I once read that patching two 12db Filters in a row results in a 24 db filter; adding even more modules (like possible in the NM) should result in a VERY exactly cut-off frequency. First question: Is this true in theory? Second question: Is it true with a NM as well? My impression is that the curve of damping frequencies starts far before the indicated point, so that the very edge of the cut-off point is lowered with each added filter. Oh my english! I can't say it better. A beautiful diagram instead:

Fig.A: "How ulrich wants it":

Fig.B: "As Is":

So, no chance to build a brick wall?

The theory behind filters is actually so complex that it's better to judge a filter by it's "sound". But let's shed some more light to the subject. A standard analog-type lowpass filter as we find them in the NM is based on selectively shifting the phase of the signal in a frequencydependant way. If the phaseshift is 180 degrees it cancels out the original signal and that frequency isn't heard anymore. In the case of a lowpass filter the shift increases from 0 degrees to 180 degrees if the frequency rises. There exists no "ideal" filter that suddenly at a given frequency shifts from 0 to 180 degrees, but a good filter tries to make that range as small as possible. What we call the cornerfrequency or cutoff frequency is the point where the signal is attenuated -3 dB. Below that frequency the passband should have no or a very small "ripple", above that frequency it should drop as fast as possible. There are some types of filtercharacteristics like Bessel, Butterworth and Chebishev. These are preprogrammed in the filteralgoritms and control things like the flatness of the passband, the corner of the cutoff frequency and the curve of the cutoff-slope. I suppose that these are responsible for the slight difference in sound between synthesizer filters of different makes. The bad news is that a sharper cutoff slope gives a bigger ripple in the passband, more irregular phaseshifts, etc.

The brickwall filter you want would have no ripple, a sharp corner and go down straight into oblivion at once. Such a filter can be made digitally in the form of an analysis/resynthesis filter or a transversal filter. The first type analyzes the frequencycontent of the signal by means of a so called Fourier-transform. This gives two "graphs" representing the loudness and phase of each frequencycomponent in the signal. By deleting some of these components in the graphs and using this information to do a revers Fourier-transform a brickwallfilter can be made. But for negligable loss in quality you need a row of NM's to do the required calculations in realtime! I heard a Capybara system with 28! DSP's can do that job quite icely.

For transversal filters it's the same story. But besides lot's and lot's of DSP-power they also need a lot of memory to temporarily store the processed material and the so called impulse-response". Both types also have a definite timedelay of maybe several msec, but if properly implemented no phases are shifted. So we might only see this kind of filter in the NM3 or NM4....... If I remember well Cool Edit Pro has such filters to experiment with. But not realtime of course.

There are some tricks, of course. If you connect the lowpass and the highpass output of the 12 dB filter to a crossfade module, we get a new type of filter with controllable curve. In the midposition the filter is actually a notch or bandreject filter. This notch becomes less noticable if the control is moved toward the lowpass or the highpass input, but it is still there. When cascading a few 12 dB filters with this outputmodification we can adjust these notchpositions in a way that they help to make the slope decrease faster just after the cutoff frequency. Raising the resonance just a bit helps in sharpening the corner. For best results a calculation should be made, but that's a terrible complex one as all filters in the cascade will have different settings, compensating for each others deficiencies, so better forget about that.

The real question is: why a brickwall filter. It hardly pays off to do all this complex stuff for a sweeping filter. The audible effect would be negligable.

But for filtering audiomaterial from the inputs it might be nice to experiment with it. So what you can do is cascade a D-type filter and a 12 dB filter and connect the lowpass and highpass outputs of the 12 dB filter by means of a crossfade module. Raise the resonance of the 24 dB filter to a setting of say 24, this increases the corner but also gives a peak. Now adjust the frequency and LP and HP amount of the 12 dB filter that the notch both corrects the peak of the 24 dB filter a bit and makes the slope fall down a bit faster just after the cutoff frequency.

So a common sense approach would be to mimic the effect of a brickwall filter. After a 24 dB filter we might find out that a strong frequency component in the suppressed band is still audible. By using the described notch filter to suppress that component, LP:HP something like 2:1 to start with, frequency adjustment by ear, the filter will appear to be much sharper.

So:

1. the soft slope of the 12 dB compensates for the increased resonance of the 24 dB, giving a sharper corner but reducing the resonance
2. the notch of the 12 dB suppresses the strongest unwanted strong frequencycomponent in the rejectband of the 24 dB
3. the frequencysetting and the X-fade setting of the 12 dB influence each other, the frequency shouldn't be set to high or it won't compensate the higher resonance, the X-fade controls the position of the notch (over a limited range)

In the attached example you can experiment with the settings. A strong high harmonic is added to a sawtooth. Playing with the X-fade control and the frequency control on the 12 dB filter will hopefully makes things clearer.

A diagram could be:

Another approach is using a combfilter, useful to remove one frequency and all of it's harmonics from audiomaterial. It's essentially a delayline with a negative feedforward. Very popular with spies to remove backgroundsounds from longrange microphone recordings. In the examplepatch it is used to completely remove one of the notes from the chord. The phaseresponse of the combfilter however is so poor that the remaining material sounds distictly different. Not really suited to remove the vocals from a song, but when swept it gives a nice phasing effect.