AudioCircle
Audio/Video Gear and Systems => The Lab => Topic started by: art on 30 Mar 2019, 03:47 pm
-
Introduction
Some of you may be aware that after a 3-year hiatus, we are starting to build equipment again. I could write a tome on what went on in those years, and why it took so long, and why it was a mess (a lot of which was out of my, and seemingly anyone's control), but that isn't the point. The point is the only stuff that was not boxed up and stored away was all of our gear to measure phase noise. Not that I needed any practice, as this is what I did, back when I had a "real job". So, this is not some new fascination with me. Since I did have an extensive background in this stuff, not to mention we now have gear that doesn't take up an entire 6' rack*, it was a way to keep the doors open. All we needed was a small testing gizmo, an external frequency reference, some test jigs, and something to measure. Which I did.
During that time, I have measured hundred of these teeny-tiny little clock parts. We sold tested and sorted ones to various manufacturers. We also worked with some of the largest companies around, that make "frequency control products", to help them refine products they intended to bring to market. We tested lots of samples, that other "high-end" companies (and some Pro Audio ones as well) were interested in using in their products.
It is important to understand some of these obscure matters before we get started. There are also some technical matters that need some background, and that will be the next post. So, let's get started!
* = Ironically, even though HP/Agilent/whatevertheyaretoday makes a phase noise gizmo that can sit on a desk, it still costs around $90k! Just like it did back in the 70s and 80s, when it took up an entire rack. How can that be? Lucky for us, there are affordable knock-offs. Limited in what they can do, but still very powerful gizmos.
-
OK, so what is jitter?
Well, there are 2 ways of looking at it. One is in the time domain, and the other is in the frequency domain.
To some of us, it seems that folks who do "digital" design only look at the time-domain element. Meaning how much a clock signal causes the data to be shifted, back and forth, relative to when it is supposed to happen. In turn, they might accuse those of us who look at the frequency domain don't understand what jitter is all about.
Let me answer that with this anecdote..................
At my last "real job", there was an infamous incident where I threw a VP of our company out of my office, for snidely asserting that I needed to get some education on how digital stuff worked. All that you need to know is that I was not fired, and he never got in my face again. (Yes, actual event. But, it should not surprise anyone I had a history of taking on upper management, and never lost my job. Hard to fire someone who got results. Results that paid their salary.)
Ok, now that you know where I stand, let's start looking at how jitter is measured.
First, I would like "you guys" to read this link:
http://www.rakon.com/component/docman/doc_download/225-phase-noise-and-jitter-in-crystal-oscillators?Itemid= (http://www.rakon.com/component/docman/doc_download/225-phase-noise-and-jitter-in-crystal-oscillators?Itemid=)
Once you read it, some of what I am going to bring up will make some sense. You will not have to memorize all of it. Just get a rough overview of what they are explaining.
Such as:
This can be analysed by matching the slopes of the real Phase Noise plot to those
shown in the idealised Phase Noise plot of Fig. 9.
For this particular plot:-
Flicker corner of Buffer Stage ~ 5 kHz
Loaded Q of the Crystal ~ 170 Hz (Q ~ 38k)
Flicker corner of Oscillator Transistor ~ 12 Hz
Random Walk ~ 0.1 Hz (extrapolated)
Two things to take from this:
1.) Phase noise ("jitter" measured in the frequency domain) has lots of components. It isn't some number some gizmo spits out. Numbers, without context and understanding are just numbers. And, when it comes to jitter, pretty much meaningless, without any context.
In the case of "jitter", the context has to be the frequency range over which the jitter is measured. And there is the problem. Because what that range is must be defined before you know whether or not something is "low jitter" or not. (More on this in a bit.)
2.) See that odd reference...................."random walk ~ 0.1 Hz (extrapolated)"?
Let's talk about that, because we are going to see a lot of that, in the plots I am going to post.
Let me just interrupt things to say if you are going to come here to argue and tell me nothing below 20 Hz or 1 Hz or whatever Hz is audible, and I don't know what I am talking about and this is all a waste of time, etc., etc., and more etc., then just stay out of the discussion. I have heard all of that crap so many times I can not stand to hear any more of it. So, do us all a favor and go watch "March Madness", and not make a mess of this. (Asking questions is ok, because you are all going to have questions, and if I was not willing to field questions you would not be seeing any of this. But stirring up crap, because it exposes your ignorance is not my problem, is not going to anything I will address.)
So, assuming "you guys" have looked at that plot, you will notice the phase noise plot stops at 10 Hz. Why? Well, for the obvious reason! It is easier, faster (and therefore cheaper) to not go below 10 Hz. Not because what goes on down below it doesn't matter. It is because it is a pain.
Now, if you have lots of time on your hands (because everything else is boxed up), so what if it takes a lot of time? Besides, all of those years spent previously measuring this stuff, you know how it important it really is.
Even the folks who do not measure below 10 Hz know it is important. A certain company that makes "frequency control products" worked with us so that they could find a correlation between their 10 Hz measurements and our 1 Hz measurements. It just takes too long for them to do it, but I can assure you that they are very interested. Granted, 99.9999% of their customers won't care. But, they wanted to make some inroads into making clocks that would be really good for digital audio. Which is a balancing act. Cell phones are driving that market. So, if it does sound better, there will be a market for it. But, it has to be as cheap as the products that stink. Which is tough to do. (In this case, the plan was to find a way to tweak their existing process, to make a specialty product, with higher price, obviously, that would appeal to our segment of the industry. Turns out there is not just one knob to turn, to crank out those better parts. Which is why it became an on-and-off-again endeavor.)
So...............if you don't measure below 10 Hz, then how do you know the "random walk" kicks in around 0.1 Hz?
Experience.
Or so I thought.........................
We'll see this in the plots. But, before that...................more background stuff.
-
OK, more background........................
The guys who make "frequency control products" all have the same way to publish their "jitter" specs.
Jitter frequency measured with f>1 kHz.
What does this mean?
Well, they only measure jitter that exists from 1 kHz and up.
Why?
Because 1.) it is easy and cheap to do, and 2.) it makes for a very low number, and therefore the weenies in the purchasing department (not to mention brain-dead engineering management, who only go into management since they were crappy engineers) all think they are buying a quality product. And if the people that work for them are just as ignorant...................
You get the idea.
But, what is in that f>1 kHz jitter?
Look at the phase noise plots, in that link I suggested that you all read. (You did look at it right? It isn't that long.) What is it?
[Thermal noise (Johnson noise)
i.e. kT buffer amplifier noise, resistor noise and Shott noise.
IOW, the noise floor. Has nothing to do with the crystal.
"So, if it is independent of the crystal, why do they even spec it?"
See above.
You will note that the really good parts will have a table of the actual phase noise. Most do not go below 1 kHz. Some will go down to 10 Hz. You have to look very hard for one that goes down to 1 Hz, but they do exist.
So, here is the latest spec sheet, for a part that we use. I have to note that this is the latest and greatest, right on their site. I say this because for several years there was one that showed a ridiculously low phase noise plot.
Well...................sort of............
Sort of because some will actually do those ridiculously low phase noise numbers. And folks believed they all would! (HA!) The truth is they are really more like what they are now showing on their site.
https://www.ndk.com/en/ad/2013/001/index.html (https://www.ndk.com/en/ad/2013/001/index.html)
(The one we are talking about is the SPXO.) (Yes, the other one is pretty darn good. And look at how big the package is, and all the mess it will be to mount one in a product!)
Let's see where was I..........................uh.......... ........oh, yeah, the "published" vs "real world" specs.
So, in the case of this one part (that we sell sorted versions to folks who can afford them), there is a certain amount of them that will really do what the supposedly real spec sheet says. (Wish I could find a copy of it. Well, take my word for it. It does exist. I may have copy. And no, most do not even come close.)
Now the ones that do not come close.......................if one were to only measure the phase noise (or "jitter" above 1 kHz, they would all measure fantastic! But, a lot don't. We have hundreds of them that are not good enough. (We may sell them, some day, in lots of 25 or 50 or whatever, for next to nothing on ebay. Maybe. Maybe not.)
OK, time to split this up, and talk about the jitter number problem..............
-
OK, the jitter number problem....................
As you should have garnered by now that the are lots of components in the jitter. And this is just from the clock! Yes, there are other sources, mostly crap on the supply voltages, from all the dividers, multiplexers, etc., etc., etc. And, in the case of good ol' SPDIF, you have all the sub-code crap.
"Why do people still use that crap? What good is it?"
You can transmit it longer than 12', which is what USB can do.
"Yeah, but now you can do LAN or wireless."
Yes, but none of that is the point. The point is you have essentially 2 sources of jitter: the clock and all the other crap that creates noise on the power and/or ground lines. I only mention SPDIF because you can see all the crap it creates, very easily, in the "modern" methods of measuring "jitter". And being able to point out the crap is essentially the point of "modern" jitter numbers.
If you are willing to accept that jitter is some number, that is undefined, and comes from some computer-controlled gizmo that reads what goes into a sound card, and spits out some plot that has a magical number attached to it.
Not saying that number, as ill-defined as it is, doesn't contain useful information. It does. It just presents a very incomplete picture of what the jitter really is.
Because it can not measure all that "close-in" jitter. You know, the phase noise that is less than 10 Hz in frequency offset. It wasn't designed to do that, and it can not do that. The only way to do that is to actually measure the clock itself.
Look, if you want to believe that the clock is not important, and only the internally-generated crap is relevant, well it is a free country. But, if you don't want to know what goes on below 10 Hz or so, because "it can't be audible", then why measure anything. (I'm guessing because you have found a way to make your "jitter number" look better than the next guy's "jitter number" and that is all that you really care about.)
Fine. It is a free country. We are glad that 99.99% of the industry doesn't believe us, because there is no way we could supply parts to them all. (Although that would be a really interesting problem to have.)
So, for the rest of you........................on to the fun and games! Background complete.
-
So, here is a phase noise plot of some gizmo we are starting to build. Which maybe no will will ever buy, but so what?
(http://ar-t.co/data/New%20AR-T%20pn%202.png)
"What the.................................? What is all that crap? I can not make any sense out of it. And what does 'DIY' supposed to mean? I'm confused."
So, was I, when I first powered this thing up.
Remember that part of "random walk ~ 0.1 Hz"? You know, where it usually is, 99.9999% of the time. So, why the ^$^&$+Y#YU#***#^$ is it around 5 Hz?
I have no idea.
"I thought that you said that you have measured hundreds of these things. Are you are just now finding out they stink?"
I have. And a lot do stink. But all were measured in a test fixture, in a controlled environment.
But what the..................and what does the "DIY bit mean?
Simple. It was measured the way a lot of "DIYers' might build something. You know, they build it, and listen to it. And that is that.
"OK...............and manufacturers do something differently? I'm so confused."
(Don't worry about that last part, ok.)
Well, maybe that isn't 100% true. There are manufacturers that make gizmos that are part of something else, which might be a computer or built into something else or will be used with something that is used with something.........................iow, not a big fancy metal box that looks all shiny that you can show off to your non-audiophile buddies (so they can wonder what it is with your obsession) but instead is just something sitting there. Calling that "DIY style.
(At this point, a thought comes to mind. We have sold these clock parts to some places that make these kind of computer thingie gizmos. Yet, they never re-order. I think I now know why....................................)
So, at this point, you ask how do I know it is "random walk"?
Uh, because I know what random walk looks like, even if it takes place a 5 Hz instead of 0.1 Hz, like it should. (Why? I would be lying if I had the slightest idea. Ok, maybe I really am an idiot!)
Or, because I fixed it, the way I normally fix it. So, at that point we go from the top plot (oh, good grief, that is horrible!), to the second one, which looks pretty good.
"But what does any of this matter to any of us? It is just some plot. You know what it means, but so what? Does it have any real significance to anything we care about?"
Yes, of course it does! If it didn't none of us would be here.
Now to the even better stuff.....................(I promise)
So, a lot of you should be wondering does any of this tell us anything about how it will sound?
YES!
OK, forget about all the crap above 1 kHz. Not only because I have explained how the noise floor is not important, but because this is a "budget" product, there is no test point to directly measure the clock. I have to measure the output, and let the s/w manipulate everything, to know what the clock is doing. (IOW, all that exists above 1 kHz is pure garbage. In fact, there is a TON of SPDIF crap above 10 Hz, as well. The s/w has a way of ignoring that, if you tell it do that.)
Well, since all of them have good phase noise, down around 10 Hz, I can safely say they will have a clean top end and tight bass. IOW, a lot of the "digititis" will be gone. But, it still won't sound right. (How exactly, I can not say, as I have never seen random walk that high. Let's skip over that, for now, ok?)
So, if you look at the bottom plot, the one that is what the finished product will look (and sound) like, you will see there is no significant amount of random walk. You will recall, from post #1 that random walk increases at 40 dB/decade and the f^-3 noise increases at 30 dB/decade, we are still in that part of the curve.
Which will lead me to state, with 100% certainty...........
Once you get rid of all of that random walk crap, this unit will sound as good as any "analog" system. IOW, not only will the bass be tight, the top end will be clean, and free of grit and glare, but this unit will actually produce a "3-D sound stage".
How do I know this? Lots of testing clocks, and just as much listening to various ones, and making notes, to correlate measured phase noise to sound perception.
Since we consider this research to be proprietary, I am not going into details how we did it. Accept it or reject it, your choice.
Same thing goes for how we fix the random walk problem. Sorry, guys.
Seriously, if I told you how we fix it, this would turn into the biggest food fight on any forum, possibly ever. If you think the stuff we have said in the past is controversial, well this one takes the cake.
Most would not believe us, and therefore call us liars. Some would try it, and it wouldn't do diddly-squat, because unless you have clock that has a phase noise number, below a certain level, well the easiest way to explain it is the inherent noise will swamp any random walk noise. you simply must have good clock in order for the elimination of the random walk to matter.
Anyway, maybe we can at least all agree if you only measure what goes on above 1 kHz..................
(Do any of you guys honestly think all 3 versions sound alike? If you do, well............not much to talk about, right?)
Next..................another new crappy thing we are building, and does it do the same stupid thing!
Stay tuned. Or not.
-
Oh, I forgot something!
This may give you guys some clues. Or not!
Here is what the frequency difference looks like, for the 3 versions. Notice which one looks the most stable?
(Don't ask what "linear residual" is. This isn't a class on regression theory. Not that I would even understand it............................)
(http://ar-t.co/data/New%20AR-T%20fd.png)
"I thought that you said drift doesn't matter, last week or so, in some other thread?"
Well, yes, but this is not necessarily "drift".
Remember the old days of "wow" and "flutter". Kinda sorta the same thing. Only different.
Different in the range that they exist under. Same thing here. This is too fast to be considered drift. Trust me, ok?
Enjoy!
Back in a while....................
-
One more thing that I forgot..................(don't ever get old)................
You will notice in the second plot it states that the jitter was measured between 1 Hz and 1 kHz. Can't measure above 1 kHz, in this example, as previously stated. In the next set.......................we are going to be brave and measure all the way out to 100 kHz! Hotcha.
-
Ok, as promised..................some other POC that we make. Since this is a bit more expensive, it has a test spigot, so I can measure the clock directly. And, since some folks are really concerned about what goes on at 100 kHz offset and how the jitter numbers are different, etc., we are going to go out that far.
(http://ar-t.co/data/AR-T%20product%20pn.png)
Well. Several things are obvious. I think.
Going out to 100 kHz means little, if anything.
These clock parts are not happy, without "the fix".
And.....................wait a second..................what happened to the simple fix, and what is this "extra fix"?
Well, I reversed the order that they were applied in.
"Uh, I am so confused now..................I have no idea what you are talking about."
Don't worry...............a lot of folks are convinced none of us here know what we are talking about!
So, here is what happened, and for a reason.
I actually measured this one first. And I thought it was broken! (Could be..............new PCB design, untested..............who knows, right?) So, I did what would essentially be the final version, for the right way. Then, I added "the extra". As you can see, the "extra" really doesn't help any.
But......................
if you can not do it the "right way" and all that you can do is the "simple fix", then these parts really need it. But, it still needs help, so just do it the right way, unless there is no way to do it.
No more clues! (Go ahead..............hate us................we're used to it.)
OK..............and just to show that you just need to do it the "right way", here is that weird frequency difference plot:
(http://ar-t.co/data/AR-T%20product%20fd.png)
See, it works the right way, and doesn't need any help.
But just in case someone goes mucking around the inside of one of these................the extra, just in case!
Ok, all for now. Coming up (in a day or so).................what happens if you measure one of these at 22 MHz, and let the s/w divide it down to 11 MHz, or you one of those nasty flip-flops to do the divide by 2 in the test fixture. I know....................you are all on pins and needles.
"No we're not!"
Don't blame you. My brain hurts. All for now. Enjoy!
-
So how would this technology apply to audio equipment (I assume a DAC)? For example I have an iFi iDSD Pro which has a plug for accepting an external clock. Would something like this just plug into that, or would more extensive surgery be required?
-
Uhm.................this gizmo (both of them, actually) are USB-SPDIF converter thingies.
So, no.
As to what the gizmo you have, and what kind of external clock it would take................I have ZERO idea. Not my realm.
-
OK, looks like it is designed to take an external 10 MHz clock. You can buy a really good one, for around $30, on ebay. But, then you have to put it into a box, supply it with 12 V @ 1 A or so, and connect it to a BNC jack, so that you can connect it to the iFi gizmo.
-
OK, thanks.
-
Actually..............now that I looked on ebay...................the days when 25 guys were selling them, for $30 are now the days of 5 guys selling them for $50.
-
OK, in yet another example of "why getting old sucks", it turns out I already did make a plot of what happens when you divide by 2 in the s/w, or the test fixture (h/w).
(http://ar-t.co/data/divide%20by%20example.png)
OK, should be pretty easy to understand. In the top plot, I run the 22 MHz signal into a D f-f, to get it to 11 MHz, because I know what a good clock looks like. In the bottom one, the 22 MHz goes right into the phase noise testing gizmo, and I tell it to divide by 2. From that, we can guesstimate how much a crappy little picogate adds.
The noise at 200 Hz is around 3 db higher. Yawn................
(Everything else is the same.)
OK, in all fairness, if the reference oscillator had a noise floor of -160 dBc (which good ones do), then you would see more detail. But this reference was chosen for its 1 Hz offset number (<-102 dBc), and not its noise floor. So, can't go below -150 dBc, but no one here is having a fit over that minor detail.
(BTW...............yes, it is one of those ebay specials that I have mentioned above. Darn good part, for that little amount of money.)
-
Here is a fair demonstration of what "random walk" can do to an expensive ($100 or so, if they would only sell you one) double-oven SC-cut crystal oscillators. IOW, about the best you can get. The reason I say "fair" is that for it to present a clearer picture it would have needed to be run over a longer time span, to average out the noise way down in the sub-sonic region. Also, looking at all of the data, it would have better better if we used a different supply. Probably, on the day these were taken, I used whatever supply I could grab that would work.
And, to be honest, I didn't think these would suffer the random walk problem. (Yeah, I should have known better.........................)
The top unit is one of those ebay specials I mentioned above. The next 2 are samples that someone managed to get, for 48 kHz clocks. (256x and 512x) I used the s/w to compare them to the 10 MHz reference, to make it easier to see how they stacked up.
(http://ar-t.co/data/morion%20x3%20pn.png)
As you can see, the break point is around 0.3 Hz, which is not far from where it should be. And you can see (not as clearly as I hoped) that there is a 10 dB difference, one decade lower than the break point. (Remember, the curve goes from 30 dB/dec to 40 dB/dec when the random walk kicks in.)
OK, so what is the point?
1.) Even really, really good, and really, really expensive parts can be affected by this. Luckily, it is easy to mitigate. Whether an individual clock is good enough to merit the effort is another story. To be honest, when I run across a sample that shows no promise, I shut the test down. So, I am sure even the crappiest parts will show the effect. Assuming its intrinsic noise is not so high that you will never see it. And even if you do see it and can fix it, the clock will sound so bad that you will never notice the difference.
2.) All clocks do not sound alike. This should surprise no one. There are lots of factors that go into that. If they all sounded the same, then why would one company put "HEAR THE DIFFERENCE!" in their spec sheet? Because you can hear the difference. (Whether or not those parts actually measure what they claim they do.................well, I better keep my mouth shut.)
(As an aside, yes, I have taken some of those apart. Don't ask why. I'll just say none of my "fixes" seem to help. But, as a result of the tests made on the 2 parts shown in this thread..................after that point we measured ALL test samples in a homemade environmental chamber. So, should not surprise anyone the "fixes" did not help those $30 parts, as we only tested them in a tightly controlled environment.)
3.) What goes on in the sub-sonic region DOES matter. This what I have a hard time convincing people. Folks want to believe in order to "hear the jitter" that you have to actually be able to "hear the jitter". IOW "what does jitter sound like and how will I know I am hearing it and if it is below 20 Hz then I can't hear it and you are nuts."
You don't hear the jitter! You hear its effects.
Which are 1.) nasty, bright, hard, brittle, rough sounding high end and 2.) loose, flabby, tubby, poor bass.
Or, to sum it up: "digititis". Get a good clock (and by good...............well, look at the clocks in the crap we make....................that gives you an idea what a "good clock" for audio is.
"And just how do YOU know this, and no one else does?"
Well, because we have more spare time on our hands, to not only measure stuff to death (and more importantly, the gear to do it), and because of our background, we knew how important it was.
let me digress a bit..................
Anyone remember some gizmo, that went on some old Philips-based CDPs called the Simply Physics IsoDrive? Anyone remember?
Remember what it did?
It tightened the bass.
It cleaned up the highs.
No one knew how. Even the guy who did it. (He would never admit it, all he would admit was that he felt there had to be some sort of improvement in making the platter rotation more stable. And he was right. He made it more stable, in terms of rotational stability. IOW, he took out some of the "wow and flutter" in the platter. Which means he lowered any modulation caused by minute changes in the rotational stability.
Which meant....................yeah, you got it...............he lowered the jitter.
(At this point, someone will conjure up the image of the character "Miller" from "Repo Man". Except we are not talking about UFOs and time travel. it just seems like it!)
So, back to the subject......................yes, none of us can hear what goes on below 1 Hz, but that does not mean that we can not hear the effects of jitter, that low in frequency.
We know, for a fact, that you can. We have demonstrated in many times. (In fact, we went through a 6-month phase trying to figure out why things sounded one way one day, and something else the next. Yes, we blamed everything but.................our "friend", sub-sonic jitter.
Seriously, once you get that down (to a certain "magical" number), the soundstage becomes what audiophiles expect soundstage to be. But, you first have to get all of the other "jitter" down to a certain level before any of that comes into play.
Trust me. We put a lot of effort into it. Believe us or not. Your choice.
Our 'net connection is doing its daily "you really don't want to be online, do you" routine. This may have some typos, and I will check for that later. Posting before it get lost in the ether.
-
Pat, first it's good to see you posting again!
Second, I haven't had time to digest everything here yet, but I've started and so just wanted to let you know that I appreciate the amount of effort that goes into just writing up a discussion like this, much less the work behind it.
Thank you!
(Now can you work this into a USB to I2S converter? :) Yeah, I know... not intended for external connections, multiple conventions but no standard for connector type/pin-out, etc..)
-
Art's website is:
https://ar-t.co/PRODUCTS.html
gab
-
(Now can you work this into a USB to I2S converter? :) Yeah, I know... not intended for external connections, multiple conventions but no standard for connector type/pin-out, etc..)
There is another forum that I post on (that has NOTHING to do with audio, which is mainly why I post there) that has an odd practice called "blue font". I am one of its leading abusers!
Blue font is for sarcasm.................
So, imagine this in blue................
You could buy what we already make, and tap into the I2S output of the USB controller chip.
Send me a PM about this, ok?
-
OK, here is an interesting group of plots.
We make it our policy to not share data from tests that were performed, for profit, from one of our customers. If we shared everything we did, for free, then why would folks pay us? (A lot of folks don't understand why ANYONE pays us money, but, hey, if you need to know, and you don't have the gear, or experience to do it, what else are you going to do?)
In this case, I can make an exception.
1.) I honestly forget who we did this for, as it was not anyone we had worked with, and never worked with again. (They were probably depressed by the results. Can't blame them: they had us stop before we finished all 10 samples.)
2.) The outfit that made these parts is out of business, so I can present them as an illustration, and not as something you can run out and buy.
(http://ar-t.co/data/tellurian%20pn%20x6%201%20Hz.png)
The first thing you might notice is that there is no #5. There is a reason for that. You can probably guess!
It was so bad......................if you have no way to measure this, how do you know if you get stuck with one of those. Since we do not make DACs, we do not measure what comes out of a DAC. Whether it would show up or not, well, can't honestly answer that one. Let's just say it will sound like crap, but these guys were a large enough company that they had no way of listening to every unit that came off the line.
But..............
Look at #7!!!!!!!!!!!!!!!!!!!
That one is so good, that it borders on the level of performance of an SC-cut crystal. Which can cost close to $100. And this is from a generic-looking HC-49 part.
Of the 6 units shown here, there is around 25 dB difference. 20 dB if you drop the really good one. With a spread that large, I would have a hard time advising a customer to use them. But, once you add in the dog, well, no way!
Don't know what the other 3 would look like, as they told me to shut it down, and to make sure I marked #7. You can bet they used that one for their lab standard, to compare other ones to, in listening tests. I can promise you that one will sound incredible.
One more thing about these.
If you only look at phase noise, at 100 Hz, because it is relatively fast and easy to do, you really can't sort out the not-so-good ones out as easily.
(http://ar-t.co/data/tellurian%20pn%20x6%20100%20Hz.png)
As an aside.............the worst unit measures below -100 dBc @ 10 Hz offset. Some outfit that I know makes a fairly expensive part that...............going by published data............theirs (and not mine)..............that isn't spec'ed that good.
I'll shut up at this point, and let y'all ponder that one. Mainly because I would like to stay on friendly terms with them.
-
Great thread, Pat, which pretty much convinces me that the standard DIY approach of ordering a couple of clocks off of the worldwide web and replacing existing ones in a DAC or USB interface is a complete crapshoot.
On a positive note, it means I can just use the vacuum cleaner to pick up the ultra-tiny NDK clocks that have disappeared somewhere on the floor near my work area. :wink:
-
Well, yes!
Look, we all have the same problem. Even the guys who make the crystals/clocks.
There is no way to do it cheaply. Everyone wants a clock that is "lights out", but only costs $1. It can't be done.
Likewise, for the folks who make these parts............
A lot of them have the capacity to "grow their own rocks", but the bosses say it is too expensive. "Just go buy some blanks, from China, and we'll grind 'em." They only grow their own, for the specialty parts. Sometimes, the bosses say to just buy the complete part, and we will put our name on them. All of this is viewed as nothing more than a commodity, so it has to be done as cheaply as possible.
This will help to segue into something I am planning to bring up. I do work, in an unofficial way, with one or more companies that build these parts. They know that I have the equipment, knowledge, and more importantly, the time, to look into some of these subjects more deeply than they do. One in particular measures their parts at 10 Hz offset. They routinely send me samples to measure, since I do so at 1 Hz offset. I send the samples and the data back to them, so they can try to correlate what I find with their internal measurements. Sometimes they go as far to grind the top off of the parts, and probe them to see what they can find. (Seems like more work to me, but if that is how they do it, well, it is their business.) I have a hunch that one of these ventures led to a part that is available, that is shown in the data sheet of a certain DAC part. (So I have been told. All I know is that the part is available, but only if you know the p/n and can buy them buy the reel.) (And that is the other problem. Even if you know the p/n, like the NDKs, you have to buy an entire reel. Who is going to do that, other than a large company? Or some goofball consultants, who think they can make money selling sorted ones.)
More later............
-
Well.................had a reply to your question, and between ATT "DUH-verse" and the power company................it is gone!
It is one of those days. When I calm back down, I'll try to reconstruct it. Of course, it will be something totally different, but you won't know that!
(And speaking of DUH-verse, which goes down all the time, when it is up, I can not connect to any of my sites. Mail, ftp, website, you name it...............nada. If only I was in charge of that what ISP we use.) (Who am I kidding. The alternative is probably just as bad. Maybe worse.)
-
Great thread, Pat, which pretty much convinces me that the standard DIY approach of ordering a couple of clocks off of the worldwide web and replacing existing ones in a DAC or USB interface is a complete crapshoot.
On a positive note, it means I can just use the vacuum cleaner to pick up the ultra-tiny NDK clocks that have disappeared somewhere on the floor near my work area. :wink:
Lol. Same here.
And thanks for the thread, Pat.
-
OK, here is condensed version of the first part of the reply I was working on.
What do we do?
Well, not much! There is no way to get the level of performance, that allows digital audio to reach its maximum potential, without a fair amount of effort and expense.
And it is not only us who are in this situation. The folks who make the crystals and/or clocks that can be used are facing the same issues. While most of them have the facilities to "grow their own rocks", the bosses are not interested in that. "Just go and buy some blanks, from China, and we will grind 'em ourselves. On second thought, no, forget that. Just buy the whole dang part, from China, and we will put our name on it." They only go the full route of grinding their own blanks (still probably sourced from China) for their top-of-the-line products. It simply costs too much to do anything else.
The reality is all of these parts are viewed as a commodity. Translation: cheap. Dumbed down specs. Jitter at frequency >1 kHz. Noise floor. Looks good on the spec sheet. Industry is used to that spec. The purchasing guys think they are buying a quality product, and everyone is happy.
Except us.
Some of us, through arduous trial and error, have found that jitter matters, and the jitter that matters is not the jitter on the spec sheet.
And what is wrong with any of that? We all have several digital audio products, with generic $1 clocks. Which is why they pretty much all sound the same. But, maybe 20 years ago, those of use who had a background in this stuff started to discuss the possibility that jitter at 10 Hz was important. There was no doubt, among us, that only measuring the noise floor was useless. But, measuring at 10 Hz offset was something that took some effort.
(As you can see in the most recent post, going by even 100 Hz numbers miss the mark.)
Eventually, the 10 Hz idea started to gain traction. I know of one of the "big boys" that measures their batches at 10 Hz offset. (I know this, because they frequently send me samples, to look at, since they know I have the time, equipment, and more importantly, the desire to look more closely. When I return the samples, they usually grind the top off of some of the parts, and probe the unit, to see if there is any correlation to their internal data, and the results I come up with.)
But, some of us were still not happy, and that is why we started looking at 1 Hz data. Eventually, for reasons I will gloss over, we found that it was necessary to go down to 0.1 Hz.
Which brings us up to now.
(Going in a slightly different direction, from the original twilight zone ether of the 'Net. So, new post for that.)
-
"Why does it cost so much to buy something that you have tested and sorted, that only cost you $1?"
Well, for starters.................we do not make as much as we should, to make it viable. We can only buy the parts we need, for our internal use, if we buy an entire reel. (Either 1000 or 2000, depends who we buy them from. Some places want 3000. Like we can ever use or sell that many.)
So, since we a lot of these parts to unload, what to do?
First, we have to sort them, to get a rough idea of the potential yield. OK, maybe we only have to do a fraction of the reel to get a good idea, but that takes a lot of time. We now know how many we are going to get from a reel.
I am not going to give the exact number, but the yield percentage is single digits. Which means if we buy a reel of 1000, then at least 900 are of no use to us. So, right off the bat, a $1 part is now at least a $10 part.
Use your imagination, and you can probably guess why we sell the really good ones for $30. Not counting the time to do it.........................the cost of the equipment................and what-not, but..............
If we sold then for what they realistically cost to test, sort, etc., then we should probably charge $50. Which may or may not leave much of a profit margin.
Kinda pricey, eh? YEAH!
Would I pay $50 for something that good?
Maybe not. I would probably shell out another $25-30 and buy an SC-cut crystal that would be much better. If you are going to spend that kind of money, it better be good.
At $30.........................well, you can buy a part from Digikey or Mouser that is probably decent. I wouldn't use them, because I know they are not good enough. But, if you need something better than the bog standard $1 special, what alternative do you have?
None. Unless you spend at least $75 or $80. Oh, did I mention that is if you buy several of them. I know of one company (they used to make crystals for microwave oscillators we used back when I had a real job), that have an odd policy of "Well, tell us what you need, and how good, etc. and how many, and we will get back to you and let you know if we are interested." (Yes, I am dead serious! That is how they operate!) Other outfits will say it costs x, and take y weeks, but since they do work for the military...........................well, ha! Never on time. Never on budget. It is a way of life with them. Yes, you get what you wanted. Just takes forever and costs a lot more than promised.
So, here we are. Back at Square One.
As an aside, yes, we sell sorted parts. And they cost a lot. We probably should charge more, but since there is another popular part that costs around $30, that kinda sorta limits how high we can go.
And what about all of the ones we deem are not good enough?
Good question!
We could sell ones that are not up to our standards, but are probably better than aforementioned $30 part, for maybe $20, and pick up some money. But, unless you are a manufacturer, and you only need one (or maybe 2) pairs, is it worth saving a few bucks to buy "almost as good"?
No, probably not.
Eventually, we are going to have a massive "mix and match" sale, where we put all of the ones that we don't want, mix them in with the "almost as good", and probably some "you really don't want to use these dogs", and blow them out, in large batches (say 25 or 50 pieces). At that point you guys decide for yourself how much do clocks REALLY sound different!
Some will get lucky. Will be a darn good bargain. Some not so lucky. But, it won't cost you and arm and a leg. Remember, those parts that we buy? Yeah, you can't buy just one. Except for some outfits that somehow got their hands on some, and sell them one at a time. (For more than $1, I might point out.)
OK, enough of that "behind the scenes" stuff. Next post will be some interesting parts, that will show what can be done, for a buck or so.
Sort of!
-
OK, here are some parts that show what can be done, at decent price.
I do not believe these parts are still in production. Even though they are a SM part, they are a bit on the big side (3x5 mm), and are just a crystal. Most manufacturers do not want to mess with crystals. You have to design the oscillator. And make sure it reliably starts under a wide range of conditions. Takes more parts, which means more cost and "real estate". IOW, easier to buy a clock chip, and leave the oscillator design to dorks like me.
The bigger problem is the cell phone market is driving this industry. Small size, low cost, and all that entails. While there is a movement to better audio, their idea of better is far from our idea of better. Even for companies like ESS, our segment of the market is pretty darn small, and not worth much effort.
Having said all of that...................
(http://ar-t.co/data/TXC%207A%2050%20MHz%20x9.png)
As you can see, these guys do a very good job of repeatability. Darn near all of their products that I have looked at exhibit these qualities. Yes, there have been some that have a larger spread. I have reason to believe they take my data, and examine it, to see if there is a certain "knob that needs to be tweaked", in the manufacturing process. In truth, it is not that simple. But, these guys seem to be willing to give it a try. (I suspect they want to surpass the company ahead of them, in terms of global sales. We are talking hundreds of millions, annually, so it is encouraging they are looking for any edge, to move up.)
Anyway, these are a roughly $1 part. It can be done. Not easy, but it can be done.
In terms of how they would stack up, to what we consider to be "yeah, you need this level of performance", I would rate these as very good. (Ok, too bad they are not an audio sampling frequency multiple. But, if you are using an ESS DAC, in an async mode, these would be something I would look at. If you can find them...........)
To put it another way, if you were to reference these parts to 45/49 MHz, and looked at the 10 Hz numbers, they would be a tad better than the aforementioned $30 specials. Actually, I would rate them better. We have other metrics we use to evaluate clock performance. Unfortunately, we consider it to be proprietary, and even if I did share it....................let's just say unless you work with this stuff, all day long, it would be a lot of gibberish. You'll just have to trust me on this last part.
OK, at this point someone will want to ask "Well, you ever do is talk about phase noise numbers, and this, that, or some other offset frequency. Now you want to tell us that it isn't important, and that you have the functional equivalent of some secret Masonic handshake?"
Well, no. But, phase noise is an easy number to understand and grasp. Just takes time to measure below 1 Hz offset. But, there are other properties that are important. While I have stated that drift is not important (it isn't), there are short-term "funnies" that clocks do, and finding out which ones are important, which ones are not, and what level of the important ones took a lot of work. And, honestly, isn't easy to explain.
Ok.................next up.....................stuff that will really make me more unpopular than I already am! (How can that be possible? Just wait.)
-
OK, here is a bog standard $1 clock, a 5 x 7 4-pin package, that you can buy at any number of places. Don't know who made it. Have a pretty good idea, looking at the number on the case. But, it isn't about them. It is about what you find in 99.99% of your typical audio product.
(http://ar-t.co/data/Mystery%2022%201%20kHz_100%20kHz.png)
Yeah, not so hot. Ok, not dreadful, but let's just say this is why digital audio sounds like digital audio, and not what we call "audio".
Now, if you believe that clocks are spec'ed by "jitter" numbers, you know, the ones that start at 1 kHz and go up to 100 kHz (or wherever), well it says 170 fSec. Ok, big deal. Couldn't pay me to use this part. Ok, maybe in my ham radio, if I still did ham radio. (No, probably not, because I probably have a 100 or so parts around the shop much better than this one. And there is a reason they are just "sitting around the shop".)
But, what if you decide to be brave and extend the jitter frequency down to 100 Hz? Some folks are that brave!
(http://ar-t.co/data/Mystery%2022%20100%20Hz_100%20kHz.png)
Hey, how 'bout that: it doesn't change! So, maybe those folks aren't really that brave, after all.
Let's get braver....................10 Hz!
(http://ar-t.co/data/Mystery%2022%2010%20Hz_100%20kHz.png)
Oh, my..................the jitter doubles. No way we can put that in the data sheet!
But, what if you get really, really, really brave, and go down to 1 Hz? (Heresy!)
Uh...............saving that for later. You'll see why. (Heh, heh, heh!)
-
OK, everyone still here? Everyone ready to see how many people in the industry I can get to hate me? (Assumes they don't already. Most probably do, so what is a few more?)
First off, I can see ABSOLUTELY NO REASON to use this next part. Yet, it is really popular. (Just goes to show how advertising in trade journals pay off.)
If you look at the datasheet, us skeptics just shake our head and say "WHY?" Why go to all the bother? OK, I get it. Put a nice rock, in a chip, and phase lock it to some oscillator. This way, we can put all of our efforts into a really good rock, and benefit from it. And not have to grind dozens of different frequencies.
So, here is the data sheet:
https://www.silabs.com/documents/public/data-sheets/si530.pdf (https://www.silabs.com/documents/public/data-sheets/si530.pdf)
And the diagram of what goes on inside:
(http://ar-t.co/data/530.png)
So....................let's say you are one of the guys who thinks that all jitter comes from all the junk that is created by muxes, dividers, etc., etc., etc. So why in the world would you use a clock that is full of that crap?
(Ok, not saying any of that is not a problem. It just isn't the biggest problem, when it comes to digital audio. Yet, a lot of folks pick a clock source that has those problems already built in. So, you are behind the 8-ball, before you even start.)
So................drum roll......................here is this marvel of modern design. Overlaid with our bog standard $1 special. And what is possible, from the right vendor. For roughly that same buck.
(http://ar-t.co/data/3%20samples%201%20kHz.png)
Oh my goodness....................that is truly awful! Yes, much worse than what you can buy for $1! So why do so many of you use these things? (You should see some of the hate mail I receive, from manufacturers, who take strong exception to my measurements. Of course, their main retort is "your measurements are crap, and you have no idea what you are talking about.................you are the world's biggest know-nothing idiot.............these parts have really good measurements, so stick it in your ear and don't ever bother me again, got it?")
Ok, got it!
I can hear the howls of derision already..................
"We use one that is selected for low jitter'. So, there!"
OK, maybe you do. It might get rid of all that crap around 1 kHz. (Some of you may recall what happens when I stick an extra divider in the measurement chain, to get the frequency down to where our equipment can measure it. It is in that area where you will see those effects.)
But....................if you look at the 49 MHz sample.....................
It also runs through that "extra divider", and we can see how it stacks up to the marvel of technology. (Actually, our extra divider adds crap in around 100 Hz. So, how do you explain all the crap from 1 kHz to 10 kHz? Only one explanation! Who can guess what it is?)
So, using our standard, of jitter from 1 Hz to 1 kHz, this is what we get. It is over 10 pSec. While the other 2 parts are in single digits.
OK, maybe I am being too picky, and not fair enough. Maybe if we look at jitter from 100 Hz up, and not the stuff down around 1 Hz, the area under the curve will be less. And let's see if the nasty crap above 1 kHz will be mitigated.
OK.................here it is:
(http://ar-t.co/data/3%20samples%20100%20Hz.png)
Oh, ok! We are back to close to that magical femtosecond range! (Remember, in the previous post we saw the 1 kHz and 100 Hz numbers were pretty much the same.)
"See, I told you that you were an idiot!"
Yeah, whatever..................
So, in the sake of fairness, let's assume someone is brave and goes down to 10 Hz!
(Isn't this exciting?)
"Hell, no!"
(http://ar-t.co/data/3%20samples%2010%20Hz.png)
OK, now that we are looking at more of the true jitter, and not the silly noise floor, the divergence starts to show up. (The reason why the 3rd sample, in the second plot is a bit higher than the 2nd sample, which is clearly not as good, is the effects of that divider, and having the s/w manipulate the data. The latter skews the noise floor. You can see it if you look closely.)
So, now we get a better picture of the relative values. The 3rd sample pretty much stays the same, and the crud added by the extra divider (which is also in sample 1, remember) is not significant. Sample 2 is about double its previous, and sample 1 also is roughly double. Any wonder no one publishes jitter below 1 kHz? Any wonder now why it means pretty much nothing?
(I had planned to show jitter from just 1 kHz, to 100 kHz, but it somehow did not make it to the thumb drive I thought it was on. Oh, well. If you really want to see it, let me know. But, haven't we seen enough? Do I need to get more manufacturers mad at me!)
Well, at this point, best to wrap things up for a while. You guys have a lot to think about.
"Are you going to share with us any of the new hate mail that you get?"
No, probably not. If folks know I am dying to show how many of the emperors are buck nekkid, they may keep their pie holes shut. Think of the entertainment us dorks in Texas will be depriving ourselves of!
(I think I just ended a sentence with a preposition. Better stop before the resident English teacher sees it, and raps the back of my hand with a metal-edged ruler!)
Enjoy.................
-
Guess what.....................found the "spec sheet" for the NDK NZ2520 parts.
(http://ar-t.co/data/NDK%20NZ2520%20spec.png)
Yeah, some will perform that well. But a lot will not.
What do you want for $1?
-
Pat
3 questions from an idiot;
1 Temp
Knowing that temp variations affect jitter, how does it affect jitter. Is it he variation over a short period of time? Like have the furnace or AC turning on and off and the clock is not covered well. Or is it the base temp, like having the house at 68F in winter and 72F in summer. Do we care about the base temp once the temp has stabilized?
2 Power
A clean supply to the clock is crucial, we know this. How does this translate? Is there a PSRR number , or similar, for clocks based on frequency? Assuming we don't want to use a 3 pin regulator is it because the high frequency stinks or is it because the output voltage can drift vs temp?
3 Oscillator
At another forum there is much talk about buying good crystals and they are using the crystals in different oscillators. Being 30+ years since I've used my FCC license or looked at RF oscillators, can you explain some the whats, whys, and the don't do this unless you're a huge idiot....
Mike
-
Pat
3 questions from an idiot;
You are not an idiot. You are not posting this on that hobbyist forum!
1 Temp
Knowing that temp variations affect jitter, how does it affect jitter. Is it he variation over a short period of time? Like have the furnace or AC turning on and off and the clock is not covered well. Or is it the base temp, like having the house at 68F in winter and 72F in summer. Do we care about the base temp once the temp has stabilized?
It doesn't affect jitter. It affects stability, which is too slow to be considered as jitter.
The thing that is more important is that you keep the oscillator powered up and actually oscillating. Figure that it takes 10-20 minutes for most crystals to settle in, once they are powered up. The enable pin, on dang near every commercial part, turns everything off. This is to save power. Which, in our applications, isn't a concern. Some will argue that having two clocks oscillating, even if they are well isolated, is a no-no.
OK, let's say it is. You need to find a better way to isolate them, so that both are on and oscillating.
2 Power
A clean supply to the clock is crucial, we know this. How does this translate? Is there a PSRR number , or similar, for clocks based on frequency? Assuming we don't want to use a 3 pin regulator is it because the high frequency stinks or is it because the output voltage can drift vs temp?
The guys who make oscillators keep their regulator circuits a closely-guarded secret. This is because they only measure the jitter above some frequency, usually 1 kHz. So, if that is how you measure jitter, then yes, it makes a big difference.
But, since you asked........................
Here are two of our crappy products, that are mainly different in that one is powered by a wall wart, with a 5-legged $1 gizmo for the clock reg. The more expensive one has a traditional "leen-yer" power supply, and a discrete reg that we cooked up.
(http://ar-t.co/data/102%20x%202.png)
If you look in the range from around 100 Hz to 1 kHz, you can see about 2 dB of difference. You will have to trust me, and believe me when I state that is the difference in the regs.
One of our dealers would kvetch that not only could he hear the differences, in the regs, he would insist he could hear the difference in the caps we used on the regs.
OK, let's say he can. We can not measure it, so we tend to say it is insignificant, if the difference does exist.
Of course, he claims our test equipment is not sufficient to do its job.
Ok.......................
As for reg drift.................
Someone that you probably know built some regs that used a mish-mash (color-wise), of different color LEDs, to get the right voltage for the clock. I think we were able to determine we could see drift, wrt temperature.
Maybe.
We would see more drift when the A/C cut on and off. Which is why the new building has a storage area that doubles as the "environmental chamber". Which means it is tucked away in a remote corner, where there is no HVAC system.
3 Oscillator
At another forum there is much talk about buying good crystals and they are using the crystals in different oscillators. Being 30+ years since I've used my FCC license or looked at RF oscillators, can you explain some the whats, whys, and the don't do this unless you're a huge idiot....
Mike
Well, let's say they are the idiots, but for other reasons.
First, you need to have the right crystal. A lot goes into that. And sometimes, just maybe sometimes, you can find some sub-$1 POC crystal that will surprise you. But, you have to be lucky, and then some. And it sometimes it takes more than luck.
The sub-$1 POC crystals we used to build our clocks with were able, on occasion, to yield great results. But, they were not an ideal configuration. It made things tricky. You had to operate on the right part of the impedance curve. You had to have the drive just right. And, believe it or not, the Q had to be just right. Too low................yeah, forget it. No way it was going to work. Too high, and it would never work. (It was at that point I decided we needed to find a better alternative.)
So, if you have the wrong crystal, you either have no chance, or you get really lucky.
If you really do have the right crystal, they are a lot more forgiving. You can do most anything to them, and they seem to work fine.
Of course, knowing that the idiots are spec'ing these crystals, and they don't really know what is good, well, all bets are off.
(I can say this because I am 99.9999% certain none of the "idiots" have the background that I do, and since they are the first ones to tell me that I am the one who is really the idiot........................well, I did design oscillators, for a living, at one time. And not some computer/innerwebs related tasks. 'Nuf said.) (Ok, probably too much, but some of those folks have seen fit to malign me, when I am not around.)
At this point, I will point out that the place I worked at built a 6 GHz SSB microwave network. (No, I did not design the master oscillator. That was done by some place in Austin, that probably had ties to Charles Wentzel's outfit.) Anyway, the point is this system had to hold 1 Hz stability, coast-to-coast. And this was at 6 GHz.
Not an easy task. (Ok, somehow it was phase-locked to the 100 kHz LORAN-C network.)
Anyway......................when we were packing up to move, I found the phase noise plot of the system. This was measured at the 6 GHz clock, so that we could use the "90 degree phase shift method", of measuring phase noise.
That clock put most audio clocks to shame.
OK, granted, I think we paid $40k, for the master clock, at each repeater. So, for that price, it better be good.
But, it was 6 GHz.
Not 24.576 MHz.
Anything else?
-
So, what might this mean for the average audiophile like me? Let’s take just one example. My very best digital recording is a Mercury SACD of Rachmaninoff’s 3rd Piano Concerto played by Byron Janis. The original recording was made in 1960.
But I also have the same recording on vinyl. I had a group listening session a couple of weeks ago in which I played both versions. The result: everyone preferred the vinyl version.
“Lot’s more involved here other than just digital vs. vinyl” you might say, and you would be right, of course. And the differences were not huge, but overall consistent with my daily experience. When I want to really listen to music, I find myself reaching for my digital only when I do not have the same piece of music in my record collection.
Are you saying that I could (maybe) get a lot closer if I upgraded the clock in my DAC?
-
In theory, yes.
The problem with the theory is that it is not easy to implement.
1.) Some frequencies are a lot harder to make quality clocks than other frequencies.
2.) It isn't cheap, to make a good clock.
3.) It isn't practical for most manufacturers to spend the money to get a really good clock. At least not if they want to hold the price down. Everyone only wants to spend $1, for a clock. It isn't going to happen. (Unless you are as stupid as we are, and buy some gizmo to measure an entire reel of $1 parts. Assuming you have >$10k to buy the gizmo. And know what to do with it. Which is another story. I know of one crystal manufacturer that has the $33k one that ours is based on. They have never figured out how to make it work. I know, because they told me so. But, they did not want to invest the time or money to pay me to show up and show them how to use it.)
4.) Some designs are harder to fix than others. Ones that use a DAC chip that does everything async has its own issues, when it comes to sound quality. Yes, some of those can sound very good. There are some folks who can not stand how they sound. Possibly due to their personal prejudice, based on their understanding of how they operate. Another one of those "well, in theory...................blah blah blah" things.
5.) And the big one..........................
99% of the industry (maybe more) has no idea what a good clock really is. You can read any number of forums, and see any number of threads, all discussing x pSec (or x fSec) of jitter. None of which mean diddly-squat. Unfortunately, the manufacturers are of the same mind set. Some are clueless that things can sound better. Some are convinced they are already at the pinnacle of performance. Some just don't want to hear about it.
Some tell me to shut up and go away.
Most of them just are not interested. Either because they know it will cost more money, they are unaware how much better they can make their products, or they flat-out just don't care.
Look...................there is one company, that makes fairly decent part, that most companies can buy for around $20, in the quantities they would use them in. (I haven't checked the price lately, so it may even be a bit less than that.) But, how many use them? Not many that I have seen.
But, even if they do.........................
How can I put this in a way that will not get various parties upset....................?
When these parts were introduced, a manufacturer that we know bought several pairs, and had us measure them. Around 3/4 of them met spec. From what I recall, the total number of units we measured for them was around 2/3 of which we would consider as "decent". IOW, sticking one of them into a product would give a level of performance that would set them apart from the competition. But, would still leave something to be desired. (What those same parts would measure today, I can not say. Your guess is as good as mine whether or not they got better, stayed the same, or the folks in China who supply the crystals are dumping whatever they can on them, and get away with it.)
So, there is a starting point. If you want to attempt to make your DAC sound better, you can try one. Assuming it will fit, they come in the right frequency, and someone can help you if you can not perform that task yourself. You will only be out around $30 to try one. (In theory!)
A lot of folks have gone back to vinyl. I understand why.
-
Art, thanks for your reply.
Looks like I have some digging to do.
1. Does my Ayre QB-9 use a DAC chip that does everything async?
2. What clock is in there now?
3. How hard would it be to replace?
4. Where does one find the magic $30 replacement?
-
There is an external clock system that M2Tech made many years ago to drive their XMOS USB to S/PDIF, I2S, AES/EBU interface box.
There was an external switch that selected the clock family that was to be used for what you were playing. A bit more manual than most probably like. It used a 50 Ohm coaxial connector output with RF cable between the two units.
I have one here. Is there a way to measure the unit for a possible upgrade?
-
Art, thanks for your reply.
Looks like I have some digging to do.
1. Does my Ayre QB-9 use a DAC chip that does everything async?
Don't know........................Charlie is no longer with us, and I haven't talked to them much since then.
2. What clock is in there now?
See above. I do know Charlie paid a lot of attention to the clocks, as everything else.
3. How hard would it be to replace?
See above.
4. Where does one find the magic $30 replacement?
Nothing magic about it.
Mouser...................Digikey....... ........others...................
http://www.crystek.com/crystal/spec-sheets/clock/CCHD-957.pdf (http://www.crystek.com/crystal/spec-sheets/clock/CCHD-957.pdf)
-
There is an external clock system that M2Tech made many years ago to drive their XMOS USB to S/PDIF, I2S, AES/EBU interface box.
Oh..................those guys...................(if it is who I am thinking of)
How can I say this diplomatically?
One of our dealers sent us his USB-SPDIF converter, made by those guys, when he signed on with us.
"Well, did you measure it? What can you do to fix it?"
"Uh, yes...............and nothing. Do you want me to throw it away for you?"
There was an external switch that selected the clock family that was to be used for what you were playing. A bit more manual than most probably like. It used a 50 Ohm coaxial connector output with RF cable between the two units.
I have one here. Is there a way to measure the unit for a possible upgrade?
Of course! The question is do you really want to spend the money for us to measure it? Since you are not a manufacturer, I guess we could give you a break, on the price. But, if it is what I suspect that it is, I would not feel great about taking your money.
Let's put it this way.......................you can probably buy what is inside it for around $30, from Mouser or Digikey. (Translation: if you can take the lid off, you will know for certain. And that won't cost anything!)
-
Ok, I will take the panels off and look what is inside.
-
Nothing magic about it.
Mouser...................Digikey....... ........others...................
http://www.crystek.com/crystal/spec-sheets/clock/CCHD-957.pdf
This is a decent oscillator at 45 and 49MHz, but not at 24 or 22MHz.
Try this one instead:
https://www.mouser.com/datasheet/2/94/CCHD-575-18036.pdf (https://www.mouser.com/datasheet/2/94/CCHD-575-18036.pdf)
Steve N.
-
Art, thanks for your reply.
Looks like I have some digging to do.
1. Does my Ayre QB-9 use a DAC chip that does everything async?
It's not the DAC chip that is asynch, it's the USB interface circuit. This was the old asynch technology from Gordon Rankin, using a TAS1020B chip, called Streamlength. I have used the TAS1020B in earlier products, the Off-Ramp 3 with my firmware coming from another company. The newer technology of my Off-Ramps 4-6 sounds better, however I did go back a upgrade the decoupling caps and clock in the Off-Ramp 3 for several customers and it is quite good now. I don't do mods to other companies products anymore.
The latest version is XMOS-based and supports DSD. It seems to use two CCHD957 oscillators at 49 and 45MHz. You will be hard-pressed to find better than these off-the-shelf.
They also changed DAC chip from the Burr-Brown DSD1796 to the ES9016S at some point.
Read more at https://www.audiostream.com/content/ayre-qb-9-dsd-dac#8osL6QBPDSSmsA1j.99
2. What clock is in there now?
I think you may have to open it up and examine it to see what is used for the clock. I believe I used 12MHz in the Off-Ramp 3. For XMOS, you will find two oscillators, at 24.576MHz and 22.5792MHz or 49.152MHz and 45.1584MHz.
3. How hard would it be to replace?
Any good technician can do this if he knows SMT rework.
4. Where does one find the magic $30 replacement?
Mouser.com, Digikey.com
-
So, what might this mean for the average audiophile like me? Let’s take just one example. My very best digital recording is a Mercury SACD of Rachmaninoff’s 3rd Piano Concerto played by Byron Janis. The original recording was made in 1960.
But I also have the same recording on vinyl. I had a group listening session a couple of weeks ago in which I played both versions. The result: everyone preferred the vinyl version.
Means nothing. The is no doubt in my mind that you have very high jitter in your digital system. Bring this down and reduce the distortion in your DAC and you will rival or exceed the vinyl sound quality.
Want to hear REALLY low jitter? Try a Sonos followed by a Synchro-Mesh reclocker into a Metrum Pavane DAC (no reclocking).
Steve N.
-
This is a decent oscillator at 45 and 49MHz, but not at 24 or 22MHz.
Try this one instead:
https://www.mouser.com/datasheet/2/94/CCHD-575-18036.pdf (https://www.mouser.com/datasheet/2/94/CCHD-575-18036.pdf)
Steve N.
Looking at the two datasheets the noise floor for the first unit is 7dB lower noise at 10Hz than the second datasheet for the 22MHz and 24MHz parts. Why would it be better?
-
This is a decent oscillator at 45 and 49MHz, but not at 24 or 22MHz.
Sort of right.....................
Those parts should NOT have the same phase noise numbers. There should be a 6 dB difference, between the 22/24 ones and the 45/49 ones.
Try this one instead:
https://www.mouser.com/datasheet/2/94/CCHD-575-18036.pdf (https://www.mouser.com/datasheet/2/94/CCHD-575-18036.pdf)
Steve N.
Uh, no thanks!
Only if you worship at the altar of low noise floor. Which most of the industry does. Which is also why they are wrong.
-
Looking at the two datasheets the noise floor for the first unit is 7dB lower noise at 10Hz than the second datasheet for the 22MHz and 24MHz parts. Why would it be better?
Because the industry is convinced they are experts, and noise floor is the most important spec.
(IOW, see above post.)
-
Looking at the two datasheets the noise floor for the first unit is 7dB lower noise at 10Hz than the second datasheet for the 22MHz and 24MHz parts. Why would it be better?
At 10Hz, the 575 is -101dB, the 957 is -98 dB, both at 24.576MHz. The 575 is better at this frequency, 3dB better.
It's the phase jitter that is highest that will be audible, not that at 100KHz which is the "noise floor". Nobody can hear -165dB or -170dB.
Steve N.
-
Nothing magic about it.
Mouser...................Digikey....... ........others...................
http://www.crystek.com/crystal/spec-sheets/clock/CCHD-957.pdf (http://www.crystek.com/crystal/spec-sheets/clock/CCHD-957.pdf)
Well, well, well. That just happens to be the clock I bought a little while ago for some cheap dac diy'ing (along w/ your Newava recommendations). I was pleasantly surprised by what I heard. I was skeptical I'd hear much daylight between it and the clock it replaced - the above-mentioned Crystek 575 (and before that, a Fox Xpresso). But the 957 sounds noticeably better to me.
-
It's the phase jitter that is highest that will be audible,
?????
Oh, really.........................
not that at 100KHz which is the "noise floor". Nobody can hear -165dB or -170dB.
Steve N.
Then why do guys like you rely so heavily on measurements that are really nothing but the noise floor? (That is what your fancy Tek 'scope does, whether you want to admit it or not.)
What you fail to realize is there are other metrics, that determine a clock's suitability, for use in a digital audio system. The phase noise is just part of it, and is the easiest to comprehend. (The folks that build these parts know all of this. They also know what the "magic numbers" are, for all of those other bizarre and obtuse metrics. Why do you think they are willing to work with such a PITA like me? Because I know this stuff, and have the understanding which ones are important and why they are important. And more significantly at what level they are important.)
But you guys feel free to brag about how many fSec of jitter that your stuff has, based on the noise floor specs.
-
?????
Oh, really.........................
Then why do guys like you rely so heavily on measurements that are really nothing but the noise floor? (That is what your fancy Tek 'scope does, whether you want to admit it or not.)
What you fail to realize is there are other metrics, that determine a clock's suitability, for use in a digital audio system. The phase noise is just part of it, and is the easiest to comprehend. (The folks that build these parts know all of this. They also know what the "magic numbers" are, for all of those other bizarre and obtuse metrics. Why do you think they are willing to work with such a PITA like me? Because I know this stuff, and have the understanding which ones are important and why they are important. And more significantly at what level they are important.)
But you guys feel free to brag about how many fSec of jitter that your stuff has, based on the noise floor specs.
The period jitter distribution is not the noise floor. All of the metrics, including phase noise plots, period jitter distribution plots, period jitter spectrum plot and listening tests are needed to fully characterize the jitter.
Steve N.
-
Did not have to remove the top. Plenty of pictures of it on the web. Could not find the part numbers DFAS13-LH with the clock frequencies, but it is from 2011. Any idea about them would be interesting.
The hiFace Evo clock unit pc board from the search:
(https://www.audiocircle.com/image.php?id=194285)
-
The period jitter distribution is not the noise floor. All of the metrics, including phase noise plots, period jitter distribution plots, period jitter spectrum plot and listening tests are needed to fully characterize the jitter.
Steve N.
That, on its own, is debatable, but is of ZERO value, when discussing clock jitter. Which is what this is about.
But, fine, you believe whatever you want to believe. I have the feeling any further discourse with you is an exercise in futility.
For everyone else.....................
If one of you guys swaps out the bog standard clock, with one that is "ok" (by the metrics ne'er-do-wells like us use), and it sounds "day and night different", what has changed? Has any of the stuff someone here likes to ramble about (because that is what they have to measure "jitter") changed? NO! The only thing that has changed is the clock, and none of those metrics are applicable to quantifying any clock.
But, what do I know..............................
And for the record......................
Yes, we have expensive gear that measures all of that other crap. I can say with 100% certainty is useless, when it comes to clock jitter. In fact, one of the best clocks (in our shop) measures like doo-doo, when those metrics are used.
Why?
Because it gives a "double-hump" histogram. IOW, it is the kind of result one would see if there is "deterministic" or "correlated" jitter present. (A good example of this would be mains-related crud getting into the clock supply, and mucking things up.)
In this case, it is due to an odd-looking, slightly asymmetrical waveform, that is created when you take a 5 MHz "rock", build an oscillator with it, and run it through a Class C amp to make a 10 MHz oscillator. (You can see what its phase noise looks like somewhere in this thread.) (Should not be too hard to figure out which one it is, since you already know it is a 10 MHz one. The last hint is how good its phase noise plot is.)
-
Oh, so it is "those guys". Yes, those guys, from the land of my ancestors.
OK, joke time.
A good buddy is an Alfa Romeo fanatic. (And not of a heritage from that place.)
"Know what the best thing about an Alfa is?"
"No. What?"
"Designed by the Italians! Know what the worst thing about an Alfa is?"
"No, please tell me."
"Built by the Italians!"
"Well, duh!"
"I see that you know Luigi and his cousins."
"Yeah, because they would be my cousins, if they were not in Milano."
OK, no more jokes....................
Spec sheet:
https://lnx.m2tech.biz/wp-content/uploads/2017/06/Evo-Clock-Two-User-Manual-PrA.pdf (https://lnx.m2tech.biz/wp-content/uploads/2017/06/Evo-Clock-Two-User-Manual-PrA.pdf)
Most important part:
Phase noise:
-98dBc/Hz (@ 10Hz)
-131dBc/Hz (@ 100Hz)
-144dBc/Hz (@ 1kHz)
-154dBc/Hz (@10kHz)
(Looks like a 24.xxx MHz clock.)
IOW, about the same as that $20-30 part previously mentioned.
Since they no longer make them, you can make something that works just as well for a lot less than what they sold theirs for. Better than what a lot of you presently have. Still leaves a lot to be desired.
(No, we have no plans to make a similar product. Folks in the Pro Audio world have approached us about this. When you talk to the places they buy their stuff from, and ask them what kind of demand they see, for this type of product................................ ...)
(Like I said, we have no such plans!)
-
art,
Thanks for the feedback.
I have one and it has both the 22.5792MHz and 24.576MHz clocks in it with the switch between the two.
Looks like no real upgrade going from what is already installed and something 8 years newer from your observation in the thread.
-
Yep, not much sense in mucking it up.
-
Because it gives a "double-hump" histogram. IOW, it is the kind of result one would see if there is "deterministic" or "correlated" jitter present. (A good example of this would be mains-related crud getting into the clock supply, and mucking things up.)
Not always. Here is my Synchro-Mesh reclocker output at the end of a 4-foot coax cable:
(https://www.audiocircle.com/image.php?id=179709)
The double-hump can be caused by impedance mismatch as well.
Steve N.
-
I posted this yesterday on the ar-t forum, but have a few more thoughts:
"I'm still listening to music, rather than devoting much time to writing, but I'd be remiss if I let too much time go by before saying that the new Legato arrived on Friday, and is one of the best audio devices that I've ever had a chance to evaluate. From the very first track, it has impressed with an organic, but crystal-clear rendition of the textures, timing and timbres of everything I chose, whether "audiophile" or just plain ol' damn good music.
I had thought that I was one of those guys who had to upsample everything to DSD1024 to, you know, make it sound extremely hi-res, but please trust me on this: Properly reproduced (and this isn't, I'm sure, limited just to a good clock) 44.1 recordings are incredibly satisfying. More to come, but if you have a chance to audition one of these, do so."
A couple of things that I've been asked:
"Wait, limited to 44.1? I've got all kinds of hi-res recordings, both PCM and DSD. Why would I want to retreat to mere CD resolution?" Well, as Pat and others have pointed out, the 22 and 24mhz clocks have better phase noise performance than their 44 and 49mhz counterparts, regardless of manufacturer or price point. And I'm now a believer that ar-t's testing and sorting abilities can produce superior digital reproduction, at least compared to anything that I've had in the past.
You can, using the software that you probably are already using (I use Daphile, a free Linux-based streaming system), convert all of your hi-res recordings on the fly to 44.1 output, and to these ears, 44.1 playback done extremely well is superior to any of the hi-res formats. In my house, and on my system.
"Yeah, but SPDIF? Hadn't we long ago abandoned that primitive tech for fancy USB to I2S gizmos (and, of course, some necessary "purifiers")?" Yes, but unless you're doing that process inside the DAC, there can be problems with things like the length of the I2S wiring that are significant, and you're still at the mercy of that first piece of the puzzle, the clock, or clocks. I've seen (and used, I'll admit) some boards that eschew the old-fashioned low speed clocks and push it to 90 and 98. It's true that you are at the mercy of the design of the receiving circuit in the DAC or other device, but Pat has spent many years perfecting the design of the transmitting side and associated hardware. (Use a long cable. :wink:)
In short, the Legato is a great audio device, one of the best that I've spent my own hard-earned dough on, and, in the specific context of this thread, it appears to be the answer to the question of how significant clock jitter is in the whole process.
-
Just want to say with great sadness that the OP 'art', Patrick DiGiacomo aka Jocko Homo (on diyaudio and diyhifi.org) passed away a month ago...