Unwanted notch (Crossover)

[stonerbox]

Hi!

Giving my first go at a split and merge frequency-thing. Trying to filter out the low frequencies from one signal and keep the lows from another, or vice versa. It is a stereo circuit and the purpose of the selector is to keep the bass in mono at all times.. plus it gives the user an opportunity to chose what bass character (of the several circuits) they want as the foundation of the distortion. What you see down below is only the left (unfinished) part of the pedal.

The issue is when I merge the sallen keys a steep notch appears at the meeting point. I can get rid of the notch by "bleeding" the filters into each other (lowering HP / upping LP) but that is not really optimal. My guess is it has to do with some sort of phase issue. Can I match the phases where the filters starts?

[Rob Strand]

What's normally done with second-order crossovers is you flip the phase of one the crossover outputs, usually the highs.
Dig up some articles and you will see.

(I should mention Q=0.707 gives a 3dB peak and Q=0.5 gives a flat response.)

I'm surprised how basic info on crossovers has disappeared from the web,

https://www.tonmeister.ca/wordpress/2012/12/31/its-impossible-to-build-a-good-loudspeaker-part-1-crossovers/

Something to keep in mind.   When you do band-splitting you only have to worry about the signal adding back together in an ideal/theoretical sense ; just like you are doing in your simulation.   For a crossover used for speakers there's zillion details which complicate things (such as driver response, phase shifts, acoustic center alignment, speaker impedances) you can ignore all aspects related to these for band-splitting.

The Q=0.5 case, which is what you circuit is,  is also called 2nd Order Linkwitz-Riley (LR2).

Just for your info, there's tricks where you can have Q=0.707 or Q=0.5774 (or other), which have peaks in the response,  but you then you make the filter frequencies slightly off and that flattens the peak.  You still need the HF inverter.   It's probably not a good idea to detune the filter frequencies to removed the notch as a way of avoiding the inverter - that can come back to bite you.

[Fancy Lime]

On the plus side, Unwanted Notch isn't half bad as a band name. I'm thinking Japanese 80's garage rock or possibly some grungy shoe gaze.

Sorry, I can't help you with your actual question. I keep my hands off signal splitting and recombination for exactly that reason.

[ElectricDruid]

On an unrelated LTSpice note, you can save yourself some work by setting up one 9V power supply and one 4.5V Vref supply. Use "Draft->Net Name" to create a new label for those voltages (So "+9V", and "Vref" or "4.5V" or "Bias" or whatever you want to call it) and connect the labels to your two voltage sources. Then when you need to connect something else to that voltage, you just attach it to a label with the same name, instead of having to create a whole new voltage source. It's exactly like the "Ground" symbol in fact, but this time you make the label, instead of using one that's a pre-defined symbol.

I hope that's clear enough. It's easier to do than it is to explain!

[PRR]

I wonder what R96 does. Besides hiss.

[stonerbox]

What's normally done with second-order crossovers is you flip the phase of one the crossover outputs, usually the highs.
Dig up some articles and you will see.

(I should mention Q=0.707 gives a 3dB peak and Q=0.5 gives a flat response.)

I'm surprised how basic info on crossovers has disappeared from the web,

https://www.tonmeister.ca/wordpress/2012/12/31/its-impossible-to-build-a-good-loudspeaker-part-1-crossovers/

Something to keep in mind.   When you do band-splitting you only have to worry about the signal adding back together in an ideal/theoretical sense ; just like you are doing in your simulation.   For a crossover used for speakers there's zillion details which complicate things (such as driver response, phase shifts, acoustic center alignment, speaker impedances) you can ignore all aspects related to these for band-splitting.

The Q=0.5 case, which is what you circuit is,  is also called 2nd Order Linkwitz-Riley (LR2).

Just for your info, there's tricks where you can have Q=0.707 or Q=0.5774 (or other), which have peaks in the response,  but you then you make the filter frequencies slightly off and that flattens the peak.  You still need the HF inverter.   It's probably not a good idea to detune the filter frequencies to removed the notch as a way of avoiding the inverter - that can come back to bite you.

Thank you for the link!
Upon researching the topic I had actually read an old thread where you recommended a inverter crossover filter. They are pretty neat but since it is only 1st order I stuck with a Linkwitz-Riley. Seriously reconsidering going with you suggestion now. That filter does not (for the most part) mess with the phase as much and at least not where the cut off points meets. At least not any visible in LTSpice. Still not optimal with a 1st order. Could it be reconfigure to a 2nd?

It is really odd nobody has successfully created a mono-left-right-switcher like this before. As many of you know, in mixing and mastering, summing down the low frequencies to mono while keeping things above 150-200hz in stereo is very very common... I bet some 70's or 80's hardware was built with that feature too at some point. Time to start digging.

I wonder what R96 does. Besides hiss.

Thanks!

[sergiomr706]

Something like a subwoofer crossover, like this one? https://320volt.com/en/18db-20hz-200hz-subwoofer-crossover-circuit/

[PRR]

> mastering, summing down the low frequencies to mono while keeping things above 150-200hz in stereo

https://www.ka-electronics.com/kaelectronics/Elliptic_EQ/Elliptic_EQ.htm

"Elliptic Filter" means something else in non-audio applications.

[Rob Strand]

Thank you for the link!
Upon researching the topic I had actually read an old thread where you recommended a inverter crossover filter. They are pretty neat but since it is only 1st order I stuck with a Linkwitz-Riley. Seriously reconsidering going with you suggestion now. That filter does not (for the most part) mess with the phase as much and at least not where the cut off points meets. At least not any visible in LTSpice. Still not optimal with a 1st order. Could it be reconfigure to a 2nd?

So in your original design the two filters are Sallen and Key types.    In order for the response to add to flat you need to add an inverter to the HPF stage.   So what I was probably referring to in the old post (I'm not sure what post it is) is you can save an opamp by using an inverting filter.   The common inverting filter is the MFB (Multiple Feedback) type.  These *are* second order filters, just think of them as an alternative to the Sallen and Key - the BF-2 Boss Flanger uses a low pass version of the MFB filter.

You can see from the following sims that the inverter makes the response add to 1 (actually 0.5 due to the resistive mixer loss).

You can also see that the response of both high-pass filters are identical.







By the way all the problems related to crossovers doing weird stuff when the add back together is nothing to do with the circuits.   It's actually a fundamental problem with filters, even mathematically they don't add back together.   The crossovers that work add to *magnitude* 1 but they don't add to give the original signal, there's always a phase shift problem.  (People have come-up with all sorts of maneuvering over the past 50 years to work around the problem but there's always something inconvenient left over.)

[m4268588]

Either one becomes 20db/dec.
I don't know how mastering process for vinyl's works.

[m4268588]

ALL RIGHT!
Divide BPF with this

[stonerbox]

> mastering, summing down the low frequencies to mono while keeping things above 150-200hz in stereo

https://www.ka-electronics.com/kaelectronics/Elliptic_EQ/Elliptic_EQ.htm

"Elliptic Filter" means something else in non-audio applications.

Look a that schematic! My eyes!!! Thank you Paul for some new filter studies.

Thank you for the link!
Upon researching the topic I had actually read an old thread where you recommended a inverter crossover filter. They are pretty neat but since it is only 1st order I stuck with a Linkwitz-Riley. Seriously reconsidering going with you suggestion now. That filter does not (for the most part) mess with the phase as much and at least not where the cut off points meets. At least not any visible in LTSpice. Still not optimal with a 1st order. Could it be reconfigure to a 2nd?

So in your original design the two filters are Sallen and Key types.    In order for the response to add to flat you need to add an inverter to the HPF stage.   So what I was probably referring to in the old post (I'm not sure what post it is) is you can save an opamp by using an inverting filter.   The common inverting filter is the MFB (Multiple Feedback) type.  These *are* second order filters, just think of them as an alternative to the Sallen and Key - the BF-2 Boss Flanger uses a low pass version of the MFB filter.

You can see from the following sims that the inverter makes the response add to 1 (actually 0.5 due to the resistive mixer loss).

You can also see that the response of both high-pass filters are identical.







By the way all the problems related to crossovers doing weird stuff when the add back together is nothing to do with the circuits.   It's actually a fundamental problem with filters, even mathematically they don't add back together.   The crossovers that work add to *magnitude* 1 but they don't add to give the original signal, there's always a phase shift problem.  (People have come-up with all sorts of maneuvering over the past 50 years to work around the problem but there's always something inconvenient left over.)

Ah, now I get it. I can't thank you enough for this, you really went above and beyond to help me.