1073 type input stage caps...why??

[HenryL]

There is a content of very high frequency noise there - have you zoomed out on a scope to see what frequency that is? Also there is some HF in the interface without the 1073. It seems like there’s a feedback loop somewhere at very high frequencies.

Hi RR. If we take it as a static frequency response curve as it theoretically is then those are all different frequencies, with wildly fluctuating levels.
I was thinking HF noise but I see it could also be a single frequency beating with the rising sweep and creating additions/subtractions to it across the upper range.

As you say, the same or very similar pattern is there in the interface without the '1073' in the loop but with less extreme fluctuations.

now I'm thinking more about the signal/noise ratio for sweep levels used in the various tests. REW urges using a fairly high (in dBFS terms) level for the interface calibration, and in the calibration plot this one has the smallest magnitude of HF chaos in the frequency response. I used a much lower level sweep signal for the mic input tests as I wanted to set the '1073' in the most 'ringy' position at -40/-45. For the line input I was using a 20dB higher sweep signal on the -20 gain position. This looks roughly consistent with the magnitudes of the peaks and dips across the 4 plots I posted above.

So if it's fixed level HF noise (or some frequency/ies) added by the interface then I should set the sweep level as high as possible and put a pot between the line out and the 1073. It seems I could also try averaging multiple sweeps, which should help reduce the effect of random elements I guess.

 

[HenryL]

Another thought - have you tried the test(s) with the 48V disabled completely by disconnecting the feed to the doubler?

Not tried it as such but didn't see anything odd previously on the scope when modding the psu. Back pocket for now I think, as I'd have to get back in to a dark corner and break things again..

 

[RoadrunnerOZ]

Hi RR. If we take it as a static frequency response curve as it theoretically is then those are all different frequencies, with wildly fluctuating levels.
I was thinking HF noise but I see it could also be a single frequency beating with the rising sweep and creating additions/subtractions to it across the upper range.

As you say, the same or very similar pattern is there in the interface without the '1073' in the loop but with less extreme fluctuations.

now I'm thinking more about the signal/noise ratio for sweep levels used in the various tests. REW urges using a fairly high (in dBFS terms) level for the interface calibration, and in the calibration plot this one has the smallest magnitude of HF chaos in the frequency response. I used a much lower level sweep signal for the mic input tests as I wanted to set the '1073' in the most 'ringy' position at -40/-45. For the line input I was using a 20dB higher sweep signal on the -20 gain position. This looks roughly consistent with the magnitudes of the peaks and dips across the 4 plots I posted above.

So if it's fixed level HF noise (or some frequency/ies) added by the interface then I should set the sweep level as high as possible and put a pot between the line out and the 1073. It seems I could also try averaging multiple sweeps, which should help reduce the effect of random elements I guess.

Yes - you may just be amplifying noise from outside the 1073 - including your interface. Turn off any fluorescent or dimmed lighting

 

[HenryL]

Yes - you may just be amplifying noise from outside the 1073 - including your interface. Turn off any fluorescent or dimmed lighting

and try not to breathe

 

[ruffrecords]

Just a thought, why are you not using a sweep to measure frequency response?

Cheers

Ian

 

[HenryL]

Just a thought, why are you not using a sweep to measure frequency response?

Cheers

Ian

Hi Ian. Those are from sweeps! Don't look much like it perhaps. It's become clearer now that the problem is the audio interface is producing some hf noise so that when I turned the REW sweep level dBFS down low to get a more suitable level for the '1073' mic input around the '-40' gain setting the sig/noise was too low and I was amplifying the relative ton of hf noise, which screwed up the freq response plotting. I've now put a crude 40dB attenuator at the input to the 1073 so I can make the sweep signal output as large as possible relative to this source noise and this is producing some much more useful looking plots, albeit they still get messy nearer to 96kHz.

( I haven't tried to match a typical mic source impedance with the attenuator arrangement as such - I'm relying on the fact that I didn't see much difference in overshoot/ringing behaviour from varying this in earlier scope-based efforts. )

 

[ruffrecords]

My mistake, it was the mention of pink that made me think you were using pink noise.

You problem is almost certainly the REW oscillator or more accurately the D/A convertor in your interface. I came across this problem when I first started using REW. The issue is accuracy of the waveform at low levels simply due to the limited resolution of the D/A. The solution is to drive the 1073 through a balanced analogue attenuator. I use a 40dB attenuator myself consisting of two 7K5 resistors and a 150 ohm resistor. The you can access the full resolution of the D/A even for very low level mic signals.

Cheers

Ian

 

[HenryL]

My mistake, it was the mention of pink that made me think you were using pink noise.

You problem is almost certainly the REW oscillator or more accurately the D/A convertor in your interface. I came across this problem when I first started using REW. The issue is accuracy of the waveform at low levels simply due to the limited resolution of the D/A. The solution is to drive the 1073 through a balanced analogue attenuator. I use a 40dB attenuator myself consisting of two 7K5 resistors and a 150 ohm resistor. The you can access the full resolution of the D/A even for very low level mic signals.

Cheers

Ian

Good to get the scenario and solution corroborated.

I've been speculating that since in theory the interface should be able to recreate a sine wave accurately up to half the sampling frequency that the problem was most likely arising from internal clocking hash escaping onto the outputs of the interface. I hadn't considered that 24bit resolution might be insufficient at -40dBFS. Would have to blow some dust off my digital theory but I thought the digital noise floor was way below that at 24bit (about -120 dBFS IIRC).

I haven't bothered making a symmetrical attenuator - should I?
I'm just using a 500R trimpot strapped across the interface output hot/cold and set to about 5 ohms. As its feeding a floating balanced transformer input then it doesn't matter which section of the attenuator resistance you tap for the output surely? (Edit: I think the attenuator may be throwing another spanner into the works now so will have to return to this...)

 

[RoadrunnerOZ]

Good to get the scenario and solution corroborated.

I've been speculating that since in theory the interface should be able to recreate a sine wave accurately up to half the sampling frequency that the problem was most likely arising from internal clocking hash escaping onto the outputs of the interface. I hadn't considered that 24bit resolution might be insufficient at -40dBFS. Would have to blow some dust off my digital theory but I thought the digital noise floor was way below that at 24bit (about -120 dBFS IIRC).

I haven't bothered making a symmetrical attenuator - should I?
I'm just using a 500R trimpot strapped across the interface output hot/cold and set to about 5 ohms. As its feeding a floating balanced transformer input then it doesn't matter which section of the attenuator resistance you tap for the output surely? (Edit: I think the attenuator may be throwing another spanner into the works now so will have to return to this...)

You woul perhaps be better using 2 x 47KΩ resistors - each in series with pin 2 and 3 with a 470Ω resistor and a 500Ω trimpot in series across the output of that and taking the output from across that pair - the trimpot has one leg and the wiper joined - this makes a variable divider network giving -40dB to -50dB approx drop

 

[HenryL]

You woul perhaps be better using 2 x 47KΩ resistors - each in series with pin 2 and 3 with a 470Ω resistor and a 500Ω trimpot in series across the output of that and taking the output from across that pair - the trimpot has one leg and the wiper joined - this makes a variable divider network giving -40dB to -50dB approx drop

I guess my question must be why would that be better?

I noticed my previous attenuator lash-up was showing more noise right down to ~5kHz in the calibration loop test (ie. without the 1073 ) as well as causing a narrow spike in the response curve at about 60kHz. Not sure if some of it was noise pickup from waving wires but anyway I've now measured the Zout of the Focusrite interface and made a new 40dB pad in a screened box with 2K2 on pins 2&3 with 47 ohms across. Just about to give it a test drive.

If I did my sums right (and happy to be corrected!) this is giving a roughly 40dB cut and adds 40 to 50 ohms to the 120 ohms Zout of the interface for an effective approx 170ohms Zout into the mic input, to provide a realistic source impedance.

Cheers

 

[RoadrunnerOZ]

I guess my question must be why would that be better?

Better than:

I'm just using a 500R trimpot strapped across the interface output hot/cold and set to about 5 ohms.

 

[RoadrunnerOZ]

Are you testing in low or high impedance mode?
Edit: on the 1073 that is - 300 or 1200Ω?

 

[RoadrunnerOZ]

According to Focusrite the output impedance of the 18i20 is 430Ω
Edit: just to clarify not sure how you’re getting 120Ω on the line output of the 18i20???

 

[HenryL]

Are you testing in low or high impedance mode?
Edit: on the 1073 that is - 300 or 1200Ω?

High / 1200, my usual setting

 

[HenryL]

According to Focusrite the output impedance of the 18i20 is 430Ω
Edit: just to clarify not sure how you’re getting 120Ω on the line output of the 18i20???

Interesting - where'd you find that?! I looked in my product manual specs and on their web site and couldn't find the info anywhere so measured it myself.

A 1kHz 1Vpk sine wave output (from Logic test osc) from a line out dropped to 0.45Vpk when a 100R load was applied. My late night maths dept made that into about 120 ohms Zout.

 

[RoadrunnerOZ]

Interesting - where'd you find that?! I looked in my product manual specs and on their web site and couldn't find the info anywhere so measured it myself.

A 1kHz 1Vpk sine wave output (from Logic test osc) from a line out dropped to 0.45Vpk when a 100R load was applied. My late night maths dept made that into about 120 ohms Zout.

Possibly your output loading is too heavy at 100Ω - maybe try 600 - 1KΩ.
User manual P32:
https://fael-downloads-prod.focusri...oads/Scarlett 18i20 3rd Gen User Guide V2.pdf

 

[RoadrunnerOZ]

High / 1200, my usual setting

Thought so - that’s why I suggested the 470+500trim

 

[HenryL]

Possibly your output loading is too heavy at 100Ω - maybe try 600 - 1KΩ.
User manual P32:
https://fael-downloads-prod.focusri...oads/Scarlett 18i20 3rd Gen User Guide V2.pdf

1kHz sine from line output adjusted to 1Vpk unloaded, then applying different loads and measuring the resulting new pk voltage on a scope:

Load, Voltage, Approx_Zout (ohms)
100R, 0.45, 122
470R, 0.8, 117
820R, 0.875, 117
2K2, 0.95, 115

FWIW I have a 2nd gen 18i20, not the 3rd gen, and it's not listed in the 2nd gen specs.
(430 sounds slightly surprisingly high to me compared to line outputs on other stuff I have eg a dbx unit I have has spec of 100R, RME 75R etc.)

 

[HenryL]

Thought so - that’s why I suggested the 470+500trim

I was aiming to maintain a typical mic source impedance into the pre73 after attenuating. Intriguing that you suggest going higher - could you clarify the rationale?

 

[HenryL]

Well I think I've got my act together with using REW now, so I'd better get some results down before someone from Australia points out the fatal flaw...

Here is an initial set of REW measurements of the frequency response of the pre73 as far as I could take it, from mic/line inputs through to line output, without any zobel components on the input tx's and then with a few example values of C and series R to get a general idea of what's happening.

The first is a set taken of the mic input at the -40 gain position, the second set is the mic input at -20 and the third set is taking a look at the line input on both -20 and -0 positions. In each mic input case the top trace is with no zobel at all. In the set of line input plots I also left in the unzobelled mic input plots for comparison, at the top.