• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

LEAF Dac Dual 9038Q2M VS Gustard X16 ! Pictures inside )

GXAlan

Major Contributor
Forum Donor
Joined
Jan 15, 2020
Messages
3,868
Likes
5,955
Should be nice that a US boy could send it to Amir and will be fixed )
What a pity that I am not in the USA because I have lots of amp to share too !
You should buy an E1DA cosmos from audiophonics.fr and run your own tests. It’s easy with the multitone tool.
 

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
Just use RCA as you’ll not gain anything using XLR out with an RCA adapter. You’ll save some money as well using RCA cables.
This is doubly true with the Leaf 9038q2m. There is no voltage difference on XLR vs RCA.
You should buy an E1DA cosmos from audiophonics.fr and run your own tests. It’s easy with the multitone tool.
1666968402450.png


Edit: Wait, I think I am figuring it out. I will start with the Loxjie D40 so we have a bseline to see how far off it is from Amir's tests, then I will do the Dual 9038q2m.
 
Last edited:

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
Okay! Here are results. Measured right channel on three DACs using the Cosmos ADC. Only did a 1kHz FFT because when I tried multitone, it was clear I was doing something very very wrong.

In all instances, flexASIO was used as the driver host in REW. All samples are 192/24, using REW as the generator.
Gustard X20 and dual 9038q2m boards were fed via LVDS input from a Singxer SU-1 as neither has a functional USB at this point in time (Gustard did not ship with one installed).

All samples adjusted for peak at -0.00 to -0.01 dBFS. All DACs were set to whatever their equivalent of fast roll-off linear is. I actually have no idea what that is on the Loxjie, so it was done as FL1 SC1.

First, the Loxjie D40:
1666991346360.png

Second, The Gustard X20:
1666991305607.png

Third, the Leaf Dual 9038q2m:
1666991371430.png



Amir's measurements show the D40 as being about 20dBFS lower in noise. This may just be a factor of my environment and test equipment. The dual board (and actually almost decade-old Gustard) put on a good showing. If we assume that the D40 is the baseline and everything else has a similar shift, then all three DACs are audibly transparent or very, very close to it.

Now, one thing I must note about my particular leaf DAC. The I/V opamps are swapped to OPA1612s which I stole from a SMSL M8. The I/V follower opamps are LME49720s (might be why the 30+kHz has harmonics since they are very high BW). I am using the stock wall wart, but I swapped the AVCC and DVDD LDOs with TI TPS7A20 3.3V units. I also adjusted the I/V to a 1k resistor on the gain stage and swapped the stock LDO input ferrites for ever so slightly larger murata units harvested from a dead monitor PCB.


Just a quick measurement addition. SINAD is basically THD+N * -1 or so I have read. My measurements show a THD+N about 14dB worse than Amir's. I'm not sure if that translates to every DAC I measured in a 1:1 fashion, but it is what it is. Here is the Loxjie D40, and here is Amir's review and measurements.
1666991115150.png
 
Last edited:

JoeSVK

Member
Joined
Feb 22, 2022
Messages
32
Likes
4
@deadwood83 so what is the benefit of removing the headphone amp board? I am not using the headphone out and if it helps with something, then why not? :cool:
 

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
so it turns out I am trash at measuring yesterday, and I had no idea what I was doing! Updated Measurements below.
@deadwood83 so what is the benefit of removing the headphone amp board? I am not using the headphone out and if it helps with something, then why not? :cool:
As more power is supplied through the switch mode converter and the LDOs on the board, each of those becomes noisier. The headphone output is driven all the time. Each opamp in the DAC is on for 100% of the time the DAC is on. Removing the ones which are not used lowers the power consumption, and should lower the noise floor a little.

Loxjie D40:
D40_1kHz.jpg

D40_Multitone.jpg



Gustard X20:
gustardX201khz.JPG

Gustard_Multitone.jpg



Leaf Dual 9038q2m:

Leaf_1kHz.jpg

Leaf_Multitone.jpg



I think the leaf is fine for what it is. Over 17-bits of usable dynamic range. I do not believe my ears can hear over 17 bits. One thing to consider is that the Gustard and Loxjie are both 2-3x the price of the Leaf, though a used Gustard is probably $300 USD or less. They used to go for $700-850 back at launch in..... 2014? I guess it;s a little long in the tooth. If they could shake the harmonics then it would be practically SOTA performance.

Similarly, the Loxjie seems to be dominated by the 3rd harmonic. ¯\_(ツ)_/¯

I think ym downstream components are all noisier than any of these DACs. Driverack PA2 is ~93 SINAD and Topping PA3s x2 are ~85 SINAD. The system sounds reeeeeeeeeeeeally incredible though, especially after I added the Emotiva sub on the PA2.
 

JoeSVK

Member
Joined
Feb 22, 2022
Messages
32
Likes
4
Thank you @deadwood83 , I will remove the headphone board then, as I am not using it. Those measurements were made with the OPA1612 in the I/V stage?
 

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
Interesting findings. I got my Xing USB DDC hooked up after tracing it back.
1667917202728.png


The firmware is programmed for very, very, very low FSR rates in USB mode. The lowest master clock I can set for any frequency is only 64Fs on 44.1kHz and 48kHz sample rates. This is still way too fast for the firmware on the STM32 on the board. It believes it is DSD256 so filtering is all messed up.

BUT I can get it to take native DSD, though it believes any sample rate is DSD256 (11.2MHz). DSD performance shows a degradation of ~8dBr THD+N. Testing with the firmware in a wacky mode is a little interesting. I have to generate a sample in JRiver, set DSP to output 4x native DSD on ASIO via JRiver, open multitone, set to external source, and then do single frequency measurements (since JRiver has no multitone).

Interestingly, they appear to set a defined ratio in the USB mode via the I2C registers but in IIS mode I am guessing they leave it in auto deterministic mode with DPLL.

Tonight I will try resoldering to the IIS input stream and see if it locks properly and engages filters appropriate to the format being sent. The ultra low ratios on USB input might be why the DAC could not meet advertised specifications, let alone datasheet specs for the chip.

1667918791557.png


I think these are firmware limited by the onboard MCU and how it sets registers for USB. I took a firmware dump to see what I could find but reverse engineering firmware is really not my strong suit.
 

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
Please ignore my last image for the circuit, as it is dead nuts wrong.
My apologies, I got mixed up in the optical switches.
Real connection diagram:

1668057291880.png


I2S is on the left and USB is on the bottom. With the correct connections, the Xing interface works wonderfully.

I tested DSD512 vs PCM. DSD was upsampled via JRiver 192kHz test tone. It actually shows a better skirt, better harmonics, but slightly elevated high frequency noise floor. The harmonics are definitely the weak point of this DAC. I'm not entirely sure how to go about addressing them. Leaf was clever and used C0G or NP0 caps on the analog signal. the X_R visible in the above photo are for VCC on the optical switches.

Green = DSD
Orange = PCM

1668057598997.png

1668057636918.png

DSD data for comparison:

1668058347912.png


PCM:
1668058204006.png


As you can see, the overall THD+N is nearly identical. What a surprise. I tried to force the 1.5MHz PCM rate supported by the Xing, but I have no application which will interact with it properly.at that sample rate. Not even Multitone would honor it, despite selecting 1.5MHz in the driver control panel. This dac is okay at multitone, but terrible for AP multitone.

1668060469484.png


It does seem to perform better at higher sample rates. A 3dB gain was had by going 192-384. However, I could not get past 24-192 on the stock xmos. TO get full performance out of (my modded) DAC I need to use the Xing. something in the sub-1k range is also messing up the noise floor. Trace imoedance or inductance maybe? Not sure.

I was going to run a Jitter test but uh........ my flying IIS wires from the Xing to the board make that rather pointless. I will need to figure out how to integrate that better.
 

JoeSVK

Member
Joined
Feb 22, 2022
Messages
32
Likes
4
You are a wizard @deadwood83 :D what op amp configuration are you running the dac on?

EDIT: I checked the posts before and saw the op amp configuration, my bad.
 
Last edited:

JoeSVK

Member
Joined
Feb 22, 2022
Messages
32
Likes
4
@deadwood83 - found a topic on audiokarma forum about op amps and there was a post about placing 0.1uF bypass capacitors, 2 per chip from + rail and - rail to the ground, to reduce errors coupling in from noisy or high impedance power supplies. Could this help in the performance? I don't know, if there are any bypass caps applied or if it would help in any matter at all.
 

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
This DAC does have such bypasses (C29, C30). Here's one whole channel's I/V stage. On the right edge of the photo it enters the 2nd voltage follower stage

1669650342688.png


We call it an I/V stage (Current -> Voltage) but it could easily be called a double transimpedance amplifier layout. The sections in orange are your two feedback resistors (one for each opposed leg of the L channel signal; R9 and R18). Feedback sets the conversion ratio of the I/V stage. In dark green you have 'Filter' generically. To the right of the output, it is a high pass filter. (Electrolyte + R2 and electrolyte + R37) This will remove the DC from the signal. If you plug in the C and R values (1000R 100uF) then you will see that any 'signal' below 1.5Hz will be cut out. Since DC looks like a 0-Hz AC signal, it is removed at this stage.

The bits in green next to the amplifiers (R19 + C2 and R20 + C8) are part of the feedback network and sets the maximum frequency operating point of the opamp. Any frequency above that pole/pole will have a decreasing amplification. You can think of it as a sort of low pass filter. Usually these values are set to be in the low-MHz range.

The non-inverting input of the opamp is tied to AVCC/2. That value is obtained by a simple voltage divider formed by R136 and R135. This is our virtual ground a.k.a. Vg. So why this value? THe datasheet has the answer.

1669651749033.png


We need to set the current mode output offset to 0 so that at idle the DAC is not sinking or sourcing current. Since the equation for current offset uses Vocm, the easiest way to ensure that happens is to set Vg = Vocm. That way we have 1000* 0 / Rdac = 0mA.

If you want max performance from your downstream devices, it is best to set the gain to max as near the DAC as possible. THe 2nd stage of the opamp stack is a voltage follower, so it has unity gain (a.k.a voltage in = voltage out). For standard XLR connections, you want about 4Vrms between both pins, or 2Vrms per hot or cold pin. This is about 5.658Vpp on either side of the opamp. Each 'leg' of the signal can put out a max of 7.7mA, and the I/V stage ratio is just a simple current across feedback resistor determines voltage. So Fb resistor value * current = Vpp. We know Vpp and we know current, so an ideal resistor value for best gain staging for optimal SNR would be about 730-Ohm.

This should give ~5.65Vpp on each leg (XLR+ and XLR-) which would combine to be 11.3Vpp across both which is 3.995Vrms. But the Leaf is set up so that RCA does not apply any gain/cut. So if you plan to use XLR then the resistor swap might be worth it (though maybe the improved SNR would not be audible). If you plan to use RCA then the factory supplied 390-Ohm resistors are about right.
 

daggernc

Member
Forum Donor
Joined
Oct 19, 2020
Messages
24
Likes
10
Some day when I grow up I want to be just like you guys!

I appreciate all the experts out here providing their time and inputs to these discussions.
 

deadwood83

Member
Joined
Sep 1, 2022
Messages
91
Likes
160
New personal best on the leaf.
1669695085627.png
 

JoeSVK

Member
Joined
Feb 22, 2022
Messages
32
Likes
4
Nice, I've used it with 2x1612 and the stock 5532, was better than the stock conf, opened up the stage, was cleaner, currently using 4x1612 and it is cleaner, the details are better, added a layer to the music with some micro details, fast roll-off minimum filter. I've tried it with 1612 and 49990 but it was meh.
 
Top Bottom