once again though, the writing speaks to a very specific thing which is, "are the internal clocks on most devices good enough?" and the answer of course, is "yes.".
the point of an external clock is to sync multiple digital devices in a studio and here the clock issue is very different.
i think that many home guys think that the reason that they aren't getting good recordings is because of the clock. manufacturers of clocks encourage this thought to sell more clocks and Mr. Lavry is just debunking that idea, because of course, the clock is NOT the problem in 99.9999% of the cases(or less). i don't think he's actually saying that thre's no reason for an external clock.
didn't we have a big discussion about this a year or so ago?
Dan Lavry on jitter and the universe
Lavry, Massenberg, Katz etc. These fellows all have excellent skillsets & knowledge within their field, a field which largely applies to the things that people here do. But just try using Katz's k-weighted scale and writing a D&B or techno tune... or for that matter Eno-esque ambient. The former would be more well accounted for in his discussion of 'heavy metal and punk' mastering techniques, the latter more for classical, but in neither case will his comments all truly apply. That doesn't mean that they don't have useful knowledge to learn from by any means, but at a certain point you have to pick up where they leave off instead of looking for verboten instructions. I'm sure most of us here are well aware of that, being artists & all that (who plays the piece as it's written?) But I thought it worth mentioning just to clarify.
Basically, keep in mind that he's coming more from the perspective of multichannel recording & mixing of more traditional types of music (traditional for our laptop day & age at least). So when you're mixing 32channels of audio that were possibly recorded together, then jitter between 4 different adat DA's is definately going to be more of an issue than when you're just transferring a single channel between your scope card and an RME to process a bit of audio, then send it back to your DAW (all digital round trip). With that in mind some of the specifics of what he's saying makes more sense.
I subscribe to Tape Op btw...
Basically, keep in mind that he's coming more from the perspective of multichannel recording & mixing of more traditional types of music (traditional for our laptop day & age at least). So when you're mixing 32channels of audio that were possibly recorded together, then jitter between 4 different adat DA's is definately going to be more of an issue than when you're just transferring a single channel between your scope card and an RME to process a bit of audio, then send it back to your DAW (all digital round trip). With that in mind some of the specifics of what he's saying makes more sense.
I subscribe to Tape Op btw...
hehe, I guess you don't use the HighTimes VU meter I made for Scope then?valis wrote:But just try using Katz's k-weighted scale and writing a D&B or techno tune...
I know Faxinadu was using it & he makes some of the best electronic music I've heard on this board, but I'm not sure if he's still using it. Btw, you're not trying to meter peaks to K-12 or K-14 scale are you?-
nightscope
- Posts: 686
- Joined: Tue Aug 21, 2007 4:24 pm
- Location: UK
"First, about clock absolute accuracy:Neutron wrote:it must be the crappy quartz clock! get an atomic one! (even though the original recording used a soundblaster as the clock source)
If you view it from the point of view of pitch, clock accuracy of the cheapest crystal is more then good enough, because the ear can not hear 1 cent of pitch deviation, not even a sudden 1 cent pitch. A cent is around 1 part in a hundred between 2 notes on the tempered scale. So you end up with somewhere around 400 parts per million or so yielding a much better accuracy then needed to hear a pitch change.
Therefore, those that try to sell you an atomic clock (such as rubidium or cesium clocks), are going to charge you a lot of money, and not do a thing to improve your audio!
What one may need is for various gears to clock together. Foe example, if you have 2 chassis of AD's and you want to use them simultaneously, you may need to lock them together. If you do not, and they 2 chassis are just slightly off, after say an hour, the difference in times can accumulate to be way too big. So we may need to lock gear together, but that does not make the absolute time is so important.
You would not care if an hour performance was slowed down by say 10msec, but you sure do not want to mix 2 tracks that are time a aligned at the start, but are off by 10msec an hour later...
What IS IMPORTENT is jitter, and that is a "sample by sample" problem. Say one sample is early (relative to where it should be) by 1 nsec. By the time it happened, it is too late to do anything about it. In fact, at about 8 inches of wire (or trace a way) you are already 1nsec "behind the time". And if you try to control jitter for too far - send a clean clock a long distance, you will pick up a lot more jitter in that transmission path, then you started with...
Regards
Dan Lavry"
ns
true, so don't send your clock signals any further than necessary.
it's simple, don't make the system more complicated than absolutely needed. this is always good advice, just like not throwing lots of plugins on every track, just because you can(that's no way to get a better sounding mix!).
the point is simple and well taken that the internal clocks are good, often a better choice because of physical limitations of clock transmission. however, if one has a studio full of devices that need to be clocked together...
it's simple, don't make the system more complicated than absolutely needed. this is always good advice, just like not throwing lots of plugins on every track, just because you can(that's no way to get a better sounding mix!).
the point is simple and well taken that the internal clocks are good, often a better choice because of physical limitations of clock transmission. however, if one has a studio full of devices that need to be clocked together...
