I just revisited my tweaks to the DSE A2667 speaker (an OK quality cheap speaker commonly available second hand in Australia for about $20). This is because I had the loan of a WooferTester 3 unit from Col, so I could measure the drivers properly, and also had been given an extra pair to experiment with (thanks Gavin!).

TLDR summaryI made one additional adjustment to the crossover that made the response even flatter, by adding a 4.4uF capacitor to the LF portion to make it a 2nd order filter, while inverting the polarity of the tweeter.. but at the cost of an increase in measured distortion at about 3.5khz.

Steps:

Measure the T/S and impedance characteristics of the woofer and tweeter
with WT3.

I was also able to save the impedance data of the two drivers as a .ZMA file for later use.

Then as I had the turntable ready I measured the MK I tweaked version for directivity while I was at it, I actually was suprised how good it was, I’d be curious to see this info as a heatmap.

This done at 2M distant, indoors, with a measurement taken every 7.5 degrees to 60 degrees, smoothed to 12/octave.

Then a measurement comparing the original tweaked version and a stock DSE A2667:

mic 2M distant
measurement done indoors. smoothed to 12/octave.
As can be seen the tweaked version considerably flatter in response, a huge dip at 8khz has been considerably reduced by the 4.4uF capacitor lowering the crossover frequency compared to the original 2.2uF capacitor.

Then I took one of the unmodded speakers and added extra wires to the drivers direct, disconnecting the crossover. I threaded these out the bass reflex port and measured the individual response of each driver, adding a 10uF protection cap to avoid burning out the tweeter.
Did this at 1M and 2M


I also exported these as text (.FRD) files from REW so I could import the data into the PCD spreadsheet.

Unfortunately the phase data from REW had “rapidly fluctuating localized phase response” – which when imported into PCD gave me jagged summed response. So it was back to Holmimpulse for measurements, which appears to measure phase properly. Later I will revisit getting phase measured properly in REW as in many ways it is more useful than Holmimpulse.
Also while I was searching around for solutions to this I found the most excellent WinPCD package – not all the features of the PCD spreadsheet are implemented, but it is standalone software that does not require MS-Excel to be installed, and it has a seperate resisable window for the system calculated frequency response which makes it easier to work with.
Jagged summed response caused by bad phase in the REW data, WinPCD screenshot:

Smooth summed response using Holmumpulse measurements with coherent data:

The author of the WinPCD package was quite helpful on the DIYaudio forums in sussing this out as well, by this time I’d been going in circles trying different smoothing of the data files before the messy phase in REW occured to me

Now that I had decent measurements done 2M from axis (except for the loss of HF response on the tweeter), adjusted WinPCD until I had a response that was flatter than the stock and tweaked response. This turned out to be by adding an addition 4.4uF capacitor to the 0.56mH inductor on the woofer part of the circuit, turning it from first order to second order. The reasoning here was to reduce the amount of overlap between the LF and HF drivers in the crossover region. Due to the alteration in phase, I also inverted the polarity of the tweeter, otherwise there was a considerable dip in the response at the HF area.
I tried a few other values, but without bumping up the cost of the crossover considerably (e.g. an addition inductor for a 2nd order high pass filter) this seemed to be the best adjustment with the components I had readily available.
Here is the WinPC screenshot with the seperate driver responses plus summed response (remember some HF loss due to damaged tweeter:

I then went and took one of the already tweaked A2667 speakers, and added the 4.4uF capacitor to the underside of the board.

Then measured it and loaded the measured response into WinPCD:

There is some variation (apart from the HF loss due to tweeter damage), but this could be due to component and driver tolerance, but measured and calculated fairly close. Whats more the response is now much flatter than the “Mk I tweak” (for future reference I’ll call this the “Mk II tweak”.

Then I went back and did more complete measurements of stock, Mk I & Mk II tweaked speakers using Holmimpulse and REW – all done at 2M on axis, with 6/octave smoothing. I took screenshots of the distortion graph in REW:
Stock crossover:

Mk I tweak:

Mk II tweak:

Stock and MK-I tweak, holmimpulse:

Stock and MK-II tweak, holmimpulse:

All three together:

conclusion
So what the above graphs tell me is I have much flatter frequency response – but at the expense of a 15dB increase in harmonic distortion clumped around the 1.5khz – 4khz area.
I’ve got one pair of speakers with the MkII tweak done, so I’ll have a listen, but so far I suspect this tweak has made some improvements in response at the expense of distortion.

Note that the final measurements were done on a different speaker to the one I did some damage to the tweeter while measuring.

extra info:
here are the screenshots of the crossover circuit portions as displayed by WinPCD, and the raw data files I was using:

DSE A2667 woofer frequency FRD file impedance ZMA file
DSE A2667 tweeter (damaged response above 14khz) frequency FRD file, impedance ZMA file

REW plots of the individual driver response, with distortion info (all on axis, 2M):

woofer only:

tweeter only, with 10uF protection cap:

tweeter only, WITHOUT protection cap, but trying from 1Khz.. this is where I think I damaged the HF response: