now we´re getting somewhere, finally we have something, that we ca agree upon, nice...
and actually i don´t know how to measure jitter exatly (...guess, i´m just to lazy to look it up...), all i´ve
heard ist, that it is fairly simple whe you have a crazy expensive osciloscope.
fair enough...
...hmm, so you don´t have a suitable Audio-anayzer...? but i thought....
... never mind. doesn´t matter...if you can find someone, who is can do the measurements, i´m happy to provide the testobject, i have a spare
rpi4, which i currently don´t use, i can ship it sweden or anywhere in europe, i´ll even include some sd-cards
with ready-to-use systems on it.
i´m confident, we can come up with a reasonably valid test setup.
no joke, i´m serious, it´s diyAudio afterall, you´re supposed to do somethting, for yourself, and not just talk about it...
what do you think ??
cheerio,
Mirko
Hi HenrikEnquist, M_Balou,
If you can arrange a measurement environment and measurement procedure, I would be very grateful.
I would like an objective assessment by a third party.
Of course, I'm willing to help.
Let me know if there is anything you need.
We can ask for help from members who agree with our ideas.
If you can arrange a measurement environment and measurement procedure, I would be very grateful.
I would like an objective assessment by a third party.
Of course, I'm willing to help.
Let me know if there is anything you need.
We can ask for help from members who agree with our ideas.
The sigma-delta modulator in most oversampling DACs ₍the actual place where DA conversion occurs) runs at master clock frequency, multiples (e.g. 8x) of the bitclock of I2S. Either the DAC (board) has its own master clock (then RPi's I2S is running in slave mode), or the DAC (chip) can/must PLL its master clock from the incoming low-frequency BCLK generated by RPi's I2S (or PLL an external clock to run synchronously with the multiple of BCLK). The PLL may produce a very different jitter than the incoming BCLK. This signal is not available on DAC pins.
IMO measuring the effect in the final analog signal is a valid method, especially for comparing various setups to find the one with best result in the analog domain.
Measuring clock jitter directly takes "just" a proper device, I just received a measurement report from Silicon Labs measuring rms jitter of 93fs.
IMO measuring the effect in the final analog signal is a valid method, especially for comparing various setups to find the one with best result in the analog domain.
Measuring clock jitter directly takes "just" a proper device, I just received a measurement report from Silicon Labs measuring rms jitter of 93fs.
Attachments
Nono I don't have an Audio Precision analyzer. I got to use one at the university for a few hours to measure an amplifier I had built. That was already 15 years ago. I don't know anyone at that department today, so I don't know if they still have the unit. I would bet that they still do (or more likely a newer one). I'll investigate. There is a high risk that it becomes a long unmanagable chain of asking someone who knows someone who knows someone...now we´re getting somewhere, finally we have something, that we ca agree upon, nice...
and actually i don´t know how to measure jitter exatly (...guess, i´m just to lazy to look it up...), all i´ve
heard ist, that it is fairly simple whe you have a crazy expensive osciloscope.
fair enough...
...hmm, so you don´t have a suitable Audio-anayzer...? but i thought....
if you can find someone, who is can do the measurements, i´m happy to provide the testobject, i have a spare
rpi4, which i currently don´t use, i can ship it sweden or anywhere in europe, i´ll even include some sd-cards
with ready-to-use systems on it.
i´m confident, we can come up with a reasonably valid test setup.
no joke, i´m serious, it´s diyAudio afterall, you´re supposed to do somethting, for yourself, and not just talk about it...
what do you think ??
cheerio,
Mirko
I have a RPi 4, but no i2s dac.
This gives me an idea... I have been thinking one step too far. I think it would make sense to start off by checking for any difference in the i2s signals from the raspberry. No dac needed, and the whole thing gets much simpler. That means we "just" need one of those highly specialized super expensive instruments.
We have a really nice spectrum analyser in the RF lab at work. I don't know if it's suitable for this exact task, but I'll ask! And if I ask nicely I can probably play with it as much as I want.
See, here is the problem, that chart starts at 100 Hz, how about 30 Hz? or 10 Hz or maybe even 1 Hz? Thats where many clocks fall apart. They look nice at 10kHz (thats the most common frequency where a time (ps, fs) is given for.
But it's the close in noise that is also a problem. It causes sidebands. And this is what is so dangerous with "femto-seconds" figures...
To measure where it really matters is a really hard task and requires serious equipment and knowledge.
If your are a designer you need to be able to measure clocks, if you are a "customer", you are interested of the analog side i.e. sideband, distorsion, noise etc.
Then there it seems to be music correlated jitter that creates the biggest problem - random jitter not so much. To add to the complex, different D/A topologies (R2R, D/S...) have different sensitivity to jitter and just to end the whole thing... different chip products have sometimes built in strategies to clean out jitter as mention above.
It's a mine-field...
//
Last edited:
Something like Compare: Phase Noise Measurement | Keysight would do
None of these are suitable for audio.
//
Actually the Silicon Labs technicians modified my clock chip configuration to reduce the output integer divider for the DAC/ADC master clocks to get above 100MHz where the measurement does not drown in the instrument noise, from the required 20-40MHz. So really not for the tens of MHz master clocks.
I would like to end to say again that the behaviour of the Pi hardware ert. clock (MASH?) is a pathological behaviour and is most unsuitable for serious audio equipment. If one has any ambition to create a well behaved product and to not start the whole project with a big problem - I would avoid Pi as the Covid19. The make excellent execution platform for signal processing but don't connect them to any audio HW.
//
//
Did you get phase noice reading for say 10Hz of your clock? Have you read Andreas thread about low jitter clock? - its often on the front page!
//
it´s nice that you agree with me in the end, leaving out the dac does make it easier, i approve!
and it would be awesome if you could use the analyser at work.
and it makes my quite envious, because we have some fancy equipment in our workshop, too.
and in the past it was totally ok to bring in your stuff and work on it, you could even take home tools
or other equipment over weekend. but some years ago we got new bosses and now we´re noz
allowed to have any fun at work anymore...
I'm thinking about the first stage of measurements, that would be to measure the bit clock without any dac connected.
The thing to compare would be symphonic-mpd vs a standard distro running mpd.
This only makes sense if I can run the standard distro without mash messing up the clock.
So then my question is, is there any sample rate where it will generate the right frequency without using mash?
The thing to compare would be symphonic-mpd vs a standard distro running mpd.
This only makes sense if I can run the standard distro without mash messing up the clock.
So then my question is, is there any sample rate where it will generate the right frequency without using mash?
I still do not understand why a PC should generate the audio clock. It provides HW device slaved to bit clock which can be simply generated from precise master clock located next to the DAC/SPDIF transmitter. The DAC should clock the sigma/delta with the master clock directly, not PLL the multiple from the I2S bit clock.
I agree, it should not.
But the claim was (if I understood correctly) that disabling MASH was only part of the improvement, and the other software changes somehow improve the quality of the i2s signals. I find that extremely unlikely, but it's something I may be able to measure (if I can get rid of MASH).
There isn't much point in measuring bit clock jitter when the Pi is slaved to a dac.