Thursday, June 18, 2009

Filters go bust — time to renew

After a lengthy hiatus — basically I took a break from doing audio and DIY stuff — I've recently (about a week before writing this) finished building a new active filter circuit.

It is soldered on a PCB and replaces an older unit that I'd built up on a veroboard. The veroboard served us well for about 5 years, but finally konked out when the jungle of wire-wrap started breaking and a thick layer of dust made the filters unstable.

Lesson learnt: always remember to put your project in a protective case to keep out the dust and miscellaneous falling objects! Don't just remember it - do it. I put mine on an antistatic sponge with good intentions, but kept procrastinating until it was too late. A consequence of my laziness was that the tweeters were destroyed by loud "clicks" and "pops" towards the end of the filter board's life. This was the ideal opportunity for some refurbishment!

I may have written about the old system before, but just recapping...

The old system:

Accuton C^2 94 midwoofers (18cm),
C^2 23 tweeters (32mm),
Jaycar 12" carbon fibre subwoofer; 4ohm.

15L satellites made from 20mm bubinga (an extremely dense and hard Australian hardwood - I managed to stall a 1200W circular saw while cutting it!!). They needed plenty of pillow stuffing and glued styrofoam to tame the resonances.
A 60L box made with 25mm marine ply and a modest amount of stuffing.

From memory...
3rd order high-pass filters at approx 2.7kHz with a low Q (somewhere between Bessel and Butterworth).
The lowpass had a much higher Q to compensate for the apparent "droop" in the C94's sensitivity around 2kHz to 4kHz and to provide a sharper roll-off so that it could attenuate the speaker's 5kHz resonance more effectively.
The midwoofer also had 4dB of baffle-step compensation below around 700Hz.
The subwoofer had a 7th order filter with a somewhat "hacked" response, vaguely resembling a 130Hz Gaussian filter with 12dB of gradual roll-off before undergoing a secondary cut-off at higher frequencies (according to my Multisim simulations anyway). I later added a 6dB step filter to partly compensate for the subwoofer's roll-off below approx 50Hz.

The new system:
Same subwoofer,
Same midwoofers,
New C44-8 high-midrange speakers. These were originally meant for a different system that I never got around to building. Plus they had the same sized frame. In they went!

I added zobel networks for the midwoofers and tweeters. While they probably don't make *much* difference, theoretically, they should improve the amplifier's damping factor at high frequencies. What usually seems to happen is that the speaker produces air resonances inside the box (to be expected), but that stored energy is then turned into electrical signals by the speaker (it is a transducer, after all). Since the amplifier is a strong voltage source, it produces large currents to maintain the correct voltages at its terminals. Thus, there is the potential for unwanted ringing and IM distortion. The zobel networks are basically snubbers that help to reduce these sorts of oscillations.

The satellites use linear phase filters with 0.05 degree equiripple and a 2.5kHz cut-off for both low pass and high pass. The calculations for the high pass filter take into account the 1-pole high-pass effect of the DC-blocking capacitors on the amplifier's output.

I chose the "linear phase" filter type in an attempt to reduce the slightly harsh sound of the previous design, which I now attribute to ringing due to high Q factors. People often seem to use Linkwitz Riley filters for all occasions but I saw that option as overkill. Given the number of poles, I simply chose a filter slope that produced negligible overshoot in simulations.

Similarly, I saw the importance of preventing audible ringing from the subwoofer filter. Since it operates at a range where people's hearing is not very sensitive, any "out-of-range" ringing around 100Hz~200Hz may actually be more noticeable than the desired signal below 100Hz! After carefully looking at some different options on TINA (a Texas Instruments sim program), I chose the same Linear Phase 0.05deg type but with 7 poles.

The midwoofers now have adjustable notch filters, which are currently tuned to 4.5kHz, approx 9dB' attenuation and Q ~=12. This seems to match the peak in the C94s' datasheet. The notches use a "simulated inductor" approach, followed by a low pass RC filter at 10kHz to reduce their self-noise.

I've already made a couple of important mods — photos and an update should be coming soon!