Baxandall Super Pair

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hello,

I'm trying to use the Baxandall Super Pair, but everytime it will oscillate, in the simulator or on the bench.

As an emitter follower it's very stable (fixed collector voltage, CCS emitter load).

But if I try to use it with a varying collector voltage (typically in a VAS, low resistor emitter load, CCS collector load), I get oscillations.

Using the Super Pair as a Cascode also results in oscillations. Measured around 50-100 MHz on the scope.

Anyone has an idea on how to tame the beast ?

I've been playing with base stoppers to no avail... need help !

The attached circuit oscillates around 100 MHz.
 

Attachments

  • spair.gif
    spair.gif
    4.5 KB · Views: 6,798
at such a high frequency this is probably a very local phenomena, possibly similar to the oscillation of a capacitively loaded single transistor emitter follower where the capacitive load is reflected to the input as a negative resistance, resulting in oscillation for some source impedances - Nelson's suggestion can be appropriate to this type of oscillation

at some point an analysis using the full set of hybrid-pi transistor parameters is useful - the source and termination circuit impedances are important in determining the stable operation region

i have built the super pair a few times in current output circuits without obvious problems, but i haven't used it in a VAS circuit with high and complex base source impedance and possibly negative impedance active load - i believe you can find several simulations i have posted that suggest it can work as a VAS, but hardware results trump simulation every time
 
Okay, I played around with this a little more in simulation. I think my earlier emitter resistor mod was some kind of fluke. It doesn't seem to help this circuit. Using a circuit similar to yours in the simulator (shown below), I got 31 dB of peaking in the voltage gain at around 53 MHz. An oscillator for sure. Then I started putting various values of series resistance into the base of the PNP transistor. I was quite surprised at how large this resistor had to be. With a series base resistor of 1k, I still get peaking, about 5 dB at 34 MHz. This is probably enough to kill the oscillation. It make me nervous though. Even with values as large as 10k, I still get some peaking.
 

Attachments

  • superpairbasestopper.gif
    superpairbasestopper.gif
    4.8 KB · Views: 5,826
jcx said:
at such a high frequency this is probably a very local phenomena, possibly similar to the oscillation of a capacitively loaded single transistor emitter follower where the capacitive load is reflected to the input as a negative resistance, resulting in oscillation for some source impedances - Nelson's suggestion can be appropriate to this type of oscillation
andy_c said:
Then I started putting various values of series resistance into the base of the PNP transistor. I was quite surprised at how large this resistor had to be. With a series base resistor of 1k, I still get peaking, about 5 dB at 34 MHz. This is probably enough to kill the oscillation. It make me nervous though. Even with values as large as 10k, I still get some peaking.

JCX’s explanation seems right to me, and as it can be seen I showed already what is happening at the input. But, if the non-zero source impedance is the cause of the problem, is it really the added impedance here the way to go? I see that there is a critical range of the impedance, but as Andy pointed out, you should probably have very high base stopper value to tame the thing this way and thus the other things would come into the consideration. In the graph I posted above the source impedance is 1k and yes, the peak shown by the small signal analysis changes its amplitude and freq as the source impedance changes (graph below).

Btw, even if used like the emitter follower, “super pair” shows more problems with capacitive loads than any voltage follower without global (classic) f/b that I simulated.

Pedja
 

Attachments

  • z_source.gif
    z_source.gif
    5.3 KB · Views: 5,301
I tried a couple more things. I set up a simulation similar to the circuit I showed in Post #5 above, except using the circuit peufeu codenamed "funky" in the "amplifier topology subjective effects" thread here http://www.diyaudio.com/forums/showthread.php?postid=286698#post286698. The "funky" circuit also shows peaking in the frequency response in the 50-100 MHz range. Then I tried a different transistor for Q14 in the diagram of that earlier post, and varied its beta to vary the amount of compensating current fed back to the emitter of the main common base amp. With very large beta in the "sensing" transistor, the compensating current is almost zero and the peaking goes away. But when the transistors are nearly identical for beta, the peaking in the response again comes back.

So I thought, "what if the very act of feeding the base current back to the emitter is what's causing the instability?". This seems possible, since two circuits that use this technique (super pair and "funky") have the same problem. But there's another - the Hawksford cascode. I've shown the circuit below. How does it do?

This circuit has 34 dB of peaking at 57 MHz. The conclusion I'm reaching from all of this is quite unfortunate: Circuits which attempt to compensate the base current, such as the Hawksford cascode, the Baxandall super pair, and the "funky" circuit suck for high-frequency stability. Depressing.
 

Attachments

  • hawksfordcascodestability.gif
    hawksfordcascodestability.gif
    4 KB · Views: 5,509
while i still would go for a thorough mathematical analysis before selling anything using this, i think i've found a correlation that makes sense - the input Q should be similar or slower speed than the output Q which explains my satisfaction with the circuit and the difficulties seen using it with higher V, slower devices in a power amp VAS

a possible approach below tames the worst case combination of andy's slow mje340 with the fast 2n3906 - i haven't looked at how much it compromizes performance

(i only kinda trust transient analysis in spice, .AC not at all, it can go wrong many ways)

(this is using trapazoidal in LT SwCadIII)

play with C2, R3 values; 500pF, 30 Ohms gives very good pulse
 

Attachments

  • cmpnd_bax_pulse.png
    cmpnd_bax_pulse.png
    40.4 KB · Views: 3,990
As an added note here, I simulated the ft of the MJE340 at a bias current of 5 mA with VCE of 4 Volts - about the same conditions as what I used in the simulation above. ft comes out to 114 MHz. So while it's not a screamer by any means, it's not a dog either. Low-frequency AC beta is about 54 under these conditions. Unfortunately, the data sheet of these devices says so little it's hard to say if the models are realistic.
 
peufeu said:
Hello,

I'm trying to use the Baxandall Super Pair, but everytime it will oscillate, in the simulator or on the bench.


I've been playing with base stoppers to no avail... need help !

The attached circuit oscillates around 100 MHz.

This kind of circuit needs some compensating capacitor. Try 470 pF between base and collector of the driver or output device ;) .
 
andy_c said:
I tried a couple more things. I set up a simulation similar to the circuit I showed in Post #5 above, except using the circuit peufeu codenamed "funky" in the "amplifier topology subjective effects" thread here http://www.diyaudio.com/forums/showthread.php?postid=286698#post286698. The "funky" circuit also shows peaking in the frequency response in the 50-100 MHz range. Then I tried a different transistor for Q14 in the diagram of that earlier post, and varied its beta to vary the amount of compensating current fed back to the emitter of the main common base amp. With very large beta in the "sensing" transistor, the compensating current is almost zero and the peaking goes away. But when the transistors are nearly identical for beta, the peaking in the response again comes back.

So I thought, "what if the very act of feeding the base current back to the emitter is what's causing the instability?". This seems possible, since two circuits that use this technique (super pair and "funky") have the same problem. But there's another - the Hawksford cascode. I've shown the circuit below. How does it do?

This circuit has 34 dB of peaking at 57 MHz. The conclusion I'm reaching from all of this is quite unfortunate: Circuits which attempt to compensate the base current, such as the Hawksford cascode, the Baxandall super pair, and the "funky" circuit suck for high-frequency stability. Depressing.

Andy, this is to be expected as the super pair has a small amount
of + feedback to the emitter of OP BJT. I haven't yet played with
super pair (will soon) but I think a suitable way to tame it may
be as shown with damping at the problem spot.
You need both R's to be able to get suitable voltage across
driver BJT. I have omitted base stoppers for simplicity, they
may still be needed but I don't think so.

Cheers,

Terry

ps - hope image attachment works

An externally hosted image should be here but it was not working when we last tested it.
 
Terry Demol said:


Andy, this is to be expected as the super pair has a small amount
of + feedback to the emitter of OP BJT. I haven't yet played with
super pair (will soon) but I think a suitable way to tame it may
be as shown with damping at the problem spot.
You need both R's to be able to get suitable voltage across
driver BJT. I have omitted base stoppers for simplicity, they
may still be needed but I don't think so.

Cheers,

Terry


Insert one R at collector of IP BJT and another at Emitter
of OP BJT. OP is connected to Emitter of OP BJT above R.
Damping R's are connected together and to CCS load at other
end. Clear as mud :)

Cheers,

Terry
 
this seem to work, i think it can be considered a RC "stopper" across the input Q2_cb; at high frequencies the RC shorts out the gain of Q2

this is equivalent to my first schematic with the loop minimized by going directly across Q2_cb instead of including V3 cascode voltage source which would introduce parasitics at 100 MHz in the real world

to regain more audio frequency performance i split the C and bootstrapped the midpoint to Q2_e, this gives the expected 40+ dB improvement of output Z over the simple cascode with the mje140 up to 40 KHz, and 20 dB better performance to beyond 1 MHz; the R3, C2,4 chain must have very low parasitics at 100 MHz, NPO surface mount would be best

(i also fixed the output current, its now 10 mA as i intended – it was only ~3mA in my earlier schem)
 

Attachments

  • cmpnd_bax_pulse2.png
    cmpnd_bax_pulse2.png
    39.7 KB · Views: 3,513
jcx said:
this seem to work, i think it can be considered a RC "stopper" across the input Q2_cb; at high frequencies the RC shorts out the gain of Q2

this is equivalent to my first schematic with the loop minimized by going directly across Q2_cb instead of including V3 cascode voltage source which would introduce parasitics at 100 MHz in the real world

to regain more audio frequency performance i split the C and bootstrapped the midpoint to Q2_e, this gives the expected 40+ dB improvement of output Z over the simple cascode with the mje140 up to 40 KHz, and 20 dB better performance to beyond 1 MHz; the R3, C2,4 chain must have very low parasitics at 100 MHz, NPO surface mount would be best

(i also fixed the output current, its now 10 mA as i intended – it was only ~3mA in my earlier schem)

JCX,

Try these mods, Q3 should have a base stopper, insert R between Q2 collector and Q3 collector, insert R between Q1 emitter and Q3 collector. Remove the compensation network.

Cheers,

Terry
 
The one and only
Joined 2001
Paid Member
andy_c said:
So I thought, "what if the very act of feeding the base current back to the emitter is what's causing the instability?". This seems possible, since two circuits that use this technique (super pair and "funky") have the same problem. But there's another - the Hawksford cascode. I've shown the circuit below. How does it do?

This circuit has 34 dB of peaking at 57 MHz. The conclusion I'm reaching from all of this is quite unfortunate: Circuits which attempt to compensate the base current, such as the Hawksford cascode, the Baxandall super pair, and the "funky" circuit suck for high-frequency stability. Depressing.

The Baxandall It does seem like a lot of trouble, but that's
because it has quite a bit of gain and inherently positive local
feedback. It's begging to be an oscillator.

As to the "Hawksford" cascode, it is inherently pretty stable,
however I would categorize it as an ordinary cascode since
the topology has been around for a very long time. I used to
buy monolithic JFET differential pairs that were internally
cascoded in a like manner. :cool:

Also beware of trusting your simulator too much.....
 
Actually I find the cascode that returns the base current to the lower transistor to be nearly equivalent to the super-pair (I agree that “Hawksford” cascode is inappropriate as this version of the connection appears long before his paper but we do need a concise way to differentiate between the subtle flavors of cascodes and Hawksford does provide a service in pointing out the difference in performance)

Below I show a cascode with the base current returned to the lower Q emitter rings but isn’t outright unstable, however in the interest of a fair comparison to the compound super-pair cascode, I have shown it with a pure current source emitter load and it spontaneously oscillates as does the compound circuit

The super-pair may be seen to be topologically nearly equivalent to the cascode with base current compensation, just follow the pairings of emitter-base, collector-emitter between the 2 Qs of each circuit – you simply have to “twist” the input Q from NPN to PNP (and eliminate the Vref) to go from cascode to super-pair

(I don’t trust my simulator, I just don’t have to leave my armchair to go for a ride)

Terry: which drawing?
 

Attachments

  • cascode_osc.png
    cascode_osc.png
    48.9 KB · Views: 3,207
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.