It is shown with 20V supplies, the clipping occurs to within <1.5V of either rail.
It is a simulation, but the real life waveform is strictly identical.

The supply voltage can range from 15 to 40V with no or minor modifications:
The current through R21 should ideally be kept between 1 and 1.5mA, and the total voltage of the zeners should approximately match the supply voltage.

The input resistance is ~10K, but the source impedance is preferably kept below 1K, because of the base current modulation caused by the bias servo.

Output impedance is <3mΩ @1KHz
THD @1KHz and 0.9*Vmax is <0.005%
Full power bandwidth is >100KHz
Slew rate: positive ~24V/µs, negative ~20V/µs

#######################################################

-All resistors can be 1/4W
-The output transistors, Q8 and Q10 obviously need a heatsink, as do the drivers Q9 and Q11. All four can be a common heatsink, or be cooled individually, or grouped in any convenient way.
-Q5 and Q6 do not normally require a heatsink unless operation at supply voltages above 45V is desired.
-No direct thermal contact is required between transistors, but the following pairs should preferably be mounted side by side: Q3/Q4, Q2/Q7, Q5/Q6, Q12/Q13.
-For D7, any small TH schottky will do, BAT81 or BAT85 f.e.
-D4 and D5 can be any 3A or 5A schottky, 1N5820 or something larger in TO220.
-Splitting D8/D9 is not required, but gives more flexibility.

-I didn't include an inductive Zobel network, because it would only be necessary in very extreme cases, but provision for it could be made on the PCB (and strapped in most instances).

#######################################################

For maniac detail nitpickers, R42 should preferably be increased to 510 ohm, but that's really icing on the cake.
In fact, it could probably be increased even more, as there is some gain stability margin, but the benefits would be quite minor, if any 

#######################################################

BIAS:
The voltage over resistor R8 is really too high, you should normally get some between 150 to 200mA through R8.

#######################################################

Bridging circlo:
https://www.diyaudio.com/forums/solid-state/189599-little-cheap-circlophone-167.html

#######################################################

Of the BC's, only one will see the full supply voltage: Q13.
At a total supply of 70V, it will see ~67V, and the BC546's Vceo is rated at 65V, thus in theory too much.
But if +/-35V really is the upper limit of your supply, you run no risk using one.
If on the other hand there is a risk your supply drifts higher than 70V, you would be well advised
to use a 100V or 80V type, like the MPSL01 or the 2N5550 for example.
Note that Q12 should preferably be identical to Q13.
Other BC's see at most half the total supply and need not being substituted.
The BD's and the 2N3019 are OK too. 


#######################################################



https://www.diyaudio.com/forums/solid-state/189599-little-cheap-circlophone-6.html


The diodes are not at all critical. Let us leave aside the 1N4148, I don't think this could cause problems.
There is one small signal schottky, the BAT54. In fact, I should have specified a BAT81 or BAT85, as it is a TH design.
It is absolutely non-critical, and any 30V - 200mA schottky will do.
There also is a 1N5xxx suitable, but at the moment, I don't remember the type number.
If you don't have a schottky available, you could even use a 1N4148.

The larger schottky's can be almost anything, BYS45, 1N5820, ... with Imax>=3A.
You could even use a fast silicon diode, BYV31 or similar: it would only affect the swing (energetic efficiency)

Quote:
I still have to do an inventory of transistors to see what I've got, but I'm sure I've got the parts
I need besides those few diodes. I am comfortable if needed to use the faster devices for the OP
transistors and tweak it for stability... another opportunity to learn.
The transistors are not critical either: they just need to have sufficient ratings for the job.
And if you confine yourself to calm, placid transistors like the 2N3055, you will not encounter any problem.
And when you're confident enough, you can begin to tweak and use faster devices.

Basically, this amplifier is designed to be tolerant and undemanding regarding the components.
There are one or two things to care about though, f.e. the collector capacitance of Q5 and Q6:
if you use high capacitance devices (>20pF), you will most likely ruin the performance.

I'll try to come up with a formal builder's guide, to make those (few) things clear.

---

A word about D6/D7: if the BATxx is not available, a possibility is to replace both diodes
by fast 1A ones like BYD33, BYV26, UF4004, etc.
This combination will be equivalent to 1N4148+BAT85 (Do not use schottkies!!!)




############################################



Input:

The input resistance is approximately 10K.
But, the source impedance must be kept well below that value.
This is because the tail current of the input pair is dynamically modulated by the bias servo, and any imbalance in the input
impedances will generate a parasitic voltage due to the bias currents of the input transistors.

This will cause a "leak" of the common mode signal into the "noble" differential path, and will increase the distortion.
For this reason, source impedances much greater than 1K should be avoided.

In practice, this means that if a volume pot is placed directly in front of the amplifier, its value should be 4.7K or less.
If the input is fed from a preamplifier, the impedance will normally be in the hundreds of ohms, which is optimal.

Q3 and Q4 can be any small signal PNP with a Vceo>Vs, BC556 or 557 f.e.
For the above-mentionned reasons, a high Hfe type is preferable, as it will minimize the effects of the bias current (suffix "C" for BCxxx).

Q1 is identical.

C6/R23 are the main compensation components of the amplifier. On some versions I have used 22 ohm for R23, on other 33 ohm,
and in the end I settled for 27 ohm, as I was not sure what the best value was. Anyway it is not critical, and either value will work.
1.5nF seems to be the optimum for C6.
Phase splitter:

Q5 and Q6 must be rated at 2Vs, and their average dissipation is 13mA*Vs.
TO5 or TO39 types are ideally suited, as their free air dissipation is generally 800mW, which is sufficient for 50V supplies.
Suitable examples are 2N3019, 2N3053, 2N1893 and 2N2219 (for low voltages).
It is important to use relatively low capacitance types (Ccb<15pF).

The zeners equalize the dissipation between the two transistors, they should total approximately Vs.

For Q13 a Vceo>2Vs is required.
Drivers:

Medium power transistors similar to the BD140 are perfect for Q9 and Q11. A Vceo>2Vs is required, and depending upon the power of the amplifier, they could dissipate up to 5W.
Output:

For powers up to 50W, 2N3055 are perfect. Vceo has to be greater than 2Vs, and they need adequate cooling. The SOAR must also be sufficient.
With such slow devices, C11 is optional and C12 can safely be reduced to 180pF.

These capacitors ensure local stability, but at the expense of the global loop's margins.
This means that being too heavy-handed is counter-productive, and excessive values will degrade the transient response.
With faster devices (Ft>4Mhz), C11 becomes mandatory, and C12 needs to be increased, up to 820pF.

The values of the capacitive Zobel look like a misprint, but they aren't: the output topology is unconventional, and these values give the best results. On some prototypes, I have also used 4.7 ohm for R25.
General:

Other transistors need no special mention, and the role of the power schottky's is simply to limit the power wasted, they have no active role.

The power supply voltage can range from less than 15V to more than 50V, provided suitable devices are used.
Only the zeners need to be adapted, and also R21, to keep the current within the 1mA to 2mA limits.

All resistors are 0.25W, and no selection, matching or thermal tracking between any components is required to obtain the advertised performances.

As with any amplifier, good layout practices are recommended, particularly for the ground and output nodes.

#########################################################


The output power is related to the supply voltages and the load impedance, in this case:

Po=(Vs-1.5)²/(2*Rload)

With the 25V supplies of the example, this translates into ~34W on 8Ω, ~30W for practical purposes, 60W on 4Ω, 120W on 2Ω and 15W on 16Ω
For 8Ω:

Po=25W Vs=22V R21=33K

Po=50W Vs=31V R21=47K

Po=100W Vs=42V R21=68K

For 4Ω:

Po=25W Vs=17V R21=22K

Po=50W Vs=23V R21=33K

Po=100W Vs=32V R21=47K


R21 CHART:

25V : 33K
25-30V : 39K
30-35V : 47K
35-40V : 56K
40-45V : 62K
45-50V : 68K


All small signal transistors except Q13 must have a Vceo>Vs, all the other, including Q13 must have a Vceo>2*Vs.

The combination of output transistors + heatsink must be able to dissipate ~Po/2 with an acceptable temperature rise.
For the lowest power levels, account must be taken of the quiescent power, which is higher than for a standard class AB: ~200mA.
That translates into heatsinks of 9°C/W for 10W, and 1.5°C/W for 100W for example.

At higher power level, the SOAR of the transistors will be the limiting factor, rather than Pdmax or Rth.

Quote:
I'm using about +-40V and I only had to change value of R21 and use higher Vceo for two small signal transistors.
Others may have to use something other than BD139/40 And depending on the output transistors, use more of them
That's about 90W for 8Ω

#####################################################

VAS: Q5 Q6

However, the collector capacitance for this transistor is slightly below the recommended range of 5 to 15pF.
This might cause minor stability issues (or not), depending on other semiconductors and other things.
The collector capacitance acts as an auxiliary self-compensation, which is why too low a value might be problematic.
If it happens, it will be in the multi-MHz range, very small and harmless both for the amplifier and the speaker,
so nothing to worry about, and the cure is simple and straightforward: simply add 4.7pF C-B capacitors. 

KSC3505 or KSC3503 for VAS (Q5/Q6) (piersma)
2SC3505 or 2SC3503 for VAS (Q5/Q6) (piersma)

simply add 4.7pF C-B capacitors.

#####################################################

Terranigma
At last I finished and listened one channel. I have modified wakibaki's pcb respecting 2sa/2sc small
signal transistors.

My setup:
2sa970 as PNP small signal
2sc2240 as NPN small signal
2sa1930 as Driver
2N3442 as output (test setup)
2N3020 as bias servo
180pF as C12 (servo compensation)
1200pF as C6 (main compensation)
31V per rail
Unbalanced signal

The 2SA970's are excellent devices, and you have to thank them for the silence of your circlophone.

#####################################################

- BC560C as input pairs was not shine on this topo. I switched to Fairchild BC556B and got better results.
- I also want to try KSA992 but the pinout is different.
- I used Silver Mica for feeback capacitor (22pf) and input capacitor 220pf and I think it has some good effect to the sound.
- I used all CMF55 resistor on important location.
- I used Vishay MKP 3u3 with back-to-back tantalum 22uf as input capacitor. Tried some other combinations and this is the best.

#####################################################

https://www.diyaudio.com/forums/solid-state/189599-little-cheap-circlophone-34.html

About Piersma posh circlo:

Waow! this circlophone is not "little and cheap" anymore: it is the "Posh Circlophone".
There are many worthwhile improvements possible over the basic topology.
The posh version includes one of the most useful: a FET input, that eliminates trade-offs between bias current modulation and input impedance.
The other modifications are quite sensible too.
I see one other point where easy improvements can be gained: replace R27 by a 2-wire current source
(I have described some suitable varieties in other topics).
This will improve the PSRR, make R22 superfluous, and allow the same circuit to work from +/-10 to +/-100V without alteration.


#####################################################


Originally Posted by terranigma

@Elvee
Is 2sc3423 a good candidate in place of 2n3019?
In theory, yes. 
But.... 
with their minuscule capacitances and high Ft, they won't tolerate anything less than a perfect layout, ideal drivers, etc.
In short, if you have the means of taming them, they will give outstanding results but in every other instance,
they will oscillate madly and uncontrollably.

Definitely not a beginner's option.

Devices having a Cob around 10pF and an Ft of ~100Mhz will give transparent results without troubles.





The 2SA1209 looks like a perfect clone of the original BD140:
http://www.rapidonline.com/pdf/81-0352e.pdf
It even retains the TO126 case, and it is cheap too.
(I didn't test them, I'll try to find a pair to make a test)


#####################################################

I have tested some more driver types.
TO126
-BD140 from Telefunken (at least, that's the way they are marked: TFK, black epoxy, + stripe of green paint on the top
       edge): OK, no problem whatsoever, even with relatively slow VAS.
-MJE350 (Motorola): OK with normal VAS (like 2N3019, 2N1893, BF819, ..).
Not stable with slow VAS (requires additional compensation): 
-BD238: idem


TO220
-BD242C: idem
-2SA968 OK under all conditions
-2SA1011: marginally unstable
-2SA1306: marginally unstable

The case of the 2SA1011 & 1306 is curious: they are quite similar to the 2SA968, yet they show signs of instabilities.
The instability manifests itself only when the transistors are completely cold, on negative output excursions
on a low load (high negative output current).
It is merely a thickening of the trace, barely visible: 100 or 200mV at several MHz, and totally unresponsive
to the usual cures, base stoppers etc.
Probably something to do with the layout.


#####################################################


Originally Posted by danielwritesbac  View Post
In regards to ♫♪discussion about 4 ohm speakers and power over 100w♫♪ I found two of the big MJL21194 on sale.
Is it suitable? If so, I was wondering what transformer to use? A 30+30vac or a 32+32vac?
The MJL21194 is suitable for 100W/4Ω.
In principle, a 25V+25V transformer would be just sufficient for that power, so with 30V you will have plenty of margin.
In fact, with 30V, you will get ~150W, rather safely since the MJL's SOA allows for it.


Originally Posted by danielwritesbac  View Post
Questions: 
1). Driver, big, fast, promotes euphonic, and durable?
Driver has to be sufficient, but not oversized, it is counterproductive.
Many types have been tested/suggested.
The BD140 is OK with the MJL's, but if you want something more "comfortable",
you can use the 2SA968, 2SB649 (tested), MJE172, MJE254, BD792 (not tested, but suitable).
The MJE15031 has also been tested, it is oversized but it works too.

#####################################################

With the MJL's, C11 should not be implemented.

If you have the choice, use SR502 schottky's instead of SR506, they have a slightly lower drop.

The schottky's role is minor, they simply improve the energetic efficiency and could be omitted without problem,
there is no need for high performance there.
A reverse voltage of less than 5V is more than sufficient, but each has to carry half of the output current,
which rules out the 1N5822.
The 1N5711 and 6263 are poor choices for the clipper: they are high voltage/low current, low conductance types, better stick with BAT8x types.


#####################################################


Here's the parts list progress so far: 
Q5/Q6 2N1893, 2SD669, 2SC2238, 2SC5171, MJE340? 
Q8/Q10 MJL21194, MJW21196 
Q9/Q11 2SB649, 2SA968, 2SA1930, MJE350 
Q12/Q13 2SC1845, KSC1854F
I would leave the MJE340 out of the list: it works, but it is not optimal, and when combined with other suboptimal components,
it could begin to cause troubles.



#####################################################

C11 is necessary when very fast output transistors are used. In your case,
it is not necessary, but it does little harm.
If you remove it, the transient response will be infinitesimally cleaner,
but otherwise the effect is barely noticeable.



As published, the Circlophone has a default gain of 27dB. That's 6dB better than 21dB.
Add another 6dB because you build a 4ohm version, you are 12dB better.
If it is still insufficient, you can easily increase the gain by changing R17 to 15K or 22K.
Increasing the gain poses little problem, decreasing it could be problematic.
The stability has been tested down to a gain of 8, I think 5 would still be possible, but this has not been tested.

#####################################################

I built a similar Circlophone to the one you plan to.

Mine runs on (has been for a few months, I think) +-40V rails, adequate heatsink+fan cooling.

The semiconductors I use are BC546/556 "C" suffix (highest hfe), 2N3019 (not heatsinked - no issues) 2SB649A,
and MJW21196. The outputs (MJW) are a beastly kind, just like the ones you are going to use.
With the supply voltage I use I know I can pump out plenty of power into 4 ohms and not worry about
anything with these transistors. Good choice. Two transistors, Q12 and Q13 (going off of Elvee's original schematic)
I changed to KSC1845F becase of the higher supply voltage.
These transistors have a different pinout and the pins were bent to compensate for the PCB. 

I matched (within +-2 hfe) and thermally coupled the input pair with a tiny zip tie and thermal compound. I have <0.1mV of DC offset.
I used 1% 1/4W metal film resistors throughout. I used polystyrent for the input low pass capacitor (C5) and C12
(I used 270pF in both places). I used polypropolene 1.5nF for C6 and I did use C11, polypropolene 10nF. 

My Circlophone is extremely dead silent. If it wasn't for the fan in the case I wouldn't know whether it was on or
off, even with my ear right against a tweeter. If i recall from reading this thread the Powerflux one may need to be slightly
modified regarding a filter due to HF noise pickup. I still have PDF's created from the wakabaki gerber files made to print to scale.
I posted them pages ago.

#####################################################

R3 is there to keep the feedback loop from going crazy and the output stage from failing in case the
fuse blows for other reasons than output stage failure, for example because of accidental output short,
or if a speaker with too low impedance is used.
Its value must be significantly higher than the speaker impedance because it is in effect parallel with
the speaker at all times (and thus dissipates power), but significantly lower than Rf so that there
isn't a huge change in the feedback loop when the fuse blows. 220 ohms to 1k is fine.


#####################################################

There are several hundred of 2n3501 located about 85 miles from here.
Motorola quality control(?) varies so much that given a 10 pack, about 20% may be down at hfe 121 same as 2n3019.
Should I get some 2n3501 for Q5, Q6?

Yes, they look like excellent candidates.

#####################################################

And my C12 is not all the way up to 820p and visually, it does look like it could help cut my lisping treble problem. Is that true?
C12 is the least critical cap in this amplifier, you can do pretty much what you like.


C4 could be styroflex, NPO, silvered mica.
C12 820pF- 1 nF opt for MKT (or styroflex)

#####################################################

Elvee, I have an extreme curiosity about KSA1220AYS from Mouser.com. Is it probably suitable?
In their lineup, it was the closest thing I could find to a Philips BD140. If that sub is is a good
idea (if it is better than generic/Motorola BD140), then it would make an entirely Mouser parts list possible in just one box.

It looks ideal: it is equal or superior to the BD140 in all respects.


#####################################################


List for RS and Mouser
D4, D5: MBR735
D7: BAT86
Q1, Q2, Q7: BC556B
Q4, Q3: BC560C (reasonably matched would be good)
Q5, Q6: ST 2N3019 (40+40vdc rails abs max, usable at 37+37vdc)
Q11 Q12: At RS, 2SC2240, At Mouser, KSC1845
Q9, Q11: At RS, 2SA968, At Mouser, 2SA1930 or KSA1220AYS
Q8, Q10: BD911 or MJ15015 or MJL21194, or MJ21194 or MJ15022

Q5, Q6 for higher voltage rails (alternate vendors): 
Current: HSD669a from AudioWind (not yet tested), 2SC3421 (not yet tested)
NOS: 2N3498, 2N3499, 2N3500, 2N3501, BF819


#####################################################

Elvee:
I have tested samples of 2SC3421 and 2SA1358 kindly provided by Daniel (VAS/phase splitter and driver).
I have just tested the 2SA1930/2SC5171 pair as drivers and VAS.
The samples kindly provided by Daniel are authentic, and they perform remarkably well in the Circlophone.
The all-plastic construction makes mounting on a heatsink very easy, and the case has a low enough thermal
resistance of its own for the VAS to be left without heatsink.
In summary, an excellent choice: high availability, good performances and ease of use: a dream.

#####################################################

output devices ratings from Elvee:

MJL - bigger case
MJW - small case

1/ 2SD1047      1.76
*****  2/ MJ(L)21194   1.6
3/ MJ(W)21196   1.45
4/ MJ15022,4    1.2
   MJE/FJA13009 1.2
5/ BD249C       1.04
6/ MJ15003      0.94
   MJ802 0.71
7/ MJ15015      0.55
8/ MJE3055      0.46
9/ TIP41        0.43
10/ 2N3055      0.35
11/ TIP33       0.34
12/ BD911       0.29
13/ TIP3055     0.16




?? 2N3773

#####################################################

Originally Posted by danielwritesbac  View Post
I have a few more curiosities. . .
The 2N5551 actually does not work consistently across all builds for Q12/Q13 (HFE and noise differences),
and as a result, there is some variety of troubles, like suddenly blown output devices, extra heat,
or just a sonic signature difficulty.

Is there, by any chance, a better commonplace device "just like" BC546B that
could manage Q12/Q13 with up to 40+40vdc rails?

The 2SC(or KSC)1845 is a perfect substitute, commonly available and suitable for up to +/-60V supplies

#####################################################

The basic requirement for the combination driver+OP is that the product of the Hfe's must be sufficient.
Using basic arithmetic, it is easy to make a quick and simplified estimation: the OP stage has to provide
a peak current dependent on the supply voltage and the load: for example, +/-40V and 4 ohm mean 10A.
The VAS has a quiescent current of ~13mA. If we don't want to exceed a 50% imbalance, this means the available
current excursion is limited to 6.5mA.
If we divide 10 by 0.0065, we get ~1540: that is the total gain required from the CFP to achieve the target power.
You can distribute that gain the way you want, an obvious method is to take the square root, in this case about 40 for each.
But the dissipation in the driver only depends on the OP's Hfe, not its own Hfe. But of course, if the OP has a low Hfe,
the driver will need to compensate for it to meet the above calculation.

Note that the calculation is rather pessimistic: not only can you use the full 13mA of the VAS, but in case it is
still insufficient, the bias servo will intervene and raise the average current level to meet the demand.
This is why the Circlophone is able to drive obscenely low loads, but the linearity is somewhat degraded in those conditions.


#####################################################

https://www.diyaudio.com/forums/solid-state/189599-little-cheap-circlophone-122.html

Inverted JFET Circlophone,

With reference to post number 331 of this threat, this is final version of the “inverted JFET Circlophone”.
Real life implementation of the attached schematic is shown in the picture.
The design is fully tested and approved by the designer of the original Circlophone, Elvee. 

Highlights of this version:
Scalable outputpower, rails up to 50V
JFET input, (virtual) absence of gate/base modulation
Improved step response of the CFP configuration
Reduced phase splitter dissipation
Original Circlophone features: 
No (very low…) cross-over distortion, semi class A
No need for thermal tracking the output devices
Idle current 200 mA, Gain : 28 dB


#####################################################

With better devices like yours, you can safely reduce C10 to 180p.

#####################################################

*Checking R21, click here > Elvee said: "Measure the voltage, divide by the resistor value, if it's 1.5mA +/-50%, it's OK" 

R21 CHART:

25V : 33K
25-30V : 39K
30-35V : 47K
35-40V : 56K
40-45V : 62K
45-50V : 68K

#####################################################

Thanks Terranigma, then what is your recommendation for Q12, Q13 on 35v rail.
I modified Alexmm pcb to take ksa1845 and my pcb are shipped.
Please note that mine is original circlophone as Elvee published. Only difference is rail voltage.
 
Elvee: Don't worry: all the transistors of the Circlo engine, including the sensors can be any
ordinary low power signal ones, like the BC547/557 etc.

#####################################################

====================

april 2013

hexfet:
https://www.diyaudio.com/forums/solid-state/209907-building-elvees-circlophone-documentation-accessories-beginner-friendly-23.html

R21 value ranges shown in layout may become problematic in some cases. I can alter it with a slight range shifting:
R21 CHART:
25V : 33K
25-30V : 39K
30-35V : 47K
35-40V : 56K
40-45V : 62K
45-50V : 68K

================

Recently converted my current darlington builds to mosfet and replaced high voltage/wattage 2sa1546 servos via 2n2905A
small signal transistors on both channels as what like you did above.
So, difference of attained sonic behavior is remarkable, as like as plugging a totally different amplifier.

Now, amp sounds more bright but seems well defined compared to earlier setup.
Some little details at familiar songs also distinguishable.
Bass is not boomy as like as darlingtons. I think I really liked the bass of
darlingtons. Low freq is not weak but different.. I heard similar bass behavior
impressions from other people with different amps regarding to mosfet-bjt comparisons.

I think I should install 2n2905A's first and then mosfets later. Although it is not clear
for me that which part caused the sonic difference mostly but I can barely state that 2n3019, 2N2905A(pnp)
type small signal transistors would be the best bet on choosing servo transistors depending on my listening expressions so far.

================

My amps run on 190mA quiescent current. I see some designers tend to bias mosfet amps
rather higher, example 200mA per mosfet. In total of 400mA bias, we may question that
"high efficient class A" concept of Circlophone. Btw, it is not that hard to do that
by decreasing values of current sensing resistors each side. 

================

Going from N-darlington to N-MOS is a very small step, it just involves adapting the 120ohm B-E resistors to a
value compatible with the G-S threshold voltage of the MOS you use.
Very minor indeed.
A drawback of this straight modification is that the output swing would be amputated of ~Vth at high current
on the positive side, something like 4V.

================

With +/-20V supplies and 2N3055's as OP, the +SR is 24V/µs and -SR 20V/µs (actually measured).
Apart from the unrealistic 2SC2922 model, the oscillation seems to be caused by a known issue: a minor
parasitic pole in the common mode loop caused by R15.
It happens with certain mixes of transistors, and the simplest fix is to eliminate R15 completely:
it has no active role and just improves ruggedness and strong EMI immunity.

================

For maniac detail nitpickers, R42 should preferably be increased to 510 ohm, but that's really icing on the cake.
In fact, it could probably be increased even more, as there is some gain stability margin, but the benefits would be quite minor, if any 

================













=================================

All small signal transistors except Q13 must have a Vceo>Vs, all the other, including Q13 must have a Vceo>2*Vs.


output circlo:

Also, if you use fast (Ft>5MHz) transistors, the capacitor C11 shown as optional on the schematic becomes mandatory to remove local instabilities, and the circuit becomes more sensitive to the lengths of cable to the transistors.

With faster devices (Ft>4Mhz), C11 becomes mandatory, and C12 needs to be increased, up to 820pF.

==========

Finally, Q5 and Q6 should have a low enough collector capacitance, less than 15pF.

Devices having a Cob around 10pF and an Ft of ~100Mhz will give transparent results without troubles.


BF819, which is new old stock, expensive and hard to find.

Many subs are possible: in TO39, the 2N1893 is perfect, and not difficult to source.

 2SC1819 and 2SC2231 are OK too (as are KSC... obviously), and many other types: basically you want a device of >100V, >1W, with an Ft in the 50 to 150MHz range, and a Cob comprised between 5 and 15pF.


I have tried successfully BF459 and BF819.

 There are two possible pitfalls with this type of transistors: they have a "knee" voltage of several volts (the Vce voltage for which the linear performances begin to degrade because the transistor "runs out of steam"), and this will somewhat limit the output excursion, and the newer types have very low feedback and/or Cob capacitances.

 This is normally considered to be a plus, , but if the capacitance is below 3 to 4pF, stability issues could begin to appear.

 The circuit relies partly on the transistor's capacitance to self-compensate, and the recommended Cob range is 5 to 15pF.

 4-20pF remains possible, this has been tested, but going beyond could lead to (minor) troubles. It also depends on Ft: if it is larger than 100MHz, lower capacitances could be tolerated, but the circuit will also become more layout-sensitive.

 2N3020-type transistors are ideal for that purpose, and if you need to go higher in voltage, you can use heatsinked 2N1893.

 Some suitable 2SC's have also been suggested earlier.


2n3501 available at mouser, $7.00 !!


2SD669 is perfect.. obsolete


==========

There is one small signal schottky, the BAT54. In fact, I should have specified a BAT81 or BAT85, as it is a TH design.

==============

The larger schottky's can be almost anything, BYS45, 1N5820, ... with Imax>=3A.

SR504 - small 5A

=============

Q3 and Q4 can be any small signal PNP with a Vceo>Vs, BC556 or 557 f.e.

 For the above-mentionned reasons, a high Hfe type is preferable, as it will minimize the effects of the bias current (suffix "C" for BCxxx).



Phase splitter:

 Q5 and Q6 must be rated at 2Vs, and their average dissipation is 13mA*Vs.

 TO5 or TO39 types are ideally suited, as their free air dissipation is generally 800mW, which is sufficient for 50V supplies. Suitable examples are 2N3019, 2N3053, 2N1893 and 2N2219 (for low voltages).

 It is important to use relatively low capacitance types (Ccb<15pF).



 For Q13 a Vceo>2Vs is required.


Of the BC's, only one will see the full supply voltage: Q13.

 At a total supply of 70V, it will see ~67V, and the BC546's Vceo is rated at 65V, thus in theory too much.

 But if +/-35V really is the upper limit of your supply, you run no risk using one.

 If on the other hand there is a risk your supply drifts higher than 70V, you would be well advised to use a 100V or 80V type, like the MPSL01 or the 2N5550 for example.

 Note that Q12 should preferably be identical to Q13.


========

--Q12,Q13 to 2N5551 (Or is KSC1845E better? (much higher hFE, but low Ic))

Those transistors are just auxiliary devices: they aren't directly in the signal path, they just ensure it's in a good condition and they have little influence on the quality.

Small signal devices are more than enough, and if they have a high Hfe, so much the better, so I would opt for the KSC's


======= terranigma:

My setup:

 2sa970 as PNP small signal

 2sc2240 as NPN small signal

 2sa1930 as Driver

 2N3442 as output (test setup)

 2N3020 as bias servo

 180pF as C12 (servo compensation)

 1200pF as C6 (main compensation)

 31V per rail


=========

1ohm resistors : 1/4 W

==========

voltages:

https://www.diyaudio.com/forums/solid-state/189599-little-cheap-circlophone-post2883608.html

============