Mark, Are you saying that 9038Q2M sounds better with SRC4392 ASRC than without? That is interesting because AFAIK the ESS DAC has internal ASRC
which actually has better (lower) jitter rejection corner frequency than 4392.
I believe in the lowest setting the 9038 has around 0.1Hz CF for jitter rejection, don't know the slope and SRC4392 is around 10Hz but the slope is quite good.
The other issue is how the digital filters compare.
WRT Cranesong DAC, I don't think it uses 4392 as the ASRC is noted as a 32 bit part. Have you had a chance to compare Cranesong with Benchmark DAC3?
The Cranesong has without a doubt a better clock. The phase noise of his 27MHz clock is superb.
Terry
The primary reason for upsampling before going into the 9038Q2M is so that the minimum phase slow transition reconstruction filter can be used without being audible. It keeps the filter transition band above the audio.
I know SRC4392 is not perfect. I did a test with some software upsampled CD rips. 4392 cannot beat Weiss Saracon in terms of sound quality, but it sounds better than not using it and probably the best hardware upsampler readily available.
EDIT: Also in terms of jitter rejection the USB to SPDIF converter I am using has decent clocks, although not great. There may not be lot of jitter to reject by the 4392 when used with that particular source sitting right next to it. Haven't tried to see if I2S is noticeably better for that link.
Next thing I want to test out is whether Q2M sounds better with its jitter correction off if I can sync the 4392 output to the Q2M BCLK. Maybe one SRC is better than two if I can get the data there on time.
The DAC in my old Crane Song HEDD 192 is not the same as the new Quantum version. The old one sounds pretty bad and I never used it. I like the sound of the ADC a lot though which is why I kept the box.
I know SRC4392 is not perfect. I did a test with some software upsampled CD rips. 4392 cannot beat Weiss Saracon in terms of sound quality, but it sounds better than not using it and probably the best hardware upsampler readily available.
EDIT: Also in terms of jitter rejection the USB to SPDIF converter I am using has decent clocks, although not great. There may not be lot of jitter to reject by the 4392 when used with that particular source sitting right next to it. Haven't tried to see if I2S is noticeably better for that link.
Next thing I want to test out is whether Q2M sounds better with its jitter correction off if I can sync the 4392 output to the Q2M BCLK. Maybe one SRC is better than two if I can get the data there on time.
The DAC in my old Crane Song HEDD 192 is not the same as the new Quantum version. The old one sounds pretty bad and I never used it. I like the sound of the ADC a lot though which is why I kept the box.
Last edited:
Apologies for the delayed response. But confirming that the circuit is correct. As long as the two resistors are the same - you will get no DC offset
Measurements on AudioScienceReview
Measurements Of Generic ES9038Q2M DAC Board | Audio Science Review (ASR) Forum
Measurements Of Generic ES9038Q2M DAC Board | Audio Science Review (ASR) Forum
Thanks for the link; as I suspected, cleaning up the power and putting a proper I/V on the back end (acknowledging Vacuphile's experiments) closes the loop tidily. Everything else points to tweaking things for the sake of.
ESS claims the Q2M can reach -120dB distortion. From the looks of it at this point it will take harmonic distortion correction to get down that far. With good hardware circuit design it should be able to get down below -110 it would seem, maybe a bit more. Don't think distortion correction is likely to give much more than right around what cdgames reported, which was about a 5dB improvement for each of the 2nd and 3rd partials. Maybe a little less than 5dB each.
However, THD/IMD nonlinear distortion is not the only type of problem that can adversely affect sound quality. There is Linear distortion particularly in the form of audible reconstruction filter characteristics including group delay (wha guitar pedal stuck-in-one-position type of phasey sound; rate of change of phase with respect to frequency). Upsampling and using a low group delay reconstruction filter improves that aspect of sound quality. There is also jitter that can produce an ugly uncorrelated-with-signal type of noise, and if bad enough can also affect L/R stereo imaging.
The point I am trying to make is that we can and should do the best we can to improve nonlinear distortion issues that detract from sound quality, and we should measure to the extent we are able our results for that.
However, it is well known that THD as a measure correlates very poorly with perceived sound quality. The same THD number can result from nearly all very low order harmonics or at the other extreme a lot of higher order harmonics. The latter usually sound much worse than the former. Similarly, although we can measure phase and group delay, we don't have a good measure that correlates well with perceived sound quality.
Because I have tried a lot of listening experiments with a reference system, with software vs hardware processing, and so forth, I keep hammering on the idea the SRC should not be ignored. Getting as much as that ugly reconstruction filter group delay out of the audio band makes for a BIG sound quality improvement, at least as compared to making big improvements in nonlinear distortion and jitter, as two examples.
Story: The DAC-3 has been here for awhile but it has been used with an old Bryston 4B power amp up until very recently. Why up til recently? Because the Q2M DAC project with headphone amp started sounding better than the reference system! That meant the time had come to retire the Bryston and replace it with something better. The better turned out to be a Benchmark AHB2 power amp. Wow, very big difference. Now the reference system sounds considerably better than DAC project and headphone amp. The result of that is I can better calibrate where the DAC project sound quality is at this point. The Q2M is never going to match the DAC-3 in performance since it is not possible to offload interpolation and reconstruction filtering to external DSP with a Q2M. Also, Q2M noise will never be as low as 9038PRO. Lastly, an LME49600 will never equal the accuracy of the AHB2 (although it can be pretty darn good, it is what the DAC-3 uses for the internal headphone amp).
Since I was able to get rid of the Silent Switcher, costs for DAC project building blocks list is reduced to only the following.
Chinese DAC $39
SRC4392 $60
Headphone amp kit $32
USB to SPDIF/I2S $28
(Probably) Arduino 3.3v Pro Trinket $10
(Probably) FTDI for Trinket $7-12 (only needed for development or debug)
Maybe $180 plus odds and ends and +-15v linear power supply
The result will sound way better than something like a DACmagic+ or pretty much anything under $1,000 (likely more than that, maybe until the Allo Katana DAC by cdgames will be pretty similar, haven't had the pleasure of listening yet). By way of comparison, the reference system DAC amd amp MSRP would probably come closer to a little over $5k. So, you can get pretty close, but not quite there, to the SQ of a very $$$ for only about maybe less than $250?
I haven't really followed other DIY projects here so I have no point of reference, but it just seems like, with the little bit of low-cost DIY-style engineering and modest-for-what-you-end-up-with parts cost, it seems like it ought to be attractive to hobbyists. So frankly, I am curious that no one has said that they want to try SRC and or LME49600 amp. Is it a matter of cost, complexity, risk, something else? Don't really want such a thing? Any replies and comments welcome, either here or by PM. I'm just trying to learn as I go in various respects, including about what DiY'ers here want or would like. Don't know if I will ever do another project here, just got started on this one out of a desire to try to help people who want a very good reproduction system at as low a cost as possible. You know, if nobody wants to build one it hasn't really been all that helpful, has it?
EDIT: Of course, what I have written was not intended to come right after DPH's post. We were probably both typing at the same time. We may disagree on some of what I have said. If so, I could only offer an invitation to come by for a vist and take a listen if ever out this way. That is all I will say on that.
However, THD/IMD nonlinear distortion is not the only type of problem that can adversely affect sound quality. There is Linear distortion particularly in the form of audible reconstruction filter characteristics including group delay (wha guitar pedal stuck-in-one-position type of phasey sound; rate of change of phase with respect to frequency). Upsampling and using a low group delay reconstruction filter improves that aspect of sound quality. There is also jitter that can produce an ugly uncorrelated-with-signal type of noise, and if bad enough can also affect L/R stereo imaging.
The point I am trying to make is that we can and should do the best we can to improve nonlinear distortion issues that detract from sound quality, and we should measure to the extent we are able our results for that.
However, it is well known that THD as a measure correlates very poorly with perceived sound quality. The same THD number can result from nearly all very low order harmonics or at the other extreme a lot of higher order harmonics. The latter usually sound much worse than the former. Similarly, although we can measure phase and group delay, we don't have a good measure that correlates well with perceived sound quality.
Because I have tried a lot of listening experiments with a reference system, with software vs hardware processing, and so forth, I keep hammering on the idea the SRC should not be ignored. Getting as much as that ugly reconstruction filter group delay out of the audio band makes for a BIG sound quality improvement, at least as compared to making big improvements in nonlinear distortion and jitter, as two examples.
Story: The DAC-3 has been here for awhile but it has been used with an old Bryston 4B power amp up until very recently. Why up til recently? Because the Q2M DAC project with headphone amp started sounding better than the reference system! That meant the time had come to retire the Bryston and replace it with something better. The better turned out to be a Benchmark AHB2 power amp. Wow, very big difference. Now the reference system sounds considerably better than DAC project and headphone amp. The result of that is I can better calibrate where the DAC project sound quality is at this point. The Q2M is never going to match the DAC-3 in performance since it is not possible to offload interpolation and reconstruction filtering to external DSP with a Q2M. Also, Q2M noise will never be as low as 9038PRO. Lastly, an LME49600 will never equal the accuracy of the AHB2 (although it can be pretty darn good, it is what the DAC-3 uses for the internal headphone amp).
Since I was able to get rid of the Silent Switcher, costs for DAC project building blocks list is reduced to only the following.
Chinese DAC $39
SRC4392 $60
Headphone amp kit $32
USB to SPDIF/I2S $28
(Probably) Arduino 3.3v Pro Trinket $10
(Probably) FTDI for Trinket $7-12 (only needed for development or debug)
Maybe $180 plus odds and ends and +-15v linear power supply
The result will sound way better than something like a DACmagic+ or pretty much anything under $1,000 (likely more than that, maybe until the Allo Katana DAC by cdgames will be pretty similar, haven't had the pleasure of listening yet). By way of comparison, the reference system DAC amd amp MSRP would probably come closer to a little over $5k. So, you can get pretty close, but not quite there, to the SQ of a very $$$ for only about maybe less than $250?
I haven't really followed other DIY projects here so I have no point of reference, but it just seems like, with the little bit of low-cost DIY-style engineering and modest-for-what-you-end-up-with parts cost, it seems like it ought to be attractive to hobbyists. So frankly, I am curious that no one has said that they want to try SRC and or LME49600 amp. Is it a matter of cost, complexity, risk, something else? Don't really want such a thing? Any replies and comments welcome, either here or by PM. I'm just trying to learn as I go in various respects, including about what DiY'ers here want or would like. Don't know if I will ever do another project here, just got started on this one out of a desire to try to help people who want a very good reproduction system at as low a cost as possible. You know, if nobody wants to build one it hasn't really been all that helpful, has it?
EDIT: Of course, what I have written was not intended to come right after DPH's post. We were probably both typing at the same time. We may disagree on some of what I have said. If so, I could only offer an invitation to come by for a vist and take a listen if ever out this way. That is all I will say on that.
Last edited:
Mark, in what I hope comes across completely conciliatory: I hope we agree that in terms of dominant forms of improvement of measurable performance, improving the power supply and addressing the I/V circuit (output buffer). These are the two most likely places to affect a significant change in overall DAC behavior, even if they're not comprehensive.
While I have a hard time seeing any time/$ ROI past that, I can acknowledge that we're going to have significant differences in opinion about clocking (-120 dB spurs is pretty dang good though!), custom resampling, etc.
While I have a hard time seeing any time/$ ROI past that, I can acknowledge that we're going to have significant differences in opinion about clocking (-120 dB spurs is pretty dang good though!), custom resampling, etc.
Then why not just get something like the Topping D50 for $250? Comes with Dual ES9038Q2M, Xmos XU208, OPA1612 opamps. The parts quality and layout is probably better than these cheap diy boards and the 100mhz clock looks relatively easy to replace by a diy'er.
Some measurements here (for what they're worth).
I love playing with cheap dac boards, but I'm not seeing the value when you have something like the Topping out there. Am I missing something?
Last edited:
Regarding Topping D50, I can't tell what it sounds like from looking at picture and reading specs. However, it may be quite good. Maybe get a little better idea with some hi-res close-up photos of the inside, good enough to read part numbers. Best to listen though.
If you think the 100MHz clock perhaps ought to be replaced to improve SQ, already I would say it would probably also sound better with upsampled input and a use of the slow minimum phase reconstruction filter. To do that would probably require another USB to SPDIF converter to feed the SRC unless its possible to attach to the right signals inside. Maybe some upgrade costs for it too?
It would be easy to do some experiments to see what could be discovered. HQ Software upsampling a very good SQ CD rip, selecting the slow filter and sending it to the D50 would probably be informative. Also informative to directly compare with something like a DAC-3. If somebody needs a good quality upsampled rip for testing, I might be able to upsample your rip a few different ways for comparison. Interesting that differences can be heard, including what sounds like some group delay from some very low THD upsamplers.
Having two Q2Ms instead of one should reduce noise by 3dB. One Q2M can already come within 2dB of a 9038PRO in terms of distortion, so probably no real advantage there.
It might even turn out this DAC upgrade project sounds better, and can have a headphone amp along with it. I wouldn't be surprised if it could, but at the same time there is no way to know for sure without listening.
If you think the 100MHz clock perhaps ought to be replaced to improve SQ, already I would say it would probably also sound better with upsampled input and a use of the slow minimum phase reconstruction filter. To do that would probably require another USB to SPDIF converter to feed the SRC unless its possible to attach to the right signals inside. Maybe some upgrade costs for it too?
It would be easy to do some experiments to see what could be discovered. HQ Software upsampling a very good SQ CD rip, selecting the slow filter and sending it to the D50 would probably be informative. Also informative to directly compare with something like a DAC-3. If somebody needs a good quality upsampled rip for testing, I might be able to upsample your rip a few different ways for comparison. Interesting that differences can be heard, including what sounds like some group delay from some very low THD upsamplers.
Having two Q2Ms instead of one should reduce noise by 3dB. One Q2M can already come within 2dB of a 9038PRO in terms of distortion, so probably no real advantage there.
It might even turn out this DAC upgrade project sounds better, and can have a headphone amp along with it. I wouldn't be surprised if it could, but at the same time there is no way to know for sure without listening.
I have one on the way. I don't know that the clock needs swapping out necessarily, but it wouldn't be diyaudio if we didn't do *some* modding, right?
If you're in Cali, I can send mine down to you to play with if you'd like once it arrives.
I am in Cali. If you send it to me and I find it sounds better with SRC, I will send it back to you with an SRC board for you try out.
I'm not following you here. If you are concerned about group delay, why not simply use a linear phase filter in the DAC?
Mostly because of the pre-ringing. Percussion instruments at low frequencies usually don't sound right to me. Slow, gentle minimum phase filtering usually sounds best whenever possible to use it. Steep transition filters are hard to get right, although there are a few examples of very clever expert engineering.
Sure some up-samplers are done better than others, but you have the exact same issues as with conventional oversampling i.e. trade-offs with stop-band attenuation, ringing etc. Unfortunately they are not a magic bullet.
I don't know how the "jitter eliminator" ASRC on the latest chips compares to the ES9008/9018 parts, but my experiments showed that at least the earlier chips benefit form a low jitter source. Years ago I fitted a DIR with analog PLL before my ES9008 and got a useful improvement. Many others reported similar experiences, so I'm not surprised it sounds better with the TI SRC part you are using.
I don't know how the "jitter eliminator" ASRC on the latest chips compares to the ES9008/9018 parts, but my experiments showed that at least the earlier chips benefit form a low jitter source. Years ago I fitted a DIR with analog PLL before my ES9008 and got a useful improvement. Many others reported similar experiences, so I'm not surprised it sounds better with the TI SRC part you are using.