This is the (v2 = improved) output stage from the ESS data sheet:
I understand that thevalues of the components around the op amp are no longer critical to avoid the hump, and no published data of 9039 DACs that I have seen exhibit the hump.
Does anyone know what the three op-amp version with slightly improved THD+N looks like? I the third op-amp maybe just for buffering the reference voltage, as a symmetrical output should have an even number of amps?
Is there a simpler way to get unbalanced output? Just using one side of the symmetrical output signal seems odd. The Chinese 9039 board seems to use only one op-amp per channel.
Does anyone know what the three op-amp version with slightly improved THD+N looks like? I the third op-amp maybe just for buffering the reference voltage, as a symmetrical output should have an even number of amps?
Is there a simpler way to get unbalanced output? Just using one side of the symmetrical output signal seems odd. The Chinese 9039 board seems to use only one op-amp per channel.
The three op-amp version is probably similar to the output stage in ES9039PRO datasheet.
Chinese ES9039Q2M boards with single op-amp per channel use voltage output mode.
Chinese ES9039Q2M boards with single op-amp per channel use voltage output mode.
You mean this (taken from the PRO datasheet)? It uses a third opamp as a servo to balance the output signals. Not sure why that would lower THD. Increase in N is probably an impedance thing. I really wonder why they didn't use another passive low pass on the output of the servo opamp.
You got any info about voltage mode? The outputs have an output impedance of about 400 R. Not ideal for voltage output, it seems to me.
You got any info about voltage mode? The outputs have an output impedance of about 400 R. Not ideal for voltage output, it seems to me.
Can one hear that difference by ear or is that only registered by measuring?
For many the purpose of DAC is to be transparent. Even if this in case of many recordings means garbage in, garbage out. Not garbage in, coloured nicer sounding garbage out.
OK. Had that same view but now I start to wonder if that transparency is noticeable with such minute differences. That is all.
The infamous ESS hump maybe had more influence? Whatever the reason was, it were not the ESS based DACs I had over for visit (till the 9039) with all their added bells and whistles that were most sonically appealing. Maybe I changed over to listening to audio devices and the transparency is only a detail amongst other factors, I don't know.
The infamous ESS hump maybe had more influence? Whatever the reason was, it were not the ESS based DACs I had over for visit (till the 9039) with all their added bells and whistles that were most sonically appealing. Maybe I changed over to listening to audio devices and the transparency is only a detail amongst other factors, I don't know.
Last edited:
We've been discussing this output stage in other threads already. Key point is that it is not DC servo, it's a full bandwidth common-mode servo. The whole thing is identical in function to a true fully differential opamp (FDA, for example OPA1633), and the same applies to the other circuit.
There also are some less obvious things happening, like the ever so slight positive feedback to tailor input impedance. Not sure if it is essential, though.
IMO the biggest problem with ES9038 (and older ESS dacs) is that distortions are highly dependent on level which is clearly visible in stepped sweeps. So much so that I would hesitate to call ES9038 transparent. ES9039Q2M (and AKM dacs) are much better in this respect.
http://archimago.blogspot.com/2024/04/detailed-thdn-vs-output-level.html
Without knowing why I already concluded that solely by listening. That is why I find the filter/features business non relevant.
Why would one bother with secondary details when it simply is not right from the start? I never heard a ES9038Q2M (which was the next big thing once) device in whatever mode, whatever filter settings and whatever opamp number that sounded to my likings whereas even cheap AKM equipped stuff may appeal. The end user/consumer in me then draws simplified but effective conclusions.
Why would one bother with secondary details when it simply is not right from the start? I never heard a ES9038Q2M (which was the next big thing once) device in whatever mode, whatever filter settings and whatever opamp number that sounded to my likings whereas even cheap AKM equipped stuff may appeal. The end user/consumer in me then draws simplified but effective conclusions.
Last edited:
Haven't thought about that before, here we get a reduction in current noise in resistors, it is very good. But on the other hand, thermal noise increases with a larger resistance.
As current noise is pink noise (1/f) and thermal noise is white noise current noise dominates at low frequencies while thermal noise at high frequencies. With good quality thin film resistors the corner frequency (i.e. equal current noise and thermal noise) is below 100Hz (maybe even at 10Hz). So paralleling the feedback resistor lowers noise below 100Hz but increases noise above that. For a company excelling in numbers game this seems an odd proposition as SINAD is likely worse with parallel resistors.
Another reason thermal R modulation, leading to THD-3. Paralleling allows to dissipate excess heat more easily
Draw the circuit and write out the equations and you will see that two 3600 ohm resistors in parallel produce precisely as much thermal noise as one of 1800 ohm, no matter whether you look at the voltage, current or available power. 1/f noise will be better; if there is no DC current flowing through the resistor, it will only show up as sidebands around the signal.
Yes, you seem to be correct. I only considered that the noise from each of the resistors is uncorrelated and thus should be added similar to other separate noise sources.