@KSTR i have done a similar CM servo on a Av=2 differential non-inverting line output. Good idea to put it around U1 here. Am I correct that you mean to match Rf1 with the injection resistors, to give the servo full authority? I’m picturing a buffer with two sample resistors off the U1 outputs. The minimum gain is no issue, just want to clarify before I draw the thing (with the center tap Rg2 as well).
oooooookay i have seen the CMR curve and now i think i know what a “0V servo” is. not a CM buffer. you mean inverting input to ground, right? and compensation is a minimum stability cap from output to ground?
Show ‘em if you got ‘em!
Output section does not have to be paralleled 1602s. OPA1622 with two 1Ks and a 2K in the middle will produce 1.4mV offset, which is fine for most folks.
Unloading the output of the composite should be advantageous regardless of what CFA (or VFA) is in-loop. The noise advantages of avoiding differential amps entirely are very enticing. There is something to be said for AC coupling the output section, ditching the DC servos, and using OPA2211 in the master position - small value PPS caps, Mohm bias, and el cheapo OPA1679 could be an option as the line driver.
Obviously if there are no 600 ohm loads to drive, the output amp can be most anything.
I learned some things from sim:
Output section does not have to be paralleled 1602s. OPA1622 with two 1Ks and a 2K in the middle will produce 1.4mV offset, which is fine for most folks.
Unloading the output of the composite should be advantageous regardless of what CFA (or VFA) is in-loop. The noise advantages of avoiding differential amps entirely are very enticing. There is something to be said for AC coupling the output section, ditching the DC servos, and using OPA2211 in the master position - small value PPS caps, Mohm bias, and el cheapo OPA1679 could be an option as the line driver.
Obviously if there are no 600 ohm loads to drive, the output amp can be most anything.
I learned some things from sim:
- you still have to have thermally coupled resistors of .1% or better for >60dB CMR at low gain.
- the sample resistor matching and the per-side Rf/Rg matching are of equal importance
- one side can have wildly different values than the other side without affecting CMR, as long as the ratio is consistent (i used 980R/490R on one side and 1020R/510R on the other side for this)
- almost 80dB CMR at low gain is attainable with an input CM offset of 20R and 100K bias resistors. 80dB CMR at high gain is not unreasonable to expect.
- CMR is generally 3dB to 4dB better with higher injection values, hence the 2K/1K matched packages per side rather than all-1K. there are diminishing returns with 5K/1K, maybe another 1dB to 1.5dB. also the 2K/1K choice enables a lower minimum gain.
- HF CMR is limited by the OLG of the servo at the wrapping amp, diminishing by >6dB at 20kHz even with an OPA211 in the servo position (45MHz).
- add another servo at the in-loop amp, and CMR is rail flat out past 20kHz. so much so, in fact, that you can use an OPA2210 for the two servos (18MHz) and the CMR curve only just begins to climb at 20kHz. OPA1602 (35MHz) is specified for cost purposes but there is a lot of flexibility in amp choice here.
- matching around the in-loop amp’s servo only affects HF CMR (within reason)
Last edited:
In the cost-and-Iq-no-object department, one could really stay on-brand here by making the output section an OPA1612 with a CFA in-loop, and for that one could simply use a THS6002, the quad that integrates a TPA6120A2/THS6012 and a dual with the same 1.7nV/rtHz noise spec but less current — still more than enough for audio line driving purposes. In that event, Rf and Rg for the output composite could be super low.
Thanks for posting your schematic and details, appreciated.
I'm still not certain if the interaction of the slave sections does anything useful. My idea (see below) is to make the composites individual black boxes and use the CM servo and differential gain only on the masters.
As for the cost-and-Iq-no-object department ;-) I'd suggest going crazy and shift rails and local GND of the individual masters and slave GND refs, from copies (fast buffers, distortion doesn't matter much) of the individual inputs.
By this, each master opamp has its inputs and output all "nailed down" at half-supply no matter the signal, from the opamp's POV, just like if it were an inverting config. No additional distortion from input CM or output voltage swing...
For fun, the servo compensation is two-pole, here. All opamps are are simple 100Mhz single pole idealized.
There is one drawback, the servo has to swing twice the input CM voltage.
Compensation for a huge range of gains might be a challenge. The servo itself is a composite, with the original master opamps now being the paralleled slaves at gain -1 to highest possible frequency (hence, no local compensation caps, rather I tried to isolate stray capacitance for a more complex gain setting network with small air-core inductors in this sketch). The slaves must be able to run a noise gain of 2x flawlessly under all conditions. Clipping behavior, again, is critical.
