()blivion
Active Member
OK. I have done some tests and tried some suggestions. The following are my thoughts, notes, and some data. Note that I'm not perfect, I'm sure there are some mistakes and false assumptions in here. Just know this is an ongoing process with room for improvement.
Decreasing switching times.
I think Q1 switching ON is a non issue under all running conditions. This is because it is always going to switch with the inductor at ~0 current and with a very low voltage across it. This is effectively "zero voltage switching". So even if it lingers in the liner region where it would be wasting power, there is no power AVAILABLE for Q1 to waste at this point in the cycle. Unless anyone has better observations and objections, I'm just going to lay this to rest right here and now.
Q1 turning OFF time is a whole different matter though. First, the switching OFF happens where ALL of the power is gathered. So efficient switching here should be considered a top priority to be sure. Second, I'm not absolutely certain, but I think ~400ns is going to be a hard lower limit for switching OFF Q1. I was reading about BJT switching when coming out of saturation, I guess there remains some charge on the base that takes time to drain. This charge makes it so even if you are driving the transistor as hard as is needed, it's still not going to come out of saturation any faster. This problem is even worse for discrete parts do to larger die size/area for base charge to accumulate in. Now there are ways to fix this, [1] [2] but these methods are always about keeping the transistor from going into saturation in the first place. This is obviously not desired because it would cause other types of loss that would be even worse.
I think the only practical way to fix the issue is to run the circuit at a low enough frequency to keep the switching ratio high. Running at 25KHz gives you a 100:1 ratio, which is plenty efficient, though it is too close to human hearing for my tastes. 50KHz halves the ratio to 50:1, which is still quite decent. And 100KHz gives you about 5% switch transitioning time, which could be rounded up directly to 5% efficiency loss. Some where between 40KHz and 70KHz seems sane to me. That, or come up with another way to fix the problem.
Battery Model.
My battery is assumed to vary from 3 volts, down to 0.7 volts, and is assumed to have around 5 Ohms internal resistance. I choose 5 ohms to model for a very dead battery, which should have quite a high internal resistance. It may be much higher, it may be much lower, I don't really know exactly what is ideal. If anyone thinks these values are unreasonable, then please let me know what you think I should model with and I may change it.
Vin capacitor BEFORE the devices on switch.
I tried this out as it seemed like a great idea. And it did kind of help with starting but not enough to change the circuits minimum Vin much. It appears to only provide enough "boosted current" for a few cycles before it drains out to following the battery. The only way to make it more effective is to make the Vin capacitance impractically large, while making Vout capacitor impractically small. This has more negative effects than positive I'm afraid.
Ringing on L1.
I 100% agree with Mr RB's assessment of this ringing. It is such a low energy that there is no point in adding parts just to deal with it. If there happens to be a "free" way to fix it, then it should be done, if only to make the waveform look cleaner. But otherwise I'm just going to close the issue right here.
Transistors
Keeping with the "use only common parts" credo, I am using "general propose" BJT's. This means 2N2222, 200+ hfE and similar parts. We can for sure get better performance from something like the BC337-40, but the majority of people looking to build this circuit are going to use whatever they have in the junk box at the time. I could instruct people to only use good high gain switching transistors, but I think it might be better to make the design work with basic small signal transistors, but have room to upgrade to better ones for better performance when needed. Also, I myself don't have any really good switching transistors to use for testing, so if this was changed, all information would remain in simulation, assuming I am the only one building the circuit.
Bypass capacitor on Q1
I am playing with the idea of adding an AC bypass capacitor somewhere near Q1. The idea is that such a capacitor would provide strong base current when oscillating, but would lower or remove quiescent current drain altogether when in discontinuous mode. I also played with putting one across R1 to give stronger Q1 base drive during high frequency oscillation, and lower drive during low speed. I did this to try and get Q1 to switch off faster, at the cost of extra Q1 base current. But as I stated above under the section "Decreasing switching times", Q1 switching OFF time doesn't seem to want to go under ~400ns no matter how strong the base drive is. I think this capacitor is only ever going to be useful for either lowering quiescent current, and/or effecting the regulation.
Regulation modes.
When playing around with the circuit, I noticed that there are distinct ways regulation can limit output.
(1) The bias on Q1 base is altered as C2 gets charged from leakage through Zd1. This causes the current level Q1 turns off at to drop as Vout rises. In other words, the current decreases proportionate to the voltage increase.
(2) The oscillation duty cycle changes from an even 50/50 to other ratios as Vout gets up to and beyond 2X Vin. This happens because a 1:1 charge/discharge duty cycle can only ever double the voltage for a given current.
(3) Vout gets near enough to Zd1 reverse breakdown that the spikes from T1 primary momentarily push through the zener. This causes different bursts of counter biasing that limit the current in a staircase pastern.
(4) Finally, Vout reaches full blown zener breakdown levels. This causes any voltage on C2 to keep from discharging through the zeners forward drop. Because of this, C2 can only discharge through Q2 base, holding the oscillator fully off.
The following is a waveform from a circuit configuration that exaggerates each different mode of operation.
View attachment 67336
This is interesting information because if we can make good use of these two different modes of operation, then we can better adjust circuit properties for particular circumstances and applications. For example, if we could skillfully avoid the current limiting mode and go right to fully stopping the oscillation, we could potentially gain efficiency because during discontinuous mode we don't have any switching loss. This would cause more noise on Vout though. Or we could do the opposite, and get less extreme ripple, at the cost of some efficiency. This is very similar to the effects of changing the running frequency up or down, so we could do that also to get each arrangement's benefits.
Load and line regulation
I can't give any load and line regulation simulation figures as I am still to novice at LTspice to be able to know how to make a varying voltage or resistance for a transient analysis. Though admittedly I only gave it about half an hour of trying. However, I have played with load and line regulation on the actual device. Load regulation is easy to understand. It is 99% dependent on the available line energy. If the voltage and current available is fairly high, then the circuit will boost and regulation will be fairly steady (± 0.1 Volts) from no load up to very near battery wattage. After the load current exceeds what the battery can fully supply, then Vout starts to sag. This is almost entirely caused by the battery's internal resistance. The more battery's you have in parallel, the lower the internal resistance, and thus the lower the voltage of each cell can be. Bad battery's tend to have low voltage AND high internal resistance. So combining them in either series or parallel will allow the circuit to boost to full the voltage. Note that the circuit doesn't seem to care if it gets it's watts from volts, or current. As long as the total load watts + system loss is on the line, it will be boosted to the proper regulated voltage. The only times this isn't true is when there are mode changes. As seen above, there are different modes of operation. The one for the transition to 2x Vin has the biggest effect on regulation. I think the total regulation, over all operating modes and conditions is about 0.3 volts. Could be better, could be worse. And all this will probably change as the circuit is tweaked more also.
Precision adjustable voltage reference, ("super zeners")
It somewhat goes against the "use only common parts" credo, but we could use a Tl431 like precision zener. The advantages is that the TL431 and similar is 100% adjustable, so we could use a trimmer pot and potentially get a tunable voltage from a single board, without changing any parts. Also, they have much more exact reverse breakdown voltages than real zeners do, making regulation a more ON/OFF thing. The disadvantage to this is that it is harder to simulate (I STILL don't have a working model for the TL431) and it would add one or two more resistors, needed to set the voltage. Also, such a change might have unintended side effects, such as the fact that a zener has a reverse AND forward mode of conduction. The TL431 may not like behaving this way, which could change the way the circuit currently operates.
That's what I got so far. Let me know what you think.
-()blivion
Decreasing switching times.
I think Q1 switching ON is a non issue under all running conditions. This is because it is always going to switch with the inductor at ~0 current and with a very low voltage across it. This is effectively "zero voltage switching". So even if it lingers in the liner region where it would be wasting power, there is no power AVAILABLE for Q1 to waste at this point in the cycle. Unless anyone has better observations and objections, I'm just going to lay this to rest right here and now.
Q1 turning OFF time is a whole different matter though. First, the switching OFF happens where ALL of the power is gathered. So efficient switching here should be considered a top priority to be sure. Second, I'm not absolutely certain, but I think ~400ns is going to be a hard lower limit for switching OFF Q1. I was reading about BJT switching when coming out of saturation, I guess there remains some charge on the base that takes time to drain. This charge makes it so even if you are driving the transistor as hard as is needed, it's still not going to come out of saturation any faster. This problem is even worse for discrete parts do to larger die size/area for base charge to accumulate in. Now there are ways to fix this, [1] [2] but these methods are always about keeping the transistor from going into saturation in the first place. This is obviously not desired because it would cause other types of loss that would be even worse.
I think the only practical way to fix the issue is to run the circuit at a low enough frequency to keep the switching ratio high. Running at 25KHz gives you a 100:1 ratio, which is plenty efficient, though it is too close to human hearing for my tastes. 50KHz halves the ratio to 50:1, which is still quite decent. And 100KHz gives you about 5% switch transitioning time, which could be rounded up directly to 5% efficiency loss. Some where between 40KHz and 70KHz seems sane to me. That, or come up with another way to fix the problem.
Battery Model.
My battery is assumed to vary from 3 volts, down to 0.7 volts, and is assumed to have around 5 Ohms internal resistance. I choose 5 ohms to model for a very dead battery, which should have quite a high internal resistance. It may be much higher, it may be much lower, I don't really know exactly what is ideal. If anyone thinks these values are unreasonable, then please let me know what you think I should model with and I may change it.
Vin capacitor BEFORE the devices on switch.
I tried this out as it seemed like a great idea. And it did kind of help with starting but not enough to change the circuits minimum Vin much. It appears to only provide enough "boosted current" for a few cycles before it drains out to following the battery. The only way to make it more effective is to make the Vin capacitance impractically large, while making Vout capacitor impractically small. This has more negative effects than positive I'm afraid.
Ringing on L1.
