990 Based ACN Summing Amp

[bjoneson]

Hoping for a cursory look and any ideas / suggestions on a 990 based summing amp. I've been at this off and on for about a year learning and think I'm ready to start prototyping. I put this together based on information and examples provided in the 990 spec sheet: http://www.johnhardyco.com/pdf/990.pdf

The basic circuit is outlined in Figure 6 of the spec sheet, with a 10k feedback resistor (assuming unity gain with 10k bus resistors), and a 91pF cap in parallel I believe to combat oscillation.

The non-inverting side of the 990s are attached directly to ground, with DC offset compensation provided by a servo circuit using an OP97FP. The servo circuit was also lifted from the 990 spec sheet.

Gain control is provided by a dual 10k Pot.

Lastly the output stage is provided by a 600:600 transformer. That particular model was chosen because it's low profile for PCB mounting.

If anyone sees any glaring issues, or something which doesn't function as intended, I'd really appreciate the input.

Many thanks!

Schematic attached.

 

[abbey road d enfer]

Wacky gain control.
Here the pots are shunting the input, which does not change gain because it's an inverting stage. Loading the input just increases noise.
Gain control could be achieved passively by putting the pots after the summing stage but you would need a buffer stage to drive the xfmr.
A best option is "active" gain control, by putting the pot, connected as a rheostat, in the NFB loop. A very basic implementation would be increasing the 10k NFB resistor to about 100k and connecting the wiper of the pot and the top across the NFB resistor.
There would be a risk of scratchy noise due to DC circulating in the pot. A DC-blocking cap is requisite.

 

[bjoneson]

Thanks for taking the time to look it over. I had a sneaking suspicion that gain configuration didn't make sense. As I look closer at the example I pulled from, it was not in a VE setup.

I did consider doing an attenuator after the gain stage, but agree this would require another stage to drive the transformer, some thing I was trying to avoid.

I'll take a look at incorporating something into the FB loop as you mentioned.

Really appreciate the feedback!

 

[bjoneson]

Ok... so hypothetically if I do something as proposed (schematic attached), gain is controlled in the feedback loop. If I understand correctly per channel gain is [rFB] / [rIN].

I'm planning value of rIN at 10k, therefore rFB @ 10k would produce "unity" gain per channel.

With the proposed design, rFB becomes the parallel resistance of the 100k resistor and the 10k pot correct?

Assuming a residual resistance of 20ohms on the pot, I believe this gives a per channel gain range as follows:

~20 (parallel resistance of "fully open pot" and 100k resistor)  / 10k = ~.0002
~9.1k (parallel resistance of "fully closed pot") / 10k = ~.91

The thing I'm not understanding is what the purpose of the 100k resistor in parallel is? What purpose does it serve other than to slightly reduce the overall resistance in the feedback loop?

As far as a coupling capacitor to avoid pot noise, that would go between the pot and the inverting input of the opamp? Obviously need to be careful not to put it in line with the DC return from the servo, correct?

Thanks again for any advice / assistance!

 

[abbey road d enfer]

The thing I'm not understanding is what the purpose of the 100k resistor in parallel is? What purpose does it serve other than to slightly reduce the overall resistance in the feedback loop?

Nothing if you don't add a DC-blocking cap.

As far as a coupling capacitor to avoid pot noise, that would go between the pot and the inverting input of the opamp? Obviously need to be careful not to put it in line with the DC return from the servo, correct?

Correct, and then, since the cap would break the DC FB, you need a resistor to restore it, with a value high enough to not interfere too much with the pot variation. It could be much more that 100k, depending on the opamp's input current.

 

[bjoneson]

I think I'm starting to track. Does the attached configuration eliminate the need for the 100k resistor, since the output of the servo circuit does not traverse the coupling cap?

 

[joaquins]

I don't know if the 100k is really necesary, since there is a dc servo to close the loop at dc, you have to be sure that at any gain the loop is closed for any freq, at low freq may get some troubles. The current in the input I think it's low enough to deal with the opamp, one option is to use a cap to couple the bus to the amp. The other, the one I'm using in my mixer, is to use DC servos at each source, so there is no (or as low as you want) DC voltage at the bus and the scratch noise shouldn't be a problem, I already used it with a couple of channels as a prototype and I didn't have troubles with that.

JS

PS:The problem with the DC blocking cap is that needs to be really big, or at lowest gains it will bump the LF, and the lowest gain when the amp is 'muted' the feedback will be the cap, and just that, the resistor in parallel and the servo will be much higher impedance, so much lower feedback. You will have some noise at low freq, (1/f till the feedback from the cap is as big as the feedback from the resistor or the DC servo). That freq must be well under 20Hz, and still the excedent noise is there. Of course you could hard mute the bus if that's the case, and if you are using it at very very low gains there is a problem in somewhere else in the chain.

 

[bjoneson]

PS:The problem with the DC blocking cap is that needs to be really big, or at lowest gains it will bump the LF, and the lowest gain when the amp is 'muted' the feedback will be the cap, and just that, the resistor in parallel and the servo will be much higher impedance, so much lower feedback. You will have some noise at low freq, (1/f till the feedback from the cap is as big as the feedback from the resistor or the DC servo). That freq must be well under 20Hz, and still the excedent noise is there. Of course you could hard mute the bus if that's the case, and if you are using it at very very low gains there is a problem in somewhere else in the chain.

The truth is the only reason I'm introducing variable gain control on the FB loop, is in lieu of adding an attenuator after gain stage, and the need for an additional opamp to drive the output. I would never actually run at extremely low gain. But I do think I understand what you mean about the different feedback paths, and non-linearity in frequency response as the gain changes.

 

[bjoneson]

Maybe simpler is better?

In the attached configuration, the only return paths in the FB loop are the 10k pot and the DC servo (allegedy low pass below 1Hz).

The DC servo should control any DC introduction regardless of feedback gain. I suppose I'm just left with potentially "scratch noise" on the pot, but wouldn't that blow right through a coupling cap anyway?

 

[joaquins]

In fact the DC servo as it is would introduce even more scratch noise, it's working to have at the output 0V, we know at the + input there are 0V, so any offset voltage will be present in - input, creating scratch. If you put the DC servo to the + input you would avoid that, but you need to be sure that your sources have no DC, could be AC coupled with a cap, or a servo amp. I did the last one and worked fine, never loaded all the 16 channels yet but I think a serious problem would be heard from a single channel, at least a bit.

(those aren't non-linearities, is just frequency response or transfer function, we call non-linearities the things that are saturating, or adding some kind of distortion, level dependent. The only thing here would be to saturate the DC servo but we don't want that, or we would be in serious troubles)

JS

 

[bjoneson]

Gain control could be achieved passively by putting the pots after the summing stage but you would need a buffer stage to drive the xfmr.

I've been scouring the web for hours trying to answer this question on my own... but, can someone help me understand why the 990 output couldn't drive the 600 ohm output transformer with a 10k pot in line for signal attenuation without the use of a buffer? I feel like this is a fairly rudimentary concept, but I can't wrap my head around it.

 

[gridcurrent]

1.  the 990 has large output current capability so why not take advantage of it ?  use a 1:2 step-up output transformer.  think "free" gain.  you can  even consider revising the summing gain structure.
2.  if the output is transformer coupled, why bother with a servo?  if the output DC is low enough no capacitor is needed.  think API 312.  if there is more DC than you are comfortable with, use a capacitor as in API 325.  that method was employed in maybe a billion successful recordings.
3.  gain control.  as mentioned, put a rheostat in the feedback loop.

 

[bjoneson]

1.  the 990 has large output current capability so why not take advantage of it ?  use a 1:2 step-up output transformer.  think "free" gain.  you can  even consider revising the summing gain structure.
2.  if the output is transformer coupled, why bother with a servo?  if the output DC is low enough no capacitor is needed.  think API 312.  if there is more DC than you are comfortable with, use a capacitor as in API 325.  that method was employed in maybe a billion successful recordings.
3.  gain control.  as mentioned, put a rheostat in the feedback loop.

I think the only real concern about offset DC would be as it relates to headroom to the supply rails. I'm realizing as I type that the offset would need to be pretty significant to bring this issue into play (especially w/ 24v rails). My choice of using the servo circuit to provide offset compensation is based on notes provided in the 990 opamp application notes for use in a VE summing configuration:

"The non-inverting input may be grounded directly, or through a resistor. The
value of the resistor should equal the DC source resistance seen by the inverting
input, which is the parallel resistance of all the input resistors (assuming they are
not AC-coupled) and the feedback resistor (RIN and RFB). When both inputs of the
990 see identical DC source resistances, the output offset voltage will be the
lowest. This resistor can result in increased noise when compared to a grounded
input. This problem can be overcome by adding a capacitor in parallel with this
resistor. The capacitor has infinite impedance at DC, so the DC specs are
unchanged. The impedance is much lower above DC, so the noise performance of
the 990 is not significantly compromised. The value of the capacitor is not critical,
with 0.1µF being a good starting point. If the non-inverting input is grounded a
compensation current can be provided to the inverting input as shown in the M-1
application note. This provides the lowest DC offset at the output of the 990, the
lowest noise, and without the potential degradation caused by the capacitor."

 

[joaquins]

1.  the 990 has large output current capability so why not take advantage of it ?  use a 1:2 step-up output transformer.  think "free" gain.  you can  even consider revising the summing gain structure.
2.  if the output is transformer coupled, why bother with a servo?  if the output DC is low enough no capacitor is needed.  think API 312.  if there is more DC than you are comfortable with, use a capacitor as in API 325.  that method was employed in maybe a billion successful recordings.
3.  gain control.  as mentioned, put a rheostat in the feedback loop.

The problem with the DC offset, is that will change depending on the amount of channels assigned, and a little DC offset through the low resistance winding of an output transformer will result in a current that may be problematic getting higher THD at the transformer, not as a lack of headroom in the amplifier. In the API 312 the gain trim resistor has a DC blocking cap, no change in the DC impedance presented at the inputs, so you are just with the DC circuit as it is, the offset could be trimmed or at least know how much it will be and will be constant. On the other hand the transformer used in the 312 has M6 core, the one in the schematics 80% Ni, which is much easily upset with DC current. I wouldn't chose that exact transformer for this circuit, I would look for something with lower DC resistance on the windings, to get lower distortion when driven by a 990.

Actually the API 512 has both, a DC servo and a coupling capacitor, the DC servo is there so the reversing polarity switch doesn't pop, which did sometimes before it had the DC servo. The coupling capacitor is no longer needed but it was left there to make the change in the circuit as minimal as possible, so no costumer would complain about it. I think I would just put the switch after the cap, but they know what they are doing.

Gain control could be achieved passively by putting the pots after the summing stage but you would need a buffer stage to drive the xfmr.

I've been scouring the web for hours trying to answer this question on my own... but, can someone help me understand why the 990 output couldn't drive the 600 ohm output transformer with a 10k pot in line for signal attenuation without the use of a buffer? I feel like this is a fairly rudimentary concept, but I can't wrap my head around it.

The problem is that the output impedance of the pot is too high to drive the transformer, and you would be driving the transformer from the pot, not the 990, the 990 would just be driving the pot and the feedback loop. you would have to account for the worst case, which is in this case something of daily use at -6dB at the pot, where you will have 2.5kΩ output impedance from the pot, so pretty bad, against the recommended 50Ω for driving this transformer, or the output impedance of the 990 which may be around a few mΩ or even less.

The resistor on the + input is good to take care of the offset if all the inputs of the summing amp are connected all the time, and when not used grounded, this means noise gain is always the worst case. I do like the idea of the servo and gain rheostat on the NFB as you are going for since it allows you to get the minimum noise gain you need at that time, so the design in this way is really well optimized in noise performance, then you need good implementation and layout, but that's another story, which isn't really easy neither. The only problem comes with the low gains again, which make the amp to be loaded by a very low impedance on the NFB loop, but as you are expecting always line levels signals when you have low gain you expect low output level, so the current isn't any bigger and no problem, and if you have a bigger input, the 990 is a tough guy which can take under 100Ω without much trouble, if the output is loaded with 600Ω you still can have about 100Ω on the feedback loop and drive it to high output levels, probably the pot himself won't be happy when that happens.

In the 990 data sheet, is recommended for summing amp applications to use an OLI on the input, that's to prevent oscillations because of the capacitance on the bus, if you have all the summing resistors close, in the same board of the summing amp, that's not needed. If you have a long bus in a mixer it's probably good to have it there.

JS

 

[bjoneson]

Joaquin - Thanks for the thorough response. That really helps me tremendously!

 

[Rob Flinn]

Had you considered putting the 990's in the Classic API ACA board ?  It has the fader after the summing stage with a buffer & transformer ouput.    I used the 990's for summing & GA2520 for the buffer. I did that & they work very well in there, & it's easy to put together.  You get nice clean summing but with an API sounding output !

 

[bjoneson]

Had you considered putting the 990's in the Classic API ACA board ?  It has the fader after the summing stage with a buffer & transformer ouput.    I used the 990's for summing & GA2520 for the buffer. I did that & they work very well in there, & it's easy to put together.  You get nice clean summing but with an API sounding output !

I'll definitely look into it.

I honestly can't believe the amount of solid, pertinent feedback I've gotten here so far. I've posted in a few other communities, and haven't gotten near the quality responses over weeks I've gotten here in one day.

I'm excited to get into more DIY projects, and it's pretty awesome to have a resource like this.

Thanks for all the help / input.

I'm going to start prototyping some stuff over the next few weeks, and I'm sure I'll run into more questions as I do.

I try to do what I can to use existing resources to learn, but it's great to know there's folks around here qualified and willing to help out.

 

[bjoneson]

I was able to get the above circuit laid down on a prototype board, and fed a 1kHz tone through through a 10k resistor.

Gain behaves as expected, with pot in the feedback loop. With a scope on the output, you can see DC introduced when adjusting the pot, and the servo recovering over the course of a second or so.

At this point I'm pretty pleased with what I'm seeing.