Oscillation due to (probably) feedback network

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'm becoming crazy playing around a parastic oscillation in a small amp I'm building: it is located around 500Khz and thay are many and many days I'm trying to eradicate it.
Simulation tells that with the components You can see in the attached schema I should have about 25° of margin, because the frequency response of the amp cross the unity gain at about
-155°.
The oscillation is about 2V Peak to Peak and changing the bias of the amp it changes its frequency but the magnitude remains aroud 2V Peak to Peak.
Today, by following some suggestions in the forum, I tried to remove the compensation cap across the feedback resistor R16 and after connecting the amp to the supply I was expecting a higher degree in the amplitude of the oscillation with a quick overheating of the power devices; quite suprisingly the amp in this configuration doesn't oscillate even if, based on the simulation, I should completely be out of any phase margin.
without the compensation cap across R16 simulation tells about -200° @ unity gain.
Moreover the amp do not oscillate (with the input shorted) even if put a 2.2uF cap acros the 8Ohm load resistor.
Do You have any comments about this?
Thank You,
Mauro
 

Attachments

  • myamp.jpg
    myamp.jpg
    76 KB · Views: 357
Other interesting things are:
- If i put a 150pf between the bases and the collectors of the drivers Q11 and Q12 the amp runs fine
- if I change the output devices form MJL4281/4302 to MJW0281/0302 or MJL0281/0302 the amp runs fine also without the 150pf across base and collectors of the drivers

the 0281s and the 0302s hava an output capacitance of 400pf while the 4281s and the 4302s have 600pf as output capacitance


I have also others channels built in the same way but for hihger supply (40v) with 3 pairs of MJL3281/1302 and the behavior is the same
 
the problem is that without c2 simulation tells that the amp should be unstable because the unity gain in the response is at -200 deg, while in the real world it runs without problems.
theory tellls that to have stability the unity gain should be reached before the phase touch -180 deg, that's why i added c2.
whit it i found in the simulation that i should have about 25 deg of margin, while under test it oscillate.
 
Adding C2 increases the loop gain at high frequencies, it is not the closed loop gain that matters. Small values of C2 gives a stabilizing effect though as it starts to provide phase lead a decade below its break frequency. Are you measuring closed loop or loop gain in the simulator?

If you do a transient simulation without C2 in the simulator, does it oscillate then? If it doesn't, then something is wrong with your loop gain measurment.

Someone on this forum posted this link:
http://eportal.apexmicrotech.com/mainsite/support/pages/app_notes.asp

It contains lots of useful stuff about stability!
 
None of those curves say it would be unstable. The unity (loop!) gain is at 300kHz where the slope difference between open loop gain and 1/beta (which is your closed loop gain at low frequencies) is still -20dB/decade.

Your method for measuring isn't correct, you need to break the loop and inject the signal to the inverting input.

The input should be grounded, put a über-huge inductor in series with the feedback resistor and from the junction a similarly huge capacitor in series with the AC signal source with amplitude 1 to ground. Loop gain is then the output voltage. (in reality divided by the inductor/capacitor junction but if you used huge enough components this will be 0dB with 0 degrees phase shift)
 
Administrator
Joined 2004
Paid Member
Hi Mambo,
Is there any reason you are using MPSA56 for input differentials? They are slow compared to the normal types normally used and the gain is low. Why not use the proper types and see what happens?

Even 2N3906 would be an improvement. 2SA970 would be a standard type to use here.

-Chris
 
Also be aware that if you are using the Onsemi models for the 4281/4302, they are wrong. Somewhere on here you will find corrected models for the 3281/1302 - use these to simulate with.

This amp looks pretty much like Doug Self's Blameless amp... only one immediate difference - the arrangement of the VAS Q9 and Q8 as a Darlington rather than as an EF.

I'd agree with megajocke on the wrong connection of Q4. Personally, I wouldn't share the reference between the VAS current source and the LTP current source. Consider using an LED as the reference for the LTP current source.

One problem I saw in simulation when playing around using a shared reference between LTP and VAS current sources, was that if clipping occurs, the VAS can saturate and draw so much base current that the current source for the LTP fails. This leads to all sorts of nasty things such as oscillation or rail sticking.
 
I replaced the reference transistor of the current source with a pair of 1n4148 but the result is the same: with C2 in place as soon as the output devices start to conduct (at around 1.7mv between the emitter resistors) oscillation comes to my scope.
Again, without C2 the amp do not oscillate

The last attempt I can try is to re-arrange the connections of the VAS so that it comes to a EF instead of a Darlington connection. I can try also to revert to a "simple" VAS: the little increase in distortion should be of no problem for the final target of the amp.
 
Those are two different issues, the current source has nothing to do with the oscillation, that is related to clipping.

Do a loop gain measurment in the simulator. I'm pretty sure C2 will decrease phase margin. The "measurement" you did in the simulator that makes you believe C2 will give you better stability is flawed. The models used for the transistors may not be correct either.

When did the simulation start to matter more than the reality? ;) To test if it is the measurment that is wrong or the models, do a transient simulation in the simulator instead and check if it oscillates in the simulator.
 
Thank You very much to all for Your help! I solved the problem: it was in the VAS and after removing Q9 and R30 the amplifier do not oscillate anymore.
It now runs ok either with or without C2, even if C2 should guarentee a greater phase margin.

Probably the Darlington layout of the VAS instead of the EF layout was suffering from some internal oscillation as soon as its whole collectore voltage was rising due to bias adjustement.

Now it is a standard VAS, built only on the KSC2690A, and the little increase in the THD is not a problem for the purpose thi amplifier has been built to.

Thank You again,
Mauro
 
Good that you got it working :) You can't be sure that C2 will make the amplifier more stable - if the capacitor is too big it will make the amp more unstable. Think of what happens if you were to put a really big capacitor there.

Still - check what happens to the input stage tail current when clipping using the simulator if you want to know what problems can arise by using the same bias network for VAS and input stage. Using two diodes and the 150 ohm resistor to the VAS won't help, the resistor is too small.
 
If someone can help me undrstanding what is happening in the loop gain I used the instructions contained in LoopGain2.ASC of the LTSPICE samlpes to plot it for my amp: i shorted the input by leaving in place the input conditioning network

The attached image is the loop gain with C2 in place
 

Attachments

  • withc2.jpg
    withc2.jpg
    85.7 KB · Views: 102
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.