You can't consider an ordinary uncalibrated DAC as the reference. Just a couple of days ago we discussed a similar case here https://www.audiosciencereview.com/...ds/e1da-cosmos-adc.27038/page-65#post-1745828
D
Deleted member 148505
A couple of days ago I noticed that during the calibration Cosmos ADCiso has no green/blue/red LEDs indication, in about 30% of cases.
I decided to retest all Cosmos ADCiso that I prepared to sell and found the same over there 30-40% of units have no LED indication at all, at least until units are cold. In my area right now pretty unusual "cold" days 23-25C, and that may explain why prepared to sell and tested at 27C units are glitching. The fact that no one so far complained about the issue even more surprised me. After couple of days I found the root cause and FW solution to fix the issue. Technically, I know the folk over here may understand what I mean, the I2C isolator chip adds 100nS of delay or 200nS round trip delay, and correctly implemented 100KHz I2C interface has 100.00% zero chance of getting any trouble with that. Even 10x times longer delay can not affect the stability, if.. I2C implemented correctly
What happening with the hardware-implemented I2C of CT7601 from Comtrue, when the ADC unit is warm the delay becomes <200nS and I2C works, however, a cold enough unit slightly increases the delay, and CT7601 I2C interface starts to lose acknowledges! Data becomes corrupted and LED's indication doesn't work. Perhaps Comrtue's I2C reads an acknowledge by mixed-up pulse edge, for example, the issue is easily fixable by adding the 47pF capacitor to the SDA line. A +/- few tens of nS are able to turn over so implemented I2C. Of course, I asked Comtrue for help because no one HW I2C register is documented in the supplied datasheet but Comtrue has been silent so far. Finally, I decided to switch I2C to the software mode, Comtrue's source has such a switch but.. that code for sure never tested at all because it had a lot of principal mistakes. So, I debugged of SW I2C and the interface started to work as supposed to be, the only data rate was way too slow 20kHz. Hence, I had to optimize the code, after I'd omitted tonnes of function() called from another function() and this one was called from one more function(), and so on, I got 33kHz of the data rate. Anyhow, after two days I prepared the FW which fixed LEDs indication issue for Cosmos ADCiso. Nobody asked me, I have no idea why ))
I decided to retest all Cosmos ADCiso that I prepared to sell and found the same over there 30-40% of units have no LED indication at all, at least until units are cold. In my area right now pretty unusual "cold" days 23-25C, and that may explain why prepared to sell and tested at 27C units are glitching. The fact that no one so far complained about the issue even more surprised me. After couple of days I found the root cause and FW solution to fix the issue. Technically, I know the folk over here may understand what I mean, the I2C isolator chip adds 100nS of delay or 200nS round trip delay, and correctly implemented 100KHz I2C interface has 100.00% zero chance of getting any trouble with that. Even 10x times longer delay can not affect the stability, if.. I2C implemented correctly
What happening with the hardware-implemented I2C of CT7601 from Comtrue, when the ADC unit is warm the delay becomes <200nS and I2C works, however, a cold enough unit slightly increases the delay, and CT7601 I2C interface starts to lose acknowledges! Data becomes corrupted and LED's indication doesn't work. Perhaps Comrtue's I2C reads an acknowledge by mixed-up pulse edge, for example, the issue is easily fixable by adding the 47pF capacitor to the SDA line. A +/- few tens of nS are able to turn over so implemented I2C. Of course, I asked Comtrue for help because no one HW I2C register is documented in the supplied datasheet but Comtrue has been silent so far. Finally, I decided to switch I2C to the software mode, Comtrue's source has such a switch but.. that code for sure never tested at all because it had a lot of principal mistakes. So, I debugged of SW I2C and the interface started to work as supposed to be, the only data rate was way too slow 20kHz. Hence, I had to optimize the code, after I'd omitted tonnes of function() called from another function() and this one was called from one more function(), and so on, I got 33kHz of the data rate. Anyhow, after two days I prepared the FW which fixed LEDs indication issue for Cosmos ADCiso. Nobody asked me, I have no idea why ))Incredible performance, just as promised at the first page of this thread.
We are now able to measure reliably to -160 dBc levels for a very reasonable price. Here is measurement with APU and Cosmos ADCiso + thermostat board. Source is good signal generator with distortion < -150 dBc. APU was at default gain resulting with -10 dBFs for a 1.7 Vrms input signal.
ADCiso board with attached thermostat board.
We are now able to measure reliably to -160 dBc levels for a very reasonable price. Here is measurement with APU and Cosmos ADCiso + thermostat board. Source is good signal generator with distortion < -150 dBc. APU was at default gain resulting with -10 dBFs for a 1.7 Vrms input signal.
ADCiso board with attached thermostat board.
Well, guidelines are practically nonexistent. Nothing at your site and Google will find only an example with standard Cosmos ADC at ASR forum. It would be helpful to add links or instructions somewhere.
Now, even on my photo I can spot connection points, but with real board in hand, it is not immediately visible (especially for someone with age induced presbyopia).
Now, even on my photo I can spot connection points, but with real board in hand, it is not immediately visible (especially for someone with age induced presbyopia).
Last edited:
Yes, if you don't care about sweepsIncredible performance, just as promised at the first page of this thread.
Also forget about measuring the ESS IMD hump with ES9822PRO.
It depends on how low IMD you looking for, for instance, APx555b is blind for that and sees no hump.
https://cdn.l7audiolab.com/wp-content/uploads/2021/10/SMPTE-Ratio.jpg
https://cdn.l7audiolab.com/wp-content/uploads/2021/10/SMPTE-Ratio.jpg
You can’t do sweeps with fixed notch anyway. OTOH, Cosmos ADCiso seems pretty fine without APU. Distortion with thermostat board and tweaking is from -133 to -135 dBc (about 0.00002%). Less than any DAC has.Yes, if you don't care about sweeps
Anyway, I’ll use this for measuring amplifiers and multitone test for IMD tells all that I need to know. Only problems are that Cosmos ADC itself and DAC used as a signal source produce more IMD hash than measured amplifier.
This is 16 Vpp at amplifier output.
Yes, but with single sine measurements HD cancellation may lead to unexpected results (i.e. too optimistic). Better to recheck with a notch to keep both feet firmly on the ground.
Yes, for single sine measurements I always use notch. And, I would recommend everyone to get latest REW beta build, as it has fix for too optimistic distortion calculation with coherent averaging.
With official REW release, coherent averaging should be used only along with option “Show phase of harmonics”.
With official REW release, coherent averaging should be used only along with option “Show phase of harmonics”.
not really, I keep playing with ES9039Q2M and I can see -150db H2 and H3, so, far not any 9822/Cosmos ADC has so low distortions yet.
BTW, if bohrok is here, I found an inexpensive and nearly perfect way to send MCLK from isolated DAC to UAC2 chip(CT7601) by 3GHz of BW 1:1 trafo.
An interesting experience was choosing the right trafo based on scattering parameters aka S-parameters, instead of a spice model.
Such types of HF parts supplies only S-parameters, and I first time was wondered how to simulate that in a spice environment. Fortunately, free MC12 has a special instrument for that and further simulation was smooth enough to make the choice for the tiny ETC1-1-13 3x3mm SMT trafo. Cosmos DACiso proto 9039q2m sync mode:
Interesting approach. However modern isolator chips have low jitter and if output I2S needs to be isolated using isolator with 3out+1in configuration does not add to the cost. Another thing is that UAC2 chip (XMOS, CT7601, FPGA, STM32, ...) anyhow adds jitter to the generated I2S so jitter free MCK at UAC2 chip is not a panacea. Fortunately DS dacs only need jitter free MCK so jitter of generated I2S is not critical. DSD dacs (or dacs using I2S without MCK) benefit from reclocking preferably at dac.
Wow! Is that really a Cosmos DACiso prototype?? Time to email Sunny?
If, for some reason, you want to extend the bandwidth of that transformer and its common mode rejection, you can add a second transformer as shown in figure 2 here:
MiniCircuits on how RF transformers work
I've built and use an isolator using this idea and this very transformer for use with our cable TV box. It effectively isolates the system from the AC currents used to power the CATV system that leak everywhere. They cause audible hum and picture degradation.
Time to email Sunny?If, for some reason, you want to extend the bandwidth of that transformer and its common mode rejection, you can add a second transformer as shown in figure 2 here:
MiniCircuits on how RF transformers work
I've built and use an isolator using this idea and this very transformer for use with our cable TV box. It effectively isolates the system from the AC currents used to power the CATV system that leak everywhere. They cause audible hum and picture degradation.
I have at least 3 working protos of that DAC, the first one on the dual 9038q2m, the second with 9039pro. A few early protos are disassembled.
9039pro gave me WOW DR(137db(A)), THD+N<130db@1kHz but quite cheap 10kHz performance yet, H3 around -110db, it is shame to me.
9039q2m has 10kHz H3 -135db, THD+N -125db@1kHz, DR 130db(A) for a single version, and DR 133db(A) for dual, THD+N -128db@1kHz, I hope.
But, what's most exciting, the cost of two pcs of 9039q2m is around 20% of 9039pro. The 9039q2m+LPF anyway outperforms 9039pro for THD+N@1kHz(130db vs 131.5db), and so crazy low noise = high DR is possible only at 10Vrms output, so 9039pro with its crazy current draw looks bloody impractical.
9039pro gave me WOW DR(137db(A)), THD+N<130db@1kHz but quite cheap 10kHz performance yet, H3 around -110db, it is shame to me.
9039q2m has 10kHz H3 -135db, THD+N -125db@1kHz, DR 130db(A) for a single version, and DR 133db(A) for dual, THD+N -128db@1kHz, I hope.
But, what's most exciting, the cost of two pcs of 9039q2m is around 20% of 9039pro. The 9039q2m+LPF anyway outperforms 9039pro for THD+N@1kHz(130db vs 131.5db), and so crazy low noise = high DR is possible only at 10Vrms output, so 9039pro with its crazy current draw looks bloody impractical.
So... Four 9039q2m's would be about the same as the 9039pro for DR and THD at 40% of the cost?
At some point, I wonder how much the high current draw hurts because of the simple IR losses within the packages limiting voltage regulation. Think about the Topping distortion when the second channel was inverted in phase.
At some point, I wonder how much the high current draw hurts because of the simple IR losses within the packages limiting voltage regulation. Think about the Topping distortion when the second channel was inverted in phase.