CLOCKS, what should we look for in next generation

[pkane2001]

1 hour ago, Superdad said:

 

That s NOT AT ALL what we are working on.  Don't even bother trying to guess.

 

OK, I2S interface over fiber, then?

[mansr]

4 hours ago, Superdad said:

Every processing chip (and even some chips that we don't think of as processors) needs a clock source.  So for the interfaces it does not matter if it is an Ethernet switch or PHY, a USB hub or PHY, the processors or FPGA associated with them, they all need a clock source.  As does any general purpose processor, be it a ARM or Intel type, or graphics or DSP.  

Such clocks will most always be at whole number frequencies, 12MHz, 13MHz, 24MHz, 25MHz being the most common.

 

Audio rate clocks as part of a DAC will never match those: they will typically (with exceptions for ASRC or very high sample rate conversions) be multiples of 22.5792MHz and and 24.576MHz.

A CPU will happily run at any clock rate up to its maximum. Old digital TV set-top boxes actually derived all clocks from the incoming data stream. Using a single clock source in a computer with standard interfaces isn't practical, however. USB requires a 12 MHz reference clock which is multiplied to produce the 480 MHz bit clock. Meanwhile, Ethernet and SATA need multiples of 25 MHz. For this reason, it is necessary (or at least simplest) to use separate oscillators for each.

 

4 hours ago, Superdad said:

And there really is not much good reason to source together the various chip clocks and the audio bit clocks since most designs are already doing their best to keep these circuits isolated from one another.

And then there's that, of course.

[jabbr]

6 hours ago, PeterSt said:

Whatever the configuration exactly is or can be made (al is always modular), I can reason out in advance that it will be a noisy mess.

The disadvantage in me is that I already deal with the electrically noisy environment in a PC at miles of distance, and with controlling that noise, controlling the DAC's behavior. And yes, the (USB) interface is sure part of that, hence it is a major source of "noise" / audible nastiness. And now you ask me to avoid the USB interface and put the whole source of further electrical noise right next to the fragile D/A process.

 

The USB interface requires processing, typically either a “dedicated” processor or an ARM +/- FPGA. 

 

Decoding the Ethernet stream is not fundamentally different in terms of processing.

 

The advances in embedded ARM along with FPGA are astonishing. Witness the “RF SoC” which includes multiple multi-GBs ADC/DAC — used for radar, cellular etc etc. These chips are used for 100-400 Gbe Ethernet (take a look at the new Mellanox NIC which has a big fat Xilinx chip right in the middle)

 

Decoding 1 Gbe and outputting PCM or DSD is child’s play for the old versions of these chips which are becoming more affordable.

 

This is the approach Merging has taken with their Ravenna implementation and which I have in my lab.

 

Clock generation chips were built for these suckers and trust me phase error is waaay more important for putting a modern equivalent of a cruise middle between the goal posts than home audio. That’s what these low phase noise crystals are made for — go ahead and show me a phase noise plot of a generated clock vs a really good crystal — there are many factors which favor the crystal and this stuff is physics and math. Huge area of interest. 

 

(Not responding to you there Peter)

 

Consider vibration isolation because you may be running into a 1/f noise issue. You are prob at the point where tuning 1/f is the limiting factor. 

[jabbr]

@PeterSt consider this: you strip down windows and have invested in optimizing your pc-> dac, and naturally because of this investment you have a bias toward windows, but with ARM/Linux and in a SOC FPGA not only do you specify the hardware but the kernel is stripped down to only the essentials. The CPU is useful to run things like the IP stack, DHCP etc and to communicate with the rendering PC which sits on the network. 

[PeterSt]

27 minutes ago, jabbr said:

but with ARM/Linux and in a SOC FPGA not only do you specify the hardware but the kernel is stripped down to only the essentials.

 

Ask the NXP rep I talked to day before yesterday (on a Farnell Warehouse showcase) what I asked him about 16 core ARM processors ...

 

Meanwhile we are working on other things (secret : with an 1.2M memory (LUT) FPGA) because we are waiting for your interface. Haha.

And we also received a most interesting development board from NXP ... But that would eliminate the interface. And because we are sports, we will wait for yours first. 

