100nF load stability issue
Marcin Slawicz
about its hf stability.
This is a Williamson circuit with UL power stage and somewhat modified front
end (John Broskie's Aikido style voltage stage with concertina phase
splitter).
The first stage is ECC82 (V1) with ~600 ohm cathode resistor and 22k plate
resistor. The amplifier open loop gain is 88.5 @ 1 kHz (38.9 dB).
Feedback loop is Rf=9.88k connected to V1 cathode (15.7 dB).
The amplifier has a toroidal OPT with estimated f3=70 kHz (measured by
someone else, however I don't know which way).
So, the lowest pole in hf range comes from the OPT (lets assume 70 kHz).
The second pole is from driver stage: 110 kHz (6SN7, gain 16, driven from
concertina plate side Rout=21k).
The third pole is at input stage: about 800 kHz.
Next poles come from output 6L6GC (1.6 MHz) and concertina splitter (3 MHz),
and can be probably ignored in a well stabilized amplifier.
According to my estimations, the uncompensated amplifier has dominant pole
(at least with nominal resistive load) at 70 kHz and the next one at 110
kHz. At 200 kHz the loop gain drops to unity with the phase margin about 30
degrees. The amplifier should be conditionally stable.
Indeed it is. However with anything but the nominal resistor at the
output the circuit starts to oscillate.
Then I used two compensation circuits:
RC lag compensation (R1=4k7, C1=390p) parallel to V1 plate resistor. This
should put a new dominant pole at f=23 kHz and zero at 110 kHz.
Lead compensation - C2=56p across feedback resistor. This should put a new
zero at 130 kHz and a pole at 11 MHz. The main goal is to dump the square
signal overshoot and ringing.
These two networks should give me a dominant pole at 23 kHz, the next one at
70 kHz from OPT (however this one can move), and the third one at 110 kHz
compensated with zero from the lag compensation network. I expect the unity
loop gain at about 90 kHz with something like 50 degrees phase margin.
The amplifier stability should be OK, however real life is much more
complex, so I wanted to test the stability the way Patrick has advised so
many times.
I used a 10k pot as the R1 and prepare the circuit to be able to easily
change both capacitors during the tests. I also used 5 kHz square wave
generator for testing.
My results are as follows:
1) 8 ohm resistive load - no stability problems.
2) No load - tested with C1=130p, 390p, 1n. Stable with R1<6k for low output
amplitude, R1<9k for high output amplitude.
3) Speaker load - stable with R1<5k5.7k5 (output amplitude dependent).
Optimal R1=4.5k.
4) Capacitor 0.22.1 uf load - stable with any used value of C1 and R1.
Optimal R1=5k5.10k (output amplitude dependent).
5) Capacitor 100n load - stable with R1<3.3k5 only.
6) Capacitor 47n load - stable with R1<1k5.2k only!
7) Any capacitor parallel with 39 ohm (or lower) load - stable with any used
R1 and C1 value.
The amplifier stability depends most on the R1 value.
C1 and C2 exact value much less contribute to the overall stability, however
C2 helps to dump pulse overshoot and ringing. Too much C2 leads to
instability though.
The values I estimated before testing (R1=4k7, C1=390p, C2=56p) seem to be
right and provide good amplifier stability with almost any kind or without
load.
The only problem is the possibility of oscillations with the 47nF or 100nF
cap as a sole load.
Should I worry about it?
Is the amplifier stable enough?
I should add, that my speakers the amplifier is to be working with, are a
kind load for it. Their impedance peak is only 21 ohm (bass region), and 7.9
ohm with any frequency above 200 Hz.
Cheers,
Marcin
Patrick Turner
Marcin Slawicz wrote:
You have more idea about the numbers than i have.
I don't bother working out all the poles or
determining which is dominant.
But you want the final full power bandwidth to be
from say 15Hz ( governed by OPT saturation )
to say at least 50kHz with a pure R load
with your NFB connected.
You will find that at 1/10 the full possible clipping voltage output
you should get about frokm 3Hz to 60kHz with an R load.
Cap loads load an amp with a load value at HF which always falls below
the rated nid F load.
So never expect a 50 watt amp that gives 50 watts into 8 ohms at 1 khz at clip
to give 50 watts at 50 kHz into 2 uF, because 2uF is an impedance
way below 8 ohms at 50kHz.
So therefore the stability tests and HF damping networks should always be set up
and tested
at 1/10 the max Vo for your amp;
you get less smoke that way.
But to a flat response into
any value of purely capacitive load is impossible
due to the increasing phase lag caused by the cap and its interaction
with the leakage inductance of the OPT.
Using a sine wave there will *always* be some square wave overshoot
with any value of cap load between 0.2uf and 5 uF, even when the amp is
optimally
damped and set up for the best R load bandwidth AND stability at any levels.
When tested with a sine wave many amps at low levels where the open loop
gain is highest because you are away from thre lower gain Ab condition
will give a peak in the sine wave response with a cap only at the output.
Typically you get a 6 db peak at 35 kHz before the response dies away at
12dB/octave
when using a 2uF cap in many good amps.
With 0.47uF, the peak will be maybe at 50kHz, 0.22 uF maybe at
65kHz; the response should be plotted for a range of cap values.
A 6 db peak is OK; it is confirmed by the square wave overshoot.
But as long as oscillations don't happen with any value of cap,
the amp is stable.
The frequency of the peak in the sine wave response becomes lower
as the C value rises, and as long as there is no rise in the response
of 1dB in the F range under 20kHz your amp is OK.
Many amps will produce a peaked response at 25kHz with 2 uF
and this indicates the leakage inductance is too high,
and there is a 3db rise in response by the time you reach 20kHz.
I usually set the amp up with a 0.22 uF cap as the sole load.
If it oscillates immediately with FB connected, i add
a small C across the FB resistor.
It may still oscillate, so I will then connect
a radio tuning cap and a pot across the anode load for V1,
and usually oscillations will stop.
I then adjust the variable RC values of the zobel network to tune the square
wave response for the minimum
ring or overshoot.
I add more C across the FB resistance, and re-do the zobel.
I then chech the sine wave response into a resisance load.
It may be too low, so more trials of FB capacitance and zobels may be needed.
I sometime place zobels across the output, 10ohms plus 0.22 uF is common,
and across wach 1/2 primary to limit the gain of the output stage at HF and to
damp the resonances of the OPT which occur at HF, something
very difficult to account for in your figurings and measurements.
Its impossible to construct a perfectly representational model
for all the LCR elements at work in a tube amp.
See my pages at http://www.turneraudio.com.au for some circuit ideas I have
used.
Solely capacitor loads should never cause more than 6 db of peaking
in the sine wave response if all has been optimised.
If there is more than 6db of peaking the amp has a low margin of stability.
Usually using a cap and an RL of the rated amount will reduce any peaking
seen when only a cap is used.
But you cannot gurantee the amp will always see an R load at HF because
many speakers are inductive as F rises to their impedance
rises to a value that is like not having any load connected at all,
and thus I like t have zobels across the windings of the OPT
to stop a rise in open loop gain at above 50kHz.
The amp is effectively a bandpass filter, and one that has a steep cut off slope
when a cap load is connected, and one with a NFB loop,
so the phase shift and gain must be tailored for stability.
And you will rarely ever get a square wave with no ring into a BPF or LPF
which has a slope greater than 1st order, or 6dB/octave.
A *brief* test at full power into 2uF should be tried.
This tests the saturation behaviour of the input/driver tubes.
With 2uF, the output tube gain will fall as F rises, so the input stages
have to provide more drive voltage to maintain the output voltage levels
until it becomes impossible due to grid current etc.
Brief means not long enough to allow the much increased tube dissipation
at HF to last long enough to ruin a tube.
Patrick Turner.
Marcin Slawicz
Uzytkownik "Patrick Turner" <in...@turneraudio.com.au> napisal w wiadomosci
news:436CE8BD...@turneraudio.com.au...
>
> When tested with a sine wave many amps at low levels where the open loop
> gain is highest because you are away from thre lower gain Ab condition
> will give a peak in the sine wave response with a cap only at the output.
> Typically you get a 6 db peak at 35 kHz before the response dies away at
> 12dB/octave
> when using a 2uF cap in many good amps.
> With 0.47uF, the peak will be maybe at 50kHz, 0.22 uF maybe at
> 65kHz; the response should be plotted for a range of cap values.
> A 6 db peak is OK; it is confirmed by the square wave overshoot.
Yes, I have to plot my sine wave response for a range of cap values. But the
square wave overshoot is, in my opinion, quite bearable. For 0.22 uF output
cap it is something like 10% amplitude without lead compensation and about
3-5% amplitude with optimat lead compensation (in my case C=47...68 pF). The
ringing usualy lasts for about 5 periods.
> But as long as oscillations don't happen with any value of cap,
> the amp is stable.
And here is my problem. The amp is stable with 0.22 uF or more at the
output. But if I want to get rid of oscillations with 0.1 uF as a load, I
have to lower my V1 Zobel resistor to 3k, and with 47 nF at the output,
further lower it to 2k or less.
Should I do it? Or maybe the 0.22 uF stabilization is enough?
Williamsons usually have this Zobel resistor as low as 1/10 of the V1 plate
resistor (i.e. 4k7). In my case it would be 2k2 then, so maybe I should do
it? According to my tests it won't deteriorate stability with any kind of
load.
>
> I usually set the amp up with a 0.22 uF cap as the sole load.
> If it oscillates immediately with FB connected, i add
> a small C across the FB resistor.
Are your amps rock stable with lower caps too?
>
> I sometime place zobels across the output, 10ohms plus 0.22 uF is common,
I have it too, but 47 ohms + 0.1 uF.
> But you cannot gurantee the amp will always see an R load at HF because
> many speakers are inductive as F rises to their impedance
> rises to a value that is like not having any load connected at all,
> and thus I like t have zobels across the windings of the OPT
> to stop a rise in open loop gain at above 50kHz.
>
My speakers have between 7 and 9 ohms at any frequency above 200 Hz (at 50
kHz too; I know, I have made them myself ;-)
But I agree. I want to make the amplifier which is happy with (almost) any
speaker.
Thank you very much Patrick. You are my real vacuum electronics master.
Marcin
Marcin Slawicz
Uzytkownik "Patrick Turner" <in...@turneraudio.com.au> napisal w wiadomosci
news:436CE8BD...@turneraudio.com.au...
>
> But as long as oscillations don't happen with any value of cap,
> the amp is stable.
>
According to Patrick's advise I decided to make my amp really
unconditionally stable. I realized there was too much sine wave hf peaking
somewhere above 50 kHz with various capacitive loads. So I decided to
further limit my open loop response to about 10 kHz. I tried to use 680p or
even 1n as the Zobel capacitor in the V1 plate circuit. This is unusual high
value, however I have only 22k plate resistor (less than most Williamsons
use - 47k).
Open loop limit frequency (resistive load) is now 8 kHz or 11 kHz (Zobel
capacitor 1 nF and 680 pF respectively).
Closed loop (1/10 clipping amplitude, R load) freq is 66 kHz and 75 kHz
respectively.
I have to measure full power bandwidth yet.
The most important: if I use V1 Zobel resistor less than 4k7 the amp is
stable with ANY tested load (33nf, 47nf, 100nF, 0.22uF, 0.47uF, 1uF, 2.2uF
included).
Using sine wave the amp has following peaks (V1 Zobel cap=1nF):
1uF load: 0.6dB @ 30kHz and 1dB @ 72 kHz
0.47uF load: 0.3dB @ 35kHz and 0.1dB @ 80kHz
0.22uF load: -2.5dB @ 185kHz (-4.5 dB dip at lower freq)
With V1 Zobel cap=680pF there are higher peaks:
2u2 load: 1db
1uF load: 4 dB
0.47uF load: 3.1 dB
There were much higher peaks (more than 6 dB) with V1 zobel cap=390pF used
before.
Which Zobel cap is better in your opinion?
Isn't the amp overdamped with the 1 nF cap?
Is 10 kHz open loop bandwidth in the 16dB feedback amp enough?
There is no rise in the response of more than 1dB in the F range under
20kHz.
The square wave response looks OK. Optimal V1 zobel R value seems to be 2k7.
R=4k7 gives a little bit less overshoot, but more ringing appears, and there
is less capacitive load stability margin.
Than you for your last advise and look forward for your next comment.
Marcin
luc.de...@free.fr
I made a williamson (gec30w improved a bit) some months ago (and im'
finishing the second one for stereo).
I studied the prototype circuit for half a year and studied of course
the HF stability.
I used a 6.8R as a load with a 0.1mfd to 0.47mfd.
Like you are doing i studied the feedback stability :
I listened for hours 470p and 1n in the feedback loop for HF stability.
And i prefered the 470p more 'alive' sound.
If i remember well with the oscilloscope the rising time with a square
wave generator at 20khz was a bit long with 1n.
But the stabiity was better and less overshoot.
How about your LF stability ?
i hope this helps,
BR,
Luc D.
Marcin Slawicz
Uzytkownik <luc.de...@free.fr> napisal w wiadomosci
news:1131627569....@g14g2000cwa.googlegroups.com...
> Hello Marcin,
>
> I made a williamson (gec30w improved a bit) some months ago (and im'
> finishing the second one for stereo).
> How about your LF stability ?
>
Well, it's OK now. I have applied all methods Patrick has described many
times in the RAT.
First of all I have worked out the V1 zobel capacitor. The value of 680 pF
gave me no more than 6 dB sine wave peak with any value of capacitor load
(actually max 4.7 dB @ 71 kHz with 1 uF load). There is no more than 1.5 dB
peak in the f<20 kHz band.
Next I found the V1 zobel resistor value which gave me no oscillations with
any kind of load and without load. It was 4k max, so I used 3k resistor, for
which I got good overshoot and ringing damping.
In the last step I checked, if my feedback shunt capacitor (47 pF) was OK.
It was. Too much feedback capacitance (over 100 pF) lead to instability in
certain conditions.
The hf stability was OK then. Some days ago I also improved the lf stability
by raising the last RC coupling frequency to 16 Hz.
Now my amplifier:
- Doesn't oscillate at lf and hf without any load connected
- Doesn't oscillate with any value of inductance connected
- Doesn't oscillate with any value of capacitance between 10 nF and 10 uF
connected
- Doesn't oscillate when the square wave signal is used in any of above
situations
- Doesn't produce bursts of hf oscillations when the voltage is sufficient
to saturate the OPT at low frequencies.
The bandwidth is (OPT saturation at LF; -3 dB at HF):
6 Hz to 75 kHz @ 1 W
15 Hz to 68 kHz @ 20 W (almost full power)
This is all I wanted to achieve. I will try to do more testing next days.
You can see my detailed report at
http://www.echostar.pl/~slawicz/concertino/concertino1.htm
Although in polish, there are a lot of pictures and diagrams there.
Marcin