@G-K
See my response #285 here:
https://www.diyaudio.com/community/...perform-apx555b-for-30000.386001/post-7345826
Victor oscillator with my passive notch and APU LNA are here (sorry it is Czech and Slovak forum only):
http://www.audioweb.cz/viewtopic.php?pid=400218#p400218
And some of my older setup here:
https://www.diyaudio.com/community/threads/low-distortion-audio-range-oscillator.205304/post-7022462
Yes, some shielding problems are visible under 1 kHz.
Martin
Edit: It seems that your APU is working good.
See my response #285 here:
https://www.diyaudio.com/community/...perform-apx555b-for-30000.386001/post-7345826
Victor oscillator with my passive notch and APU LNA are here (sorry it is Czech and Slovak forum only):
http://www.audioweb.cz/viewtopic.php?pid=400218#p400218
And some of my older setup here:
https://www.diyaudio.com/community/threads/low-distortion-audio-range-oscillator.205304/post-7022462
Yes, some shielding problems are visible under 1 kHz.
Martin
Edit: It seems that your APU is working good.
Last edited:
I managed to improve 3rd HD of my Cosmos APU by bit over 3dB by replacing the input opamp IC3 (OPA1612). Earlier I got different results depending on the ADC I was using (ES9822PRO or AK5394). After replacing IC3 I now get more or less similar results with both types of ADCs. With ES9822PRO I still get slightly better results without Cosmos APU but that may be related to APU having only unbalanced outputs (which IMO was a poor decision). Also I'm not that confident about ES9822PRO as the "ESS hump" makes it somewhat questionable as a precise measurement ADC.
Sorry for the slightly Offtopic but I can't resist 
Actually, it was measured with Cosmos APU.
The THD+N performance Cosmos DAC proto V1.4 ES9039Pro@44.1k 0dbfs 9.7Vrms balanced output. Different MCLK settings show the compromises between noise and harmonics distribution, pretty much the same as for ES9038 - lower MCLK = cleaner harmonics but higher noise. The dynamic range is 136.7db(A), crosstalk -144db@1kHz. This is purely DAC output, no LPF is involved, just ES9039Pro + one composite opamp for the left, and another one for the right channel.
Actually, it was measured with Cosmos APU.The THD+N performance Cosmos DAC proto V1.4 ES9039Pro@44.1k 0dbfs 9.7Vrms balanced output. Different MCLK settings show the compromises between noise and harmonics distribution, pretty much the same as for ES9038 - lower MCLK = cleaner harmonics but higher noise. The dynamic range is 136.7db(A), crosstalk -144db@1kHz. This is purely DAC output, no LPF is involved, just ES9039Pro + one composite opamp for the left, and another one for the right channel.
@IVX could you please show results for +4 dBu, 0 dbV too? I dont know any amplifier or preamplifier with input sensitivity of 9.7V resp. +21.95 dBu. I really dont know how much sense it makes to increase the output voltage of a DAC to such level. Next we will have output of 20 V to increase the dynamic range to another 6 dB.
Fortunately, it is very easy to test on my side, I used powerbank powered APU PCBA, and battery-powered scope to minimize common-mode disturbances.
I see the highest peak at 84MHz(-82dbfs, where is 0dbfs = 100mV), next at 36MHz -84dbfs or so, and the lowest one at 12MHz. 1.3MHz area is clean.
I tried to read the noise just from the GND pin of 3.5mm plug and the result was pretty much the same, hence, I see the common-mode noise.
I see the highest peak at 84MHz(-82dbfs, where is 0dbfs = 100mV), next at 36MHz -84dbfs or so, and the lowest one at 12MHz. 1.3MHz area is clean.
I tried to read the noise just from the GND pin of 3.5mm plug and the result was pretty much the same, hence, I see the common-mode noise.
I can't see clearly but looks like 60mV-peak? How much it changes if set 34db of gain? Did you check the actual gain is close to 34 and 60db(to make sure if THAT 1510 inputs, or protection resistors/zener arent burned)?
Also, try to short the 3.5mm plug input+ to the input- but don't short them to the GND pin, is the 1.3MHz noise the same?
Also, try to short the 3.5mm plug input+ to the input- but don't short them to the GND pin, is the 1.3MHz noise the same?
Very nice resultsSorry for the slightly Offtopic but I can't resist
The THD+N performance Cosmos DAC proto V1.4 ES9039Pro@44.1k 0dbfs 9.7Vrms balanced output. Different MCLK settings show the compromises between noise and harmonics distribution, pretty much the same as for ES9038 - lower MCLK = cleaner harmonics but higher noise. The dynamic range is 136.7db(A), crosstalk -144db@1kHz. This is purely DAC output, no LPF is involved, just ES9039Pro + one composite opamp for the left, and another one for the right channel.
I assume that the noise can be reduced by using a number of I/V's and adding the outputs instead of using one I/V per channel. But whether it makes a significant difference, I don't know.
What is the benefit of using a composite amplifier? Is it to get enough current drive or is it to reduce the distortion (or both)?
Further details would be appreciated
JensH,
sure, if you'll find the optimal opamp for the I/V stage i.e. low current noise first(about 900ohm FB resistor and 195ohm internal DAC), and low voltage noise as well. The approach requires 10pcs opamps + 1 if you need unbalanced output too.
However, as I see, I have a better chance to find just low voltage noise opamp for the 220ohm of the FB resistor. Of course, I need a high current output opamp at the same time, due to such an opamp doesn't exist, I go with composite ADA4898 for .9nV/Hz^.5 + AD8397 for 300mA+ output and tonnes of loop gain in the audio band. The approach requires 2pcs opamps(magic one with low voltage noise and >100mA output, in reality, there are 2 pairs of ADA4898+AD8397) + 1 if you need unbalanced output too. 11 vs 5 opamps, and as you can see no one chi-fi ES9039Pro DAC got >134db(A) of DR using 11 opamps brute-force approach. Need to add that APx555b widely used for such measurement is shamefully noisy, and probably simply not capable to deal with modern DACs. The older APx555 is even worse.
The simulation opa1612 promises -136.5db(A) for the brut-force variant, and the summation's opa1612 is loaded as hell with 400ohm@5Vrms i.e. here would be nice our composite again.
And one more reason to use quasi-direct paralleled outputs - my PCB size is 90x50mm, I can't fit there 11 opamps, and feed them well.
Quasi because 9039 doesn't like direct shorting outputs, it seems there are async noiseshapers, and they lost their loop gain when synced by shorting, unlike 9038 DACs.
matulo_g, FYI, I recollected how I did prove the H3 level is real but not a result of multiple self-compensations. I used R/R(good ones) divider to reduce the input Sine-gen level and see that H3 at the same level.
sure, if you'll find the optimal opamp for the I/V stage i.e. low current noise first(about 900ohm FB resistor and 195ohm internal DAC), and low voltage noise as well. The approach requires 10pcs opamps + 1 if you need unbalanced output too.
However, as I see, I have a better chance to find just low voltage noise opamp for the 220ohm of the FB resistor. Of course, I need a high current output opamp at the same time, due to such an opamp doesn't exist, I go with composite ADA4898 for .9nV/Hz^.5 + AD8397 for 300mA+ output and tonnes of loop gain in the audio band. The approach requires 2pcs opamps(magic one with low voltage noise and >100mA output, in reality, there are 2 pairs of ADA4898+AD8397) + 1 if you need unbalanced output too. 11 vs 5 opamps, and as you can see no one chi-fi ES9039Pro DAC got >134db(A) of DR using 11 opamps brute-force approach. Need to add that APx555b widely used for such measurement is shamefully noisy, and probably simply not capable to deal with modern DACs. The older APx555 is even worse.
The simulation opa1612 promises -136.5db(A) for the brut-force variant, and the summation's opa1612 is loaded as hell with 400ohm@5Vrms i.e. here would be nice our composite again.
And one more reason to use quasi-direct paralleled outputs - my PCB size is 90x50mm, I can't fit there 11 opamps, and feed them well.
Quasi because 9039 doesn't like direct shorting outputs, it seems there are async noiseshapers, and they lost their loop gain when synced by shorting, unlike 9038 DACs.
matulo_g, FYI, I recollected how I did prove the H3 level is real but not a result of multiple self-compensations. I used R/R(good ones) divider to reduce the input Sine-gen level and see that H3 at the same level.
The big roadblock in an analyzer noise floor is the need for input protection. You need some series resistance for the protection circuit to work. However that series resistance limits the noise floor. Typical value is 10K Ohms. Add to that an input attenuator scaled to 100K Ohms and you run into theoretical limits pretty quick. One trick is to use incandescent bulbs for the current limiters. You get an improvement but they are a pain and becoming obsolete.