Oscillations on blameless EF amp

jricha

Member
2015-08-14 6:57 pm
I am having some major issues with a LTspice simulation of my EF amplifier. I am getting ringing on the lower waveform and have tried everything to stop it. The phase margin for this amp is approx 45 degrees. I am starting to think it is the spice models. Can someone please shed some light on this. Thanks.
 

Attachments

  • Screenshot from 2016-06-16 14:39:41.png
    Screenshot from 2016-06-16 14:39:41.png
    125.3 KB · Views: 261
  • Screenshot from 2016-06-16 14:40:24.png
    Screenshot from 2016-06-16 14:40:24.png
    40.6 KB · Views: 255

jricha

Member
2015-08-14 6:57 pm
Thanks for that tip. It fixed the oscillation

I found with a similar circuit I got oscillations on the output too.
I put 10R in between q29 and q26 and also q30 and q25.
r43 and r44 should also be about 10R.

I guess a base stopper was needed because of the large transistor capacitances and inductances.

nigelwright--- You said you had problems like this before? Did you put the amp on a completed board? If so, how did it sound and how was the stability?

Thanks.
 

jricha

Member
2015-08-14 6:57 pm
Thanks for the tip with the onsemi models.

Use the Bob Cordell models for most things. LTSPICE's stock BC546B model is ok. For MPSA06/56 use the models from Fairchild not Onsemi (most of Onsemi's model's are crap!)

Also your output Zobel has 100pF not 100nF which looks incorrect.

Thanks for the tip with the onsemi models. The 1m resistor and the 100p cap is not a zobel network. It is the worst case speaker cable capacitance. I am building a tri-amped 3-way amplifier for some specialty speakers. A zobel network is not necessary in my situation because I am using an active crossover and will only be operating each speaker driver in its linear region.
 
I guess a base stopper was needed because of the large transistor capacitances and inductances.

nigelwright--- You said you had problems like this before? Did you put the amp on a completed board? If so, how did it sound and how was the stability?

Thanks.

Yes it was on a pcb.
I had to cut a couple of tracks to get the extra 10R resistors in.
It sounded good and once the 10r resistors were in there was no oscillation on the output.
 
Last edited:
Thanks for the tip with the onsemi models. The 1m resistor and the 100p cap is not a zobel network. It is the worst case speaker cable capacitance. I am building a tri-amped 3-way amplifier for some specialty speakers. A zobel network is not necessary in my situation because I am using an active crossover and will only be operating each speaker driver in its linear region.

That output Zobel network is necessary for amp stability, it has nothing to do with speakers.

Mike
 
Thanks for the tip with the onsemi models. The 1m resistor and the 100p cap is not a zobel network. It is the worst case speaker cable capacitance. I am building a tri-amped 3-way amplifier for some specialty speakers. A zobel network is not necessary in my situation because I am using an active crossover and will only be operating each speaker driver in its linear region.
Install an output Zobel.
Try 8r+100nF located very close to the output devices and returning to the Decoupling ground via a VERY SHORT loop length.
If using any 4ohms drivers, then try 4r+150nF
 

wg_ski

Member
2007-10-10 5:21 pm
A pure capacitance on the output of an emitter follower will cause a negative resistance to be reflected at the base. Put a resonator on the base and it's how you actually make an oscillator. Since these oscillations are local doing things to the global feedback and compensation rarely help and if they do it's by accident. Base stoppers, zobels, or both are needed. You can and often do get lucky that your particular transistor combination and capacitances are stable without doing anything specific to address this. But don't count on it.

The high impedance of a voice coil at high frequency in itself does not cause a problem. But what it does do is leave the cable capacitance, parasitic wiring, and the input capacitance of the opposite side of the push pull emitter to dominate. Making the load effectively "capacitive" which we know is bad. The zobel is a "de-Q'ed capacitior" which adds some resistance over the majority of the frequency range. Base stoppers directly attack the negative R.

Another problem with these oscillations is that they are typically parameteric - requiring a drive signal to "pump". They won't show up in a small signal loop gain analysis. Notice that it is only occurring over part of the cycle. This requires a large signal to "activate". Whether or not stability or instability would be predicted is really REALLY dependent on the accuracy of the models.
 
I am having some major issues with a LTspice simulation of my EF amplifier. I am getting ringing on the lower waveform and have tried everything to stop it. The phase margin for this amp is approx 45 degrees. I am starting to think it is the spice models. Can someone please shed some light on this. Thanks.

You should take the sensible advice given by others on stopper resistors and zobel networks.

Looking at the second image there is a small notch in the wave form along with the ringing - which suggests a crossover distortion issue.

In that regard you have not cross coupled the emitters of Q29 and Q30 with a resistor (as per Self) and adopted the alternative arrangement where these are coupled by separate resistors to the load.

R35 and R36 at 200R apiece I dare say are not providing enough bias on the output devices - for the power output of your amplifier 100R would be more usual - have a look at some other designs for ideas on this point.

I think you would do better to cross couple Q29 and Q30 emitters with 100R.

The low impedance path so created allows better discharge of stored charges in the bases of the power devices. This occurs on on opposite cycle swings and speeds up the turn off process.

In the crossover region the device pairs should have similar turn on and turn of speed in order that there be no timing gap or too much overlap.
 

jcx

Member
2003-02-17 7:38 pm
..
series few uH L||R is a good idea too for isolating amp output from cable C

you shouldn't really be concentrating on the speaker Z flattening application of "Zobel" - its use on the amp output is different

more robust networks include the RC to gnd "inside" the isolating series output L, directly loading the output Q
some even make a pi network, Self, Cordell discuss the issues

and cable C can be much higher, nF is possible even in home

measured numbers are not hard to find: https://passlabs.com/articles/speaker-cables-science-or-snake-oil
 
Last edited:
Another area for investigation is the use of a single voltage reference for the CCS elements feeding the LTP and VAS stages. There is no separation of the base currents for Q22 and Q24 which will cause unwanted cross influence under dynamic conditions.

A simple solution would be to include a stopper resistor in the base feed to Q24. A starting point for SIM purposes might be 330R.
 

Bonsai

Member
Paid Member
2003-07-25 10:44 pm
Europe
www.hifisonix.com
You need base stoppers and a Zobel.

Use a resistor tied class A drive as well.

wgi_ski has given the best advice on this one so far.

Your LTP current source is also not optimum - you are using a current mirror instead of a two transistor current source which means you will have very bad ripple rejection.
 
Last edited:
I am having some major issues with a LTspice simulation of my EF amplifier. I am getting ringing on the lower waveform and have tried everything to stop it. The phase margin for this amp is approx 45 degrees. I am starting to think it is the spice models. Can someone please shed some light on this. Thanks.

This circuit has a number of changes from the Self Blameless amplifier and a number of problems.

It would have been better to have dealt with these along the way using a model of Self's circuit for comparing changes one step at a time and finding out in whatever way if and when should problems arise.

That would avoid the job of sorting through a raft of technical replies at the end.