It is simpler than I thought to make it oscillate. Imaging putting 2 feet,1uH cable in front of the output transistor and a tinny capacitor at the output, congratulation, you've got an oscillator.
This might be what is happening.
The circuit make it a Colpitts Oscillator. Please see the detail analysis from: https://en.wikipedia.org/wiki/Colpitts_oscillator
C1 could be just the internal capacitance from the transistor.
C2 is the capacity load at the output of the emitter follower.
L could be the cable run from the input.
This might be what is happening.
The circuit make it a Colpitts Oscillator. Please see the detail analysis from: https://en.wikipedia.org/wiki/Colpitts_oscillator
C1 could be just the internal capacitance from the transistor.
C2 is the capacity load at the output of the emitter follower.
L could be the cable run from the input.
Last edited:
My goal is to figure out how to make an oscillator with a simple Emitter Follower, so that I can avoid it when building Emitter Followers.
It is not an emitter follower but a common emitter gain stage, that is why it oscillates having a voltage and current gain. just a matter where you choose the ground point. to make an oscillator you need an amplifier and a 3 element feedback phase shifter if the amplifier is inverting.
The usual approach is to add a base stopper resistor. 1 uH is a bit high for a 60.96 cm cable.
To illustrate what @basreflex wrote, these are three signal schematics (schematics with everything that's only meant for biasing left out for simplicity) of an oscillator. When you look carefully, you see that they are exactly the same, but basreflex and I find it easier to see why it oscillates for the right version than for the other two. C1 can be the base-emitter capacitance (junction and diffusion capacitance) of the transistor.
I once had a hum problem in a preamplifier. It was somehow related to the muting circuit; with a lot of experimentation, I found that connecting a small reverse diode from a node of the muting circuit to ground helped, but I had no idea why.
Years later, I bought a 150 MHz oscilloscope and found out that an emitter follower in the muting circuit oscillated. I hadn't seen that with my earlier 20 MHz oscilloscope. Apparently the capacitance of the reverse diode helped to reduce the oscillation amplitude.
The muting circuit worked off an unregulated supply with a large ripple, modulating the oscillator. The oscillation was picked up and rectified somewhere in the signal path. I added a base stopper to the emitter follower and the hum was gone, with or without the reverse diode.
Years later, I bought a 150 MHz oscilloscope and found out that an emitter follower in the muting circuit oscillated. I hadn't seen that with my earlier 20 MHz oscilloscope. Apparently the capacitance of the reverse diode helped to reduce the oscillation amplitude.
The muting circuit worked off an unregulated supply with a large ripple, modulating the oscillator. The oscillation was picked up and rectified somewhere in the signal path. I added a base stopper to the emitter follower and the hum was gone, with or without the reverse diode.
You can see the exact same structures shown above in an EF3 that help to understand why they can oscillate easily without precautions. Typically the traces around an OPS EF3 are 40-60 nH cf 4-6 cm assuming 10 nH per cm. The other one to be careful of is a cascode. I work a lot with CFA’s which are full of emitter followers and cascodes on the front end as well, requiring damper networks, short traces and compact layouts.
Re your hum problem Marcel, I’ve also seen that whereby the HF oscillation gets modulated by LF ripple and then couples into everything only to be later demodulated somewhere else in the circuit. Tough one to solve without a wide band scope.
Last edited:
Marcel and I have attended the same school, rigorously educated by Ernst Nordholt. Ernst used to have a vacation home 500m from my house here in spain.
The Cbe of MJL3281 is around 6nf.
I model it with a 2n3904 that it has less internal capacitance.
It oscillates even happier because of high fT.
C3 could be the input capacitance of the transistor of the next stage. All transistors have some input capacitance. The 600p value is from Cob of MJL3281. Thus, it is a realistic value.
The 1uH inductor at the input represents the worst case you may have.
How to tackle this problem?
Based on the theory from https://en.wikipedia.org/wiki/Colpitts_oscillator#Theory
There are 2 requirements to make it oscillate.
Condition 1: Negative impedance looking from the input.
From the wiki page, the real component of the complex impedance at the input is.
The deal breaker is the negative impedance is proportional to the gm. gm of the transistor increases with the bias current. That's why you see the most of parasitic oscillation only happens at the top or bottom of the wave form.
Putting a base stopper resistor alone is not that effective as the gm could be very big in some situation.
Condition 2: An inductor attached to the input.
The idea to avoid this is to make the input not inductive at the critical frequency with some help of the shunt capacitor. Like below, I also added some resistors to damping out the Q. They might not be in the optimal values, but you should get the idea.
1MHz drive into 600pf capacitor.
PS: In the case of the last output stage, a zobel network could also help the situation. However, zobel network is not a definitive fix. You will find lots of examples that amps with zobel network also suffer the parasitic oscillation.
I model it with a 2n3904 that it has less internal capacitance.
It oscillates even happier because of high fT.
C3 could be the input capacitance of the transistor of the next stage. All transistors have some input capacitance. The 600p value is from Cob of MJL3281. Thus, it is a realistic value.
The 1uH inductor at the input represents the worst case you may have.
How to tackle this problem?
Based on the theory from https://en.wikipedia.org/wiki/Colpitts_oscillator#Theory
There are 2 requirements to make it oscillate.
Condition 1: Negative impedance looking from the input.
From the wiki page, the real component of the complex impedance at the input is.
The deal breaker is the negative impedance is proportional to the gm. gm of the transistor increases with the bias current. That's why you see the most of parasitic oscillation only happens at the top or bottom of the wave form.
Putting a base stopper resistor alone is not that effective as the gm could be very big in some situation.
Condition 2: An inductor attached to the input.
The idea to avoid this is to make the input not inductive at the critical frequency with some help of the shunt capacitor. Like below, I also added some resistors to damping out the Q. They might not be in the optimal values, but you should get the idea.
1MHz drive into 600pf capacitor.
PS: In the case of the last output stage, a zobel network could also help the situation. However, zobel network is not a definitive fix. You will find lots of examples that amps with zobel network also suffer the parasitic oscillation.
Last edited:
The base-emitter capacitance is actually gm/(2 pi fT) - cbc. That equation includes diffusion capacitance as well as junction capacitance.
Last edited:
I may have asked this before, but do you also know Fred Neerhoff, the rigorous and somewhat eccentric former network theory professor?
Right, I haven’t looked it in details. For big output transistors, the Cbe should be at nF level.
Here are four analyses from four different authors. I suspect Marcel has even more.
_
_
I started at delft in 1972 and graduated in 1981, I had the pleasure of knowing the eccentric mr Buijze; must be something with network theory people. He shouted about resonant circuits having voltage amplification, no need to read that awful elektor magazine, or to go the shop to buy some transistors.
It's not quite as bad as it seems. At high currents, the diffusion capacitance dominates and the base-emitter capacitance becomes much larger than the base-collector capacitance - assuming that the transistor is kept out of saturation. In that case,
C1 ~= gm/(2π fT).
Filling this in in your equation,
Rin ~= -fT/(2πf2C2)
As the transit frequency of a transistor does not increase without bounds, neither does the negative resistance go ever more negative with increasing current. It stops when you reach the peak fT.
The equation predicts that the negative resistance goes to minus infinity as frequency drops to zero, but that is because the effect of finite current gain at low frequencies is not included. The low-frequency current gain of a MOSFET can go quite high, but its output resistance will also have a damping effect at low frequencies.
By the way, there is some similarity to the issue in this thread: https://www.diyaudio.com/community/...lifier-stability-and-source-impedance.420032/ It's also about a voltage follower that may get instable depending on its source impedance, but the voltage follower is assumed to be a CMOS op-amp rather than a capacitively loaded transistor. You also get FDNR-behaviour at the input then.
Last edited:
Nice. This is more useful as gm is not accurate under big current and Cbe is unknown for the most of parts.
