High-Power Circlophone

Elvee

Member
2006-09-08 2:04 pm
This thread is going to be dedicated to the specific issues encountered in the design and construction of higher power amplifiers using the Circlophone topology.

The original Circlophone is not naturally suited to high power topologies, because they require paralleling, meaning emitter resistors, which the normal C does not use.
With a single pair of output transistors, it is difficult to go very much beyond ~100W, and a BTL configuration doubles the complexity, which is not very advantageous.
Great! It looks like you're going to actually build it?
No, I am afraid I won't: I have no use for such a monster, and my speakers are only 100W anyway, and if I tried to use them at 1/10th of their power, I would have serious problems with my neighbors.
I could build one and not use it, but it would be a serious waste of resources, time and space.
I may test critical sections and schemes in isolation, and I will provide guidance for candidate builders, but that's about it.

It seems like beside mechanical issues (heatsinking), the only more complicated thing would be correct matching of the output devices.
Question - what will go wrong (beside the load being distributed not evenly between all devices) if they are not matched perfectly?
If the matching is too imperfect and there are no compensation measures, the transistor having the highest gain/lowest Vbe will take more than its share of dissipation, which will lead to an even higher Hfe and lower Vbe, etc. etc.: a thermal runaway
Not thinking about extreme cases when devices are totally unmatched,
but just slight differences?
Slight differences will be absorbed by the balancing measures
Never tried paralleling BJTs. Only Mosfets, with mixed results.
Either it worked perfectly well (matched devices of course), or
I was getting oscillating amp, with no visible way to fix it.
Tried little zobel networks, different gate stoppers, different PCB designs, etc...
Same amp with one pair would work OK.
Yes, multiple devices can turn into a local push-pull oscillator (or a multi-push, multi-pull one)

Now, let's get into the heart of the matter:
I have examined the question, and I think I have found an acceptable solution: one that is safe and effective but also preserves the Circlophone sound character.
Not everyone likes the sound of the C, but if someone builds a bigger version, he probably wants to keep all of its peculiarities, including the sound character.

It would of course be possible to just add largish emitter resistors for a safe paralleling, but the sound would be "anonymized", just like any other class AB amplifier.

The solution has three components:
-Matching of the paralleled transistors
-Use of special, PTC emitter resistors
-Mounting of the transistors on a common heat spreader to minimize inter-transistor Rth compared to the thermal resistance of each transistor to the ambience

These measures go against the Circlophone spirit of simplicity, absence of matching, tolerance and ease of mounting, which is why I shall develop them in a dedicated thread, because I do not want to confuse would-be builders and lead them into thinking that a 1kW C is going to be as easy as a 30W one.
In the meantime, here is a first preview of the method:
796834d1574445299-little-cheap-circlophone-hipocirclo1-png


The most intriguing aspect are probably the low value PTC's.
Using PTC's is advantageous, because they behave in an opposite way compared to the transistors. When the current increases, they heat up and increase their value, just where it's needed.
Thanks to that compensating behavior, they can be much smaller than a regular resistor and still have a good balancing effect.
The 1,000$ question is: where to source those miracle components?

This is DIYaudio, and we are going to make them ourselves, in the true DIY spirit, and it is much easier than you might think.

That's what I am going to describe in my next post.
 

Elvee

Member
2006-09-08 2:04 pm
The resistors are just going to be home-made wirewound resistors.
The + tempco problem is a very easy one: most common metals have a positive tempco of 0.3~0.5%/°C, which is perfectly suitable here.
Copper for example has a tempco of +0.4%/°C and is very common, but it has one disadvantage: a good conductivity, meaning a relatively long length of wire, which is unpractical.
Steel also has a comparable tempco and a much larger resistivity (and tensile strength), but where to find high-quality, consistent and easily solderable steel wires?
The answer has a name: RG179 (for example).
This coaxial cable has a stranded, silver-plated steel inner conductor, of 7 x 0.1mm:

attachment.php


Here it is, partially stripped:

attachment.php


attachment.php


The next step is to measure the resistivity (45mΩ/cm), compute the length required (19mm), and wind that length on a suitable support, here a 1.8Ω CR25 resistor:

attachment.php


attachment.php


The final step is to dip the PTC in a suitable coating (high temperature).
I have tested two types of silicone: one is a general purpose, one-component neutral RTV (transparent), dispersed in an organic solvent (toluene)

attachment.php


, and the other is a 2-component, high temperature RTV 145HT (red).

attachment.php


attachment.php
 

Attachments

  • RG179.png
    RG179.png
    51.5 KB · Views: 1,105
  • RG179a.jpg
    RG179a.jpg
    162 KB · Views: 1,103
  • RG179b.jpg
    RG179b.jpg
    165.2 KB · Views: 1,105
  • PTC1.jpg
    PTC1.jpg
    367.1 KB · Views: 1,083
  • PTC2.jpg
    PTC2.jpg
    357.9 KB · Views: 1,077
  • PTC3.jpg
    PTC3.jpg
    149.6 KB · Views: 1,091
  • RTV1.jpg
    RTV1.jpg
    248.8 KB · Views: 1,063
  • RTV2.jpg
    RTV2.jpg
    251.6 KB · Views: 1,063
I could build one and not use it, but it would be a serious waste of resources, time and space.

On a philosophical note - 95% of what we humans do in our daily lifes, is not really essential, and it is "serious waste of resources, time and space".
Even assuming listening to music belongs to "essential/important activities",
99% of people who need speakers (instead of iphone with headphones)
would be just fine with $20 chines D-class amp from amazon, instead of spending time on diyaudio..

So, in conclusion - one more amp won't really hurt anybody :)
 

Elvee

Member
2006-09-08 2:04 pm
So, in conclusion - one more amp won't really hurt anybody :)
You are right -in a way- I have crates full of such amplifiers, but this one would require one full crate of its own, where I can normally store dozens of smaller ones.
As for the spreader, and the insulation - looks like the best (expensive!) way, would be to use Keratherm sheet, or something similar..
Or maybe kapton tape (way cheaper, more fragile)?
Good and expensive could be counter-productive: the thermal compensation needs very low transistor to transistor Rth, so a spreader made of thick aluminum or better, copper, is essential and transistors can be mounted with silver compound, but the common thermal resistance plays an important role in balancing, and if it is low enough to compete with the transistor to transistor Rth, that's not good.

In addition, the area of the spreader is much larger than the sum of the individual transistors, meaning the thermal resistivity of the interface is much less important (within reason).
If the thermal resistance of the interface is between 1/5th and 1/10th of the heatsink, it will have a moderate to negligible impact on the global result, and will allow a quasi-isothermal heat-spreader necessary for a good balancing.

Normally, an ordinary silpad sheet should be OK, if the spreader is not too small.
 

Elvee

Member
2006-09-08 2:04 pm
RG 179 from Mouser:
RG_179_B/U HUBER+SUHNER | Mouser

Over $7 per meter! Impressive.
Yes, I mentioned that as an example, because that's what I had handy, but any similar product would be OK.
That said, 1 meter of cable translates into 7m of silvered-steel wire, and since an individual resistor uses a bit less than 2cm, you could theoretically make 350 resistors for $7.
Even if your production yield is 10%, each PTC would end costing 20 cents for the wire. Add the same amount for the support resistor and coating, and you are at < 50 cents, which sounds tolerable.

I made some tests on the finished products: the cold resistance is comprised between 77 and 88mΩ: less than +/-10% tolerance centered on 82mΩ, not too bad for a quick and dirty attempt.
The absolute value is unimportant: what counts is the matching, and ideally we would like 1 or 2%, which is easily attainable by better techniques and/or sorting.
I measured a 86mΩ sample at various currents:
Code:
  I(A)     R(mΩ)
   0        86
   1        92
  1.5       97
   2        107
  2.5       122
   3        148
The variation is exponential: very slow at the start, increasing its pace at higher currents.
This is understandable: the formula P=R*I² is a first reason, but in addition R increases at higher currents, increasing the dissipation.

This means that these PTC will only work "statically" at low currents (which is better than nothing), and they will become really active at 2A and above (for this example).
This explains why other measures: matching and thermal coupling are still necessary for low currents.
The PTC effect will kick in at higher currents, and prevent thermal runaway.

What about the robustness?

In free air, such a straight wire fuses at ~2A. Here, the wire is in contact with the supporting resistor, and is embedded in the coating, meaning the current capability will be significantly higher (in a naked, isolated wire, the PTC effect would become apparent at much lower currents).
I will make a destructive test later, to assess the coated version.

Why did I use such a low value resistor as mechanical support?
To help damp possible VHF resonances caused by the 10 or 20nH of the winding. Probably not important with LF transistors
 
Last edited:

Elvee

Member
2006-09-08 2:04 pm
And here is appropriate RTV silicone in smaller container.
It might be too thick in its original form. If it is the case, you can dilute it with a small amount of organic solvent to ease the application.

I have made a destructive test on one of the samples: at 3.5A, it begins to smoke, and at 4A there is brutal increase in resistance, but no actual fusing.

For these examples, the sweet spot is around 2A. For normal operation (max output current rms), each one should see a little less ~1.8A.
That way, if one of the transistor diverges and goes over 2A, it will quickly come back on track.

1.8A on one side of the PP means √2*1.8=2.54A for the output of that pair.
If we want each pair to contribute 80W to the total, this requires a voltage of 80/2.54=31.5V (rms).
On a 8Ω load, the total output power would be 124W
On a 4Ω load, it would be ~250W ==> 3 pairs of transistors.

These are just proof of concept samples, and are not very usable in practice.
In reality, the design procedure should be reversed: start from the total output power, determine the number pairs required, compute the nominal rms current in each pair, and design the PTC accordingly.
 
In reality, the design procedure should be reversed: start from the total output power, determine the number pairs required, compute the nominal rms current in each pair, and design the PTC accordingly.

I will start with the rails, since I already have 3 different PSUs:
+/- 35V DC (300W)
+/- 50V DC (500W)
+/- 75V DC (800W)
So I guess 4 pairs will be minimum for 75V. Maybe 5 pairs...
My other amp using this voltage has 5 pairs of hexfets.

Any other changes to the Circlophone?
So far I've seen TO-220 drivers, lower value for BE resistor for output devices (BTW, I guess this value will depend on number of pairs, right?).
Perhaps Zobel resistor should be 2W, feedback 10k resistor 1W ?
TC004B should be fine for VAS/cascode ?
 

Elvee

Member
2006-09-08 2:04 pm
I will start with the rails, since I already have 3 different PSUs:
+/- 35V DC (300W)
That would be around 135W/4 ohm.
Some amplifiers use a single pair for that kind of power, but 2 pairs is more reasonable.
Total rms output current = 5.81A ==>2.9A/pair 2.1A/branch.
That is very close to the example, thus a very slight adjustment would be sufficient, to get about 70 milliohm instead of 82, meaning 17mm of 0.1mm wire on a CR25-sized core.
+/- 50V DC (500W)
that's ~300W/4 ohm, meaning at least 3 pairs, 8.7A tot., 2.9A/pair = previous example.
+/- 75V DC (800W)
So I guess 4 pairs will be minimum for 75V. Maybe 5 pairs...
That's 670W/4 ohm, probably too much for the supply, or 335W/8 ohm requiring 4 pairs.
6.5A tot., 1.62A/pair, 1.15A branch. The virtual power dissipation for the nominal current and a fictitious 25°C needs to be about 250mW, thus R=0.25/1.15² = 189 milliohm, about 42mm of wire.
I am going to make a sample of that value, to check that it behaves as expected.

If you want to use the amp with 4 ohm, you need to make similar calculations for 7 pairs.

You are of course free to use a different power/pair, these are just examples


Any other changes to the Circlophone?
Probably, let me think about it there is no need to rush.
So far I've seen TO-220 drivers, lower value for BE resistor for output devices (BTW, I guess this value will depend on number of pairs, right?).
Yes it will, but the initial value was chosen with sluggish transistors like the 2N3055 in mind.
With 4MHz, more modern devices, such a low value is probably not required
Perhaps Zobel resistor should be 2W, feedback 10k resistor 1W ?
The Zobel can remain the same, it is not the same as with follower output stages, increasing the rating of the FB resistor is certainly a good idea, even if it is not strictly required for pure power reasons
TC004B should be fine for VAS/cascode ?
Yes
 
That's 670W/4 ohm, probably too much for the supply, or 335W/8 ohm requiring 4 pairs.
6.5A tot., 1.62A/pair, 1.15A branch. The virtual power dissipation for the nominal current and a fictitious 25°C needs to be about 250mW, thus R=0.25/1.15² = 189 milliohm, about 42mm of wire.
I am going to make a sample of that value, to check that it behaves as expected.

My speakers have nominal impedance 6 ohm, so I guess 300-400W peak
power, with 4 pairs, and 800W transformer will have to suffice :)

In my previous builds, CCS transistor (small NPN, replacement for R21) was running hot, perhaps with higher rails I will use MJE340 or something similar.
Also, if I was redesigning the PCB, I guess maybe I would add a trimmer in the CCS to adjust idle current.
But as it is, I think my current PCBs (still have 16 of them :)) will work just fine.

Here is a pic with both Circlophones visible.
It's amazing how cool the amp runs with these fans.
They barely move, yet they cool very efficiently.
Actually just one fan would be enough..
For the High-Power amp, I plan to also use CPU cooler fans.
 

Attachments

  • circlos.jpg
    circlos.jpg
    218.1 KB · Views: 392

Elvee

Member
2006-09-08 2:04 pm
I have made a quick, 111 milliohm test sample using 25mm of wire.

Here are the resistance vs. current values (after thermal stabilization):
Code:
Current (A)    Resistance (mΩ)
   0                   116
   0.5                 123
   1.5                 131
   2                   147
   2.5                 175
   3                   229
It begins to smoke at 3A (with ordinary silicone), but even after several minutes at that level, there is no visible damage.
The sweet spot (where resistance begins to increase more rapidly) is somewhere between 1.5 and 2A, which means that the initial extrapolation of 1.5A was correct.
Note that a great accuracy is not required: it is just a protection against thermal runaway, and as soon as the current increases beyond the average, it will become limited by the PTC effect.
Matching, both for the transistors and the PTC's is more important
 

Elvee

Member
2006-09-08 2:04 pm
I'm ordering the cable...
These resistors - do they have to be CR resistors?
Metal film is not good?
Anything having a similar form-factor is OK: MR25, VR25, PR25.
In fact, I would recommend a PR25, because they are cement-covered rather than lacquered

Also, when it comes to matching output devices - do you have some special procedure/circuit that is preferred?
What do we need to match? Hfe and Vbe?
Vbe is essential, Hfe is optional, but if you measure the Vbe with the B and C shorted, a Hfe correction will be included which is good.
Try to make the measurement at a reasonably high current, >=1A, and normalize the thermal conditions: either you leave the transistor in free air and wait for the thermal equilibrium, or you clamp it to a good heatsink.
 

Elvee

Member
2006-09-08 2:04 pm
https://www.amazon.com/MEECOS-RED-DEVIL-1352-Fireplace/dp/B002JFSVGE/

Perhaps this kind of high temp cement would work better than silicone?
Certainly, if it remains perfectly neutral wrt. the metal of the wire

https://www.amazon.com/gp/product/B075JCQPD2/

Would this silicone sheet we good enough for the spreader/heatsink
barrier?
Yes, the thermal resistivity is 0.833°Kcm²/W.
With 10cm², you would have 0.0833°K/W which is about right
 

wg_ski

Member
2007-10-10 5:21 pm
That looks like an inrush limiter. It might not do anything until the transistors are already severely out of balance.

So what would really happen if you just used regular emitter resistors? With a CFP, they are really in the “collector” circuit. Which doesn’t degenerate the stage in the same way as if they are in the emitter of the PNP. Yeah, it’s not the same, but you never get something for nothing. It may be worth trying to see what the results are. Many of the large transistor types have multiple emitter ballasts inside the die anyway, so it’s effectively like multiples in parallel with resistors.