what are OSI for this SSL_82E26 mix amp?

[gnd]

Thnx all.
I think balanced mix bus may be overkill in my case afterall. SSL mix bus seems quiet enough for my 24 channels. And I feel that ssm2210/NE5534 combination may be fine, I can compensate it manually as needed. And distortion is low. I tend to be more in favor of SSL mix bus clone at this moment.

Samuel:

To help the author of this thread with deciding on the proper topology we'd need to know if the number of channels (24 it is) is fixed or variable (perhaps I missed that information?).

Fixed 24. No switching on bus, just 2 bus output. All channels connected to bus from fader on. Channel on/off switch is prefader.


...

 

[JohnRoberts]

For any who may not be following the rationale for this question in typical negative (voltage) feedback amplifiers the stability compensation is strictly based on closed loop gain.

Specifically the higher the closed loop gain, the more the negative feedback is attenuated, allowing for proportionately higher open loop gain. The typical decompensated opamp (like 5534) is still only stable for closed loop gains of 3x or so and higher. In a dedicated application with fixed noise gain 25x for 24 input virtual earth summer we can do better.

FWIW a trick that can be used for consoles with bus assignment switches that must remain stable all the way down to unity gain is to back switch in a small capacitor to ground in series with the bus resistor, when a channel is de-assigned. This capacitor can be sized so the resistor only appears connected an octave above the audio band or higher. This will allow you to compensate the sum amp as if it was fully assigned even when it isn't, but without the audio band noise when lightly loaded.

The TRANSAMP approach as Sam advised is also friendly to variable assignment loading as the open loop gain tracks the closed loop gain for optimal performance no matter how many input are punched up.

For this specific application (fixed assignment load) just dial in the compensation for 25x.

JR

 

[Samuel Groner]

And I feel that SSM2210/NE5534 combination may be fine, I can compensate it manually as needed.

An important note with respect to compensation: T1 is shown as having no compensation capacitor (across pin 5 and 8) altough beeing used in a low noise gain configuration due to C28. The early Signetics NE5534 are known to show greater stability margins than current implementations so this worked well but may lead to instability nowadays. As solution you may need to use a 4.7 pF to 15 pF to compensate T1. Alternatively one could perhaps skip (or reduce) C28 for a fixed 24 channel configuration.

Samuel

 

[gnd]

Thnx Samuel, John, MT.

Samuel: As solution you may need to use a 4.7 pF to 15 pF to compensate T1. Alternatively one could perhaps skip (or reduce) C28 for a fixed 24 channel configuration.

Regarding compensation, what is the usual procedure to find out C values? Untill now I would feed 10kHz square wave into input, and then change compensation capacitors untill square looked as good as possible, but without ringing. Is this proper way to do it? Because I usually found that I needed to put in lower C values than were in schematics. So maybe I was overdoing it and going too high in my expectations?

What square frequency is usually used for tuning-in compensacion caps? At which square frequency can square start being rounded a bit (lets say 10% of half periood, or whatever the standard measure is)? Should it be almost perfect square up to 10kHz, or 100kHz, .... Or how long should be the rise time (rounded in-ramp) of square? Maybe 10us (10% of 10kHz square), or less? What are the usual expectations for quality console mix bus (or channel) in this regard?

thnx
gnd

 

[JohnRoberts]

There are perhaps multiple approaches that would work. I generally reduce compensation until a circuit oscillates, then add back a comfortable margin. If amplifier is exposed to a line driving the outside world, confirm that it remains stable with capacitive load on that line.

Indeed parts can appear stable with lower than expected compensation, but to insure stability under all conditions, after warming up, etc., I like to add in a margin of safety.

Specifically wrt stabilizing 5534 there was an alternate approach for low noise gains using a small feedback capacitor, which on the surface appears to work against stability but in practice swamps out input pin capacitance. There is clearly potential for interaction with network values used, and this isn't practical for many applications.

I'm not sure what to expect from the square wave response of that DOA circuit. The input LTP is degenerated which should give rise time limited response I prefer, at least for modest voltage square waves, however degeneration resistors are shunted with inductors so that degeneration may be limited in effect. Hopefully it will be well behaved for square waves.

JR

{EDIT] note: the signals seen by this summing amp will be bandpassed by channel electronics so it may be worthwhile characterizing that path's edge rates first or in combination with this exercise.

 

[gnd]

John.

I generally reduce compensation until a circuit oscillates, then add back a comfortable margin.

Looking at square wave, to what frequency would you expect square to look like square? 1kHz should surely be nicely square. What about 10kHz? Is 10kHz also supposed to be nicely square at scope, or can it start to get rounded? 100Khz square is probably too much?

Hopefully it will be well behaved for square waves.

First tests showed fine 10kHz square. I had some oscillation, because of too low caps, but fixed that.

the signals seen by this summing amp will be bandpassed by channel electronics so it may be worthwhile characterizing that path's edge rates first or in combination with this exercise

Yes, I did testing on one channel first, and I reduced two 22p capacitors. 10kHz square was heavily rounded before, but is square now. I'm still playing with new superbal input stage, it likes to oscillate. Anyway, I bought a packet od ceramic capacitors, all values from 1p to 100p, so I will play with that untill I find best values.

...

 

[JohnRoberts]

IMO the consideration for square wave "wave shape" is that rising and falling edge take on damped but properly exponential (rounded?) wave shape. How high frequency of a square wave reaches full amplitude should be limited by small signal bandwidth of audio path, not stability compensation.

JR

 

[JohnRoberts]

[quote author="mediatechnology"]BTW going back to the input protection diodes the SSM2210 has anti-zener connections as they are connected from base to emitter. But in the 82E26 (thanks for linking to my site BTW) they are base-to-base a la NE5534. Seems like the SSM2210 (LM394CN) configuration is a better approach to anti-zener.

[/quote]

Indeed when employing emitter degeneration (resistors in series) as is the case here it's possible to transiently experience Vth (differential input voltage) in excess of a diode drop without slew limiting. For example if applying an infinite rise time, rail to rail square wave, the output would take finite time to move and momentarily lag. The input differential would see the full input voltage, reduced by rest of input and feedback resistors. In this case the say 30V divided by the feedback factor of 25 would generate a slightly over 1V step.

In this scenario of rail to rail square wave, base to base connected diode protection clamps would conduct and limit the current available to charge the compensation cap to a fixed amount. That in turn would cause the leading edge of the output square wave to begin as a fixed slew limited ramp, not the preferred by me exponential curve. After the output has moved enough to reduce the input differential (via feedback resistor) to drop below the diode clamp, this rate of change will slow further and follow the familiar rounded appearance as it approaches the endpoint.

Therefore these base to base diode clamps will limit peak slew rate to a lesser level and result in different square wave or transient response for small signals that don't clamp and large signals that do (not optimal IMO).

This is IMO mostly academic, as I don't expect preceding channel audio path to deliver 30V infinite rise time transients, but this analysis suggests that little differences in how things are hooked up could matter.

Note: The inductors in parallel with emitter degeneration resistors while reducing audio band input noise will also affect this exponential transient rise time as instead of a simple RC (or delta V divided by R times C), it's a resistor in parallel with an inductor feeding the compensation cap.

Sorry if this is TMI, but a longwinded agreement with Wayne's comment. Diode clamps reverse b-e are more betta.

JR

 

[Svart]

I'm interested in what you guys think about using JFETs on the frontends of summing circuits..

In an experiment some time ago I breadboarded up some FETblokes, their associated feedback(non INV.) with a trimmer for gain matching. I AC coupled them(bipolar MUSE) and replaced my entire mix amp in the master section of the console with these. I literally took out the discrete BJT frontend and opamp backend and all supporting parts. I simply wired in the fetbloke assembly, fed noise in the unit and trimmed until I clipped the next stage and then backed off a few DB.

results: more headroom, cleaner sound and ~10db better S/N..

I didn't change anything else. In fact I was shocked. To this day I haven't bothered to figure out the what or why I just use it and love it.

opinions?

 

[JohnRoberts]

FET inputs are usually a mixed bag of pros and cons. The reduced transconductance allows for less compensation capacitance and higher slew rates. OTOH DC offsets are typically much worse than equivalent bipolar devices. Historically FETs were higher noise than bipolar, but there are some nice Toshiba parts and perhaps more that deliver 1 nV/rt. Hz which is more than adequate for summing amps.

Your experience of 10 db better S/N with FET suggests both use of good low noise FETS and the former summing amp was substandard or faulty.

JR

PS: I am not familiar with what a FET "Bloke" is so ASSume for this discussion it is a discrete opamp or equivalent.

 

[Svart]

Hey John, yes it's a DOA, sk170 front end, CFB design using darlington outputs. It's very simple and has about 200v/us slew. Being CFB, you can't use a feedback cap or the pole created turns the whole thing into an integrator mess..

I've worked a lot with CFB opamps and tend to choose them over VFB for a lot of things.

the circuit that was replaced was a typical 2x 2n440X(i can't remember if it was 4401 or 4403) with a simple 5532 following. nothing more than a cheap preamp stage with set gain. I always found that when monitored before this stage the audio was much more pleasant sounding then after the stage, thus the quick and dirty trial of the fetbloke replacement.

As for DC offsets, I took this into consideration before I installed the stage and used nichicon muse bipolars before and after the stage. I later found that I could do without the input coupling caps but in other testing I found that these caps were so good that I decided to leave them in for nothing more than lazyness and the potential safety mechanism they provide.

In fact, I am working on a design for a summing bus made entirely from this DOA design to fit my console, spurred on by the discover I mentioned earlier. I bet it trumps most anything else made for it's simplicity and for it's ability. I think I made a thread about it some time ago but I don't seem to have a link for it.. I should really get back to work on it though..

 

[JohnRoberts]

Indeed a 2sk170 will spank 2n4401/4403s, which I personally never found very useful for low noise design. I used them as inexpensive GP transistors.

I used a similar 2sk117 in an old phono preamp and it was a sweet part. 170 may be newer/slightly better, but much improvement at this point is hard to come by.

I find current vs. voltage feedback, six of one half dozen of the other. Each have their merits. As mentioned before if switching in a changing number of channels to the bus, current feedback could be useful to optimize loop gain margin.

JR