How to compensate for transistor lead inductance in power amp?

[Ionforbes]

Hi all,

I'm working on a class AB audio amp that's experiencing high-frequency oscillation of ~100mV pk-pk at ~25MHz. I'm fairly sure this is a result of the lead inductance or capacitance of the 2N3055s, since they're TO-3 devices and are connected to the PCB via 15cm wires, 16AWG for collector and emitter and 22AWG for the bases an. Simulating the circuit shows now oscillation whatsoever. Curiously, the oscillation is affected by the position of the wires. If I move the collector and emitter leads of Q2 closer, and move the base and emitter leads of Q1 further apart, the amplitude of the oscillations decreases largely, and if I touch both collectors together with my fingers as well, the oscillation goes away completely! Using a different mounting method is not an option for me, and so I'm trying to see if I can calm the oscillation with another compensation capacitor somewhere, but I can't figure out where. Could anyone provide some pointers? I should add I'm testing this with no load at the moment.

Thanks,
Ion

[bdunham7]

I would start by checking C3 and if it is good, doubling or tripling it to see if that eliminated the issue.  Then you'd have to see how much that change affects the HF rolloff to see if it is acceptable.

[bson]

It's probably not the 2N3055's that are oscillating; they have a unity gain bandwidth of a few MHz, nowhere near 25MHz.

More likely it's the predrivers (Q8, Q9) that oscillate.  They have much higher bandwidth and could probably use some compensation (across B-C). 

[mawyatt]

Emitter followers are notorious for oscillating, this usually caused by the extra phase shift in the BE junction that is usually not modeled in most simulation models.  Any additional phase delay such as base lead inductance can increase the potential for oscillation because what "appears" as a negative resistance at the base.

As mentioned by bson, since the frequency is ~25MHz, likely not the 2N3055 but the driver transistor.

A couple simple things to try are adding a series R (~100) in the base lead of the driver to make the overall base effect have a positive resistance. Also load the driver emitter with a resistor (~500) across the 2N3055 Base Emitter.

There's a couple other techniques, but try the base resistor 1st.

Best,

Edit: I didn't see you were referring to Q2 in OP schematic. The driver (2N4890) here isn't an EF but composite that makes Q2 appear as a power PNP. Adding a degeneration resistor (~30) in series with the driver emitter will reduce the "loop gain" between Q2 and the driver, also a shunt (~500) across Q2 BE will also reduce the "loop gain".

[magic]

Also load the driver emitter with a resistor (~500) across the 2N3055 Base Emitter.

Indeed, both drivers are completely unbiased. They only run on whatever base current the outputs take under quiescent condition.

[Ionforbes]

I increased the 150pf to 300pf and also added 47pf b-c capacitors to the drivers, but to no avail. I'll try the resistors and report back. I also have a picture of the oscillation.

[bdunham7]

I increased the 150pf to 300pf and also added 47pf b-c capacitors to the drivers, but to no avail. I'll try the resistors and report back. I also have a picture of the oscillation.

Is this something you are constructing or is it an existing device that you have repaired or are repairing?

As I think others have been covered, the reason the amplitude of your oscillation is only 100mV instead of breaking up entirely is that the 2N3055s themselves aren't that fast, but the stuff before them might be.  I usually run into this sort of issue when repairing an amplifier and substituting in newer transistors.  If you compensate back where C3 is, you can use a very small capacitor, but further down the line you need larger values.  If compensating back at C3 didn't help, perhaps try 1nF across R14 and R15.  I'm not discounting the other suggestions that have been made either, especially if this is a new device you are making.

Of course your other option is to just live with it.      Seriously, there are plenty audio amps out there that oscillate and nobody notices.

[Conrad Hoffman]

I notice the circuit has no inductor at the output. This is usually a coil of moderate gauge magnet wire wound on some arbitrary 2W resistor used as a coil form. It will usually be associated with a 0.1 uF cap to ground. Look at some other amp circuits. I'd start there.

[T3sl4co1l]

It's probably not the 2N3055's that are oscillating; they have a unity gain bandwidth of a few MHz, nowhere near 25MHz.

More likely it's the predrivers (Q8, Q9) that oscillate.  They have much higher bandwidth and could probably use some compensation (across B-C).

2N3055s haven't been "2N3055" for decades; it's quite possible there is sufficient gain left up there in an epitaxial, perforated-emitter, etc. type.  Note that basically any power transistor of roughly those ratings, can be sold as a '3055.  You really have no idea what you're getting.  So, I recommend against using them, as a rule.

The wire length and positioning is your key.  There is a loop combined with, probably device capacitances, making a resonator.  Dampen this loop, by reducing its inductance, or increasing its resistance, and you're set.  Resistance can be added by placing ferrite beads on relevant wires.  Although a bead on the collector is unlikely to help, as ferrite beads saturate at fairly low currents (100s mA to some A), so, it might work at idle but not under peak currents (resulting in intermittent or squegging oscillation).  A small capacitor or R+C around Q8 or Q9 B-C ought to serve a similar purpose.

Also, why such an old fashioned design?  It's weird actually, it's not even old, it's just, not great.  I mean---in the old days they would've had more economy with respect to the still-expensive transistors; for example, a bootstrap ("Q") capacitor to maximize output swing at signal frequencies (without requiring current sink/source), but this just has straight up resistors into the bias diodes (R12-R15), so it's not going to swing to the rails very well at all.    And the asymmetrical drive connection (into the middle of the diodes!?) isn't doing any favors for that.  Or the even less efficient driver (R11 ~ 1.16W!).  Not to mention the unbalanced coupling capacitors ensuring maximal turn-on THUMP!

Tim

[magic]

Q7 also runs at over 1W and it's a TO39 device and the reason for such high bias is quite unobvious. Maybe it's a typo and 2k2 was meant?
Nevermind, high Q7 bias is pointless but R11 needs to somehow be able to pull all those 4k7 resistors close to the negative rail

However, I see no excuse for R5 being that high. It should be ~200Ω for the sake of input stage balance.

[Benta]

It's probably not the 2N3055's that are oscillating; they have a unity gain bandwidth of a few MHz, nowhere near 25MHz.

More likely it's the predrivers (Q8, Q9) that oscillate.  They have much higher bandwidth and could probably use some compensation (across B-C).

2N3055s haven't been "2N3055" for decades; it's quite possible there is sufficient gain left up there in an epitaxial, perforated-emitter, etc. type.  Note that basically any power transistor of roughly those ratings, can be sold as a '3055.  You really have no idea what you're getting.  So, I recommend against using them, as a rule.

The wire length and positioning is your key.  There is a loop combined with, probably device capacitances, making a resonator.  Dampen this loop, by reducing its inductance, or increasing its resistance, and you're set.  Resistance can be added by placing ferrite beads on relevant wires.  Although a bead on the collector is unlikely to help, as ferrite beads saturate at fairly low currents (100s mA to some A), so, it might work at idle but not under peak currents (resulting in intermittent or squegging oscillation).  A small capacitor or R+C around Q8 or Q9 B-C ought to serve a similar purpose.

Also, why such an old fashioned design?  It's weird actually, it's not even old, it's just, not great.  I mean---in the old days they would've had more economy with respect to the still-expensive transistors; for example, a bootstrap ("Q") capacitor to maximize output swing at signal frequencies (without requiring current sink/source), but this just has straight up resistors into the bias diodes (R12-R15), so it's not going to swing to the rails very well at all.    And the asymmetrical drive connection (into the middle of the diodes!?) isn't doing any favors for that.  Or the even less efficient driver (R11 ~ 1.16W!).  Not to mention the unbalanced coupling capacitors ensuring maximal turn-on THUMP!

Tim

+1 to this. It's a really old-fashioned design.
Also, pay attention to R16 and R17. At those values they are typically wire-wound and thus inductive, causing trouble. I've seen this in other designs. Better to user some film resistors in parallel instead.

[T3sl4co1l]

Well, I wouldn't worry about R16/17, inductance there just rolls off the output stage gain.  I suppose inductance might oscillate with a capacitive load.  (Another good reason for the Zobel network / Boucherot cell (shunt R+C and/or series R||L) at the output, as recommended.)

Note that film and oxide resistors are typically spiral cut as well; the cutoff frequency (where inductive reactance dominates) ranges from a few MHz for wirewound, to some 10s of MHz for film.  Leaded resistors of these values are not typically resistive to much higher, but SMT chips can go into the low 100s MHz.

Tim

[Vovk_Z]

I don't see if there is a reason to use SMT type resistors as emitter resistors having 15 cm long output transistor wires. Any through-hole film or oxide resistors will work fine and are much reliable.

[Ionforbes]

Thanks everyone for your replies, it's getting a bit late here so I'll try everything later. I did put 470 ohm resistors between the bases and emitters of both 2N3055s but it increased the oscillation's amplitude.  I think this might be a good thing though, as perhaps it meant the previous unbiased driver didn't have much gain at low current
 

Of course your other option is to just live with it.      Seriously, there are plenty audio amps out there that oscillate and nobody notices.

Wait, really? I suppose it is a bit outside my hearing range 

2N3055s haven't been "2N3055" for decades; it's quite possible there is sufficient gain left up there in an epitaxial, perforated-emitter, etc. type.  Note that basically any power transistor of roughly those ratings, can be sold as a '3055.  You really have no idea what you're getting.

These are 'real' 2N3055s from CDIL, they don't specify a maximum f_t but the minimum is 2.5MHz, so I wouldn't imagine it's much higher on mine. Choice of transistor was price, it was the cheapest TO-3 and I wanted an externally mounted can for aesthetics.

The wire length and positioning is your key.

I can get it to stabilise by  moving the Q2 leads closer and Q1 leads further fron each other, but it's precarious and hacky for a permanent 'solution.'

[magic]

Offset voltage between the bases of Q3 and Q5 is something like 75mV, correct?

[T3sl4co1l]

Right, B-E resistance pulls more bias current --> more transconductance --> stronger oscillation.  Not that this is a bad thing, it does improve performance (higher bandwidth, lower distortion) -- assuming the amp isn't doing weird shit that makes performance worse (typically the oscillation causes a shift in offset or bias, hence adding distortion as the oscillation goes up and down, increasing power consumption, etc.).  Such currents are reached in normal operation anyway, so it's not like you can avoid oscillation by sweeping it under the rug...

Tim

[floobydust]

OP you'd have to add a film/disk cap across C5 because electrolytics at 10's MHz don't do much. Also your circuit grounding and layout can aggravate things at those frequencies.

I find any audio amplifier or chip amp with a lop-sided output stage - Darlington up top and Sziklai pair on the bottom is always a PITA to get stable. The voltage gain between +ve and -ve half-cycle is non-symmetrical. There's a inherent instability in the Sziklai people mention.
Q9 would tend to oscillate, National Semi called it the "bottom-side fuzzies" where there's a burst of RF occurring on the bottom half-cycle. They add a Boucherot cell 10R/0.1uF at the output to stop that.

[bdunham7]

National Semi called it the "bottom-side fuzzies" where there's a burst of RF occurring on the bottom half-cycle. They add a Boucherot cell 10R/0.1uF at the output to stop that.

Which doesn't always work when you have to sub in modern driver transistors...

[Circlotron]

Google "Baxandall diode"
Simple mod to improve a quasi complementary setup.

And definitely 10R and 100n in series across the speaker output. Some chip amps take off without this. 

[Ionforbes]

Cheers everyone for your help; a 47pF capacitor across Q9's BC junction and a few turns through a ferrite bead on Q2's base lead and the oscillation is 100% gone 

[Kleinstein]

The conbination around Q9 Q2 is definitely prone to oscillation. Some resistant or the mentioned ferrite bead at the emitter can help and a base to emitter resistor for Q2 is also a good idea.

There are more problems like the high power loss for the poor 2N2905 and 2N3054. The circuit also still has not short circuit protection.

[Conrad Hoffman]

Now, don't forget to test it under real world conditions. That means with the expected length of speaker cable, attached to a real speaker. Blow the signal way up and look for "fuzzies" on the top and bottom.