I was wondering if, either as part of Cosmos DAC or a standalone box like APU, you could further clean up the test signals with analog filters. A Sallen-Key 2nd or 4th order low pass at or below 1 kHz would already attenuate the 2nd harmonic considerably and almost completely get rid of the higher orders. It might even work with the DAC set to 10 kHz, maybe with some extra gain which would probably need to be placed between the first and second filter stages.
Or active notch filters, at least two of them for IMD and maybe with switcheable notch frequencies?
Can you give us an example of how you implemented composite op-amps in Scaler and apparently the ADC-iso? Maybe I'll cobble up an active filter with a composite op-amp one of these days.
Unfortunately for the Cosmos DAC project, 9039S at 24MHz MCLK produces so clean sine that I have no idea how to improve that for Cosmos DAC in the future. Actually, 9039S was born as an attempt to get a cleaner harmonics-free performance of Cosmos DAC proto-III. The proto-I was based on dual ES9039Q2M and got 130db(A)+ of DR and THD+N -123db, that's not really WOW and I decided to move the project to ES9039Pro. The proto II, due to 9039Pro(2 ch config) gave me DR 137db(A) and THD+N -130db, however, that chip has quite cheap high-frequency performance, 10k H3 is -110db! Of course, I can't be completely happy with that, fortunately, ESS released one more 9039 chip - Q2M, and 10x cheaper than 9039Pro. I instantly prepared proto-III on dual ES9039Q2M and got DR 133db(A), THD+N@1k about -128db, and 10k H3 -135db. The only bare THD@1k wasn't perfectly clean, lots of tiny high-order harmonics don't let me sleep well. To realize the root cause of the residual H-ord harmonics I decided to try ES9039Q2M in a simpler project. I took my old 9038S portable DAC project, and redesigned that for ES9039Q2M and OPA1612+AD8397 composite. The very first PCB version showed me an extremely clean sine performance, and the next 14 PCB iterations were needed only for fine-polishing the performance under 16ohm load + right parts selection. This is the story about why Cosmos DAC release is delayed again and instead of that I suddenly offer 9039S which I tend to consider as Cosmos DAC Lite.
Last weekend I tested the first production panel with 10pcs 9039S PCBA. Each DAC was tested with a Windows PC USB3 port + 2m good fully featured USB2 cable for close to real-life conditions. 16ohm loaded(Cosmos Load-board has 16.1ohm yet) results would be better with a shorter cable + powerbank and so on(see Archimago's review). Cosmos APU LNA 34db for the DR test, Cosmos APU notch +6db for the THD+N tests, and THD test used APU notch 0db.
32/96 DR AVG = 130.32db(A), No load 32/96 0dbfs THD+N AVG -125.66db, 32ohm loaded 32/96 0dbfs AVG THD+N -124.77db, 16ohm loaded 32/96 -.5dbfs AVG THD+N -122.935db. Interestingly, for the kind of Sine_gen mode(MCLK = 24MHz 1db higher noise but less harmonics) THD compensation coefficients are nearly zero(between 0 and 4), and all samples have at least one channel with all harmonics <-150db!
Last weekend I tested the first production panel with 10pcs 9039S PCBA. Each DAC was tested with a Windows PC USB3 port + 2m good fully featured USB2 cable for close to real-life conditions. 16ohm loaded(Cosmos Load-board has 16.1ohm yet) results would be better with a shorter cable + powerbank and so on(see Archimago's review). Cosmos APU LNA 34db for the DR test, Cosmos APU notch +6db for the THD+N tests, and THD test used APU notch 0db.
32/96 DR AVG = 130.32db(A), No load 32/96 0dbfs THD+N AVG -125.66db, 32ohm loaded 32/96 0dbfs AVG THD+N -124.77db, 16ohm loaded 32/96 -.5dbfs AVG THD+N -122.935db. Interestingly, for the kind of Sine_gen mode(MCLK = 24MHz 1db higher noise but less harmonics) THD compensation coefficients are nearly zero(between 0 and 4), and all samples have at least one channel with all harmonics <-150db!
regarding the composite, in Cosmos Scaler I used one I found in that forum https://www.audio-perfection.com/forum/showthread.php?tid=87
However, I tried a classic composite as well with the same results because the bottleneck isn't Aol but all complex of nonlinearities together.
However, I tried a classic composite as well with the same results because the bottleneck isn't Aol but all complex of nonlinearities together.
