More Steinberg UR12 Black measurements

[AnalogSteph]

(Initial observations start in this thread, with more on input noise here.)

RCA L --> XLR mic in (nominal Zin = 4k, Zout = 600R).
-12dBFS at max output + min input gain becomes -3.49 dBFS in.
The difference between min and max gain settings is 43 dB. (Spec: Gain 10-54 dB, so we're only 1 dB off at one end.)

Frequency response via REW 5.31, RCA L out --> XLR mic in, 48-96-192 kHz, min gain in and ~14:00 out, plus 192 kHz (max input gain)/(min input gain) trace arithmetic:

Variations with sample rate on the bottom end seem to be down to the ADC's DC offset removal filter, while on the high end there's mostly the DAC filtering at play if you compare the CS4270 datasheet. 96k becomes effectively the flattest because the double speed DAC filter has a boost of up to +0.8 dB up there, partially compensating the effect of external output filtering. I think the ADC side is much flatter.

The difference between min and max input gain is surprisingly little. I don't think many of us are going to care if 10 Hz is 0.3 dB more down, and on the high end I got 0.169 dB at 79 kHz / 0.193 dB at 86 kHz. If I have my math right, the latter indicates an effective GBW of about 2 GHz for the preamp. That's definitely not a 1-opamp job, more like an instrumentation amp.

So this could really be a classic Yamaha D-PRE (think ESP P66-ish) input after all... but why is it so noisy then? My estimate for EIN at full gain with an assumption of a -54 dBu input sensitivity is about -118 dBu, 24 kHz BW (-64 dBFS at 48 kHz), or roughly -121 dBu(A), not exactly earth-shattering performance.

EDIT: Here's a whole sack full of RMAA loopback measurements (mono mode, mic in or line in, various levels and corresponding input gain / output level settings). The maximum dynamic range I could get was 97.4 dB(A) on mic in (L) and 102 dB(A) on line-in (R), though distortion at this point is not too pretty and exceeds 0.1% (I do think the unbalanced output may be provoking some extra common-mode distortion in the preamp in the form of elevated H2 in particular... H3 is pretty constantly around -90 dBFS throughout, at any kind of gain level). This dynamic range is about in line with what a CS4270-based device should be getting, although the mic in falls a bit short and also comes in a few dB behind the big brother, the UR22 mkII.
Dynamic range at maximum gain on mic in comes out as about 65.5 dB(A) or 64.5 dB unweighted (20k), so my estimate of a -64 dBFS noise floor wasn't bad at all. That output amp must be pretty quiet then.

Note: The device seems to reset and be redetected each time the sample rate is changed. This can upset RMAA in ASIO mode quite significantly, so I eventually made sure to set sample rate manually via the Steinberg control panel first. I am pretty sure that different sample rates for ADC and DAC are not supported.

 

[AnalogSteph]

EDIT: Here's a whole sack full of RMAA loopback measurements

Guys, I swear I had edited this link to correct it. *grmbl* *misc. muttered expletives*
Here's the actual file.

Since we're here anyway I might as well post a selection of results:

Don't be fooled by the distortion on the "mic maxgain" result, that's noise-limited:

This is what we get with input gain at 12 o'clock:

And this at minimum gain (i.e. highest input level):

A bit of the H2 might be related to output loading as well, as things look a bit cleaner on the line-in:

And this is how things clean up by 10:45 in / 14:00 out:

And 12 o'clock in / 13:30 out:

And finally, max gain on the line-in:

BTW, if nothing noted, input level was adjusted to -1.0 dB +/- 0.2 dB.

I also did some testing in REW again with a line-out --> line-in (R) loopback, changing nothing but input gain this time (i.e. not compensating on the monitor level):

This time, my maff gives about 35.4 MHz of GBW.
The above maxgain/mingain graph is G(f)/Gn. Gn = +40 dB = 100. I'm reading about -1.485 dB (~= 0.843) at 66 kHz.
Our OLG A(f) is
A(f) = Gn * f * (G(f)/Gn)/[1-(G(f)/Gn)]

It seems like my GBW estimates may be a tad on the high side as I don't think they would have put a part this fancy in there for a Hi-Z input (which is what this is actually supposed to be). Maybe a TL07x or similar as a buffer followed by an NJM2068 at best, rated 27 MHz at 10 kHz (and probably more like 19 MHz if the NJM8068 datasheet is any indication). Well, maybe not a TL07x, given that the specified max input is +8.5 dBV and that would be about +/-3.75 V p-p, which seems hard to pull of with a TL07x on +/-5 V (which would go into inversion below -2 V or so). I don't see any elevated 1/f noise levels either like you would typically get with a JFET input part. Maybe a 4565?

A rough quantication of line-in (Hi-Z) knob position vs. level (estimated absolute gain):
max (17:00) = 0.0 dBr (+40 dB)
13:00 = -20.2 dBr (+19.8 dB)
12:00 = -23.7 dBr (+16.3 dB)
10:45 = -28.4 dBr (+11.6 dB)
min (07:00) = -39.4 dBr (+0.6 dB)

Same for mic in knob position vs. level (estimated absolute gain):
max (17:00) = 0.0 dBr (+53.1 dB)
16:00 = -8.3 dBr (+44.8 dB)
15:00 = -15.0 dBr (+38.1 dB)
14:00 = -20.6 dBr (+32.5 dB)
13:00 = -24.7 dBr (+28.4 dB)
12:00 = -27.6 dBr (+25.5 dB)
11:00 = -30.0 dBr (+23.1 dB)
10:00 = -35.2 dBr (+17.9 dB)
09:00 = -39.8 dBr (+13.3 dB)
08:00 = -42.8 dBr (+10.3 dB)
min (07:00) = -43.1 dBr (+10 dB)

Here's a plot for this one (don't lose it ):

I just noticed that my RCA --> XLR adapter cables are wired to pins 2 and 3, so no wonder I didn't have any issues with ground loop noise on the mic-in.

A bit of absolute level calibration:
The 1 kHz signal that gives -1,0 dBFS on the mic-in according to REW measures as 0.727 Vrms unloaded. The mic input itself should be seeing about 0.63V of that (Zout =600R, Zin=4k - I confirmed Zout with a 2.195k resistor in parallel, giving ~619 ohms, close enough). That's -1.75 dBu, so 0 dBFS is -0.75 dBu.

We know we have a gain range of 43.1 dB, so 0 dBFS at full gain would be about -43.85 dBu at the input.

This would be equivalent to -42.63 dBu from a 600 ohm source or -43.53 dBu from a 150 ohm source.

So our EIN comes out to -43.5 dBu - 64 dB =... -107.5 dBu unweighted, maybe 2-3 dB less A-weighted.
Wait, what?! That would be terrible.

What did I do wrong here? Nothing, as far as I can see... The output level is entirely realistic at about 14:00 on the monitor volume (I'd expect 6-10 dB below maximum, which is 2 Vrms, and if you do the math we're right in the middle at about 8 dB down). Likewise, maximum input level and gain range all check out. No wonder the noise from the output stage is not worsening things appreciably, input noise is the better part of 3 µV. Shocking, absolutely shocking. I will have to investigate.

 

[AnalogSteph]

Screwdriver time!

I found +/-6.0 V analog rails generated by a Maxim-y looking chip (AK0 / 9C0). The P48V boost converter uses an NJM2374A PWM controller.

3 of 4 opamps are NJM2068, with an NJM4580 being responsible for the headphone output. Speaking of which, each channel gets a massive 220µ/35 V output coupling cap and two 180 ohm resistors in parallel for a 90 ohm output impedance.

The brains of the operation is an NXP LPC1820FET100, with a Microchip USB2422 USB hub / controller thing.

Now the interesting part, our microphone input stage. Here are the preliminary results of my reverse-engineering efforts:

Yes, each input leg is going through a 10µ/50V coupling cap, then another 47µ/25V (clearly someone did not trust just the first one) and 2k2 in series. The first UR22 was a bit notorious for killing components when phantom power was involved (e.g. hotplugging/unplugging), so I can understand using the 2k2 in series, especially when you're too cheap for some clamping diodes (though it looks like the **C3875GRs may be serving such a function). Still, it messes up EIN performance bigtime. BC859C rbb' of about 300 ohms (according to its SPICE model) doesn't even matter. I suspect the input impedance spec is incorrect, too.

Let's do the math (approximately anyway):
1. Each leg sees a source impedance of about 310 + (4k7/2||33k/2||2k2) ohms ~= 1.373 kOhms. A bit of creative use of my opamp noise calculator (assuming beta = 500, so Ib ~= 2000 nA) yields -117.8 dBu worth of input noise referred to the input transistor bases (in 20 kHz at 295 K).
2. The signal also gets attenuated, by roughly 6.85 dB. So that's actually about -110.95 dBu input-referred. Still not what I'm seeing but getting there.

Jeez, how does one salvage this mess? I was expecting to see some sort of construction problem, not an intentionally gimped input that would have been too hot and endangered by the Phantom Menace originally.

I think what happened was that they may have copied the circuit from something like the MG10XU, subbed in a 10k gain pot instead of the original 20k, but forgot that this would increase their minimum gain. I think I saw some 33k resistors around the 2068, just like in that mixer. Why they didn't just alter the values of a few passives to compensate is a mystery to me. Low-end response would still be fine if the 56 ohm resistor were to be approximately halved. Now you'd have to be dealing with both the original issue and their hacky "fix".

 

[AnalogSteph]

Reverse-engineering continues:

I may have gone slightly overboard with this one.

After all of that, I couldn't be arsed to figure out where our mystery supply comes from, I suppose it's a low-noise something or other. EDIT: I have now confirmed it appears to be +6V. The PCBs appear to have more than 2 layers and some sort of conformal coating, which makes it hard to trace things at times.

Don't even try to simulate this as-is. I just recycled an old simulation of ESP P66. Not sure whether you can get SPICE models for all of these components.

And no, this isn't actually my anthem of the day.

EDIT: BTW, I think they literally turned on loopback mode for the measurement in the iXBT review, i.e. what we're seeing there is digital loopback of the signal with analog noise from the mic input (and perhaps some analog signal leakage) mixed in via the ADC. I cannot imagine that a CS4270 (THD+N -87 dB DAC, -95 dB ADC) could ever manage 0.0005% of THD (-106 dB) in loopback.

So what do we do with this now? It occurred to me that it would be easy enough to just solder some thin film 33ks in parallel with the existing ones once the situation on the input side has been figured out (well, for those into microsoldering anyway, which unfortunately is going to be a requirement). Likewise, one could solder 100Rs in parallel to the 2k2s, not sure where to find +/-6 V in the vicinity for some clamping diodes though.

I'd have to simulate what would happen if one were to tack on some MMBT5087s in parallel to the BC859Cs instead of replacing them outright. Either way, these oldies would seem to be a suitable replacement once the rest has been optimized for noise (low rbb' at ~40 ohms, medium-high beta, decent fT, low Vce_sat).

 

[AnalogSteph]

I'd have to simulate what would happen if one were to tack on some MMBT5087s in parallel to the BC859Cs instead of replacing them outright. Either way, these oldies would seem to be a suitable replacement once the rest has been optimized for noise (low rbb' at ~40 ohms, medium-high beta, decent fT, low Vce_sat).

Hmm, Bob Cordell's model lists an rbb' (RB) of 193 ohms, now that's not good if this sort of thing can happen. In that case even a trusty 3906 would be better. Looks like it could get to about -124.9 dBu unwtd 20-20k like that (with 100R series resistors), so about... -126 dBu A-wtd? Not much of a difference to 4403, but higher beta. It's a proven standard circuit so finding EIN performance in a normal range after optimization isn't actually all that surprising.

Simulated distortion does degrade substantially with an unbalanced source at high levels. With 1 kHz at 727 mVrms (unloaded) from a 150 ohm source, H2 rises from -98 to -85 dBr, with harmonics decaying more slowly as well.

On a side note, I haven't seen anything on the inside that would look different from a regular UR12, it's just the front panel and housing that are painted a different color. I'd have been somewhat miffed had I bought this at full price - mind you, things have moved on since the UR12 first came out at the end of 2014 (which would also help explain why it slots in somewhere between the original UR22 from 2013 and the mkII from 2017 performance-wise). At least parts quality appears to be good (e.g. the XLR input is a Neutrik, Nichicon caps etc.), even if soldering is a bit variable.

Simulated 0 dBFS output level for the micpre comes out to about 3.2 Vrms, so it looks like they have an attenuator for the CS4270 (which can take 2.8 Vrms at +5 V) as suggested in the datasheet. I'm measuring about 6.3k to ground, so maybe 3k6/2k7?

The 48V boost converter is funny, it seems to be followed by a voltage doubler (Cockcroft-Walton). Makes sense as the chip can only take 40 V. The feedback network suggests 24.6 V (280k/15k) so I guess attaches to pre-doubled output? There's a 33µ / 63V on the output, followed by an RC of 68R + 10µ. No hints of why it would seemingly be affected by input voltage though. I think it's mainly a rumble because CMRR down there is letting up (effectively we're looking at 8.24 µF with both caps in series, with heaven knows what kind of tolerances - 22-47 µF would be more typical).

Simulation for LTspice is attached, shouldn't require anything but Bob Cordell's models.