Bus Assign switches, with 3dB attenuation

[bjoneson]

I'm continuing work on a summing amp project. The whole unit will consist of 24 inputs, with a gain pot, and 2 busses (L/R). I don't have a need or desire to have pan pots, but would like to be able to independently assign each channel to the L bus, R bus or both. This in and of itself is quite simple, but I'm trying to determine if there's away to apply 3dB of attenuation when both buses are assigned.

The concept I came up with is to use a DPDT push button for bus assignment (one for L, one for R), and using the spare pole / throw to introduce 3dB of attenuation on the opposite side when it's engaged. The only way I can think of to make that work is to switch a shunt resistor in a voltage divider in and out for attenuation.

The math is complex, and since this doesn't follow any of the "normal" cases for L-Pad or T-Pad attenuators, there doesn't seem to be any nice formulas to plug in to.

The truth is, I'm not even sure this is technically feasible, and achieving 3dB of attenuation could be made as simple as turning the channel gain pot counter clockwise a little bit when both busses are assigned. But alas... I'm a glutton for details.

The channel amp I'm planning to use is the gar2520 (from classic API) which can hypothetically drive loads down to 75 ohms. The busses will be virtual-earth type with 4.7K feedback.

Hoping someone can validate if this is even a feasible approach to solve the problem, and if so give some advice on calculating resistor values.

Thanks in advance for any help / input!

 

[joaquins]

I wouldn't go like that since you don't need to have the 150Ω load you are having in this case, even if the 2520 can handle it with no problem the higher current could introduce unnecessary noise into the rails and wasting power only for having the 3dB attenuation option.

You have two options, the first one would be a higher impedance pad, accounting for it's impedance in both positions to make it work properly. The other would be to change the resistor feeding the bus, if you want the 4k7 bus then 4k7* 1.41=6k64, then  you want to add a series resistor to the 4k7 of 1k94. 2k is easy to find E24 value, which gives you an attenuation of 3.08dB. It's in series with the 4k7 and shorted by the other switch when the other channel is off. I think I would go with that one, you also get lower noise gain with this approach since the impedance on the bus is lower at each channel assigned to L+R so it shouldn't make things worse, rather better, wouldn't be a lot but still, I prefer unnoticeable better than unnoticeable worse, and in this case you are saving 2 resistors per channel.

JS

 

[bjoneson]

The other would be to change the resistor feeding the bus, if you want the 4k7 bus then 4k7* 1.41=6k64, then  you want to add a series resistor to the 4k7 of 1k94. 2k is easy to find E24 value, which gives you an attenuation of 3.08dB. It's in series with the 4k7 and shorted by the other switch when the other channel is off.

That was originally what I was thinking, but I'm not sure I understand how I could switch the 2K series resistor in and out. For example, when the L Bus switch is "assigned", there's a 4k7 bus resistor to the L Bus. I can't think of any conceivable way to "introduce" a 2k resistor in series with this with another SPDT switch (1 side of the R Bus assign).

 

[joaquins]

You said DPDT earlier, and you probably want to use DPDT as stated in the scheme to be able to switch the resistor to ground so it isn't flying around when not used, could introduce some noise, I don't know how much would affect you in a stereo bus, probably more of a problem when there are more bus around...

There are ways of doing this with SPDT, you have one end of the switch at the opamp output, the other switches between the L assign bus resistor and the other to shorting the 2k resistor.

In fact you can't do what your schematic say with SPDT, maybe you are confused with DPST, in which case you would need a decider as you posted but probably with higher impedance, just scale it up, maybe 1k/2k, and you probably want to make a little less attenuation for more precise match for the 3dB target. In fact you are already at -3.5dB with the values you shown and it does follow pretty good the L-pad attenuator math, at least in a first approach, since you never have more than 100Ω resistance before the 4k7, so you are about 2% error neglecting the L-pad impedance in the channel gain.

JS

 

[bjoneson]

You said DPDT earlier, and you probably want to use DPDT as stated in the scheme to be able to switch the resistor to ground so it isn't flying around when not used, could introduce some noise, I don't know how much would affect you in a stereo bus, probably more of a problem when there are more bus around...

There are ways of doing this with SPDT, you have one end of the switch at the opamp output, the other switches between the L assign bus resistor and the other to shorting the 2k resistor.

In fact you can't do what your schematic say with SPDT, maybe you are confused with DPST, in which case you would need a decider as you posted but probably with higher impedance, just scale it up, maybe 1k/2k, and you probably want to make a little less attenuation for more precise match for the 3dB target. In fact you are already at -3.5dB with the values you shown and it does follow pretty good the L-pad attenuator math, at least in a first approach, since you never have more than 100Ω resistance before the 4k7, so you are about 2% error neglecting the L-pad impedance in the channel gain.

JS

Thanks again for taking the time to reply.

To clarify the switching topology... each bus assign switch, is DPDT. From a practical perspective, I'm looking at each DPDT switch as 2 SPDT switches mechanically linked. One half of the switch does the typical job of assigning the signal to the bus. The other side of the same switch does the job of attenuating the signal going to the opposite bus.

For clarity I've attached another schematic, showing the topology with no attenuation involved.

I'm with you on scaling up, was just trying to keep the L-Pad divider as low impedance as possible to reduce the impact and error on the summing bus.  I picked the 100 Ohm "worst case", for the channel amp, since it's a bit above the maximum drive capability for the opamp. Though I agree, that's a fair bit of current flowing, and has the potential to impact the power supply requirements, which I'm trying to keep low.

Another alternative I have is to use a single DP3T switch with L,C,R positions. the L position feeding the L bus via 4k7, the C position feeding both via 4k7s, and the R feeding the R bus via 4k7.  a DP4T would also allow for an "Off" position. Rotary DP4Ts are a bit pricey (and i need 24 of them). I also have a space consideration. I'm trying to pack 4 channels into a single module (about the size of a 500 series. There's just not a lot of real estate left on the front panel, since I'm already planning 10k pots for "faders".  I could do a slider DP4T (which are pretty inexpensive) , I'm just worried about the potential crosstalk in that type of switch.  Though as I look at it, what I have proposed isn't much better in that regard.

I may just roll with the simplest option (shown on the bottom of the attached schematic) and simply live with the lack of "pan law" when assigning a channel to both busses.

 

[joaquins]

Ok, on your new schematic, add 2k resistor in series with your 4k7 resistor, before or after, it doesn't matter. Then use the spare switch on the other side to short it out when needed (when this other side is on)

That should do it.

If you still want to scale up you can totally rule out the error since you know all the variables, and accounting the output impedance in each case of the attenuator in series with the summing resistor to know the actual gain of it would give you the exact figure. I don't see the point on doing so, but it's totally possible, and not so hard, you could simulate it and tweak the values in each position (with the pan or and off) till you get 3dB difference between them.

JS

 

[bjoneson]

Ok, on your new schematic, add 2k resistor in series with your 4k7 resistor, before or after, it doesn't matter. Then use the spare switch on the other side to short it out when needed (when this other side is on)

That should do it.

Yes! That totally makes sense, and is *way* simpler that what I had originally put together. It also puts *way* less load on the channel amp.  I'm not too worried about "scaling up" for more accuracy on the attenuation.  If I'm calculating correctly this put's things at right about 3.1dB of attenuation, which is plenty close enough for me. I'd like to keep the 4.7k base for bus resistors, since it keeps the bus impedance nice and low, which is good from a noise perspective. I think this design strikes a good balance between that and not loading the channel amps to the hilt.

I really can't thank you enough for taking the time to converse on this. It's truly remarkable in this day and age how folks around the world are able to share knowledge and insight with one another.

I'll be posting additional information as this thing continues to develop.

Thanks again!

-Bob

 

[abbey road d enfer]

There is a simpler solution if the bus are VE.

 

[bjoneson]

There is a simpler solution if the bus are VE.

In the back of my head, I thought something like this may work, though I'm not 100% certain how the math works on it. Since each bus is at zero volts, It appears like the opposite channels act as shunts in a divider for one another.

I'm going to have to try to wrap my head around it.

Thanks for the quick schematic!

 

[joaquins]

Ok, on your new schematic, add 2k resistor in series with your 4k7 resistor, before or after, it doesn't matter. Then use the spare switch on the other side to short it out when needed (when this other side is on)

That should do it.

Yes! That totally makes sense, and is *way* simpler that what I had originally put together. It also puts *way* less load on the channel amp.  I'm not too worried about "scaling up" for more accuracy on the attenuation.  If I'm calculating correctly this put's things at right about 3.1dB of attenuation, which is plenty close enough for me. I'd like to keep the 4.7k base for bus resistors, since it keeps the bus impedance nice and low, which is good from a noise perspective. I think this design strikes a good balance between that and not loading the channel amps to the hilt.

I really can't thank you enough for taking the time to converse on this. It's truly remarkable in this day and age how folks around the world are able to share knowledge and insight with one another.

I'll be posting additional information as this thing continues to develop.

Thanks again!

-Bob

You are welcome, it's always a pleasure to help someone in this community, I wait for updates.

There is a simpler solution if the bus are VE.

Saul Walker recommended not to leave attached resistors not in use to the sum bus, and grounded instead if possible, to avoid crosstalk and picking up noise, I don't know how much influence would it have on each case. It's an simpler way of doing so, I think the crosstalk wouldn't be a problem here since only one stereo bus is used, I probably move the switches after the resistors so they don't end up floating with one leg on the bus as an antenna. As I said, I don't know if this would make a difference at all, I'm pretty sure even less in a small project, maybe in a big mixer the sensitivity of this point is quite high to add anything useless to it which may add some noise.

JS

 

[bjoneson]

There is a simpler solution if the bus are VE.

I think I figured out the math on this...

So, we get 9dB of signal attenuation at the divider node created by the 4.7k, and the parallel impedance of the 2 5ks (2.5k)

From that point the the VE amp provides a gain of 6 dB (10k feedback against 5k bus resistor) netting us 3dB of attenuation.

When one bus is unassigned, the 4.7k and 5k appear in series against the 10k feedback netting us right near unity.

Did I get that right?

I'm still not positive this will result in the lowest noise / crosstalk,  but it is definitely simpler, no doubt.

And it definitely opened my eyes.

 

[abbey road d enfer]

Saul Walker recommended not to leave attached resistors not in use to the sum bus, and grounded instead if possible, to avoid crosstalk and picking up noise,

That is a strange recommendation, since grounding the resistors increases noise gain of the summing amp.  X-talk may be an issue but proper layout with ground plane and guards usually takes care of that.

 

[abbey road d enfer]

There is a simpler solution if the bus are VE.

I think I figured out the math on this...

So, we get 9dB of signal attenuation at the divider node created by the 4.7k, and the parallel impedance of the 2 5ks (2.5k)

From that point the the VE amp provides a gain of 6 dB (10k feedback against 5k bus resistor) netting us 3dB of attenuation.

When one bus is unassigned, the 4.7k and 5k appear in series against the 10k feedback netting us right near unity.

Did I get that right?

That's it.

I'm still not positive this will result in the lowest noise

Noise gain will be about 3dB higher.

/ crosstalk,

Indeed X-talk will be higher. For general stereo program it shouldn't be an issue. It would matter if you were making dual-track programs where one channel is different than the other, like radio quizz or bi-lingual.

 

[joaquins]

Saul Walker recommended not to leave attached resistors not in use to the sum bus, and grounded instead if possible, to avoid crosstalk and picking up noise,

That is a strange recommendation, since grounding the resistors increases noise gain of the summing amp.  X-talk may be an issue but proper layout with ground plane and guards usually takes care of that.

No, maybe I expressed myself wrong, or at least confusing. The preferred option was to switch the resistor at the bus end, rather than the feeding amp end, and if possible to tie it to ground. SO you switch between ground and the bus, always 0V at this point so any cap in there wouldn't see the change and no resistors floating around.

Leaving the unused resistors attached to ground and the bus would be a big error, working as shunt, increasing as you said the noise gain.

JS

 

[abbey road d enfer]

Saul Walker recommended not to leave attached resistors not in use to the sum bus, and grounded instead if possible, to avoid crosstalk and picking up noise,

That is a strange recommendation, since grounding the resistors increases noise gain of the summing amp.  X-talk may be an issue but proper layout with ground plane and guards usually takes care of that.

No, maybe I expressed myself wrong, or at least confusing. The preferred option was to switch the resistor at the bus end, rather than the feeding amp end, and if possible to tie it to ground.

OK, I understand. That has been seldom used, as it requires the switches to be very close to the bus connector. Failure to do so would leave the bus floating as an antenna across the PCB.
I know it has been done with CMOS switches.

 

[joaquins]

Saul Walker recommended not to leave attached resistors not in use to the sum bus, and grounded instead if possible, to avoid crosstalk and picking up noise,

That is a strange recommendation, since grounding the resistors increases noise gain of the summing amp.  X-talk may be an issue but proper layout with ground plane and guards usually takes care of that.

No, maybe I expressed myself wrong, or at least confusing. The preferred option was to switch the resistor at the bus end, rather than the feeding amp end, and if possible to tie it to ground.

OK, I understand. That has been seldom used, as it requires the switches to be very close to the bus connector. Failure to do so would leave the bus floating as an antenna across the PCB.
I know it has been done with CMOS switches.

He did recommended CMOS for that, could you explain how would be that failure? the bus open unterminated when no channels connected? Or just the trace going to the front panel hard switch if you don't use the CMOS switch?

Thanks.

JS

 

[abbey road d enfer]

He did recommended CMOS for that, could you explain how would be that failure?

Not a failure, but the intended goal to minimise capacitive coupling to the bus would be defeated with a PCB track (connected to the bus)  running across the board.

  Or just the trace going to the front panel hard switch if you don't use the CMOS switch?

Yes.

 

[JohnRoberts]

It sounds like you need a 3 position switch.

In general you want to keep the noise gain of the sum bus low.

regarding back grounding bus resistors  Yes, this reduces potential crosstalk (if grounds are clean) but increases bus noise gain so is a trade off.

I have used CMOS to route signals inside a console but they are not trivial so too complex** and expensive for low level console functions that don't need to be automated or switched simultaneously in a group. I used CMOS to switch an entire bank of modules between tracking and monitoring electronically with one master switch. 

JR

*** I used basic CMOS transfer gates but located them between the input/feedback resistor network and the op amp. This way  the nonlinear characteristics of the solid state switches are all but ignored. I couldn't measure any distortion using early 1980's bench test equipment, but not simple. 

PS: Thank you for spelling bus correctly.

 

[bjoneson]

PS: Thank you for spelling bus correctly.

I've been reading Doug Self's and aside from a brief rant on the spelling of "bus", there's a whole section devoted to bus assignment schemes.

In general he recommends keeping the bus resistors on the channel amp side of the switch, and shorting to ground when not in use. By switching the resistors off the bus when not in use, noise gain minimized by the ACN amps.  And it keeps a constant / consistent load on the channel amps.

By grounding the bus resistor at the channel when not in use, there's no signal to create capacitive coupling at the switch.  Though as was mentioned earlier, poor pcb layout can defeat the advantage, since you essential have a wide open bus wire running around the pcb when the busses are unassigned. Care needs to be taken to ensure it isn't run in close proximity to any "hot" channel traces.

In the design I think I've landed on (shorting 2k in series with 4.7k for attenuation), there is still a hot signal on the 2nd pole of the switch from the opposite divider, but I believe that signal would be attenuated  and therefore would not create significant crosstalk. It also likely depends on the physical construction of the  switch as to the proximity of the 2 poles.

Thanks again for all of the fantastic feedback and discussion. Little by little I feel like I'm learning and being able to have this type of dialog is really helping me understand the practical implementation of a lot of the theory.

 

[joaquins]

Thanks for the extra information, it does makes sense now. The only problem I still see in this approach is the extra load in the channel amp, since it will be always loaded as all the bus are assigned, giving extra current, higher THD and maybe the need of using higher values of summing resistors than preferred, resulting in higher noise. This is a problem when you have many bus and you probably end up using no more than 2 or 3 assigned to a single channel. If you have 24 bus mixer I see this could be a problem.

Then, could I use a guard trace around the resistor and switch pins that will be floating instead of grounding so I prevent crosstalk and the other undesired effects? I would still consider the case all the resistors connected but only as a worst case rather than the normal condition, so the amp won't clip prematurely when all the bus are assigned but higher THD would be the only concern.

JS