Nice work and good detective analysis.
A couple of things puzzle me in the scope shot showing the oscillation. Firstly, why is there apparent LF roll off, the component values we have shouldn't do that at 1kHz. And secondly the initially 'ringing' looks for to low a frequency but be normal instability.
The burst of HF that comes later can be typical of wiring and layout issues (in my experience)
A couple of things puzzle me in the scope shot showing the oscillation. Firstly, why is there apparent LF roll off, the component values we have shouldn't do that at 1kHz. And secondly the initially 'ringing' looks for to low a frequency but be normal instability.
The burst of HF that comes later can be typical of wiring and layout issues (in my experience)
sim racing!
seems a waste to have a cascode beta enhanced VAS, output triple and only 22 dB 20 kHz loop gain
I think Bob overdoes the diff pair degen, I reduced it 10x
I more than make up the audio frequency linearity with 2-pole compensation, almost 40(37) dB added 20 kHz gain
which means ~ 70x less diff input V for the same 20 kHz output
and of course much more loop gain as you go lower into conventional music 3-5 kHz power bandwidth
seems a waste to have a cascode beta enhanced VAS, output triple and only 22 dB 20 kHz loop gain
I think Bob overdoes the diff pair degen, I reduced it 10x
I more than make up the audio frequency linearity with 2-pole compensation, almost 40(37) dB added 20 kHz gain
which means ~ 70x less diff input V for the same 20 kHz output
and of course much more loop gain as you go lower into conventional music 3-5 kHz power bandwidth
Attachments
Last edited:
I went ahead and did some loop gain plot with the Tian probe trying to see what the stability margin we could anticipate. Here is my observation.
1. As is
I started out based on Mooly's LTspice schematic in post #20, with an extra 4R7 damping resistor across the output coil, and an extra 0.1uF capacitor across the Vbe multiplier.
The result doesn't look too good with gain margin at about mere 10dB, and some peaking in the amplitude there about.
2. C4-R24 network removed.
Saw major improvement in gain margin, now at 35dB, plenty, yet the peaking is still there.
3. damping R-C network put in at pre-driver transistors.
Loop gain Peaking is tamed, gain margin now at 33dB, still excellent.
A few years ago I did some simulation trying to get a handle about the behavior of a 5-pair 3EF output. See this post. Was not surprised to see the notorious peaking in the high frequencies. That simulation shows how the peaking respond to different source impedance, and that it can be tamed with a R-C shunt at its input. That seems to be why the R-C network at the pre-driver transistors worked.
Amazing! The snubbers work! Seems stable at the mo, can set the bias to any value and it's fine. 1kHz square wave shows ripple but no oscillation, still needs some some work and optimising but this has done the trick! Thanks!
Some further feedback.
The base dampers are usually placed in the driver bases.
On the pre-drivers, you can insert a low value resistor (3.3 to 4.7 ohms works fine, although I have used as high as 10 Ohms). You then decouple the pre-driver collectors to ground with a good quality electrolytic capacitor of 5-10uF and a parallel film of 0.1uF.
This makes the EF3 absolutely bullet proof wrt stability. See Cordell where this is discussed in some depth.
I don't use any capacitor in parallel with the driver tie resistor anymore. Base storage in modern sustained beta devices (fT ~ 30 MHz) is low enough not to be of concern - different story of course on the old 21193/21194 devices (fT ~ 4MHz).
Re the ripple on your square wave tops - it can either be due to the output inductor OR your speaker cable resonating with some sort of capacitive load. The other possibility is that your rail localized decoupling is inadequate.
The base dampers are usually placed in the driver bases.
On the pre-drivers, you can insert a low value resistor (3.3 to 4.7 ohms works fine, although I have used as high as 10 Ohms). You then decouple the pre-driver collectors to ground with a good quality electrolytic capacitor of 5-10uF and a parallel film of 0.1uF.
This makes the EF3 absolutely bullet proof wrt stability. See Cordell where this is discussed in some depth.
I don't use any capacitor in parallel with the driver tie resistor anymore. Base storage in modern sustained beta devices (fT ~ 30 MHz) is low enough not to be of concern - different story of course on the old 21193/21194 devices (fT ~ 4MHz).
Re the ripple on your square wave tops - it can either be due to the output inductor OR your speaker cable resonating with some sort of capacitive load. The other possibility is that your rail localized decoupling is inadequate.
Last edited:
post43 mentions base stoppers for the CCS...................
Has anyone suggested a base stopper for Q6? There was a discussion somewhere on the forum on the active current source and its propensity to break into oscillation sans the base stopper in the control (Q6) transistor.........
Not directly for this amp but interesting:http://www.diyaudio.com/forums/parts/35821-some-noise-measurements-leds-zener-diodes.html
higher Vz gives lower normalized noise and with 1% R we expect better than -30 dB CM to Diff conversion in the mirrored diff pair shown at 0.3 uV in a 20kHz BW the 12V Zener noise needent be an issue
Other studies show that it's better to stay below 6v with zeners to avoid the avalanche breakdown and its noise.
See especially the summary in following article:
Noise Behaviour of Zener Diodes
More studies:
Junction Noise Measurements
Gergard tests
Morgan Jones
Last edited:
Its not just the ripple on the squarewave tops, what about the apparent lack of LF response.
To get something similar needs a response like this.
Looking at your sim diagram, the ringing is the output coil and output cap combo.
If you remove the output load cap, the ringing will stop.
Square wave slopes - will need to take another look.
I fiddled the sim (load/coil etc) to deliberately get the ringing but what I was really
aiming to show was the LF roll off, the slant that is apparent on the top and bottom
of the trace and trying to get it to look like Boscoe's actual scope shot.
Maybe the Rigol is set to AC coupling.
Maybe the Rigol is set to AC coupling.
That could be it.
wind your own is the usual diy solution
plenty of inductance calculators online
solenoid or toroidial
rember that the mag field reaches out a good distance from a solenoid's open ends
this can couple to circuit wiring loops, and ferromagnetic material interacting with the external solenoid field can cause distortion
plenty of inductance calculators online
solenoid or toroidial
rember that the mag field reaches out a good distance from a solenoid's open ends
this can couple to circuit wiring loops, and ferromagnetic material interacting with the external solenoid field can cause distortion
the scope has been set to DC coupling.
Then either the input coupling capacitor or the unity gain nfb capacitor is undersized,
assuming your generator waveform looks normal direct.
I did not see any coupling caps - assumed it was DC coupled.
The actual circuit built (before all this) is the attachment in post #1.
From the tilt, it seems to be around a 100Hz corner.
If the tilt is not there just after the input coupling cap, then the
problem must be the nfb cap. Maybe it's 10uF instead of 100uF.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
- Home
- Amplifiers
- Solid State
- Please help with issues I'm having with my Bob Cordell amplifier implementation.