♫♪ My little cheap Circlophone© ♫♪

[EdwinR3]

Originally Posted by Elvee

Some transistor models are not included, so I couldn't run the simulation,

The models I used are all included in the file Cordell.lib that is included in the zip archive. I have included the used models in the file and changed to standard models if available.

but anyway the likely cause of the problems is the lack of a DC blocking capacitor in the output, and/or a low/absent winding resistance for the transformer.
I wasn't able to track the actual series resistance used by the simulation: I saw it equaled Rip, which then equaled Ris, but then I lost the track.

Rprimary = 0.15R, Rsec=4252R

Even if there is a finite series resistance, it is probably not a good idea to drive the transformer directly: normally, this type of transformer is ungapped and has a high permeability, and the winding resistance is low.
All this means that even a negligible output offset voltage will be sufficient to saturate the core.

The transformer is a C-core with double bobbins. At the two gaps I inserted a small piece of plastic, so probably 2x0.2mm gap is there (I must verify if the gap is included in my transformer-model)

It will also cause a high dissipation in one of the OP transistors, because the amplifier will try to impose this OS voltage to a low resistance.

Ok, good to know.

Try inserting a large cap in the output, this should fix the problem.

Just tried it with output caps from 1µ to 200 µ and the dissipation is restored to the level of a load resistor. Unfortunately the output is also reduced.

BTW, making R23 ~=0 will probably cause problems. Even if they don't appear in the sim, the reality will be less forgiving.

This was left over from a test to see if it made any difference. In sim with higher values or R23 I'm getting a lot of oscillations.

Thanks for the advice, I'll try to further understand what is going on inside the loops (I'am not an electronics guy, just trying to understand and finding a suitable amp for an ESL, other than buying an Ncore).

Edwin

[Elvee]

Originally Posted by EdwinR3

Rprimary = 0.15R, Rsec=4252R

OK, let's say you want to keep the offset current into the primary under 10% of the quiescent current, thus Ios<15mA, which I already find uncomfortably large.
This requires an offset voltage smaller than 2.2mV.
It is doable, by using good quality hand-picked input transistors.
Some builders have achieved under 1mV offset with this method.
But anyway, that is not something I would recommend: without any protection, a powerful subsonic signal that is slightly asymetrical could lead to the saturation of the core, with disastrous consequences.

The transformer is a C-core with double bobbins. At the two gaps I inserted a small piece of plastic, so probably 2x0.2mm gap is there (I must verify if the gap is included in my transformer-model)

2 x 0.2mm looks pretty large, it will seriously reduce the magnetizing inductance, and this could become problematic at very low frequencies.
Anti-saturation spacers are more like 0.05mm.

Just tried it with output caps from 1µ to 200 µ and the dissipation is restored to the level of a load resistor. Unfortunately the output is also reduced.

The value needs to be very much larger than that: it should be 4700µF or something like that.
If you don't like the idea of a big E-lytic in the signal path, there are alternative ways to arrange the feedback path.
The transformer too could benefit from inclusion in the FB.

This was left over from a test to see if it made any difference. In sim with higher values or R23 I'm getting a lot of oscillations.

Thanks for the advice, I'll try to further understand what is going on inside the loops (I'am not an electronics guy, just trying to understand and finding a suitable amp for an ESL, other than buying an Ncore).

I recommend you use the standard Circlophone as a starting point: it is stable, both in sim and in reality.
You can then make incremental changes from that base to see the effect of your mods.
This is probably the safest approach, since the sim is not very good at predicting subtle, marginal modes of oscillation between the different loops.

[Elvee]

Including the input capacitor to balance the input LTP results in an offset of 7µV (that's in sim, obviously).

The offset current falls to 3.68mA, but this means this is incorrect:

Originally Posted by EdwinR3

Rprimary = 0.15R, Rsec=4252R

These values yield a primary resistance of ~2mΩ

[EdwinR3]

I'm going to have to test your patience with we, since I'm a beginner in the world of amplifiers (and electronics...). I build some segmented wire stator electrostatic speakers and made the step-up transformers for them, but I’m still struggling with the driving site of the system.

Originally Posted by Elvee

OK, let's say you want to keep the offset current into the primary under 10% of the quiescent current, thus Ios<15mA, which I already find uncomfortably large.

The quiescent current being the collector current from the output transistors when there off ?
The description in futura-science states for "warm" class AB a quiescent current of 150 to 200 mA, so 10% of it could make 15mA. Why is this to large ? Risk of saturation of the core, with hereby larger currents in primary and output transformers ?

Originally Posted by Elvee

This requires an offset voltage smaller than 2.2mV.
It is doable, by using good quality hand-picked input transistors.

Here you already lost me. Is this 2.2mV a result of the simulation or can it be calculated from the Ios, or measured on prototype ?

Originally Posted by Elvee

But anyway, that is not something I would recommend: thout any protection, a powerful subsonic signal that is slightly asymetrical could lead to the saturation of the core, with disastrous consequences.

Normally I would drive it with a HP filter at 120Hz, so powerful subsonic signal is unlikely, but I see your point.

Originally Posted by Elvee

2 x 0.2mm looks pretty large, it will seriously reduce the magnetizing inductance, and this could become problematic at very low frequencies.
Anti-saturation spacers are more like 0.05mm.

This was a choice I made to try avoiding DC offset problems and to linearize the core, since the lower signals will be filtered out. In sim the air gap is included in the transformer model. With 2x0.2mm I find no more saturation with DC (not at 0.01Hz and 20Vrms)

Originally Posted by Elvee

The value needs to be very much larger than that: it should be 4700µF or something like that.

This one I could verify with the sim. I found out that inserting a resistor of 0.6 to 0.8 Ohm on the primary side gives the same result. What puzzles me is that the capacitance of an ESL is always quoted as making a it hard task to drive for an amp, yet I would need to insert an even larger capacitor to protect the amp.

Originally Posted by Elvee

If you don't like the idea of a big E-lytic in the signal path, there are alternative ways to arrange the feedback path.
The transformer too could benefit from inclusion in the FB.

Indeed this is contrary to "standard audiophile tradition". I always understood that E-lityc caps are causing a lot of distortion. It should be a non-polarized type as well I suppose.
I know of the principle to include the transformer in the FB loop, but I have no idea how to do this. It occurs to me that the base for the development of the circlophone was to remove the transformer from the original concept. Maybe for an ESL amp, it is logic to stay with the original topology.
Including the trafo in the FB, means that the amp must be integrated in the ESL (this is possible but will require some reworking). I'm not so sure about the including of the HV in the amplifier, unless the FB only includes the primary of the trafo.

Originally Posted by Elvee

I recommend you use the standard Circlophone as a starting point: it is stable, both in sim and in reality.
You can then make incremental changes from that base to see the effect of your mods.
This is probably the safest approach, since the sim is not very good at predicting subtle, marginal modes of oscillation between the different loops.

I did restart from the schematic of page1 and made some gradual modifications, based on following ideas :
1. at 20Khz, the input is limited. When the supply is increased this limitation was removed. This also seems to reduce distortion (I went to +/- 54V).
2. The increased rail voltage exceeds the voltage-range for the input transistors Q3/Q4 so I swapped the BC560b (Vce=45V), to BC556B (Vce=65V). This might be less low-noise, but I don't know if it matters here.
3. Q13 is changed to a 2SC3601 to exceed Vce > 2xVsupply)
4. Q8/Q10 are replaced with MJE21194, with higher power rating.
4. Based on on off the last schematics of (Keantoken I believe), I played with some capacitors parallel to emitter resistors of Q1/Q2/Q13. These capacitors do not seem to have a influence on the bias-servo, but reduce some oscillations at high frequencies (20 tot 100 Khz).
5. With these modifications and the insertions of a primary side resistor of 0.66 Ohm, I'm getting very good results (THD 2-20 of 0.004% from 160Hz to 20KHz on HV side of transformer, so transformer distortion include), phase margin of > 90° and gain margin of 28dB.

I'm still having some doubts, because a test of square wave stimulation gives me always very strange results. (the output is always either positive or negative, but never symmetrical. The same problem I have with the original circlophone with a load resistor, so it is probably something I do wrong.


Sorry for the long rant, but I’m eager to learn more about this "simple" amplifier (simple as in parts, I still do not understand all the different loops and interactions).

If this gets to far of topic, let me know,

Edwin

[keantoken]

The square source is a great big failure on Linear's part regarding LTSpice. The simulator solves the operating point according to what is the initial value, so you end up starting the simulation with the input cap charged wrong. There is no complete way to work around this.

[logician]

Can you not use two SQW GEN in series with one having some delay time to the required shift from 0v and a long width, thereby setting a voltage from which the main pulsed generator then operates?

[Elvee]

Originally Posted by EdwinR3

The quiescent current being the collector current from the output transistors when there off ?

Yes

The description in futura-science states for "warm" class AB a quiescent current of 150 to 200 mA, so 10% of it could make 15mA. Why is this to large ? Risk of saturation of the core, with hereby larger currents in primary and output transformers ?

Two things have to be considered here: generally, designers do not allow more than ~25mV offset for a general purpose amplifier (directly driving the voice coil of a speaker), which for 8Ω translates to ~3mA, because otherwise the OP works in an asymetrical way at low power level, changing the harmonic profile, etc
With a transformer, the problem is exacerbated because of the much lower resistance of the primary.
But with a transformer, other specific problems do appear, specially in those having a high quality magnetic material of high permeability.
Very small bias currents are sufficient to cause saturation, resulting in severe distortion.

Here you already lost me. Is this 2.2mV a result of the simulation or can it be calculated from the Ios, or measured on prototype ?

It is based on the info you provided: 2.2mV/0.15Ω~=15mA

This was a choice I made to try avoiding DC offset problems and to linearize the core, since the lower signals will be filtered out. In sim the air gap is included in the transformer model. With 2x0.2mm I find no more saturation with DC (not at 0.01Hz and 20Vrms)

If it does not require an excessive reactive current from the amplifier at low frequency, then it's OK.
It is simply a thermal problem: the Circlophone will happily drive severely reactive loads at high currents, but operating in those conditions will cause high dissipation in the OP transistors.

This one I could verify with the sim. I found out that inserting a resistor of 0.6 to 0.8 Ohm on the primary side gives the same result. What puzzles me is that the capacitance of an ESL is always quoted as making a it hard task to drive for an amp, yet I would need to insert an even larger capacitor to protect the amp.

What counts for the amplifier is the reactive power it has to provide, ie the product of current and voltage across a capacitor.
The ESL panel sees the full (transformed) output voltage, and the full output current: this means a high reactive power.
When you add a proper coupling capacitor in series, this capacitor sees the full current, but no voltage at all, since it is calculated to have a negligible effect at audio. The product of something (the current) by ~0 (the voltage) is ~0, thus no problem there.

Indeed this is contrary to "standard audiophile tradition". I always understood that E-lityc caps are causing a lot of distortion. It should be a non-polarized type as well I suppose.
I know of the principle to include the transformer in the FB loop, but I have no idea how to do this. It occurs to me that the base for the development of the circlophone was to remove the transformer from the original concept. Maybe for an ESL amp, it is logic to stay with the original topology.
Including the trafo in the FB, means that the amp must be integrated in the ESL (this is possible but will require some reworking). I'm not so sure about the including of the HV in the amplifier, unless the FB only includes the primary of the trafo.

The transformer is a much larger source of distortion than Elytics, and if you don't think including it in the FB is necessary, then you have even less reasons to have qualms about the capacitor.
Ideally, it should be bipolar, but in practice for voltages well under 100mV, a polarized lytic will work transparently and indefinitely.

I did restart from the schematic of page1 and made some gradual modifications, based on following ideas :
1. at 20Khz, the input is limited. When the supply is increased this limitation was removed. This also seems to reduce distortion (I went to +/- 54V).
2. The increased rail voltage exceeds the voltage-range for the input transistors Q3/Q4 so I swapped the BC560b (Vce=45V), to BC556B (Vce=65V). This might be less low-noise, but I don't know if it matters here.
3. Q13 is changed to a 2SC3601 to exceed Vce > 2xVsupply)
4. Q8/Q10 are replaced with MJE21194, with higher power rating.
4. Based on on off the last schematics of (Keantoken I believe), I played with some capacitors parallel to emitter resistors of Q1/Q2/Q13. These capacitors do not seem to have a influence on the bias-servo, but reduce some oscillations at high frequencies (20 tot 100 Khz).
5. With these modifications and the insertions of a primary side resistor of 0.66 Ohm, I'm getting very good results (THD 2-20 of 0.004% from 160Hz to 20KHz on HV side of transformer, so transformer distortion include), phase margin of > 90° and gain margin of 28dB.

It looks good, but be prepared for some tweaking in the real circuit: the dynamic aspects are not covered very reliably by the sim, and some adaptations may be necessary.

[krokkenoster]

Originally Posted by Elvee

Yes

Two things have to be considered here: generally, designers do not allow more than ~25mV offset for a general purpose amplifier (directly driving the voice coil of a speaker), which for 8Ω translates to ~3mA, because otherwise the OP works in an asymetrical way at low power level, changing the harmonic profile, etc
With a transformer, the problem is exacerbated because of the much lower resistance of the primary.
But with a transformer, other specific problems do appear, specially in those having a high quality magnetic material of high permeability.
Very small bias currents are sufficient to cause saturation, resulting in severe distortion.



It is based on the info you provided: 2.2mV/0.15Ω~=15mA



If it does not require an excessive reactive current from the amplifier at low frequency, then it's OK.
It is simply a thermal problem: the Circlophone will happily drive severely reactive loads at high currents, but operating in those conditions will cause high dissipation in the OP transistors.



What counts for the amplifier is the reactive power it has to provide, ie the product of current and voltage across a capacitor.
The ESL panel sees the full (transformed) output voltage, and the full output current: this means a high reactive power.
When you add a proper coupling capacitor in series, this capacitor sees the full current, but no voltage at all, since it is calculated to have a negligible effect at audio. The product of something (the current) by ~0 (the voltage) is ~0, thus no problem there.



The transformer is a much larger source of distortion than Elytics, and if you don't think including it in the FB is necessary, then you have even less reasons to have qualms about the capacitor.
Ideally, it should be bipolar, but in practice for voltages well under 100mV, a polarized lytic will work transparently and indefinitely.




It looks good, but be prepared for some tweaking in the real circuit: the dynamic aspects are not covered very reliably by the sim, and some adaptations may be necessary.

Matey's I wonder how this will look like in solder wikre pcb and componente
Krokkenoster

[Elvee]

Originally Posted by krokkenoster

Matey's I wonder how this will look like in solder wikre pcb and componente
Krokkenoster

Directly driving a transformer from the output of the amplifier looks quite risky.

A much better option is to insert a DC blocking capacitor, and to rearrange the feedback to include both the transformer and the Elytic in the loop:

[danielwritesbac]

Here it is attached.
It looks like this sim is useful to explore quite a few different options of refining the sensor behavior.
Personally, I was never able to figure out which mods were best, except that the integrated buffer always looked pretty good.