New Discrete OpAmps

[Ptownkid]

sounds good to me, no pun intended.

 

[clintrubber]

First of all, thanks for sharing !

Your design looks great for a conventional (virtual ground inverting) console summing bus amp with a lot of inputs,

If this is indeed in relation to the second opamp design with the complementary inputs (the thread-progress at least suggests so),
then something feels a bit strange.
Aren't we mainly using compl. inputs to accomodate large input-swings ? (which doesn't happen in 'virtual ground summing').

IIRC using complementary inputs (espec. when indeed not really needed) can introduce more hassle than is avoided.
Details may lead to a different final conclusion for a certain situation, but that's at least how I've remembered it: don't use a complem. input when you don't need it.

Regards,

  Peter

 

[JohnRoberts]

First of all, thanks for sharing !

Your design looks great for a conventional (virtual ground inverting) console summing bus amp with a lot of inputs,

If this is indeed in relation to the second opamp design with the complementary inputs (the thread-progress at least suggests so),
then something feels a bit strange.
Aren't we mainly using compl. inputs to accomodate large input-swings ? (which doesn't happen in 'virtual ground summing').

IIRC using complementary inputs (espec. when indeed not really needed) can introduce more hassle than is avoided.
Details may lead to a different final conclusion for a certain situation, but that's at least how I've remembered it: don't use a complem. input when you don't need it.

Regards,

  Peter

I don't mean to hijack this thread. Yes #2- Complementary inputs can also be looked at like paralleling a pair of differential inputs so you get the input noise voltage benefit of paralleled devices (noise voltage drops 3dB, noise current increases 3 dB), but with a first order cancellation of some errors. Input bias current is an obvious one. I just like the symmetry of the approach, for no other reason than it feels right. The .5nV/rt Hz will be respectable for the typical summing bus application with high noise gain.

Virtual earth summers in real world consoles really do need to be proper differential amps because ground in a 6' long chassis is a concept not a voltage node.

JR

 

[clintrubber]

I just like the symmetry of the approach, for no other reason than it feels right.

Topology sure looks good, fully agreed. But I guess it's the same as with pretty women,
good looks may&will come with hidden problems 


The obvious potential problem with compl. input stages is the variation of gm over the CM-input-range.

http://ecad.tu-sofia.bg/et/2005/pdf/Paper155-D_Dimitrov.pdf :

The straightforward method for achieving a rail-to-rail input stage is to connect in parallel
n-channel and p-channel differential pairs. It is highly desirable that the circuit
performance is independent of the input common-mode (CM) voltage. The problem
with the complementary input stages is that in the middle of the CM range, both the P
and the N part operate and the total transconductance is twice as high as the value
when the CM voltage is close to any of the rails and only one of the pairs operates.

Certain designs stabilise this variation:

http://www.imec.be/esscirc/essderc-esscirc-2003/papers/all/132.pdf

http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/4/27322/01214729.pdf?arnumber=1214729


Note that in case of an uncompensated gm-variation over the CM-range it doesn't
matter much for a virtual earth summer since the CM-variations are small/absent.
I initially thought there was another mechanism ongoing, but I recalled that wrongly; instead it was this gm-variation.

But then again there's little need for a rail-to-rail input capability; most discussed benefits can still be had by 'conventional' means.

Regards,

  Peter

 

[bcarso]

This is somewhat analogous to the problem with "rich" class AB output stages, as explicated by Self.  He suggests going class A all the way if you are going to be that concerned with crossover distortion, if I've read him corrrectly.

In Samuel's case, though, the gm of the bipolar complementary input pairs doesn't vary that much with CM swing---nothing near like the CMOS structures of R-to-R amps.

 

[jdbakker]

Very nice designs, some food for thought there (as posted elsewhere).

Samuel, do you know of a good source for the FPN560/660? None of the usual suspects have them in stock.

Is the relatively slow MJE1x0 output stage stable with inclusive Miller compensation? I'd be worried about that, especially with capacitive loading on the output.

JDB.

 

[JohnRoberts]

I just like the symmetry of the approach, for no other reason than it feels right.

Topology sure looks good, fully agreed. But I guess it's the same as with pretty women,
good looks may&will come with hidden problems 


The obvious potential problem with compl. input stages is the variation of gm over the CM-input-range.



Note that in case of an uncompensated gm-variation over the CM-range it doesn't
matter much for a virtual earth summer since the CM-variations are small/absent.
I initially thought there was another mechanism ongoing, but I recalled that wrongly; instead it was this gm-variation.

But then again there's little need for a rail-to-rail input capability; most discussed benefits can still be had by 'conventional' means.

Regards,

  Peter

Indeed there is no r-r or significant CM swing requirement for a virtual earth summer, but the first order cancellation of input bias current can be useful when using very low Rbb devices biased up for significant current density.  Since there is a noise cost to using identical impedances at + and - inputs, using conventional topology would just about require an added servo or trim to deal with DC errors.

And it looks pretty... Pretty women can be smart... the only difference is they don't have to.

JR

 

[Samuel Groner]

Do you know of a good source for the FPN560/FPN660?

I got mine from Mouser some time back; I'm not sure if these parts are even discontinued too..?

Is the relatively slow MJE1x0 output stage stable with inclusive Miller compensation? I'd be worried about that, especially with capacitive loading on the output.

It is stable for all tested conditions without substantial capacitive loading. Indeed the later improves with standard compensation, but for most practical audio applications stability appears to be sufficient. I'm more concerned about the local stability in the SGA-LNA-1 where there are two followers within the loop. In fact there is some HF-wobbling there during fast transients which makes me think if perhaps I should use standard compensation. There's no resulting obvious instability though.

Samuel

 

[clintrubber]

Virtual earth summers in real world consoles really do need to be proper differential amps because ground in a 6' long chassis is a concept not a voltage node.

Indeed there is no r-r or significant CM swing requirement for a virtual earth summer, but the first order cancellation of input bias current can be useful when using very low Rbb devices biased up for significant current density.  Since there is a noise cost to using identical impedances at + and - inputs, using conventional topology would just about require an added servo or trim to deal with DC errors.

I'm not after de-rail-to-railing this thread either, but one could then say that the mentioned requirement for a proper diff-summing-amp also realizes the identical source-impedances...  shifting the previously more required first order cancellation of input bias current of the compl.input now towards the nice-to-have category.

Regards,

  Peter

 

[Ptownkid]

Are there any subs for the LNA that are a little easier to get?

 

[JohnRoberts]

I'm not after de-rail-to-railing this thread either, but one could then say that the mentioned requirement for a proper diff-summing-amp also realizes the identical source-impedances...  shifting the previously more required first order cancellation of input bias current of the compl.input now towards the nice-to-have category.

Regards,

  Peter

In my experience no. In differential (not balanced) bus topology the ground side of the differential bus is often executed using very low impedance resistors to keep johnson noise down. The relatively large impedance difference between + and - inputs would generate an error voltage with significant bias current.

Since you are basically connecting one bus of the differential to a bunch of local grounds, driving low impedances to that bus is not a concern.  For input noise considerations, this parallel non-inverting input R is added to the R of the inverting side. For a true "balanced" bus topology using equal values Rs you could see roughly 2x total R and noise current contribution could then become more of a factor. 

Of course there are multiple ways to skin any cat.. so YMMV.

JR

 

[Samuel Groner]

I've updated the first post to include the third design.

Are there any subs for the SGA-LNA-1 that are a little easier to get?

Difficult. For Q1-Q4 2SC4547E/2SA1085E (or any of the lower voltage versions) come to my mind but I'm not sure if they are easier to source. For the output stage there should be more choices, check for sufficient power dissipation (i.e. same tall TO-92 package) as well as similar or higher fT and hFE.

Samuel

 

[BradAvenson]

Maybe look at the TN6717 and TN6727 parts.  I think the FPN560/660 parts have been discontinued. 

 

[Ptownkid]

Thanks Brad, I'll do that

 

[jdbakker]

Maybe look at the TN6717 and TN6727 parts.

Those have much lower hFE than the FPNx60 parts, and possibly a lower fT as well. This may make a difference in DOA performance, especially with the inclusive Miller compensation.

JDB.
['s not my circuit, but other than the smaller footprint I see little advantage in the TN67x7 parts over the more traditional MJE1x0 outputs]

 

[bcarso]

Someone in here (sorry to have forgotten---was it Ed Anderson?) was kind enough to send me some samples of a Sanyo comp pair that might be suitable if you can get your hands on the high beta bracket ones.  Fast as blazes, and the constancy of beta with collector current (often described improperly as "linear beta"), until you run into current-crowding effects, is impressive.

The P/Ns are 2SC3596 and 2SA1402.  I'm not sure how hard they are to get.

 

[thermionic]

I've updated the first post to include the third design.

Are there any subs for the SGA-LNA-1 that are a little easier to get?

Difficult. For Q1-Q4 2SC4547E/2SA1085E (or any of the lower voltage versions) come to my mind but I'm not sure if they are easier to source. For the output stage there should be more choices, check for sufficient power dissipation (i.e. same tall TO-92 package) as well as similar or higher fT and hFE.

Samuel

How about Toshiba 2SA970 and 2SC2240 ? They can be found easily from RS in 'Magnatec' guise.

http://www.toshiba.com/taec/components2/Datasheet_Sync//50/6451.pdf

http://www.toshiba.com/taec/components2/Datasheet_Sync//50/6461.pdf


Justin

 

[tv]

>>>>>> The relatively large impedance difference between + and - inputs would generate an error voltage with significant bias current.


But this could be remedied with feeding the + ("active" ground) input via additional capacitor and having equal impedances (ok.. resistances) on both + and - (summing node) opamp inputs, right? Assuming we have bjt-input opamp....

 

[JohnRoberts]

>>>>>> The relatively large impedance difference between + and - inputs would generate an error voltage with significant bias current.


But this could be remedied with feeding the + ("active" ground) input via additional capacitor and having equal impedances (ok.. resistances) on both + and - (summing node) opamp inputs, right? Assuming we have bjt-input opamp....

Yes, for the case of permanent assignment loading. If individual inputs are switched onto the bus it gets more complicated. I guess you could cap couple all the low Z sends, and the low Z R in master, but this is messy, and a lot of parts.

JR

PS: My apologies again for the hijack.

 

[jdbakker]

I've updated the first post to include the third design.

I missed that. Another very interesting design; thanks for sharing.

Two questions:

- why did you reduce the VAS emitter decoupling capacitor so much (HVA C2:220p vs SOA2 C1:1n)?
- is Q8 included just to tame clipping behavior (cf this thread), or is there some other/additional reason ?

JDB.
[still tweaking; presently trying to decide whether some bias current drift is tolerable if it reduces amplifier noise]