Is DEQX a game changer?

Andrew (Drewan) thanks for the clarification. That would certainly explain why your group delay plot appears as it does. Also, I note that the excursions in the plot appear to be greatest in the vicinity of what is probably the crossover point between the two drivers. Which makes sense, as in that vicinity the measurement mic would be receiving comparably strong outputs from two different drivers which are at two different physical locations.

Also, since DEQX was providing for you a lot of the optimizations that would be provided by the manufacturer in the case of speakers such as mine and Bruce's, the criticality of your speaker measurements being as accurate as possible (i.e., taken outdoors) figures to have been increased correspondingly.

Best regards,
-- Al

Yes Al your interpretations are correct. It's also worth pointing out that my graphs are the 'sum' of tweeter and midrange readings taken concurrently for each speaker so I could produce comparative graphs to your own (I am using six channels of amplification in this setup and these graphs relate to four of them).

When I look at the individual midrange or tweeter plots, the Group Delay variances are quite a lot lower than the summed result you see. I am also able to view each measurement graph per driver and that helped me during the prototyping phase of developing the speakers to arrive at both the measurement and the sound I was after.

In each measurement I took, it was gratifying to see how close the plots of each speaker became. I had specified matched pairs of drivers when I purchased them and the workshop cutting the various speaker panels programmed the CNC to very fine tolerances based on my drawings. The finished speakers measured to be virtually identical even before DEQX calibration (not to imply that they sounded good until after correction).

I've performed the experiment I indicated I would be doing to investigate the glitch at 0.7 ms which was evident in the measurements I posted the other day, as well as in the measurements I described taking from other distances.

What I did today was to measure one speaker in its normal position, from a distance of 30 inches, with and without a boom on the mic stand. The resulting impulse response plots made clear that the glitch had indeed been caused by a reflection from the mic stand. However, in comparing today's results with and without the boom, there was little or no difference in the step response at or near that point in time. Also, despite the much greater room reflection content of today's measurements compared with the ones I made the other day with the speakers in the center of the room and with acoustic panels placed as shown in the photos I posted, aside from the deep bass region the frequency response plot derived from today's measurement with the boom matched remarkably well with the frequency plot taken from the same distance the other day (without the boom, with the speaker in the center of the room, and with the acoustic panels in place).

So, thankfully, I can proceed without having to re-do the measurements, which in my case involved a laborious 5 hour effort. :-)

Best regards,
-- Al

The group delay plot should be relatively smooth post windowing, and should not have any "spikes", if you have those it is a sign of reflections corrupting the measurement.

I’ve spent some time studying the measurements I previously described having taken, and observing on the computer screen the results of applying various window durations to them. As a result I’ve chosen two specific correction filters to evaluate sonically in the coming days. I’ve uploaded jpg files depicting those filters, which can be viewed here. There are ten files, five for each filter, depicting the corrections in terms of frequency response, group delay, impulse response, step response, and phase response. In interpreting them, be sure to take into account the scales marked on the vertical axes. (To see the markings clearly, click on the image thumbnails to expand them). The measurements, window durations, and correction limits that were used in creating these filters are described below.

The ways in which I narrowed down the many possible correction filters to these two are as follows:

As you may recall, I had performed measurements of each speaker at distances of 30, 36, and 42 inches, with the grilles removed, and also at 36 inches with the grilles in place. The grilles appeared to make essentially no difference.

I found that both 36 inch measurements of the right speaker had a huge group delay spike at about 420 Hz, which was not helped significantly by smoothing, and which did not appear in the 30 or 42 inch measurements of that speaker, and which did not appear in any of the measurements of the other speaker. I have no idea why that occurred, as the two speakers and the mic were positioned identically when they were measured, within perhaps 1/8” or less. The only variable that seemed to be present, which in turn seems very unlikely to have anything to do with that spike, is that the design of the speakers is such that their rear surfaces, rather than being parallel to the front baffle, or being otherwise identical between the two speakers, are mirror-imaged at an angle such that the side of each speaker that is closest to the other speaker (in their normal positions) is 1.5 inches shorter than the other side. In any event, due to that spike I eliminated the 36 inch measurements from consideration.

I experimented on the computer with windowing of the 30 and 42 inch measurements at three different points, each of them just prior to what appeared to be significant reflections or increases in reflections, at about 13, 17, and 21 milliseconds. The 21 ms window resulted in significant frequency response wiggles in the 500 Hz to 1000 Hz area, so I eliminated that choice.

The 13 and 17 ms window durations provided results that looked fairly similar, but 13 ms (which terminated just prior to what I’m pretty certain was a ceiling reflection, based on its timing) looked slightly more promising. So that’s what I went with, for both measurement distances, that also being exactly what Andrew (Drewan77) had suggested the other day after looking at the measurements I had posted.

The (approximately) 13 ms window duration (actually 13.1 ms for the 30 inch measurement and 13.2 ms for the 42 inch measurement) corresponds to 7.4 ms after the initial sound arrival for the 30 inch measurement, and 6.7 ms after the initial sound arrival for the 42 inch measurement.

Regarding the correction limits I set, I used the default amplitude limits, which in turn were not called into play at all within the frequency response limits that I set. For both correction filters, Filter 1 corresponding to the 30 inch measurements and Filter 2 corresponding to the 42 inch measurements, I set frequency response limits such that corrections were only performed between about 400 Hz and 10.5 kHz. Those choices being made taking into account suggestions from both Nyal Mellor of Acoustic Frontiers and Alan Langford of DEQX to be conservative in dealing with the top octave, and to avoid correcting further into the bass region than seems reasonable in relation to the window duration. With the latter determination being made in the manner I described in my post dated 6-22-15. And, also, taking into account a presumption I made that both the high frequency and low frequency limits should be chosen such that abrupt discontinuities in frequency response are not introduced at the limit points.

Finally, in deriving the correction filters all parameters which I haven’t mentioned were used at their default values, including 100% smoothing.

Best regards,
--Al