multiband splitter and combiner

wondering how easy it would be to make a box that splits the stereo input signal into 3 separate bands. Lows, mids, and highs and them combines them back into stereo before going into the mastering rig.  A 3 way crossover can do that. But what about summing those signals back into a stereo signal.  A passive summing device(will not use summing mixer as it is redundant) with volume adjustment, Then a make up gain amp?



any thoughts or ideas?

Summing is easy; passive or active doesn't really make a difference with only three inputs.
Beware of the fact that the resulting signal will not be minimum-phase, except if you use 1st-order or complementary x-overs (6dB/octave).

Forgive my ignoraNCE, but isnt linkwitz-riley phase coherent if recombined . ..  I thought that was the point  . .


  Once upon a time I have had fun with the demo of Roger Nichols Digital Spizer plugin for this. Shame it seems to be mothballed . . . .


 

abbey road d enfer said:

Summing is easy; passive or active doesn't really make a difference with only three inputs.
Beware of the fact that the resulting signal will not be minimum-phase, except if you use 1st-order or complementary x-overs (6dB/octave).

Click to expand...

cool thanks for the tip.  off to the drawing board.

strangeandbouncy said:

Forgive my ignoraNCE, but isnt linkwitz-riley phase coherent if recombined . ..  I thought that was the point  . .

Click to expand...

4th-order and 8th-order LR have phase curves that provide unity summing in the crossover region, but the overall transfer-function can be mathematically reduced to an all-pass filter, i.e. it exhibits phase variation and no amplitude variation. The matter of audibility of phase variation is debatable, but it is easy to show that the combination of phase-shift and deficiencies or limitations in the signal path can have measurable and predictible effects.
Just think that the crest factor of a pulse (read transient) will be changed because the temporal location of harmonics is shifted. Phase-shift changes the crest level of a signal, so has effects on THD, IM,...
That's the reason why all multi-band limiters have an output limiter on top.

There is a frequency dividing topology that doesn't exhibit the phase funnies, namely subtract a HPF from unity, to get the LPF response. While this will recombine to one, more cleanly, you sacrifice filter slope in one of the two passbands.


Google "derived" or "subtractive" crossovers.  image from http://sound.westhost.com/articles/derived-xovers.htm

JR

Thanks for pointing that one out JR - cool idea! Off to do some reading.

Thanks J.R. that is interesting. Now to read and expand my mind.

Here's a schematic of an example:

http://users.telenet.be/Rogy/IEMFinalizer/Multiband%20comp/Multiband%20XO%20schem.pdf

You'll find some sims of the resulting curves here:

http://users.telenet.be/Rogy/IEMFinalizer/Multiband%20comp/

Regards,

Rogy

It seems we had another thread discussing this a while back. Another topology that might be useful for dividing the bandpasses for smoother recombining, might be the common two-pole state variable (based around opamp integrators).  Once again the LP, BP, and HP outputs should combine nicely, while the slopes will be only one pole between bandpass outputs.


http://www.analog.com/library/analogdialogue/archives/41-10/phase_relations.html

JR

Didn't Manley make a box that did that?  I could swear I remember seeing one,  just can't find it now.....

Got it,  it was Drawmer:

http://www.drawmer.com/products/multi-band-processing/three-sum-multi-band-interface.php

JohnRoberts said:

It seems we had another thread discussing this a while back. Another topology that might be useful for dividing the bandpasses for smoother recombining, might be the common two-pole state variable (based around opamp integrators).  Once again the LP, BP, and HP outputs should combine nicely, while the slopes will be only one pole between bandpass outputs.
JR

Click to expand...

In fact no, they don't combine nicely. As it is, there is a 3.5dB hole in the summed response, which can be corrected by applying some gain to the MID band. Another possibility is to change the Q of the filter, halving the resistor from positive input of 1st opamp to gnd. It changes the response of the HP and LP filters with a 1.2dB overshoot.
But even then, the end result would not be satisfactory. The MID output is fairly wide, with 6dB/octave asymptotes. Under significant gain reduction of the MID band, the reconstructed response exhibits just a quite narrow notch, which is certainly not what's expected.  When the HI or LO bands are compressing, the response becomes almost unpredictable.
Check the attached graph. Cyan is the recomnined signal under no GR. Green is with 6dB GR on MID. Red is 6dB GR on HI.

