Having observed what I thought might be the typical diffraction ripples in the measured amplitude responses of these speakers, I figured it might be nice to round-over the front baffle side edges in an attempt to ameliorate the problem. My woodworking skills are very basic and even if I could have when originally building the speakers, routing the edges to any useful radius would surely have weakened the structure, so I figured now that it’s too late anyway “why not just bolt on some curved bits?” The subjective improvements were quite profound.
This is about the front vertical corner edges of the cabinets which are a cause for concern. I’ve seen a trend in commercial speakers of late toward curved sidewalls tapering to a narrow rear panel or a rounded back panel, but where sharp corners remain at the front, I see no acoustic purpose in such designs.
The phenomenon of the baffle step response insofar as it relates to these speakers presented as a 6dB rising response from around 200Hz up to around 1kHz. On top of that, there’s the typical 2dB blip at around 1kHz and random ripples above that.
Here is a simulation of the unfiltered midrange driver response on the baffle – as if reproducing a full-range sweep:
Bear in mind, that was just a simulation of the rectangular baffle with the midrange driver in it’s actual offset position. It excludes the crossover and tweeter. It illustrates how the theory would have it.
Here’s an illustration (that I pinched and adapted) showing how refraction around sharp baffle corners causes interference patterns from a tweeter. This is just at one fixed frequency (going by the wavelength c.f. my baffle width – probably around 11kHz):
For a sweep, the result is better illustrated in the amplitude response.
These are in-room near-field on- and off-axis amplitude responses measured some time ago before attaching any extensions:
That includes the crossover and tweeter. Clearly the 15° off-axis response is the best/flattest and the speakers are set up to take advantage of this.
I deliberately applied smoothing to show that the 1kHz blip seems quite consistent. The larger consistent ripples around 500Hz are simply the near-field room response – not speaker resonances.
It’s the blip and ripples from the midrange driver as well as ripples from the tweeter’s refraction around the sharp edges that the rounded extensions aimed to minimise. These are highlighted by the red box.
As can be seen, the 6dB rise had already been pretty much compensated for (electronically). The blip could almost be addressed by applying a notch filter, but being smack in the middle of the midband I didn’t want to go there. There’s already enough going on electronically, so something physical and simple (and which looked cool) seemed like a fun idea.
Fabricating the extensions
These things could probably have been made of timber, but I liked the idea of powder coated steel and an old buddy has a small metalwork factory unit and had an ideal 600mm off-cut of 112mm OD structural tube, so I drew something up and he made two sets of these out of it:
It was some job cutting the pipe lengthwise into four segments with an angle grinder, but a lot cheaper than the set-up costs for laser-cutting. The spot welds were just to hold the braces in place before tidying up and linishing:
Bolting them on
This is an aside really. I got some of these threaded inserts for M6 bolts:
They’re a bit of a PITA really. There are Allan key alternatives, but I couldn’t find any without a flange that would sit proud of the wood surface. These had to be slightly recessed. A flat blade screwdriver of the exact size was required to prevent gouging and slips. I didn’t have one, so I tried this idea:
It’s just a bolt with a reversed Nylok nut on it and it worked well.
The holes in the cabinets were carefully drilled with a collar around the drill bit to ensure I didn’t drill too deeply:
Anyway they’re staying on:
The radius is about 56mm. It would have been “interesting” routing that into the timber.
The tops are chopped at 30° and the bottoms at 45° for aesthetic reasons only.
They are mounted on a thin neoprene pad and the mounting bolts have rubber O-rings so that the extensions can move very slightly in a dampened way. They are “dead” when tapped.
Comparative near-field responses
Here is how one speaker measured about 1.5 metres in front of the baffle and slightly off-axis:
I suspect that further off-axis, the difference would be even more obvious.
Blue trace = extensions attached. Red trace = no extensions.
The microphone was not moved between sweeps.
IR window is 3ms to remove the first reflection (off something about 1m away – i.e. the floor).
Well it looks surprising, but kind of unsurprising at the same time. It’s unsurprising that the 1kHz area is flattened. What is surprising is that the flattened response is a result of an overall increase (by about 2dB) at either side. Also the slight apparent rising response across the midrange has flattened by increasing the lower end of it. The woofers (which cross at 150Hz) were not activated for these measurements. In other words the wings raise the general response at either side of 1kHz peakiness so that the volume can be turned down by 2dB for a reduction in overall distortion. Indeed for the tweeter’s entire range, electrical input can be reduced by 2dB for the same acoustic output level.
The speakers are toed about 15° from the line of sight from the favourite seat. The microphone was not quite on that line (perhaps only 5° off axis – my mistake) and at tweeter height (which is at ear level).
I sat and listened to various kinds of music for several hours before making these measurements. My impression was of a very much improved centre image and overall soundstage. I do not know how the flattened response might explain that however.
Now just in case the viewer thinks my speakers have a god-awful response on the basis of the ripply looking graph above, here is one shown with a more conventional scale. Here with extensions attached and with a wider IR window:
Cursor not shown there, but it was at 1kHz. I.e. the figures in the legend apply for 1kHz only. Again – near field with LR4 150Hz high pass and woofers switched off. At 80dB playback, all distortions (ignoring THD which is just a sum of all the others) are at least 8dB below the noise floor of the room which is around 33dB.
I never really paid too much attention to that obvious rise in the tweeter’s response above about 13kHz because I cannot hear above about 12kHz anyway. A simulation of a simple first order low pass filter to flatten it resulted in too much phase shift down around the mid/tweeter crossover frequency so it shall remain – inaudible and all.
Addendum: Actually it could just be that the microphome needs recalibrating. It’s getting on a bit.
Here (taken from the listening seat prior to attaching the wings) to show bass response with woofers and coffee table subs engaged (two front speakers driven):
The extensions flatten out the peakiness centred around 1kHz by increasing the amplitude response at either side. Subjectively, with the extensions attached to both speakers in a stereo pair they improve the illusion of a great centre image.
Addendum (15 Dec 2016)
I may have an explanation for the perceived improvement in the soundstage – reduced lower midrange distortion.
Here is the distortion graph for the above near-field measurement without the wings attached:
As can be seen, the cursor is at 300Hz and THD is 0.268% there (predominantly 3rd harmonic).
With the wings attached and without moving the microphone or making any other changes there is an obvious general reduction in the lower midrange distortion:
- Interesting observation: The above is an astonishing performance from a system exploiting no DSP whatsoever. All crossovers, baffle step compensation and driver alignment is performed by analogue circuitry! Indeed this is vastly superior to anything I have seen on the ‘net from the likes of DEQX, Acourate, Rephase and JRiver for example.
Again – near field with LR4 150Hz high pass and woofers switched off. These measurements were repeated several times to make sure the traces were consistent, which they were.
At 300Hz, the THD has reduced to 0.18% (mainly 3rd harmonic reduction) despite the fact that the wings increase the SPL by 2dB around that frequency! This is a massive improvement of some 33% (discounting the extra volume).
I did not choose 300Hz for any particular reason other than that the trend is clearly in the lower midrange and not the upper midrange or tweeter bands.
I have no explanation for the 3rd harmonic vanishing like that. Some investigation is wanted.
Then again, the improved image might be due to a “softening” or de-emphasis of early reflections off the previously sharp vertical edges. I really don’t know.
Addendum (27 May 2017)
The 2dB level boost at either side of the 1kHz peak is equivalent to about a 1.25 voltage ratio or 1.6 times the energy size. No doubt this makes me turn down the volume by 2dB to approx. 60% of the original energy input or approx. 80% of voltage. I’m thinking that this must reduce overall distortion to explain the dramatic overall improvement to the system that these extensions have provided.
Addendum 17 Dec 2017
I made some new Floyd Toole-inspired polar measurements at his recommended 2m radius, 1/24 octave smoothing and IR-windowed (to 7.9ms) to remove unwanted room reflections and here it is:
From top to bottom that’s on-axis, 15°, 30°, 45° and 60°.
Not related to the baffle extensions, but interesting nonetheless is the extremely well behaved directivity with on-axis flatness to ±1.2dB from the LF limit of the applied gating to 20kHz (including that near-ultrasonic resonance from the ScanSpeak tweeter). The off-axis performance is even better than the Genelec 8030A studio monitors and completely devoid of their off-axis mid-frequency suck-out caused by woofer beaming. → The high-frequency off-axis performance is just as good and without waveguides! And no audible resonances. The 15kHz blip is easily removed with a DSP notch filter applied in the playback software, but is inaudible anyway. I think that even Toole might be impressed with that! Where’s the crossover point again? 😀