My ES9038Q2M DAC board

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


When i make time , i wil report how the sound is.


I have the board.
 

Attachments

  • y1.jpg
    y1.jpg
    544.3 KB · Views: 1,274
Well, I found the answer myself on an Ebay ad

VOLUME: Please remove the short circuit resistance when you need the volume function, and connect a 10K single potentiometer condition. This function uses the internal ADC of the microcontroller to control the volume of the ES9038Q2M with 64 levels.

MCLK1: ASRC setting, if the ON position is 1, the ASRC is enabled. The ASRC is enabled by default. Since we use the synchronous clock, there will be no asynchronous oversampling.

MCLK2: clock selection, ON position is to choose 45.1584/49.152MHZ, 0 position is 22.5792/24.576MHZ

DF1/DF2: Internal FIR filter selection 00-apodizing, 10-brick wall, 01-slow, 11-fast
 
My board arrived and it looks almost identical to Democles.
681827d1526816254-es9038q2m-dac-board-y1-jpg




The bulk caps are different - mine are 5x Jamicon 1000uF 35V. The electros near the op amps are also different - branded Rubycon 3MO246 10uF 25V non polar. Rubycon's catalogue doesn't have "non-polar" caps because they refer to these as bi-polar. They look like these

An externally hosted image should be here but it was not working when we last tested it.


Apart from that, the two boards seem to be identical.
The board has two bridge rectifiers - both marked MB40M. The datasheet says they are nothing special with a forward voltage of 1.1V and 50pF input capacitance. It doesn't give recovery time.
The PCB is clearly multiplane so I assume the ad is correct and it is 4 plane, and it's obvious the analogue and digital sections have their own ground planes so that's good news.
The power regulation for the digital section has an LM317 with a 7.5R resistor so it seems to be a constant current source of 166mA. This feeds a shunt regulator for a 7.5V supply to the four LT3042 regulators. This is very good news - the CCS + shunt as a pre-regulator should get the best out of the LT3042.
The XO for the Bravo SA8804 is marked as 12.000 (mHz) and "748". If anyone recognises the brand, please let me know what it is. Thanks!
An externally hosted image should be here but it was not working when we last tested it.


The op amps power is +/-12.5V from a 317/337 combo. It has 220R between output and base and 2kR between base and ground, in parallel with a tantalum cap. The output cap is 220uF Panasonic FC.
 
Last edited:
This is the analogue output circuit. I have a bunch of different op amps to roll in this circuit, and I'm going to use a dual TPA6120 headphone amp with differential input and gain of 3.7 to listen to the DAC on HD650 cans. This DAC seems to have very little opportunity/need for modifications, except the power for the op amps.

An externally hosted image should be here but it was not working when we last tested it.


Besides the brand of the XO, another thing I'm stuck on is the chinese print screen next to four empty through holes - in the top right corner near the DSD and Lock LEDs. The first character is "setting" but I don't know the second one. It doesn't mean filter or volume. DMM readings show very large resistsances between all the different holes, except the furthest ends which is 10K between them, so I guess this is the volume controller described in the Ebay ad.
An externally hosted image should be here but it was not working when we last tested it.
 
Last edited:
Some of the people who were most excited about using LT304x regulators for critical audio, such as AVCC, have since abandoned them. They say LiFePO4 batteries and supercap power supplies sound better.

The reason, we think, is because people were so impressed with the graphs of low noise density numbers. Pretty much everyone neglected to consider it isn't noise density that matters, it is integrated noise over the frequency band that can adversely affect the audio circuitry. LT304x regulators have significant 1/f noise below 1kHz and integrated it adds up. If used for AVCC it intermodulates with the audio signals and increases distortion as well as noise.

The LT3045, IIRC, might be LT3042, one of the data sheet shows a suggested application circuit to address the LF noise. It uses an LTC6655 as a reference instead of the LT304x internal reference, then uses LT304x and a buffer and error amplifier. That is at least a big step in the right direction, seems to me.

Regarding the schematic in the above post, Vref should normally be AVCC/2. The voltage divider above is shown as 10k over 6.3k. Why run Vref lower than normal, if that is what is happening?
Also, the 47 ohm IV input resistors are a bit unusual. They could increase distortion. What are they for?
Last question, IV feedback resistors appear sized for maximum voltage swing, but that gives more distortion at low signal levels. Those resistors are usually much smaller so the IV opamps can operate fully class A, with associated reduced distortion.

EDIT: I would like know what brand of clock is on the board, too. Be interesting if anyone happens to know.
 
Last edited:
Hi,


Yes, I saw the comments about LT3042 in another es9038q2m thread, and it seemed many people were happier with an op amp buffer after the reg. I didn't see anyone tried a CCS+ shunt before it tho - this would relieve the reg of much of it's load regulation demands - which is pretty much what a buffer would do too I think. It seems it's about getting a dynamic bass so that's quite easy to assess audibly, and the op amps' slew and output impedance are part of that too. AD8620 would be a worth a try here.


I haven't measured the Vref voltage yet - curious about that too. The ESS application note has the feedback resistors at 2.7K and this board has 2.4K so this would reduce gain, and then with the 47R the gain is lower still. I'm not sure I follow your thought that this would worsen distortion or push it into clipping. I think the opposite. Have I missed something?
 
I hooked it up to a 50VA transformer 0-9 and 15-0-15, both about 1A, and measured voltages. It's all as you might predict based on the circuits. Vref is 1.27V so if that is the midrail, full scale output can't be over 2.54V at the op amp input. If it's more than that then I need to change the resistors for Vref - I guess it needs 1.5V to 1.6V for Vref.

Page 52 datasheet says the output R of the DAC is 774R +/-11%, Vopp is 0.906xAVCC (measured at 3.28V so 2.97V), and full scale output is 1000*Vopp/Rdac. 1000*2.97/774= 3.83mA +/-11%. That's a bit weak imo. but perhapos I was unrealistically expecting the multiple DAC output current. Anyway, mid rail of 2.97V is 1.485V.

The app note circuit has an inverting op amp with 2.7K negative feedback so gain is 2700/774=3.48 +/-11%. This DAC has 2.4K and an additional 47R in series so gain is between 2400/898=2.67 and 2400/727=3.30. i.e. 2.98 +/-12%. That's roughly 17% less, with the usual margin of error.

Given gain is lower, full scale output will be lower so I think Vref can be lower too. 1.27V versus 1.485V is actually 17% lower - same as above. So I think the designer's done Vref correctly to keep the op amp inputs exactly how they want to be at full scale. I might be wrong but it seems I don't need to change any of the Vref resistors at this stage.

The series 47R seem necessary not only to lower the gain but also because the output lines from the DAC chip to the op amps are not short enough to have no capacitance/inductance and the 47R will help isolate the chip output from the effects of the traces.

Seems like this DAC has been thoughtfully designed. But then again, I may have got my maths wrong so .... ;-)

I forgot to say that I built an ES9018 years ago and found that regulator noise wasn't nearly as important as output impedance - that was the key to making the DAC as dynamic as it wanted to be. I guess the ES9038Q2M has that trait also, which is why lithium batteries and super caps would work well, with the latter working because of the lowish current output of this chip. The ES9018 also suggested that sync mode is best, which is why I particularly liked this DAC - the SA8804 low jitter source also allows sync mode and I don't have to have cables from an external I2S device adding jitter. I wish i could find a datasheet for the SA8804 - no luck so far.
 
Last edited:
Usually, AVCC is set to 3.3v, and Vref is half of that 1.65v.

The 47 ohm makes the IV not as much an ideal (offset) virtual ground.

The IV feedback resistor should be small enough that the the IV opamp never swings more than between maybe +2.5v at the max to +0.5v at the min. Maybe even a little less, or maybe more such as +2.75v and +.25v. Up to wherever distortion starts to creep up. The idea is to keep the opamp internal output stage from going though Class AB cross over. That is likely where a lot of low level distortion comes from, such as with decaying cymbal tails. If it stays a bit positive all the time then the opamp is biased 100% Class A, where distortion is always close to the minimum.

However, changing the IV resistor changes the IV stage RC time constant, so the feedback cap would need adjustment. It should be a C0G or NPO cap, and the corner frequency should probably be up around 500kHz which may seem rather high, but its sometimes even a bit higher for that first pole. The dac output might like it better better that way because the offset virtual ground effect stops working when most of the IV stage current flows through the cap rather than the resistor. It may like the virtual ground effect working at higher frequencies for lowest distortion or something. Haven't really had a chance to study it to see why the pole usually seems to be so high.

The remainder of the output stage that you need to add will probably benefit from more filtering. There isn't enough in the first few LP poles as you have now.

I would have concerns about the clocking scheme too, but not much to do about that I guess. Not easily, for sure.
 
Last edited:
I replaced the screw terminals with an sma and fed it an spdif signal from a CM6631A. Output from each op amp channel always measures just under 2.3VACC at 0db. The mid point is 1.15 so that's enough under Vref that the op amp output is always above 0. This coincides with output offset of 0.17V.

The PNP transistor is 64.7 degC even a full current demand from the digital section, so I'll add a small heatsink to aid dissipation. I tested the Jamicon 1000uF - voltage loss is fine but esr is 0.22 and it was 868uF so I'll replace all of them.
 

Attachments

  • P_20180923_124726.jpg
    P_20180923_124726.jpg
    420.8 KB · Views: 328
Last edited:
This is the analogue output circuit. I have a bunch of different op amps to roll in this circuit, and I'm going to use a dual TPA6120 headphone amp with differential input and gain of 3.7 to listen to the DAC on HD650 cans. This DAC seems to have very little opportunity/need for modifications, except the power for the op amps.

An externally hosted image should be here but it was not working when we last tested it.


Besides the brand of the XO, another thing I'm stuck on is the chinese print screen next to four empty through holes - in the top right corner near the DSD and Lock LEDs. The first character is "setting" but I don't know the second one. It doesn't mean filter or volume. DMM readings show very large resistsances between all the different holes, except the furthest ends which is 10K between them, so I guess this is the volume controller described in the Ebay ad.
An externally hosted image should be here but it was not working when we last tested it.

设置 is to program the mcu , the others are right . xo is taitien , low phase noise xo . the sa8804 is the best DIR , can output a mclk with jitter below than 15ps .
 
Thank you! I found this on Taiten's site - they make XO specifically for Hi-Fi and the phase noise plot is impressive.



Do you have a datasheet for SaviAudio Bravo SA8804?


phase-noise.jpg
:
I have been over the whole board checking, measuring and testing and I'm very pleased with the quality - thank you. The PCB is very easy to work with so it's been painless to make a few minor changes. It's now ready for op amp rolling. I have AD8599, OPA1612, OPA2211, OPA1642, LME49720, LM4562, NJM2114, JRC4558, OPA2132, OPA1602, NE5532.... I guess I will find the flavour I like best :)

An externally hosted image should be here but it was not working when we last tested it.

For the digital power, I fitted 2x Panasonic FM 1,800uF 16V before the CCS=shunt, and FM 1000uF 16V after it. I put small heatsinks on the 317 and PNP.
For analogue power, I fitted 2x Panasonic FM 1,500 25V before the 317/337. 317/337 now have 120R between output and base, and base has 2x LM329 to ground. Output is +/-15.4V. The output capacitor is Sanyo 100uF 25V ESR 0.19 so it will stay stable. Its leakage is less than FC.
Vref filter capacitor is now Wima MKS 4.7uF.
SPDIF input is through an SMA socket.
 
Last edited:
My board arrived and it looks almost identical to Democles.
681827d1526816254-es9038q2m-dac-board-y1-jpg




The bulk caps are different - mine are 5x Jamicon 1000uF 35V. The electros near the op amps are also different - branded Rubycon 3MO246 10uF 25V non polar. Rubycon's catalogue doesn't have "non-polar" caps because they refer to these as bi-polar. They look like these

An externally hosted image should be here but it was not working when we last tested it.


Apart from that, the two boards seem to be identical.


I forgot to say that while going over the board I noticed 3 other differences. Democles has 0R resistors on the DAC output, I have 47R. I also have 127R on the op amp output and Democles has 30R. I also have a different film cap in the spdif signal path after the transformer.
 
I forgot to say that while going over the board I noticed 3 other differences. Democles has 0R resistors on the DAC output, I have 47R. I also have 127R on the op amp output and Democles has 30R. I also have a different film cap in the spdif signal path after the transformer.

you could change the 47r if you want to use oR , after

there are alotof you can do on the board . also you can add a dirver for a output transformer , or a amp lpf .

i can send the datasheet to you , please give me you email .
 
Thank you - I've sent a private message with my email address.

47R is fine - it isolates the chip from the the traces and it has very little shot/thermal noise (about -150dB). Norton's theorem means the op amp sees it as part of the I/V structure. Overall, I think it is a better choice than 0R for this PCB but perhaps 15R or 22R might be best.

I am very pleased the output is differential with one RC low pass filter. I might lower the frequency of the filter pole to 120kHz-150kHz, and I plan to add a transformer and a TPA6120 amp with a potentiometer volume control so the DAC is always at 0dB. The output current isn't very strong so I won't use any digital volume control.

As you say, the design creates many options :)
 
Last edited:
Some of the people who were most excited about using LT304x regulators for critical audio, such as AVCC, have since abandoned them. They say LiFePO4 batteries and supercap power supplies sound better.

The reason, we think, is because people were so impressed with the graphs of low noise density numbers. Pretty much everyone neglected to consider it isn't noise density that matters, it is integrated noise over the frequency band that can adversely affect the audio circuitry. LT304x regulators have significant 1/f noise below 1kHz and integrated it adds up. If used for AVCC it intermodulates with the audio signals and increases distortion as well as noise.
.
I just read the LT3042 datasheet and saw these graphs, which show that with Cset of 4.7uF (smd, low leakage, tantalum) integrated noise is not a problem at all and with a larger value up to 22uF it creates an even lower 10Hz spectral noise, and this has almost no effect on integrated noise. The datasheet says that Cset also improves PSRR and transient response, the latter of which is what AVCC needs if it's anything like the ES9018. So I'm wondering how people came to the conclusion that 1/f noise was the culprit of deficient bass. Transient response seems far more likely. In this post, it seems that the cap itself at Cset is extremely important, not the 1/f noise. I'll have another look at that thread and see if I can find more.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.
 
Last edited:
I just read the LT3042 datasheet and saw these graphs, which show that with Cset of 4.7uF (smd, low leakage, tantalum) integrated noise is not a problem at all and with a larger value up to 22uF it creates an even lower 10Hz spectral noise, and this has almost no effect on integrated noise. The datasheet says that Cset also improves PSRR and transient response, the latter of which is what AVCC needs if it's anything like the ES9018. So I'm wondering how people came to the conclusion that 1/f noise was the culprit of deficient bass. In this post, it seems that the cap itself at Cset is extremely important, not the 1/f noise. I'll have another look at that thread and see if I can find more.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

yes , so i use a 22uf
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.