 

 

[jabbr]

33 minutes ago, PeterSt said:

 

Ask the NXP rep I talked to day before yesterday (on a Farnell Warehouse showcase) what I asked him about 16 core ARM processors ...

 

Meanwhile we are working on other things (secret : with an 1.2M memory (LUT) FPGA) because we are waiting for your interface. Haha.

And we also received a most interesting development board from NXP ... But that would eliminate the interface. And because we are sports, we will wait for yours first. 

 

 

Ah well, I’ve been thinking that supplying the interface alone would be far too expensive for me to deliver — not interested in supporting forever etc for any reasonable price, and besides the interface itself isn’t that hard — what is hard is programming the SoC with all the voodoo. 

 

What doing sing your own interface enables are modifications to the DAC itself  —— oh and little hint guys: why oh why couldn’t the CPU work directly off a 24.576 or whatever clock?

 

I think the interesting part comes after the FPGA when isolation etc and there are some really interesting opportunities here. So let me say that creating the Ethernet -> PCM/DSD (no not I2S because that’s interleaved) is basically a learning excersize for doing more interesting things after the FPGA ... the DAC clock runs the whole shebang

 

[zephyr24069]

On 11/8/2017 at 7:52 PM, JohnSwenson said:

I've been thinking about writing a primer on crystal oscillators and digital audio and this looks like the perfect place to put it. I promise I will leave out all the complex math that most articles are filled with. I'm NOT going to go into how it all works, since most people don't care, just what makes them different and how that matters for audio.

 

A crystal oscillator is a combination of a special piece of quartz crystal and an electronic circuit, the combination produces periodic signal at a specific frequency, several things can change this frequency:

 

Thickness of the quartz piece, this is the primary determining factor in the frequency

 

Temperature of the crystal,  this is called the temperature coefficient (TEMPCO for short), it is the change in frequency for a small change in temperature. It is not constant but changes with temperature, this is the TEMPCO curve. All TEMPCO curves have a temperature where the TEMPCO is zero, this is called the inflection point. If you run the crystal at this temperature, small changes in temperature produce no change in frequency, THIS point is where you want to run a crystal oscillator. If the temperature is far away from this point a small change in temperature makes a big change in frequency, you do not want to be here.

 

Capacitance across the crystal, all crystal oscillators need some capacitance across the crystal to work, changing that capacitance changes the frequency.

 

Power flowing through the crystal. The oscillator circuit works by running power (in the form of an AC signal) through the crystal, changing the power changes the frequency.

 

TEMPCO is THE most important characteristic besides the thickness, so a lot of crystal oscillator design has to do with this.

 

Now on to "cut", this is how a slice of crystal is cut out of a block of quartz. This is all very complicated so I won't go into the details, just to say there are many ways to do this and the exact cut determines the properties of the oscillator.

 

The most common cut (BY FAR) is called the AT cut. Almost all the oscillators in your electronics devices use the AT cut. The primary reason for this is that the inflection point of its TEMPCO curve is at 25-35C, right around "normal" room temperature, especially in a box where the electronics warm it up slightly. With this cut you usually do not need to apply any temperature stabilization since it is at a point where a change in temperature makes a very small change in frequency.

 

The other cut we need to talk about is the SC cut, this is used in OCXOs, I'll talk about that later. This cut has much higher Q than the AT cut, which means much lower phase noise, BUT in order to get that the inflection point of the TEMPCO curve is at 95C. THIS is why an oven is needed, not so much to stabalize the temperature but to get the crystal to the inflection point where a change in temperature makes an extremely small change in frequency. The slope of the TEMPCO cut around the inflection point is much shallower than the AT cut, so a given change in temperature makes a much smaller change in frequency, IF it is at 95C, outside of that and it is worse than an AT. So you ONLY want to use an SC cut in an oven.

 

So what aspect of this is really important for digital audio? Most oscillator spec sheets spend a lot of time talking about their long term stability. It turns out crystals will change frequency over time (called aging). Some applications need this, digital audio does not. A 1 part per million change in frequency over  years time is completely irrelevant. Another spec that is important for some application is the TEMPCO, how much the frequency is going to change as the heater turns on and off. Again, irrelevant to digital audio. What DOES matter is phase noise. I'm not going to go into any detail on this but that is what matters. It is not a single number but a graph, you have to see the graph to really get an idea of what it is.

 

The manufacturers are starting to realize this and are now making some fairly inexpensive AT cut crystals with extremely low phase noise. They don't have great aging or great TEMPCO but they DO have great phase noise.

 

There are three common crystal oscillator configurations you will come across in digital audio:

 

XO - basic simple crystal oscillator, always uses an AT cut crystal, susceptible to the ambient temperature (remember that 25-35C) changes a fair amount over the years, has a huge range of phase noise from one model to the next. Anywhere from $0.35 to $25.

 

TCXO- Temperature Compensated Crystal Oscillator. Standard AT crystal with a temperature sensor that feeds a voltage variable capacitor across the crystal. In order to have a large enough "pull range" to handle large changes in temperature the crystal is modified so the frequency changes a lot with a given capacitance change. Unfortunately this radically increases the phase noise of the crystal. Thus TCXOs are about the worst clock you can use for digital audio. You get much better temperature stability, which you don't care about, in exchange for much worse phase noise which you DO care about. A very bad trade off. So if you see  a digital audio device with a TCXO, stay away.

 

OCXO Oven Compensated Crystal Oscillator. The oscillator sits in an oven that brings its temperature to 95C. Most writing you find on the net will say this is to stabilize the temperature, but the real reason is to bring an SC cut crystal up to 95C where its built in TEMPCO is zero. This gives extremely low frequency change with temperature, but the SC also has MUCH lower aging than the AT AND much lower phase noise than the AT. Thus the OCXO is great for both systems that require extremely low drift but also systems that require extremely low phase noise.

 

The problem is that OCXOs are not cheap, $100 and up (WAY UP). The cheapest OCXOs have about the same phase noise as the best AT cut XOs, for about 4 times the price. So for digital audio at least a low end OCXO is not particularly useful. You have to get in the $300 range to get OCXOs with significantly lower phase noise. As you go up from there you can get WAY better phase noise, but you really have to pay for it. So when looking at OCXO specs, all you need look at is the phase noise, all the stuff in PPB etc is irreverent. Don't waste money on getting the best in those specs. If a manufacturer just shows the PPB numbers and doesn't give phase noise, stay away.

 

Another thing that has been talked about is "atomic clocks". The "inexpensive" ones (less than $10k) are rubidium. These have EXTREMELY low long term drift, but very bad phase noise. There is NO reason to get one of these for digital audio. Sometimes a rubidium oscillator is paired with an OCXO, the rubidium "disciplines" the OCXO, this gives the best of both worlds, but if you spent the same amount of money on just the OCXO you could get much lower phase noise which is what matters.

 

In the next installment I'll go into frequency synthesizers and how recent changes are changing the landscape of clocks for digital audio.

 

John S.

 

 

John,

 

This is EXCELLENT...I look forward to the next installment!

 

mark

[zephyr24069]

On 11/8/2017 at 8:08 PM, rickca said:

There's a TCXO switch and router right here.  They also make an OCXO version.

http://thelinearsolution.com/

The OCXO should be superior no matter what to a temperature-compensated CO implementation. Oven-stabilized is the way to go IMHO...

[zephyr24069]

On 11/8/2017 at 6:40 PM, beerandmusic said:

 

below is some of the mumbo-jumbo going on in the 190+ page sotm thread....

 

Based on the Hans channel raving review of the SOTM, and all the talk about the SOTM lately, and other things i have read and heard about at recent audio meetings about clocking.....I am a believer in recent advancements inre clocks....

 

My main question is there something we should look for in marketing of future network players that would suggest they have embraced these new advancements in layman's terms.

 

For example, i know if i buy a new usb dac, that i would likely want it to have "galvanic isolation"....what could/should i look for in the purchase of a network player inre clocks?

 

 

 

------

austinpop

Clock Distribution and Termination

 

I wanted to update this thread with some information I have gleaned from recent discussions with Kenji Hasegawa-san of Cybershaft, May from SOtM, as well as @zephyr24069, who has provided me very valuable advice with regards to clocks.

 

Important Considerations with Clock Distribution

 

With all the interest in the SOtM sCLK-EX and master reference clocks, it's important not to lost sight of the basic tenets of clock distribution. At the frequencies of interest here, in the 10-54 MHz range, a clock cable acts as a transmission line. Transmission lines are defined by their characteristic impedance. For the SOtM, Mutec, and Cybershaft being discussed here, they are designed to use cables with 50 or 75 ohm characteristic impedance. It is also necessary for clock cables to be terminated at both ends, with the correct impedance (50 or 75 ohm). Proper termination is necessary to minimize reflections, which occurs on the transmission line if there is an impedance mismatch.

 

Point to Point

 

We have almost exclusively been discussing point to point clock connections in this thread. Examples of this are:

Cybershaft clock <-> tX-USBultra master clock input via 50 ohm BNC

Mutec Ref 10 <-> Mutec MC-3+ USB via 75 ohm BNC

sCLK-EX clock point <-> modded component (switch, router, sMS-200, etc) via 50 ohm SMB

In all these cases, it's important to know that the source and target connections are internally terminated. In scenario 1, for example, the Cybershaft output is terminated with a 50 ohm impedance, as is the master clock input of the tX-USBultra. I have confirmed this with both Cybershaft and SOtM. The Mutec Ref 10 uses an internal distribution amp to provide 8 independent clock outputs, but each of these is internally terminated, and expects a properly terminated target as well.

 

In case you are wondering, SOtM handles the termination internally when you send them components to mod. For each sCLK-EX clock point, both the source and the target are appropriately terminated.

 

But what about the case, where you're a cheapskate like me, who doesn't want to spring for multi-output clocks like the Ref 10 or the upcoming SOtM sMS-OCX10? Can I use a single high-quality clock like the OP-14 and drive multiple outputs?

 

This is where things get interesting.

 

Daisy Chaining

 

Yes, you can daisy chain, but you have to consider the entire length of the cable across the chain as a single transmission line, and you must have termination on each end equal to the characteristic impedance of the cable. For this to work, you have to have devices whose inputs can be selectively configured to be terminated (i.e. have an input impedance of 50 or 75 ohms as needed), or unterminated, in which case the input impedance of the input is very high. In the quote from Kenji below, he calls this "Hi-Z."

 

Regarding distribution of clock T plugs, it is possible under certain conditions.

For example, in the case of a daisy chain

 

1.

Clock output(50ohm) ------T--------T--------T-(50ohm)Terminator

                                      (Hi-z)   (Hi-z)    (Hiz)

                               Device1   Device2   Device3

2.

Clock output(50ohm) ------T--------T- non

                                     (Hi-z)   (50ohm) 

                                  Device1   Device2 

 

( Hi-z = High impedance) 

Be sure to have one 50 ohm impedance at the end and the other relay point must be Hi-z.

Therefore, T plug can be used only when the clock input can be set to Hi-z.

For reference, I attach a connection diagram of CH presision CH and D1. Please be aware that each relay point is Hi-z.

If, for example, there are 50 ohm points in two places, the impedance will be 25 ohm, the current will overload and destroy the output circuit of clock due to long-term use.

SOtM tX - USBultra can not set the input impedance to Hi - z. So SOtM tX - USBultra does not use T plug in the situation.

Also, due to the use of T plugs, noises of each device can interfere and clocks that can be transmitted can not be transmitted. I do not recommend using T plug.

The best way to distribute the clock is to use an ultralow noise distribution amplifier.

However, I could not find a low-cost ultra low noise distribution amplifier of less than $ 1000 from the world market.

We are developing to sell low-cost distributors by next spring.

 

Pay particular note to his point about overloading the output circuit of the clock due to it "seeing" a lower impedance than designed. The second consequence of not paying careful attention to termination is that impedance mismatches cause reflections, that degrade the signal integrity of the clock, which negates the whole point in the first place.

 

The other point to note is that he discourages the use of daisy chaining, and promotes the use of an ultra low noise distribution amplifier. I got the same feedback from May and I believe (although I'd have to check) Mutec also discourages daisy chaining. With the Ref 10, it's self evident due to the provision of 8 clock outputs!

 

So what does this all mean

Don't daisy chain, if at all possible. Even without the other complications, daisy chains necessitate longer cable lengths, which degrades clock quality

If you must, design your chain to ensure that all devices in the chain can be configured to be unterminated, and that there is proper termination of the right impedance on both ends, using tees and terminators.

At least for the SOtM Ultra components, SOtM does not currently provide a way to switch between terminated and unterminated. I'm pretty sure that upon request, they can configure a device whichever way you want it.

For me, this means that if I get to the point where I have more than one component with a master clock input that I want to drive, I'll want to spring for a Ref 10 or OCX-10. Ouch!

This write-up by austinpop is excellent....

[beerandmusic]

18 minutes ago, zephyr24069 said:

This write-up by austinpop is excellent....

I forwarded the info to LUMIN....who knows, maybe they will listen, maybe they won't, but i would like to see more competition in network players.

 

 

[beerandmusic]

On 11/9/2017 at 12:34 PM, jabbr said:

 

The take home message should be that it is the phase error plot which is the important spec, not whether there is an “OCXO”. If you would like 1 single number to look at: take the dB/Hz @ 1Hz offset ... many phase error plots don’t give you 0.1Hz.

 

Turns out that this number correlates with 1/f noise. That’s why good power supplies are important because if the power supply has high 1/f noise, then the clock will suffer. The power supply can’t just be “low noise” but has to have low 1/f noise.

 

In any case if I were to rip out and replace the clock that @PeterSt put in his NOS1a DAC and replaced it with the SoTm sclk-ex then I would be an idiot. I have no idea if the clock is an OXCO.

 

To summarize can you clarify in layman's terms....

 

We should want a power supply with "LOW" 1/fnoise and a clock with "LOW" phase error  (that OCXO is not important)...

 

Define "LOW" for clock and power?

Is this a spec they will market?  Should they if it is important, and justify cost?

 

 

 

 

 

 

 

[jabbr]

2 hours ago, beerandmusic said:

 

To summarize can you clarify in layman's terms....

 

We should want a power supply with "LOW" 1/fnoise and a clock with "LOW" phase error  (that OCXO is not important)...

 

Define "LOW" for clock and power?

Is this a spec they will market?  Should they if it is important, and justify cost?

Yes.

 

The Crystek 957 is a reasonably low cost XO which publishes phase error plot... admittedly this may be under ideal circumstances (e.g. great PSU) but say -100 dBc/Hz @ 10hz is a reasonable definition for "LOW" in these days +/-

 

For analog circuits including PSU: http://www.analog.com/en/analog-dialogue/articles/understanding-and-eliminating-1-f-noise.html

I don't know that PSUs typically supply this number.

 

But yes, if you are looking to "upgrade" something, both of these measures are ones to improve.

[Superdad]

17 minutes ago, jabbr said:

For analog circuits including PSU: http://www.analog.com/en/analog-dialogue/articles/understanding-and-eliminating-1-f-noise.html

 

That's a very good article Jonathan, thanks!  (What makes it "good" is that I actually understood most of it. )

[zephyr24069]

3 hours ago, beerandmusic said:

 

To summarize can you clarify in layman's terms....

 

We should want a power supply with "LOW" 1/fnoise and a clock with "LOW" phase error  (that OCXO is not important)...

 

Define "LOW" for clock and power?

Is this a spec they will market?  Should they if it is important, and justify cost?

 

 

 

 

 

 

 

Phase noise specs at 1Hz, 10 Hz and 100Hz are all significant regardless of whether we are talking about a quartz crystal or rubidium oscillator.  There is a comment in a slightly new post on this thread that says -100 dBc/Hz at 10 Hz is considered 'low' in terms of phase noise; while it is, higher quality OCXO and Rb-based implementations for better quality audio result in the range of -115 to -130 (and better) dBc/Hz @10Hz are the norm these days.  Examples are Esoteric (both with their OCXO or oven-stabilized Rb implementations sourced from various high quality vendors), Cybershaft (OCXO only), BVA and others; not meant to be an exhaustive list,...just examples.

 

Another key point of analysis is the Allan Deviation as well as Allan Variance behavior of such a circuit/overall clock assembly. See this link for more details;

 

https://en.wikipedia.org/wiki/Allan_variance

 

Here are the phase noise specs. of the clock I am using today for 10 Mhz which has proven to be amazing for output of high-quality and very musical results;

 

1 Hz	-121.8 dBc/Hz
10 Hz	-136.4 dBc/Hz
100 Hz	-142.7 dBc/Hz
Calculator rms phase jitter 1Hz-100Hz; 29 Fs

 

My prior clock (also excellent for high-end musical output) was;

-114.7 dBc/Hz

-133.5 dBc/Hz

-141.6 dBc/Hz

Calculator rms phase jitter 1Hz-100Hz: 48 Fs

[jabbr]

That's all just terrific. Of course the transmission line length has effects: https://dl.cdn-anritsu.com/en-au/test-measurement/files/Technical-Notes/Technical-Note/MP1800A_EE1100.pdf

 

of course this is well documented. Now what is the phase error when the clock is converted to something usable for audio? ... and of course the question about the phase error of the DAC logic itself or whatever is being clocked. 

 

the problem is that the clock can't improve the phase error of the clocked circuit better than the best it can be, just as a low 1/f PSU feeding a high 1/f transistor can't make the transistor better than it can be...

 

issues such as: https://ac.els-cdn.com/S1877705812008636/1-s2.0-S1877705812008636-main.pdf?_tid=e19ed90e-ce44-11e7-834b-00000aacb35e&acdnat=1511218278_747e5edcf586778a9501c9ff0176ae6c, https://www.researchgate.net/profile/Rui_Aguiar/publication/4297963_Predicting_noise_and_jitter_in_CMOS_inverters/links/00b4951bed2dfe7cc4000000/Predicting-noise-and-jitter-in-CMOS-inverters.pdf

[beerandmusic]

1 hour ago, zephyr24069 said:

Phase noise specs at 1Hz, 10 Hz and 100Hz are all significant regardless of whether we are talking about a quartz crystal or rubidium oscillator.  There is a comment in a slightly new post on this thread that says -100 dBc/Hz at 10 Hz is considered 'low' in terms of phase noise; while it is, higher quality OCXO and Rb-based implementations for better quality audio result in the range of -115 to -130 (and better) dBc/Hz @10Hz are the norm these days.  Examples are Esoteric (both with their OCXO or oven-stabilized Rb implementations sourced from various high quality vendors), Cybershaft (OCXO only), BVA and others; not meant to be an exhaustive list,...just examples.

 

Another key point of analysis is the Allan Deviation as well as Allan Variance behavior of such a circuit/overall clock assembly. See this link for more details;

 

https://en.wikipedia.org/wiki/Allan_variance

 

Here are the phase noise specs. of the clock I am using today for 10 Mhz which has proven to be amazing for output of high-quality and very musical results;

 

1 Hz	-121.8 dBc/Hz
10 Hz	-136.4 dBc/Hz
100 Hz	-142.7 dBc/Hz
Calculator rms phase jitter 1Hz-100Hz; 29 Fs

 

My prior clock (also excellent for high-end musical output) was;

-114.7 dBc/Hz

-133.5 dBc/Hz

-141.6 dBc/Hz

Calculator rms phase jitter 1Hz-100Hz: 48 Fs

 

I am just curious.....how do you and jabbr and so many other "electronic geeks" on this forum know all this "stuff"....

are you audio engineers?  what are your products?  Of just DIY type guys? 

 

(grin)

 

 

 

PS- thanks

 

 

[austinpop]

2 hours ago, jabbr said:

That's all just terrific. Of course the transmission line length has effects: https://dl.cdn-anritsu.com/en-au/test-measurement/files/Technical-Notes/Technical-Note/MP1800A_EE1100.pdf

 

of course this is well documented. Now what is the phase error when the clock is converted to something usable for audio? ... and of course the question about the phase error of the DAC logic itself or whatever is being clocked. 

 

the problem is that the clock can't improve the phase error of the clocked circuit better than the best it can be, just as a low 1/f PSU feeding a high 1/f transistor can't make the transistor better than it can be...

 

issues such as: https://ac.els-cdn.com/S1877705812008636/1-s2.0-S1877705812008636-main.pdf?_tid=e19ed90e-ce44-11e7-834b-00000aacb35e&acdnat=1511218278_747e5edcf586778a9501c9ff0176ae6c, https://www.researchgate.net/profile/Rui_Aguiar/publication/4297963_Predicting_noise_and_jitter_in_CMOS_inverters/links/00b4951bed2dfe7cc4000000/Predicting-noise-and-jitter-in-CMOS-inverters.pdf

 

I envy your grasp of the technical aspects of this stuff. Bravo!

 

Like @zephyr24069 's previous reference clock - I use a reference clock with phase noise of -114 dBC/Hz @ 1Hz. It drives the frequency synthesizers in the SOtM sCLK-EX, for which phase noise specs have never been published - boo, hiss!

 

I do understand that what the Lord giveth with low phase noise at the OCXO source, He taketh away with transmission line length, reflections, etc. Still it is important to understand that the net effect is still an outstanding improvement in sound quality!

 

It IS important to understand the nuances on clock distribution, as well as the importance of excellent PSUs for clock generators, and I have written about it - albeit not at this level of expertise. But it should not detract from the key point that clock improvements of this nature do improve the SQ tremendously.

 

I can't stop listening to my system these days - it is so scary good!

[PeterSt]

4 hours ago, beerandmusic said:

I am just curious.....how do you and jabbr and so many other "electronic geeks" on this forum know all this "stuff"....

are you audio engineers?  what are your products?  Of just DIY type guys?

 

I have a secret for you : the normally educated EE doesn't know much about these things. Say that this is because audio(philia) is not a real target (there is no education for it). Telecommunications come close but still requires quite different skills (but at least they have the measurement gear).

So it it experience, experience, experience. Day in day out don't do much else. And I tell you : if i'd investigate once per year the status of "oscillators", it could take me 2-3 days before being up to date again (and possibly learn that nothing is new). Or, "know" from the figures zephyr just gave, what product he is talking about.

 

Didn't I tell you to start investigating and come back in a couple of months ? Well, that is the answer to your question.

And it all springs from a hobby. A profession it is for almost nobody. 

[Ralf11]

it is still foundational for excursions into audio

 

similar to an aerospace engineer who retires early and starts making carburetors for old sports cars (they are really good carburetors too!)

[JohnSwenson]

I spent years doing this in my day job in the semiconductor business. Take one of the most difficult, 64 arm processors, a large complex switch fabric handling very high speed serial streams between processors. The clocking for this was incredibly tight, they called me in to design and implement the clock network for this.

 

As with everything in this space the power for this is very critical, no way can you meet the spec using the "normal" power distribution network used for the "processing" elements. It takes a whole separate network driven by its own ultra low noise LDOs, special very low noise PLLs . The whole thing is a differential clock distribution where the whole thing has to be balanced to within a fraction of a micron. Complex shielding systems are employed to protect the clock signals form other stuff going on nearby. The whole thing including the power networks and PLLs has to be spiced multiple times to make sure  it is doing what it is supposed to.

 

This took a lot of work to get it right (many months of work), but it worked perfectly the first time.

 

One thing very different about doing this on a chip rather than a board, there is no way to go in and analyze the signals on the chip, measure the jitter etc., all you get is if the system works or not. So it generally tends to be over engineered to make sure it will work, but that makes it even harder to design.

 

So yeah I actually did learn this stuff on the job. There was never any official learning for this stuff, you pick it up as you need it wherever you can. I wound up getting different parts from different people over time as I worked on different projects. There was no course on "ultra low jitter clock networks for CMOS chips". Just bits and pieces from different sources and putting it all together to make things work.

 

John S.

[jabbr]

15 hours ago, beerandmusic said:

 

I am just curious.....how do you and jabbr and so many other "electronic geeks" on this forum know all this "stuff"....

are you audio engineers?  what are your products?  Of just DIY type guys? 

 

(grin)

 

 

 

PS- thanks

 

 

 

There are many ways to learn. The real clock stuff has traditionally not been audio related rather radio, GPS, satellite etc. there is a "time-nuts" mailgroup devoted to these topics. http://www.leapsecond.com/time-nuts.htm

 

This thread is absolutely awesome: http://www.diyaudio.com/forums/digital-line-level/261651-tempered-master-clock-building-low-phase-noise-jitter-crystal-oscillator.html and the primary way I became more interested in the details of clocks for audio.

 

There are also many resources online as well as textbooks.

 

A big interest of mine has been the relationship between 1/f noise and 1/f offset phase noise.

 

I try to back up statements with online references.

[d_elm]

A book about jitter and phase noise available in Jan 2018 may be interesting.

 

https://www.cambridge.org/core/books/understanding-jitter-and-phase-noise/E7A2EBDE1575B47793F67FCA25218614

[zephyr24069]

16 hours ago, beerandmusic said:

 

I am just curious.....how do you and jabbr and so many other "electronic geeks" on this forum know all this "stuff"....

are you audio engineers?  what are your products?  Of just DIY type guys? 

 

(grin)

 

 

 

PS- thanks

 

 

First off,...I'm an audio and music nut and don't work for any manufacturer and never have. My day job is something sales related

and extremely technical. I only wish I could make a career of high-end audio as it is so much fun but alas, that is not the case...

 

The short answer is that I knew NOTHING about any of this in 2006 when I bought into the marketing hype of 'ppb is better than ppm' with the purchase of a rubidium oscillator-based Esoteric G-0s master clock after noting that JVC, Esoteric and others made heavy use of master clocks as did many other studios in their remastering and mastering projects.  I happily used the G-0s for 3+ years thinking I had the 'most accurate' clock and musical playback; everything was arguably very precise and rock solid in many ways but it was a touch

analytic in sound after adding this clock,...that stated I rolled with it for 3+ years with the thought that I was supposed to like it as the

'ppb' marketing hype was some very strong mojo !

 

Later in 2010 for a variety of reasons I sold my G-0s to an audiophile in Mexico that is still happy with it and using it to this day.

A year later, I bought an Esoteric 'less precise, i.e. 'ppm' rated' G-03x, based upon an OCXO implementation.  A funny thing happened

in that I like the musical playback far better and it was much more realistic than with the Rb-based much more expensive G-0s. Surprise then turned to frustration as I was wondering how I could have bought into hype that cost me roughly $8K more years

before and relative to 2006/2007 'bucks' was actually more costly.  

 

In 2011/2012 I started to read anything I could find on word clocking, master clocking, types of clocks, phase noise and jitter

and a variety of other topics. I also talked and corresponded with some very smart people in high-end audio and also

telecom/satellite industries both of which make heavy use of sophisticated clocking solutions for different reason.

Over the last couple of years I spent time off and on continuing to try to learn more and experiment with different types

of clocks (Rb, OCXO, dual), cables (both 50-ohm and 75-ohm) from a variety of sources/manufacturers and to talk in-depth

with several vendors of all this for audio to learn more on the subject and its interrelationship to the mastering,

remastering, playback of the  music that drives me...

 

This has been a labor of love so to speak and I definitely have not mastered the topic.  I like to share what I've learned

and learn from others.  It's clear though that I'll be busy learning more from anything John S., Elberoth, Grimm, Cybershaft,

and others have to say or write on the subject for many years to come.

 

FWIW...

 

[zephyr24069]

1 hour ago, d_elm said:

A book about jitter and phase noise available in Jan 2018 may be interesting.

 

https://www.cambridge.org/core/books/understanding-jitter-and-phase-noise/E7A2EBDE1575B47793F67FCA25218614

Very cool...thank you!  This will be one very interesting book!

[austinpop]

12 hours ago, PeterSt said:

 

I have a secret for you : the normally educated EE doesn't know much about these things. Say that this is because audio(philia) is not a real target (there is no education for it).

 

As a nominal PhD/EE (although my research was in networks) I would agree with you wholeheartedly!

 

On the whole EE subject - I should clarify that Electrical Engineering (what I did in my Bachelor's) and Electronics Engineering are actually somewhat different fields. And then, when you get into audio engineering, I'm not sure that's even a degree program anywhere. This is why you get into this realm of opinions. The link between engineering metrics and sound quality is still tenuous, even after all these decades. Rather than causality, we often only have correlation.

 

This is a fertile breeding ground for the cult of the personality.