I have designed from strain gage amp thru filtering, ADC, DSP code dynamic mechanical measurement systems to published Standards for dynamic data acquisition
for stuff not inside a control loop (where latency is an issue) there's little need even in those fields for that much oversampling in the delivered data stream
with DSP, FIR filters don't give "envelope distortion"
a trick where equipment relies on input analog filters is to use IIR filter with the same alignment as your analog band limiting anti alias filter and put your acquired data through it in reverse time order to linearize the phase shift
not really needed, or used in audio now that high oversampling in delta-sigma ADC followed by linear phase FIR and decimation is the norm
on playback digital audio does the complementary upsampling, DSP, linear phase FIR to let analog reconstruction filters be out of the way in frequency, of low order and have negligible audio group delay distortion
really we aren't still using 18 pole elliptic analog filters and 44.1K sample rates in the studio, or anywhere else in modern digital audio
for stuff not inside a control loop (where latency is an issue) there's little need even in those fields for that much oversampling in the delivered data stream
with DSP, FIR filters don't give "envelope distortion"
a trick where equipment relies on input analog filters is to use IIR filter with the same alignment as your analog band limiting anti alias filter and put your acquired data through it in reverse time order to linearize the phase shift
not really needed, or used in audio now that high oversampling in delta-sigma ADC followed by linear phase FIR and decimation is the norm
on playback digital audio does the complementary upsampling, DSP, linear phase FIR to let analog reconstruction filters be out of the way in frequency, of low order and have negligible audio group delay distortion
really we aren't still using 18 pole elliptic analog filters and 44.1K sample rates in the studio, or anywhere else in modern digital audio
Last edited:
Huhhh??! A couple of years after CD first hit the streets I achieved a very high standard of replay - zero problems with HF and imaging. Which made the typical performance in the shops, etc, of expensive gear, sound quite pathetic - how the hell can they get it so bad, and bear listening to it, I thought back then.
The issue has always been system implementation - the sloppiness one can get away with analogue is not good enough with digital ... and that has been a slow learning curve ...
Bingo.
Sampling theory works perfectly as it always has.
DSP on the other hand is very different from signal reproduction. To make an envelope follower, you need to rectify the signal. This causes lots of extra harmonics not present in the original signal. Hence the need for oversampling.
Actually, you probably have. It's called sin(x)/(x) or sinc envelope distortion. Since, when enough samples of a single frequency signal are averaged, a response compensation can be applied to flatten the response characteristic in amplitude and time. Works great for signal analysis, given a large enough time window. Accuracy anywhere near the Nyquist response limit is dependent on these restrictions.
However, for envelope analysis or reasonably accurate audio, there is no such window, no averaging is possible if the proper instantaneous response is to be realized in a PCM system. As I mentioned before, the problem becomes worse near the Nyquist limit where the average error approaches 4 db, that is the average of the samples ranging from near zero amplitude response to full scale amplitude response.
This amplitude variation error, as I have mentioned, is only reduced linearly as the sampled frequency and sampling frequency ratio is increased beyond the Nyquist threshold. This, I believe is a prime contributor to Redbook's crappy HF and soundstage characteristics.
I doubt that anybody can disprove sinc envelope error or deny its application to any format that limits PCM response to Nyquist regardless of interpolation methods or the type of filter (all sharper LP types tend to average in the passband near the corner frequency, independent of passband group delay variations). If anybody believes they have, perhaps they should patent their idea and show the world of DSP engineers doing envelope analysis how they've had it wrong all this time.
Last edited:
Thanks for explaining.
But as jcx has mentioned, this can be compensated for.
See: www.hit.bme.hu/~papay/edu/Conv/pdf/FlatResponse.pdf
And modern converters use oversampling anyway.
Okay, yes, that I'm familiar with. I was the thinking that, perhaps, you were alluding to something akin to modulation envelope distortion in carrier based systems, and which produces distortion spectra. I don't think of sinc envelope frequency contouring - which is due to the sample-and-hold aperture operation of most quantizers - as distortion since it is correctable via EQ. While all digital filter interpolating DACs either digitally EQ the sinc envelope and/or feature an high enough oversampling ratio to narrow the sample-and-hold aperture enough to push the envelope droop beyond the baseband signal spectrum. In addition, NOS DACs are noted by many for their subjective ease and musicality, yet typically leave the sinc envelope droop totally uncorrected.
Last edited:
Hi and interesting. I have a question.
This applies only to the CD format or to higher rez formats as well ?
I guess that also in the old times pros have noticed the limits of the CD format and decided for the 48/16 format instead (i.e. DAT format).
Thanks and regards, gino
I don't believe any of this stuff about envelope distortion. In order for the proposed distortion mechanisms to be audible, the ear would need to be able to hear beyond the Nyquist frequency.
I think worries about envelope distortion are equivalent (by time-frequency duality) to the classic audiophile argument that the ringing in the impulse response of a brickwall digital filter creates "pre-echoes" that interfere with the ear/brain's time of arrival detection. There is no proof that this effect is audible either, though.
I think worries about envelope distortion are equivalent (by time-frequency duality) to the classic audiophile argument that the ringing in the impulse response of a brickwall digital filter creates "pre-echoes" that interfere with the ear/brain's time of arrival detection. There is no proof that this effect is audible either, though.
This has nothing to do with having to be able to hear beyond the Nyquist frequency. The fact is that sinc envelope distortion causes the digital MF & HF audio representation to include arbitrarily time varying clock sampling byproducts with amplitude results that are perfectly audible within the range of human hearing, but particularly at higher frequencies. It is only convenient for analysis using averaging that the sinc envelope distortion has a predictable frequency response characteristic, on average, not as some coverup for marginal digital audio standards.
I have done spreadsheet analysis of equal period sampled PCM that shows that sinc distortion level approximately is inversely proportional to the ratio of sampling rate vs sampled frequency. If somebody has an ax to grind with this, I suggest they go argue with the DSP engineers who insist a ratio of a minimum of 30 is necessary for reasonably accurate envelope analysis and *then* come back here with a little substance to back up their denials if they have managed to prove their point with the DSP guys.
You can consider sinc distortion as being somewhat like worst case terminal analog tape dropout problems but with the advantage that it can be smoothed over for the distortion spectrum world by smearing the error out in time using filter averaging or interpolation.
Sinc distortion can have a particularly disastrous effect on imaging. Imagine a spectral component in the Left channel that is sampled at the lowest sinc amplitude and when propagated to the right channel is sampled at the highest sinc amplitude. Now imagine that happening all throughout the higher end of the audio spectrum in a way that appears audibly to be random. This example shows how it is little wonder that redbook is known for its crap imaging and trashy hi end.
I have done spreadsheet analysis of equal period sampled PCM that shows that sinc distortion level approximately is inversely proportional to the ratio of sampling rate vs sampled frequency. If somebody has an ax to grind with this, I suggest they go argue with the DSP engineers who insist a ratio of a minimum of 30 is necessary for reasonably accurate envelope analysis and *then* come back here with a little substance to back up their denials if they have managed to prove their point with the DSP guys.
You can consider sinc distortion as being somewhat like worst case terminal analog tape dropout problems but with the advantage that it can be smoothed over for the distortion spectrum world by smearing the error out in time using filter averaging or interpolation.
Sinc distortion can have a particularly disastrous effect on imaging. Imagine a spectral component in the Left channel that is sampled at the lowest sinc amplitude and when propagated to the right channel is sampled at the highest sinc amplitude. Now imagine that happening all throughout the higher end of the audio spectrum in a way that appears audibly to be random. This example shows how it is little wonder that redbook is known for its crap imaging and trashy hi end.
Last edited:
Your making a bold claim here.
Can you give us some real proof that you can hear this phenomenon? Real proof would be (ao) an ABX report with the files used for the positive result. And a description on how you could tell the files apart. This so we can try to get a positive result to.
Thanks in advance.
As soon as you provide it, and others can verify your results, I'll be screaming to the world envelope distortion is the devil.
Nothing bold about it. It's rigorously true mathematically.
There's no doubt that sinc envelope distortion is a real world effect that has been consciously ignored by the meter readers since Sony's 'perfect sound forever' advertising campaign.
I am in no way saying that all digital is 'bad' or necessarily even inferior to all analog. I'm saying that redbook CD has always been a marginal standard for high quality audio, just by looking at its transient response.
The train left for high sample rate 24 bit digital audio in the professional field nearly 25 years ago. It's anomalous to me that video based media such blu ray and even some dvd's have higher quality sound tracks than CD is capable of.