I certainly wasn't saying that music would suck if it was metered in Katz's systems.<~Shroomz~> wrote:hehe, I guess you don't use the HighTimes VU meter I made for Scope then?valis wrote:But just try using Katz's k-weighted scale and writing a D&B or techno tune...
Most of the Drum&Bass tunes I have for DJ use could be used to establish a K-10 or K-8 scale (probably K-10 as that leaves room for the -8rms bunch to "push it" a bit) as they tend to be very flat & saturated. Meaning if you were to take the genre as a whole and establish the baseline from it, rather than establishing a baseline and pushing the genre towards that. Even D&B producers are aware of the whole loudness discussion though, so don't mistake my meaning here.
Katz was attempting to establish customs for musical styles that were relevant to the circle of people he works with (and with good reason). The parallel between that & the jitter discussion from Lavry is that Lavry considers jitter to be an issue when syncing multiple digital devices, especially during the AD stage. What he didn't state outright is that most of his crowd (and most of the Tape-op crowd) are recording multichannel performances where the jitter would not only interfere with the reproduction of a single channel of audio but cause some real nastiness across digital devices that are constantly swimming into & out of phase with each other on the sample level. Ie, when a good external clock actually matters (in his discussions).
Btw, the downtempo & electronica that I tend to enjoy listening to when I'm on my own time fits quite fine into a K-12 calibrated meter.
that's right.valis wrote: What he didn't state outright is that most of his crowd (and most of the Tape-op crowd) are recording multichannel performances where the jitter would not only interfere with the reproduction of a single channel of audio but cause some real nastiness across digital devices that are constantly swimming into & out of phase with each other on the sample level. Ie, when a good external clock actually matters (in his discussions).
sure it would be...at0m wrote:Good idea. Or would that be too confronting, Tom?nightscope wrote:...Well, you could always ask him. Then you'd find out. He's answered everyone's questions in great detail so far. Here's the link:-
http://www.cockos.com/forum/showthread.php?t=19675
ns:]
how can I confront to something that isn't even mentioned ?
I never questioned his statements (external versus internal) quoted above - in fact they are pretty obvious
but as I lack the required measurement gear for hands-on verification, I have to rely a bit on the logic chain...
imho it's reasonable to assume not all clocks (both versions) are created equal.
And as expensive gear usually comes with clock specs, while inexpensive stuff doesn't, I humbly assume there may be something worth hiding...
That 'something' is a candidate for improvement, which is equally banal.
Yet the topic isn't about the 'best (or most effective) clocking device ever' but about 'sound quality influenced by internal /external clocking'.
If I was to change anything about my system in the clock domain, I'd rather have it 're-clocked' by an RME card, as their whitepapers convinced me and it's an affordable solution.
cheers, Tom
-
nightscope
- Posts: 686
- Joined: Tue Aug 21, 2007 4:24 pm
- Location: UK
More from Dan @ Reaper forum.
Q = So would 2 ADs running together for a shorter period of time (say 5-8 mins) encounter this same timing issue or would it be to small of a discrepancy to detect.
A = "Generally speaking, when you have say multiple AD's in one chassis, you operate all channels with internal clock and that is optimal, no time matching issues...
But if you have separate chassis, it is wise to have them clocked together.
To answer you question, one has to know a few facts and have a "calculator at hand".
Say you have 2 AD's. Each one has an internal clock of say +/-50ppm accuracy (ppm is parts per million). Then the combined worse case time drift is when one is at +50ppm and the other at -50ppm, a total of 100ppm (you do not want to take a chance and assume anything other then worst case, you may get lucky but you may not).
So lets take say 5 minutes and see what 100ppm will do:
100ppm is .0001
5 minutes is 300 seconds
.0001*300 = .03 seconds
That is a lot of delay accumulated over only 5 minutes. You will of course want to have much better clock accuracy, and in fact it gets pretty hopeless very quickly. Say you have 2 clocks with 1ppm each (2ppm combined), and you calculate the time mismatch for an hour. You end up with 7.2msec.
At 2ppm for 5 minutes you get 600usec. Of course, most crystals will not yield 1ppm stability over 1 degree of temperature change.
I said that you do not need to fall for the “atomic crock” (I mean atomic clock), because you do not need an overall system accuracy of a few nsec per year (nsec is nano second which is a billionth of a second). But there are also a couple of other reasons why you need to synchronize multiple chassis of AD’s.
First, the data receiver end may need to have the same rate on all channels (that is not always the case).
Second, in some cases (when some of the sound is shared between channels), you want to have some control over the relative timing between tracks. If each of the channels is completely separated from the other (each mic is pointing at a different source), then there is no issue. But if say 2 mics pick up some of the same sound, then, when you mix (add) the channels, you want the tracks to have the same delay. If you do not, you will have a “comb filter effect” – an unwanted “EQ” of peaks and deeps, on the portion of the music that is “shared” by the mics. If the tracks “slide” over time, you have a “variable comb filter effect which is much worse. And for comb filtering, even 10usec can start taking a toll.
Say you have a 10KHz tone – thus the cycle time is 100usec. If you add 2 channels with the same tone, you will double the amplitude, But say you delay one channel by 50usec (half cycle). When you add you get total cancelation instead of double the amplitude…
It does not take long to accumulate 100usec. Then when the drift gets to be 200usec you get "double amplitude". Another 100usec drift, and you have full cancelation....
So if you get say 30msec in 5 minutes (the above 100ppm example), your "double then cancel effect" happens once per second for 10KHz, twice per second for 20Khz...
Regards
Dan Lavry"
ns
Q = So would 2 ADs running together for a shorter period of time (say 5-8 mins) encounter this same timing issue or would it be to small of a discrepancy to detect.
A = "Generally speaking, when you have say multiple AD's in one chassis, you operate all channels with internal clock and that is optimal, no time matching issues...
But if you have separate chassis, it is wise to have them clocked together.
To answer you question, one has to know a few facts and have a "calculator at hand".
Say you have 2 AD's. Each one has an internal clock of say +/-50ppm accuracy (ppm is parts per million). Then the combined worse case time drift is when one is at +50ppm and the other at -50ppm, a total of 100ppm (you do not want to take a chance and assume anything other then worst case, you may get lucky but you may not).
So lets take say 5 minutes and see what 100ppm will do:
100ppm is .0001
5 minutes is 300 seconds
.0001*300 = .03 seconds
That is a lot of delay accumulated over only 5 minutes. You will of course want to have much better clock accuracy, and in fact it gets pretty hopeless very quickly. Say you have 2 clocks with 1ppm each (2ppm combined), and you calculate the time mismatch for an hour. You end up with 7.2msec.
At 2ppm for 5 minutes you get 600usec. Of course, most crystals will not yield 1ppm stability over 1 degree of temperature change.
I said that you do not need to fall for the “atomic crock” (I mean atomic clock), because you do not need an overall system accuracy of a few nsec per year (nsec is nano second which is a billionth of a second). But there are also a couple of other reasons why you need to synchronize multiple chassis of AD’s.
First, the data receiver end may need to have the same rate on all channels (that is not always the case).
Second, in some cases (when some of the sound is shared between channels), you want to have some control over the relative timing between tracks. If each of the channels is completely separated from the other (each mic is pointing at a different source), then there is no issue. But if say 2 mics pick up some of the same sound, then, when you mix (add) the channels, you want the tracks to have the same delay. If you do not, you will have a “comb filter effect” – an unwanted “EQ” of peaks and deeps, on the portion of the music that is “shared” by the mics. If the tracks “slide” over time, you have a “variable comb filter effect which is much worse. And for comb filtering, even 10usec can start taking a toll.
Say you have a 10KHz tone – thus the cycle time is 100usec. If you add 2 channels with the same tone, you will double the amplitude, But say you delay one channel by 50usec (half cycle). When you add you get total cancelation instead of double the amplitude…
It does not take long to accumulate 100usec. Then when the drift gets to be 200usec you get "double amplitude". Another 100usec drift, and you have full cancelation....
So if you get say 30msec in 5 minutes (the above 100ppm example), your "double then cancel effect" happens once per second for 10KHz, twice per second for 20Khz...
Regards
Dan Lavry"
ns
he assumes a plus/minus discrepancy, but only adds the errors...nightscope wrote:More from Dan @ Reaper forum.
Q = So would 2 ADs running together for a shorter period of time (say 5-8 mins) encounter this same timing issue or would it be to small of a discrepancy to detect.
...
Say you have 2 AD's. Each one has an internal clock of say +/-50ppm accuracy (ppm is parts per million). Then the combined worse case time drift is when one is at +50ppm and the other at -50ppm, a total of 100ppm (you do not want to take a chance and assume anything other then worst case, you may get lucky but you may not).
So lets take say 5 minutes and see what 100ppm will do:
100ppm is .0001
5 minutes is 300 seconds
.0001*300 = .03 seconds
That is a lot of delay accumulated over only 5 minutes...
if the clock drift is statistically even distributed, the balance would be close to zero...
an accumulated error will only show up if the drift follows a pattern, but under any circumstances the +/- nature of the events will provide some leveling.
Only trust a statistic you faked yourself
cheers, Tom