Is DEQX a game changer?

Thanks. I may try using some of the GIK panels to do something similar around the mic and compare that to the measurements without. Unless that seems like a boneheaded idea to some of you folks wiser than I.

Caution: Very long post :-)

Well, it seems that my forward progress has taken a step backward, although the step backward has resulted in discoveries that will hopefully be beneficial in the end, to me and to others who may read this.

After evaluating four different speaker correction profiles on numerous recordings, all based on measurements taken with the acoustic panels I purchased placed around the measurement microphone, I’ve concluded that even though those measurements, when viewed on impulse response plots over a reasonable time scale (e.g., 30 ms), looked considerably better than the measurements taken with no panels and with the panels placed around the speaker being measured, the panels were doing more harm than good. And I say that even though there were no points on the impulse response plots at which reflection amplitudes appeared to look greater with the panels placed around the mic than in the other plots, and at the great majority of points reflection amplitudes looked significantly smaller with the panels placed around the mic compared to the other two cases.

As will become clear later in this post, what I probably should have done was to place the panels up against the nearest reflective surfaces (the fireplace on the left and the large antique radio/phono on the right, as seen in my system description photos), rather than surrounding the mic with them on three sides, at a fairly close distance.

I’ll first say that one of those four profiles (one of the first two I tried, with the impulse response truncation window terminated 7.2 ms after the initial sound arrival, and corrections only performed between 600 Hz and 10 kHz), on most but not all recordings sounded distinctly better than bypass mode and than the other three profiles I tried (all of which had the window terminated about 18 ms after the initial sound arrival, with corrections over a somewhat broader range of frequencies).

However I noted especially on the last two profiles I tried that the image was shifted considerably to the right, even though bypass mode (as always) was perfectly centered. And that right channel boost (the difference being in the area of 2 to 6 db depending on frequency, and occurring primarily between about 200 Hz and 1 kHz) could be clearly seen when the correction profiles were viewed in the DEQX software, with the profiles for the two speakers placed on the same graph and the scale of the vertical axis suitably adjusted.

I also found while doing further experiments with the software that the volume difference between the two channels increased dramatically in proportion to the duration of the truncation window, which really puzzled me at first, and also increased in proportion to the distance of the measurement microphone from the speaker (meaning also that it increased as the distance between the mic and the panels behind it became smaller). And the volume difference was worse for correction profiles created from the measurements made with the panels surrounding the mic than for correction profiles created from the measurements made with the panels surrounding the speaker. And, as mentioned above, I found that the issue occurred primarily (although not exclusively) in the area of 200 Hz to 1 kHz, especially around the middle of that area, with the volume difference varying considerably at different frequencies.

After a lot of study of frequency response plots, impulse response plots, and step response plots, I concluded with a fair amount of certainty that the cause of the different corrections for the two speakers was that I didn’t have the panels placed in precisely the same locations when I measured the two speakers. The reason for the slightly different placements being that since I was making measurements with the panels surrounding the speaker as well as with the panels surrounding the mic, and it happened that I did the measurements with the panels surrounding the speaker last, I moved the panels aside when the first speaker being measured was moved away from the center of the room, and the second speaker was moved into that position.

Upon very close examination, the consequences of that can be seen in terms of slight differences between the timing of the wiggles of the impulse response measurements for the two speakers in the area of about 3 ms after the initial sound arrival, and can also be especially seen in the form of a roughly 1 db difference occurring at that same instant between the step response plots of the two speakers, in the cases of the measurements taken at 3 and 3.5 foot distances which I used for the correction profiles.

So, I wondered, if the issue was being caused by reflections from the panels occurring just 3 ms or so after the direct sound arrival, why would the consequences of those reflections in the correction profiles get worse as the truncation window was extended much further out in time, for instance from 7 ms to 18 ms and beyond? I’m not totally certain, but I believe the answer to that is inherent in the mathematics of the Fast Fourier Transform, some variation of which I assume is what the software uses to convert between the time domain (impulse and step responses) and the frequency domain (frequency responses).

Now if I were to redo the measurements while making a point of placing the panels at precisely the same locations for both speakers, I could evidently eliminate the inter-channel differences in the corrections. However, the fact that slight differences in panel placement caused dramatic differences in calibration profiles between the speakers would seem to say that even if I were to achieve identical profiles for both speakers, that identical profile would reflect (pun intended) significant adverse effects of the panels. So for that reason, in addition to the effort that would be involved in re-measuring the speakers, I’m not planning to do that. Instead I’m now planning to simply try some correction profiles that I’ll create based on the measurements I’ve already taken with no panels in place. If those don’t work out well, then I’ll consider re-measuring the speakers, with the panels much further from the mic and probably placed against the reflective surfaces I mentioned earlier.

Also, if I were to try to correct the inter-channel differences using the equalization capabilities of the DEQX, given the extensive variations of those differences as a function of frequency I suspect that the effort would be extremely time-consuming, and would probably result in a less than ideal set of complementary colorations.

So although my efforts have had a bit of a setback here, it’s probably a good thing that I didn’t make a point of placing the panels in exactly the same positions for the measurements of the two speakers. If I had done so I probably wouldn’t have discovered any of this, and I would very conceivably have ended up deriving less benefit from the DEQX than I hopefully will, eventually. So, undaunted, I shall persevere and carry on. Due to various unrelated upcoming activities, my next significant update will probably be in about a week. Meanwhile, just using the DEQX in either bypass mode or with the one correction profile I mentioned as being superior to the others, provides (despite a bit of channel imbalance in the case of the latter) a modest but notable improvement on what I previously had.

Best regards,
-- Al

A very interesting summary Al and I sympathise with the challenges you face

At the risk of repeating myself (apologies), are you absolutely sure that there is no chance of taking an outdoor speaker measurement?

I originally made several attempts to achieve a good indoor measurement but none came close to those taken outdoors. Once I saw the cleanness of the resulting plots, any further speakers or subs were always measured this way - including manhandling a huge Miller & Kreisel sub into the garden and then up onto the measuring platform which was a great deal of effort

It does make a significant difference and I believe that is why music created from the resulting calibrations sounds so remarkable. I am rather a 'perfectionist' so I couldn’t have rested knowing that I wasn't getting the maximum performance from DEQX

Yes, this can involve a lot of time and commitment (in my case around three hours from start to finish, moving the gear etc) but provided you use a high number of averaged measurements (I recommend 1.4s/96K x at least 18 sweeps) and the day is completely dry without much wind, the measurements only need to be taken once and are extremely accurate, virtually anechoic if done well. Any random birdsong or other faint wind noises appear to be completely disregarded by the averaging feature of the software

Somewhere I may have photographs of the setup but they are not on this computer so I will have to look for them. If anyone is interested I would try to find them plus some of the resulting measurement plots and post them as images on 'my page'

Thanks Al and Andrew. I guess I'll hold up on buying the panels for a bit. I'll wait for future reports.

Andrew, it may be that I will bite the bullet and schlepp my whole rig outside one day. Gotta tella ya though ... I am concerned about damaging gear by doing this.

As I've already mentioned, my rig is tucked away in my basement/sound-room. At a minimum, I'd have to haul up the steps: two 100 pound speakers; an 80 pound amp; and the DEQX. And what about my 110 pound sub?? I assume I can leave the other gear downstairs.

I have posted a couple of screenshots under 'System' at the foot of my posts for anyone with a DEQX processor to compare the plots

The measurement shows how 'clean' the frequency plot may look relative to an indoor reading that others may have taken and the impulse response illustrates that there is only a tiny reflection from the microphone itself

(The measurement is a raw image from the mid-high range speaker before any correction & at 0% smoothing. Indoor calibrations tend to use up to 100% smoothing and a closer windowing, both of which reduce accuracy and have an impact on the sound in my experience)