♫♪ My little cheap Circlophone© ♫♪

the Ccb participates to the overall compensation, and if it disappears something else will have to make for it.

This is not necessarily true. Ccb is miller compensation. The VAS compensation is shunt compensation. One probably works better than the other, and has less performance cost. Furthermore, the application of Miller compensation to the VAS when not cascoded is lopsided, creating an imbalance which causes Q3 to output more current, with Q4's gain lowered by the miller.

Dan has the most recent sims, maybe he can do a quick demo.
 
More specifically, 8x lower distortion, than without a cascode.

OK, I made a quick test: without any cascode, distortion is 0.0045% (+/-20V supply, 32Vpp out on a 4Ω load).
With a theoretically ideal cascode, it becomes 0.0035%
With Daniel's cascode, it is 0.0044%
With Ken's cascode, it is 0.0040%

Ken's version is marginally superior, but more importantly it doesn't harm anything and plays the role of the zener.
 
In my simulations the improvement was at most 2x less THD. Maybe Dan's figures are from his high-bias simulations... The effect depends a lot on VAS component, implementation and so on. I noticed in Dan's schematic the VAS was cascoded at 16V - which could explain his better THD specs, but this really limits the negative swing. VAS devices which aren't overpowered don't benefit much from high cascode voltages (hint hint). May I suggest what is potentially the most reasonable, well-performing and hazardless option, attached.

Maybe NXP has models for the BC337-40, I haven't checked. The BC550C might work better but with less Ic margins.
 

Attachments

  • Circlophone_FullCascode.png
    Circlophone_FullCascode.png
    119.7 KB · Views: 524
OK, I made a quick test: without any cascode, distortion is 0.0045% (+/-20V supply, 32Vpp out on a 4Ω load).
With a theoretically ideal cascode, it becomes 0.0035%
With Daniel's cascode, it is 0.0044%
With Ken's cascode, it is 0.0040%

Ken's version is marginally superior, but more importantly it doesn't harm anything and plays the role of the zener.

does it really matter? can you actually hear the difference?
 
I like to make it a practice to go for a superior design even in the face of diminishing returns when cost is not prohibitive. After all, how do we get better at what we do when we aren't striving for improvement? There has to be some testing ground for creativity and experimentation. It may not sound better but a gain in skill is always a good thing, and helps us in the other endeavours where it actually matters.
 
Hello All,

After all that brain storming, I need to ask something about what's going on..

* CFP's drawbacks seem can't be deflated without heavy compensation?
* Buffered approach makes Circlophone showing its real capabilities?
* Is two transistors VAS approach what we need after all about?

In my simulations the improvement was at most 2x less THD. Maybe Dan's figures are from his high-bias simulations... The effect depends a lot on VAS component, implementation and so on. I noticed in Dan's schematic the VAS was cascoded at 16V - which could explain his better THD specs, but this really limits the negative swing. VAS devices which aren't overpowered don't benefit much from high cascode voltages (hint hint).

Is this for buffered variations from Daniel? If not, could you elaborate that why your latest pick hasn't any buffer please? I mean, which one seems best fit for you, playing with LTP or VAS?

May I suggest what is potentially the most reasonable, well-performing and hazardless option, attached.

Your involvement with this topic brings some interesting ideas for topology. That's what I like about benefits of innovative spirit.

Also, dunno if you know this, in LTSpice, Tools->write bitmap to clipboard and then paste into MSPaint will give you an instant schematic with no need for cropping window borders. Too bad it doesn't work in Linux.

Tools -> Write to a .wmf file works under Linux for exporting schematics. An average image viewer application can open and export to other formats if desired.
 
Hello All,

After all that brain storming, I need to ask something about what's going on..

* CFP's drawbacks seem can't be deflated without heavy compensation?
The correction I have given is not too heavy, and gives an acceptable transient response.
* Buffered approach makes Circlophone showing its real capabilities?
Buffering doesn't change the basic Circlophone's capabilities, but it allows for higher source impedance.
If you stick with low impedance (<5K) sources, the "natural" Circlophone remains the best option
* Is two transistors VAS approach what we need after all about?
That's a completely different issue. Cascoding brings some marginal linearity benefits, but has no impact on the input requirements or characteristics.
 
And anyway, by accumulating numbers of marginal improvements, you end up with better designs.
1
Keantoken's buffer, last seen at the input in post 1016 is one such improvement for enhanced tolerances and everything easier and probably safer too, since substitute parts have less problems if you've got the buffer, and you can run the amp almost as fast as the outputs if you want to set it up that way. Of course it isn't for copying post 1; however, when operating voltage and part selection starts to vary (such as for power enhancement and/or parts availability), then that is where the buffer is mighty helpful. Not everyone has either the landlord next door or 101db efficient speakers, so we could safely assume that individual builds are expected to vary and therefore we could also assume that the buffer is a good thing. Right?
2
Earlier, you said you were looking for something to make the sensor easier or more tolerant. On that same schematic from post 1016, the sensor-anti-freakout cap C18 is another possibly useful improvement and it can be set the same value as (or larger than) C1 in most cases. It should allow a wider range of values and components without misfiring the sensor. What do you think of it?
CFP's drawbacks seem can't be deflated without heavy compensation?
Not true, but you can say it this way: CFP's drawbacks can't be deflated without specific compensation.
And this means that either you have a scope handy or else you make an exact copy of Elvee's schematic at exact same voltages with exact same parts. Unfortunately, substituting makes the compensation invalid, and the CFP version isn't as tolerant. SO, if the published modestly low power version isn't what you need, then you're up that proverbial creek without a scope.
The CFP version needs additional individualized schematics (possibly numerous) representative of each change, such as voltage and vas parts will need little adjustments for success. . . which would be a different schematic every time, but that is the way to do a CFP build without a scope, if someone already used a scope to verify that exact same build at the exact same voltage and with the exact same parts, thereby identical copies work.
In a nutshell, the CFP build requires tighter tolerances.
Is this for buffered variations from Daniel? If not, could you elaborate that why your latest pick hasn't any buffer please?
The buffer is Kean's, as you can tell by the BC560C army doing an excellent job at small signal. Its removal from the schematic was my fault--I removed it because the buffer is too helpful to the simulator. The buffer helps tolerances (opposite of the CFP), and I did not want the buffer's help while determining component values. Instead, I wanted to work with the values and then re-add the buffer again, at the last moment, in order to maximize its effectiveness. The 2 transistors, 1 cap and 3 resistors should be easy to add at any time.
 
Last edited:
1
Keantoken's buffer, last seen at the input in post 1016 is one such improvement for enhanced tolerances and everything easier and probably safer too, since substitute parts have less problems if you've got the buffer, and you can run the amp almost as fast as the outputs if you want to set it up that way. Of course it isn't for copying post 1; however, when operating voltage and part selection starts to vary (such as for power enhancement and/or parts availability), then that is where the buffer is mighty helpful. Not everyone has either the landlord next door or 101db efficient speakers, so we could safely assume that individual builds are expected to vary and therefore we could also assume that the buffer is a good thing. Right?
I repeat: the buffer changes essentially one thing: the input impedance.
It also very marginally degrades the linearity, but that's completely anecdotal, and at source impedances above 1K, this degradation will be cancelled and reversed.
Other than that, the buffer changes nothing, unlike the CFP which also adds gain.

2
Earlier, you said you were looking for something to make the sensor easier or more tolerant. On that same schematic from post 1016, the sensor-anti-freakout cap C18 is another possibly useful improvement and it can be set the same value as (or larger than) C1 in most cases. It should allow a wider range of values and components without misfiring the sensor. What do you think of it?
Tampering randomly with the schematic can be very dangerous:
*The sim only shows you what you look at
*The sim tells only half of the story: discrepancies between simulation and reality do persist, and you could have nasty surprises when you build what you have simulated.

As an illustration, see below what you destroy by slowing down the quadrature loop by a factor ~100 with C18: at frequencies as low as 10KHz, Q3 shuts down completely, thereby ruining completely the non-switching OP.
You gain a marginal improvement somewhere, but you break completely something else far more important.

This is not to say there is no room for improvement, everything is always perfectible, including the Circlophone, but I did my homework, and if you change something you have to be extremely careful not to damage something.

Not true, but you can say it this way: CFP's drawbacks can't be deflated without specific compensation.
And this means that either you have a scope handy or else you make an exact copy of Elvee's schematic at exact same voltages with exact same parts. Unfortunately, substituting makes the compensation invalid, and the CFP version isn't as tolerant. SO, if the published modestly low power version isn't what you need, then you're up that proverbial creek without a scope.
The CFP version needs additional individualized schematics (possibly numerous) representative of each change, such as voltage and vas parts will need little adjustments for success. . . which would be a different schematic every time, but that is the way to do a CFP build without a scope, if someone already used a scope to verify that exact same build at the exact same voltage and with the exact same parts, thereby identical copies work.
In a nutshell, the CFP build requires tighter tolerances.
No, that is not really the case: the CFP version has good stability margins, and is as tolerant as the base version.
However, even with optimum values, the transient response is less clean, but this has nothing to do with sensitivity.
The buffer is Kean's, as you can tell by the BC560C army doing an excellent job at small signal. Its removal from the schematic was my fault--I removed it because the buffer is too helpful to the simulator. The buffer helps tolerances (opposite of the CFP), and I did not want the buffer's help while determining component values. Instead, I wanted to work with the values and then re-add the buffer again, at the last moment, in order to maximize its effectiveness. The 2 transistors, 1 cap and 3 resistors should be easy to add at any time.
If the buffer provides improvement, that's the symptom something is going wrong somewhere, see above.
 

Attachments

  • Circlop10K.png
    Circlop10K.png
    121.6 KB · Views: 520
Ah apparently, I had thought about a switching subwoofer amp (and it is possible to make the current precisely track the waveform). My bad!
That cap, C4, should probably work if it is small enough, perhaps somewhere near 330p; however, it is also possible that it is unnecessary. Thank you for checking it. No need to spend more time on that particular cap. We'll try something else more suited to non-switching amp. :)
Thanks again!
 
0.0013% thd

Kean worked hard to straighten me out and then we worked for most of the night on this new simulation. I didn't manage to contribute much other than some device selection. The drivers could be BD140, KSA1220 or 2SA1930 with practically the same results, and many different parts are usable for vas. See attachment.
 

Attachments

  • Circlo-Kean.gif
    Circlo-Kean.gif
    25.4 KB · Views: 514
Kean worked hard to straighten me out and then we worked for most of the night on this new simulation. I didn't manage to contribute much other than some device selection. The drivers could be BD140, KSA1220 or 2SA1930 with practically the same results, and many different parts are usable for vas. See attachment.
Here again, fiddling with the quadrature loop has nasty side effects: since C2 is larger and has no series resistor, the loop becomes unstable and requires phase advance capacitors C21 and C22 to be stable again, and the gain of the loop at high frequencies is ~0 which means the effectiveness of the correction decreases quickly with frequency to the extent that the non-switching action lost from several KHz.
 
Did you try it? C22 has the same effect as a resistor in series with C2. C21 helps with stability and makes switching smoother for the lower output. In my simulations the version with 220n for C2 and a 68R series resistor was unstable and the lower output had strange overshoot. This version seems better behaved, even to several KHz.
 
Combination approach works

Here again, fiddling with the quadrature loop has nasty side effects: since C2 is larger and has no series resistor, the loop becomes unstable and requires phase advance capacitors C21 and C22 to be stable again, and the gain of the loop at high frequencies is ~0 which means the effectiveness of the correction decreases quickly with frequency to the extent that the non-switching action lost from several KHz.
Thank you sir.
Error: The simulator had (again) assumed an absolutely perfect cap at C2.
I returned C1.
There was no need to remove it since it is helpful for tolerance if it isn't too big. After re-deploying C1 (100p or smaller for reduced glitching), then some of the new compensation was oversize. C2 went to 330n (range 220n to 470n). Not shown, C21 went to 470p.

I got these values worked out from adding a resistor series to C2 to represent a lossy imperfect cap. The sim freaked out with instability as usual, whenever that resistor is present. Then I added capacitance at C1 until even a lossy red polyester dip cap could work at C2 (and I hope we use a better cap instead). And lastly, I removed the destabilizing resistor from C2.
Success!!
Practical cap losses were swamped and surely no need to add any. With the sim set at 1k, 10k, and 20k, it is the same each time--no switching. Distortion raised to 0.0014% but stability should be like you wanted (see attachment)? If that won't do it, try C1 at 100p which is almost big enough to be bad but rather 100p there is the make break point. 47p is less distortion.
Here it is:
(a streamlined combination of both compensations)
 

Attachments

  • Circlo-Kean.jpg
    Circlo-Kean.jpg
    176.3 KB · Views: 461
Last edited: