using MiniDSP UMIK-1 to Calibrate a Behringer ECM-8000

note: article incomplete, I need to tidy up the graphs and clarify what the lines are.

I recently purchased a MiniDSP UMIK-1 calibrated USB measurement microphone to help with speaker tuning. I already have a Behinger ECM-8000, but the Behringer isn’t calibrated, while the MiniDSP comes with a calibration file, you just download from their site keyed on the serial number of your unit.
As you can see, the two units come in similar, but not the same packaging, and also look quite similar:


I had been using the “generic” ECM-8000 calibration curve, but there is a fair bit of info out there that there is quite a bit of variance between different examples of the ECM-8000.
So why not try and calibrate the ECM-8000 using an already calibrated Mic? I know there will be inherent loss of precision during the process , but the result may still be more accurate than the generic file.

Equipment used:

Laptop with Homimpulse
Behringer Xenyx 1204FX Mixer
Behringer UCA-202 USB interface
old Windows XP laptop
Homimpulse 1.4.2.0 for measurement

Sound source is my “under construction” active stereo speakers with the Behringer UCA-202 connected via SPDIF output. The microphones place in a mic stand about 1M from the speakers. I’m not actually concerned too about the response of the speakers now, just that the measurements remained consistant between holmimpulse runs, which they did.

Stage 1:
As a sanity test I’ll calibrate the Microphone against itself. If I can generate something resembling the calibration curve by comparing measurements with and without the calibration file applied, I should be in the right ballpark.

Here are the homimpulse measurements with and without the calibration file, smoothed to 5 per octave to make the differences easier to see:

 umik-with-and-without-calibration-smoothed-to-5-for-visibility

Here is the “no calibration” measurement, divided by calibrated – using the Holmimpulse frequency domain manipulation functions.. turns out division (uncalibrated / calibrated) gave me the curve back.

I used the raw response (no smoothing or gating) data as the source of the calculation, then smothed to 20 for visibility, phase displayed just to show mostly flat.

This is superimposed on the actual calibration curve supplied by miniDSP. To line them up exactly I had to add 2.8 dB to the calculated cuve.

umik-1-nocalibrated-divided-by-calibrated-offset-smoothed-20

From this I can be fairly sure the method I’m about to use with the Behringer ECM-8000 will give me a usable calibration curve.

Stage 2:
Calibrate the ECM-8000

For this I measured the ECM-8000 the same way, through the Behringer mixer and USB sound card. I’m not concerned that much about lining up the dB magnitude as it’s arbitrary due to the level settings on the mixer (chosen to have decent headroom without clipping), so this is only a relative frequency response calibration, not an absolute dB sound level.

Below is the ECM-8000 measurement, smoothed for readability to 5/octave, and superimposed on the uncalibrated UMIK-1 measurement. The ECM8000 measurement has also been reduced by 6dB to line up closely with the UMIK measurement for comparison. They are quite close.
umik-1-nocalibr-ecm8000-no-calibr-smoothed-5

Calculated ECM-8000 calibration curve (red). It has been offset by -6.5 to line up with 0dB, and the UMIK-1 calibration file and generic ecm-8000 calibration file are also shown (blue and green).
Smoothed to 20, although the imported calibration curves obviously are even smoother.

ECM-8000-calced-calib-with-umik-and-generic-smoothed20

With Holmimpulse I can export as a calibration file. The file exported is 1997 lines compared to the 560 lines of the UMIK-1 calibration file, so possibly overly detailed compared to the source data – but using this calibration file with Holmimpulse, I do a frequency response sweep and compare against that done with the UMIK-1:

And the calibrated/uncalibrated curves are quite close.. with the UMIK-1 curve a bit different.. so I can conclude this method probably not too useful, although the results look good.