Chapeau ou non? (PCM5102MK non-HAT DAC review)

[AnalogSteph]

I think what may be happening is that the PCM5102 output stage has a bit of a hard time driving the low input impedance of the E2x2 mic input (which is the XLR portion), especially if it's a first batch specimen, hence the high distortion. Adapting to TRS (for the line input) would have been the better choice, it would also give you a lot more headroom on the recording side.

The excessive periodic ripple may be down to some sort of windowing settings / clock sync issue. If the ADC and DAC are not perfectly in sync you generally have to make accommodations for that. Neither the PCM5102 DAC nor the E2x2 ADC (AK4621) should have periodic ripple much exceeding +/-0.01 dB, in fact the AK4621 also has a "classic" sharp ADC filter with no more than +/-0.005 dB.

In case of the M-Track Solo the ripple is accurate and a combination of PCM29xx ADC and DAC digital filters (see datasheet, about +/-0.05 dB for each). The larger magnitude makes it a lot more obvious.

 

[Linards]

@AnalogSteph , Thanks!
Very good and useful points. I think you might be on to something!
Will get TS to RCA cables after Easter to check this and post results here. ^_^

 

[AnalogSteph]

Will get TS to RCA cables after Easter to check this and post results here. ^_^

Ideally, what you want is something like #18 from RaneNote 110 (you could get even fancier in construction if you know that DAC's output impedance, but this will generally work OK). You seem to have had few ground loop issues, but anyway.

 

[Linards]

I did PCM5012A board injustice - it seems to preform practically up to the spec.
@AnalogSteph pointed to that low impendence inputs of E2X2 interface that I was using could cause problems with measurements (DAC starts "driving" interface and thus distorts; see higher volumes in original post).

I had some extra patch TRS cables to cut and RCA screw terminals to experiment with wiring: red wire of TRS is Tip, white - Ring and shield (bare wire) - ... well Shield. (+) terminal on RCA is tip and (-) ring (?).

Tip (red) is connected to tip (+) terminal. Other 3 configurations were tested connecting to (-) terminal:
1) Ring (white)
2) Shield (bare)
3) Ring + Shield

Additional to that, it is possible to enable INST input, by pressing button on interface. Without press of the button is LINE input. Key difference tested here is influence of input impedance:

a) INST = 1M Ohm
b) LINE = 6K Ohm
*c) MIC = 1.5K Ohm (*used in original test)

I ran tests with 1 kHz tone @ -1dBFS (98% volume in Volumio). To be consistent with original post - I used minimum gain on ADC.
Here is setup for 3) configuration:

For reference, here is -1dBFS MIC input measurement:

And here are new test results:

1) RCA = T & R; Huge problems with ground. Interface clipped when trying INST.

2) RCA = T + S; Grounding seems to be OK.

Practically spec. TD+N using INST.

3) RCA = T & R+S: No 50 Hz mains leakage at all. Best grounding option (?)

Some 50 Hz mains appear, but at low level.

From these figures, it seems to me, that INST (highest impendence) input and T & S connection for RCA should be used for further measurements.

Any other thoughts, Guys?

 

[AnalogSteph]

The PCM510x datasheet says

(3) Output load is 10kΩ, with 470Ω output resistor and a 2.2nF shunt capacitor (see recommended output filter).

So you may want to provide a parallel 10k load in INST mode to fully verify performance. Or alternatively, add ca. 2.0 kOhm resistors in series with R and S to bring up line input impedance, at the expense of a bit of level. The same approach would work with the microphone input, using 4.3 kOhms in series each. (In fact, a number of interfaces are implementing their line-in exactly like that.)

Instrument inputs are generally unbalanced, hence why T & R + S did not get rid of the 50 Hz harmonics when using INST (unbalanced, R grounded) but did when using LINE IN (balanced). Hence also why RCA --> T+R was an epic fail, you basically left S floating.

 

[Linards]

@AnalogSteph , Thanks for feedback!
Indeed I could add resistors. ^_^

Which got me thinking - perhaps it makse sence to add (higher?) resistance to emulate input impedence of an amp? In the end of the day it will be used with one. What I have seen with little search is something around 40K Ohm or something like that.

Is there any standart for amp input impedance? Or 10K Ohm is a standart for DAC testing (lower limit of amp impendence)?

 

[AnalogSteph]

I don't think balanced line inputs are ever higher than 10 or 20 kOhms. 50k is the traditional standard in hi-fi but you also see 20k and 10k. Driving 10 kOhms should not be a challenge in this day and age. A lot of better opamps will drive 3 kOhms quite easily.

If you need something to comfortably drive the 1.5 to 4 kOhms of a balanced microphone input, a headphone output (/ amp) should do the trick.

 

[gfx_1]

Thanks for measuring. I use cheap hat with an optical and coax digital output on a Raspberry Pi 3B.
That should keep the noise down a bit.

 

[trungdtmc]

My mini DSP 2Ways

 

[Linards]

@trungdtmc , looking good, going strong!
I can identify, 2xPCM 5012A boards, ADAU1401 DSP, programmer board and Amanero (not hooked up). What are USB to I2S and other (green) board in your setup?
Also, could you share what is your project/plan for this setup? ^_^

 

[trungdtmc]

The blue board like the amanero is the qcc3034 module to test for some of my friends, and the smaller green one is the df1706e, which I use for old DAC circuits like the PCM58p...
I built this circuit as a mini 2 way dsp for my bi amp bookshelf

 

[NTTY]

Hi @Linards, thank you for this review, it must have taken an unreasonable amount of time to write, great job.

I wanted to ask about this specific measurement and your comment about fact the noise floor was low (-135dB):

I am not familiar with the software you used but if, as I can see on the print screen, the FFT length is 1M, then the noise is attenuated by 10log(1M/2)=57dB. This is the FFT gain.
So if that representation is correct, then what we see here as noise floor around -110dBr is in reality 57dB higher.
What the software reports as -135dB noise floor, might be the performance of your capturing device, running without gain adjustment at the input (level -40.9dBFS?), but not the actual performance of the DUT.
There might be something I missed here, and I’m eager to learn.
Thanks in advance for your reply.

————
Flo

 

[Linards]

Hi, @NTTY! Good questions there.
I also think that figure of noise floor you mentioned (-135dB) could be atributed to ADC capturing device. SNR spec of PCM5102A is 112dB and it seems that just by analyzing / looking at the spectrum in multitone test it is the same ball-park.
I cannot comment on implementation of FFT in MA software, but from user perspective all I can say, that number of averages (8 here) also plays a role when generating this plot. Maybe @pkane can chip-in and comment on details of FFT gain / noise floor in MA?

 

[pkane]

Hi, @NTTY! Good questions there.
I also think that figure of noise floor you mentioned (-135dB) could be atributed to ADC capturing device. SNR spec of PCM5102A is 112dB and it seems that just by analyzing / looking at the spectrum in multitone test it is the same ball-park.
I cannot comment on implementation of FFT in MA software, but from user perspective all I can say, that number of averages (8 here) also plays a role when generating this plot. Maybe @pkane can chip-in and comment on details of FFT gain / noise floor in MA?

The noise number in Multitone is computed by removing signal and harmonic/IMD distortion artifacts and then summing the remaining FFT bins. In effect, the FFT size shouldn't matter, but the number of averages (and the type of averaging selected) can certainly have an effect. Here's an example:

 

[NTTY]

@Linards and @pkane: thank you both for your detailed replies, this is really appreciated.

@pkane: your example is clear. But in the print screen of Linards, I don’t understand how the noise can be measured at -135dB. I suppose I need to get used to play with your obviously very nice software.

 

[pkane]

@Linards and @pkane: thank you both for your detailed replies, this is really appreciated.

@pkane: your example is clear. But in the print screen of Linards, I don’t understand how the noise can be measured at -135dB. I suppose I need to get used to play with your obviously very nice software.

Can’t really say by looking at the plot. Coherent averaging could explain it, if that was used for example.