I 100% agree with Mr RB's assessment of this ringing. It is such a low energy that there is no point in adding parts just to deal with it. If there happens to be a "free" way to fix it, then it should be done, if only to make the waveform look cleaner. But otherwise I'm just going to close the issue right here.
Transistors
Keeping with the "use only common parts" credo, I am using "general propose" BJT's. This means 2N2222, 200+ hfE and similar parts. We can for sure get better performance from something like the BC337-40, but the majority of people looking to build this circuit are going to use whatever they have in the junk box at the time. I could instruct people to only use good high gain switching transistors, but I think it might be better to make the design work with basic small signal transistors, but have room to upgrade to better ones for better performance when needed. Also, I myself don't have any really good switching transistors to use for testing, so if this was changed, all information would remain in simulation, assuming I am the only one building the circuit.
Bypass capacitor on Q1
I am playing with the idea of adding an AC bypass capacitor somewhere near Q1. The idea is that such a capacitor would provide strong base current when oscillating, but would lower or remove quiescent current drain altogether when in discontinuous mode. I also played with putting one across R1 to give stronger Q1 base drive during high frequency oscillation, and lower drive during low speed. I did this to try and get Q1 to switch off faster, at the cost of extra Q1 base current. But as I stated above under the section "Decreasing switching times", Q1 switching OFF time doesn't seem to want to go under ~400ns no matter how strong the base drive is. I think this capacitor is only ever going to be useful for either lowering quiescent current, and/or effecting the regulation.
Regulation modes.
When playing around with the circuit, I noticed that there are distinct ways regulation can limit output.
(1) The bias on Q1 base is altered as C2 gets charged from leakage through Zd1. This causes the current level Q1 turns off at to drop as Vout rises. In other words, the current decreases proportionate to the voltage increase.
(2) The oscillation duty cycle changes from an even 50/50 to other ratios as Vout gets up to and beyond 2X Vin. This happens because a 1:1 charge/discharge duty cycle can only ever double the voltage for a given current.
(3) Vout gets near enough to Zd1 reverse breakdown that the spikes from T1 primary momentarily push through the zener. This causes different bursts of counter biasing that limit the current in a staircase pastern.
(4) Finally, Vout reaches full blown zener breakdown levels. This causes any voltage on C2 to keep from discharging through the zeners forward drop. Because of this, C2 can only discharge through Q2 base, holding the oscillator fully off.
The following is a waveform from a circuit configuration that exaggerates each different mode of operation.
View attachment 67336
This is interesting information because if we can make good use of these two different modes of operation, then we can better adjust circuit properties for particular circumstances and applications. For example, if we could skillfully avoid the current limiting mode and go right to fully stopping the oscillation, we could potentially gain efficiency because during discontinuous mode we don't have any switching loss. This would cause more noise on Vout though. Or we could do the opposite, and get less extreme ripple, at the cost of some efficiency. This is very similar to the effects of changing the running frequency up or down, so we could do that also to get each arrangement's benefits.
Load and line regulation
I can't give any load and line regulation simulation figures as I am still to novice at LTspice to be able to know how to make a varying voltage or resistance for a transient analysis. Though admittedly I only gave it about half an hour of trying. However, I have played with load and line regulation on the actual device. Load regulation is easy to understand. It is 99% dependent on the available line energy. If the voltage and current available is fairly high, then the circuit will boost and regulation will be fairly steady (± 0.1 Volts) from no load up to very near battery wattage. After the load current exceeds what the battery can fully supply, then Vout starts to sag. This is almost entirely caused by the battery's internal resistance. The more battery's you have in parallel, the lower the internal resistance, and thus the lower the voltage of each cell can be. Bad battery's tend to have low voltage AND high internal resistance. So combining them in either series or parallel will allow the circuit to boost to full the voltage. Note that the circuit doesn't seem to care if it gets it's watts from volts, or current. As long as the total load watts + system loss is on the line, it will be boosted to the proper regulated voltage. The only times this isn't true is when there are mode changes. As seen above, there are different modes of operation. The one for the transition to 2x Vin has the biggest effect on regulation. I think the total regulation, over all operating modes and conditions is about 0.3 volts. Could be better, could be worse. And all this will probably change as the circuit is tweaked more also.
Precision adjustable voltage reference, ("super zeners")
It somewhat goes against the "use only common parts" credo, but we could use a Tl431 like precision zener. The advantages is that the TL431 and similar is 100% adjustable, so we could use a trimmer pot and potentially get a tunable voltage from a single board, without changing any parts. Also, they have much more exact reverse breakdown voltages than real zeners do, making regulation a more ON/OFF thing. The disadvantage to this is that it is harder to simulate (I STILL don't have a working model for the TL431) and it would add one or two more resistors, needed to set the voltage. Also, such a change might have unintended side effects, such as the fact that a zener has a reverse AND forward mode of conduction. The TL431 may not like behaving this way, which could change the way the circuit currently operates.
That's what I got so far. Let me know what you think.
-()blivion
Last edited: