Markw4, 122dbA-10dbfs=112dbA=-109db but AP gave me THD+N -100db@-10dbfs and .0003%@-3dbfs. I think AP see there some USB common-mode HF noise and I found level where -.5db step is enough to change THD+N reading from .0005% to .001%. My notch+ADC see nothing there in 10-24000Hz. Another obvious problem is THD+N formula which is not the same for AP vs arta or spectraplus, or WS etc. No idea why about THD+N so big mess. Did you see AP reading THD+N = .0001% with 9038Q2M yet? I didn't see <.00025% with AP's loopback .00015%(that's limited only by AP's THD+N analyzer because its generator has <-130db), even if notch+ADC shows me -123db for 2nd and 3rd harmonics. I think I need to lease APx555 anyhow..
Last edited:
Huh, I found the reason for the mismatch and it was about ground loops, all of my toys(AP/ADC/Notch and DUT) were tied to USB ports of single PC. Now I can adjust 2nd and 3rd harmonics with 9038q2m regs#22-25 up to THD a bit less than .0001% even .00007%(my ADC+notch si enough to see about .00005% from AP's generator where is <.00003% guaranteed) but level .0001% looks more or less stable 
However, AP reading stop changing at about .00024% even A-weighted but I believe it is due to noise and AP's THD analyzer resolution limit together.
However, AP reading stop changing at about .00024% even A-weighted but I believe it is due to noise and AP's THD analyzer resolution limit together.
As I already reported here, the best clock I found is an Xtal. It was hard to get Xtal for 100MHz and I found one in quite big case 7x5mm i.e. it's not suitable for USB->DSC->HPA dongle. I prepared one more ES9038Q2M USB->DAC->HPA proto on bigger PCB(#9038X) where I implemented dual foot-print Xtal/Osc. I tried there 100MHz Xtal 7x5mm but it generates on 33.333MHz yet )) Shop sold me Xtal has no any datasheet about it, SE9038Q2M is also too laconic, so I've no idea how to run that Xtal with SE9038Q2M in 3rd overtone. Mark, the product mentioned here as #9038S not intended for further amplification, this is USB->DAC->HPA interface for balanced 4 wire tied headphones with about 350mW@32ohm, and my measurement was performed with such load or directly on planar headphones. Of course, it is possible to arrange kinda adaptor from #9038S 2mm 4 poles jack + USB GND to XLR->Amp but I consider #9038S as USB->DAC->HPA interface tiny one and precise simultaneously.
PS: what about my question: Did you see AP reading THD+N = .0001% with 9038Q2M yet? Why I keep asking because I think it is not possible if SNR worse than -126..-129dbA
PS: what about my question: Did you see AP reading THD+N = .0001% with 9038Q2M yet? Why I keep asking because I think it is not possible if SNR worse than -126..-129dbA
They say that up to -129dB DNR and -120dB THD+N is possible. At other times they have said -120dB distortion, and -122dB noise. Probably the lowest DNR value would come from using dual mono dac chips. In terms of distortion, I think it is possible to hit their numbers with careful design. However, some things (jitter being one example) are better measured in other ways than just THD or THD+N. IMHO, even THD doesn't mean much: If there are higher order harmonics above the 3rd, sometimes adjusting the HD compensation registers so as to leave a bit if 2nd and 3rd harmonics can help improve perceived sound quality by masking the higher order harmonics, again, IMHO.
Last edited:
Wow, this thread has some amazing activity, a lot way beyond my knowledge, anyway, I thought I'd update my activity.
I had two single channel balanced to unbalanced converters arrive from China (Balanced to Unbalanced XLR to RCA Single Ended OP-AMP Preamplifier Board CD Play | eBay) so I decided to bypass the onboard analogue and feed the DAC/DACB to these converters after 6.2k resistors (cutting all links around these).
These converters come with 2x5532 which I replaced with LM4562s, removing the +/- 15V from the DAC board and powering these instead. The DAC powered by a separate transformer and Dexa NewClass D 5v feed which is then dropped down to 3.3v using 3 LDOs (the standard 1117 for the ancillaries, 2xLT3045s (LT3045) which power the clock & AVCC, the other VCCA and DVCC) in a similar fashion to pixel28.
I'm very impressed with the sound so far, I'm only testing on my 2/3rd systems (Marantz MA22 preamp to Rotel RB970BXII or a vintage Marantz 2285B).
My next tasks are to look at replacing the crystal and changing the AVCC power source to an op-amp based one.
When using an Arduino to reprogram the DAC registers, what does that actually entail?
I had two single channel balanced to unbalanced converters arrive from China (Balanced to Unbalanced XLR to RCA Single Ended OP-AMP Preamplifier Board CD Play | eBay) so I decided to bypass the onboard analogue and feed the DAC/DACB to these converters after 6.2k resistors (cutting all links around these).
These converters come with 2x5532 which I replaced with LM4562s, removing the +/- 15V from the DAC board and powering these instead. The DAC powered by a separate transformer and Dexa NewClass D 5v feed which is then dropped down to 3.3v using 3 LDOs (the standard 1117 for the ancillaries, 2xLT3045s (LT3045) which power the clock & AVCC, the other VCCA and DVCC) in a similar fashion to pixel28.
I'm very impressed with the sound so far, I'm only testing on my 2/3rd systems (Marantz MA22 preamp to Rotel RB970BXII or a vintage Marantz 2285B).
My next tasks are to look at replacing the crystal and changing the AVCC power source to an op-amp based one.
When using an Arduino to reprogram the DAC registers, what does that actually entail?
Okay. IIUC what you did there, that would leave the dac outputs operating in voltage mode. Voltage mode makes THD *much* worse than if using current mode (opamp I/V conversion).
Not ideal. Clock and VCCA could be grouped together to save using one regulator. DVCC should be on its own regulator.
And of course, very good for sound quality to use an opamp regulator for AVCC. In that case a clean +3.3v reference is needed, and the +-15 or other rail voltage used for opamps should be suitable for making opamps sound their best (same quality as one would want for the output stage). As we try to point out from time to time, film cap power rail filters can work for that, and Nazar reguators may be very good low cost option to consider instead.
Sound quality should keep getting better and better as the modding progresses
What I usually do is let the dac board MCU get the dac booted up and running, then I take over control of the I2C bus and read/write numbers from/to the dac registers to make the dac perform as desired.
It takes a data sheet for the dac chip to know what the various registers do, and since data sheets are confidential documents under ESS policies we cannot post confidential details here.
Someone may find publicly known information about controlling Sabre dacs in places like the Twisted Pear threads, perhaps in publicly posted software (such as Dimdim's Sabre controller project), and perhaps in other places.
General things such as master mode, asynchronous mode, ASRC, DPLL bandwidth, etc., can be discussed in general ways, so long as particular register numbers and particular register control bits are not talked about.
We can talk about syntax for writing bits and what bits do in registered I2C devices. We can talk about number systems such as decimal, hex, binary, etc., and how they are related, notated, how to convert between them, and so forth.
In other words, I cannot tell you to write some number X to some register Y to make the dac do this or that thing. That would be getting too specific for public discussion.
People can ask questions about notation, numbers and base conversions, how to understand the idea of bits as control inputs, how bits may be manipulated, and lots of similar kinds of things.
Or, for people who find the above prospects overwhelmingly complicated, there are options like Dimdim's Sabre controller project where software has already been written and any confidential info from some data sheet is hidden from users so no legal problems are created.
It should also be pointed out that ESS is usually pretty good about giving out data sheets to people who request them through an ESS distributor. However, getting data sheets does require signing an non-disclosure agreement. Emailing an ESS distributor and asking for an NDA form should get things rolling. For people that do have NDAs in effect with ESS and who have legal copies of data sheets, such people do not appear to be prohibited from communicating with each other in private regarding info they are entitled to have, IMHO only, so long as all confidential information stays confidential and does not become public. For official clarification of any NDA rules, please contact ESS and or their distributor.
Last edited:
Yesterday I mentioned there was another piece of gear here for a few days. That second item is a pair of Audeze LCD-4 headphones.
They certainly are way, way more detailed and accurate than my old Sennheiser HD 600 headphones. Especially so in the bass and low to mid part of the midrange. Don't know if cymbals sound exactly as they ought to. The HD 600 cans are like listening to mud in comparison to the LCD-4's. However, the latter are big, and kind of tight fitting. Not ideal for all listening situations, but really good for hearing small details very clearly and with much lower distortion than most other transducer types. Some people still say the high end Stax are better and that is likely so, but short of that the LCD-4's may be the next best thing.
Interestingly, both my high end audio designer friend and my son have compared both of the dacs using both sets of headphones with both headphone amps and arrive at the same dac judgement: Both prefer the sound of my modded dac#2 (now that the somewhat complicated grounding situation has been improved a lot) over DAC-3.
All the more reason for me to get to work trying to come up with an improved clock divider scheme. I keep thinking about how I might go about it, and I am putting together a 3rd test dac with that purpose in mind.
By the way, the THD-4 headphones and the Pass HPA-1 will be leaving soon. I think I will probably miss the headphones more. However, they are much too expensive for me (and for most people, I would reckon).
They certainly are way, way more detailed and accurate than my old Sennheiser HD 600 headphones. Especially so in the bass and low to mid part of the midrange. Don't know if cymbals sound exactly as they ought to. The HD 600 cans are like listening to mud in comparison to the LCD-4's. However, the latter are big, and kind of tight fitting. Not ideal for all listening situations, but really good for hearing small details very clearly and with much lower distortion than most other transducer types. Some people still say the high end Stax are better and that is likely so, but short of that the LCD-4's may be the next best thing.
Interestingly, both my high end audio designer friend and my son have compared both of the dacs using both sets of headphones with both headphone amps and arrive at the same dac judgement: Both prefer the sound of my modded dac#2 (now that the somewhat complicated grounding situation has been improved a lot) over DAC-3.
All the more reason for me to get to work trying to come up with an improved clock divider scheme. I keep thinking about how I might go about it, and I am putting together a 3rd test dac with that purpose in mind.
By the way, the THD-4 headphones and the Pass HPA-1 will be leaving soon. I think I will probably miss the headphones more. However, they are much too expensive for me (and for most people, I would reckon).
eziitis, it is THD but not THD+N. Top one plot is ADC after 40db notch and THD reading is .009%, hence, due to 40db = 100x the actual THD = .00009%. The same time AP reading THD+N I've never seen <.0002%, and I don't believe if someone else did with SNR -122-123dbA. About the sound, I don't want publish my opinion yet, I'm an engineer that's it. I've an agreement with Z-reviews youtube channel to review two of my models, let's see.
the TRUE digital one
the TRUE digital one
Last edited:
Hah, I found the good example explains why THD+N is not possible to be <.0002% when SNR -122dbA. Link#1 Qutest - The Award-Winning Desktop DAC | Chord Electronics output noise is exactly the same as I have got i.e. 2.6uVRMS A-weighted(AP shorted input noise is 500nVRMS A-weighted). Qutest has 3VRMS output, hence its SNR -121dbA i.e. a bit worse than my #9038S which has 3.4VRMS output @THD+N=1% SNR -122.2dbA(to be honest, it could be 1 db worse if the DUT being connected one of 6 USB ports of my PC).
Link#2 Review and Measurements of CHORD Qutest DAC | Audio Science Review (ASR) Forum independent measurement with APx555(the most precise AP analyzer today, about $30000). As you can see Qutest DAC has the same THD level as #9038S -120db =.0001%, however, APx555 THD+N reading again .0002% that corresponds -114db. So, if somebody will claim that he got THD+N .0001% with SNR -122-123dbA it will be a mistake or BS ))
PS: probably the comparison of Qutest vs #9038S is not exactly actual due to my #9038S sample is individually tuned but Qutest probably not. However, if #9038S also will be tagged as $2000 USB DAC+HPA, I ready to tune every one sample by myself. Well, ok, even $200 I agree )) BTW, untuned #9038S preliminary retail is $70
Link#2 Review and Measurements of CHORD Qutest DAC | Audio Science Review (ASR) Forum independent measurement with APx555(the most precise AP analyzer today, about $30000). As you can see Qutest DAC has the same THD level as #9038S -120db =.0001%, however, APx555 THD+N reading again .0002% that corresponds -114db. So, if somebody will claim that he got THD+N .0001% with SNR -122-123dbA it will be a mistake or BS ))
PS: probably the comparison of Qutest vs #9038S is not exactly actual due to my #9038S sample is individually tuned but Qutest probably not. However, if #9038S also will be tagged as $2000 USB DAC+HPA, I ready to tune every one sample by myself. Well, ok, even $200 I agree )) BTW, untuned #9038S preliminary retail is $70
Last edited:
I've remeasured SNR for #9038S with battery powered Android phone as USB streamer(WIFI = off, BT=off, Neutron player 24/44.1 output, no idea why but that player works free on my Android! FYI, Googleplay isn't available in china), and got such result: YouTube
SNR continuously breathes from -122.9dbA to 123.8dbA and seems I know why, the DAC working in the Performance mode where host MCU reading I2C every 500mS or so, to optimally readjust internal registers ES9038Q2M according to audio data Fs rate. I think if I stop that I2C scanning the SNR will be over 123.5dbA. Interesting that any my wireless mouse movement affects the reading SNR, and you can see that in my video ))
SNR continuously breathes from -122.9dbA to 123.8dbA and seems I know why, the DAC working in the Performance mode where host MCU reading I2C every 500mS or so, to optimally readjust internal registers ES9038Q2M according to audio data Fs rate. I think if I stop that I2C scanning the SNR will be over 123.5dbA. Interesting that any my wireless mouse movement affects the reading SNR, and you can see that in my video ))
Last edited:
Ok, ok, I'll see what I can do here
Sorted
My +/-15V toroidal has plenty of capacity, I'll look at replacing with something a little quieter than the current 317/337 (AUDIOWIND Power Supply Board Kit, PCB, Based on LM317 & LM337 IC. AUDIOWIND SP-2, AUDIOWIND MD-A240/LM)
Sound quality should keep getting better and better as the modding progresses
What I usually do is let the dac board MCU get the dac booted up and running, then I take over control of the I2C bus and read/write numbers from/to the dac registers to make the dac perform as desired.
It takes a data sheet for the dac chip to know what the various registers do, and since data sheets are confidential documents under ESS policies we cannot post confidential details here.
Someone may find publicly known information about controlling Sabre dacs in places like the Twisted Pear threads, perhaps in publicly posted software (such as Dimdim's Sabre controller project), and perhaps in other places.
General things such as master mode, asynchronous mode, ASRC, DPLL bandwidth, etc., can be discussed in general ways, so long as particular register numbers and particular register control bits are not talked about.
We can talk about syntax for writing bits and what bits do in registered I2C devices. We can talk about number systems such as decimal, hex, binary, etc., and how they are related, notated, how to convert between them, and so forth.
In other words, I cannot tell you to write some number X to some register Y to make the dac do this or that thing. That would be getting too specific for public discussion.
People can ask questions about notation, numbers and base conversions, how to understand the idea of bits as control inputs, how bits may be manipulated, and lots of similar kinds of things.
Or, for people who find the above prospects overwhelmingly complicated, there are options like Dimdim's Sabre controller project where software has already been written and any confidential info from some data sheet is hidden from users so no legal problems are created.
It should also be pointed out that ESS is usually pretty good about giving out data sheets to people who request them through an ESS distributor. However, getting data sheets does require signing an non-disclosure agreement. Emailing an ESS distributor and asking for an NDA form should get things rolling. For people that do have NDAs in effect with ESS and who have legal copies of data sheets, such people do not appear to be prohibited from communicating with each other in private regarding info they are entitled to have, IMHO only, so long as all confidential information stays confidential and does not become public. For official clarification of any NDA rules, please contact ESS and or their distributor.
I've got the 9038Q2M's datasheet and have had a look at this so I can see there's plenty to play with.
Noise isn't the issue. One problem with LM317/LM337 is they will not fully attenuate power ripple from getting into the sound. Using two stages of LM317/LM337 in series can fix that, or so can other approaches.
Of course, low noise and low ripple still won't make opamps sound their best. Film cap filters can do it. Not sure what the minimum workable value is, but somewhere around 100uf of film caps (maybe half that value would suffice, but not tested yet) in parallel from each +-15v rails to ground will do quite well. None of the individual film caps should be greater than 33uf and at least 20uf of the total for each rail should be made up from 10uf or smaller film caps. That keeps power impedance low at frequencies that affect opamp sound quality. Most likely, Nazar regulators can do more or less the same thing too for lower cost. How power filtering would be done in that case would be different. Nazar regulators use a lot of feedback to keep impedance low, much like Jung/Didden regulators do. So, phase shift caused by caps on the power rails has to be limited/controlled. I have been told by multiple people that the Nazar circuit is a good one for audio circuits, and although I have not tried them myself yet, I am inclined to believe the good reviews. However, I can't personally recommend them over film caps until I try Nazar for myself. (I recently found out that ADM7150 isn't so good for ES9038Q2M AVCC even though it is used a lot by some people for ES9038PRO AVCC. I need to be careful about what it sounds like I am recommending until I or someone I trust has done a direct comparison. So far, I am not aware of a direct comparison between film caps and Nazar.)
Forgot to ask. Have you tried to load the output ADM line additionally? Perhaps this would have a good effect on the sound, but I cannot draw hasty conclusions. ES9038PRO uses much more current than Q2M and ADM response
the change of load is much better.
Even with the transistors in the TO92, they were very hot (not surprisingly). Now I put the BD139 / BD140 and it became better. But I still have to put the radiators, because they heat up to 65-70C.
And there was a very interesting situation. I left the DAC on and suddenly I felt the smell of burning. I looked inwards and the smoke came from the resistor in the nazar regulator again. I felt very sad, because again for unknown reasons, the output transistor burned down. I rang it, the emitter-collector junction was burned. I checked the entire circuit and then replaced the transistor. But what was my surprise when the resistor began to smoke again! I rang the transistor again - and it burned again. It turns out that it burned a 10uf capacitor, which was soldered to the transistor legs. I still think why this is the case. Maybe due to the fact that I bent the transistor legs with a soldered capacitor? Or because of overheating. And maybe because of the pulsations (this is unlikely).
I also thought about how my transfer Amanero is now arranged. I think the earth is made terribly. It comes from the outlet in my computer. Then over the USB cable to the Amanero board. There is an RC circuit from a resistor of several tens of ohms and a capacitor (That is why I initially isolated the USB case from the case of the entire DAC) And then it goes to the ground wire going to the ES9038 board. It turns out that the earth is not connected to the body at all. More precisely, it is connected, but through the contact spots of RCA connectors that touch the case. I think this is very bad.
Maybe I should isolate the PC ground from the DAC. I'm confused.
I also plan to power the Amanero board from the Nazar regulator. What do you think is a good idea? There you need 3.3v, they are taken from the USB of the computer and are converted to 3.3 using a converter. I heard that the Amanero board has 2 modes of operation - from an external power source and from USB. But for this you need to flash the board. Although I do not see any difference between this. This is known only to the designer Amanero.
Last edited:
Not yet. I can only do so much myself. It would be great to have more people working on settling issues such as that one.
Not sure how much you know about electronics? When the regulator is running okay with a resistor load only, it might be a good idea to measure the DC and the AC voltages across each component. Then do the same when it is connected to the dac as a load. Do the same again if it starts to show signs of excess heating. In some ways it will not be as good as also using a scope to capture waveforms relative to ground potential, but it could be helpful information. We are kind of stuck with figuring out how to use test instruments since we usually can't directly sense what is happening in a circuit. Then, besides measuring and calculating to see if measurements make sense, we can also devise experiments that can tell us more of what we need to know.
Regarding grounds, there are some common chips used for galvanic isolation I2S and I2C signals in dacs.
https://www.mouser.com/datasheet/2/368/Si840x-1397948.pdf
https://www.silabs.com/documents/public/data-sheets/Si866x.pdf
We have not used parts like that in the dac modding project so far, but there are some already built into JLSounds USB to I2C board (which is a direct competitor with Amanero). I2SoverUSB - I2S over USB Audio
Since Amanero boards don't have isolators built in, typically they would be used on an additional board located somewhere in between the USB board and the dac board. Both TP and Iancanada offer USB dac clocking boards/solutions with isolators that can be used with Amanero.
Separate power supplies are used for the dirty and clean sides of the isolators. The dirty side can be powered by USB power in some cases, or from a dedicated dirty power supply in the dac chassis.
Also, isolators do add some jitter. Often a reclocking circuit follows the isolators at some point in the circuit before reaching the dac. That circuit is intended to remove jitter produced by the isolators and or from the USB board. How well those relcocking circuits work at high clock speeds and or in certain dac topologies is not something I have seen carefully studied. Despite that possible concern, the isolators are commonly used in many high end dac products.
Last edited: