Many amps have a series CR network (I believe it’s also called a Zobel network) somewhere in the input stage, for the purpose of creating a HF roll off for stability. Some shunt the anode load of the input tube. Some, like the ST-70 (which I’m specifically interested in) incorporate this network between the input and the phase splitter, shunting to ground.
I’ve read that sometimes the value (pole) of this network is a little too aggressive, such that valuable HF in the audio range is impacted. So my questions are these.
What is the pole of the ST-70 network, 82pf/18k?
If necessary, how would I adjust the component values to raise the pole?
FWIW, I am just a total (and maybe a little obsessive) novice hobbyist, immersed in tweaking.
Thanks,
Dave
I’ve read that sometimes the value (pole) of this network is a little too aggressive, such that valuable HF in the audio range is impacted. So my questions are these.
What is the pole of the ST-70 network, 82pf/18k?
If necessary, how would I adjust the component values to raise the pole?
FWIW, I am just a total (and maybe a little obsessive) novice hobbyist, immersed in tweaking.
Thanks,
Dave
That RC is (AC wise) across the 270k plate resistor, so the 270k sets the stage's rolloff point
along with the 82pF, rather than the 18k. With a scope and generator, you can see how varying
the C affects the ringing. You may be able to reduce its value somewhat, perhaps by half,
with a given amplifier load before significant overshoot or ringing begins.
The input pentode also affects the rolloff point somewhat. The 18k should remain constant,
since it mainly smooths the phase response, by limiting the amount of hf gain reduction.
Be very careful, oscillations at high power will destroy loudspeakers. Use a dummy load.
Last edited:
Whether you shunt the plate load (connecting plate to B+) or connect from plate to common 'ground', from an AC perspective they are equivalent. Makes no difference, though by habit I put it across the plate load.
A generalized rule of thumb (someone produced a fairly reasonable math proof which I read but cannot recall where it came from) is to size the resistance of the RC network to 10% of the value of the plate load. So with the ST-70 having a plate load of 270K, you could start out with a 27K resistor. From there, I tweak the desired capacitance value by scope with a square wave input. Produces very decent rise times with low overshoot. Sure, you can putz around with other R values, but 10% is a very good starting point. C makes all the difference on the scope.
A generalized rule of thumb (someone produced a fairly reasonable math proof which I read but cannot recall where it came from) is to size the resistance of the RC network to 10% of the value of the plate load. So with the ST-70 having a plate load of 270K, you could start out with a 27K resistor. From there, I tweak the desired capacitance value by scope with a square wave input. Produces very decent rise times with low overshoot. Sure, you can putz around with other R values, but 10% is a very good starting point. C makes all the difference on the scope.
This is called setting a dominate pole, or "slugging".
Usually it's necessary in an amplifier with an output transformer
and overall negative feedback.
Shelf networks were becoming a well known stability aid back in the 1940's:
Corrective Networks for Feedback Circuits, V. Learned, Proc. IRE, July 1944.
http://dalmura.com.au/projects/corrective%20networks.pdf
Stabilizing feedback amplifiers, T. Roddam, WW Mar 1951
http://www.americanradiohistory.com/Archive-Wireless-World/50s/Wireless-World-1951-03.pdf
Corrective Networks for Feedback Circuits, V. Learned, Proc. IRE, July 1944.
http://dalmura.com.au/projects/corrective%20networks.pdf
Stabilizing feedback amplifiers, T. Roddam, WW Mar 1951
http://www.americanradiohistory.com/Archive-Wireless-World/50s/Wireless-World-1951-03.pdf
The attached image shows the network I'm trying to figure out. Which resistor is used for the frequency calculation? 18K, results in a roll off frequency of 107883. 220K results in 8826. Significantly different. One is well out of the audio band, the other isn't.
Hey Ray, I went back and read your response so you probably already answered my question. Please verify that the resistor which establishes the roll off frequency is the input tube's anode load, and not the one downstream of the cap.
Thanks,
Dave
Yes, approximately the input stage's roll off -3dB point is 1/(2Pi x 270k x 82pF) = 7.2kHz.
The input pentode's Rout also affects that frequency, but this formula is good enough to start.
Last edited:
Thanks.
The calculation with a 40pf cap works out to be about 18kHz, way more appropriate for HI FI, and probably still offering good HF stability.
If you're keen then the assessment process includes:
Determine stage gain at midband with feedback.
Determine shelf start corner frequency ~ 1/(2π.220k.82pF) = 9kHz, assuming pentode effective plate resistance is >> 220k
Effective plate resistance may significantly reduce above nominal 9kHz due to loss of feedback.
At high frequency above shelf stop corner f then 220k//18k = 17k
Determine drop in stage gain given that internal plate resistance may have changed and external plate load has changed from 220k to 17k.
Shelf stop corner frequency ~ 1/(2π.17k.82pF) = 114kHz.
Stop to start frequency ratio is 114:9 or 13.
Determine max phase increase.
The open-loop gain will start falling above 9kHz due to the shelf network and level out above 114kHz.
Determine stage gain at midband with feedback.
Determine shelf start corner frequency ~ 1/(2π.220k.82pF) = 9kHz, assuming pentode effective plate resistance is >> 220k
Effective plate resistance may significantly reduce above nominal 9kHz due to loss of feedback.
At high frequency above shelf stop corner f then 220k//18k = 17k
Determine drop in stage gain given that internal plate resistance may have changed and external plate load has changed from 220k to 17k.
Shelf stop corner frequency ~ 1/(2π.17k.82pF) = 114kHz.
Stop to start frequency ratio is 114:9 or 13.
Determine max phase increase.
The open-loop gain will start falling above 9kHz due to the shelf network and level out above 114kHz.
Last edited:
Once the nfb loop is closed around the output transformer,
the 7kHz pole may be better for stability. It must be enough
lower than the other circuit HF poles to keep the phase shift low.
Last edited:
Amplifiers will suffer phase shift at high frequencies.
In an amplifier with feedback, that phase shift may be enough to cause the feedback signal to look like positive feedback, and once the feedback has become positive then if the gain at that frequency is greater than one, then it will oscillate.
If you just put a frequency limiting cap across the load resistor then at the frequency where the cap impedance = the anode load resistor, the gain will be about 1/2 BUT you will have also introduced 45 degrees of phase shift. That will not help a stability problem and in fact might make it worse.
The answer is the series connected resistor and capacitor across the anode load resistor. The stability network.
This reduces the high frequency gain but leaves the phase response mostly unchanged.
The half response frequency will be when the the series connection of the resistor and the capacitors impedance = the anode load impedance that is the 18K + the 82pF impedance is = the 270K anode load resistor. (or was that 220K - check above)
BUT
The 45 degree phase shift will happen at the frequency where the 82pF impedance is equal to its series connected 18K - That is at an approximately ten times higher frequency.
The thing that many get wrong is the value of that series resistor in the stability network. It should ideally be not more than 1/10th of the value of the anode load resistor so that the phase shift happens at a much higher frequency than the gain roll off. That is, to do its job properly.
Cheers,
Ian
In an amplifier with feedback, that phase shift may be enough to cause the feedback signal to look like positive feedback, and once the feedback has become positive then if the gain at that frequency is greater than one, then it will oscillate.
If you just put a frequency limiting cap across the load resistor then at the frequency where the cap impedance = the anode load resistor, the gain will be about 1/2 BUT you will have also introduced 45 degrees of phase shift. That will not help a stability problem and in fact might make it worse.
The answer is the series connected resistor and capacitor across the anode load resistor. The stability network.
This reduces the high frequency gain but leaves the phase response mostly unchanged.
The half response frequency will be when the the series connection of the resistor and the capacitors impedance = the anode load impedance that is the 18K + the 82pF impedance is = the 270K anode load resistor. (or was that 220K - check above)
BUT
The 45 degree phase shift will happen at the frequency where the 82pF impedance is equal to its series connected 18K - That is at an approximately ten times higher frequency.
The thing that many get wrong is the value of that series resistor in the stability network. It should ideally be not more than 1/10th of the value of the anode load resistor so that the phase shift happens at a much higher frequency than the gain roll off. That is, to do its job properly.
Cheers,
Ian
Last edited:
It is quite normal for genuine hi-fi to sometimes have an open-loop frequency response which is narrower than the audio band, so don't get too worried about keeping the pole up near 20kHz. A solid-state circuit could have an open-loop pole well below 1kHz!EL34Dave said:
No, the gain will be about 0.707 at the 45 degree point.gingertube said:
No, it creates a 'bump' of phase shift which could approach 90 degrees at the geometric mean of the two corner frequencies (if they are sufficiently far apart). The advantage of the CR network is that it leaves the extreme HF phase shift unchanged.
No. Reactance and resistance doesn't add like that.
That is (approximately) the second point where phase will be 45 degrees. There is also a lower point, roughly where the cap reactance is equal to the anode load resistor. In between these two points the phase will be more than 45 degrees.
I haven't tried Googling 'CR step network' but I am sure that if I did I would find lots of useful information, including a diagram of how the phase varies.
Not desiring to be catty, but I wish more designers would understand the above several posts before proceeding.
Now I must react negatively!
Yes, this is true, but that is precisely the reason for some elevated high order harmonic products present - too little NFB at high frequencies!
The ideal is to design an amplifier circuit with a flat open-loop response over the audio band, thus with NFB attenuating all audible products. This can be no mean task but is possible even with high NFB semiconductor circuits.
The phase-lead capacitor over the feedback resistor should not be forgotten, it is part of the stability matter. That can often make the difference for full audio band NFB. Again all component values will depend on the particular design and are interactive, but basics remain the same.
Now I must react negatively!
Yes, this is true, but that is precisely the reason for some elevated high order harmonic products present - too little NFB at high frequencies!
The ideal is to design an amplifier circuit with a flat open-loop response over the audio band, thus with NFB attenuating all audible products. This can be no mean task but is possible even with high NFB semiconductor circuits.
The phase-lead capacitor over the feedback resistor should not be forgotten, it is part of the stability matter. That can often make the difference for full audio band NFB. Again all component values will depend on the particular design and are interactive, but basics remain the same.
I guess it depends on what you mean by "ideal". Flat open-loop response limits the feedback which can be applied at lower frequencies, where signals are greater. The basic point is that feedback, including the open-loop response, must be designed rather than either avoided or merely slapped on as an afterthought. I am sure we can agree on this.Johan Potgieter said:
Definitely not, it will ring like a bell (unless you remove the overall nfb).
As I said, I do not have experience with ST70, but the guy who has, swears.
Did you try?
If I recall, it rings for around half a mS, at a frequency of about 8kHz.
Sounds awful that way, too. Hafler's team knew what they were doing.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.