Dave's Speaker Pages

KEF (Celestion) F15 2-Way Mid-Woofer(!) Modifications/Tweaks

February 2008

Why the exclamation point in the description? Well, I have a hard time calling a 5.25" driver a mid-woofer. It's more like a midrange. The high cutoff of this 2-way system (maybe 60Hz) is barely in what I'd call mid-woofer territory. This thing needs a sub-woofer unless it's stuck on a bookshelf. OK, now that I've got that off of my chest, on to tweaking. I had not planned to do so much testing with this driver. Some discussions going on at the diyaudio board prompted much of it. I really had only planned to use the surround tweak, but some comments and measurements related to added mass and Mortite piqued my interest. The measurements I've made provide, I think, additional support for there being only an added mass and some damping involved. There has yet to be any concrete and valid evidence of any other mechanism involved when any kind of mass is added.

The driver is an ordinary one. The whole system is inexpensive (read cheap), but I liked the style of the box. I figured that I'd use the boxes if the drivers weren't much good. The tweeter is not much, that may be the weakest part of the system. The midwoofer has issues as you'll see, but they can be improved a fair amount, more than I expected.

The Driver

The diaphragm is a fairly typical polypropylene with what looks like a phase plug. The latter is actually a raised dust cap that is probably used to give it better top end extension. The surround is also typical, being a raised, half-roll type. It's made of some kind of rubber. One thing you'll notice is some felt on the baffle. These were placed as soon as I bought these. They remained in-place and untouched throughout the testing.

The frequency response of the raw driver shows some serious issues, not surprising for the cost of the whole system. What you'd likely notice first is the most serious one, the dip at 1.7KHz. This is right in the middle of the area where our ears are the most sensitive. It's also one that isn't easily controlled without complicating the crossover. The other issues are the resonances higher up. These can be pushed into the stop-band to some degree, but not enough to make them benign. They also will affect the distortion profile, especially that part related to the non-linear motor distortion, regardless of the crossover. The best approach is to try to control the frequency response through cone modification (tweaks).

Measurement Conditions

The entire time that tests were conducted, nothing changed with the measurement system at all. I followed my standard test procedure, with the exception that the driver was left mounted on the baffle of the original system box rather than being mounted on my 2m x 2m baffle. I did this because I had no insert cut out for my large baffle and because I had started this only to improve the driver a bit, not test extensively. The measurements were made with the mic on the woofer axis @1m. The actual power applied remained constant using LAUD, the mic was never touched nor was the box moved. All applications of tweak material were applied with the driver and box in place. The window start and stop time markers for the FFT of the impulse were never changed. A "square" window was used for all FFTs. The square window provides the best detail in the low end with the possible introduction of spurious high-frequency anomolies that can occur due to the MLS technique and the window transition. However, for a midrange this is not as critical as it would be for a tweeter. Repeated tests were done that showed very good consistency in the high end of the FFT. No smoothing was used. The higher the Q of any resonance, the more "hidden" it will be either a smoothing or a tapered window were used. All measurements were made using a train of eight impulses that are averaged by LAUD to improve the signal-to-noise ratio.

One last point. For every measurement I make for which I keep the frequency response, I save the impulse response of both the mic and the probe inputs. This allows me to re-load any measurement made over time and re-generate an FFT or CSD for any setting I might want, even years later. I can always re-generate any output made previously for whatever reason or, for example, generate a CSD if I had not done so originally.

Determining the Source of the Problems

This was easy for the first anomolie at 1.7KHz. I needed only to see the frequence response curve knowing the rough diameter of the diaphragm. This is the very common cone/surround mismatched impedance issue. The cost of adding the damping compound probably makes it too expensive for the intended market price point. The tweak I use is simply using the modification made on more expensive units that have this issue. It's been used for decades and is a simple answer to this particular problem.

The other problem area is the set of peaks at 4.6K, 6.1K and 7.5K. It's interesting to note that the difference from one to the next is close to the same delta, about 1.5KHz. Having no access to expensive test equipment such as a laser interferometer, I can only guess as to the specific resonant mode and where it's occuring on the driver. My guess is that it is associated with the "phase plug" style dust cap (here on referred to only as a dust cap). Further mods shown later shed more light on it.

There are also many very tiny bumps in the frequency response between 2K and 4K. These will be shown to be directly related to the surround termination as well. What's not clear is how much is strictly the main cone and how much is the dust cap reacting to resonances of the main cone, since any main cone resonances will become part of what energizes the dust cap. Since the cone and dust cap are inherently closely coupled, there's no way to know precisely what resonances are strictly related to one or the other. The same holds for any similar driver such as a full-range driver with a whizzer cone.

New Untreated Driver

Tweaking Part I

My intent when I started working with this driver was to apply the tweak that I first used and documented on my page on the Insignia. This tweak is, as stated earlier, intended to do what manufacturers have done for decades when trying to get the optimum response from a particular driver. The cone/surround junction is critical to good driver response. It's a complex impedance and therefore frequency dependent. This is manifested in much of the non-linear nature of the frequency response seen above (not to be confused with non-linear distortion), particularly in the dip at 1.7KHz. Applying multiple coats of "The Wet Look" to the surround at the intersection of the roll and flat section on the cone side (shown in later photos) does just what is intended. The results of this tweak are shown below. The comparison is of the raw untreated driver to first one, then three applications of The Wet Look.

New Untreated Driver vs. Applications of The Wet Look

New Untreated Driver vs. Three Applications Normalized

The improvement is significant. The dip at 1.7KHz, while not fully controlled, is reduced close to 3db to the point that its influence with a crossover in place will be minimal. The side benefits of the application are that the upper resonances are partly controlled while the tiny bumps noted between 2K and 4K are essentially eliminated altogether. Curiously, a resonance seems to be introduced at about 9.5Kz, though it's small and will be insignificant with a crossover in place. No further applications were made to the surround, though it may be possible to get more improvment and it will certainly be at the point of diminishing returns.

Rather than duplicate the description and photos of the application technique, please see my page on the Insignia woofer tweak on how to apply it.

Tweaking Part II

This is where it became interesting again. I knew what to expect of the surround tweak, but had never tried to damp the cone directly. I don't recommend applying any irreversible mods to a driver cone, since each one is so different. However, I did experiment with some distributed mass, some that proved beneficial and some detrimental. The final and best configuration surprised me. Before that, I'll show the majority of the configurations tested and the measured response of each.

There are too many tests and photos to go into detail on them all. Instead, a table follows with the photo accompanied by the measurements to the side. Some photos encompass several configurations, but the description in each measurement should (I hope) make it clear what was used for it.

A large number of dots configurations outlining geometric shapes were tested. I tried a square, a triangle, a pentagon and more. My thinking was that odd-numbered damping points would be the most beneficial. After some disappointment I recalled the damping "ears" used by Accuton in their drivers. I had always been curious about the fact that they used only two, opposite each other. This prompted me to try just two dots. The results were so much better that I stayed with that.

Cone Mod Photos and Graphs
The graph above is for the configurations shown to the left and above left.
No picture for these graphs. The sequence was first placing 4 dots in a square pattern on the cone close to the surround. Each subsequent test was made by doubling the dots and spacing them equally, halfway between the adjacent pair. The sequence was 4, 8, 16 and finally 32 dots. The last is shown in the picture below.

These graphs are for the configurations that start with the 32 dots shown in the picture to the left. The rest are in sequence below.

CSD Comparison

CSD plots are pretty much useless for purposes of design. The frequency response is sufficient. No software uses the CSD, it all uses the FR. However, the CSD does show the relative levels of resonances. The caveat is that if any two CSDs are compared, they can only be directly compared for those that are taken by the same system with the same software, the same software settings (window type, start and stop time markers, etc.) and the same physical setup (mic and driver positions). I've included some CSDs for those interested in them.

CSD Plots

New, Raw, Untreated Driver

Three Applications of the Surround Treatment

Surround Treatment + 32 Dots Distributed Near the Surround Junction

Surround Treatment, Former Ring + 3 Lines (Scan-Speak Style) + 3 Dots

Surround Treatment + 2 Dots on Cone and Dust Cap

Surround Treatment + 3 Angular Positions of 2 Dots on Cone and 2 Dots on Dust Cap


As it was for the Insignia, the surround tweak is one of the most useful going if the driver has an impedance mismatch as this one exhibited. It doesn't require any elaborate patterns and doesn't take much time. The three coats that I applied took, from start to finish, about one hour. Successive coats can be applied soon after the previous one has started to dry. The benefit cannot be had any other way and has little in the way of negative consequences.

Adding the dots is a bit more problematic in that each driver is going to react to any and all added mass differently, much more so than it will for the surround tweak. The really good thing about it is that it is totally reversible. Painted lines or other distributed mass applications are not likely to be nearly as effective for many if not all drivers from what can be seen here. No driver should be modified with added mass unless it can be measured afterwards to document the changes and ensure that it's not just shifting resonances. At worst, resonances may be worsened or new ones introduced.

The best result is arguably the one shown below. It's the surround application with three sets of different dot pairs on the cone and a pair on the dust cap. Each pair was placed at different spacings from former to surround. I chose them because I hoped to attack the resonances at different points for each. This seems to have worked and is how they will remain. With a new crossover in place I don't expect there to be any loss in system sensitivity other than what I may change for my own purposes.

The Chosen Configuration


My initial goals were to improve the raw response of the midwoofer with the application of the surround tweak. It worked a bit better than expected. The surprise was the improvement with the addition of a few pieces of mass/damping material strategically placed.