Thanks IVX for your feedback. I will consider your suggestion about modifying the existing model, however, from looking at pictures on the internet the oscillators are really close to the ES9822 ... Might I ask you if you had some specific technique in mind for removing the NZ2520s? I was thinking about a hot air gun but I reckon it would be quite challenging to remove the oscillators without also heating the ES9822 ...
Regarding the thread I linked to it seems to me that it started as an illustration of the difference brought about by improving the clock jitter/phase noise and then, as bohrok2610 discovered the differences between the decoupling capacitors the focus shifted. Anyway, my intention was only to illustrate the background for my question ...
Cheers, Jesper
Maybe I don't understand something here. You are using the Cosmos to analyze the jitter of a DAC I think. If you are looking at the analog out the jitter of the ADC is not likely to be an issue as long as its good. I may experience use a long sample at high sample rate. Use averaging, chherent averaging if possible. Use a non related clock at a different sample rate. This avoids any injection lock or picket fence issue that would hide synchronous sidebands or jitter.
A passive notch should enhance the resolution by increasing the dynamic range.
I think measuring at the analog output is the only relevant place since all the internal issues of clock distribution etc. are present. And those are the real system performance limits that a better clock won't improve.
A passive notch should enhance the resolution by increasing the dynamic range.
I think measuring at the analog output is the only relevant place since all the internal issues of clock distribution etc. are present. And those are the real system performance limits that a better clock won't improve.
If the purpose is to look at close-in phase noise (e.g. +-10Hz around the fundamental) the clocks of dac and adc should be coherent. Otherwise FFT spectral leakage will result in high noise skirts that effectively hide all differences between clocks or Vrefs. That is why Cosmos ADC as is does not really work for this type of measurement. But of course for wide band J-test measurements clocks can be non-coherent.
With coherent clocks it is possible to see differences in close-in phase noise between dac chips.
E.g. here is ES9038Q2M (orange) vs. AK4493 (green) with more or less same Vref and clocking.
Attachments
Last edited:
What you are looking at there is the additive phase noise of the internal stages of the DAC and the clock distribution to drive both DAC and ADC with a common clock. The jitter in the source clock will cancel out to some degree. Spectral leakage will also come from different window functions and may be real and hidden with a common clock. I may try this if I can find the time to see what happens.
Its interesting but I'm not convinced its significant, especially if the first bit of noise is at -130 dB. How many samples in the fft? How many averages? Effective bandwidth? Could the analog stages after the DAC introduce the phase noise you see?
Its interesting that the AKM is a little better than the ESS.
I doubt its physically possible to maintain that phase noise with transducers and air in the chain. It would be an interesting experiment.
Its interesting but I'm not convinced its significant, especially if the first bit of noise is at -130 dB. How many samples in the fft? How many averages? Effective bandwidth? Could the analog stages after the DAC introduce the phase noise you see?
Its interesting that the AKM is a little better than the ESS.
I doubt its physically possible to maintain that phase noise with transducers and air in the chain. It would be an interesting experiment.
gentlevoice, if you can't trust your soldering tool, you can try to cover the PCBA with aluminum foil with the window over the OSC.
You can also try Chip-Quik:
https://www.chipquik.com/store/index.php?cPath=2200&osCsid=v8ji0g1hnsijo5b7fuv9g4cr61
https://www.amazon.com/chipquik/s?k=chipquik
That stuff can work like magic. We used it to remove clocks on dac boards way back when.
There are a number of youtube videos on how to use it.
https://www.chipquik.com/store/index.php?cPath=2200&osCsid=v8ji0g1hnsijo5b7fuv9g4cr61
https://www.amazon.com/chipquik/s?k=chipquik
That stuff can work like magic. We used it to remove clocks on dac boards way back when.
There are a number of youtube videos on how to use it.
A modest propopal: why can't the next addition to the Cosmos family be a USB soundcard, i.e. a DAC and ADC that share the USB interface and the isolator? This would result in overall cost saving (one bridge and isolator saved) and improved performance (no ground loop and common clock)?
Hi all ... many thanks for your helpful replies 
However, I do not want to digress too much from the topic of this thread (IVX's Cosmos ADC, etc.) so just briefly:
From pictures on the internet it looks as if the Cosmos oscillators are 45/49 MHz versions (enabling 768 kHz Fs I suppose). Given that I would like to use Andrea Mori's 22/24 MHz (and maybe lower frequency) oscillators I would then have to include clock doublers/quadruplers etc. which adds some more complexity. I think I will have to consider things - and also if there may a feasible workaround for this.
Again, thanks for considering
Jesper
However, I do not want to digress too much from the topic of this thread (IVX's Cosmos ADC, etc.) so just briefly:From pictures on the internet it looks as if the Cosmos oscillators are 45/49 MHz versions (enabling 768 kHz Fs I suppose). Given that I would like to use Andrea Mori's 22/24 MHz (and maybe lower frequency) oscillators I would then have to include clock doublers/quadruplers etc. which adds some more complexity. I think I will have to consider things - and also if there may a feasible workaround for this.
Again, thanks for considering
Jesper
I haven't made any claims about the audibility. IMO this is similar to chasing ppm distortions or >120SINAD. Low close-in phase noise is an indication that there are no significant faults in the clock or Vref implementation. Besides a DAC having high noise skirts (e.g. due to poor Vref) can still have SINAD > 120dB. This just goes to show that high SINAD does not tell everything.
IME there are practical problems with these measurements. ADCs have limited resolution (about 20 bits). Also it is difficult to have low enough DAC Vref noise that does not overshadow clock noise.
These were about 15 minutes recordings at 44.1 kHz so there are over 32M samples. Results can be viewed at e.g. 32M FFT with no averaging or 4M FFT with 8 averages and in both cases even without FFT windowing as that is possible with coherent clocks. Bandwidth is not very relevant. The horizontal scale on the graph is +-10Hz around the fundamental (11015Hz to 11035Hz).
The resulting phase noise probably includes lots of noise sources. E.g. DAC/ADC Vref, DAC/ADC clock, DAC output stage, ADC input stage. However as ADC is the same the differences come mainly from DAC.
A brief comment on noise: Correlated noise and uncorrelated noise are two different things. They should look different on an FFT. Correlated noise is produced by some process like multiplication in the time domain. Uncorrelated noise is additive. Most of the noise produced in a dac output stage would presumably be additive, and thus uncorrelated. However, noise that becomes intermodulated with a music signal or test signal would be correlated. It is correlated noise that shows up as noise skirts. Uncorrlated noise shows up as a change in the overall noise floor.
Also, phase noise is not equivalent to noise skirts. Noise skirts are the sum of phase noise and amplitude noise each of which are intermodulated with the signal. Phase noise in the frequency domain only corresponds with random timing errors in the time domain. Correlated amplitude noise in the frequency domain corresponds with multiplicative amplitude errors in the time domain, such as those produced by Vref (which is multiplied by the the music signal or test signal inside a dac as viewed in the time domain).
Some additional information attached.
Also, phase noise is not equivalent to noise skirts. Noise skirts are the sum of phase noise and amplitude noise each of which are intermodulated with the signal. Phase noise in the frequency domain only corresponds with random timing errors in the time domain. Correlated amplitude noise in the frequency domain corresponds with multiplicative amplitude errors in the time domain, such as those produced by Vref (which is multiplied by the the music signal or test signal inside a dac as viewed in the time domain).
Some additional information attached.
Last edited:
Please do this with a DSP in the middle so us active speaker builders have something besides the terrible ADAU1701 to use! ...preferable something that can be mounted to a plate inside the speaker...
two independent USB devices(9038S/D2Cosmos ADC), a pair has ADC/DAC has 10mS round trip delay at 32/192k ASIO4ALL. In terms of sonic speed, it is 3.4m, so there should be no problem. Actually, acoustic tests don't need extra precision at all, hence, any laptop's built-in audio would be fine.
I agree, 10 ms delay is something that loudspeaker measurement software like ART can deal with. But how constant is it? The two USB ports and devices establish their own timing. How do they stay in synch?
Do you have any comments about my proposal of a siamese Cosmos, i.e. a DAC and ADC that live in the same box and share their power supply, insulation, oscillator and USB bridge?
Do you have any comments about my proposal of a siamese Cosmos, i.e. a DAC and ADC that live in the same box and share their power supply, insulation, oscillator and USB bridge?
I agree that the timing error of two converters is negligible for the outcome of audio measurements, but a common power supply becomes noticeable as an influence in the measurement. It is also fundamentally necessary to have a DAC with a very powerful preamplification and a very good power supply for measurements of distortion in the bass (20 Hz to 200 Hz), as typically, DACs in this frequency range exhibit the component with the poorest characteristics. Therefore, it is welcomed if there is a Cosmos DAC that can play with very low distortion from 20Hz to 10kHz under a load like that of the Cosmos ADC. I am eager to see whether the Cosmos DAC will be the first to achieve this goal with a 5V power supply of max. 15W