I am looking for a patch that has multiband distortion. e.g. quadrafuzz filter is exactly the type I am looking for. It is not only great for a guitar but I use it to tweak and give variation in techno beats. You can boost the Kick tremendously with this kind of filter.

The QuadraFuzz does have a lovely sound; it would be great to emulate on the Nord. Good for beats, bass, guitar, vocals ....

Presumably:

4 channels, each begins with a bandpass (high frequency = low frequency of next channel, so you'd need 5 knobs) After that, an individual distortion on each channel. I think the Quadrafuzz offers a variety of distortion types, and different curves, though these settings apply accross all channels I think the audio is processed independently. There's also a good degree of compression, and 1 parameter of the dynamics can be set separately for each channel. I think it's the curve too, but maybe it's just a level control.

With 4 channels you can do some kind of stereo spread for a fatter sound too.

Anyone else have any ideas?

Eh, no, to make four channels you need three crossover frequencies, and hence three knobs. Something like the attachement.

The second option means you can create a dirty bass sound without the lower frequencies taking over the compression; and the top filter can be nice and bright whilst cutting hiss. Depends on your source.

I have tried the 4fuzz.phc patch but it is not really a quadrafuzz.

Quote from paia:

This sophisticated fuzz designed by Craig Anderton splits an instrument signal into four specially equalized channels and fuzzes each channel separately for the smoothest, cleanest sustaining sound with minimum harshness. Fuzz outputs are normalized to a single output but are also available at individual output jacks. Panel controls include separate level and resonance controls for each channel, effects loop send and receive jacks and foot switch input. If you're tired of the same old fuzz and want something that will let you realize an entire palette of distinctive sounds ... here it is. link: http://www.paia.com/quadrafz.htm

I have heard it once and it sounds simillar to the vst plugin from qbase.

If you have a quadrafuzz, I suggest you try to make a NM patch that sounds like one. It would be interesting to see what you come up with.

I remember a thread about trying to implement a quadrafuzz with the NM. This somehow implies that we can correctly separate a signal into frequency band using the Modular filters. If I remember well, filters that are accurate for that purpose need to be zero-phased filters. Is that correct? Are the NM filters such filters?

The filters in the NM are of the IIR or Infinite Impulse Response type. These filters use feedback, in fact the basic active element in such a filter is an addition of a certain amount of the current input value with a certain amount of the last output value, where the balance between the amounts defines the cutoff frequency. In the last output value there is still a certain amount present of the output value before the last one, and decreasing amounts of the output values before that one.

What happens in a IIR filter is not unlike an object hitting a rubber block with a certain speed, after the object hits the block the resistance of the block will increase as the object gets deeper in the block and the speed of the object will decrease until it stops. The rubber block is like a resonant filter as much of the kinetic energy of the moving object is temporarily stored and then released to the object to make the object bounce back. This can be compared to the resonance effect of a IIR filter. For a short moment the object is in contact with the block before it is bounced back, this slight time delay before hitting the block and be bounced back is in a way comparable to the phase response of a IIR filter.

In a lowpass IIR filter low frequencies get less phaseshift as high frequencies. The phaseshift can be used to increase the resonance of the filter by applying feedback. A single filter section has a 90 degree phaseshift at the cutoff frequency, other frequencies have a different phase shift. If four section are in series the phaseshift for the cutuff frequency becomes 360 degrees and if feedback is applied it will start to resonate. With two sections resonance is also possible by inverting the feedback signal, inverting a signal means a 180 degree phase shift, well not really a shift but a phase inversion, but that the filter doesn't know, it just adds up 90+90+180.

So, rule of thumb is IIR filters always have phaseshifts, the steeper the cutoff the more the total phaseshift will be. But every frequency in the signal has a different phaseshift. If filtering is applied this phaseshift cannot be heard. Eg imagine a sawtooth wave of 220 Hz is filtered with a 24 dB LP filter at 1000 Hz. The ear does not hear the actual phaseshift of say the third harmonic. Or that the fourth harmonic has slightly more phaseshift compared to the third harmonic. It is only when the output of the filter is added to or subtracted from its input, or the output of another filter set to a diffent cutoff frequency, that frequency components start to boost or cancel out each other and the sound changes. Also when a nonlinear waveshaping technique like distortion is used, the different phaseshifts of the frequency components in a filtered signal may alter the sound depending on the type of distortion.

There are filters that do not have this phaseshifts, FIR filters or Finite Impulse Response filters. These filters work with a block of memory where the last couple of hundred samples are stored. There is another table withthe same length, holding numbers that represent in some mathematical way the behaviour of the filter. In fact this data represents the impulse response, compare it to a description of how the rubber block will react when hit by an object. The actual filtering is done by multiplying each sample in the input sample block by a corresponding parameter in the impuls response table and adding all multiplication results together. A filter with zero phase response for each frequency component can be made by calculating with a 'symmetrical' impulse response parameter table. However this only means that there is no phaseshift between the frequency components after the signal is filtered, but the output of the filter has a time delay equal to the table length of the input memory block. So no phase shift but latency. If on a 96kHz samplerate system like the NM a LP FIR filter is implemented that should go as low as +/- 12Hz one needs a block of memory of 96000/12 = about 1024 locations for the input and the same 1024 locations for the parameters. For a zero-phase filter one needs 2048 multiplications and additions, a multiplication and addition to a previous value is luckily a single instruction in a DSP. But it is still about 2000 DSP instructions for every input sample. This is at least about 250% patchload for such a filter. And for such a zero-phase filter the latency or total input->output delay is about 100 msec!!!

If a manufacturer claims he uses 'zero-phased filters' of the FIR kind in a _realtime_ multiband distorter and there is no input-output delay on the realtime signal, then we have a very special situation that will surely change our world completely and would get immediate attention from every government agency in the world. Most likely the manufacturer would be immediately abducted by the CIA and locked away from the world. The reason being that this manufacturer would have invented a technique to reliably predict the future! And although he would be able to predict only the amount of time that represents his input memory table, it would still be predicting the future. And by adding more RAM he would be able to predict more of the future. Imagine what it would mean if such a technique would fall in 'the wrong hands'.

So, there is a trade off between DSP power used, latency and phaseshift. For realtime applications the choise is for a IIR filter, but for non-realtime applications latency and calculation time is no issue, so only there a FIR filter is the obvious choise.

To create a 4-band distorter one needs a LP filter for the lowest band, two BP filters for the middle bands and a HP filter for the highest band. Preferably the bandfilters should be real bandfilters with a low cutoff point, a high cutoff point and the same slope as the LP and HP filters. One might think that this would call for a bunch of LP and HP filters but such a filter can actually be made with only three LP filters. The basic idea is that if there are two LP filters, one set to 500 Hz and the other to 1000 Hz, a 500 Hz to 1000 Hz bandfilter can be made by subtracting the output of the 500 Hz filter from the output of the 1000 Hz filter. By additionally subtracting the output of the 1000 Hz filter from the input signal a 1000 Hz HP signal is created. So to create a filterbank with n bands one only needs n-1 LP filters. Adding all band outputs together again will reconstruct the exact original signal. The only thing of importance is that all filters should have the same resonance setting and the passband gain should be unity to get the HP output right.

The problem with distortion and a fuzz based on a feedback clipper circuit is that these amplify the signal which is a problem when a high band must distorted and there is a strong bass in the signal. As the original amplitude of the bass is in regular cases much higher than the amplitude of the original high signals the high band tends to reintroduce an unwanted bass signal in the distortion circuit. My solution is to use an extra HP signal before the BP and HP bands are fed into their distortion circuits. By crossfading between the bandsignal before the extra HP filter and the output of the distortion the distortionlevel for than band can be conveniently set. The two last Patches are examples of a quadrafuzz, one with 12 dB bandfilters and the other with 24 dB bandfilters. On the 12 dB version the resonance can be controlled as well, if used this creates a different sort of timbral control. On the NM 24 dB filters thats is not possible as the amplification is changed in some strange nonlinear way when the resonance is increased. The problem here is especially the HP band, which has also to do with the fact that straightforward digital IIR filters behave not really well when the cutoff is in the highest regions and the LP output is subtracted from the input signal.