I will try the CAP suggestion.
Microcontroller separation: Unfrotunately I can't tell.
At the beginning I made all the changes in one step. (Power complete separation 7808+6pcs LP5907 + OPA change) and these modifications in total had an enormous change to the original sound - which was very disappointed.
Sorry that I can't tell the effect of this singular step.
(The next big result derived from the CAP changes.)
Sz.
Replacing the clock oscillator with an ultra-low jitter type helped sound quality similar to doing the AVCC power mod, and the IV current mode with differential summing output stage. The other thing that helps a lot is upsampling the digital input.
Excellent advice on AVCC power and output stages can be found in some ESS documents, IMHO. They have a downloads page, and although some of the information probably dates back to the ES9018, it still seems to be applicable. I recommend studying what they have to say here: ESS Technology :: Downloads In particular, the "maximizing dad performance for every budget" document is very useful, again IMHO. The old evaluation board schematic shows how they got the output stage reference voltage from AVCC.
The upsampling board I use and recommend and that helps sound quality a lot, IMHO, uses an SRC4392 chip which also helps with jitter reduction: Buy src4392 board and get free shipping on AliExpress.com
I recommend a good headphone amp for listening tests. Having something that is low enough distortion can help a lot to be sure about differences that are heard. A cheap $32 ebay board can be modded for very low distortion. Somewhere in this thread I posted pictures of one way to do the mods to make it a good companion for the DAC: 1PCS LME49720NA+LME49600 headphone amplifier KIT | eBay
In addition to AVCC power quality having a big effect on sound (and I am hoping eventually someone will try building and comparing some different AVCC circuits including the ESS recommended one, maybe LT3045, LifePO4 batteries, capacitor upgrades, etc.), the +-15 volt power for the opamps needs to be very clean, and it helps a little for the clock to be on its own regulator since it is sort of an analog device. The digital circuitry doesn't seem to need a lot of attention to special power supplies, IME.
In summary, there is:
*high quality external linear power supply
*AVCC on-board power regulator (ESS recommended one is good)
*Clock upgrade, and maybe dedicated clock power voltage regulator
*opamp IV stages with differential combining to make single-ended outputs
*upsampling SPDIF or TOSLINK to 192kHz or 96kHz, maybe with minimum phase, slow-transition reconstruction filter (That filter sounds good and is probably the best choice with upsampled input. There are technical reasons explained in the thread or we could discuss again if there is interest.)
Optional:
*Modded headphone amp for low-distortion mod comparisons, and of course for music enjoyment.
And if driving the DAC from a computer:
*Set computer default audio sample rate to the same sample rate as the file you want to play back, otherwise the OS will real-time resample you music using a low quality algorithm. I can post pictures of how to it in Windows from the Control Panel. Unfortunately, it has to be changed every time you want to play something from a different sample rate file.
However, if using Foobar with ASIO drivers, and an extra soundcard not the Windows default sound device, it is sometimes possible to avoid automatic resampling of music playback.
Most or all of the above items have been discussed in the thread, in some cases multiple times. There are close-up, in-focus pictures of many mods. We can cover any of it again if there is new or recurring interest.
One other thing I didn't mention above includes making an Arduino 'shield' interface to the control the DAC over I2C bus. The SRC4392 hardware sample rate converter can also be controlled over I2C. I have written some programs to do various functions with both ES9038Q2M and SRC4392. One of the most recent includes read/write access to any of the Q2M registers and adjustment of the double precision harmonic distortion compensation registers. The double-precision read/write signed-integer conversion is done in the program.
As it happened I used a 3.3v Pro Trinket Arduino to avoid any concerns about level translation, although that could be handled easily enough too. But, at the moment the code runs on that particular platform.
As it happened I used a 3.3v Pro Trinket Arduino to avoid any concerns about level translation, although that could be handled easily enough too. But, at the moment the code runs on that particular platform.
Last edited:
I see I have some possiblities to further develop.
I think you know the 80-20 rule. So I see I put 20% and achieved 80%, still the rest requires a huge effort. I will decide what next and to be done.
really thanks a lot for your SUM. I will check the former posts - but it is a good intention and direction for me.
Sz.