I'm still not certain if the interaction of the slave sections does anything useful. My idea (see below) is to make the composites individual black boxes and use the CM servo and differential gain only on the masters.
As for the cost-and-Iq-no-object department ;-) I'd suggest going crazy and shift rails and local GND of the individual masters and slave GND refs, from copies (fast buffers, distortion doesn't matter much) of the individual inputs.
By this, each master opamp has its inputs and output all "nailed down" at half-supply no matter the signal, from the opamp's POV, just like if it were an inverting config. No additional distortion from input CM or output voltage swing...
For fun, the servo compensation is two-pole, here. All opamps are are simple 100Mhz single pole idealized.
There is one drawback, the servo has to swing twice the input CM voltage.
Compensation for a huge range of gains might be a challenge. The servo itself is a composite, with the original master opamps now being the paralleled slaves at gain -1 to highest possible frequency (hence, no local compensation caps, rather I tried to isolate stray capacitance for a more complex gain setting network with small air-core inductors in this sketch). The slaves must be able to run a noise gain of 2x flawlessly under all conditions. Clipping behavior, again, is critical.
whoa, that is wild that i utterly misunderstood you and still came up with something functional.
is the servo acting on everything below 800kHz? one of those poles is ~16MHz and one is 20x lower if I read correctly. it’s nice to see an example of a two-pole being done in a single feedback network; i have heard of such things but typically the one-pole-per-input method is all i draw.
i dunno if your drawback is a concern - input CM is uV range for both AC and DC. input is AC coupled because phantom power. the only high CM I can think of would be an AC power line running in parallel with the mic cable.
i’d love to see the whole composite so i can better grok where you’re going with this. i can see why you want to separate the slaves.
is the servo acting on everything below 800kHz? one of those poles is ~16MHz and one is 20x lower if I read correctly. it’s nice to see an example of a two-pole being done in a single feedback network; i have heard of such things but typically the one-pole-per-input method is all i draw.
i dunno if your drawback is a concern - input CM is uV range for both AC and DC. input is AC coupled because phantom power. the only high CM I can think of would be an AC power line running in parallel with the mic cable.
i’d love to see the whole composite so i can better grok where you’re going with this. i can see why you want to separate the slaves.
hey, you’re on LTspice…maybe plug in an ADA4898-2 for the master and an AD815 for the slave? i’ve got peak output from the master down around 1.3V (26Vpk / 2 / 10), so the 4898-2 bandwidth would probably be unencumbered if we’re talking about the same gain staging. or a coupla 797s if i’m barking up the wrong tree.
ADA4898 for the servo, or would an 8597 do?
ADA4898 for the servo, or would an 8597 do?
Would you mind recommending a current-and-voltage limiting approach for my v2 schematic? If I understand correctly the concern is if the user is not listening to the output (or it’s shunted) and the level at the input is so high that the output section has already been driven >6dB into saturation? This is most likely to happen if a user mistakenly plugs a line output into the mic preamp input.
I’d alternately take a link to a clipping mitigation primer for composites if that’s easier. I’m guessing it would start with a specific value for the buildout R after U1, and continue with a zener something or other in the U1 local loops?
Despite lack of a specific CFA choice at this juncture, it’s fair to say (even if there is no Av=2 output section and the composite is the line driver with Riso directly following) that there is no way either CFA in the composite will ever be asked to put out more than 13Vpk, and if that needs to be 12.5Vpk (24dBu differential) then so be it. From my research it seems that all the likely slave candidates are capable of at least 12.5V on 15V rails.
If i’m thinking about this wrong, please do correct me.
Last edited:
Here is Jan’s post with the complete circuit for his Groner/Polak DIP project. I think limiting the master amps to 650mV would be fine (1.3Vpk x 10 x 2 = 26Vpk), or 1300mV if there’s no Av=2 output section (2.6Vpk x 10 = 26Vpk). I’m assuming that is done with a specific resistor value in place of the 1K. Some people would gravitate to a soft clip circuit because it’s a mic amp and that is considered cool. I think a hard clip at maximum value is best for headroom.
A quad diode package exists for such an event, the BAV99S.
Original note:
Didden schemo:
A quad diode package exists for such an event, the BAV99S.
Original note:
Didden schemo: