You make a good response. However, it seems that you may be overlooking the key implementation details. To have low conversion jitter, the reclocking signal must itself exhibit low jitter. For that, the oscillator circuit on which the reclocking signal is derived must feature low phase-noise. All oscillator circuits exhibit some amount of phase-noise, so the notion of what's low enough phase-noise, and so, low enough clock-jitter depends on the question of audibility threshold. Even so, it still will some non-zero amount of jitter.
What can be worse is that, unless a DAC and it's bit-source utilizes some form of closed master-slave clock control loop, it must rely on some form of phase-locked loop for clock recovery. Which adds another layer of jitter on top of that caused by the phase-noise of it's oscillator circuit.
As you pointed out, what constitutes sufficiently low jitter depends not only on it's amplitude, but it's spectrum. Also on, to what degree the jitter is correlated with the signal it is converting. Easily, not a trivial task to satisfactorily address.
He, I wrote a quite long post about asking you Magnus to describe the cause and effect in the chain of nodes that you are thinking about and how "noise" is transfered. Maybe first define noise (density) and the difference between 2 different clocks in a more technical way than just "noisy"... but I scrapped it...
//
//
Back to the original question: found this clock on mouser, looks decent (not as good as CCHD-575 but can't find that in 25Mhz). It even mentions Ethernet (up to 100G) and PCIe as suitable applications.
https://www.mouser.se/datasheet/2/268/MX55-57-Low-Jitter-Crystal-Oscillator-DS20005972A-1374963.pdf
Now if I only knew how to connect it (I assume I have to bypass the xtal and its supporting electronics) 🙂
https://www.mouser.se/datasheet/2/268/MX55-57-Low-Jitter-Crystal-Oscillator-DS20005972A-1374963.pdf
Now if I only knew how to connect it (I assume I have to bypass the xtal and its supporting electronics) 🙂
Both parts of this statement are wrong. Accuracy is a specification of frequency, which breaks into absolute frequency accuracy, long and short term frequency stability. It is not affected by external electronic noise.
Part 2: Ethernet is inherently noise immune. It has to be to work reliably. Switches do not introduce noise, and actually cannot. It's a balanced differential transmission system.
While technically true, it's not about if they do have phase noise or they don't, it is entirely about how much. You've never begun to state that specification, so the discussion is now meaningless.
Nope, wrong again, for reasons previously stated.
How much less phase noise does it have compared to the stock clock?
Frankly, that's only one very minor gap in your knowledge here.
And thus, is not worth discussing. If you can't define a threshold (and you can't) there's nothing to talk about.
So what? Every system has a non-zero amount of noise an distortion, and frankly, at levels orders of magnitude more audible than jitter. Just measure the distortion, any kind, of any speaker, or the ambient noise of any listening room.
This is outside the discussion of phase noise and jitter caused by network transmission.
But trivial to check on. You may never know how much is too much, but you can very, very easily check to see if it's there, and make comparative tests to see if anything is improved after changes. What you cannot do is depend on sighted, biased, subjective listening tests to determine this.
Yes, gnd to shell.
Not sure what you mean by having 2 grounds and moving them. I'd, for this step, change nothing and just solder the ground to the shell of the crystal.
After that, change the battery supply ground to another place, most likely close to the regulator you think draws most current (if there's 2, otherwise just to the one regulator there is).
The ic that needs the clock, usually has an input and an output. If you can find the datasheet that's be of great help. Things you need to know are:
What pin should be connected to the external oscillator?
What's the maximum input voltage of the clock input?
What's the best ground connection for the external clock?
Last one's easy: take the earlier described and used pad
Maximum voltage is usually normal cmos level (3v3), but best to make sure by a datasheet.
The right input: usually you need the datasheet as well. You could blow the input if connection is made at the wrong place.
Curious if soldering the crystal shell makes any difference in your opinion. Good luck!
Before and after what? And specifically what aspect?
You'll have to be more specific. Are you looking for before/after of the goofy mods the OP is trying to perform? I've already said how that can be done at no cost. If you're looking for data transmission integrity, you need something like a managed switch that monitors and keeps logs.
I would however welcome anyone to measure anything here. It would be a giant leap forward for the thread.
I only have one ground, but I take that from the motherboard/PCIe through the pins on the connection. But I would like to cut those ground connection and in effect make the card "floating" (ground will be connected to the voltage regulator board but thats it).
But that might mess with the PCIe signalling (or it might not).
You don't happen to live in Stockholm, Sweden? If so, you are welcome to come listen to my "goofy" mods and compare them to a regular switch. I can promise you one thing; you would be surprised 🙂
One thing you should definitely NOT do is cut ANY ground going from the motherboard to the card.
I'm not gonna explain, too many options what might break down. In some cases even the cpu might get damaged. Please don't.
Have a look at what I attached...
1. the photo shows how to position and wire up the LDOVR 1A 3.3V regulator
a) solder solid core copper wires to +3.3V DC rail and to the ground fill.
b) slide (open the eyelets first!) the LDOVR reg on to these wires, and then secure the reg on that big black IC. Solder the wires to be as short as possible.
c) use one of these: 35SEPF22M (22uF/35V). Solder this cap to OUT+ and OUT- outlets on reg board. Keep the legs as short as possible. You could actually, slide both the capacitor legs and the copper wires at the same time, through the eyelets, and then solder the whole lot in one go. The eyelets diameter is large enough. This will ensure the shortest wiring. NOTE: that LDOVR reg needs to see some capacitance at its output to perform at its best. Just be sure to keep the capacitor legs as short as possible.
2. Once you move to DAC xtals/ oscillators, always go with NDK; much better than the competition. See the document I attached. Also, go straight to SDA types, which have much better phase noise at crucial frequencies of 1Hz, and below. They are difficult to find in quantities of 1, but search the net and you'll find them. See the pdf attached!
The above will keep you occupied for a while. Continue playing, experimenting, reading and...learning. The mod I suggested will give you very good results. Let us know what you think...
Good luck,
Nick
PS
I just saw mterbekke's post... do not cut any ground copper fills...
Attachments
Last edited:
Oh boy. After all these remarks, you still think most of these good intended remarks about what you can not do, shouldn't do, have to do, has no point, convincing you of the meaningless "because that's how networks work" and trying to "educate" you, you think you could prove to them the results by ear?
Haha, of course not!
Never trust your ears!
The OP tried to present this DIY-endeavour as a project he was interested in, just for the fun of it. Instead of helping the guy out with knowledge how to approach, implement and execute best and do the things he wants to accomplish, all I read are remarks from guys who either talk about their own opinion (theoretical standpoint, this is practice), ridicule, see black and white (isolation and balanced signalling mitigate all forms of any "noise"(whatever noise means)) to measurements only possible if access to extremely expensive equipment, which can only generate reliable data if operated by an expert?
This is clearly not the intent, nor fruitfull for any of us here and I don't see the point.
If the OP's goal would be, for instance, to make a commercial product and go about like this, advertising it as the best NIC ever and charge 100's of dollars/euro's for ir, making snake oil, I get the idea of ridiculing somewhat.
Either way, this should be a place to learn, experiment and help out.
You should.
Yet, he ignored all the advice by people who understand the inner workings of the physical layer of an ethernet interface and the electronic theory on how these devices function.
True, but it doesn't mean you can ignore the laws of physics, electronic circuit theory and standards on how equipment and devices need to conform with to communicate over ethernet networks simply because it doesn't conform to a personal belief system.
The OP would have saved us all a lot of time by just understanding the information that was being imparted rather than refuting it.
The IEEE 802.3 standard is your friend.
Sometimes it helps to know the rules before you break them, other times you surprise yourself.
Have you played around with ferrites before?
Have you played around with ferrites before?
Cheers for the help, I will do the capacitor fix but think I will leave the reg board as it is (its so nicely glued there on the top) 🙂 I use solid OCC coppar (Neotech), leftover from some cables I made, awg 18, so its not like an inch of extra cable will matter much
I use an NDK already, in my DAC (RME ADI-2 has this one NZ2520SDA(Ultra Low Phase Noise Type)/Crystal Clock Oscillators/NDK ) but I could not find the SDA in 25 Mhz, but NZ2520SD (without A) can be ordered from Japan though, see link below:
NZ2520SD-25.000000M-NSA3449E 日本電波工業 | タイミングデバイス | 水晶発振器(SPXO) | Nihon Dempa Kogyo (NDK) - チップワンストップ 電子部品半導体通販サイト
Last edited:
No, but I read that they can help. I power my digital stuff from batteries with short cables so don't think they will do much in that case.
- Status
- Not open for further replies.