Measurements of an ES9023 DAC

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Several members commented on the quality of audio achievable with an inexpensive DAC chip like the ES9022 or ES9023. I'm impressed with the subjective quality and objective measurements as well.

I started out at 44.1/16 for the measurements and quickly realized that the board is much better than this is capable of demonstrating so 24 bits was the way to go. I also could only find a 24 bit JTEST wav file at 48KHz sample rate so that is what I used. Measurements at 44.1 are for all practical purposes identical…

The analyzer is a UPV from R&S running 256K point FFT with no averaging. The one I have does not include the digital interface option so I use a PC running DR Jordan Designs’ function generator as the source and a Highface USB to SPDIF converter in 48KHZ 24bit using ASIOforall v2. Measurements that used two tones (IMD and DNR) required ASIO to be disabled so they are at 192KHz to make up some of the noise floor lost running without ASIO and driving with two function generator instances.

The first picture is THD at 1KHz and 2V RMS. At a little less than 0.006% it is not earth shattering but very respectable.

The second picture is IMD 18K+20K.

The third picture is DNR at 1KHz with a 10KHz 2V RMS tone. With a noise floor of a microvolt or so RMS the -125 db 1KHz peak is readily visible peeking above the noise. That is ~21 bits of useful resolution. This DAC is dead quiet with a completely black background to the music!

The fourth picture is of the 48/24 JTEST. The -133db peaks are inherent (they are essentially the same with either a single 12KHz tone or with the JTEST signal) and represent jitter of around 3pS. There is almost no measureable increase of data correlated jitter for this board with the Highface.

I notice even low levels of jitter in cymbal ring decay and in the percussive sounds around piano and plucked string instruments. If the cymbals in a good recording sound soft and fuzzy, if the ring decay seems attached to some other sound source, or if the ring decay seems disconnected from the cymbal location it really grates at my preception of the music. The same holds true for the upright bass. Since the fundamental tone and the string plucking/rattling sounds that naturally occur are octaves apart it is a good indicator for my ears. If the string sounds are separate in physical space or out of time from the attack or ring of the note, it really sticks out. The realism of reproduced music for me is strongly tied to how well these attack and decay sounds are placed in the image and in time.

The last picture is of the DAC board itself. I have a final revision I want to make to the PCB to remove some passives and filters I added for testing purposes, as well as to accept different capacitors in the output filter. Other than that it is pretty well finished for now.
:cheers:
Dave
 

Attachments

  • 1KTHD.jpg
    1KTHD.jpg
    105.5 KB · Views: 2,460
  • 18K20K.jpg
    18K20K.jpg
    124.8 KB · Views: 2,360
  • 1KDNR.jpg
    1KDNR.jpg
    112.6 KB · Views: 2,333
  • Jtest_Full.jpg
    Jtest_Full.jpg
    104.7 KB · Views: 2,263
  • 9023_small.jpg
    9023_small.jpg
    147 KB · Views: 2,413
Last edited:
Very interesting plots, thanks for sharing. :)

I notice even low levels of jitter in cymbal ring decay and in the percussive sounds around piano and plucked string instruments.

How do you know you're hearing jitter on these decays? IME I hear artifacts on some DACs, but its not jitter. Are you saying its jitter because you can't think of what else it might be? If you are hearing jitter when the jitter products are below -100dB that's no mean feat.

The last picture is of the DAC board itself.

Looks interesting - may I ask why two transformers, and toroids too? Also is that a mains inlet filter?
 
Dave is using WM8804 SPDIF receiver with 12MHz crystal, and 50MHz XO for the ES9023. See here for more details :

http://www.diyaudio.com/forums/digi...ng-new-ess-vout-dac-es9022-4.html#post2612326

We have a similar setup with WM8804 and ES9022 (Citizen Xtal with mica load caps), but we have also experimented with 48MHz Crystek XO feeding the ES9022 direct, and at the same time feeding the WM8804 after dividing by 4 using 2 high speed single gate Flip Flops from On Semi. Our experience is that the one-clock-drives-all-approach has much better clarity and details.

There are some differences in our approaches and choices of components, and Dave and I have been discussing AB tests to optimise both our implementations. But I am too busy right now and is hindering his progress. Sorry about that, Dave.

Maybe you should also consider publishing some of your objective and subjective comparisons with the ES9018 evaluation board.

I also recently received feedback from someone who had a balanced 9022 module from me, and he clearly preferred the 9022 to his Buffalo II + various IVs. I have asked if he would also publish his impressions some time.

I am the first to say that I want to build discrete for my own reference system, and the 9022/9023 is not the ultimate DAC available on earth. But when done right, it sounds really damn good, and Dave's different approach has shown that neither of us have really found the optimum of this yet. I sincerely hope that more people will join in to experiment, rather than just copy what we are doing. Only that way can we compare notes and results to discover the remaining potentials of these chip.

;)

One of the test that we plan to do soon is a direct AB comparison between 9022 and 9023, using the same setup.


Patrick

.
 
Last edited:
Good job! Where do we sign up for these?
Thanks! No where yet... I am not quite finished and I would really prefer to have some other ears listen to it first. It has also only been connected to two a handful of different systems to audition. I want to hear it in other setups to see if it holds up as well as it has in the limited set so far.

Interesting job! What is the DIR chip of that dac?
Ian
Thank you. It uses a Wolfsen WM8804 in hardware mode. It directly supports 44.1 to 192KHz with the exception of 176.4KHz. I did not want a micro for ths DAC so it seemed a small price.
Did you try spdif sources with different clock jitter?
Ian
I tried a TeraLink, Musiland Monitor 01, and the highface USB to SPDIF interfaces for the function generator. These are just what I happen to have. Each had subtle, measureable differences in noise floor(broadband, random jitter), intrinsic jitter spikes (those there all of the time) and how they reacted to JTEST. There were instances where one worked clearly better than the others in a certain mode. I have not listened to any of these other than briefly to the Hiface. All of the DAC listening tests were done with CDs.

Doing an objective measurement comparison that followed with a blind listening test (by more than one person, and excluding the person doing the measurements of course to minimize bias) would make a very interesting read... Someone with a local “audio support group” should take this on!

More in the next post…
Dave
 
Very interesting plots, thanks for sharing.
My pleasure.
How do you know you're hearing jitter on these decays? IME I hear artifacts on some DACs, but its not jitter. Are you saying its jitter because you can't think of what else it might be?
Good question. I am confident in the conclusion of jitter being the culprit for several of the improvements I made along the way.

It is obviously a subjective measure as I described it, but I was able to back it up with circuit improvements during development that reasonably correlated to measurement improvements. One still baffles me though and I didn't find a correlation in measurements.

I built two boards at each step and always did A/B testing, both subjectively and objectively.
If you are hearing jitter when the jitter products are below -100dB that's no mean feat.
Agreed. My conclusions are based largely on the engineering process and on time spent with a formal education in music. Knowing how these instruments really sound in many different settings is kind of a blessing and a curse…

My hearing is pretty good, but I figure I am hearing the differences caused by time inaccuracy and intermoduation effects in the complex music sounds, not by the direct spurious spikes we can see in the simple repetitive sinewave measurements.

The beauty of listening to instruments like the cymbals in my experience is the absolute predictability of the decay if they are allowed to ring naturally. Top hat is great because of its mix of mid and high frequency components. I love that sound in good recordings. Similarly the upright bass fundamental frequencies in the mid range of the instrument are such a long wavelength (yet still directional) compared to the transient of plucking or the rattle of the string when played loudly that it makes any disparity in the timing or placement of these audible cues easier to pick out.

Looks interesting - may I ask why two transformers, and toroids too?
Absolutely. My intention was to fit it into a small enclosure to match a headphone/pre amplifier I started some time back. I also was trying to get the lowest possible noise from an AC line supply. The toroids reduce the magnetic induced noise and were easily mounted to the PCB.

The requirement for the second toroid is that I included a JFET follower buffer configured as a second order output filter instead of the RC (the R is internal to the 9023) described in the dataheet. I wanted to be able to drive a pot or other volume control directly... The power supply topology was specifically tailored to get the best performance out of the JFETs I chose and required a seperate supply.
Also is that a mains inlet filter?
Yes. It is an off the shelf unit available from Digikey. I wanted a filtered inlet, as well as one that was shielded - the proximity to the follower power supply is fairly close.

Dave
 
Hi Patrick,

Thanks for the link!
There are some differences in our approaches and choices of components, and Dave and I have been discussing AB tests to optimise both our implementations. But I am too busy right now and is hindering his progress. Sorry about that, Dave.
No problem at all. I am also finding it difficult to tear away from the day job (read all day and most of the night) to do this fun stuff. It is just about 1:00AM here now... We will get to it at some point. I look forward to more data.
Maybe you should also consider publishing some of your objective and subjective comparisons with the ES9018 evaluation board.
I started taking some measurements to share but ran out of time. The demo board has a few strengths compared to my version of the 9023 but it really sounds and measures poorly in other ways. The time accuracy I spoke about in the previous post for example is subjectively very good on the 9018 demo board. The sound is harsh though and the apparent sound field small compared to the 9023 board.

Don't get me wrong, I think there is potential in the 9018 and plan to build one of my own at some point with a similar approach to that used with the 9023. The 9023 just does so much right for a budget build.

I am the first to say that I want to build discrete for my own reference system, and the 9022/9023 is not the ultimate DAC available on earth. But when done right, it sounds really damn good, and Dave's different approach has shown that neither of us have really found the optimum of this yet. I sincerely hope that more people will join in to experiment, rather than just copy what we are doing. Only that way can we compare notes and results to discover the remaining potentials of these chip.
I second that!

One of the test that we plan to do soon is a direct AB comparison between 9022 and 9023, using the same setup.
ASAP…

Dave
 
Agreed. My conclusions are based largely on the engineering process and on time spent with a formal education in music. Knowing how these instruments really sound in many different settings is kind of a blessing and a curse…

Sure, comparison with the sound of real instruments is a dead giveaway there's some distortion happening. I prefer voice myself because my own musical experience covers that. But the changes you've been getting probably parallel my own. Whilst I'm not yet sure about time inaccuracy, I am convinced that intermodulation effects are crucial and don't show up on repetitive sinewave testing as you seem to say.

Absolutely. My intention was to fit it into a small enclosure to match a headphone/pre amplifier I started some time back. I also was trying to get the lowest possible noise from an AC line supply. The toroids reduce the magnetic induced noise and were easily mounted to the PCB.

Yeah I agree for radiated magnetics they're amongst the best. However on another measure - conducted noise - they're probably the worst owing to their relatively high primary-secondary capacitance. They can easily be an order of magnitude worse than an EI type on this measure.

Yes. It is an off the shelf unit available from Digikey. I wanted a filtered inlet, as well as one that was shielded - the proximity to the follower power supply is fairly close.

I hope the mains earth in your set up is jolly clean then, not polluted by PCs or LCDs with noisy SMPSUs. :D Myself I prefer not to rely on the earth quality (it sucks where I am) so much in filtering mains common-mode noise.
 
Great Job Dave!! what did you end up with in the filter pps? or np0? i cant see close enough to pick out of those 1210 size are panasonic pps or perhaps rubicon hybrid polymer/film. I think you'll find you take a pretty different tack with the 9018, the demo board is not a good indication of this chips sonic abilities.
 
Sure, comparison with the sound of real instruments is a dead giveaway there's some distortion happening. I prefer voice myself because my own musical experience covers that. But the changes you've been getting probably parallel my own.
Sounds like we are on similar paths. Are you building a DAC based on these?

Whilst I'm not yet sure about time inaccuracy, I am convinced that intermodulation effects are crucial and don't show up on repetitive sinewave testing as you seem to say.
Completely agree on intermoduation distortion. At one point I included a capacitor on the output as an RFI shunt but there was something about the sound that bugged me. Like a slight fuzz in female vocals at moderate listening levels. Almost like a Laryngeal leak or buzz. When I took the time to simulate it in circuit the intermoduation distortion products in the few kilohertz range were increased and easy to find on the real hardware. Removing the capacitor proved a positive, audible and measureable improvement.

What leads me down the path with time accuracy are two personal observations about high frequency information outside of our audible range but that make noticeable difference in the perceived sound quality of music reproduction for me. I noticed at some point that I strongly prefer audio amplifiers and preamps with wider bandwidths over systems limited to the audible range. By comparison it sounds as if the music is truer to being in a room with the instruments. Similarly one of the last changes I made with the 9023 design was comparing an oscillator with decent performance to one that was quite a bit better. The only differences I could measure objectively were increased non-harmonic spurs and slightly decreased normal harmonic content above the audible range with the lesser oscillator. It had a sound signature that made it more difficult to pinpoint where the instruments I use for location cues actually were in the sound field. With the better oscillator, the focus is much better. The objective measurements show an increase in the normal harmonic distortion products but the low level spurious stuff was mostly eliminated.All of this is above 20KHz. The normal harmonics are also related in phase with the fundamental where the spurious stuff is not.

My conclusion from all of this is that since the ears are incapable of "hearing" frequencies above somewhere in the 17-20KHz range but capable to perceive arrival time differences between the two ears outside of this audible bandwidth, time accuracy matters and bandwidth above 20KHz is necessary for reproduction accuracy in stereo or multi channel audio.

Yeah I agree for radiated magnetics they're amongst the best. However on another measure - conducted noise - they're probably the worst owing to their relatively high primary-secondary capacitance. They can easily be an order of magnitude worse than an EI type on this measure..

Agree. This is in part why I use the inlet filter. Im not yet ready to share everything I did for normal mode rejection from earth ground, but it seems to work. I also went to great lengths with the power supply designs to maximize rejection into the hundreds of MHz range. From a few Hz and up the supplies' rejection are > -100dB and increase to two notches targeted at the 12MHz and 50MHz oscillator frequencies. I actually missed the 12MHz notch. It came in at a little better than 9MHz. Close enough though...

It's all overkill but the output noise floor speaks for itself.

I hope the mains earth in your set up is jolly clean then, not polluted by PCs or LCDs with noisy SMPSUs. :D Myself I prefer not to rely on the earth quality (it sucks where I am) so much in filtering mains common-
mode noise.
Nope, quite noisy here too. I took a different approach to addressing this that appears to work well. It needs more testing in other situations to ensure it does not cause ground loop hum, but it has been flawless so far.

Dave
 
Great Job Dave!! what did you end up with in the filter pps? or np0? i cant see close enough to pick out of those 1210 size are panasonic pps or perhaps rubicon hybrid polymer/film. I think you'll find you take a pretty different tack with the 9018, the demo board is not a good indication of this chips sonic abilities.

Thanks qusp.

They are currently the 0805 Panasonics. The larger caps you see are for an experimental shelving circuit that is falling off in the next revision. I will be changing the footprint to accommodate up to 1210 sized parts in the higher voltages. Since this is an LPF (the caps are in parallel) I'm not sure if it matters but I want to try. It would also be easier to solder leaded caps to the larger pads for experimentation purposes.

I'm with you on the 9018. I look forward to building something unique to support that chip and see what it is really capable of. I am slowly collecting parts for the BII/D1 and hope to complete it this year. It will be nice to have comparative measurements and time to listen to it as well.

I don't know that the 9022/23 could be beat for the money though!

Dave
 
Sounds like we are on similar paths. Are you building a DAC based on these?

No - I heard one a couple of months back and it had at least one very endearing quality - huge depth and coherence of sound stage. But on voice it introduced a veiling effect which I wasn't partial to. I've pretty much abandoned low-bit DACs these days for a couple of reasons. One, we're already talking about - intermodulation. They generate very strong out of band signals and its tough to keep those signals from interfering with the audio. The other issue which all are subject to ISTM is noise modulation (ESS acknowledges this and claims theirs is better but still not perfect), and its this effect which I hypothesize is responsible for the voice veiling I heard.

Completely agree on intermoduation distortion. At one point I included a capacitor on the output as an RFI shunt but there was something about the sound that bugged me.

Caps as RF shunts recently came up on John Curl's thread. I mentioned my own experience of them sounding bad (on current out DACs though). Nobody seemed to share that - Charles Hansen said his experience differed but was unwilling to share details. Since the ESS9023 is voltage out its too different a case to make comparisons.

What leads me down the path with time accuracy are two personal observations about high frequency information outside of our audible range but that make noticeable difference in the perceived sound quality of music reproduction for me. I noticed at some point that I strongly prefer audio amplifiers and preamps with wider bandwidths over systems limited to the audible range. By comparison it sounds as if the music is truer to being in a room with the instruments.

My hypothesis for this would be the wider bandwidth amp has lower susceptibility to out of band interfering signals so generates fewer in-band intermod products. IME correct grounding helps improve soundstaging for the same reasons.

Similarly one of the last changes I made with the 9023 design was comparing an oscillator with decent performance to one that was quite a bit better. The only differences I could measure objectively were increased non-harmonic spurs and slightly decreased normal harmonic content above the audible range with the lesser oscillator. It had a sound signature that made it more difficult to pinpoint where the instruments I use for location cues actually were in the sound field.

I've noticed some DACs do give 'pin point locations' for e.g. voices. This is not like real sound - listening to a real singer I have noticed no 'pin point' location effects. With real instruments, real sound sources they do sound a bit diffuse, definitely no sharp edges which is what sometimes happens with DACs.

Agree. This is in part why I use the inlet filter. Im not yet ready to share everything I did for normal mode rejection from earth ground, but it seems to work. I also went to great lengths with the power supply designs to maximize rejection into the hundreds of MHz range.

When you are ready to share details, such will be very interesting :) I've offered some ideas about wide band mains filtering on my blog. I haven't found anyone else so far talking about mains filtering up to 100's of MHz - you're the first.
 
One, we're already talking about - intermodulation. They generate very strong out of band signals and its tough to keep those signals from interfering with the audio
One reason I chose a 2nd order LPF. I did not have space for a 4th order.
My hypothesis for this would be the wider bandwidth amp has lower susceptibility to out of band interfering signals so generates fewer in-band intermod products. IME correct grounding helps improve soundstaging for the same reasons
Good point. I also agree on grounding. It took a while to get it right for the DAC board alone.

My context has been the source for a while and I have been thinking about analog audio in the time domain. In retrospect it makes good sense that mid frequency jitter modulation in the digital to analog conversion would cause distortion similar to (or just like...) IMD in the same band in and above our hearing range, although not necessarily harmonically related. Perhaps this and the time accuracy I described are the same. If the instrument sound is not accurate from transient attack through to the end of its decay the brain may not be able to put the image together cohesively. This is certainly true if the left and right channel are not adding theses artifacts identically.

I've noticed some DACs do give 'pin point locations' for e.g. voices. This is not like real sound - listening to a real singer I have noticed no 'pin point' location effects. With real instruments, real sound sources they do sound a bit diffuse, definitely no sharp edges which is what sometimes happens with DACs.
"Pin point" was a poor choice of words from me. The intention was more about the perceptive act of locating the instrument in the sound field rather than a concrete descriptive of the sound. They should not sound like they are coming from everywhere and unable to be localized. They also should not move around with the other sounds. Cymbals again are a good measure of this performance to my ears as the ring decay will tend to attach itself to other sounds if there are issues like those we have been discussing.

I use instruments with percussive attacks for judging imaging and the time accuracy metric I suggested. Percussion, plucked strings, and piano. I agree - voice 'pin point' location is not realistic to me either. In good recordings it should be localized to its relative source location in the sound field (not seem like it is everywhere) and not move around as the music dynamics and mix of other instruments change.

I like to think of the percieved sound field in three dimensions as well. The depth dimension may well be closer related to time perception in the brain than to stereo imaging processing of the audio in it. Maybe not but an interesting topic for discussion elsewhere.

When you are ready to share details, such will be very interesting I've offered some ideas about wide band mains filtering on my blog. I haven't found anyone else so far talking about mains filtering up to 100's of MHz - you're the first.
I am interested in your thought on mains filtering and will take a look.

My purposeful vagueness my have confused two details in the previous posts - The power supplies themselves have been designed to reject into the hundreds of MHz range so that noise passed through the transformers is rejected. The ground noise approach is a little unconventional. Take the two separately.

If it proves to work as well as it seems to, I'll share some details.
Dave
 
Hi Hp,
about your DJitter measurement...

the base signal is usual 16 bit (no dither) -6db with a LSB toggle (square wave)... and then please zoom in to get the narrow modulated symmetric waves (if any). May look at my web about my measurement using a RME BabyFace..

I used a 48KHz 24 bit version of the JTEST signal for the graph in the original post. This ensures the harmonics from JTEST are under the noise floor. It looks like the fundamental is -3dB.

Here are three additional graphs - first is a 48/24 JTEST capture zoomed in. Second is the 44.1/16 JTEST. Last is a zoom of a 44.1/16 JTEST capture.

Dave
 

Attachments

  • 441_16JtestZoom.jpg
    441_16JtestZoom.jpg
    122 KB · Views: 802
  • 441_16Jtest.jpg
    441_16Jtest.jpg
    113.6 KB · Views: 809
  • 48_24JtestZoom.jpg
    48_24JtestZoom.jpg
    122.2 KB · Views: 922
Hi Hp,


I used a 48KHz 24 bit version of the JTEST signal for the graph in the original post. This ensures the harmonics from JTEST are under the noise floor. It looks like the fundamental is -3dB.

Here are three additional graphs - first is a 48/24 JTEST capture zoomed in. Second is the 44.1/16 JTEST. Last is a zoom of a 44.1/16 JTEST capture.

Dave

What I still missing is a complete zoom in into the +/- 10...100Hz region.. while your bin freq. is about 0.2Hz. And use also some averaging (may 8 times). Compare also with my measurement (currently presented on my web) where I used a 2^25 FFT size. Be also careful not to overdraft with the level. The basic measurement is by J. Dunn - 6.02 dBFS.

Second on your IM measurement: the level is here also very high near dBFS and this caused some IM products 2KHz -78 dBFS and 16K / 22K -92 dBFS.

You where also writing about some output filter caps to get ride of RF. the better design would be to not have/feed any RF on the analog part. This requires a complete redesign. I did this with my DAC project where the DAC & analog part where in complete different & closed copper boxes O;)

Additional if the DAc or the analog part is sensitive to IM product, may consider to bench it using an IM cluster 16..18Khz.

Hp
 
Hi HP,
What I still missing is a complete zoom in into the +/- 10...100Hz region.. while your bin freq. is about 0.2Hz.
Pictures of captures at +/-400Hz and +/- 100Hz are attached below. What are you looking for?


And use also some averaging (may 8 times). Compare also with my measurement (currently presented on my web) where I used a 2^25 FFT size.
The R&S HPV analizer I have has some limitations. I cannot do greater than 256K points and averaging in the FFT window is not implemented. It is what I have so I use it... At some point I will probably upgrade.


Be also careful not to overdraft with the level. The basic measurement is by J. Dunn - 6.02 dBFS.
The JTEST files I have were not generated by me. Can you provide JTEST wave files with the appropriate level? 16 and 24 bit files at 44.1 and 48KHz would be appreciated. 60 seconds in length is fine.


Second on your IM measurement: the level is here also very high near dBFS and this caused some IM products 2KHz -78 dBFS and 16K / 22K -92 dBFS.
Each tone is set at -6dBFS for that graph. What are you advocating for levels of each tone?


You where also writing about some output filter caps to get ride of RF.
The purpose of the caps in that experiment was not RF from the DAC but from external sources and the interconnects. It was directly at the output after the JFET LPF and current limiting resistor. Implementing faraday shields and other design changes to mitigate RF at the design level are extreme overkill for a design based on a $2US DAC chip. Just my opinion here.

Additional if the DAc or the analog part is sensitive to IM product, may consider to bench it using an IM cluster 16..18Khz.
I am unaware of this testing method. Can you describe it further?


Dave
 

Attachments

  • 441_16JtestZoom_400.jpg
    441_16JtestZoom_400.jpg
    137.9 KB · Views: 756
  • 441_16JtestZoom_100.jpg
    441_16JtestZoom_100.jpg
    134.9 KB · Views: 754
Hi,

>> I cannot do greater than 256K points and averaging in the FFT window is not implemented. It is what I have so I use it... At some point I will probably upgrade.

>> The JTEST files I have were not generated by me.


>> I am unaware of this testing method. Can you describe it further?

IM Cluster : 15kHz ...16kHz range using 6 frequencies some with 100Hz and some with 200Hz separated

All are invited to evaluate my SW ;) while all noted items above, are known implemented features ...

Hp
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.