So been finalising a little.
1. The booster maths ties Ipeak * frequency * L = C * duty cycle
2. The inductor maths 1/Ltotal =1/L1 +1/L2 + ... for inductors in parallel. (ie Ltotal = L1*L2/(L1+L2) )
The previous PCV2 inductors coped with high current, providing 560uH each with little saturation impact, the down side is they have zero EMI shielding and act as BBC Crystal Palace radio mast transmitters.
The switch to shielded AGP-series inductors sees a lower 460uH and 30% inductance drop at about 4A, so although they can take 12A each their inductance would drop considerably at that point.
Doing some quick booster maths - the previous setup
75kHz, 24Vin, 260Vout, 600mA and 0.1Vripple gives us a 90% duty cycle, with 23uH minimum with a requirement of 13Apeak and a minimum of 80uF Cout. The diode seeing a good 13+Amps. I Use 260Vout to give some room for regulators but so far we've been seting this to 230V for all the sims..
The PCV2 inductors max was 7Arms. Hence two in parallel may drop the inductance to 290uH total, the current handling is 14A. Hence two were chosen originally.
The AGP sees both lower 'usable' average current and lower inductance - 470uH which in parallel is 235uH, so either peak current has to increase (nope!) or frequency (easier).
If we can target a max working current of 2x4=8A (with 24A peaks acceptable) then we can use two AGPs. The inductance still exceeds the minimum required (235uH vs 80).
So pushing up the frequency to 95kHz with a 90.7% duty cycle, we see the minimum inductance required (L) drop to 17uH for 13A. With our relationship from point 1 at the start.. we can that for Ipeak * L *C = C, we don't worry about C, so if we want to drop Ipeak we need to increase L. So if we have 235uH vs 17uH minimum, we're 10x that so we can reduce the current. 1.3A in an ideal world so it's likely to be more.. in reality that's going to be higher but the simulation shows it's very happy running at 4A per inductor - 2x is 8A. So we're dropping only a small amount of peak.
Now plugging the frequency back into the calculations.. we see we only need about 60uF minimum Cout for 0.1V ripple.. So happy if that's 100uF.
The AGP series plots show it's very happy up to a good 300kHz before it slowly starts increasing, so it's stable at 95kHz.
So the simulation for 95kHz and a max of 8A on the current sense limiting sees the system apply push the desired 230V 600mA giving the 3080 regulator space for 200V 1mVripple I want.
For giggles I already know that given the duty cycle, freq that the system is probably balanced for the choice of components and current limits. When you look at the maths, adding another inductor won't work, adding more frequency helps a little but in the end because duty cycle is Vin/Vout pushing 400V from 24V means a 94% duty cycle regardless of frequency with 20A.. >50% duty cycle is a danger zone from a stability perspective and we're already implementing slow compensation. At this point, going high in voltage/current from 24V is probably worth switching from a simple Booster to a different topology. This aligns with the industry norms, where we see switching to half-bridge with active secondary rectification etc.
It's possible to switch to 48V for 400V, the duty cycle naturally drops to 88%, and additional work on the IC side so a boost could work. All depends on the inductor voltage and insulation ratings.
1. The booster maths ties Ipeak * frequency * L = C * duty cycle
2. The inductor maths 1/Ltotal =1/L1 +1/L2 + ... for inductors in parallel. (ie Ltotal = L1*L2/(L1+L2) )
The previous PCV2 inductors coped with high current, providing 560uH each with little saturation impact, the down side is they have zero EMI shielding and act as BBC Crystal Palace radio mast transmitters.
The switch to shielded AGP-series inductors sees a lower 460uH and 30% inductance drop at about 4A, so although they can take 12A each their inductance would drop considerably at that point.
Doing some quick booster maths - the previous setup
75kHz, 24Vin, 260Vout, 600mA and 0.1Vripple gives us a 90% duty cycle, with 23uH minimum with a requirement of 13Apeak and a minimum of 80uF Cout. The diode seeing a good 13+Amps. I Use 260Vout to give some room for regulators but so far we've been seting this to 230V for all the sims..
The PCV2 inductors max was 7Arms. Hence two in parallel may drop the inductance to 290uH total, the current handling is 14A. Hence two were chosen originally.
The AGP sees both lower 'usable' average current and lower inductance - 470uH which in parallel is 235uH, so either peak current has to increase (nope!) or frequency (easier).
If we can target a max working current of 2x4=8A (with 24A peaks acceptable) then we can use two AGPs. The inductance still exceeds the minimum required (235uH vs 80).
So pushing up the frequency to 95kHz with a 90.7% duty cycle, we see the minimum inductance required (L) drop to 17uH for 13A. With our relationship from point 1 at the start.. we can that for Ipeak * L *C = C, we don't worry about C, so if we want to drop Ipeak we need to increase L. So if we have 235uH vs 17uH minimum, we're 10x that so we can reduce the current. 1.3A in an ideal world so it's likely to be more.. in reality that's going to be higher but the simulation shows it's very happy running at 4A per inductor - 2x is 8A. So we're dropping only a small amount of peak.
Now plugging the frequency back into the calculations.. we see we only need about 60uF minimum Cout for 0.1V ripple.. So happy if that's 100uF.
The AGP series plots show it's very happy up to a good 300kHz before it slowly starts increasing, so it's stable at 95kHz.
So the simulation for 95kHz and a max of 8A on the current sense limiting sees the system apply push the desired 230V 600mA giving the 3080 regulator space for 200V 1mVripple I want.
For giggles I already know that given the duty cycle, freq that the system is probably balanced for the choice of components and current limits. When you look at the maths, adding another inductor won't work, adding more frequency helps a little but in the end because duty cycle is Vin/Vout pushing 400V from 24V means a 94% duty cycle regardless of frequency with 20A.. >50% duty cycle is a danger zone from a stability perspective and we're already implementing slow compensation. At this point, going high in voltage/current from 24V is probably worth switching from a simple Booster to a different topology. This aligns with the industry norms, where we see switching to half-bridge with active secondary rectification etc.
It's possible to switch to 48V for 400V, the duty cycle naturally drops to 88%, and additional work on the IC side so a boost could work. All depends on the inductor voltage and insulation ratings.
So the coil craft inductors should arrive today 
Seems stocks of caps are scarce etc..
I've got a plan B - I have a 5A-9A 13.8V power supply (a maplin cheapie) from years ago. I had a look inside and it's transformer delivers around 18-22Vac but doesn't have EMI filtering. A simple NPN & op-amp acts as a regulator down to 13.8V.
This could, with a voltage doubler provide 600mA of 200V based on tapping the transformer - enough to get the amp build moving, along with a SMPS 5A supply that could be used for the heaters.
I can then continue with the SMPS but not be completely dependent on it.
I may, with a switch to the amp design, end up needing to split current across B+ and B- rails instead. Eitherway - it's possible todo this with both supplies.
Seems stocks of caps are scarce etc.. I've got a plan B - I have a 5A-9A 13.8V power supply (a maplin cheapie) from years ago. I had a look inside and it's transformer delivers around 18-22Vac but doesn't have EMI filtering. A simple NPN & op-amp acts as a regulator down to 13.8V.
This could, with a voltage doubler provide 600mA of 200V based on tapping the transformer - enough to get the amp build moving, along with a SMPS 5A supply that could be used for the heaters.
I can then continue with the SMPS but not be completely dependent on it.
I may, with a switch to the amp design, end up needing to split current across B+ and B- rails instead. Eitherway - it's possible todo this with both supplies.
Last edited:
So a cheeky little test whilst I'm working late.
Switching to 48V input, with the same parallel inductors and max 90% DC. With he same 370ohm load.
This is running about 2A RMS per inductor. 4A at 48V total ~192W.... and 155W output.
Just running a test into a 5000 ohm load, I'm interested in seeing the maximum voltage this will do with reduced current...
Switching to 48V input, with the same parallel inductors and max 90% DC. With he same 370ohm load.
This is running about 2A RMS per inductor. 4A at 48V total ~192W.... and 155W output.
Just running a test into a 5000 ohm load, I'm interested in seeing the maximum voltage this will do with reduced current...
48V to 500V 100mA into 5Kohm.. 85W in and 50W out
That's with the same current limiting in place.Now I'm not sure I want 500V anywhere near the inductors!
I'll certainly be soldering cables rather than using screw fits!Coildcraft give no statement on the inductor maximum voltage.. I suspect that it would arc over internally.
Last edited:
So after more simulations, I'm certainly considering moving to 48V.
The benefit is that for the 90% max duty cycle, it will support a wider range of B+ voltage. The down side is that I have to switch caps on the front end to low ESR 100V (£1.40 each but even lower 23mOhm) rather than the 50V. Not a big issue, there's choice to fulfil that role - the IC needs it's own 24V supply from the 48V, the fun there is the supply needs to be stiff enough to stop oscillation as the switching starts as the B+ hits 16V (if it drops below 12V then it shuts down). Putting a resistor works but it needs a cap to then cover the immediate sag. Once the thing is running it's all happy as Larry. Lastly there's the small issue of current inrush I need to sort. I'd assume that SMPS would throw it's toys out of the pram and refuse to start with a 200A peak 20uS possible. Low ESR..
The benefit is that for the 90% max duty cycle, it will support a wider range of B+ voltage. The down side is that I have to switch caps on the front end to low ESR 100V (£1.40 each but even lower 23mOhm) rather than the 50V. Not a big issue, there's choice to fulfil that role - the IC needs it's own 24V supply from the 48V, the fun there is the supply needs to be stiff enough to stop oscillation as the switching starts as the B+ hits 16V (if it drops below 12V then it shuts down). Putting a resistor works but it needs a cap to then cover the immediate sag. Once the thing is running it's all happy as Larry. Lastly there's the small issue of current inrush I need to sort. I'd assume that SMPS would throw it's toys out of the pram and refuse to start with a 200A peak 20uS possible. Low ESR..
Last edited:
http://ww1.microchip.com/downloads/...W-Vienna-PFC-Reference-Design-DS50002952B.pdf
This is interesting it's using the same AGP series inductors as differential mode filters. Only they're putting 300-400V through them on each phase. They use a different inductor (40A+) for the boost of each phase. However this really confirms that the inductors will be happy as Larry with 48V and acting as a boost inductor.
Will order the caps and other parts.. along with some amp parts..
This is interesting it's using the same AGP series inductors as differential mode filters. Only they're putting 300-400V through them on each phase. They use a different inductor (40A+) for the boost of each phase. However this really confirms that the inductors will be happy as Larry with 48V and acting as a boost inductor.
Will order the caps and other parts.. along with some amp parts..
The Mrs has just bought a Circut Joy so I'm currently investigating if, with a little poking, the little vinyl cutter will work for my PCB creating needs - I can then etch and dremel any through holes and create vias with wire initially.
Only a couple of issues initially (a) the PCB with is limited to 130mm, the width of the device and (b) the device can't handle some aspects of .SVG plot files, this caused part of the print to work but the remainder of the SVG to be essentially piled up at the 0,0 coordinates. I think there is a way to simplify the SVG output by Kicad so that the shape paths don't confuse the little 'joy' (there's a tool in GitHub that can simplify SVGs).
The PCBs quoted for local prototyping were in the region of £250 a pop! So this seems a sensible way to make an initial prototype before ordering from china.. I don't want to fork out for another flatbed cutter/router/plotter.. I have my own gadgets I want
Only a couple of issues initially (a) the PCB with is limited to 130mm, the width of the device and (b) the device can't handle some aspects of .SVG plot files, this caused part of the print to work but the remainder of the SVG to be essentially piled up at the 0,0 coordinates. I think there is a way to simplify the SVG output by Kicad so that the shape paths don't confuse the little 'joy' (there's a tool in GitHub that can simplify SVGs).
The PCBs quoted for local prototyping were in the region of £250 a pop! So this seems a sensible way to make an initial prototype before ordering from china.. I don't want to fork out for another flatbed cutter/router/plotter.. I have my own gadgets I want
This is a lovely document from onsemi: https://www.onsemi.com/pub/Collateral/HBD853-D.PDF
CCM is quieter, but this shows examples including CCM of a 270W implementation of active power factor correction and shows the THD for the implementation.
Later this gives a very good bridge less example running at 800W without needing a synchronous IC (~page 100).
Note that this would mean the power supply would need a transformer further up if we're to abide by forum rules on discussions of isolated power supplies only.
CCM is quieter, but this shows examples including CCM of a 270W implementation of active power factor correction and shows the THD for the implementation.
Later this gives a very good bridge less example running at 800W without needing a synchronous IC (~page 100).
Note that this would mean the power supply would need a transformer further up if we're to abide by forum rules on discussions of isolated power supplies only.
Last edited:
Next up is an interesting point about inductors and "coil whine": Measures Against Acoustic Noise in Power Inductors | Solution Guide | TDK Product Center
So I've been looking at modularising the design.
Next step is making a PFC which will be a CCM fixed frequency - now I'm not sure if I can get a Cuk topology PFC but that may be an option given 300W should be doable. It may even be possible to interleave them.
Cuk is a topology that is low noise for both input and output - although it was patented so I suspect people aren't using it due to that. Add to this some additional anti-noise measures such as synchronous switching, etc this may be a good thing for some of the power supplies. Still need to research and make a model.
The good news is that most of the research I've done on the booster is reusable.
Next step is making a PFC which will be a CCM fixed frequency - now I'm not sure if I can get a Cuk topology PFC but that may be an option given 300W should be doable. It may even be possible to interleave them.
Cuk is a topology that is low noise for both input and output - although it was patented so I suspect people aren't using it due to that. Add to this some additional anti-noise measures such as synchronous switching, etc this may be a good thing for some of the power supplies. Still need to research and make a model.
The good news is that most of the research I've done on the booster is reusable.
So I have a PFC and multiple Cuk output boosters all running at 200kHz in LTSpice.
First thing to say is how shocklingly low the power use is. After the initial energisation of the caps and inductors, the resulting mains power use is exceptionally low. (Sure 200kHz would result in higher switching losses so having a low capacitance high speed mosfet and diodes with low reverse losses are key to reducing this)
The noise is way above the audio band and the resulting power is lower in noise than I was expecting - partly due to the Cuk's use of capacitors that results in a smoother output. However also SMPS are very transparent - if there's 100Hz noise from the initial bridge then you'll see it unless you essentially make a 390Vdc low ripple supply.
Twin simple EMI common mode chokes result in almost no noise back to the mains too and with the PFC it's possible to limit current to a respectable normal amount.
Now I could tie all the frequency generators for the ICs together, but it's occurred to me that actually staggering them or leaving them to their own devices may help to spread load at the cost of spread noise.
Still working on it, it's more complex than a simple linear supply, however I'm already starting to see there's almost repeated modular blocks that can offer enough flexibility to allow reuse.
First thing to say is how shocklingly low the power use is. After the initial energisation of the caps and inductors, the resulting mains power use is exceptionally low. (Sure 200kHz would result in higher switching losses so having a low capacitance high speed mosfet and diodes with low reverse losses are key to reducing this)
The noise is way above the audio band and the resulting power is lower in noise than I was expecting - partly due to the Cuk's use of capacitors that results in a smoother output. However also SMPS are very transparent - if there's 100Hz noise from the initial bridge then you'll see it unless you essentially make a 390Vdc low ripple supply.
Twin simple EMI common mode chokes result in almost no noise back to the mains too and with the PFC it's possible to limit current to a respectable normal amount.
Now I could tie all the frequency generators for the ICs together, but it's occurred to me that actually staggering them or leaving them to their own devices may help to spread load at the cost of spread noise.
Still working on it, it's more complex than a simple linear supply, however I'm already starting to see there's almost repeated modular blocks that can offer enough flexibility to allow reuse.
Last edited:
https://authors.library.caltech.edu/63101/1/00017970.pdf
A very interesting paper on step down cuk with voltage dividers.
A very interesting paper on step down cuk with voltage dividers.
Just exploring GANFET such as the TP90H050: High Voltage GaN Design Resources - Transphorm
They have a buck/boost eval board (half bridge) 600V 2.4KW in a small footprint. Note the board needs a controller and secondly with the 900V device that's even more interesting.
I'll model up something as a half bridge and compare against a 390V->12.6V flyback model. Possibly look at a half bridge cuk too.
Running high frequency is great but the switching losses at 200+kHz are far higher.
They have a buck/boost eval board (half bridge) 600V 2.4KW in a small footprint. Note the board needs a controller and secondly with the 900V device that's even more interesting.
I'll model up something as a half bridge and compare against a 390V->12.6V flyback model. Possibly look at a half bridge cuk too.
Running high frequency is great but the switching losses at 200+kHz are far higher.