16 Channel analog summing bus

[fallout]

I found this link on another forum for an 8 channel analog summing bus. I was wondering what modifications would need to be done to make this 16 channels. Would it just be a matter of changing the 22k resistor?

http://forsselltech.com/8chsum_1.pdf

Thanks!
-Jay

[ssltech]

That design has some odd aspects. -These days it's usual to use op-amps in a virtual-earth application. That design has a fader between the summing point and the gain stage... basically an all-passive front end with a simple make-up gain stage... noisy and prone to crosstalk/bleed. -It's essentially how I built the first mixer that I ever designed and built myself, and from that design I learned the pitfalls... I'd never do it that way again!

You could just keep adding more passive inputs on the front end, but they will interact noticeably due to the lack of a virtual ground at the summing point.

-Not how I'd do it at all...

Keith

[gyraf]

That, or changing the 1K resistors (setting the output gain) to 470R. Both ways you get 6dB more gain to compensate for your higher number of loading channels.

But altering the output stage has the advantage of keeping your input impedances up - a figure that would be lower if the summing resistors were made smaller.

Btw, I think it would be wise with some 1uF polyester capacitors between the OPA604's and the pots. Otherwise, the input offset voltage of the opamp will cause the pots to be scratchy all too soon.

Jakob E.

[PRR]

Leave the 22K as-is; it is fine. Just build more input networks.

At 8 or 16 inputs, any mix network has large losses unless you are really using all the inputs. Do not build a mix network any bigger than you need. With tis topology, a 16-in mix network has a "gain" from a single input of 1/15 or -23.5dB. Put your lead track in at 0dB and it comes out at -23dB. If you put 16 tracks in at 0dB, they come out about 0dB; but you rarely want all tracks at the same level. In practice the output of this with 10 or 15 various inputs will be around -10dB.

Actually it will be lower because his pan pot takes a high 5.5dB of loss at center. So with all pots UP, loss is 29dB.

If you do the "obvious" things: set input and master faders and pans half-way, the total loss is about 58dB! With Audio-taper pots set about "7", it is "only" 40dB.

You need to run this type of network with the input faders (5K Aud) mostly near the top. I don't see why these are Audio taper uless you have tracks that are way-hot on the 16-track tape and you want them subliminal on the mix.

The master faders between mix network and make-up amplifier seem wrong. As Keith says, it makes a low signal lower and buries it in the mud. The only way this will be happy is to keep the master faders ALL the way up, using them only for the fade-to-black at the end of the song. Even so, I don't see why they are shown as LOG pots. The useful range between mix network loss and pot taper will give too-low levels except in the 8-10 zone at the top of the fader.

Assuming you leave the Master pots way-up, Keith's comments about virtual-earth mixing are not so very important on a many-input mix network. If you have just two inputs being mixed, with a simple passive mix network then when you move one fader, the load and level on the other fader changes. Interaction. Virtual earth mixing eliminates that, and is probably nice for consistent levels up to 8 inputs. But at 10 or 16 inputs, it is easy to keep the interaction down to 1dB, which is small. Virtual earth does not improve the noise more than a trace (an insignificant trace with many inputs). But what it does do is "prevent" you from putting a fader between the mix network and the make-up amp. You can work the summing amp at fixed-gain (a 22K feedback resistor) and put a fader and main-out amp behind it, or you can use a variable pot in the summing amp feedback path.

[rafafredd]

how about this one?

http://www.forsselltech.com/schemati...ng%20Buss2.pdf

[ssltech]

That's a bit more to my taste, -A virtual earth, with a DC-killer servo and the added noise cancellation of "Quasi-Balanced" remote source ground sensing.

On the original circuit, just feeding a signal panned dead-center at 0VU output on left & right channels, I rather suspect that even with the other (unused) inputs turned down, fiddling with the panpots from the "dead" outputs would interact to the point of noticeable "steering" the signal away from dead-centre. Virtual ground circuits (such as this second example) accomodate this without that interaction.

A second benefit of virtual ground is that the gain is proportional to the number of connected inputs (as I think it has to be with both topologies) but with 'Virtual Earth' it 'automatically' adjusts if you disconnect or add more inputs. -thus if you add a simple "disconnect" switch for unused inputs, the gain (and consequently the noise) diminished significantly. -You'd have to manually adjust the gain to match if you added or subtracted input networks from the original interactive circuit.

Keith

[rafafredd]

Iīve been thinking about building one of this to use as an analog mixing board.

Rsum and Rfeedback are set to 15K. And there is anote saying that both should be set at 22K for 48 channels.

So, 15K is good for how many channels?

Would you be able to calculate Rsum and Rfeedback for a 16 channel unit?

Also, what about this unit feeding line inputs? Would it need an output transformer or any impedance matching?

Each Forsselltech opamp is U$65.00, So I think I would be able to put a 16 chanell unit together for not more than US$300... (if I can go without output transformers)

And about gain implementation? Would it be possible to use a pot or a resistor network on Rfeedback or should I add an attenuator or something on the output?

I know, one more time, LOTS of questions!

Many Thanks

[fallout]

Rafafredd, do you plan on making it stereo? I was speaking with alk509 on 'IRC' and said if we used 2 it could be used in stereo as well.. I'd like to make a 24 input unit and be able to use the inputs as either 12 stereo pairs or in mono, sort of like the Dangerous 2 Bus. Would it be possible to somehow use the pairs in mono by using a switch? BTW you all should hop on IRC sometime, take a look at Scenaria's post...

-Jay

[BjornZetterlund]

Is the JFET 990 really a good choice as a summing amplifier, given that it has a FET input? I thought at least in terms of noise, normally you see bipolar op-amps chosen for this role, and that the summing amp generally follows similar design requirements as mic-pre's.

I'd be very happy if someone could shed some more light on how you decide on the requirements of the actual summing amp. I replaced the mix summing amp on my Soundcraft 800B (which was basically a 5532) with two OPA604's, each with a BUF634(?) 250 mA current booster in the feedback loop. I never really arrived at a good design approach for this, but in very crude terms I thought that the worst case scenario must be a sine wave of say 30V p-p driven into the 10k resistor of each channel, all of which feed the virtual earth. Assuming 48 channels all fed with the same (in phase) signal, when this signal is at a 15V peak, the total current into the virtual earth is 1.5mA*48 = 72 mA; therefore that is the maximum current that the op-amp will have to sink or source. That is obviously a hopelessly unrealistic case, so how do you arrive at a more realistic estimate?

And as regards the feedback resistor, on the Forssell page it is the same value as the 'input' resistors, but on some designs it is less than this, to ensure that the summing amp doesn't run out of voltage swing as it tries to pull the current through it. Once you arrive at the maximum current, this is straightforward to work out unless there are details here that I'm ignorant of.

To tie all this together, there is a gentleman on this forum (forgot who again...sorry!) who has a desk that is modified so as to have patchable summing amps, and he simply patches in whatever mic preamp he likes depending on what sound he is looking for. I think I understand how this works, but surely some pre's would be less suitable than others in this regard? For instance, could you use a tube pre like the Gyraf G9 on the drum subgroups to get a nice fat sound, or would it have insufficient current drive?

That brings up the last (promise!) question which is what would you set the gain to? Presumably maximum; the JFET 990 has 67dB max gain which is way less than a 5532. My electronics theory knowledge assumes an op-amp with infinite gain for virtual earth circuits, but I assume less gain just means the virtual earth is less perfect and fluctuates more with the signal it sees? And how does that affect the sound? (...ok, last paragraph was one question in three parts!)

Sorry to be a pain, and thanks for any replies!

Bjorn

[cuelist]

Originally posted by rafafredd:
So, 15K is good for how many channels?

In this case (the schematic from Fred Forssell) your main concern is the loading of the pan-circuit. The mix resistor (Rsum) loads the pan-pot and consequently affects the panning character.

As a 'rule of ', set the feedback resistor to the same value as the Rsum or at half this value to get 6db attenuation for increased headroom. You can make that up later in the chain. Usually its mixamp, fader, post fader buffer (providing the makeup gain).

I have seen as low as 2k mixing resistors, the lower you go, the less noise. The CADAC theatre consoles, often with 100+ channels use 3k6. Mackie uses around 3-4k too. On the other hand, the mix amp must have enough current drive to feed the correspondingly low feedback resistor.

There are other issues with summing amps. The virtual earth summing point is a favourite input for RF garbage. Put a small ferrite bead in series with "-" input, add a low value resistor in paralell with an inductor if you really want to keep the cellphones out!

Mats