A new discrete current-feedback op-amp design -- RFC

[jdbakker]

EDIT 20090831:

Revision 1.2 uploaded. Seven transistors eliminated at the cost of ~1dB 3rd rise at 1kHz (but still ~-170dBc), otherwise no major spec changes.

EDIT 20090828:

Revision 1.1 uploaded. 3rd and higher are now 20dB lower, otherwise no major spec changes.

Original post:

[now it's ready to be posted]

Hi all,

I've finished a first draft of the current-feedback op-amp that is to form the first stage of my discrete mic pre work-in-progress.

Have a look at the schematics. (EDIT 20090828: this is the original version).

Preliminary specs:
(running on +/- 24V supplies, driven by a 100R source, into a 100R load, set for an amplification of 21x (~26dB))
Noise figure: ~1.5dB
Dynamic range: ~130dB between 20Hz-20kHz noise floor and clipping
Output swing: +/-21V without load, +/-20V with 100R load
Slew rate: +40/-110 V/us
Distortion: rhymes-with-lies claims that with a 500mVRMS input signal at 1kHz the 2nd is at 128dB below the fundamental, and the 3rd and higher are 150dB down. At 10kHz those figures are 109dB and 130dB, respectively. I expect that in real life noise pickup and thermal distortion will dominate (never mind that I have no equipment to measure 300nV harmonics in the presence of a 10V fundamental...)
Frequency response: 3dB down (rel 26dB) at 6MHz, unity gain at 24MHz. Apart from 0.4dB in-band gain peaking the AC response is monotonically decreasing to 1GHz (I looked no further, and I expect anything over a few tens of MHz is going to be dominated by parasitics)
Overshoot when the amp is slew-limited is about 30% of the step response; recovery is within 1us. Negative clipping shows 300mV overshoot; no phase inversion or other funny business.

And yes, that's enough transistors to build two or three traditional DOAs.

My main problem in designing this monster was handling negative clipping. Contrary to voltage feedback designs where you know that the maximum current into your VAS stage is set by the differential pair tail current source, a current feedback op-amp will happily dump as much current into your negative input (and thus the VAS) as your feedback network impedance allows. For this design, that's about 100mA in differential configuration. I had to implement a three-stage clip control: D18/D20 work to keep Q4 out of saturation, with D20 bootstrapped by R6. Similarly, D21/D22 are a Baker-ish clamp for Q16, and for really serious overloads the drop across R11 drives input transistor Q1 into saturation.

This brings me to one of my worries: when Q1 saturates, its base current increases, changing the amplifier's input impedance. This may lead to...funky responses in a reactive source. How much good is it to build an amp that clips gracefully if it produces, say, ringing in a mike/cable assembly? Worse, what'll happen to a ribbon mic? Granted, for this to happen the ribbon mic would have to somehow send a >5V signal down the line, but still. I need your input here; if this is a real problem I could implement a simple current limiter around the input transistor, but that does increase distortion some.

I have considered a current source in place of R7. While this increases VAS impedance and open-loop gain and lowers distortion even further, higher order products become relatively more present. Besides, an active current source will have more noise than the present solution.

Samuel: thanks for the hint on getting better output Z for cascodes. Like you've said elsewhere VAS output current swing has a strong impact on VAS distortion, and your suggestions have helped a lot in that area.

The 2N4403 on the input is just a placeholder. If you feel you need that last bit of noise performance a 2SA970 or 2SB737 should work fine. I might try a FET to see what happens. For the rest of the small-signal transistors I would recommend the BC550C/560C over the '547C/557C.

The output drive is deliberately beefy to allow for a low-Z attenuator between this stage and the next one. There's no overcurrent protection yet, but I'm not sure that it's needed for something that always drives a known load (plus I'd be loathe to add two more diodes, that would make this design too complex).

And yes, that's a lot of diodes, but I've pencilled in duals in SOT-23. These are twice as fast to place (as there's only half as many), plus getting the orientation/polarity wrong on a 3-pin part is much harder than on a 2-pin part.

Like in regular power amps, I am considering feeding the output transistors from an unregulated supply.

I need to tweak the biasing for the cascoded diamond buffer some. Apropos cascodes: I've used them as much to isolate high-dissipation positions from transistors where the actual VBE matters (ie: the current sources). More apropos cascodes: I've found that it's time to stop tweaking and start publishing/building when you get the urge to start cascoding your cascodes.

Thoughts? Suggestions? Flames?

JD 'but where to buy that 1kF capacitor?' B.

 

[jdbakker]

...and I've spotted the first bug. I know why SPICE didn't complain, but I should have caught this sooner. Oh well, sleep first, revised version later.

JDB.

 

[Samuel Groner]

Thanks, great to see this design!

I expect that in real life noise pickup and thermal distortion will dominate.

I'd be surprised if with all those cascodes thermal distortion would be particularly significant; however, (standard) SPICE does not model voltage dependence of Early voltage. At these levels we're considering here this will probably be the dominant error source in distortion simulation (I've seen cases where real live showed 60 dB higher distortion due to this). While in theory Early effect in Q4/Q5 is well cancelled with the arrangement shown transistor mismatch will considerably limit the improvement.

A couple of remarks after a quick look:

* The seven-diode string level shifter imposes a considerable temperature coefficient for Q1 collector current. I'd consider a zener instead, which is also a more elegant solution with respect to parts count.
* While the output-Z enhancement for Q4 is IMO an appropriate solution I'd use an emitter referenced cascode for Q6; this is both simpler (3 transistors less) and there is no transistor matching need.
* To get anywhere near the theoretical impedance levels at Q4 collector you need a tripple emitter follower for the output stage. Skip Q10/Q15/R14 and use a PNP follower driven from Q20 emitter to drive Q11/Q12. Both simpler and better--should not have much effect on stability.
* I'd consider the use of better output power transistors. There are a couple of not that difficult to get parts which have much higher hFE and fT.

Samuel

 

[jdbakker]

Samuel, thanks for your feedback.

* The seven-diode string level shifter imposes a considerable temperature coefficient for Q1 collector current. I'd consider a zener instead, which is also a more elegant solution with respect to parts count.

As drawn a 10 degree temperature rise lowers Q1 current by 3%, if I replace the diodes by a 3V9 or 4V3 zener this is cut down to 1%. I mostly wonder how much impact even a 3% current decrease has on the performance of the amp. Noise figure, hFE or fT will not be changed noticeably, input/output offset will change some but not enough to worry about. Am I missing something here? While I agree that on the face of it replacing 4 double diodes with one zener is a definite decrease in parts count, I already use the double diode type elsewhere so on the whole build time and parts procurement effort are likely to go up some. I'm willing to do that if it's worth it, of course.

* While the output-Z enhancement for Q4 is IMO an appropriate solution I'd use an emitter referenced cascode for Q6; this is both simpler (3 transistors less) and there is no transistor matching need.

Which three transistors would I be able to eliminate? I still need Q2/Q8/Q7 to bootstrap/cascode the diamond buffer input pair (which increases input Z of that stage by a factor of 10); the best that would happen is that D15/D16 could go, but I fear that adding a separate bias network for the emitter referenced cascode will cost more parts on the whole.

* To get anywhere near the theoretical impedance levels at Q4 collector you need a tripple emitter follower for the output stage. Skip Q10/Q15/R14 and use a PNP follower driven from Q20 emitter to drive Q11/Q12. Both simpler and better--should not have much effect on stability.

I'll look into that, but I'm not sure how much the gain is. When driving 30Vpp into an 100R load @1kHz, the largest load on the collector of Q4 is the compensation network drawing 10uApp which is thirty times more than the current into the base of Q20. Besides, wouldn't I still need Q10/Q15/R14 to bias Q12?

* I'd consider the use of better output power transistors. There are a couple of not that difficult to get parts which have much higher hFE and fT.

I'm very open for suggestions there. I had a brief look at 'better' devices in TO-126/TO-225/TO-220 but couldn't find any that clearly beat the familiar MJE17x or BD13x parts.

By the way, the error I spotted last night is in Q21, which is now pushing Q20 dangerously close to saturation. I have a few obvious fixes which I'm planning to sim when I get some time.

Thanks again,

JDB.

 

[Samuel Groner]

As drawn a 10 degree temperature rise lowers Q1 current by 3%.

True, first thought the effect would be more pronounced.

Which three transistors would I be able to eliminate?

I mis-read the schematic--did see Q11/Q12 as diode-connected transistors. So you actually have a triple follower already. If you'd like to reduce the complexity I'd recommend that Q11/Q12 is converted to a single-ended PNP follower. This would then allow a considerable parts reduction as only one CCS is needed. But the complementary approach has its advantages as well of course.

I'm very open for suggestions there.

I'd recommend the KSA1220AY/KSC2690AY, available from Mouser; IME their higher fT/lower C can make a noticeable difference in stability. Also note the lesser dependence of hFE on collector current which helps for low distortion. KSA1381AF/KSC3503AF essentially offer TO-92 speed but are somewhat more limited with respect to power.

Besides, an active current source will have more noise than the present solution.

Did you actually check the effect? While an active load has higher current noise than a resistor it also reduces the effect of voltage noise of the following stages.

BTW, have you checked if CM distortion is significant here? Another place where the imperfect modelling of Early effect might cause misleading results.

Samuel

 

[jdbakker]

I'd recommend the KSA1220AY/KSC2690AY, available from Mouser; IME their higher fT/lower C can make a noticeable difference in stability. Also note the lesser dependence of hFE on collector current which helps for low distortion. KSA1381AF/KSC3503AF essentially offer TO-92 speed but are somewhat more limited with respect to power.

Ooh, those look very nice indeed, plus they share the same E-C-B pinout as the MJE/BD transistors so people who can't source the Fairchild parts have a fallback. It would appear that Digi-Key carries the latter parts as 2SA1381/2SC3503. DK also has a few very interesting Panasonic parts in TO-126, all of which sadly are in Maintenance or Obsolete status (ie: Not Recommended For New Designs).

Besides, an active current source will have more noise than the present solution.

Did you actually check the effect? While an active load has higher current noise than a resistor it also reduces the effect of voltage noise of the following stages.

I haven't studied it in detail, just observed that a transistor+resistor will always be noisier than a resistor by itself. Besides, while a back-of-the-envelope calculation shows that an ideal current source in place of R7 would increase DC open-loop gain by 20 to 30dB, the same back of the envelope shows that at 1kHz my current plan still has almost 100dB of OL gain, mostly limited by the Miller compensation.

As for voltage noise: gm * (R7||Q16) is what, 32dB? The VAS/buffer would have to be really noisy to have any impact on Ein. (This of course also applies to my No Nasty Noisy Current Sources-argument, and I'd have to emphasize that my main reason not to use a current source is its apparent effect on the harmonic distortion spectrum).

BTW, have you checked if CM distortion is significant here?

I have, but not extensively. In the final amp I plan to have an optional JFET cascode for Q1. I was thinking about the 2SJ74 for this purpose but could not find any suitable models.

I'm starting on the diff I/O amp next; do you see any reason why I should deviate from the current VAS/buffer (ie: everything but Q1/R11)? Of course the belt-and-suspenders clip protection can go (D21/D22), and I may experiment with a current source here in place of R7 to better accommodate the OCM circuitry, but other than that driving and bandwidth requirements are the same. Better copy what's working than having to start again from scratch, no?

Thanks again,

JDB.
[yes, working is a very bold qualifier for something that so far only exists within a simulator, but you know what I mean]

 

[Samuel Groner]

The VAS/buffer would have to be really noisy to have any impact on Ein.

