Signal path matching

[Tinkerer]

When we have 2 signal paths that we want to subtract from each other, it is very important to have both signal paths matched.
Every small difference, when amplified later results in significant noise.

What are the different ways to achieve an exact match?

What can be called common mode signal or noise and how can we reduce it?

[Carl-NC]

Most of the matching requirements are for getting the propagation delays equal. Caps are the biggest issue. They need to be hand-matched, 1% or better. You can also split a cap into multiple pieces. Ferinstance, if you need a 470nF cap use (10) 47nF caps instead. You get a statistical matching improvement of sqrt(10) so if you buy 2% caps you end up with an overall 0.62% tolerance.

Resistors, obviously use 1%.

Opamps: I use duals and split the dual between the channels. Opamps on a common die will match pretty well.

Some designs use JFETs for either the sampling switch or a clamping switch and those need to be hand-matched. I prefer the 4066 and other (better) variants. Again, switches on the same die tend to match pretty well.

For noise:

Use good power planes and bypassing at each chip. Depending on the design, you might want to guard-ring particular opamps. Lay the channels out very symmetrically with an adequate buffer zone to any nearby noise sources. Route clock lines symmetrically and preferably on the bottom layer. I do everything I can to route all signal lines on the top layer.

[Tinkerer]

Most of the matching requirements are for getting the propagation delays equal. Caps are the biggest issue. They need to be hand-matched, 1% or better. You can also split a cap into multiple pieces. Ferinstance, if you need a 470nF cap use (10) 47nF caps instead. You get a statistical matching improvement of sqrt(10) so if you buy 2% caps you end up with an overall 0.62% tolerance.

Resistors, obviously use 1%.

Opamps: I use duals and split the dual between the channels. Opamps on a common die will match pretty well.

Some designs use JFETs for either the sampling switch or a clamping switch and those need to be hand-matched. I prefer the 4066 and other (better) variants. Again, switches on the same die tend to match pretty well.

For noise:

Use good power planes and bypassing at each chip. Depending on the design, you might want to guard-ring particular opamps. Lay the channels out very symmetrically with an adequate buffer zone to any nearby noise sources. Route clock lines symmetrically and preferably on the bottom layer. I do everything I can to route all signal lines on the top layer.

Thank you very much for the good advise.

How about diodes. How close do they match?

[ivconic]

Generally (without knowing true specs of signals); band pass and unity gain, compressor maybe, at the end, just before subtracting.
That's why direct sampling is always better choice.

[Carl-NC]

How about diodes. How close do they match?

That's a good question... I really don't know.

[Teleno]

Thank you very much for the good advise.

How about diodes. How close do they match?

Use a transistor array, short the collector-base junction and use the emitter-base junction as the diode. Some inexpensive matched pairs here https://nl.mouser.com/Semiconductors...sistor&FS=True

There are unconnected schottky diode pairs such as the HMS-2825

[Tinkerer]

Use a transistor array, short the collector-base junction and use the emitter-base junction as the diode. Some inexpensive matched pairs here https://nl.mouser.com/Semiconductors...sistor&FS=True

There are unconnected schottky diode pairs such as the HMS-2825

Thank you for the feedback.
Here is the way I look at it:
General opinion is that it is useless to connect 2 diodes in parallel, because their conductivity does not match, so the diode with the lower conductivity carries all the load.
So when I separate the positive signal path and the negative signal paths with diodes, I assume that there is a possibility that the 2 signals might not be 100% identical after the diodes.

How should I go about for testing the diodes? For matching?

Teleno, thank you for the links. Matched pairs sounds like an excellent idea.

[Tinkerer]

Generally (without knowing true specs of signals); band pass and unity gain, compressor maybe, at the end, just before subtracting.
That's why direct sampling is always better choice.

sorry for my ignorance: what is a compressor?

[ivconic]

In audio application is known as "compressor", aka limiter.
Limiting dynamic range at some extent; but providing steady signal for further processing.
Since compression is adjustable; two separate signals can be "compressed" and totally equalized before any mixing, subtracting etc.
As i said i don't know detailed specs of those two signals, you never explained.