KEF 107 frequency response Measurements

Ok, I found the 1/3 octave setting, and the curves look much easier to compare. Thankfully, I still like the looks of the DX25 the best. It eliminates the original tweeter's dip at 3-4kHz, and has a nice flat response above that.

The AC25 actually looks a lot better with the 1/3 octave smoothing. My ears must not have liked the extra energy in the 10kHz region, but other than that it looks almost as good as the DX25.

The XT25 still looks very good in terms of its flatness, but again the off-axis response and lower output (and higher price!) makes it a non-contender for me. I could imagine that someone with more crossover design experience than me might be willing to adjust things for better balance with the midrange, but that adds even more to the price. . .
 

Attachments

  • KEF T33.pdf
    35.3 KB · Views: 46
  • AC25.pdf
    35.7 KB · Views: 15
  • XT25.pdf
    35.5 KB · Views: 12
  • DX25.pdf
    35.6 KB · Views: 25
  • All Tweeters.pdf
    81.3 KB · Views: 39
On-axis and 30-deg off-axis curves with 1/3-octave smoothing. No midrange.
 

Attachments

  • DX25 off-axis.pdf
    53.1 KB · Views: 16
  • XT25 off-axis.pdf
    54 KB · Views: 12
  • AC25 off-axis.pdf
    53.1 KB · Views: 16
  • KEF T33 off-axis.pdf
    52.8 KB · Views: 21
I can see that by those graphs, the DX25 does more closely resemble the "shape" of the response curve of the original T33, than the other alternatives...

HOWEVER- one thing I see... with ALL the Vifa alternatives... there's SIGNIFICANTLY more output at 1KHz than with the T33. This may require some crossover tweaking... a bit of an increase in the crossover frequency, or an increase in the slope at crossover. Without doing this, I would worry about there being a significant increase in energy between 800 Hz and 2KHz... which may or may not be objectionable.

This crossover tweaking might get involved, though- as my "intuition" tells me that it might want a slightly smaller inductor (to ground) in the highpass filter in the crossover. That can be hard to test... other than just substituting a smaller inductor, wholesale.

Regards,
Gordon.
 
Thanks for your comments Gordon, but can I ask you to clarify? I don't see a noticeable difference at 1k between any of the tweeters (that is well below the 2.5 crossover point). With the midrange in the circuit, all the curves look identical until 2.5k.

I agree there is a significant difference in the 4k area, but to me the increased energy there looks like a good thing, since it is more in line with the rest of the frequency range.

And Pete, thanks for the link! I have seen that, and a few others that compare the tweeters. It is interesting that Zaph's charts show the DX having decreased output in the 10-20K range, while the XT is flat. In my measurements, though (which only go up to 15K), the DX is almost ruler flat up there, while the XT seems to have a very mild roll-off, if anything. I wish I could figure out why I can't get readings above 15K!
 
On another note, I'm now thinking about adding a very small 1.5db fixed L-pad to the DX25. It seems to me this will get even closer to a good match with the midrange. I don't know how Zilch saw this just from looking at the charts, but I'm hearing it now, and it's starting to make sense in the charts as well.

I thought people weren't supposed to be able to hear level differences under 3db very well, but all of these very subtle differences seem to make a world of difference to me in getting the balance just right.
 
Did you measure the DC resistance of all these tweeters?

The attenuation level at 1 kHz is interesting, the T33 seems to have much more, Perhaps there is not much of an input impedance rise at resonance with the T33?

Look at the curves of the tweeters alone around 1 kHz, there is much more output from most of the replacements you tried. I was not sure if you matched Fs, but if you look at Zaph's curves you can see that the DX and XT have a much lower Fs, and as expected, much more low freq (LF) extension. While this might still blend just fine with the mid, it means that these tweeters will have more LF displacement and hence distortion.

The AC has a higher Fs that I think is closer to the T33, and it was a better match for this reason, it also seems to have better off axis response. I would have just padded the AC and perhaps tried some diffraction control if a flush mount was not possible. I might consider modding the DX to provide a closer match to the T33, it probably has a chamber that you might try filling to raise Fs.
 
Hmmmm, things are starting to get interesting. . . .

I guess I missed your point the first time around, Pete, and it took Gordon's reiteration for it to sink in with me.

OK, so if I understand correctly, the increased energy of these tweeters around 1K might contribute to increased distortion in that area?

I would have thought that the difference between -33 db and -45 db wouldn't matter too much, but I guess I'm wrong on that account. It doesn't seem to have much effect on frequency response, but, if I'm understanding you correctly, it could cause some interference with the midrange, causing some sort of distortion?

Anyways, I'm posting a chart of all four tweeters on the same graph to compare. The AC, despite its closer matching Fs, actually has the highest output in that area (the red curve), so it would seem to be the worst match in that one respect. The XT is closest, with the DX in the middle. All of them, as Pete and Gordon pointed out, are significantly higher than the original tweeter.

Wait a sec, here's an idea. For these measurements, the T33 was flush mounted, in a chamber that was considerably damped. All of the other tweeters were fitted into the mounting space, but they all stuck out about a 1/2" or so (I couldn't modify the mounting plates since I had to return them to the retailer). They were therefore not fitted into the enclosure, and did not have the benefit of an airtight seal with the damped chamber. Could this account for the difference in the response in that area?

Now that I have the DX mounting plate modified in order to be properly flush mounted to the enclosure, I suppose I should take some more measurements. This time I plan to take measurements from a few different locations, averaging the responses in order to minimize room reflections. Also, I would like to measure with a small L-pad on the tweeter, to take off the touch of extra treble energy that I'm hearing.
 

Attachments

  • Tweeters sans midrange.pdf
    84.9 KB · Views: 26
OK, I went back to do more measurements today with more care taken in setup to minimize reflections etc, and with more attention in general.

Please note that the db level is different on the two different days that I measured. So, for the first two graphs, ignore the difference in levels, and look for the relative shape of the curve.

First, with the DX25 mounted flush, and the midrange out of circuit, there is an improvement. The 1K point is still relatively high, but the rolloff slope in general is a lot closer to the original tweeter. In particular, it starts rolling off earlier than when it wasn't flush mounted, and at about the same point as the flush-mounted KEF tweeter (1st graph). This means that the rising slope I measured earlier in the 2-5k range is much flatter with the flush-mounted DX25 (2nd graph).

It is still putting out significantly more energy than the original tweeter in the 1-3K range, though. What sort of problems could this entail (I'm still a bit fuzzy on that), or is it sufficiently buried by the crossover?

OK, now on to the full-range graphs from today.

I decided not to add a small L-pad after all, since I did a test with the grills on, and its looking a lot better. I had been doing all testing and listening with the grills off up until now, but the grills do a good job of padding down the 5-7K hump (graph 3). So now with the 3-5K rise and the 5-7K hump cured, I'm liking these even better. The only thing is that there is now a bit extra in the 14k region, and a tiny dip in the 5k region, but I'm not sweating that too much.

My ears are telling me to leave the grills on, since the treble was just a touch too tizzy for me before, and it is now sounding smoother. Based strictly on the curves, though, anyone care to give a second opinion on the grills-on vs grills-off problem?

I've also included off-axis response curves taken at 10-degree intervals, both with and without the grills on (the swiveling heads really helped make these measurements easy). They seem to do pretty well up to 30-degrees, before HF response gets really poor.

Anyways, I think I'm about done with this adventure, unless anyone wants to elaborate a bit more on what might be going on with the 1kHz tweeter output. Is there a way to listen or test for the distortion that might be going on? I'm not hearing any problems right now, but I don't really know what to listen for.
 

Attachments

  • no midrange comparison.pdf
    70.8 KB · Views: 13
  • DX25 flush mount comparison.pdf
    54.1 KB · Views: 24
  • 107DX grill comparison.pdf
    52.6 KB · Views: 17
  • 107DX off-axis, with grills on.pdf
    128.9 KB · Views: 8
  • 107DX off-axis, with grills off.pdf
    129.2 KB · Views: 9
Nice work, I don't think anyone would complain about those curves if your goal is flat response. Systems were being desinged in those days to take the grille into account so this might be a factor.

Driver displacement increases as frequency decreases in tweeters, just as it does with woofers. Large excursions at 1 k can cause distortion (HD, IM, all types) in the passband. The designers used a particular XO slope for a reason and 10 dB is a significant difference. It is easily explained by the lower Fs of the DX tweeter. It seems that the XO amplitude response dominates the tweeter rolloff response with regard to total system summation so that the total system output does not suffer. This is good, the flat response indicates that the driver integration has been preserved. I notice that you're measuring off axis, a better test is to test the vertical axis since this is where phase differences between the tweeter and mid will show up more. The goal here would be to determine if the DX perserves good FR for both seated and standing listening positions and to determine if the new tweeter generally provides the same off axis response.

You could also try inverting the polarity of the tweeter both with the T33 and with the DX to see if you get a notch at nearly the same frequency.

Also note that the highly complex crossover might have compensated for the tweeter impedance rise at Fs and the mismatch might cause some strange behavior regarding the system input impedance. I believe this system also had conjugate load matching to provide a reasonably flat input impedance.

You've done a nice job, I'm just covering all the bases since you seem to be asking.
 
Thanks for the detailed response, Pete!

I think I should be able to measure the system impedance pretty easily with Fuzzmeasure, and then compare that to the stereophile review measurements (It's for the 107/2, but at least I should get an idea about whether it is in the ballpark or completely bizarre).

The vertical axis measurements should be doable too, but I have to say that I'm done for the day, so those will have to wait for another time.

The distortion stuff I have no idea how to test for. Is this do-able for a modest home setup, or does that require specialized equipment?

I'm happy with the results so far, and I don't want to get too obsessive, but I do also want to make sure there's no funny business going on. It may have to wait a while, but I'll eventually want to investigate this further.
 
Oh, and did I mention the bass? The original chart I posted had it about flat from 22Hz-100Hz in the nearfield, but I no longer trust that, since I think the curve was being flattened out by excessive volume levels.

I used that RoomEQWizard program that Carl posted a link to earlier in order to get more accurate bass response curves. I haven't figured out how to make a pdf of them yet, but it looks to be -3db at 32Hz, with a very slow roll-off after that, to -8db at 20Hz. This is with the active circuitry in the "Kube" equalizer that comes with these speakers set to Extension=18 / Q=.3 / Contour=0. I set the Q way down because while higher settings give better extension, they also give a big hump at 50Hz. With Q at .3 and extension set to the lowest level, I like the compromise.
 
Also note that the highly complex crossover might have compensated for the tweeter impedance rise at Fs and the mismatch might cause some strange behavior regarding the system input impedance. I believe this system also had conjugate load matching to provide a reasonably flat input impedance.

I would be surprised if this was NOT the case. As mentioned, KEF didn't leave much of anything to chance.

Unfortunately, the conjugate compensation that would make for flat impedance vs. frequency for the original T33, most likely will NOT do so, with another tweeter with different resonant points (and impedance "peaks" at different frequencies than the original)... and the only way to really "set things right" may be to rework the compensation network, to match the new tweeter.

HOWEVER- that said, if the impedance isn't doing harmful things (i.e, big dips that might cause amp problems), then it may or may not be a significant issue, from a functional standpoint... it would require testing with different amps (say, solid state amps and tube amps, due to their different output impedance) to make this determination...

Regards,
Gordon.
 
Oh, and did I mention the bass? The original chart I posted had it about flat from 22Hz-100Hz in the nearfield, but I no longer trust that, since I think the curve was being flattened out by excessive volume levels.

I used that RoomEQWizard program that Carl posted a link to earlier in order to get more accurate bass response curves. I haven't figured out how to make a pdf of them yet, but it looks to be -3db at 32Hz, with a very slow roll-off after that, to -8db at 20Hz. This is with the active circuitry in the "Kube" equalizer that comes with these speakers set to Extension=18 / Q=.3 / Contour=0. I set the Q way down because while higher settings give better extension, they also give a big hump at 50Hz. With Q at .3 and extension set to the lowest level, I like the compromise.

One thought here, if the "donuts" are dead, bass will not respond as predicted. It may go deeper, but will be less "tight", and will definitely have much different slope than intended on the bottom. To clarify, the "donut" is as important to the cone's suspension as the spider and outer surround is. I've learned this refurbing a few pairs of KEF 104/2.

Also an interesting exercise with the tweeter issue would be to try placing some felt around the outside edge of the tweeter since it is sticking out farther from the old housing, just to keep "baffle weirdness" in check.
 
Thanks for the detailed response, Pete!

I think I should be able to measure the system impedance pretty easily with Fuzzmeasure, and then compare that to the stereophile review measurements (It's for the 107/2, but at least I should get an idea about whether it is in the ballpark or completely bizarre).

The vertical axis measurements should be doable too, but I have to say that I'm done for the day, so those will have to wait for another time.

The distortion stuff I have no idea how to test for. Is this do-able for a modest home setup, or does that require specialized equipment?

I'm happy with the results so far, and I don't want to get too obsessive, but I do also want to make sure there's no funny business going on. It may have to wait a while, but I'll eventually want to investigate this further.

Nothing wrong with sitting back and enjoying them. You might put on a noise source, or interstation FM noise and actually check the stand up/sit down performance by ear.

I believe that the most straight forward approach to the 1 kHz issue is to attempt to match the amplitude response, then we know that the tweeter XO is working as designed. This is more of an issue at high SPLs, and perhaps it's not an issue at all. On the other hand people cook tweeters all the time, and that added drive at 1 kHz will contribute to VC heating.

Again, nice work, enjoy them!
 
One thought here, if the "donuts" are dead, bass will not respond as predicted. It may go deeper, but will be less "tight", and will definitely have much different slope than intended on the bottom. To clarify, the "donut" is as important to the cone's suspension as the spider and outer surround is. I've learned this refurbing a few pairs of KEF 104/2.

Also an interesting exercise with the tweeter issue would be to try placing some felt around the outside edge of the tweeter since it is sticking out farther from the old housing, just to keep "baffle weirdness" in check.

Donuts and surrounds have all been replaced with correct foams. It was a pain to do! The vc gap is very tight on these woofers. They donuts, though, were actually very simple to replace. The only tricky thing was to make sure no rotted foam got into the VC gap when removing the old donut.

I did stick some damping in the gap between the tweeters and baffle when I was trying them out, but now everything is flush and good.
 
OK, so I did a few off-axis tests. This was a quick and dirty test, but I was able to notice a fairly deep 6db notch at about 2500Hz (the crossover region) when I got to about 20 degrees above tweeter axis (much higher than I could imagine listening). I also got a much shallower notch at about 400Hz. Below tweeter axis everything stayed pretty good until about 30 or 35 degrees.

The 2500Hz notch is understandable, and is consistent with Stereophile's measurements on the 107/2. Also consistent with stereophile is that the below-axis response is better than the above-axis response.

What I don't understand is the shallow notch at 400Hz. You can also see a bit of this in the horizontal axis measurements I posted above. The woofers cross over at 150, and they are mounted vertically, so it couldn't be woofer beamy-ness. Well, at least it's shallow enough not to be too big of a worry.

Most importantly, though, the 800-2000Hz area seems fine.

My impedance measurements didn't work out well, since I can't seam to get a readable chart from fuzzmeasure. And besides, the constant impedance magic goes on in the external "Kube" processor that hooks up through the Preamp's tape loop, so I have no idea how to measure impedance with the Kube in the equation. It seems to be doing fine with my amplifier right now, so I will just cross my fingers for the time being and hope everything is ok.

Allright, now on to room response measurements with the REW software that Carl posted earlier. . .
 
Donuts and surrounds have all been replaced with correct foams. It was a pain to do! The vc gap is very tight on these woofers. They donuts, though, were actually very simple to replace. The only tricky thing was to make sure no rotted foam got into the VC gap when removing the old donut.

I did stick some damping in the gap between the tweeters and baffle when I was trying them out, but now everything is flush and good.

Very good. Another couple things I've learned from re-doing 104/2:

1. Even if all the cabinet seams look ok from the inside, it's worth the time to run a good bead of wood glue down all the seams, especially the two sealed chambers. If your cabinets are the more coarse variety of chipboard, consider this a requirement!

2. Assuming the 107 mounts the woofers the same as the 104/2, keep in mind that the woofer bolts/studs only hold the edge of the woofer in place, it is the force-canceling rod that pulls the woofers toward each other, and makes them air-tight to the cabinet. Over time, and with the tension, the distance of the woofer baffles compared to the length of the force-canceling rod may have changed. To make SURE the woofers seal, I add a strip of 3/8" wide, 3/16" thick foam weather-stripping ON TOP of the original KEF seal, erring on the side toward the basket, not the edge, if there's a choice. This one thing helps a LOT, because in the 104/2 at least, the studs that hold the woofer edges are shouldered, and the thickness of the original seals is about the same height at those shoulders. Put another way, it may be that in some spots the seals don't quite reach the baffle, no matter how hard you torque the mounting nuts down.
 
Very good. Another couple things I've learned from re-doing 104/2:

1. Even if all the cabinet seams look ok from the inside, it's worth the time to run a good bead of wood glue down all the seams, especially the two sealed chambers. If your cabinets are the more coarse variety of chipboard, consider this a requirement!

2. Assuming the 107 mounts the woofers the same as the 104/2, keep in mind that the woofer bolts/studs only hold the edge of the woofer in place, it is the force-canceling rod that pulls the woofers toward each other, and makes them air-tight to the cabinet. Over time, and with the tension, the distance of the woofer baffles compared to the length of the force-canceling rod may have changed. To make SURE the woofers seal, I add a strip of 3/8" wide, 3/16" thick foam weather-stripping ON TOP of the original KEF seal, erring on the side toward the basket, not the edge, if there's a choice. This one thing helps a LOT, because in the 104/2 at least, the studs that hold the woofer edges are shouldered, and the thickness of the original seals is about the same height at those shoulders. Put another way, it may be that in some spots the seals don't quite reach the baffle, no matter how hard you torque the mounting nuts down.

Thanks for the tips!

I'm not eager to open these up again right now, but next time I'm inside I will follow your advice on these points and make sure everything is solid and tight. I double-checked the woofer seals when I reassembled them, but it sounds like your method is pretty foolproof. (And yes, from your description it sounds like the construction is very similar to the 104/2).
 
Back
Top Bottom