This was more interesting to me than most all of the debate over whether the speakers are junk or pretty decent and an exceptional value....
Thanks for your interest in the mods.
But back to your mods. I'd be very interested in seeing the crossover schematic both before and after your mod, and including that series crossover for the two tweeters. In a subsequent post of yours you showed the individual on-axis responses and concluded that the the super-tweeter contributes very little below 10 kHz. But perhaps things are different in the off-axis response? I.e., it might be that the tiny tweeter has substantially better dispersion, such that in the listening window response or whatever, its contribution is much stronger than what your graph implies, possibly even stronger than the bigger tweeter depending on the off-axis angle.
I don't intend to post the schematics, out of respect for the engineer(s) who developed the crossover, and out of fear of any liability I may face if a competitor reverse engineers the design, etc, etc. If you are interested in the schematic, you can buy a pair and open them up like I did. Knowing that the tweeters use a series crossover is a big hint to figuring out what's going on in there.
Your anti-diffraction mod is interesting. You provided response plots to show the before/after effect. You did this for both the crossover mod and the anti-diffraction mod, and to give you the benefit of the doubt, you would have done these measurements with just one or the other of the two mods, such that you avoided confusing the effects of the two mods.
You don't have to give me the benefit of the doubt. Post 9 shows the impact of the frame by itself.
https://www.avsforum.com/forum/89-speakers/3068362-improving-sony-ss-cs5.html#post58076502
Diffraction occurs over a broad range of frequencies starting when the wavelength is small relative to the enclosure or the baffle. However the 1st diffraction peak occurs at a fairly well-defined wavelength where the wave propagated at the corner is in phase with the direct wave. The wave propagated at the corner is a wave associated with a "soft reflection", where a 180-degreee shift in phase occurs. Thus, the longest wavelength at which constructive interference can occur (i.e., the first peak in the diffraction ripple) is the wavelength at which the distance from the center of the driver to the edge of the cabinet or baffle is equal to one-half wavelength. In other words, the 1st peak in the diffraction ripple, for a driver centered horizontally on the baffle, occurs at wavelength matching the width of the baffle. In this case, 7", which corresponds to frequency of roughly 2 kHz. Some reinforcement will occur below this frequency and some above, however there just isn't any way that the sharp response peak at about 1.1 kHz can be attributed to diffraction. The broad elevated response that you described, from 900 to 2.2 kHz, is almost certainly a feature of the driver itself. Diffraction may contribute in the upper half of that plateau, and at 2 kHz especially, but this is the upper end of the plateau.
There's a few things going on here. First, my measurements aren't entirely consistent with Amir's. I didn't see a 4 dB jump between 800 and 1000 Hz. The increase I measured was more on the order of 2 dB. I used gating only in my measurements, no averaging. The difference there is either between our measuring tools, or between our samples, or some contribution from both. I certainly don't have a comparable measurement system. Lifewire measured something like a 3 dB jump there, with unknown amounts of averaging. Dennis measured about 2 dB AFTER the frame was in place, which may agree with Amir's 4 dB jump with no frame. It's not really important, but when I refer to the trends, I tend to refer to (or be influenced by) my own measurements.
Diffraction doesn't explain the significant discrepancy Amir measured between 800 Hz and 1000 Hz, so there is something going on with the output of the driver he measured. On the other hand, I simulated the effect of diffraction of a generic 3 way with something close to these crossover points and driver sizes and positions. In simulation, the effect of diffraction becomes positive at around 600 Hz, and doesn't go negative again until about 2700 Hz. With that in mind, diffraction DOES contribute to the plateau from 900 Hz to 2.2 kHz.
I recall reading a few years back the writeup that Linkwitz did on baffle diffraction, which is probably still on the site (it is still active). He had something to say about the effectiveness of rounded edges on cabinets. I read through it too quickly and this was probably more than five years ago, but on the off chance that I recall what he wrote, the rounding starts to become effective where 1/4 of the wavelength is shorter than the radius of the round-over. Since you used a pipe with 4" diameter (approximately), this implies that the longest wavelength at which it would be even modestly effective should be 8", or about 1.7 kHz. Of course this effect is a gradual thing, but the point is that rounding the corners doesn't have much effect until reaching higher frequency where the 2nd peak, 3rd peak, etc., of the ripple are found, but not the 1st peak, which is always the strongest and the only one that can't be neutralized by placing the affected driver 1/3 of the distance from one edge to the other edge. Thus, even if the sharp peak at about 1.1 kHz or the broad elevation from 900 to 2.2 kHz were genuinely diffraction effects, the mod you did would almost certainly not have any significant effect at these frequencies.
The impact of the frame on-axis is two-fold. It spreads the loss of energy at the edge over time, but in this case it also extends the edge. It's acting like a wider baffle. This has the measured effect of increasing the output from 400 Hz (which was the gating frequency) to 800 Hz by a couple of dB relative to the 7 inch baffle alone. My simulations show the theoretical frame elevating all frequencies from 100 Hz to 1000 Hz, which is pretty similar. The frame also has the measured effect of decreasing output from 900 Hz to around 2.5 kHz. My simulations show a decrease in output from 1000 Hz to about 2.6 kHz, which again is very similar. If you look at Amir's measured on-axis frequency response, and consider raising the output from 100 Hz to 800 Hz, while decreasing the output from 900 Hz to 2.5 kHz, that's a big win.
But additionally, the frame smooths out the response, and creates greater similarity across azimuth. Without the frame, there's a measured difference between on-axis and 60 degrees off-axis of 5 dB at 2 kHz. My simulation predicts a spread of 4.6 dB. With the frame, the measured discrepancy is 2 dB, and the simulated discrepancy is also 2 dB. In both the measured and simulated plots, you can see the on-axis response and the 60 degree off-axis response both converge on the 30 degree off-axis response when the frame is added. If you look at Amir's Horizontal Directivity plot, you'll see that from 800 Hz to 10 kHz, the 30 degree response is better than the on-axis response.
I am including 2 simulation plots that show 0, 30, and 60 degree frequency response with and without the frame. I didn't do anything to fix the baffle step inherent in the simulations, which I think is informative for the comparison. Also, no irregularities of the drivers used were included. I believe my simulations are over-estimating the dispersion of the drivers some (which exaggerates the interference, particularly at higher frequencies), but I haven't taken the time to fix that for this discussion. I only want to show that the theoretical effects and the measured effects have great similarity at the frequencies you questioned.