Selective signal cancellation

[ricko_uk]

Hi,
I have a number of complex signals (i.e. like audio not repetitive waveforms) that go through different paths on the PCB so they have different delays and phase shifts. Up to 200KHz.

Some are added together and from those added together I have to sometime cancel out one of them (i.e. fully subtract it) from the summed final signal.

What are the best methods?

I am thinking of an inverting amplifier (for the signal I want to remove) which is then fed into the summing amplifier. Is that correct?

But I think that approach does not account for time delays (every signal goes through different paths/components). So it might not cancel the signal completely (or as much as possible).

So how do I shift that signal back or forth in time to cancel it out as close as possible to 100%?

Thank you

[fourfathom]

Since the propagation speed of a typical circuit board trace is about 6 inches per nanosecond, with your 200 KHz signals you can probably ignore any trace length differences.

As for subtracting the signals, yes you can use an inverting amplifier to subtract a signal at the summer, but if possible it might be easier to just disconnect that particular signal from the summing junction.

[ricko_uk]

Thank you fourfathom,
the signal to be subtracted is a reference signal that is also contained/mixed into other signals. That is why I need to subtract it and cannot just disconnect it.

What I was thinking about was not so much the trace delays but the delays/phase shifts that all sorts of parassitic (and sometimes component) capacitance introduce. Sometimes because there is some filtering of noise, that might slightly affect the delay/phase shift.

Any suggestions on how to shift a signal in time (just analog i.e. without using DSP)?

Thank you

[fcb]

It sounds like it's a phase shift rather than time delay you are looking for.  Could you use an all-pass filter to phase-shift your clean path to match the processed path?

[ricko_uk]

Thank you fcb,

a bit rusty on the analog... I thought a phase shift is the same as a time delay? What is the difference?

With an all-pass filter, how much phase shift can I add? And is it fixed or can I vary it?

Thank you

[fourfathom]

What I was thinking about was not so much the trace delays but the delays/phase shifts that all sorts of parassitic (and sometimes component) capacitance introduce. Sometimes because there is some filtering of noise, that might slightly affect the delay/phase shift.

Parasitic delay/shift/attenuation effects will depend much on the source impedance of your signal source (the source that is driving the traces).  Unless your source is high-impedance the parasitics in your frequency range should be minimal.

Deliberate filtering will obviously have an effect, and in order to be able to subtract one signal component you will need to have identical filtering in all paths (at least as far as the filtering affects your frequencies of interest).  If the filtering is only affecting high-frequency noise, then you can apply an additional low-pass filter at the summation output which will further attenuate any un-cancelled noise.

[Siwastaja]

"time delay" (or "group delay") and "phase shift" are conceptually the same but the actual numbers and units are different.

For example, if a 1Hz signal has 180 degrees of phase shift, the time delay is 0.5 seconds. If a 10Hz signal has the same 180 degrees of phase shift, the time delay is 0.05 seconds.

In reality, both time delay and phase shift are functions on frequency, not single numbers, so you plot them.

What follows is, if you want constant time delay at different frequencies, the phase shift is not a constant but a linear slope.

Using LTSpice, for example, you can do bode plots plotting either phase shift or group delay, whatever interests you more for particular application.

[pwlps]

I am thinking of an inverting amplifier (for the signal I want to remove) which is then fed into the summing amplifier. Is that correct?

You can do it with a single opamp configured as a difference amplifier:

https://www.electronics-tutorials.ws/opamp/opamp_5.html

But you will get a better precision using an integrated difference amplifier like the AD8271, AD8276 or similar.
https://www.analog.com/en/analog-dialogue/articles/deeper-look-into-difference-amplifiers.html

[bson]

Any suggestions on how to shift a signal in time (just analog i.e. without using DSP)?

Your parasitic phase responses aren't going to be linear, meaning there is no fixed delay.  Instead you need an allpass filter and a trim procedure to cancel out the parasitics.  The first step is to look at the phase response of the path you want to cancel it from, then model an allpass filter that is a reasonable match.  This may well turn out to be impracticable.  A VNA that can get well below the usual 300kHz or 100kHz is your best friend here - like the trusty old HP 3577A.

Edit: of course not parasitics at these frequencies, as much as filter responses.  Mathematical modeling is silver, but a real-world measurement is gold.

[fourfathom]

ricko_uk, it would help if you could tell us the relative levels and frequency characteristics of your input signals, and the amount of cancellation you require.  Also the amount and type of noise filtering you need.

What is the nature of this reference signal?  If it is well-defined you might be able to do some sort of synchronous filtering, designed for an extremely narrow notch.  This may not work if your other signals of interest also have components at the same frequency as the reference that you want to preserve. 

[Marco]

I have a number of complex signals (i.e. like audio not repetitive waveforms) that go through different paths on the PCB so they have different delays and phase shifts. Up to 200KHz.

If the phase shift from a few parasitics can make that much impact, the impact from a subtraction circuit which relies on resistor matching for subtraction accuracy is going to be far worse.

You might be able to use AD8293 if the fixed amplification works for you. MAX4208/MAX4209 are a bit tight in bandwidth.