> What I get with this arrangment is the summing I so desire, but the directs on the back of course are getting some of the other tracks bleeding in.
Draw it out and see why that happens.
Each source has a source resistance, not necessarily the same as the nominal load. For true-600 ohm gear of modern vintage, the source resistance may be 50-100 ohms, but for less professional stuff (and some stuff that claims to be pro, and most really old stuff) the source resistance may be 470, 600, or even 1,000 ohms.
I have assumed 1K ohms sources, 10K mix resistors, 12 inputs, infinite "direct" (bridged) loads, to make the math obvious. And I have napkin-drawn it unbalanced, though that really changes nothing in this analysis.
Say that Input 1 has a 1V loud signal, Input 2 has "the sweet sound of silence" (or a soft track) that you want to keep clean of the Input 1 signal.
The sneak path from Input 1 to Input 2 has two "split points": the mix network, and the output of Input 2. At each point the power can go several ways, and we can simplify each split into a 2-way split with tolerable accuracy.
Split 1: In round numbers, the mix network has loss of about 1/10. So the 1V on Input 1 lands at the mix bus at about 0.1V. This hits the "back" of all the mix resistors and leaks "toward" all the sources.
Split 2: Input 2 has source resistance of 1K. This with the 10K mix resistor gives loss of about 1/10.
So Input 1 arrives at Input 2 at (1/10)*(1/10)= 1/100 or -40dB of original strength. Assuming Input 1 is 1V, then 0.010V of Input 1 signal lands at the Input 2 jack and "direct" output.
What to do?
Suppose the source resistances were more like 100 ohms. We still have 1/10 loss in the mix network, but the back-feed path is the 10K mix resistor into the 100 ohm source, or 1/100 loss. Total loss is 1/1000 or -60dB down. In many situations that won't be a real problem.
But it can be. If you can get the source resistance down to 10 ohms we now have -80dB leakage.
> I can tolerate additional loss
There is another way. The 1/10 loss in the mix network "could" be higher. For illustration, a 100-input mix network or a reasonable network loaded in 100 ohms would have 1/100 loss. Combined with 100 ohm sources we get the additional 1/100 loss for total -80dB crosstalk, with 10 ohm sources we have -100dB crosstalk which really should be ample.
Simply loading the mix network reduces crosstalk by reducing signal. In many cases we can't afford much loss of signal because we are close to noise-level already.
This is really buying crosstalk the hard way: S/N is going down as fast as crosstalk. You have to be WAY above the universal noise to go for that path. NYDave's 10dB loading is a reasonable thing to try, and sure the cheapest quick-fix. But trying to get 20dB or 30dB less crosstalk that way will bite you like a snake (hissssss).
Active mixing has the indisputable advantage of reducing crosstalk without significantly reducing signal or S/N ratio. A common op-amp in the common summing-amp hook-up will show dynamic input impedance well under 100 ohms, under 10 ohms at lower frequencies (limited by non-infinite gain and bandwidth). Or to put it another way: mix-net crosstalk is now equal to the amount of feedback, not the number of inputs. Using 5534 or TL071 the crosstalk at 5KHz may be -60dB, plus whatever mix/source-resistance loss you have.
If you are stuck with a high-resistance source, high-resistance mix resistors help. They don't change the crosstalk in the mix-net, but increase the back-talk loss into your unavoidably high source resistance. A 10-in mixer, 1K sources, and 1Meg mixing will have 1/10,000 or -80dB crosstalk. However 1Meg mix resistors have 10 times the noise voltage as 10K mix resistors. However, the 10K mixer has self-noise lower than a cheap amplifier (and lower even than 5534), so the increase of noise may not be bad.
If crosstalk is more important than minimalist design, the path is clear: put buffer amps everywhere. The source boxes have resistance partly for stability in long cables but also for short-protection. Put a TL07x input buffer on every channel. Feed your "direct" output through a few hundred ohms for safety; feed the mix network directly from the op-amp pin. The output impedance of common op-amps will be an Ohm or less over most of the audio band, so the back-talk from the mix-network is swamped 1/10,000 plus the mix-net loss. Adding active summing gets you to a computed -140dB (but we have not considered inductive or capacitive leakage or ground resistance; -100dB would be exceptionally good in real life).
Draw it out and see why that happens.
Each source has a source resistance, not necessarily the same as the nominal load. For true-600 ohm gear of modern vintage, the source resistance may be 50-100 ohms, but for less professional stuff (and some stuff that claims to be pro, and most really old stuff) the source resistance may be 470, 600, or even 1,000 ohms.
I have assumed 1K ohms sources, 10K mix resistors, 12 inputs, infinite "direct" (bridged) loads, to make the math obvious. And I have napkin-drawn it unbalanced, though that really changes nothing in this analysis.
Say that Input 1 has a 1V loud signal, Input 2 has "the sweet sound of silence" (or a soft track) that you want to keep clean of the Input 1 signal.
The sneak path from Input 1 to Input 2 has two "split points": the mix network, and the output of Input 2. At each point the power can go several ways, and we can simplify each split into a 2-way split with tolerable accuracy.
Split 1: In round numbers, the mix network has loss of about 1/10. So the 1V on Input 1 lands at the mix bus at about 0.1V. This hits the "back" of all the mix resistors and leaks "toward" all the sources.
Split 2: Input 2 has source resistance of 1K. This with the 10K mix resistor gives loss of about 1/10.
So Input 1 arrives at Input 2 at (1/10)*(1/10)= 1/100 or -40dB of original strength. Assuming Input 1 is 1V, then 0.010V of Input 1 signal lands at the Input 2 jack and "direct" output.
What to do?
Suppose the source resistances were more like 100 ohms. We still have 1/10 loss in the mix network, but the back-feed path is the 10K mix resistor into the 100 ohm source, or 1/100 loss. Total loss is 1/1000 or -60dB down. In many situations that won't be a real problem.
But it can be. If you can get the source resistance down to 10 ohms we now have -80dB leakage.
> I can tolerate additional loss
There is another way. The 1/10 loss in the mix network "could" be higher. For illustration, a 100-input mix network or a reasonable network loaded in 100 ohms would have 1/100 loss. Combined with 100 ohm sources we get the additional 1/100 loss for total -80dB crosstalk, with 10 ohm sources we have -100dB crosstalk which really should be ample.
Simply loading the mix network reduces crosstalk by reducing signal. In many cases we can't afford much loss of signal because we are close to noise-level already.
This is really buying crosstalk the hard way: S/N is going down as fast as crosstalk. You have to be WAY above the universal noise to go for that path. NYDave's 10dB loading is a reasonable thing to try, and sure the cheapest quick-fix. But trying to get 20dB or 30dB less crosstalk that way will bite you like a snake (hissssss).
Active mixing has the indisputable advantage of reducing crosstalk without significantly reducing signal or S/N ratio. A common op-amp in the common summing-amp hook-up will show dynamic input impedance well under 100 ohms, under 10 ohms at lower frequencies (limited by non-infinite gain and bandwidth). Or to put it another way: mix-net crosstalk is now equal to the amount of feedback, not the number of inputs. Using 5534 or TL071 the crosstalk at 5KHz may be -60dB, plus whatever mix/source-resistance loss you have.
If you are stuck with a high-resistance source, high-resistance mix resistors help. They don't change the crosstalk in the mix-net, but increase the back-talk loss into your unavoidably high source resistance. A 10-in mixer, 1K sources, and 1Meg mixing will have 1/10,000 or -80dB crosstalk. However 1Meg mix resistors have 10 times the noise voltage as 10K mix resistors. However, the 10K mixer has self-noise lower than a cheap amplifier (and lower even than 5534), so the increase of noise may not be bad.
If crosstalk is more important than minimalist design, the path is clear: put buffer amps everywhere. The source boxes have resistance partly for stability in long cables but also for short-protection. Put a TL07x input buffer on every channel. Feed your "direct" output through a few hundred ohms for safety; feed the mix network directly from the op-amp pin. The output impedance of common op-amps will be an Ohm or less over most of the audio band, so the back-talk from the mix-network is swamped 1/10,000 plus the mix-net loss. Adding active summing gets you to a computed -140dB (but we have not considered inductive or capacitive leakage or ground resistance; -100dB would be exceptionally good in real life).
