adamdea
You are not a sound quality evaluation device
Thanks. I am away on holiday for a couple of weeks but will PM you on my return.
"The more I learn, the less I know..."
- Thread starter greg788
- Start date
adamdea
You are not a sound quality evaluation device
Can you perhaps explain the difference between pre-echo and pre- ringing to me. I have to confess that I have never understood it and would like to know.
I can understand that pre-ringing is the inevitable consequence of band limiting a signal using an LP filter. Is pre echo also caused by LP filters? How does it differ?
I was under the impression somewhere that pre echo was a problem with sharp filtering in the midband and was one of the big problems with perceptual codecs.
[I am pretty sure that Werner really understand the difference though. FWIW I would be very surprised if anyone around these parts understands the mathematics involved better than him.]
adamdea
You are not a sound quality evaluation device
Neither- I am very dubious about the part where he claims that the noise modulation in s- d dacs can be easily identified by lots of listeners. I am distrustful in the absence of published professional tests that there is a real widely perceived problem for which this new miracle technology is the cure.
It reminds me of a TV ad for some washing powder or something which was supposed to solve the supposed problem that your newly washed clothes might smell unpleasant when you ironed them. I have never, outside the context of this ad, heard any mention of this "problem". Must admit though that it's not an area in which I claim even enthusiastic amateur interest.
Ciunas Audio
Trade: Ciunas Audio
Pre-ringing = Gibbs Phenomena "The Gibbs phenomenon involves both the fact that Fourier sums overshoot at a jump discontinuity, and that this overshoot does not die out as the frequency increases." http://en.wikipedia.org/wiki/Gibbs_phenomenon
So it is only at jump discontinuities i.e up at high frequencies of the order of 20KHz
Linear phase filters have an unavoidable side-effect = Pre & post echo at all frequencies. Post-echo is not such a problem as it is acoustically masked by the main signal impulse, pre-echo is a problem.
It is mistakenly called pre-ringing!!
http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf
Careful, this is a paper from Cirrus - is it marketing & should be "treated with a pinch of salt"
There are academic papers at the end if you are more inclined! Where are linear filters used extensively? As digital interpolation filters on DAC outputs. So I'm reckoning that these designers either don't believe the pre-echo information as presented in JJ's PPT or they don't know about it.
My point is that the elements I highlighted from JJ's PPT are not well known & already well established knowledge or indeed agreed upon as being the only version of how this works. JJ is careful to state in other PPTs & this one that these are his models & not universally agreed. Indeed, he is a good example of the title of this thread as are most of those who have an expertise & knowledge of a deep nature!
Ciunas Audio
Trade: Ciunas Audio
Ok, so you don't dispute the actual phenomena of noise modulation in S-D DACs, just it's audibility. That's perfectly reasonable!
martin clark
pinko bodger
In what way? Gibbs Phenomenon artifacts only comprises components entirely above the sampling rate cut-off; i.e both pre- and post 'echo' artifacts are inaudible, unless you want to claim useful hearing response above 40Khz.
And it's not an echo - simply an artifact of a sampled system, an image 'complement' of stuff above cut-off.
(PS no contest from me on noise floor modulation, one area definitely worth further study)
Ciunas Audio
Trade: Ciunas Audio
Did you read my post from which you quoted?
martin clark
pinko bodger
Yes. It and the Cirrus link doesn't add anything useful to the Gibbs phenomenon discussion.
There may be reasons to prefer other ways round, apodizing or similar, but I would bet those reasons really come down to non-linearities in the realities of the whole conversion process - not the basic maths underlying the process.
There may be reasons to prefer other ways round, apodizing or similar, but I would bet those reasons really come down to non-linearities in the realities of the whole conversion process - not the basic maths underlying the process.
Ciunas Audio
Trade: Ciunas Audio
It's nothing to do with Gibb's phenomena - it's NOT pre-ringing, it's pre-echo of linear phase filters. I take it that you disagree with JJ Johnston about pre-echo & it's audible effects?
As was levelled at me recently, why discard what is an valid explanation & instead look for a mystical explanation - this one even more ill-defined "non-linearities in the realities of the whole conversion process"?
martin clark
pinko bodger
Aha, sorry , stepping back - yes there can be issues with filters, but that depends on exactly which function is chosen. By no means are all equal. Quite separate from Gibbs -yes. (sorry jumped in on that and then went back a few pages)
And 'non-linearities ' is not meant to be mystical, quite the opposite. Most of what lets all digital audio down is the real-world engineering/cost compromises on the analogue side, including the conversion of D-A; some of which can prove grossly non -linear compared with the underlying (even if 'economical') digital implementation. Often no reason to look further.
And 'non-linearities ' is not meant to be mystical, quite the opposite. Most of what lets all digital audio down is the real-world engineering/cost compromises on the analogue side, including the conversion of D-A; some of which can prove grossly non -linear compared with the underlying (even if 'economical') digital implementation. Often no reason to look further.
Ciunas Audio
Trade: Ciunas Audio
Thanks, yes, OK, equiripple linear phase filters which are commonly used on oversampling DACs, agree?
Yes, I agree, but in the context of what we are talking about, it was a stretching for another explanation rather than the one that was on the table.
An interesting paper worth reading from Julian Dunn, unfortunately 1998, which touches on some of this "Anti-alias and anti-image filtering:
The benefits of 96kHz sampling rate formats for those who
cannot hear above 20kHz." http://www.nanophon.com/audio/antialia.pdf
adamdea
You are not a sound quality evaluation device
I'm grateful for your input on the terminology but I do note that confusingly the cirrus paper you link to refers to "ringing or pre echo". I am very cautious about the consistency of terminology in the digital audio field after eventually discovering that there is no fixed or consistent meaning to the terms upsampling and oversampling .Pre-ringing = Gibbs Phenomena "The Gibbs phenomenon involves both the fact that Fourier sums overshoot at a jump discontinuity, and that this overshoot does not die out as the frequency increases." http://en.wikipedia.org/wiki/Gibbs_phenomenon
So it is only at jump discontinuities i.e up at high frequencies of the order of 20KHz
Linear phase filters have an unavoidable side-effect = Pre & post echo at all frequencies. Post-echo is not such a problem as it is acoustically masked by the main signal impulse, pre-echo is a problem.
It is mistakenly called pre-ringing!!
http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf
Careful, this is a paper from Cirrus - is it marketing & should be "treated with a pinch of salt"
Where are linear filters used extensively? As digital interpolation filters on DAC outputs. So I'm reckoning that these designers either don't believe the pre-echo information as presented in JJ's PPT or they don't know about it.
My point is that the elements I highlighted from JJ's PPT are not well known & already well established knowledge or indeed agreed upon as being the only version of how this works. JJ is careful to state in other PPTs & this one that these are his models & not universally agreed. Indeed, he is a good example of the title of this thread as are most of those who have an expertise & knowledge of a deep nature!
I hope you will find the following helpful in turn. Consonantly with the title of the thread, this is basically what I know I don't know:
if the phenomenon you are referring to is the impulse response characteristic of linear phase filters then you are presumably aware that these filters are used because they are the only ones which maintain phase integrity. In fact the sampling theorem strictly requires nothing less than a sinc function filter which is onviously Lp. It is a logical if perhaps practically useless consequence of the sampling theorem that if you could have a perfectly bandlimited signal and could reconstruct it from its samples using a perfect sinc function filter, there could not be any pre-echo (pre ringing?)
This is really deep water because the bandlimiting filter and reconstruction filter interact. I have with some patient help from experts, managed to achieve fleeting glimpses of fragments of understanding. I know that Werner has actually calculated the convolution of Anti alias and reconstruction filter functions to model the interactions. I have nGever seen an audiophile magazine publish anything nearly that informative on these issues.
Iirc the seminal craven paper on the apodising minimum phase filter pointed out that you only need one in the whole chain. - if there is no embedded ringing (echo? I'm pretty sure craven called it ringing) in the anti alias (pre sampling ) filter then an LP reconstruction filter will be ok. Equally if the anti alias filter was minimum phase, then having an mp reconstruction filter could be a bad idea because of cumulative phase errors and post ringing which whilst supposedly better than pre- ringing, is not wholly benign
As I understand it early digital recording and transfers did indeed use analog anti aliasing filters which were minimum phase or something like it.
Assuming there is an audible problem with LP filters in the 18-22khz
Ish range , that does not necessarily mean that using any other filter is better. The phase errors created by an MP filter might be worse.
In any event I think you should be a little more cautious as I strongly suspect that the pre echo he was referring
to was the effect of steep filters right in the most audible band. These may arise eg in perceptual codecs or drc filters. The phenomenon can be more or less serious depending on the steepness and location of the transition band and the amount of attenuation.
It is AFAIK common ground that at the least an LP filter with a transition band between say 80- 96 kHz could not have audible pre echo (if that is the right word). Iirc Bruno putzeys suggested that 4khz transition band is fine eg 18-22 kHz which will not encroach of the frequency range of most grown up hearers.
There are sound reasons why Digital engineers still use LP filters at least sometimes although it is now fashionable to offer a choice. I don't think its because they are ignorant of the issues, although maybe some people are. I do however find the continued use of the half band lp filter surprising. But that's for another day.
I would be interested to know if anyone else has any views either on what pre echo ( as opposed to pre ringing ) means and what JJ may have been referring to.
Ciunas Audio
Trade: Ciunas Audio
Thanks Adamdea, I'll let others take up the digital audio theme & express their views.