I agree that the problem is to a great deal insignificant for a decent discrete implementation with high-gm input stage. ICs are struggling though, particularly if the second stage includes lateral PNPs. The low-frequency voltage noise of opamps such as the LT1028 is significantly worsened by second stage noise contributions.

My main reason not to use a current source is its apparent effect on the harmonic distortion spectrum.

I must admit considerable doubts about the validity of this simulation result as I do not see any reason why this should be so. The magnitude of the harmonics above the third or forth depends strongly on the exact modelling of parameter dependence such as hFE vs. collector current, junction capacitance vs. collector voltage etc. There is little reason to believe that standard modells are particularly good at this. At least with the SPICE implementation I use more distortion (at overall low levels) often just means that the numerics is more challanged...

Do you see any reason why I should deviate from the current VAS/buffer (ie: everything but Q1/R11)?

As I've indicated above I think the output stage is a bit too complicated and I'd check simpler solutions. But this might just be a matter of tast.

Apropos cascodes: I've used them as much to isolate high-dissipation positions from transistors where the actual Vbe matters (ie: the current sources).

Note however that hFE also shows a rather strong dependence on temperature. With the standard (base-referenced) cascode this still leads to a output current modulation, although surely of lower magnitude than without cascode.

Samuel

 

[Samuel Groner]

Forgot to add: For the bootstrapping of Q11/Q12 you could use the output.

Samuel

 

[Samuel Groner]

Sorry to nag on the output stage again, but perhaps something like this would be useful:

I1/I2 (my schematic) would be replaced by Q5/Q4 (your schematic), and I3/I4 could be used for cascode enhancement of again Q5/Q4. Saves two CCS, solves the Q20 saturation problem and perhaps even provides better ouput swing symmetry (didn't think it through though).

It is a bit more difficult to provide thermal compensation, but its doable.

Samuel

 

[jdbakker]

[addressing first the easy bits, then the harder ones]

By the way, the error I spotted last night is in Q21, which is now pushing Q20 dangerously close to saturation. I have a few obvious fixes which I'm planning to sim when I get some time.

One such obvious fix is to connect the base of Q21 to the collector of Q8. This increases VCE of Q20 to one diode drop, give or take. I had to do some sims to see if the added delay would mess up my HF response, but that looks OK too.

Forgot to add: For the bootstrapping of Q11/Q12 you could use the output.

I originally had it that way, but that runs into trouble when high loads cause voltage drops across R18/R19 pushing Q11/Q12 into saturation. Also at my standard 1kHz 30Vpp-out test the impedance of the Q11/Q12 stage drops by 40%, and that gets worse for higher frequencies; plus third and higher harmonics go up a few dB. (Still, if I were space constrained I agree this modification would be among the first I'd consider).

Sorry to nag on the output stage again,

Trust me: I don't see your feedback as nagging. Any holes that you (or anyone else) can shoot in my plan now saves me headaches in proto building!

I1/I2 (my schematic) would be replaced by Q5/Q4 (your schematic), and I3/I4 could be used for cascode enhancement of again Q5/Q4. Saves two CCS, solves the Q20 saturation problem and perhaps even provides better ouput swing symmetry (didn't think it through though).

Looks interesting, although I expect that output swing is more likely to be symmetrical in the diamond buffer due to the NPN/PNP-pairing (although my single emitter follower Q20 may well spoil that).

It is a bit more difficult to provide thermal compensation, but its doable.

...and that there is my main worry. My design is easy to get/keep thermally stable by using the same parts for Q11/Q14 and Q12/Q13, and screwing Q11/Q13 and Q12/Q14 together. However, your suggested plan looks very useful for my work-in-progress complementary DOA, as there I am allowed to use dual SMD transistors in either SOT-23-6 or SC-70-6.

Apropos cascodes: I've used them as much to isolate high-dissipation positions from transistors where the actual Vbe matters (ie: the current sources).

Note however that hFE also shows a rather strong dependence on temperature. With the standard (base-referenced) cascode this still leads to a output current modulation, although surely of lower magnitude than without cascode.

True, but there too the expected dissipation in Q4/Q19 and Q5/Q8 is pretty much matched, so the base current cancellation should affect temperature-induced changes as well. This does not hold for the other cascodes, of course.

My main reason not to use a current source is its apparent effect on the harmonic distortion spectrum.

I must admit considerable doubts about the validity of this simulation result as I do not see any reason why this should be so. The magnitude of the harmonics above the third or forth depends strongly on the exact modelling of parameter dependence such as hFE vs. collector current, junction capacitance vs. collector voltage etc. There is little reason to believe that standard modells are particularly good at this. At least with the SPICE implementation I use more distortion (at overall low levels) often just means that the numerics is more challanged...

I don't disagree, and I share your concerns. SPICE is very good at getting its user to chase ghosts. However, I have seen this effect (ie: an increase of OL gain decreasing absolute THD but increasing the relative proportion of higher-order products) occurring in RF amplifiers, albeit at much higher levels than -150dBc. And while SPICE does have a nonrealistic outlook in some areas, this effect is pretty reproducible even after I made changes in the output buffer stage, plus when I would change the signal levels the distortion spectrum would change proportionally in the way I would expect. Still, it's not impossible that at -150...-170dBc (25...29 bits) I am hitting a floating point rounding artefact. And when all is said and done, we're talking about what happens on the dozens-of-nV level when the amp is driving 10VRMS into a 100R load, while the expected output-referred noise in 20-20kHz is 4-5uV. Maybe that's the point to stop tweaking and to start building.

Thanks again,

JDB.
[now where did I leave that bag of BC850/860s?]

EDIT: have you (or anyone else) ever come across thermal mass/thermal time constant data for standard SOT-23 transistors?

 

[Samuel Groner]

My design is easy to get/keep thermally stable by using the same parts for Q11/Q14 and Q12/Q13, and screwing Q11/Q13 and Q12/Q14 together.

I agree that your plan is simpler with this respect, but note that the different collector voltage/current of Q11/Q14 and Q12/Q13 still gives rise to a slightly different Vbe tempco. To support my suggestion I could add that negative feedback might be used to set quiescent current of the output transistor pair (see the class A power amp by D. Self).

True, but there too the expected dissipation in Q4/Q19 and Q5/Q8 is pretty much matched, so the base current cancellation should affect temperature-induced changes as well. This does not hold for the other cascodes, of course.

Yes, Q4 and Q5 are temperature compensated. For Q10/Q18 there'd be the possible fix to reference the cascode to the emitters of Q15/Q17. Costs a resistor and two dual diodes. Would often qualify as overkill, but perhaps in this context? There's probably no easy fix for Q8/Q19.

BTW, couldn't we operate Q3/Q6 at one Vbe lower collector voltage? Safes at least one dual diode in the lower diode string and gives slightly more headroom.

An increase of OL gain decreasing absolute THD but increasing the relative proportion of higher-order products.

I wouldn't be so sceptical if there actually were an increase in o/l gain--but as far as I understand 1 kHz should be well above the dominant pole, so no significant change occurs.

Have you tried a feed-forward cap from Q16 base to Q3 base? Found that often helpful to reduce gain peaking above unity gain.

Have you ever come across thermal mass/thermal time constant data for standard SOT-23 transistors?

Not that I'd remember.

Samuel

 

[jdbakker]

[N]ote that the different collector voltage/current of Q11/Q14 and Q12/Q13 still gives rise to a slightly different Vbe tempco.

...but probably not to the extent that it'll cause thermal runaway, right? I must admit I've only done a manual first-order analysis on this.

To support my suggestion I could add that negative feedback might be used to set quiescent current of the output transistor pair (see the class A power amp by D. Self).

Are you referring to the one discussed in Self on Audio? I don't own a copy, and Google Books' Limited Preview was only of limited help.

For Q10/Q18 there'd be the possible fix to reference the cascode to the emitters of Q15/Q17. Costs a resistor and two dual diodes. Would often qualify as overkill, but perhaps in this context?

I'm still not sold on the emitter-referred cascode; I am worried that the loss in PSRR (even with the biasing R replaced by an R-C-R filter) outweighs the gain in output impedance. And while it may be hard to believe from someone who's just proposed a 21-transistor DOA, I indeed doubt the possible gain in performance is worth it.

There's probably no easy fix for Q8/Q19.

I could connect the base of Q10 to the emitter of Q19, creating a cascade of cascodes. Probably not worth it, especially as I'd need to re-investigate the effect on clipping behaviour.

BTW, couldn't we operate Q3/Q6 at one Vbe lower collector voltage? Safes at least one dual diode in the lower diode string and gives slightly more headroom.

Sadly no. As drawn VCE for Q4, Q19 and Q10 are approximately equal; if I were to remove one diode then Q10 would saturate during negative clipping, which slows clipping recovery and can cause ringing or oscillations. A similar story goes for the upper diode string.

An increase of OL gain decreasing absolute THD but increasing the relative proportion of higher-order products.

I wouldn't be so sceptical if there actually were an increase in o/l gain--but as far as I understand 1 kHz should be well above the dominant pole, so no significant change occurs.

Very good point. In that light I have no further theories ("R7 linearizes VAS input impedance" ... but how would that impact what I'm seeing?).

Have you tried a feed-forward cap from Q16 base to Q3 base? Found that often helpful to reduce gain peaking above unity gain.

I tried a few feedforward caps in an earlier version, but I found that the gain peaking I was seeing at the time was more effectively eliminated by C4/R15. All I'm getting now is a very smooth bump starting at 500kHz which tops out at just below 0.5dB at 3MHz; is that anything to be concerned about?

One thing I am still concerned about is this:

This brings me to one of my worries: when Q1 saturates, its base current increases, changing the amplifier's input impedance. This may lead to...funky responses in a reactive source. How much good is it to build an amp that clips gracefully if it produces, say, ringing in a mike/cable assembly? Worse, what'll happen to a ribbon mic? Granted, for this to happen the ribbon mic would have to somehow send a >5V signal down the line, but still. I need your input here; if this is a real problem I could implement a simple current limiter around the input transistor, but that does increase distortion some.

Thanks again,

JDB.

 

[Samuel Groner]

...but probably not to the extent that it'll cause thermal runaway, right?

No, the problem is probably of academic interest only. I've never seen thermal runaway to happen, not even with poor designs.

Are you referring to the one discussed in Self on Audio?

It's in both of his books. Basically a simple diff amp which compares the voltage drop across the emitter resistors (which is constant for class A) against a voltage reference. The output is used as error voltage to adjust bias.

All I'm getting now is a very smooth bump starting at 500 kHz which tops out at just below 0.5 dB at 3 MHz; is that anything to be concerned about?

No. The feed-forward cap I was talking about acts more in the 10-100 MHz region and helps to increase gain margin by increasing phase margin of the local Miller loop. In stuborn cases it may even prevent local instability of that loop.

I suggest that you include the part for a prototype layout and then see if there's any need for it. Additionally I'd include base stopper resistors for Q13/Q14 (you never know when a follower decides to oscillate in the 50 MHz region--guess what I'm doing right now), a series RC network in parallel with the feedback resistor (helpful for reducing square-wave overshoot) and a series RC network across the inputs of the two amplifiers (comes in handy if the entire preamp decides to oscillate at high gain, e.g. with open input or strongly inductive sources).

One thing I am still concerned about is this:

Personally I've never cared too much about overload behaviour as I tend to set my levels low enough. But there are surely other oppinions.

Samuel

 

[jdbakker]

Additionally I'd include base stopper resistors for Q13/Q14 (you never know when a follower decides to oscillate in the 50 MHz region--guess what I'm doing right now)

Oh yes, I know about that one. For through-hole parts like the output transistors I generally use ferrite beads on the base leads, but a small series R is probably more reproducible for builders. An added advantage of this topology is that R16/R17 double as base resistors for the output pair.

[...] a series RC network across the inputs of the two amplifiers (comes in handy if the entire preamp decides to oscillate at high gain, e.g. with open input or strongly inductive sources).

Good one. I think my input RFI network should present a well-defined input Z in such circumstances, but four more pads is cheap insurance.

Personally I've never cared too much about overload behaviour as I tend to set my levels low enough.

Normally I would agree, especially since the amp doesn't start acting up until voltage spikes of 5V or so are applied, but I want to make this mostly foolproof for other builders. On the other hand some of the other projects here have similar behavior when overdriven hard enough. A simple fix would be to have a diode string across the inputs, but I need to evaluate what that does to my low-low distortion figures.

JDB.

 

[Samuel Groner]

If you accept a limited input CM range it should be easy to add a ground-referenced clamp which is capacitively bootstrapped from the emitter of the input transistor. For a purely differential clamp things are more difficult. HF source impedance of dynamic mics may reach 1 kOhm, so voltage-dependent capacity will be an issue without bootstrapping.

Samuel

 

[abbey road d enfer]

Hi all,
I've finished a first draft of the current-feedback op-amp that is to form the first stage of my discrete mic pre JD

Coming late on the thread, but on the original schematic, Q1 can work only if the output sits at quite a high positive voltage. There should be some current injected in Q1's emitter from a positive source.

 

[abbey road d enfer]

Besides, an active current source will have more noise than the present solution.

While an active load has higher current noise than a resistor it also reduces the effect of voltage noise of the following stages.
Samuel

Can you elaborate on that? My understanding is that an active load will increase the open-loop gain somewhat and then reduce the overall nonlinearities of the whole structure when feedback is applied. Is that what you mean?

 

[jdbakker]

Coming late on the thread, but on the original schematic, Q1 can work only if the output sits at quite a high positive voltage.

Note that the input is biased at -2.5V (first parameter in the SINE()-statement, not exactly glaringly obvious I agree). Q1 VBE is ~0.7V, R3 has 1.8V across it (@5.2mA), the output sits at 0V give or take a few mV. The negative input bias has the desirable side effect of keeping a positive voltage across the phantom blocking electrolytics even if P48 is switched off.

I experimented with having a separate current source provide the emitter current for Q1, but the more symmetrical load on the output stage wasn't worth the extra complexity or noise.

If you accept a limited input CM range it should be easy to add a ground-referenced clamp which is capacitively bootstrapped from the emitter of the input transistor. For a purely differential clamp things are more difficult. HF source impedance of dynamic mics may reach 1 kOhm, so voltage-dependent capacity will be an issue without bootstrapping.

After some more thinking and tinkering I've decided to add just a two-diode clamp between B and E of Q1 to keep its B-E junction from zenering. Going into saturation doesn't harm the input transistor, and in saturation the dynamic impedance looking into the base is ~150Ω which I doubt will be an issue for any mic. I'm still open to be persuaded otherwise, though.

JDB.

 

[jdbakker]

Posted an updated schematic. Changed the output transistors to the ones Samuel suggested, increased output stage bias to 30mA, gave in and added a current source to the VAS. SPICE wants me to believe that at 1kHz 30Vpp out third and higher are below -170dBc. The weird rise in higher harmonics is gone after rebiasing the output stage, BTW.

Unless anyone has any major issues with this one I think I'll build one (probably a diff version, as that's what I need in the end). Depending on output device hfe I may need to up Q11/Q12 current to say 5mA, but that doesn't change the PCB layout.

JDB.
[now let's see what SPICE thinks about the diff version...]

EDIT: Yet another typo fix

 

[Gordy]

I can add nothing to the technical aspects of this thread, however I just wanted to drop in and say thanks for sharing such an interesting project and good luck with the design.