I think you know the 80-20 rule. So I see I put 20% and achieved 80%, still the rest requires a huge effort. I will decide what next and to be done.
really thanks a lot for your SUM. I will check the former posts - but it is a good intention and direction for me.
Sz.
Regarding the 80/20 rule, I would say my experience with this DAC is you put in 20% and got may be to 30%-40% at most.
When I started I hoped to get to 80% quickly with a few mods. Didn't turn out that way. It turned out that all the mods of different things each help similar amounts. The only one that didn't help a lot was the separate clock voltage regulator. Only thing I can say is the 80/20 rule doesn't seem to work for some things, and this is one. Sorry.
When I started I hoped to get to 80% quickly with a few mods. Didn't turn out that way. It turned out that all the mods of different things each help similar amounts. The only one that didn't help a lot was the separate clock voltage regulator. Only thing I can say is the 80/20 rule doesn't seem to work for some things, and this is one. Sorry.
Did you try Salas BiB to power the osc? Powering the osc, the 3.3 V Salas BiB shunt improved my DAC a lot.
By the time I got the the clock power, most of the power on the board was already cleaned up a lot. Probably not much more to do by then. The separate clock power regulator did smooth out the highs slightly. But the DAC was already quite good sounding and measuring very low distortion by then. The clock had already been upgraded to Crystek too. Possibly it has better PSRR, don't know.
LifePO4 batteries is an interesting solution .
we find 3.2v or 3.3v batteries on ebay but it's difficult to find chargers for 3.2v LifePO4
Some of the automatic chargers work with them, such as this one:
Amazon.com: Universal Battery Charger, EASTSHINE S2 LCD Display Speedy Smart Charger for Rechargeable Batteries Ni-MH Ni-Cd AA AAA Li-ion LiFePO4 IMR 10440 14500 16340 18650 RCR123 26650: Electronics
You do have to be careful not to let them discharge too much. From looking at some LifePo4 battery data sheets it looks like you have to stop using them and recharge before they get down to 2volts. There are limits to charging voltage, although don't remember exactly at the moment what the numbers were. If charging them on an automatic charger I would just look to make sure it correctly identifies them and doesn't confuse them with another battery type. Shouldn't be a problem when they are discharged but when fully charged the voltage can be a little higher than normal until a very small amount of discharge. It might be then that a charger could confuse them for a higher voltage battery. The above observations are from watching my own charger with some new batteries. The package the batteries came in said to charge them before use, but they were already charged to the max as it was (although I didn't know it at the time), and more than LifePo4 batteries usually are. I think charger thought they were 3.7v batteries rather than 3.2v so I took them out and decided to ignore the instructions on the package to charge before use. After seeing what happened with the charger, I measured the voltage it was a little above 3.3v, but again don't remember exactly how much more, it might have been 3.35v or so. Same for the other new ones that had not been in the charger even briefly. Apparently it was enough to fool the automatic charger with the first battery I tried. Maybe I should have measured first before following the instructions on the package. At least I took the battery out of the charger pretty quickly, and apparently no damage.
Actually, a search turned up the post where I had some pictures to show of the mods I did for for my HPA. Please let me know if this is helpful or if I need to clarify anything: http://www.diyaudio.com/forums/digital-line-level/314935-es9038q2m-board-63.html#post5402398
EDIT: The only thing I would say about the picture of the test setup at the time was that I was testing a little switching supply when I took the picture. I decided it failed testing and went back to a linear supply. So, just to be clear, I do not recommend the power supply shown in the picture. I do however, stick with my opinions about the headphone amps. The opinions are my opinions with with a very clean linear supply, or any other supply I tested. Unfortunately, I can't go back and edit the picture at this point.
Last edited:
Simple Voltage Regulators Part 1: Noise
no data on this particular type of battery but it does appear batteries are not as clean as many think provided you implement your SS reg properly.
On the noise question: Right, I think I brought up before that batteries produce some noise. I was assured in an arm-waving sort of way that LifePo4 batteries have very low ESR and therefore should be low noise. To the extent the equivalent circuit model holds, they should be very low-noise. However, I don't know how well the simple model holds and it could be the batteries are more noisy than some may believe.
It would be great if someone would like to compare (1) the ESS opamp AVCC circuit with (2) LT3045 modules and with (3) batteries that people tend to see as attractive options. Maybe a (4) shunt regulator would be good to compare too. And maybe they are all fine or some fine, I just don't know and I would like to see a comparison done over a very short time period before aural memory decays.
Also, you guys know the trick not to be fooled by loudness differences, right? In case anybody does't know about this, even very tiny loudness differences as little as .1dB are enough to make one thing sound better than another. So, what I always do when comparing to files, sources, caps, amps, whatever, it turn the volume knob up and down and listen for what changes with loudness and what doesn't. Soundstage generally does. Depth, energy, punch, bass compared to midrange, highs compared to mids, etc., all change with loudness. Also happens with nonlinear processing like compression (not file compression, but a compressor that alters average level compared to peak level).
Anyway, the only way I know of is to learn to listen for the things about a sound that don't change when the volume knob is turned. Distortion produced by some process occurring before the volume control does not change, for example. It's still there just the same. Just so long as the distortion isn't added after the volume control. When I say distortion, I don't mean only nonlinear distortion, but also linear distortion. Group delay is a type of linear distortion that does not change with volume, for example. The frequency response differences that change with volume are due to Fletcher-Munson effect, although the famous, published everywhere Fletcher-Munson curves are wrong. The research was repeated later by others who found the curves were somewhat different, enough different to not use the old curves. Also, the curves are statistical averages for how people hear, and you may not be exactly average. If there is something about frequency response that is different when comparing two sounds that turning up and down the volume leads you to believe Fletcher-Munson doesn't or can't account for it, you might be right. Just have to be very careful about it. It is very easy to for the effect to throw you off is all.
It would be great if someone would like to compare (1) the ESS opamp AVCC circuit with (2) LT3045 modules and with (3) batteries that people tend to see as attractive options. Maybe a (4) shunt regulator would be good to compare too. And maybe they are all fine or some fine, I just don't know and I would like to see a comparison done over a very short time period before aural memory decays.
Also, you guys know the trick not to be fooled by loudness differences, right? In case anybody does't know about this, even very tiny loudness differences as little as .1dB are enough to make one thing sound better than another. So, what I always do when comparing to files, sources, caps, amps, whatever, it turn the volume knob up and down and listen for what changes with loudness and what doesn't. Soundstage generally does. Depth, energy, punch, bass compared to midrange, highs compared to mids, etc., all change with loudness. Also happens with nonlinear processing like compression (not file compression, but a compressor that alters average level compared to peak level).
Anyway, the only way I know of is to learn to listen for the things about a sound that don't change when the volume knob is turned. Distortion produced by some process occurring before the volume control does not change, for example. It's still there just the same. Just so long as the distortion isn't added after the volume control. When I say distortion, I don't mean only nonlinear distortion, but also linear distortion. Group delay is a type of linear distortion that does not change with volume, for example. The frequency response differences that change with volume are due to Fletcher-Munson effect, although the famous, published everywhere Fletcher-Munson curves are wrong. The research was repeated later by others who found the curves were somewhat different, enough different to not use the old curves. Also, the curves are statistical averages for how people hear, and you may not be exactly average. If there is something about frequency response that is different when comparing two sounds that turning up and down the volume leads you to believe Fletcher-Munson doesn't or can't account for it, you might be right. Just have to be very careful about it. It is very easy to for the effect to throw you off is all.
Last edited:
Pretty sure links were provided to measurements/assessments done by others on this forum. They are by no means a magic bullet, there are many options - but they have been discussed at diyaudio at length over the years.
I can see where all this is heading, so I'll unsubscribe.
I'm sorry, I don't understand that
It is not meant to head anywhere. Maybe you could point us to some of the work that was done before on batteries. I try to remember as much as I can but I am certainly not perfect. Also, when I said arm-waving I was NOT thinking of you. It was just an impression I was left with at some point. It could be that after that you did point to some measurements. I don't remember.
Well, if I have made myself persona non grata, it was inadvertent. Sorry.
But, I will go ahead and risk digging myself into an even deeper hole.
The things I mentioned above apply mainly to monaural aspects of listening comparisons. Stereo brings another dimension where timing and time delays are important, so I would like to say a few things about making comparisons of two sounds with respect to stereo.