I'm more interested in psychoacoustics as I believe that a thorough understanding & accurate model of hearing is the first step to achieving "realistic" audio playback. For instance, I was interested in JJ's statements on a center channel being necessary for realistic depth cues!
Lest we get complacent & accept everything that is in that JJ presentation as final, there are other researchers with different viewpoints & experimental support. Some interesting papers can be found here http://www.neuroscience-of-music.se/engH.htm
I'm more interested in psychoacoustics as I believe that a thorough understanding & accurate model of hearing is the first step to achieving "realistic" audio playback. For instance, I was interested in JJ's statements on a center channel being necessary for realistic depth cues!
Lest we get complacent & accept everything that is in that JJ presentation as final, there are other researchers with different viewpoints & experimental support. Some interesting papers can be found here http://www.neuroscience-of-music.se/engH.htm
martin clark
pinko bodger
Good link, thanks.
adamdea
You are not a sound quality evaluation device
Thanks I'll have a read.
The third centre channel idea is intersting- i think meridian provide for this in their processors under the name Trifield. Stereo is not necessarily the best system - its just the one we have.
My own view is that it is a tragedy that ambisonics never caught on. If everything was recorded that way we could all have up- down, back- front Surround sound from a recording which would only have 4 channels. We could each have as many speakers as we could manage.
Werner
pfm Member
An echo is an event that is heard separate from its mother event. Ringing is not separated.
Pre-echo is normally not existent, although there was a time in the 80s when digital filters were designed the wrong way, which indeed caused a real pre-echo. Lagadec documented this extensively, Dunn unearthed this in the 90s (at which time it was no longer a problem).
I am almost certain, from the context, that JJ was referring to pre-ringing. If a linear phase filter has its transition frequency in the audible band (that is below 20kHz) and if it transitions so fast as to have an impulse response significantly longer than the response time of the ear's bark band (i.e. band filter) around the transition frequency, then the pre-ringing becomes audible and very objectionably so.
This was a problem the first designers of practical perceptual coders had to address, as such coders invariably rely on filtering the signal in a number of bands, i.e. with a series of transition frequencies all over the audible band.
Pre-ringing of AA and AI filters at 20kHz or higher is much less of an issue. In fact, there is not even proof today that it is a real-world issue.
That is what is in JJ's slides, should one care to read and understand it.
If you know the answer to:
1) what is one of the fundamental problems of two-channel equilateral triangle stereo replay?
2) what is one of the dominant mechanisms for conveying (a hint of) depth in recordings?
Then you'll see that a center channel indeed brings a massive improvement. Hardly news, this, have a look at the first stereo film theater systems and also at RCA and Mercury
recording techniques in the 50s.
sergeauckland
pfm Member
The creation of phantom images from two loudspeakers is the biggest issue with stereo as we have it today. The impression of depth comes from the relative loudness of near and far sounds, but also from the frequency response drop off at higher frequencies the further back the sound source is, due to the differential absorbtion of higher frequencies by the air. Furthermore, in a real sound-field, depth perception is also helped by the diffuse nature of the sound field, whereas at home, in stereo, we are listening in somewhere between the near-field and far field., with different amounts of diffusion than the recording.
Unless the recording has been done naturally, with just a pair of microphones, or better still, a single Soundfield microphone, the chances of there being any real depth to the recording is pretty unlikely.
To me, the best future for home HiFi is in trying to recreate the Soundfield at a recording. However, modern music doesn't lend itsef to that, as it's constructed track-by-track, close miked and pan-potted so if Stereo is ever to get beyond what we have now, it requires a complete change in the way music is made and recorded.
A few years ago, I heard a demonstration at the AES show in Munich of a recreation of a soundfield which involved some 200 separate DSP channels, each driving it's own small loudspeaker. This was set up by the Fraunhoffer Institute using Lawo hardware. It was pretty impressive albeit complete impractical domestically.
Tha's where I think Stereo should be going, trying to recreate the soundfiled at a recording, but I expect it will go the same way as Ambisonics.
S.
adamdea
You are not a sound quality evaluation device
Thanks Werner that's what (rather imprecisely) thought. I have in mind that to be experienced as 2 separate sounds events have to be separated by something like 4 or 6 ms. Does this mean that to produce a pre echo that the impulse response of a filter would have to produce audible ringing 176 samples (44x4) before the impulse?
Ciunas Audio
Trade: Ciunas Audio
But nobody said the experience was two separate sounds - I think the idea of time-smear is important - the slide title is the main context for the slide text.
Edit: OK, having re-read that slide, you are probably correct Werner, it is audio codecs that he is talking about in terms of pre-echo & pre-ringing!
But I believe all this discussion does show that even looking at only the mechanical operation of the ear, there are important areas that are still to be worked out. Thi si snot even taking the feedback & other neural mechanisms into account!
Werner
pfm Member
You may want to read the presentations that deal with a possible model of loudness.