I use a 1000uF 35V for C3. There is <1mV of DC offset. I'm not sure what you're worried about.
The cap that blew for me was a 35V bypass capacitor (C9). And it was because I had forgot to replace them with higher voltage types when I installed the proper power supply that I am now using (+-40V). Stupid mistakes are made when you're more focused on bending leads of Q12 and Q13 to rearrange the pins for the PCB (KSC1845F - Higher Vceo, different pinout than BC546).
I have monitored DC offset continuously throughout cold and warm temperatures, a wide range of output volume, and it stays around 1mV or less. I do use a matched input pair. I have been running it for a while now.
The subject certainly deserves some attention.
Here, this capacitor is protected by back to back diodes: one limits the reverse bias to a maximum of 0.5~0.6V, and the other prevents the eventuality of a freak event triggering a latch-up.
Regarding the polarity, the bias current from the transistors clearly shifts the odds in the good direction, and it is unlikely a mismatch in the input transistors could reach more than 25mV, even with unmatched devices.
At temperatures of ~25°C, aluminium electrolytics can tolerate 0.5V of reverse bias indefinitely. Not something I would reccommend, but in any case we are very from this value here.
Even if the capacitor leaks, its leakage resistance would need to go below ~10K to begin to have adverse effects, and anyway, if the worst happens and the cap becomes shorted, the offset voltage will be multiplied by the gain of the amplifier, which should normally not damage the speaker.
I am not sure applying the "correct" bias to back to back capacitors is a good idea: such large capacitors have a non-zero leakage current, in the nanoamps range, and this current tends to fluctuate randomly.
Applied to the 10K feedback this current could inject some VLF noise into the amplifier, making the cure worst than the problem it is supposed to address.
As for the film cap in parallel: actual measurements tend to show adverse effects rather than improvements: see here for actual data:
http://www.diyaudio.com/forums/powe...lm-caps-electrolytic-caps-17.html#post2257381
Heavy Class A, where? This crosses only 1A, for 3A peaks.
Granted, never allowed turn off, but only 150mA so wasted.
Cooking 25W per 2n3055 at idle. 50W a $$$ burning fireball?
Might look JLH at first glance, but a much tamer AB variant.
Maybe you didn't spot shunt regulated current source Q8?
Classic
overbias in the center is solved here...
Q8 is my "second control loop in quadrature" we spoke of earlier.
While the main loop is hopefully holding linear control over voltage,
Q8 is holding Schottky curved control over output stage currents.
The objective to enforce quasicomplimentary square laws...
Granted, never allowed turn off, but only 150mA so wasted.
Cooking 25W per 2n3055 at idle. 50W a $$$ burning fireball?
Might look JLH at first glance, but a much tamer AB variant.
Maybe you didn't spot shunt regulated current source Q8?
Classic
overbias in the center is solved here...Q8 is my "second control loop in quadrature" we spoke of earlier.
While the main loop is hopefully holding linear control over voltage,
Q8 is holding Schottky curved control over output stage currents.
The objective to enforce quasicomplimentary square laws...
Last edited:
Mmm... quadrature with a limp then: for Q8 to act on Q7's current, it first has to change Q6's current, which is sensed by the FB because it alters the output level, and makes the FB loop restore the equlibrium via the (noble) horizontal path.
Quite the opposite: it is very clever and elegant.
But IMHO, it looks like a beautiful solution to a non-existent problem....
I simulate 0.86A upper, and 1.07A lower, idle...
Base current helps the upper work the load, so
collector current asymmetry is to be expected.
True crossing isn't exactly where the collectors
in the first screenshot appear to cross equal.
But Schottky currents are perfectly symmetrical.
Nothing deceptive about the second screenshot.
Base current helps the upper work the load, so
collector current asymmetry is to be expected.
True crossing isn't exactly where the collectors
in the first screenshot appear to cross equal.
But Schottky currents are perfectly symmetrical.
Nothing deceptive about the second screenshot.
Attachments
Last edited:
Here is a quick, informal test I improvised to check the effect of bias on E-lytics.
The capacitors were standard, run of the mill 470µ/16V, new and unused, 5~10 years old.
Prior to the test, they were reformed at 16V during half an hour.
They were then connected to the test set-up, with the switch in position 1.
The output was monitored on a digital scope, speed 20ms/div, vertical sensitivity 100mV/div.
When the trace was centered in a division, the trace never crossed the lines during 1~2 minutes observation.
The switch was then turned to position 2. After stabilization and offset adjustment to correct the effect of the leakage current (not shown on the schematic for clarity), the trace was observed again, in the same conditions.
This time, there were moderately frequent crossings of the 100mV limits.
As an unscientific estimation, I would say the amplitude was increased by ~25%.
The set-up was left in the same condition for 2 hours, and the trace observed again.
This time, crossings still occured, but very infrequently compared to two hours earlier.
My conclusion is that there is a minor but observable increase in the VLF noise when bias is applied to an electrolytic.
This of course is a one of observation, with one model of cap, and for a limited period of time, and it is partly qualitative.
It could be that differences vanish completely after two weeks, but in some conditions, there is a detectable effect (I hope this will not bring grist to the mill of "break-in adepts").
The capacitors were standard, run of the mill 470µ/16V, new and unused, 5~10 years old.
Prior to the test, they were reformed at 16V during half an hour.
They were then connected to the test set-up, with the switch in position 1.
The output was monitored on a digital scope, speed 20ms/div, vertical sensitivity 100mV/div.
When the trace was centered in a division, the trace never crossed the lines during 1~2 minutes observation.
The switch was then turned to position 2. After stabilization and offset adjustment to correct the effect of the leakage current (not shown on the schematic for clarity), the trace was observed again, in the same conditions.
This time, there were moderately frequent crossings of the 100mV limits.
As an unscientific estimation, I would say the amplitude was increased by ~25%.
The set-up was left in the same condition for 2 hours, and the trace observed again.
This time, crossings still occured, but very infrequently compared to two hours earlier.
My conclusion is that there is a minor but observable increase in the VLF noise when bias is applied to an electrolytic.
This of course is a one of observation, with one model of cap, and for a limited period of time, and it is partly qualitative.
It could be that differences vanish completely after two weeks, but in some conditions, there is a detectable effect (I hope this will not bring grist to the mill of "break-in adepts").
Attachments
Keep LTP collector voltages equal? Maybe useless, I dunno...
Not as-if I had any actual "thought process" going on.
To the naked eye, it looks random, but I'd need to make a Fourier analysis to be positive.
Anyway, most of the noise is caused by the opamp, and this is random, and it certainly has the capacity to mask the characteristics of the capacitor's noise.
If I find some time next week, I'll try to make further tests and also use a lower noise amplifier than the LF411.
Member
Joined 2009
Paid Member
Elvee, I knew I liked this design but as I start to realize how this works, well now I'm starting to consider it as the closest solution to a long search I've been on. I've been looking for a non-switching Class AB without switching x-over distortion (whatever that means). I've sworn off LTPs for various reasons but the way they are employed here with insensitivity to device matching gets my thumbs us. It also avoids all the limitations of matching complementary devices at the output too.
However, I was also interested in a hybrid - tube VAS up front. I wonder if it would be possible to use this (circlophone) approach to realize a unity gain power buffer without gnf (except for the bias servos).
edit: p.s. can you post the LTSpice file for this ?
However, I was also interested in a hybrid - tube VAS up front. I wonder if it would be possible to use this (circlophone) approach to realize a unity gain power buffer without gnf (except for the bias servos).
edit: p.s. can you post the LTSpice file for this ?
Last edited:
Unity buffer is very simple with JLH+SRPP, merely change Q1 from NPN to PNP.
2n2905 or MJE350 might work...
Other methods described here return common mode feedback to the tail of a
LTP. Are gonna have a much harder time if a LTP isn't there (sworn off why?)...
Of course, if you made a LTP of a twin triode, you could still use any of the
other solutions. Its an autobias for the triodes as well.
2n2905 or MJE350 might work...
Other methods described here return common mode feedback to the tail of a
LTP. Are gonna have a much harder time if a LTP isn't there (sworn off why?)...
Of course, if you made a LTP of a twin triode, you could still use any of the
other solutions. Its an autobias for the triodes as well.
Member
Joined 2009
Paid Member
but this is Class A only though right ?
well it started off because I needed a change, seems LTPs are everywhere. But Then there's the question about the accuracy of the LTP as an error amplifier relying on matching of two devices when a Singleton has no such limitation. And the LTP sacrifices balance between odd and even harmonics in favour of odd harmonics. Lastly, LTPs require more compensation capacitance than the Singleton and compensation is an ugly word most places. The 'series LTP' or Rush cascode seem a better option.
But I have to agree, the LTP does a couple of things really nicely - dc-offset and thermal drift are nicely handled.
Unfortunately, no p-type triodes available...
Ken and I have posted a number of circuits that might more or less fit your requirements ( have you looked at Unigabuf, that's a really quirky one).
I also have a huge number of other ones in my archives, I'll have to dig to see whether I can find something more appropriate.
But be aware that a good part of the "magics" of the Circlophone finds its origin in the association of same sex OP transistors with drivers having an horizontal symetry (LTP's), and a common current servo.
The result is transcends the mere sum of the ingredients, and if you remove one of them, the magic will not work.
With the LTP's, the current servo can act on both output devices simultaneously, and doesn't interfere at all with the signal path.
It would be difficult to achieve the same result otherwise (vertical [N-P] symetry could in principle do it but practical results are disappointing).
I'll think about it anyway.
Here is the .asc file
Member
Joined 2009
Paid Member
Thanks for the file Elvee.
Don't get me wrong - I'm not suggesting removing the LTP from this design - I don't see how it would work without it. What I like is that you don't find the device balance in the LTP to be critical - which I thought it was. Or do you mean it's not critical for operation of the common-mode servo but is still critical for good accuracy of gnf voltage feedback error amplifier ??
Don't get me wrong - I'm not suggesting removing the LTP from this design - I don't see how it would work without it. What I like is that you don't find the device balance in the LTP to be critical - which I thought it was. Or do you mean it's not critical for operation of the common-mode servo but is still critical for good accuracy of gnf voltage feedback error amplifier ??
Device balance in the driver (or VAS or predriver, call it what you like, it doesn't fit conventional classification, in short the lower LTP) is completely unimportant.
In the input stage, it is like any other amplifier for one thing: the offset voltage will be copied unchanged at the output, but unlike conventional amplifiers, the unbalance will not cause further degradation, the linearity will remain unaffected.