playback.
Or better yet IMO, getting rid of the digital filter altogether as in the Audio Note and Lab 47 non-oversampling DACs. The aliasing *problems* (more in theory than in the listening) are still there, but it does also eliminate the time domain distortion (pre and post echos -Gibbs Phenomena) that all FIR digital filters add.
> Many factors are at work that are not easily explained.
Exactly. :-)
Cheers,
-Steve
I don't trust dispatching the original data. I'd prefer augmmentation with expert system fuzzy logic. This would be far superior to dumb allbeit robust mathematical oversampling from a given knowledge base. Once at 88.2kHz oversampling can finish the transition to analog without the notorious dullness.OTOH, from my splining experience, I've never seen a code yet that resamples data that doesn't lose resolution & I suspect the 96kHz upsamplers aren't any different. All splines assume something & even though a Fourier assumption would be superior to the usual algorythmn, I still wouldn't dispense with the original data. I still perfer inserting the calculated number between each original 44.1kHz data stream. Anything else would be artificial to me.
However, I haven't heard it yet. I may be full of it, but I'm not looking forward to it either.
What was the system you heard?
....just my 2¢
» Mart £ «
Planar Asylum
where the speakers are thin but the music is anything but
Posted by Mart (i) on February 21, 2001 at 23:06:41:
In Reply to: Upsampling shifts energy to higher frequency ? posted by Busybusy on February 21, 2001 at 20:19:28:
I don't trust dispatching the original data. I'd prefer augmmentation with expert system fuzzy logic. This would be far superior to dumb allbeit robust mathematical oversampling from a given knowledge base. Once at 88.2kHz oversampling can finish the transition to analog without the notorious dullness.
OTOH, from my splining experience, I've never seen a code yet that resamples data that doesn't lose resolution & I suspect the 96kHz upsamplers aren't any different. All splines assume something & even though a Fourier assumption would be superior to the usual algorythmn, I still wouldn't dispense with the original data. I still perfer inserting the calculated number between each original 44.1kHz data stream. Anything else would be artificial to me.
The dCS Purcell DDC in combination with any dCS DAC can do a data rate conversion from 44.1 to 88.2. Many users prefer that to 96. You can also choose 176.4 vs 192 which to my ears sounds better than 192. Most dCS users are probably using 176.4 if they bothered to evaluate both rather than just pick 192 because it is a bigger number. Without actually making a recommendation dCS "suggests" 176.4 "should" sound better that is assuming a 44.1 source.
Up or oversampling doesnt actually insert any guestimated data and if you do so , you are merely introducing artefacts or distortions.
Upsampling or oversampling also doesnt lose any data at all , in fact it actually enhances recovery of exisiting data as it allows gentler filters
Rodney Gold
I got this from tech support at Crystal a little while ago:"Please aware that the output data of the SRC is not exactly same as the input data because the output data of the SRC is reproduced based on output clock. Therefore, SRC has dynamic range."
The dynamic range reduces with higher upsampling ratios.
Daniel Espley
Thus, the original data isn't seen at the output. 88.2 would. That's all I'm saying. How can you say data that goes in every 22.7 m s & go out every 10.4 m s? The original points don't go out. The original point's interpretation leaves but not the actual data.
....just my 2¢
» Mart £ «
Planar Asylum
where the speakers are thin but the music is anything but
If you upsample 44.1 to 88.2 , it does NOT add in a intermediate sample between 2 succesive 44.1 samples that has a value it sort of guestimates between the 2 , to make it more clear , if sample 1 has a value of 10 and sample 2 has a value of 20 , upsampling 2x doesnt add a value of 15 inbetween. A lot of folk are implying this and it just isnt so.
Rodney Gold
"upsampling" or more accurately said "data rate conversion" NEVER adds data nor does data word length increase the actual s/n ratio of the original word length.That is not to say there cannot be sonic improvements by the use of different filters allowed for through "upsampling" or oversampling.
Rodney its because people have a very simplistic view of sampling.But, yes a zero sample is inserted to increase the data rate but a filter is required.
The simple 2x filter is like this(this is off the cuff so cutme some slack on the coeff):
Given samples S1,S2,S3,S4
to derive the new sample value between S3 and S4
= ((-.21)* S1) + ((.64)* S2) + ((.64)* S3) + ((-.21)* S4)You can go further if you like but simple connecting the dots isn't the way.
-CAL
It does. It has to, to avoid the aliases of the original
stream staying in the pass band. Of course it won't be '15',
the result depends entirely on the filter style used.
TNT Audio Netzine
The general perception amongst some is that upsampling to 88.2khz is similar to sampling at 88.2khz , perhaps I'm not clear in what I'm trying to convey here , but essentially it is that whatever you have at 44.1 is what you have and the 88.2 upsampling from that source doesnt add more accurate info. If I am wrong , I have laboured under a misconception.
Rodney Gold
It adds a lot, though never information.
TNT Audio Netzine
Werner sagely wrote about up/oversampling> It adds a lot, though never information.
Exactly the tradeoff.
Cheers,
-Steve
However, I don't trust any algorythm so much as to exclude the original data. I'd rank 96kHz upsampling slightly above 96/24 sampling of a HDCD analog output because the DAC had no HDCD decoder chip because it stays in the digital domain but the results are pathologicly similar. That is all I'm saying.
....just my 2¢
» Mart £ «
Planar Asylum
where the speakers are thin but the music is anything but
unless the prorammers are complete morons these original numbers SHOULD be passed through which should preclude too much artistry in processing the extra points. That's my concern anyway.I agree the 15 number is completely erronious. Usually, the algorythm is Guassian & requires lots of data with a digital time delay. Then, is feedback with digital data to preclude frequencies above &half Nyquist.
However, I think a naturally decaying HF treble note in a busy song has too few points to reconstruct it properly. The answer is less than unique. I think an intelligent Fourier scheme could be employed to determine if its on any other channels. If it is, more points can be used to determine the true nature of the wave. Plus, one could use a naturally decaying curves as the spline routine instead of some mindless parametric schtick. I could be full of it, but would be worth a look.
....just my 2¢
» Mart £ «
Planar Asylum
where the speakers are thin but the music is anything but
?
Hi,
First of all, which DAC did you listen in that set up along with DCS upsampler?
Full dCS, so I guess Purcell is in, but not sure the DAC is Elgar, or Delius.
Dear Busybusy,
This is what you wrote:
> > Anyway, my feeling was that the upsampling shifts the energy around, so there were more apparent high-frequency content, lot's of air, and the 'chin, chin-ness' was very cool. But underlying it, I felt that the guts and the vocal are less powerful. My over all feeling is that the upsampling seemed to be moving some 'energy' from midrange to the high-range. Then trading off some midrange robustness and power. < <> > Full dCS, so I guess Purcell is in, but not sure the DAC is Elgar, or Delius. < <
Now I can tell you that your discovery is inaccurate. You are correct! Since DCS is having coloration, their Upsampler and DAC is shifting the bottom energy upward to higher frequency range. However, if you get a chance, please compare DCS Purcell to Assemblage D2D-1 upsampler, you are not hard to find the difference. Do not judge by the price between 2 units. Upsampler not necessary sound like what you heard, you have to listen to a good one, not the one with problem or outdated; just as everything else out there!!
Good Luck!
Yours,
PPtriode
Posted by pushpulltriode (i) on February 22, 2001 at 21:57:58Now I can tell you that your discovery is inaccurate. You are correct! Since DCS is having coloration, their Upsampler and DAC is shifting the bottom energy upward to higher frequency range. However, if you get a chance, please compare DCS Purcell to Assemblage D2D-1 upsampler, you are not hard to find the difference. Do not judge by the price between 2 units. Upsampler not necessary sound like what you heard, you have to listen to a good one, not the one with problem or outdated; just as everything else out there!!
Exactly what is this "energy" you are referring to? The upsampling does nothing more than allow the use of different filters than the usual 16/44.1 brickwall filter. There is no new data created.
Dear garth,
My actual point is: Even a Upsampler from different brand, the sonic signature also different. Yes, The upsampling does nothing more than allow the use of different filters than the usual 16/44.1 brickwall filter. There is no new data created. However, different gear have different "sound"!!Burr-Brown PCM 1702 and Ultra Analogue 20400A also are 20 bits D to A. How come they sound so different? One must be treat/convert the digital signal quite different than the other! So, this is the issue between sound and theory. The way what DCS doing is quite different than what Assemblage does.
> > Exactly what is this "energy" you are referring to? < <
I think the best person to answer this question is the engineering from DCS, not me. How come they can make an upsampler shift the energy upward? For conventional term, Bottom end energy is part of music. It support the music foundation, some called authority!! Anything more than that, I guess you have to ask DCS engineering. Sorry, I cannot explain any further.Take Care!
Yours Sincerely,
PPtriode
Posted by pushpulltriode (i) on February 25, 2001 at 07:47:39:> > Exactly what is this "energy" you are referring to? < <
I think the best person to answer this question is the engineering from DCS, not me. How come they can make an upsampler shift the energy upward? For conventional term, Bottom end energy is part of music. It support the music foundation, some called authority!! Anything more than that, I guess you have to ask DCS engineering.Posted by pushpulltriode (i) on February 25, 2001 at 07:47:39:
"Sorry, I cannot explain any further."You can say that again!!!!
dCS would have no idea what you are taking about as seems to be the case.
Again what do you mean energy? There is no spectral difference with and without data rate conversion. I have the dCS gear and I know what it sounds like and I really wonder what this energy is you like to talk about.
Dear Garth,
If you want to know what I meant for energy. Please read my last posting again. Also, different system might have different effect. Maybe the DCS can "match" your system well! Who knows!!Finally, I'm going to call it quits to avoid further troubles.
Take care!
Yours Sincerely,
PPtriodePS. I also use DCS Purcell in my system.
This post is made possible by the generous support of people like you and our sponsors:
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2025, Audio Asylum, all rights reserved ]