There's no doubt that sinc envelope distortion is a real world effect that has been consciously ignored by the meter readers since Sony's 'perfect sound forever' advertising campaign.
I am in no way saying that all digital is 'bad' or necessarily even inferior to all analog. I'm saying that redbook CD has always been a marginal standard for high quality audio, just by looking at its transient response.
The train left for high sample rate 24 bit digital audio in the professional field nearly 25 years ago. It's anomalous to me that video based media such blu ray and even some dvd's have higher quality sound tracks than CD is capable of.
I'm not doubting the maths at all, its a real effect.
I'm doubting the audibility. Its up to you to make me think otherwise.
Maybe I mentioned in this thread or an other, but there's only 1 peer reviewed published paper that has demonstrated an audible difference between redbook and hires formats. They had to use extremely steep filters, with a 500Hz transition band, and ear splitting 120dB spl's to get a barely positive result. Not a really strong case in favour of hires formats as a delivery format.
I suggest you read "Elementary and Basic Aspects of Digital Audio" by Barry Blesser in the first AES conference on digital audio 1982 before carrying these misconceptions further. He actually mentions aperture correction specifically later in his full paper on converters.
Last edited:
ditortion
Ok But reading data sheets about an ultimate dac by
outputs nearest -60bB the THD+NOISE becomes nearest 2%
and for me these means the Fourier components 2nd 3nd etc..
have a significantly distortion or cancellation and is more difficult for human brains the sounds reconstruction .
These problem is more reduced on stylus data sheet than
have low distortion on Fourier components in the same conditions .
Ok But reading data sheets about an ultimate dac by
outputs nearest -60bB the THD+NOISE becomes nearest 2%
and for me these means the Fourier components 2nd 3nd etc..
have a significantly distortion or cancellation and is more difficult for human brains the sounds reconstruction .
These problem is more reduced on stylus data sheet than
have low distortion on Fourier components in the same conditions .
I think the noise would be more significant. Have you looked at distortion spectra at -60dB? Easy to measure for yourself, or if you don't want to bother, here's some data at -90dB on a cheap CD player (figures 2 and 3).
Marantz CD5004 CD player Marantz CD5004 CD player Measurements | Stereophile.com
You really need to provide technical references for what you are claiming, especially the part about the sinc envelope producing abritrarily varying clock byproducts. So, while I'm not a DSP engineer, I'm also not totally unfamilar with the subject. The sinc envelope effectively functions as a frequency transfer mask, shaping the system frequency response, but I'm pretty sure that it does not produce new spectral components. It is no more a distortion mechanism than is an EQ network.
The sinc envelope isn't simply a function of sampling frequency, it is a function of the quantizer hold period. What high oversampling ratios effectively do is to narrow the hold period, which reduces the effect of the sinc envelope on the baseband signal. With that in mind, however, it is easy to see that clock jitter then would modulate the sinc envelope, so perhaps, that was what you were trying to say. But then, the root problem would be clock jitter, not the sinc envelope. I suppose it's conceivable that the sinc envelope could be significantly magnifying the audible effect of the clock jitter. In which case, it may be easier to neutralize the audible effect of the sinc envelope than to further neutralize residual clock jitter.
Except that NOS DACs typically feature the greatest amount of sinc envelope contouring, yet are felt by many to have the most natural sounding imaging, soundstaging and treble. I don't wish to be argumentative. I would love for you to have identified the key to fatigue free digital playback. I just don't see how the sinc envelope could be the problem, aside from the intriguing possibility of audible envelope modulation via clock jitter.
Last edited:
As I just did some reading, it seems that there are multiple solutions to the aperture effect or as thoriated calls it envelope/sync/sinx/x distortion.
1 simple eq
2 use a resampler
3 oversampling
Basically its a non existing problem.
1 simple eq
2 use a resampler
3 oversampling
Basically its a non existing problem.
If someone asked you to do that, exactly what would you do? Just so I don't waste my time. Probably, unless your system is pretty lousy, you're hearing it all the time, IAC.
Sorry. I've clearly stated that none of those will have any effect on sinc envelope error. Think about it for a second. If it were that simple, why haven't DSP engineers done these things? The answer is that they have no effect on correcting the envelope distortion.
- Status
- Not open for further replies.