Here goes:
Stereo imaging is partly due to volume differences, but also in live performances and in recordings made with simple two-mic setups it turns out that timing differences very much affect perceived location in a stereo field.
What I do is sit about 3 feet from the speaker and with the speakers about 3 feet aparte from each other. That is near field listening setup often used for mixing records. For stereo imaging it is important to be in the near field where you hear sound from the speakers directly and as free as possible from any room reflections or reverberation. Reflections off the floor and tables, etc., can all interfere with near field listening.
Sounds in a stereo field have a position that is supposed to be perceived as coming from between the speakers. It is possible for that illusion to work very well, but timing accuracy is critical. I close my eyes and listen. In the worst case the sound seems to be coming from two distinct speakers. As stereo imaging gets better a sound starts to be perceived as coming from somewhere between the speakers. With my eyes closed I try to point my finger at the spot between the speakers where some instrument or sound seems to be coming from. I note how how clearly defined that location appears to be. I also note its apparently width. With a very good DAC the point is space where a sound seems to be coming from can very narrow, maybe only 1 inch wide or less. Sometimes a sound can seem to be spread out across a large part of the distance between the speakers. Time jitter can affect that perception, and human hearing includes some ability to localize a sound source laterally to a degree that requires timing differences much smaller than corresponds to the highest frequencies humans can hear. In other words, for hearing, the relationship between time domain perception and frequency domain perception is not entirely linear.
Anyway, the idea is to learn to listen with eyes closed to estimate sound source location precision and perceived width of the sound source. While that may relate to something some people might call a stereo sound stage or something, my feeling is that it is better to try to learn to hear it and describe in physical terms that engineers can understand and that can probably be measured with instruments. Understanding in technical terms and in measurable terms is important because it helps designers understand what they need to work on. And it can help to convince some skeptical type engineers that the perceived difference is real and they should probably give the credence to the report. It also is a way when doing one's own comparisons that tries to separate out loudness and Fletcher-Munson type effects from the equation. It is listening for something that does not change with volume.
But, I will go ahead and risk digging myself into an even deeper hole.
The things I mentioned above apply mainly to monaural aspects of listening comparisons. Stereo brings another dimension where timing and time delays are important, so I would like to say a few things about making comparisons of two sounds with respect to stereo.
Here goes:
Stereo imaging is partly due to volume differences, but also in live performances and in recordings made with simple two-mic setups it turns out that timing differences very much affect perceived location in a stereo field.
What I do is sit about 3 feet from the speaker and with the speakers about 3 feet aparte from each other. That is near field listening setup often used for mixing records. For stereo imaging it is important to be in the near field where you hear sound from the speakers directly and as free as possible from any room reflections or reverberation. Reflections off the floor and tables, etc., can all interfere with near field listening.
Sounds in a stereo field have a position that is supposed to be perceived as coming from between the speakers. It is possible for that illusion to work very well, but timing accuracy is critical. I close my eyes and listen. In the worst case the sound seems to be coming from two distinct speakers. As stereo imaging gets better a sound starts to be perceived as coming from somewhere between the speakers. With my eyes closed I try to point my finger at the spot between the speakers where some instrument or sound seems to be coming from. I note how how clearly defined that location appears to be. I also note its apparently width. With a very good DAC the point is space where a sound seems to be coming from can very narrow, maybe only 1 inch wide or less. Sometimes a sound can seem to be spread out across a large part of the distance between the speakers. Time jitter can affect that perception, and human hearing includes some ability to localize a sound source laterally to a degree that requires timing differences much smaller than corresponds to the highest frequencies humans can hear. In other words, for hearing, the relationship between time domain perception and frequency domain perception is not entirely linear.
Anyway, the idea is to learn to listen with eyes closed to estimate sound source location precision and perceived width of the sound source. While that may relate to something some people might call a stereo sound stage or something, my feeling is that it is better to try to learn to hear it and describe in physical terms that engineers can understand and that can probably be measured with instruments. Understanding in technical terms and in measurable terms is important because it helps designers understand what they need to work on. And it can help to convince some skeptical type engineers that the perceived difference is real and they should probably give the credence to the report. It also is a way when doing one's own comparisons that tries to separate out loudness and Fletcher-Munson type effects from the equation. It is listening for something that does not change with volume.
Last edited: