Tom, I imagine the meaning is 'unlimited' - and it certainly runs to 10s or 100s of MHz depending on the capacitances of the part.
I agree on the advisability of ferrites on the drain as well as gate leads, but there are (as usual) precautions to be aware of:
- some beads are conductive! watch out for neighbouring device leads when placing them.
- If the tubular beads are used in audio circuits, I recommend fixing the beads firmly. If you don't, ambient vibration can actually induce noise on the lead hosting the bead! Took me a while to trace the bizarre impact on the sound of an amp from this effect.
- for this reason, I prefer chip-ferrites like the Murata BLM21A series. Choose the current-level required first, then pick a value the gives high losses at 1 to 10MHz especially. These give high loss at 100MHz naturally.
- chip beads must be mounted on a circuit board, or some plain FR4 with 'tracks' cut into it. The end-caps snap off if you try to wire to them.
- Chip beads are not inductors, (they store very little energy) but they dissipate power at high-frequency. They don't form a LC tank circuit, with resonances, like an inductor does.
Murata and Fair-Rite have application notes that give a lot of very useful information on the choosing & use of ferrites and suppression generally.
I agree on the advisability of ferrites on the drain as well as gate leads, but there are (as usual) precautions to be aware of:
- some beads are conductive! watch out for neighbouring device leads when placing them.
- If the tubular beads are used in audio circuits, I recommend fixing the beads firmly. If you don't, ambient vibration can actually induce noise on the lead hosting the bead! Took me a while to trace the bizarre impact on the sound of an amp from this effect.
- for this reason, I prefer chip-ferrites like the Murata BLM21A series. Choose the current-level required first, then pick a value the gives high losses at 1 to 10MHz especially. These give high loss at 100MHz naturally.
- chip beads must be mounted on a circuit board, or some plain FR4 with 'tracks' cut into it. The end-caps snap off if you try to wire to them.
- Chip beads are not inductors, (they store very little energy) but they dissipate power at high-frequency. They don't form a LC tank circuit, with resonances, like an inductor does.
Murata and Fair-Rite have application notes that give a lot of very useful information on the choosing & use of ferrites and suppression generally.
Here's a snippet from the Murrata datasheet.
Is it because of mosfet properties that frequencies of 10-100MHz are of interest? I mean, what about AF up to 10MHz?
One could lower the resonant/operating frequency by paralling ten ferrite beads for instance. Or is another filtering technique more useful, if applicable.
An externally hosted image should be here but it was not working when we last tested it.
Is it because of mosfet properties that frequencies of 10-100MHz are of interest? I mean, what about AF up to 10MHz?
One could lower the resonant/operating frequency by paralling ten ferrite beads for instance. Or is another filtering technique more useful, if applicable.
Hi Jaap, when the FET oscillates, it has usually formed a single-stage amplifier-oscillator with an effective LC load circuit formed from parasitic inductance (wiring length) and capacitance (stray wiring plus device parasitics).
Voltage swings on the gate are 180 degrees out of phase from the drain swings already, and if drain swings are further phase-shifted by the LC networks to give another 180 deg shift, AND these signals are effectively coupled back to the gate at a frequency which gives an effective gain of 1, we have the conditions for oscillation.
In practice, the variability of the wiring parasitics (how long is the wire, how close to gate wiring, wire diameter, etc etc) and the FET parasitics (capacitances) means that the self-oscillating frequency can be from 1MHz to 100MHz or more.
The Murata beads all give high losses at 100MHz, so we are covering more possibilities by having high loss at 1 to 10MHz, as well.
Other ways to reduce the chance of oscillation:
- reduce the gate wiring length. Think of this as the 'input antenna' if you like. Adding a resistor (stopper) near the gate forms an RC pole with the gate capacitance, and reduces effective gain at risky frequencies. Everyone knows that a stopper must be close to the gate (or grid), but the reason for this is to reduce the effective length of the 'input antenna'. The stopper can be undermined by connecting other components directly to the gate, so please attach gate-protection zeners at the other side of the stopper.
Adding a ferrite bead, or a chip-bead to the gate enhances the suppression of troublesome frequencies.
Meanwhile, the drain wiring may be thought of as a 'Sender Antenna' emitting energy that is picked up by the gate (having been phase shifted by parasitics). Adding a ferrite here helps suppress high frequency emissions, and reduces the effective loop gain of the FET amplifier at high frequency.
The above notes apply mostly to a common-source amplifier. With a source follower the cause of oscillation is often that the effective Real Part of the complex-impedance at the gate becomes negative at some frequency. Suppressing source followers usually only requires a stopper resistor, though a gate bead won't hurt.
Voltage swings on the gate are 180 degrees out of phase from the drain swings already, and if drain swings are further phase-shifted by the LC networks to give another 180 deg shift, AND these signals are effectively coupled back to the gate at a frequency which gives an effective gain of 1, we have the conditions for oscillation.
In practice, the variability of the wiring parasitics (how long is the wire, how close to gate wiring, wire diameter, etc etc) and the FET parasitics (capacitances) means that the self-oscillating frequency can be from 1MHz to 100MHz or more.
The Murata beads all give high losses at 100MHz, so we are covering more possibilities by having high loss at 1 to 10MHz, as well.
Other ways to reduce the chance of oscillation:
- reduce the gate wiring length. Think of this as the 'input antenna' if you like. Adding a resistor (stopper) near the gate forms an RC pole with the gate capacitance, and reduces effective gain at risky frequencies. Everyone knows that a stopper must be close to the gate (or grid), but the reason for this is to reduce the effective length of the 'input antenna'. The stopper can be undermined by connecting other components directly to the gate, so please attach gate-protection zeners at the other side of the stopper.
Adding a ferrite bead, or a chip-bead to the gate enhances the suppression of troublesome frequencies.
Meanwhile, the drain wiring may be thought of as a 'Sender Antenna' emitting energy that is picked up by the gate (having been phase shifted by parasitics). Adding a ferrite here helps suppress high frequency emissions, and reduces the effective loop gain of the FET amplifier at high frequency.
The above notes apply mostly to a common-source amplifier. With a source follower the cause of oscillation is often that the effective Real Part of the complex-impedance at the gate becomes negative at some frequency. Suppressing source followers usually only requires a stopper resistor, though a gate bead won't hurt.
Last edited:
If you make a power-FET amplifier, and find that it oscillates at 1MHz or more (higher for small signal devices), the cause is most likely self-oscillation.
But sometimes, we build an amplifier and find that it oscillates much lower than 1MHz - maybe 100kHz, say. In this case the oscillation conditions have been created by a loop of two or more devices - a multi-stage oscillator.
These problems are fixed by rolling off the gain of the fastest of the two stages:
Say you have a heavy power transistor driven by an op-amp, and feedback taken from the transistor all the way back to the opamp inputs. This circuit may well oscillate, because the opamp reacts too quickly for the transistor to respond, and a 'tail-chase' ensues. The fix: compensate the op-amp with a C feedback network around the opamp only - to reduce the bandwidth of the op amp.
In most cases, ferrites are unable to help with these problems, and RC compensation must be used.
But sometimes, we build an amplifier and find that it oscillates much lower than 1MHz - maybe 100kHz, say. In this case the oscillation conditions have been created by a loop of two or more devices - a multi-stage oscillator.
These problems are fixed by rolling off the gain of the fastest of the two stages:
Say you have a heavy power transistor driven by an op-amp, and feedback taken from the transistor all the way back to the opamp inputs. This circuit may well oscillate, because the opamp reacts too quickly for the transistor to respond, and a 'tail-chase' ensues. The fix: compensate the op-amp with a C feedback network around the opamp only - to reduce the bandwidth of the op amp.
In most cases, ferrites are unable to help with these problems, and RC compensation must be used.
Thank you for the in depth answer Rod. Recently I was looking for a way to separate the power part from the steering part on this shunt regulator (SSHVSR):
The advance would be obvious: getting the heat out via a back side heatsink while maintaining accurate measurement at the cool running front end. One way would be the introduction of Kelvin connections, the other way -which I tried- would be extending the connection to R7. There I went off, probably because of the problem you addressed above. Would a bead cure my lil' problem or is more involved for that?
An externally hosted image should be here but it was not working when we last tested it.
The advance would be obvious: getting the heat out via a back side heatsink while maintaining accurate measurement at the cool running front end. One way would be the introduction of Kelvin connections, the other way -which I tried- would be extending the connection to R7. There I went off, probably because of the problem you addressed above. Would a bead cure my lil' problem or is more involved for that?
IF there is oscillation at HF to VHF [1 to 100MHz], suspect self-oscillation. Apply ferrites to all the gates, and to Q3 drain.
If there is lower frequency oscillation, then suspect a loop oscillator between Q3 and Q2. This will certainly have its phase margin worsened by adding long wire to R7.
BTW, I assume that the emitter of the Q2 should be swapped in the drawing with the collector.
IF that is so, and the oscillation is below 1MHz, you can apply frequency compensation to the Q3-Q2 pair by adding a little C between R7 and R11. Maybe 100pF [Wima FKP2 1000V] to start, and try higher values if it is stubborn and refractory.
The transient behaviour of the regulator may be degraded by higher values of capacitor, so as usual we have a tradeoff. If the wiring can be reduced to 50mm, the chances of success are higher!
Indeed, Q2 should have its emitter to the positive output.
I was thinking about 30cm or more (crossing from one corner of the case to the other).
That would be too troublesome to implement?
The charm of the circuit lays in the lack of an extra power supply,
besides the good taste with which Nick created the reg.
Could a new front end (probably opamp based) stretch that distance?
I was thinking about 30cm or more (crossing from one corner of the case to the other).
That would be too troublesome to implement?
The charm of the circuit lays in the lack of an extra power supply,
besides the good taste with which Nick created the reg.
Could a new front end (probably opamp based) stretch that distance?
It can be done, but the bandwidth of Q2's stage would need to be reduced seriously - using the capacitor R7 -> R3. But the regulator's dynamic performance will suffer, sadly.
Whether it is worth doing depends on the kind of load - but if you are using a shunt regulator, I suspect that dynamics are important.
You can try adding caps, and measuring the transient performance, if you wish to see where the tradeoff lies.....
Whether it is worth doing depends on the kind of load - but if you are using a shunt regulator, I suspect that dynamics are important.
You can try adding caps, and measuring the transient performance, if you wish to see where the tradeoff lies.....
Dynamics are very important, indeed.
Three years ago I had a 2A3 powered with this regulator.
It burned off 10 watts from the 330V 0,1A I supplied.
Reproduction was top notch (compared to LC filtering)
but thermal drift was the enemy I could not beat.
In the mean time several regulators have showed up (a.o. Jans')
but all at the expense of high component count or hard to obtain parts.
I'll give it a try next days and show results (if any
).
Three years ago I had a 2A3 powered with this regulator.
It burned off 10 watts from the 330V 0,1A I supplied.
Reproduction was top notch (compared to LC filtering)
but thermal drift was the enemy I could not beat.
In the mean time several regulators have showed up (a.o. Jans')
but all at the expense of high component count or hard to obtain parts.
I'll give it a try next days and show results (if any
).
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.