Upload folder is full; I'll post the image later.

abbey road d enfer said:

JohnRoberts said:

It seems we had another thread discussing this a while back. Another topology that might be useful for dividing the bandpasses for smoother recombining, might be the common two-pole state variable (based around opamp integrators).  Once again the LP, BP, and HP outputs should combine nicely, while the slopes will be only one pole between bandpass outputs.
JR

Click to expand...

In fact no, they don't combine nicely. As it is, there is a 3.5dB hole in the summed response, which can be corrected by applying some gain to the MID band. Another possibility is to change the Q of the filter, halving the resistor from positive input of 1st opamp to gnd. It changes the response of the HP and LP filters with a 1.2dB overshoot.
But even then, the end result would not be satisfactory. The MID output is fairly wide, with 6dB/octave asymptotes. Under significant gain reduction of the MID band, the reconstructed response exhibits just a quite narrow notch, which is certainly not what's expected.  When the HI or LO bands are compressing, the response becomes almost unpredictable.
Check the attached graph. Cyan is the recomnined signal under no GR. Green is with 6dB GR on MID. Red is 6dB GR on HI.

Click to expand...

I have used state variables for a number of different applications (parametric, crossovers, etc), but never this. It seems the pole spacing for the BP skirts is arbitrary. Are you summing all three outputs?

You can force the combination to sum to one, if you ground the + input on the left most state variable opamp, and add a unity gain inverter stage between the BP integrator and the first opamp - input.

Then by definition the HP,BP,and LP sections will sum to unity, in the first stage opamp. This will also bring the BP output into the same polarity as the HP and LP.

JR

JohnRoberts said:

abbey road d enfer said:

JohnRoberts said:

It seems we had another thread discussing this a while back. Another topology that might be useful for dividing the bandpasses for smoother recombining, might be the common two-pole state variable (based around opamp integrators).  Once again the LP, BP, and HP outputs should combine nicely, while the slopes will be only one pole between bandpass outputs.
JR

Click to expand...

In fact no, they don't combine nicely. As it is, there is a 3.5dB hole in the summed response, which can be corrected by applying some gain to the MID band. Another possibility is to change the Q of the filter, halving the resistor from positive input of 1st opamp to gnd. It changes the response of the HP and LP filters with a 1.2dB overshoot.
But even then, the end result would not be satisfactory. The MID output is fairly wide, with 6dB/octave asymptotes. Under significant gain reduction of the MID band, the reconstructed response exhibits just a quite narrow notch, which is certainly not what's expected.  When the HI or LO bands are compressing, the response becomes almost unpredictable.
Check the attached graph. Cyan is the recomnined signal under no GR. Green is with 6dB GR on MID. Red is 6dB GR on HI.

Click to expand...

I have used state variables for a number of different applications (parametric, crossovers, etc), but never this. It seems the pole spacing for the BP skirts is arbitrary. Are you summing all three outputs?

You can force the combination to sum to one, if you ground the + input on the left most state variable opamp, and add a unity gain inverter stage between the BP integrator and the first opamp - input.

Then by definition the HP,BP,and LP sections will sum to unity, in the first stage opamp.

Click to expand...

If you do that, you get the response I was talking about earlier, with 1.2dB hump for LP and HP and 0dB at center for the BP.

This will also bring the BP output into the same polarity as the HP and LP.

Click to expand...

BP output is in quadrature with the LP and HP. Both polarities of BP produce the same amplitude output (phase is different) when recombined.

There are two issues there:
Unity summing, which, as we've just seen, can easily be achieved by tweaking some components or the topology.
Response under gain-reduction, which shows large interaction, because the HP and LP corner frequencies are the same (and same as the center freqiuency of the BP). The BP output is relatively narrow according to mathematical definition - about one octave, depending on the chosen allignment, but it spreads audibly over a spectrum of about 5 octaves. In contrast, the effect of applying GR to the BP creates a narrow notch that is almost insignificant.

abbey road d enfer said:

JohnRoberts said:

You can force the combination to sum to one, if you ground the + input on the left most state variable opamp, and add a unity gain inverter stage between the BP integrator and the first opamp - input.

Then by definition the HP,BP,and LP sections will sum to unity, in the first stage opamp.

Click to expand...

If you do that, you get the response I was talking about earlier, with 1.2dB hump for LP and HP and 0dB at center for the BP.

This will also bring the BP output into the same polarity as the HP and LP.

Click to expand...

BP output is in quadrature with the LP and HP. Both polarities of BP produce the same amplitude output (phase is different) when recombined.

There are two issues there:
Unity summing, which, as we've just seen, can easily be achieved by tweaking some components or the topology.
Response under gain-reduction, which shows large interaction, because the HP and LP corner frequencies are the same (and same as the center freqiuency of the BP). The BP output is relatively narrow according to mathematical definition - about one octave, depending on the chosen allignment, but it spreads audibly over a spectrum of about 5 octaves. In contrast, the effect of applying GR to the BP creates a narrow notch that is almost insignificant.

Click to expand...

#1, I said same polarity not the same phase... yes they are in quadrature, but same polarity (only 90' apart, not 270').

#2.. If I have a simple inverting opamp, with one input resistor and three feedback resistors, each FB resistor coming from one of my three individual bandpasses, how can these three bandpasses, not sum to equal the input signal?

Any two will exhibit phase interaction in the transition region, but all three are forced to sum perfectly by the nature of how simple summing opamps and feedback networks work.

Of course after you alter these bandpasses, and recombine with different levels or different anything, the perfect sum will be compromised, but that is the nature of any multi-band dynamics processor, with the exception of the de-esser topology, that only band splits the side-chain, while applying gain changes wide band to the full audio signal. 

JR

PS: I prefer the de-esser approach in general since it doesn't interfere so much with the integrity of the audio waveform, but I consider this an effect so smash away... .

JohnRoberts said:

#2.. If I have a simple inverting opamp, with one input resistor and three feedback resistors, each FB resistor coming from one of my three individual bandpasses, how can these three bandpasses, not sum to equal the input signal?

Click to expand...

The schematic you posted in answer #9 does not, because the 1st opamp has different gain on its pos and its neg inputs.

Of course after you alter these bandpasses, and recombine with different levels or different anything, the perfect sum will be compromised, but that is the nature of any multi-band dynamics processor,

Click to expand...

What I mean here is that having the poles of the HP and LP interfering with each other has consequences that go beyond simply "compromising the perfect sum". It's unfortunate the upload folder is closed, but I can show that reducing gain by 6 dB on the LP actually produces a 1.5dB hump 1/3 octave below center frequency.

abbey road d enfer said:

JohnRoberts said:

#2.. If I have a simple inverting opamp, with one input resistor and three feedback resistors, each FB resistor coming from one of my three individual bandpasses, how can these three bandpasses, not sum to equal the input signal?

Click to expand...

The schematic you posted in answer #9 does not, because the 1st opamp has different gain on its pos and its neg inputs.

Click to expand...

Yes, I just lifted a generic SVF schematic to post... and in response to your earlier comment. suggested grounding the + input.

Of course after you alter these bandpasses, and recombine with different levels or different anything, the perfect sum will be compromised, but that is the nature of any multi-band dynamics processor,

Click to expand...

What I mean here is that having the poles of the HP and LP interfering with each other has consequences that go beyond simply "compromising the perfect sum". It's unfortunate the upload folder is closed, but I can show that reducing gain by 6 dB on the LP actually produces a 1.5dB hump 1/3 octave below center frequency.

Click to expand...

Even if this does start out summing to unity, once you mess with the individual levels dynamically, it won't anymore, but what band splitting approach does?  I guess you can argue that this will act differently than others, and in your judgement it is worse. I won't argue that any approach is good or better than another.


JR

PS: yes if you imbalance the level of the HP and LP quadrature outputs they won't cancel, since they need to be equal level  "and" (+90') - (-90') = 180' to null.

JohnRoberts said:

Even if this does start out summing to unity, once you mess with the individual levels dynamically, it won't anymore, but what band splitting approach does?  I guess you can argue that this will act differently than others, and in your judgement it is worse.

Click to expand...

It is not dependant on the filter technology, actually. This is just a consequence of having the HP and LP characteristic frequency too close to each other.

I won't argue that any approach is good or better than another.

Click to expand...

I think almost everybody would concur with me that a system where the level increases when some GR is applied is seriously flawed. It's not a matter of personal taste.

PS: yes if you imbalance the level of the HP and LP quadrature outputs they won't cancel, since they need to be equal level  "and" (+90') - (-90') = 180' to null.

Click to expand...

I agree, is it your explanation for the hump? Yes. The hump is there because in the crossover area, the HP and LP have significant output, but being out of phase, they cancel nicely, making the BP the main contributor. When either the LP or HP is reduced, there is a build-up of energy, which is quite counterintuitive.