Thanks, Ivan. I am slightly confused: Cosmos DAC
Proto I was 2x 9039Q2M and had 130 dB DR
Proto II was 9039Pro and had poor 10 kHz H3
Proto III used the then newly realeased 9039Q2M - maybe there was a typo about the DAC in the Proto I?
About composite, are you referring to the PDF in post 1 of the link? That is only accessible if one registers in the Russian forum. Can you explain how it differs from classic composite?
Proto I was 2x 9039Q2M and had 130 dB DR
Proto II was 9039Pro and had poor 10 kHz H3
Proto III used the then newly realeased 9039Q2M - maybe there was a typo about the DAC in the Proto I?
About composite, are you referring to the PDF in post 1 of the link? That is only accessible if one registers in the Russian forum. Can you explain how it differs from classic composite?
capslock You are damn right 
9038Q2M of course, thank you for the correction, actually, I decided to post that txt on the 9039S product page..
PS: the composite looks like this: -IN of 1st and 2nd opamp are tied together, Output of the 1st opamp goes through the R/C voltage divider(but not R/R as a classic one to get extra Aol in audio band) to the +IN of the 2nd opamp.
9038Q2M of course, thank you for the correction, actually, I decided to post that txt on the 9039S product page..PS: the composite looks like this: -IN of 1st and 2nd opamp are tied together, Output of the 1st opamp goes through the R/C voltage divider(but not R/R as a classic one to get extra Aol in audio band) to the +IN of the 2nd opamp.
Last edited:
So, basically a variation of this? (from LT's classic AppNote 21 https://www.analog.com/media/en/technical-documentation/application-notes/an21f.pdf)
Speaking entirely for myself, I would spend extra for a Cosmos DAC in the traditional/classic Cosmos housing over the portable dongle version. For use on a lab bench, the connector arrangement in the Cosmos series is easier to deal with than the dongle package. The dongle package is ideal for the intended audience, but I'd like separate balanced and unbalanced outputs for each channel.
Perhaps if you want to add a feature, you could add in the USB isolator. Not everybody will be using a Cosmos ADCiso in their test set-ups. Plus, I could imagine ground loops between an amplifier under test and the computer could influence measurements if there is no isolation in that part of the system. (Of course, you'd want to be able to test that too...)
Perhaps if you want to add a feature, you could add in the USB isolator. Not everybody will be using a Cosmos ADCiso in their test set-ups. Plus, I could imagine ground loops between an amplifier under test and the computer could influence measurements if there is no isolation in that part of the system. (Of course, you'd want to be able to test that too...)
Not quite what I would make from Ivan's description. Below is my interpretation. The low pass instead of the resistive divider used in the old National Semiconductor app note and the Neurochrome is a neat idea (though the Brainstorm works without any resisitive divider: https://zelfbouwaudio.nl/forum/viewtopic.php?t=22898&sid=70857f49bf17df243f8e7d84720d994e). I am pretty sure U1 needs a C or a series RC to provide local feedback (dotted line).
With the two op amps sharing the - input and hence the feedback, U2 essentially becomes a unity buffer. If it weren't for the capacitive divider, it would need to be unity gain stable, or in other words, if U2 is unity gain stable, maybe you can get rid of the capacitive divider?
I don't really see the advantage of having the two share the feedback network and jump through hoops to make the thing stable. Why not give U1 its own feedback set at somewhat lower than the overall feedback?
Keep in mind that it is supposed to be an IV circuit (unless I'm mistaken here and the composite HP driver comes after the IV proper).
They are not sharing the feedback. The master OpAmp increases the error voltage gain for the slave OpAmp, as seen between its input pins of the latter. In both circuits.
In your circuit, the slave's -IN connected to the main feedback divider just means that the master's output follows the input voltage (+ the small correction voltage on top). If that -IN were grounded, the master's output also is kept at ground (+ correction voltage). Doesn't make much of a difference.
Last edited:
Well, I was thinking of the composite amp in Cosmos Scaler and presumably ADCiso. You're right, in a DAC circuit it could be an I/V.
Master = U1, slave= U2? If so, and if gain = 20, the LF output of the master will be 20x input + some error. The slave's output will also be 20x input + a smaller error. What both cuircuits?
it is available but need to ask Sunny first, she hasn't put that item into Aliexpress yet.
I finished the product description today https://e1dashz.wixsite.com/index/9038s
Looks a bit aggressive ))
I finished the product description today https://e1dashz.wixsite.com/index/9038s
Looks a bit aggressive ))