I want to make a SMPS of around 2400 to 1800 W that is lower noise and EMI and more efficient than the usual.
I started a "3 phase" thread but 3 phase is not widely available and that limited the interest.
Since my preferred option is to build the three phase system out of a series of one phase modules anyway, I decided to start this thread dedicated to just the standard one phase SMPS component, usable by anybody.
For modules of around 800 W the basic default is usually a rectifier, boost PFC, bulk capacitor, then switcher for isolation and step down, then more rectification.
These days the switcher is often an LLC to enable it to soft switch.
That all works OK but is less efficient than possible because the multiple conversion steps all add losses.
A better option seems to be to combine the PFC and switcher, and maybe eliminate the separate input rectifiers too.
There's a lot of ways to do this and I have read many but can't find much of an overview or principles to choose between the options.
Some of the issues are
Anyone tried to build an advanced circuit with better efficiency and lower noise?
David
I started a "3 phase" thread but 3 phase is not widely available and that limited the interest.
Since my preferred option is to build the three phase system out of a series of one phase modules anyway, I decided to start this thread dedicated to just the standard one phase SMPS component, usable by anybody.
For modules of around 800 W the basic default is usually a rectifier, boost PFC, bulk capacitor, then switcher for isolation and step down, then more rectification.
These days the switcher is often an LLC to enable it to soft switch.
That all works OK but is less efficient than possible because the multiple conversion steps all add losses.
A better option seems to be to combine the PFC and switcher, and maybe eliminate the separate input rectifiers too.
There's a lot of ways to do this and I have read many but can't find much of an overview or principles to choose between the options.
Some of the issues are
- Overall efficiency
- Control
. Can a standard IC controller be used?
. Duty cycle versus variable frequency etc. - Control behaviour
. Buck has simple response, many others have RHP zero that complicates feedback - Stress on the transistors
. Some circuits stress the transistors more severely with peak current or over volt transients at turn off/on. - Ripple on the output.
. Cuk looks nice here - EMI coupled to the input
. Cuk also nice here - Soft-start
. Easier with some circuits - Circuit simplicity.
. Obviously prefer simplicity, what's the trade-off? - DCM/BCM/CCM
. Optimisation of current mode.
Anyone tried to build an advanced circuit with better efficiency and lower noise?
David
Last edited:
Cuk stresses the cap too.
You’ll probably find the switching for 1800W pushes you to full bridge. Also once you start trying to shift switching noise up the frequency ante then switching loss becomes a problem.
It’s possible todo a isolated cuk. Only issue for making your own smps is really the transformers/inductors.
You’ll probably find the switching for 1800W pushes you to full bridge. Also once you start trying to shift switching noise up the frequency ante then switching loss becomes a problem.
It’s possible todo a isolated cuk. Only issue for making your own smps is really the transformers/inductors.
Yes, but mains rated PP capacitors are readily available and have very low ESR so the losses are low.
And they are not expensive so I don't see any major issue here.
Each module is only ~800 W because there's one per phase (in my use).
Maybe neither full or half B.
I don't want to shift the noise, it seems better to keep the frequency lowish and losses down.
Some circuits vary frequency dependent on load, these need to stay out of the audible band.
But for duty cycle control I expect 50 kHz to 100 kHz is fine, pretty low by modern standards so I don't think switch losses will be a major issue.
Best wishes
David
One point you’re missing - DCM vs CCM. You’ll get more efficiency it’s DCM but harmonics/noise increases. CCM on the other hand is quieter but less efficient.
In the duty cycle/freq you’ll also want to decide if you’re altering frequency and also you’ll probably need to adjust duty cycle on the fly to minimise overshooting voltages. Lastly you’ll want to think about loop time - larger the output cap the slower the voltage feedback control loop.
In the duty cycle/freq you’ll also want to decide if you’re altering frequency and also you’ll probably need to adjust duty cycle on the fly to minimise overshooting voltages. Lastly you’ll want to think about loop time - larger the output cap the slower the voltage feedback control loop.
Yes, I have had a look at that, I should have had it on the list.
Boundary conduction mode is common with boost PFC, is simple to control.
Looks attractive for other applications too.
LLC is usually described in frequency domain, simple converters with variable duty cycle often in the time domain.
One of Cuk's proposal's is a mix of resonant fixed half sine On times and variable Off times, so both frequency and duty cycle vary but it seems easiest to think of it in the time domain.
Best wishes
David
You may want to look at 4 pin and paralleling. Even using WIMA DC link 80uF 4pin it helps reduce ripple putting a few 20uF 4pin in parallel.
LLC is hard to design for - you will need todo the maths based on knowing the output parameters. You’ll not get a one size fits all.
Lastly on the high power - it’s more than just putting in parallel, you will need to look at current balancing between the supplies. Meanwell’s medical supplies have an optional link between parallel supplies for this purpose.
I think the best path is start with the simplest single phase for a specific use that you will use it for. You’ll have a lot of pages of maths to work through so having specific target will help alot.
Well, the whole point is to eliminate DC link capacitors with a simpler, more efficient architecture.
And the LLC too😉
Cuk proposes a direct, one step PFC and transformer isolation > https://www.power-mag.com/pdf/issuearchive/46.pdf
At the moment I study the classic Cuk isolated converter.
The input inductor acts as an EMI filter and the output inductor filters noise out.
While the power transfer capacitor helps with soft start.
So it couldn't be much simpler and still retain functionality.
Once I have a feel for the classic I will have another look at his latest idea in the above link.
Main output is about ±100 V.
Also want lower volt subsidiary supplies and probably a floated Mosfet driver supply too but these can come latter, at least in priority.
In practice they will be extra taps on the transformer secondary so need to be finalised before I start.
There is some baseline load from the Class H amplifiers and Class B+ bias that probably actually helps with stability.
Best wishes
David
Last edited:
… keep it simple 🙂 personally I’d looked at a simple single output to start. Unless you’re using the secondary taps for sensing. You’ll have enough on your plate given the current spikes. You’ll want transient filtering from the mains (ie large spikes).
I certainly try to keep it simple conceptually and start simple with simulations.
But once I have the concept worked out then it shouldn't be too hard to add extra secondaries on the transformer.
It's similar to what Hypex do in their SMPS, main outputs, auxiliary low volt supplies and a floated FET driver supply.
I could, of course buy one of theirs but where's the fun in that?😉
Best wishes
David
Please note that ripple cancellation in a Cuk converter involves winding of transformers with predictable / repeatable leakage inductances, much nicer on paper than in reality.
Yes, but even without the ripple "cancellation trick" the Cuk still has the benefit of ripple reduction from the inductors in both the input and output side.
I see any cancellation as a bit of a bonus.
Best wishes
David
Any converter topology (including Cuk's) that first stores energy somewhere (say inductor) and later transfers it to the output (say capacitor) inherently tends to have these control issues related to the RHP zero. Sorry to have missed this in the earlier post.
Well... buck stores energy in the inductor then transfers it to the output capacitor but doesn't have an RHP zero.
So I don't know how to formulate the condition exactly.
But yes, it's quite common, I think Cuk even has a second RHPZ somewhere.
If PFC is included in a one step converter then there is probably a constraint that the loop time constant has to be more than mains period or else we mess up the PFC.
That is so much slower than the RHPZ limit that the RHPZ hardly matters
So I suspect it's not a major problem in this application, even a linear supply with no feedback at all works OK.
Just put it up there as a reminder and discussion point.
Best wishes
David
Dave Zan said:
Within every switching cycle of a boost / flyback converter, the capacitor must wait before it can be charged (to allow inductor charging), and it's this delay that manifests itself as an RHP zero in the complex frequency domain. Now, enter the Cuk converter, you have found yourself another pair (of inductor and capacitor) that needs to be taken care of.
However, the buck converter whose capacitor could be charged along with the inductor, doesn't suffer from this delay (or the RHP zero) altogether.
Last edited:
But do note that the presence of RHP zero(es) does not make a converter altogether useless. The SEPIC topology, very similar to Cuk's, has been successfully used in air conditioners.
My point was that inductors and capacitors do not inherently cause an RHP zero.
Plenty of LC filters that are minimum phase.
So we need to be careful of exactly what the conditions are that cause the the RHP zero.
For example, a Miller compensated amplifier usually has an RHPZ caused by feed forward thru the Miller capacitor, but it can be eliminated.
One of the Cuk converter's RHPZ appears to be inherent, I am not sure about the other one.
I have read some notes that imply it can be eliminated but without details of how.
Yes, I am not too concerned, the plan is a rail tracker for each individual Class H amplifier anyway.
I have done the LTSpice simulations for this and it is quite unaffected by power supply variations.
This is because the tracker is a buck down converter, chosen because it is minimum phase so no RHPZ and can be well controlled by fast feedback.
So my lack of recent posts was not due to the RHPZ issue.
Instead I have spent the time on the search for an efficient circuit.
The usual Totem Pole PFC circuit does not work well with the Cuk converter.
And there are common mode noise issues with some of the other attempts to remove the b. rectifier.
Manufacturers don't mention this sort of problem in their application notes, they just skip over inconvenient issues so it took some time to work it out for myself.
I have at least a basic circuit that looks OK, low noise and low loss but requires some duplicate parts (transfer capacitor and switch transistor).
It saves on rectifier diodes however so not really more parts.
I feel there should be a better way, exploit the isolation transformer and pull/push on the two ends.
I don't know if I haven't seen it because most PFC circuits work into a DC link capacitor and don't have this option, or just because it isn't possible.
Still more days to think before Covid lock-down eases up.
Best wishes
David
Last edited:
Dave Zan said:
They don't, but that also assumes that the filter is LTI, which is unfortunately not the case for many converters (buck is an exception), as their switches keep opening/closing (changing the circuit structure) numerous times in a second.
In fact, this is exactly the basis for all the non-linear (modern) control techniques that have been proposed for power converters. These methods also tend to be more robust, when compared to the linear ones. The following is the thesis of one of Cuk's (and Middlebrook's) students, on the sliding-mode control of power converters.
https://thesis.library.caltech.edu/3700/3/venkataramanan-r-1986.pdf
Last edited:
Surely Buck also opens and closes it's switch, so is no more LTI than any other switch mode power supply?
All the basic converters can be approximated with Middlebrook/Cuk's technique to use the State Space Ave. and treated as LTI.
So I don't believe the RHPZ comes from any failure of LTI, just as there are perfectly LTI systems that have RHPZ.
Thank you, even if I don't think it is needed here, I am still interested in control theory and always keen to learn more.
I am about to start on LTspice for my new (?) one step PFC and switcher/isolation circuit.
Will soon see if it my ideas actually have any truth in them...
Or if I have just re-invented Cuk's later work.
Best wishes
David
Last edited: