Got Focusrite Scarlett 2i2 (2nd gen) for measurements. What's next?

I've just got Focusrite Scarlett 2i2 (2nd gen) USB audio interface. The intended use is measurements of audio circuits. So far I've done basic loopback tests (posted here). Since it is the first time I measure audio signals I need some guidance where to start to make my results meaningful.

The first question is what kind of probes do I need? I have a couple of 1/4" TRS terminated cables to start with. Since the inputs/outputs of the 2i2 are balanced it probably makes sense to use balanced probes instead of single ended. Is there a recommended probe construction to start with? So far I plan to measure line level circuits and headphone amps. I may also measure power amps and speakers later which means I will need switchable probes with different ratios (1:1, 1:10 etc.). Also should I worry about proper probe's impedance termination?

Hope to have some pointers!

Merry Christmas!
Oleg
 
Moderator
Joined 2002
Paid Member
Hi

It is probably a good idea to have some protection for your investment. This would involve using back to back zeners across the card's input and to common, to ensure your maximum levels are never exceeded.

Capacitor coupling for both input and output are a good idea to prevent amps with DC offset at either end sending it down the card and taking out the circuitry. A large capacitor (I use 80uF bipolars here) will ensure there is no sonic impact. The card can work with a variety of source and load impedance, no further corrections should be needed in your probes.

Finally, at the card's inputs you would be best advised to have some sort of rotary attenuation or selection to tailor in input to a variety of possible input levels - an amp at clipping will be putting out a lot more voltage than a preamp. To achieve all of this you can create a small jig with the required circuitry and have the card permanently connected to it. Then it can have binding posts and other terminals to connect various equipment. To begin testing attenuation must be maximum and then you can bring up the trimmer to the level you need for a good measurement.

Balanced inputs will always make more sense for rejection of 50Hz hum. However I'm not sure the Scarlett range uses properly balanced inputs, maybe pseudo-balanced. This works well for common-mode noise but does not reject the ground noise. In which case you might end up with harmonics of mains frequency in your results.

I looked at your other thread, and note that THD will rise with sample rate, this is a fact with all DACs. For most meaningful measurements you should be testing at 24/48.
 
I just realized that I have already almost all I need for the input protection attenuator. I have a bunch of R-2R attenuator PCBs as well as input selector PCBs and other staff like arduino interface, encoders etc. I think I'll also build in a variety of loads for line stages and headphone amps measurements and control all of it by the arduino.

One question remains, what input impedance for the attenuator should I use? The 2i2 interface has rather high input impedance thus not to add more noise to the measurements I guess I should not go higher than 10k attenuator input impedance (which results in output impedance in the range of 0 to 4k), or should I select constant output impedance of let say 5k for the attenuator to keep the noise low? Or should I increase the input impedance up to 1M and use an op-amp buffer after the attenuator?
 
I noticed that the sweet spot for the generator output of only -16 dBFS is severe limitation in measurements. If the output level is increased even a bit the THD rises so quickly that there is no way to test the amps at full swing. Therefore I have a question if anybody tried changing the line out op-amps to lower noise ones with higher max voltage swing before THD starts to degrade? As discussed here this interface probably uses NJM4565 at its line outputs and line inputs. May be OPA1612 would be a better option here with its lower noise and higher undistorted voltage swing?
 
Last edited:
Disabled Account
Joined 2010
I noticed that the sweet spot for the generator output of only -16 dBFS is severe limitation in measurements. If the output level is increased even a bit the THD rises so quickly that there is no way to test the amps at full swing. Therefore I have a question if anybody tried changing the line out op-amps to lower noise ones with higher max voltage swing before THD starts to degrade? As discussed here this interface probably uses NJM4565 at its line outputs and line inputs. May be OPA1612 would be a better option here with its lower noise and higher undistorted voltage swing?

Are you sure distortion originates from line out and not from overdriven input amp?
 
The THD rise could be from the generator, the non-linearity of the ADC when dealing with large signals (doesn't need to be overdriven), or both.

I would try a 1Vrms signal from the generator, use a resistor voltage divider to drop signal to 0.2Vrms and measure. If the THD is better than measuring 1Vrms directly, then the ADC is contributing to the increased THD.
 
I just measured the behaviour of the 2i2 generator output. I used my OPA1622 headamp with the gain of 2 to attenuate/amplify the output to the same level (seen by the 2i2 input). The resulting spectra are attached. If I set the generator to -16dBFS and amplify the output to -22.7 dBFS (seen by the 2i2 input) I get a small third harmonic indication but not more. If I set the generator output to -3dBFS and attenuate the level to -22.6dBFS (seen by the 2i2 input), I get relatively high second and third harmonics. I interpret it such that the generator is responsible for the rise of the THD at higher output levels.
 

Attachments

  • 3dBFSAttenuated.png
    3dBFSAttenuated.png
    32.6 KB · Views: 1,740
  • 16dBFSamplified.png
    16dBFSamplified.png
    33.5 KB · Views: 1,752
I agree that looking at the numbers the difference is minor but looking at the difference in terms of present/absent harmonics above the noise floor is encouraging. I'll try to further amplify the output to reach higher absolute levels and see if it brings further improvements.

Since it is now clear that the problem is with the output level, the next question is which parts are really responsible for the THD rise: are these the output op-amps or the D/A chip itself? I guess there is no easy way to test it without swapping the op-amps. I am fine to experiment with my unit but I have no idea how to open it. The 1st generation interface has screws at the back and TRS sockets have fixing nuts but the new unit has no visible screws or nuts to start with. Does anybody have an idea where to start opening?
 
Have fun searching the sweat spot with the right opamp.

In my case with my own sound card, the sweat spot requires a specific output level and input level, as both the generator and the ADC are contributing to the THD, so there is an optimal compromise.
 

Attachments

  • Loopback THD.png
    Loopback THD.png
    49.3 KB · Views: 1,122
Have fun searching the sweat spot with the right opamp.

In my case with my own sound card, the sweat spot requires a specific output level and input level, as both the generator and the ADC are contributing to the THD, so there is an optimal compromise.

Your result is spectacular. What sound card are you using if I may ask?
My Focusrite 2i2 showed its best when generator out is set to -16dBFS (around 0.185 Vrms) and input gain is increased to -16dBFS(it's 0dBFS is at around 14Vrms). Even then the THD I get is around 0.00029 %.

I also researched a little more about the AD/DA chip CS4272 (which people claim is inside) and indeed it might be the DAC which is responsible for the THD limit.

Anyways, I would be interested to look inside of my unit but as I wrote earlier, I have no idea how to open it :confused:
 
Last edited:
Just another Moderator
Joined 2003
Paid Member
Hi Oleg, I found with the Focusrite 2i2 first gen that it is important to output to a balanced input to get good distortion performance, for input from SE source do not tie ring to sleeve just connect tip and ring.

I made up a circuit on breadboard using a couple of LM4562's to covert the balanced output to SE and that helped a lot.

The way to tell if it is necessary is to do a loopback with full balanced TRS cables and then do another loopback with cables that are either TS to TS. ie single ended output to single ended input. If the performance drops substantially then you ideally need both balanced to SE for output and SE to balanced for input, though as I said doing TS on the input with the balanced to SE on the output worked quite well for me.

Tony.

edit: attached the circuit I did for the balanced to unbalanced 0.1% resistors advised. This certainly helped with the gen1 2i2 it may not be necessary with the gen2...
 

Attachments

  • bal_to_unbal.png
    bal_to_unbal.png
    17.1 KB · Views: 1,116
Last edited:
Thanks a lot Tony! I'll resolder the connectors since I recently made a couple of balanced to SE cables to use with the 2i2 interface and, indeed, tied ring to sleeve... I also noticed significant amount of mains harmonics if using my bal-SE cables, so I'll definitely give your suggestion a try. Going all balanced is another possibility.
 
Just another Moderator
Joined 2003
Paid Member
It seems like the correct thing to do but certainly on the gen1 it drops the performance substantially!

I used 12V SLA batteries for the powersupply on the balanced to unbalanced converter, note that the other advantage is it doubles the output magnitude, so effectively 6db more signal :)

Tony.
 
I think gen 2 is not much different from gen 1, except that there is no obvious way to open the unit anymore:)

6dB extra is always a good thing:) I have an assembled low noise PSU, a bunch of quality op-amps and suitable small PCBs (see my signature), so I'm nearly ready to try bal-SE and SE-bal route. I've also started putting together an attenuator for the 2i2 input, so all these bal-SE/SE-bal can go in there.
 
Just another Moderator
Joined 2003
Paid Member
Note I did put 100nf decoupling caps on the opamps in the real world implementation ;)

One difference looking at your graphs is the noise floor. definitely lower than my gen1. Below are the first measurements I took (not optimized) comparing balanced loopback with SE loopback. As you can see the distortion goes through the roof with the SE.

Also attached is measurement made with the balanced to SE circuit with with se to TR cable (third pic) you can see the improvement! Note that the third one was done with different levels probably after I had more idea what I was doing ;) Unfortunately I can't find a straight balanced loopback done at the same time as the third pic, but the fourth one is probably more representative of the real performance...

Tony.
 

Attachments

  • balanced.png
    balanced.png
    22.4 KB · Views: 1,122
  • unbalanced.png
    unbalanced.png
    22.4 KB · Views: 933
  • 2i2_distortion_test_bal_to_se.png
    2i2_distortion_test_bal_to_se.png
    22.6 KB · Views: 918
  • 2i2 loopback_distortion.png
    2i2 loopback_distortion.png
    18.7 KB · Views: 385