>
keep crosstalk levels as low as possible.
Just to be clear what the problem is:
I've shown two sources V_left=1Volt and V_right=0Volt. I used DC batteries because my sim can show DC results on the plan. Since I have neglected reactances, this is valid.
I've picked values that are maybe realistic but "embarrassing".
The signal from V_left reaches Lout at almost full level, OK. The signal at the mono mix Mout is almost half a volt, which is correct for these inputs.
The Right signal at Rout should be Zero. Instead we have 42mV of the
Left signal or about 26dB below the level of the Left source. Why? Left signal flows through R3, past the buffer, through R4, to the Right side. Here it is sucked-up by R2, the output impedance of the source. (Also by R7 load resistance, but normally R2<<R7 so we can neglect this; I'm going to neglect many small errors.)
There are some obvious improvements. If R3+R4 are larger, leakage onto R2 is less. But the buffer loads R3||R4. The buffer you cited has ~80K input resistance, so R3 R4 could be 10K easy, but going to 100K will cause some drop of level into the buffer (which may be tolerable if the following stage has gain in reserve). Also huge values of mix resistor will get you into noise issues.
Note that if we add a resistor from buffer input to ground, not all the current in R3 passes to R4 and over to the right side. If the buffer input were 2 ohms, crosstalk would be about (2/4700)*(470/4700)= -87dB! But buffer input would be about 2/4700= -67dB re: the input, so we'd need a ton of makeup gain. This is the advantage of a Feedback Mixer: NFB forces the summing node to near-zero impedance, while the overall gain can be unity or whatever (even if you cut inputs on or off the bus).
Note that if R1 R2 were dead-nuts Zero, there would be no crosstalk, and 2 ohms would give -73dB which is often tolerable. What are realistic output impedances?
Most chip opamps have naked output resistances of 10Ω to 500Ω, but this is reduced by feedback. At 50Hz we may have 100,000:1 of feedback, so 100Ω reduces to 0.001Ω which is pretty small. But an amp with finite gain-bandwidth working at some gain may only have 10:1 of spare gain at 20KHz, so output resistance could be 10Ω. This could give -66dB crosstalk, which might be OK.
But... most chips get unstable if they see capacitance, such as cables. Or they run hot if the idiot shorts the output. So there is almost always an actual resistor between the chip and the jack, OTOO 22Ω to 470Ω. (That's where I got my embarrassing example.) If you
know your source is very low-Z, you may be able to figure a mix with tolerable crosstalk. But most boxes don't tell you the actual output impedance, only the minimum load. (Even if there is a number on the sheet, it may be wrong.) And you may need this to work with various sources.
Since you have Rod's emitter-follower open, one thing you can do is use two EFs to buffer the sources, mix in R3 R4, and buffer the output. Three buffers. Crosstalk won't be zero, but better than -60dB, and -80dB if you put about 5K between each input buffer and its source. You could use just one transistor as an inverter with gain and do Feedback Mixing, but crosstalk might be around -40dB; you need a pretty good amp to make the active mixing really clamp the summing node, and before long it is easier to grab a chip.
I neglected reactances (except to imply that most amp outputs become inductive at the top of the audio band). Some boxes have series capacitors. These may be picked for "full bass" in a ~10K load, which means the cap reactance may be much more than 22-470 ohms at 20Hz. This means increased crosstalk in the bottom octaves. If you mix your bass mono (gets both stereo speakers booming in phase), that may not matter. But if you have significant stereo bass, bass crosstalk foils your effort.
