FabFilter

I can reproduce this easily. I've tried with a 1kHz xover freq, 24 dB/oct filters, Q=1, linear phase max.

SPAN shows the residual spectrum as follows:

• Q: a notch at 1kHz, a maximum at around 2 kHz with a max level around -102 dB or so.
• Q2: a "mountain shaped" spectrum, max at 1kHz with a level around -42 dB (perfectly audible).
• Q2 with Q lowered to 0.707: interestingly things improve and the spectrum shape looks more like the one in Q, notch at 1kHz, max. at around 2kHz with a level around -66 dB or so. Better but still audible.

Maybe there's something about the way the Q value is translated to the filter shape in 1 and 2?

Ok, further investigation reveals some significant differences between Q1 and Q2.

You will notice that in Q1, when you specify a certain cutoff frequency in a 24 dB/oct filter, that corresponds to the -6 dB point, not the -3 dB. It seems that here the filter is implemented as a 4th order Linkwitz-Riley, which is basically two 2nd order Butterworths in series. It's the most commonly used filter used in loudspeaker crossovers because, and this is the important part, it sums to a completely flat amplitude response. If you think about it, it makes sense that you need the xover freq. to be at -6 dB; if it were at -3, the sum would have a +3dB bump around the xover freq.

Now, I guess that for reasons of consistency, in Q2 the cutoff freq is always at -3 dB, whatever the filter slope. Lowering the Q to 0.707 does bring the slope down to -6 dB at the cutoff, but as noted the problem isn't solved completely, perhaps because of rounding errors?

Note that in Pro-Q (both 1 and 2) for high cut and low cut filters, the Q value corresponds to sqrt(2) times de actual, mathematical Q value (the Q of a Butterworth filter is 0.707, not 1, and for a Linkwitz-Riley, it's 0.5, not 0.707), which is convenient in a way but I think it would be nice (for Q2.02 maybe?) to have the option of specifying LR shapes for implementing crossovers. How about a slope drop down list that looks like this:

6 dB/oct
12 dB/oct
12 dB/oct LR (or XO or whatever)
18 dB/oct
24 dB/oct
24 dB/oct LR
...

Where for those the filter is an exact LR shape with the cutoff freq at the -6 dB point?

Please can you specify what platform you are using. PC or Mac and plugin VST or AU?

Thanks

I'm using the 64bit vst on 64bit Windows 7 and 64bit Reaper.

Cabirio, that's a great in-depth analysis there! I wonder if maybe the hi and low cut filters in Q2 were designed with a certain musical practicality in mind over mathematical accuracy? I'll have to compare Q and Q2's "low cut" on my next mix.

Hi guys,

If you'd like to use an EQ tpo create crossover filters, those filters need to have special properties... for example, the filter needs to have a -6dB gain at it's cutoff frequency, so that adding LP and HP at that point will result in a 0dB level. In short: not every EQ filter can be used as a crossover filter!

In Pro-Q v1, we had implemented the 24 dB/oct and 48 dB/oct filters as 'doubled' 12 and 24 dB/oct filters. Due to this, they can 'accidently' be used a crossovers :-)

In Pro-Q 2, we've implemented 'true' higher order filters (at any slope, the gain at the center frequency is -3 dB/oct.

So to use Pro-Q 2's filters as crossover filters... just double them! In other words: you can re-create Pro-Q v1's 24 dB/oct filters by simply using 2 12 dB/oct filters in Pro-Q 2.

So simple! I've just tried it with the same parameters as above using two 12 dB/oct with Q=1 and the residual spectrum is pretty much flat and down at -156 dB or thereabouts, so it actually works much better than Q1 with a single 24 dB/oct filter.

Cheers,