Hi
I would like to think that the cross conduction issue has been fixed after the long discussion some time ago : http://www.diyaudio.com/forums/solid-state/258208-cross-conduction-complimentary-pairs.html
To save you reading all that, even though it should be pinned, it is the collector resistors of the driver transistor in the complimentary pair that make all the difference. Most designs use 100R or 120R but I have used 33R. You can even go lower.
My drivers are BC327/337 and the power transistors are MJE15028G/15029G, 3 pairs of each so total of 12 transistors. The BC327/337s seem small but are not, they get to about +20C above ambient, and with an aluminium plate glued on them I have brought this down to +10C over ambient.
The output transistors get very hot and require heatsink and two fans. Considering this is a portable device that fits in a medium size enclosure, the heat generated is a problem, but so far manageable. I have the heatsinks and fan in such orientation so that the hot air is expelled without coming into contact with other components, to avoid overheating everything else. Normal operation is at around 65C-70C on the heatsink.
On simulation each output complementary pair and its emitter resistor spend around 4W - multiply by 6 and we get 24W spent on the board and 23W on the load.
I would like to think that the cross conduction issue has been fixed after the long discussion some time ago : http://www.diyaudio.com/forums/solid-state/258208-cross-conduction-complimentary-pairs.html
To save you reading all that, even though it should be pinned, it is the collector resistors of the driver transistor in the complimentary pair that make all the difference. Most designs use 100R or 120R but I have used 33R. You can even go lower.
My drivers are BC327/337 and the power transistors are MJE15028G/15029G, 3 pairs of each so total of 12 transistors. The BC327/337s seem small but are not, they get to about +20C above ambient, and with an aluminium plate glued on them I have brought this down to +10C over ambient.
The output transistors get very hot and require heatsink and two fans. Considering this is a portable device that fits in a medium size enclosure, the heat generated is a problem, but so far manageable. I have the heatsinks and fan in such orientation so that the hot air is expelled without coming into contact with other components, to avoid overheating everything else. Normal operation is at around 65C-70C on the heatsink.
On simulation each output complementary pair and its emitter resistor spend around 4W - multiply by 6 and we get 24W spent on the board and 23W on the load.
My initial reaction is that this looks like there may be room for improvement.
You only have an 8 V peak output, however the transistors are optimised for audio amplifiers, 120V breakdown on the 15028/15029 but not super fast.
Can you not use faster and lower loss transistors?
You also have a bit of an optimisation problem on Vcc.
Some of your losses are substantially constant, you need a few volts minimum for Vce, for instance, so these become proportionally less as the rails increase.
And so do resistive and transformer losses, even if minor.
But you probably have a practical desire to keep to a convenient battery pack.
Did you evaluate alternate Vcc? Maybe smaller batteries in series.
Then just reduce your transformer step-up ratio.
Best wishes
David
You only have an 8 V peak output, however the transistors are optimised for audio amplifiers, 120V breakdown on the 15028/15029 but not super fast.
Can you not use faster and lower loss transistors?
You also have a bit of an optimisation problem on Vcc.
Some of your losses are substantially constant, you need a few volts minimum for Vce, for instance, so these become proportionally less as the rails increase.
And so do resistive and transformer losses, even if minor.
But you probably have a practical desire to keep to a convenient battery pack.
Did you evaluate alternate Vcc? Maybe smaller batteries in series.
Then just reduce your transformer step-up ratio.
Best wishes
David
Last edited:
What Dave Zan said.
You do not need an oscillator driving a linear amplifier (that's why it's so inefficient), but:
1) a fixed frequency power inverter
2) that power inverter output (which by itself may be switching transistors , pure square wave or worse: squarewave with much less than 50% duty cycle, which means Class C) drives a high Q resonant tank which kills harmonics and turns that horrible waveform into a sinewave (remember, fixed frequency) and after that it is further filtered by extra LC filters until output has no harmonics, period.
Exactly what the RF guys do.
Don't think of this as an audio amplifier working above the Audio spectrum but as a VLF radio transmitter.
Investigate Radio transmitters at an RF Forum, not at DIY Audio.
The SMPS or Class D guys here do work with such frequencies, but their goals are others.
FWIW typical RF transmitters are made by design, no matter wheter they are powered by a 7.2V battery pack or +2500V , to drive 50 ohms, which is a typical coaxial line impedance (or 75 ohms) so by definition they are already close to the 100 ohms you need, scaling up for the difference is easy.
As an example of what I'm talking about, here's a 30/100W RF transmitter , in this case producing around 500 kHz output ... here you practically have the answer to what you need, just needs scaling down (tuning) to 100kHz and matching to a 100 ohms load instead of a 50 ohms one.
Notice that the output transistor runs UNbiased and is heavily overdriven, it works in highly efficient Class C .
Then output frequency is heavily filtered by L2/L3/C2/C3/C4 which are more than "just LC filters" , they are tuned filters at the fundamental frequency, much stronger than what you imagine.
Remember that RF transmitters are required by Law to attenuate harmonics (i. e.distortion) by 50 dB, 60 or better, to avoid interference on other bands.
This transmitter puts out 100W with 24V and 30W with 13V , so it's practically what you need.
And is driven by a very simple CMOS oscillator.
You do not need an oscillator driving a linear amplifier (that's why it's so inefficient), but:
1) a fixed frequency power inverter
2) that power inverter output (which by itself may be switching transistors , pure square wave or worse: squarewave with much less than 50% duty cycle, which means Class C) drives a high Q resonant tank which kills harmonics and turns that horrible waveform into a sinewave (remember, fixed frequency) and after that it is further filtered by extra LC filters until output has no harmonics, period.
Exactly what the RF guys do.
Don't think of this as an audio amplifier working above the Audio spectrum but as a VLF radio transmitter.
Investigate Radio transmitters at an RF Forum, not at DIY Audio.
The SMPS or Class D guys here do work with such frequencies, but their goals are others.
FWIW typical RF transmitters are made by design, no matter wheter they are powered by a 7.2V battery pack or +2500V , to drive 50 ohms, which is a typical coaxial line impedance (or 75 ohms) so by definition they are already close to the 100 ohms you need, scaling up for the difference is easy.
As an example of what I'm talking about, here's a 30/100W RF transmitter , in this case producing around 500 kHz output ... here you practically have the answer to what you need, just needs scaling down (tuning) to 100kHz and matching to a 100 ohms load instead of a 50 ohms one.
An externally hosted image should be here but it was not working when we last tested it.
Notice that the output transistor runs UNbiased and is heavily overdriven, it works in highly efficient Class C .
Then output frequency is heavily filtered by L2/L3/C2/C3/C4 which are more than "just LC filters" , they are tuned filters at the fundamental frequency, much stronger than what you imagine.
Remember that RF transmitters are required by Law to attenuate harmonics (i. e.distortion) by 50 dB, 60 or better, to avoid interference on other bands.
This transmitter puts out 100W with 24V and 30W with 13V , so it's practically what you need.
And is driven by a very simple CMOS oscillator.
I'd use a MOSFET H-bridge to generate the square wave. ADP3120A and its buddies from OnSemi are really nice little MOSFET drivers, super cheap, simple to use, really fast, and they can really kick some butt (>1A gate drive). Add some adequate MOS, and resonant circuit as said above.
Is this circuit dependent on proper tuning with the aerial in order to (a) produce the required power and (b) achieve the efficiency?
I ask because my load is variable, from around 80R to maybe even 150R. There will be no way to "tune" anything, the output stage will be required to cope with whatever load you throw at it.
If it has to cope with anything thrown at it then Class E is out. Class F may be out too. You may be back to Class D again.
Currently it is being powered by home made battery packs, each being two 12V 5000mAh batteries in series to produce +/-12V at 5000mAh. Current draw is from 1.5A to over 3.5A as the batteries deplete and the Vcc drops. Cutout is at +/-10.5V. Fully charged batteries are at 12.3V each.
One of the cross conduction issues is "sticking to the rails" (I am not sure of this old school expression), and the closer the output gets to the rails the more serious the danger of cross conduction and overheating. That is why in my implementation I can only reach +/-7.5V even though the rails maybe at +/-10.5V - if I try to go closer there are problems even with the 33R collector resistors. I have to make do with 1:10 on the transformer rather than with something smaller.
I have an updated design in the wings. It uses lower emitter resistors from 0.2R to 0.1R, it has much wider tracks for the heavy currents to flow through and it uses a larger transformer, from RM12 to RM14, for fewer losses.
I also discovered that the ferrite material N87 is much better at 200KHz than the N97 I was using up to now.
But I am not committing just yet, hence asking here.
I will look into different transistors, would you have a suggestion to replace the MJE15028 (ft=30MHz)? So far I have found the much faster 2sc5707 /
2sa2040 at 300MHz ft and rise and fall times 5-10 times better than the MJE15028, but their packaging is a bit problematic (no hole to mount it on heatsink)
I think this is a lot of your problem.
The maximum instantaneous dissipation in an output transistor occurs when it has half the rail across it (and half across the load).
Most text books do a calculation for a sine wave to find the amplitude with the maximum total dissipation over time.
Your 8 volt peak has an RMS value of 5.6 V, very close to half your 10.5 rail under load.
That must be close to worse case (I have skipped over the calculus details because it's not a very sensitive function)
So more rail Vcc would help, probably considerably.
From memory 42 V DC is considered "Low volts" in the electrical code.
Not sure on how they classify +- supplies but I would try for at least +-24 V if you can and preferably +- 36 V.
A few power tool companies make 24 and 36 V lithium ion battery packs for portable tools that would be less heavy than your Lead acids.
Otherwise two or three Lead acids in series.
I assume you have used the CFP structure in an attempt to reduce the wasted headroom to the rails.
As you have noted there are oscillation problems that are soluble but maybe cost you more volts than just an emitter follower.
An EF with a boosted rail for all but the output transistors would be another option, complicates your battery however.
This will help, but back of envelope calculation makes me think these are small losses compared to improvement from the raised rails.
These look like an improvement too, I assume there are clips for these because they are a power transistor so someone must have a heatsink for them.
But I really think that ultimately the way to do it is a switch mode system of some sort, even if it's called it "Class C" (or whatever)
JMFahey's idea to look at low/medium frequency RF is excellent, not least because there is a DIY community with a tradition of helpfulness.
Sorry I don't know their nomenclature, my "3 step sine wave approximation" is probably called "Push-Pull class C" or some such.
Ask a few ham enthusiasts and come back and tell us all the latest 😉
Best wishes
David
EFs are just less prone to cross conduction than CFPs. Use a low Rbe - maybe even 6.8 or 10 ohms. And don't drive the drivers to saturation (don't clip) so they don't have any excuses either. Heat from cross conduction quickly overcomes any efficiency gain from incrementally higher output being fully driven.
The '3 step sine wave" turns on the postive half cycle for 120 degrees, waits 60 degrees, then turns on the negative for 120 degrees, waits 60 degrees, then repeats. Low pass filter and you have a sine wave. Those power inverters that "look" like car amps do it this way - but they don't bother with the output filter. Most loads don't care. They do modiulate the duty cycle somewhat for voltage regulation.
The '3 step sine wave" turns on the postive half cycle for 120 degrees, waits 60 degrees, then turns on the negative for 120 degrees, waits 60 degrees, then repeats. Low pass filter and you have a sine wave. Those power inverters that "look" like car amps do it this way - but they don't bother with the output filter. Most loads don't care. They do modiulate the duty cycle somewhat for voltage regulation.
I am thinking aloud - answering ...
The 2 * 12V batteries were chosen after deciding that 12V batteries and PSUs are the easiest to find.
I use LiIon batteries which are 3.7V per cell (nominal) and the most common are the 3 cell batteries giving 11.1V per pack. Fully charged around 12.3V and "depleted" I stop them at 10.5V.
The closer to the rails the more efficient we become, for a fixed power output. And the further away from the rails, the more power we spend on the output stage as heat.
The range for Vcc is 12.3V-10.5V. It is very hard to set the right bias to work equally well at 12.3V and at 10.5V. Typically I set the min bias possible at 10.5V (checking output curve on load on the scope) but then at 12.3V the bias is much higher.
Because the output wave is fixed in amplitude, at 10.5V we spend less power than at 12.3V (17 Watts vs 30 Watts) - almost double!!!
Because this is a portable device I am also very limited on the heatsink size.
Solutions:
Get faster output transistors to reduce cross conduction issues (remember we are at 200KHz) and achieve a larger amplitude on the output. The MJE15028/29 is 30MHz and at 50W they do get hot, all 6 of them (heatsink at 70C) - I cannot find anything in a TO-220 case to be faster. The transistors I mentioned before (2SA2040-2SC5707) are only 15W each, would not bother trying them at all (would be much harder trying to stay within SOA at 70C than it is with the MJE15028/29)
Maybe try the Darlington route (as per wg_ski), see if they are better in this respect? Rod Elliot says they are worse in all respects especially thermal runaway.
Or maybe try a MOSFET stage, using Lateral MOSFETs to be safe, ALF08P16V and ALF08N16V ? Biasing is different for MOSFETs and I do not use negative feedback at all, and from what I see, MOSFET amp designers use negative feedback in order to solve the biasing issues between N-channel and P-channel...
2) Increasing the Vcc does not help on its own - we need to increase the amplitude of the sine wave too otherwise we are back to square one. Therefore savings in power can be made by increasing the amplitude of the wave RELATIVE and PROPORTIONALLY to the Vcc. Some extra savings due to "fixed" costs can be made by having say 18.5V over 11.1V, but they are small in comparison (less turns on the transformer for one, less power spent on the emitter resistors). The hassle of getting a different LiIon pack - adapters, chargers, not sure it is worth it.
Well, of course.
Quick back of envelope, rounded numbers.
At the moment
11.2 V nominal rail. 8 V peak. 5.6 V RMS across the transistor, one unit of current per transistor (per 1 amp for comparison) 5.6 W
New
36 V nominal 33 V peak 23.3 V RMS so just under one quarter the current. 12.7 V across the transistor so with one quarter the current, ~3.1 W.
You cut your power dissipation in the transistor almost in half for the same power into the transformer.
Then there are minor improvements because the currents are lower so trace resistance losses are less, transformer improvements and so on.
Total improvement in power dissipation should be close to 100%.
Plus the improvement in efficiency.
Worth it?
Best wishes
Few more ideas.
FETs are attractive but would require a bit of a rethink.
Bootstrapped drive? Self discusses this, used in car amps with limited rail volts.
Since you have a transformer output you can easily use push-pull like the old tube amps.
So you don't need complementary pairs, you can use 2 of the more efficient N-Fets.
Last edited:
But I am not using a fixed power supply. I am using batteries that vary from 10.5V up to over 12.3V.
The biasing scheme has to be adequate and work without distortion at 10.5V but then becomes way to high at 12.3V. This results in total heat at around 17.4 Watts at 10.5V and maybe 30-35W when the battery is full at 12.3V.
Had I be using a larger battery, for example a 5 cell, at nominal 18.5V, it would vary from 17.5V to 20.5V, and then the power/heat would vary from 12Watts to 50Watts or more!!!
Therefore I now think the first problem to solve is a biasing scheme that adjusts on varying Vcc, keeping the biasing current more or less constant.
My current biasing scheme consists of two transistors connected as diodes and mounted on the heatsink, bypassed by a trimmer. It works for temperature, but not for voltage.
I will start another thread to discuss this.
The biasing scheme has to be adequate and work without distortion at 10.5V but then becomes way to high at 12.3V. This results in total heat at around 17.4 Watts at 10.5V and maybe 30-35W when the battery is full at 12.3V.
Had I be using a larger battery, for example a 5 cell, at nominal 18.5V, it would vary from 17.5V to 20.5V, and then the power/heat would vary from 12Watts to 50Watts or more!!!
Therefore I now think the first problem to solve is a biasing scheme that adjusts on varying Vcc, keeping the biasing current more or less constant.
My current biasing scheme consists of two transistors connected as diodes and mounted on the heatsink, bypassed by a trimmer. It works for temperature, but not for voltage.
I will start another thread to discuss this.
The battery variation will be the same percent for either scheme.
So the numbers are still valid as nominal (typical) values, to show the approximate improvement.
If you plan a substantial rework then you may as well move to a switcher/class C scheme.
No bias problem at all, since there's no bias, and no power wasted on bias pretty much solves your heat problem.
Best wishes
David
I have done some work on higher voltage supplies. In theory (simulation) the higher voltage supplies produce less heat, but most op-amps can accept max +/18V and can produce +/-15V under load. OK there is one OPA445, will need to try it to see how it performs at 200KHz. So then I tried the LM3886 which has lots of advantages over a discrete design (very good quality, tiny PCB, well protected for over temperature etc) - however it starts to roll off heavily aftewr around 30KHz and cannot do much at 200KHz under load.
I will try the OPA445 next and take it from there.
I will try the OPA445 next and take it from there.
- Status
- Not open for further replies.