Do you think this phase-selectable splitter will work???

[Praying_V]

Hi everyone.  I know RG has an article with a phase switching buffer in it.  I want to do something a little different however, and I want to double check that I'm not missing something.  Here's what it would look like:



The first dual-opamp is splitting the signal into two channels.  I'm thinking that I need these buffers here for high input impedance.  Then, each channel goes to another buffer.  With the switches in the positions shown, its an inverting amp with a gain of 1.  With the switches flipped, signal now goes to the non-inverting input, and the feedback resistor is shorted.  Now its a non-inverting amp with a gain of 1.

Well, how does this look?  How do my cap & resistor values look?  I figure the caps should be as large as possible, maybe larger than I'm showing here, because any low-end roll-off would be bad (I want to use this for bass).

*edited for a typo

[earthtonesaudio]

The first section(s) look okay, you can probably get away with a smaller cap, even for bass.  A .01uF cap with a 1M resistor to ground gives a cutoff of about 16Hz, plenty low.

As for the second stages, it looks like it would work just fine in non-inverting mode, but I'm not sure if it would work as well in the inverting mode.  It depends on the opamp you use, but generally when you use them as an inverting buffer, the non-inverting input is either connected directly to the half voltage supply, or through a resistor equal to or smaller than then inverting input resistor.  Check the datasheet for the op-amp you're using, it should have a diagram for an inverting buffer.

[R.G.]

Works, but is unnecessarily complicated.

The first dual-opamp is splitting the signal into two channels. I'm thinking that I need these buffers here for high input impedance.

One opamp works fine for driving two following opamps. No real "splitting" happens other than driving two inputs, which one opamp can do fine. It also adds the extra opamp input current and noise. The current may not matter, especially for JFET input opamps, but the noise might well.

With the switches in the positions shown, its an inverting amp with a gain of 1.  With the switches flipped, signal now goes to the non-inverting input, and the feedback resistor is shorted.  Now its a non-inverting amp with a gain of 1.

With the input resistor to the inverting input open and signal to the non-inverting opamp, the opamp is still a gain-of-one buffer. The value of the feedback resistor, whether zero or a megohm does not matter to gain. It does matter to noise. 10K for both input resistor and feedback resistor is a good compromise when you are driving the input from another opamp, so go to 10K for both resistors on both output sections.

Eliminating the feedback resistor switching lets you use a SPDT switch. Changing to the alternate circuit I used lets you use an SPST switch.

How do my cap & resistor values look?  I figure the caps should be as large as possible, maybe larger than I'm showing here, because any low-end roll-off would be bad (I want to use this for bass).

Caps are evaluated for frequency response by using the well known, and well-used in this forum, RC rolloff formula: F = 1/(2*pi*R*C)

At the input, if you're using JFET opamps like the TL072, the 1M input bias resistor is all there is to R. If you use something like the NE5532, the opamp itself can have an input resistance as low as 100K. Let's assume the TL072 for simplicity. Then

F = 1/(2*3.14*1E6*22E-6) = 1/ 138.16 = 0.007237, or about 7 MILLI-hertz. Yep, you got the bass frequencies on that one. 

4-string bass goes down to 42Hz; hearing quits at 20Hz. To get to 20 Hz with a 1M input resistor only needs C = 1/(2*pi*F*R) = 7nF. So any cap at the input bigger than 7nF will work to an octave below normally-tuned bass.

On the second stages, the input impedance changes when you switch. When it's in the noninverting position, same math as above holds. When its in the inverting position, the input resistance of the opamp goes to 100K, the input resistor. Then the frequency response with a 22uF only goes down to 70milli-hertz. You can get away with any cap bigger than 70nF, so a 0.1uF/100nF works great.

On the output, you can't do the math that well, as you don't know what load it will see. If a guitar amp, then 22uF into a 1M guitar amp plus the 1M pulldown is response down to 14milli-hertz. If into a 10K effect input, only down to 0.7Hz, arguably just fine. If into a fuzz face, it's 10X bigger than the 2uF cap on the FF input, so it simply cannot limit frequency response. It makes the FF sound bad for reasons having to do only with the FF, not the splitter.

Does that help?

[Praying_V]

Moving in the right direction?  I cut out the unnecessary opamp, removed the redundant switches, and lowered those resistors to 10K each.  Should I be shorting the 1M resistor to Vref when in "inverting mode"?

About the caps, I thought that when a signal runs through several consecutive amplification stages there is a cumulative frequency roll-off effect with each additional coupling cap.  I mean, if my signal goes through three 22uf caps here, don't I have to treat them as if they're in series when making frequency calculations?  And I thought that other buffers in my signal chain would also have caps in series, further raising the low cut-off frequency...  Am I way off about this?

[earthtonesaudio]

I don't think you need to worry about cumulative frequency roll-off with multiple caps and multiple stages.  Something you might need to worry about (but mostly only in high-gain circuits, not this one in particular) is phase shift through the components.  It's easy to accidentally make a phase shift oscillator when you have multiple high gain stages cascaded through lots of R's and C's.



You could probably reduce the value of the 1M resistors going to Vref on the two output op-amps, maybe to 10k or so, for slightly less noise.

[R.G.]

Moving in the right direction?  I cut out the unnecessary opamp, removed the redundant switches, and lowered those resistors to 10K each. 

Yep, getting better.

Should I be shorting the 1M resistor to Vref when in "inverting mode"?

That would lower noise a bit, but I don't think it's worth it. Try this instead.

Make Rfeedback and Rin both 20K. Then add another resistor in series with the + input of 20K, leaving the 1M to bias from the + input. Now tie both the 20K resistors going to the inputs together. In the absence of any switch, a signal going into the combined inputs sees a gain of -1 through the path to the inverting input, and a gain of +2 from the + input, so the net result is the signal has a gain of +2-1 = 1. If you set your switch to short the 1M bias resistor, the gain through the + side is now zero, and through the inverting side is -1. The resistance for frequency considerations is 10K for a gain of -1 and 20K for a gain of +1. The switches can now be SPST, or you can use SPDT and use the opposite side to run LEDs for indicators.

About the caps, I thought that when a signal runs through several consecutive amplification stages there is a cumulative frequency roll-off effect with each additional coupling cap.

A little knowledge is a dangerous thing. You're worried about a second order effect (rolloffs in series) when you can't compute the first order effect of a single cap and a resistor. Learn the basics, then build on them.

That being said, yes, the rolloffs do accumulate, but not in a simplistic way. And your caps cause rolloffs so far below what is necessary that you can ignore them.

I mean, if my signal goes through three 22uf caps here, don't I have to treat them as if they're in series when making frequency calculations?

No. Each R-C rolloff is separate, but they each have a sequential effect. If they're identical, instead of the -6db point being the effective rolloff, the -2db point for a single section will be the -6db frequency for the cascaded sections. But you have to be able to do the math (i.e. RC-rolloffs and "db" calculations) to nail that point.

And I thought that other buffers in my signal chain would also have caps in series, further raising the low cut-off frequency...  Am I way off about this?

No, but you're vastly overcompensating, using hugely more capacitance than you really need. Not that that's necessarily going to sound bad; but it can cause you problems with turn on times as the caps ramp up to their operating positions, and coupling through long periods of DC offset. If you happen to put in a signal with an unintentional DC offset on it, the input with a 22uF/1M rolloff of 7milli-hertz will setting in five time constants of 22 seconds, or 110seconds. Far better you set the input rolloff for 10 Hz (0.22uF/100K) and have it settle in 0.11seconds.

Something you might need to worry about (but mostly only in high-gain circuits, not this one in particular) is phase shift through the components.  It's easy to accidentally make a phase shift oscillator when you have multiple high gain stages cascaded through lots of R's and C's.

Maybe. That's called motorboating at these low frequencies, and can only happen if you have feedback. The feedback almost has to be through the power supply if you don't do it intentionally; there is no feedback in the schemo, so the power supply is the only path. Generally power supply oscillation is at high frequencies, or if you have some unusually poor decoupling.

[Ben N]

1. Why do you need both channels to be phase-selectable? After all, phase is relative, and performance is better in a non-inverting opamp.

2. If you are using a couple of duals, you could use the remaining opamp as a simple (unbuffered) mixer, if desired.

[Praying_V]

Hey Ben,
I want to be able to invert both, because ultimately my bass player will have three chains being mixed back together.  Maybe not all at the same time.  But still, since one chain will have a fixed phase, I want to be able to invert the other two.

You're worried about a second order effect (rolloffs in series) when you can't compute the first order effect of a single cap and a resistor.

Woah, that was below the belt!  I can calculate the frequency response of an RC filter,  but i had thought that I had to take every single capacitor that my signal would see between guitar & speaker into account.  I figured, whats the harm in using the biggest caps I had?  But I see now, the answer is in the really long time constant.  So when you want to pass everything above 20Hz, is it always wise to use the smallest capacitance that will get the job done?

Thanks everyone!  I'm gonna go start soldering...

[R.G.]

I want to be able to invert both, because ultimately my bass player will have three chains being mixed back together.  Maybe not all at the same time.  But still, since one chain will have a fixed phase, I want to be able to invert the other two.

You're exactly right on that one.

Woah, that was below the belt! 

Sorry - it wasn't meant to be. I misestimated based on what I read. No offense meant.

I can calculate the frequency response of an RC filter,  but i had thought that I had to take every single capacitor that my signal would see between guitar & speaker into account.  I figured, whats the harm in using the biggest caps I had?  But I see now, the answer is in the really long time constant.  So when you want to pass everything above 20Hz, is it always wise to use the smallest capacitance that will get the job done?

In many cases you'd be fine, and there is some rationale for being sure that you're maybe an octave or a decade lower than you need to be so that the resistors instead of the perhaps imperfect electro cap dominates the response at lowest-audio. But really big time constants can cause you problems with offsets and slow turn on. I've stubbed my own toe on that from time to time.