I was
not "targetting (sic) scientific perfection". I was doing a job as a qualified professional electronic engineer working in a studios complex dealing with a safety issue.
I would have been happy with a reliable +/- 3dB, but it was almost impossible to get that. I take my hat off to anybody who does a good job of headphone measurements. If you read Amir's headphones reviews, you will see how difficult he finds it to get consistent results with some devices!
Your idea may be valid and possibly useful, but the variances you are posting are evidence that your methodology is sufficiently flawed that you should
not draw any conclusions.
If I was you and wanted to make this a useful approach. I would:
- Find a way to reliable couple the two devices in a way that they cannot be disturbed even by 1/10th of a millimetre (since that is the tolerances you will need to work with) e.g. by gluing them together;
- Make sure the coupled devices are mechanically isolated from their surroundings (holding them together with your hands is absolutely certain to mess up any measure);
- Ensure the coupled devices are a long way from any electrical interference sources;
- Fix the cables so they cannot be moved and don't pass mechanical perturbances to the coupled devices;
- Measure the signals using 400Hz and 1kHz and ensure both devices are giving the same output +/- 0.1V
- If you are using a noise source, pick a good one which has excellent randomness and run each test for several minutes
- Alternatively use REW and do frequency sweeps
- Repeat each test 20 times and create an averaged plot
- Look for any outliers, they can be interesting