Advice for a choke for blocking parasitic switching spikes

[artis]

I am looking how to minimize switching interference on an smps that is powering an AB class amplifier.
Measuring at the amplifier output there is about 0.8 to 1v parasitic high frequency oscillation, then measuring on the DC supply rails there is also a parasitic ripple around 0.8v amplitude, I cannot read the waveform exactly (my older scope can't focus on very spikey waveforms) but they appear to be small very "sharp" spikes seemingly at or near the switching frequency of the smps (around 50 Khz)
I already have ceramics parallel to secondary filter caps , also snubbers across the secondary DC output.I tried wrapping the DC output wires that supply the amp around a ferrite choke, toroidal ferrite two turns per core. This seems to remedy the situation a little by "rounding" the spikes into a bit lower amplitude sine-like oscillations at the amp output, audibly it help by almost eliminating the sharp crackling noise but still the scope shows an ugly output.

I wonder how to design a choke that can deal with such spikes as best as possible?
Any advice is welcome.
thanks.

 

[berkeman]

It would be good to see a schematic and a drawing/pics of how it all goes together. There are lots and lots of considerations for dealing with noise in SMPS systems.

That said, is this an audio amp? What power levels are you working with? It is often a good idea to follow the SMPS with linear regulators to smooth the output power supply rails. This is common in audio and RF systems, for example. You can use low-dropout linear regulators to minimize the voltage headroom you need between the SMPS outputs and the power rails of your amplifier...

 

[Rive]

"sharp" spikes seemingly at or near the switching frequency of the smps (around 50 Khz)

50kHz is a bit low. Could you please tell us the type of the control IC of the SMPS circuit?
Or, if it is a complete unit, then the type of the unit? If it is something DIY style, then a schematics or some photos?

 

[artis]

It's a diy thing. the secondary smoothing capacitances are larger than in the schematic as I was kind with capacitors.
I saw the schematic online, originally it had optocoupler for output regulation but it didn't work too well and made the frequency unstable so I disabled that, for a moderate load works fine.

yes its an audio amp, not exactly a small power headphone one rather a 2x500 watt class AB topology, so on large speakers can't hear the parasitic oscillations but still the ugly image from my scope sits in my head.

This is more of a learning thing for me than trying to get audio perfection as I have other amps.
I was just wondering could I maybe make those chokes somehow function better than they do, maybe I need to add a short circuited secondary on the chokes with a small capacitor in series so for lower frequencies it would seem as open circuit and for higher frequencies which are essentially my spikes it would seem like a resistance?

 

[berkeman]

the secondary smoothing capacitances are larger than in the schematic as I was kind with capacitors.

Did you use low ESR smoothing caps? Just using larger capacitors does not necessarily reduce high-frequency spike noise.

originally it had optocoupler for output regulation but it didn't work too well and made the frequency unstable so I disabled that, for a moderate load works fine.

So it is operating without feedback now?

I saw the schematic online,

Where did you get the transformer? There are transformer versions for SMPS that help to reduce noise coupling from primary to secondary significantly.

 

[Rive]

It's a diy thing.

Well, that's nice. May we get some pictures?

 

[berkeman]

Yeah, good point @Rive -- @artis how did you wire this up? Is it on a carefully laid out 4-layer PCB, or on a hand-wired breadboard?

 

[dlgoff]

I am looking how to minimize switching interference ...

Sounds like switch contact bounce. A simple fix might be going to the source, the switch. You might try putting a 0.1 μf capacitor across the switch's contact terminals.

 

[Baluncore]

@artis. Your smps has two capacitors in parallel for each of the output rails. You should place an inductor between those two capacitors, to make a Pi filter in each rail.

You may benefit by placing some low ESR ceramic or polyester capacitors across the first capacitor in each Pi filter. That will reduce the highest frequency components that are otherwise not reduced due to the high ESR of electrolytic capacitors.

 

[artis]

It is indeed working without feedback, but when I did a stress test it dropped only a few volts when loaded with 1kW of constant dummy load so that's fine for me.
I have 12 1000uF/160v nippon chemicon caps at the secondary, 6 for each +- rail, I decided to split up the secondary in two for each of the amplifier channels. The caps themselves well as good as one can get an electrolytic for a decent price tag what else can I say about them.
pcb is a single sided through hole mount one, tried to keep switching components close and traces short but for the first design it was a bit trial and error so some of the layout could have certainly been better. Overall not the prettiest looking thing.

I see your advice about the Pi filter, it's a pitty I did not think about this in the first place because there is very little space there and in order for me to rewire the pcb I would have to cut traces between the caps which is not the biggest problem , I've done it before experimenting , space might be a problem.Dlgoff is also right surely every parasitic in an smps comes from the swithching devices and associated circuitry, you mean putting a spike suppressing cap across where exactly? The drain and source of each Igbt?
Yes by the way I'm using IGBT's because for the same pricetag they gave me much more power reserve. (higher specs)
Sadly I don't have a good enough scope otherwise I could probe the primary waveform which would give me more insight, now I can just see the secondary, but I think it's good enough for this example.

 

[Baluncore]

The caps themselves well as good as one can get an electrolytic for a decent price tag what else can I say about them.

You can be kind to an electrolytic by paralleling it with lower ESR capacitors such as ceramic or polyester. If you reduce high frequency currents in an electrolytic cap you reduce resistive losses, lower the temperature and lengthen the life.

 

[Rive]

Overall not the prettiest looking thing.

That does not matter, we do not ask for pictures to have our daily eyecandy
Both the transformer and the layout (external wiring included) can be the source of lots of problems. Right now we are working blind. Even those miracle teledoctors gets more than that to start with before providing their blessings...

by the way I'm using IGBT's

You mean, instead of the power MOS switches in the primary? Can you please give us a type?
Generally, it is not advised to change the key component in the circuit just because the new one has higher numbers...

 

[artis]

Well I looked at the specs before replacing them , yes the half bridge switches are IGBT instead of Mosfet. Every parameter was basically better except maybe maximum frequency but these new IGBT's have a freq. of 1Mhz and so more so driving at 50 khz should be no issue, also the gate charge was the same if not lower. I do not recall the model of the devices and in the smps they are covered by a metal push bracket that keeps them attached to the heatsink so I can't see.

I will see what I can do with a photo, well all in all there is one definite shortcoming because I had to change the original layout the larger ETD59 core and bobbin would not fit nicely so I had to make shot wires from the primary to the switching transistor traces.By the way I wonder , how can it be that these low capacitance polyester, styrene or propylene or ceramic caps have lower ESR than large , large surface area and capacitance electrolytics? Is it all about the dielectric or the fact that electrolyte introduces additional resistance even though it helps to fill the pores and increase the capacitance?

 

[Baluncore]

Every parameter was basically better except maybe maximum frequency but these new IGBT's have a freq. of 1Mhz and so more so driving at 50 khz should be no issue, also the gate charge was the same if not lower.

IGBTs work well with high-voltage low-current switching while MOSFETs are better for low-voltage, high-current. That is because the IGBT has a higher junction voltage drop than does a MOSFET while conducting.

By the way I wonder , how can it be that these low capacitance polyester, styrene or propylene or ceramic caps have lower ESR than large , large surface area and capacitance electrolytics?

It is because the electrolytic capacitor is made from two strips of metallised film, rolled together. External terminals connect to the film in one place only. Lower profile caps use longer narrow strips rolled together that have high effective metal resistance.

Ceramic caps are made as a laminated stack with alternate laminae connected in parallel to the terminal electrodes. There are high ripple-current "pulse capacitors" available that have very low inductance and resistance with polymer insulation.

 

[artis]

I would love to hear @dlgoff's explanation of where to put the cap or some other more finalized ideas?
Maybe you all got upset for me not providing my pictures which if they are so necessary I can do, although you can take my words at their value when I said the whole smps is a bit crude, rude and unprofessional.
basically I would be happy to find a way to squeeze out the best results from this one without major modifications because I feel that this being my first fully functioning high power smps is sort of a done deal and I wouldn't like to go over it once more, since I have done it already so many times.

the large caps are mains filter caps, the vertical pcb with the Ic's on it is the gate driver module, the wires attached to the IGBT's are gate drive wires the other power leads are soldered to pcb traces on the main pcbthose traces then go to the smps traffo, seen in the pic.then there are bridge rectifiers after the traffo and 12 capacitors for smoothing, after that the DC is fed through fuses (seen on top) to the amplifier boards.
Tried to keep the switching traces short but in the future will have to still improve on that.

 

[Baluncore]

I would experiment with a Pi filter between the smps as it is and the amplifier. The Pi input and output would have ceramic and polyester capacitors, the Pi output also needs one electrolytic. The Pi filter would use toroidal inductors. Such a circuit might connect to the existing fuse holders, and have duplicate fuses. Only the Pi filter ground connection needs to be arranged. If it improves things sufficiently you can try removing some of the extra electrolytics from the SMPS output and replace them with the Pi filter components.

 

[artis]

hmm , typically i see this Pi filter being implemented as part of the smps output stage , so it comes in between the filter caps , are you saying I could achieve the same result putting the filter on the already filtered DC "rails" that go to the amp boards?
Technically it should because the parasitic waveform in any case is AC and a filter should work with AC anyhow, right?

if this would turn out to be successful it would help me to know how to make better smps circuitry in the future. after all this is exactly the reason why many "audio people" fear smps as a power supply more than the devil fears the cross... :D
although modern well made smps are I think superior, at least from my own experiments the audio seems more dynamic , probably due to the fact that on higher power levels the supply voltage sags much less on demand than it would with a linear supply, but this is beyond the thread.

 

[dlgoff]

I would love to hear @dlgoff's explanation of where to put the cap ...

Here's a circuit image compliments of stackexchange.com. C1 connected to SW2 terminals. Only 0.1μf instead of 10 μf.

 

[artis]

in my case as it's a half bridge I would have to put the cap across each of the two switches ?

I wonder how does the cap reduce the switching spikes? because when the switch is closed it seems it is charged by the potential across the switch contacts then as the switch opens the cap gets discharged through the switch and as the switch closes (presumably causing the spike) the spike is absorbed as the charging current of the capacitor?

 

[Baluncore]

I wonder how does the cap reduce the switching spikes?

I see no way that a cap across a switch will reduce supply output noise. If anything it will increase switch power due to discharge of cap when switch closes.

 

[Rive]

Adding snubbers to switches and (bridge) diodes is a common practice when it is about excess EMI (after the layout-cleanup is already done). It is just that for this device it would be extremely hard to pinpoint the actual source of the spikes, since everything has coupling with anything else.

What I would do is to 'flatten' the design first, then look around with a two channel scope - one channel for the noise, as reference: the other is a few turn small coil as a 'sensor'.

I have to admit that I have no idea what I would do with limited equipment.

 

[Baluncore]

Snubbers are used to protect semiconductor switch components, they are NOT used to reduce switch fundamental output noise. RC snubbers across a switch limit dv/dt to reduce false triggering, while series L reduces di/dt to prevent hot-spots forming that can punch through a PN junction.

SMPS efficiency comes from fast clean switching, where the voltage appears across the inductor, not the switch. The output switch component is selected for lowest capacitance. The last thing you need to do with a SMPS is increase the switching time by adding a parallel capacitor. That is because the switch should always be fully ON with zero voltage, or fully OFF with zero current. Power is dissipated in the switch during slow transitions when the product of switch voltage and current can be significant.

The title of this thread refers to parasitics, but parasitics are almost certainly not the cause of the reported fundamental output noise. A MOSFET gate can have parasitic oscillations if driven by a low impedance gate driver without a gate series resistor. Those parasitics are at a much higher frequency than the SMPS switching fundamental. They tend to lengthen the switch transition time which can rapidly heat the switch.

Fundamental output noise needs to be removed by a Pi filter stage at the output. That filter needs input capacitors with very low ESR. The Pi input capacitor ripple current flows through the series resistance of the input capacitor, alternating between recharge and discharge at the SMPS fundamental frequency. That alternation doubles the AC ripple voltage on a simple SMPS output, and leads to a lower noise Pi filter solution that removes switching noise.

 

[artis]

Well @Baluncore explanation sounds reasonable to me that an extra capacitor will be an extra current sink to charge and then more current to discharge as the switch opens.

When still in assembly I probed the gate drives of my gate driver board and they had good looking square waveforms although IIRC there were some minor spikes at the closure of the switch but they were some 2 volts in amplitude as since that is insufficient to conduct a mosfet/IGBT I left them there as removing them seemed complicated.
the switches themselves run rather cool even under heavier loads which indicates to me the switching is going on properly and the switch isn't half open or half closed for prolonged periods which would cause more heating in the heatsink.

@Rive well when the smps board was out of the box i measured the spikes and they were the same, thinking logically they don't come from the filter caps nor from the rectifier diodes so they come from the transformer and since the transformer doesn't switch anything they obviously come from the switches and the current interaction with the inductor/transformer I think.

Sound like the Pi filter might be worth a try and the most reasonable choice for me given remaking this smps would essentially mean building a new smps.

Got some good proven Pi filter schematics with part markings etc? @Baluncore, thanks.

 

[Baluncore]

In post #1 you describe narrow 1V spikes on the amp output and on the supply rails. That might be ground noise on the oscilloscope ground lead, so I would want to also look at the ground connections throughout the system to see if there is ground noise before installing more supply filters. Those filters will need a reliable ground reference so you might need to filter the ground connection, or remove duplicate ground connections that cross-connect the amplifier to the SMPS.

All grounds are not the same. The ground network topology is important. Arrange for the switching currents in the SMPS to flow in minimum area return loops. Take the output and the ground connection from the loops at the quietest differential point. Use twisted pairs for longer connections to eliminate EMI.

The narrow spikes reported have only high frequency energy components so toroidal cores followed by ceramic caps should block the spikes. Since the power supplies are expected to deliver high peak DC current you will need to consider and avoid saturation of the chokes at peak loads. You might wind both polarity chokes, bifilar on the same core, but I would first question if the amplifier supply currents track or if significant supply return current flows through a common ground connections.

 

[artis]

there is only one ground wire (4mm) coming from the smps to the speaker output , there at the output two additional smaller ground wires go to each amplifier board, a star type connection. that is all the ground there is.

the spikes are not due to scope or other ground issues because the spikes were sharper at the smps output and also amplifier output then after putting ferrite chokes and simply wrapping each DC rail some two turns around the choke now the spikes are sharp as they were at the smps DC output but bit lower in amplitude and rounded at the amplifier output, at the amp output after choke installation they now look more like high frequency sines with a sharper rise and fall than a classical sine.
Also i have tested the amp boards years ago with a mains transformer supply and there was no oscillation or unwanted sound due to any sort of oscillation, my conclusion based on what I know is that it is most likely the smps output contains these spikes and the amplifier being class AB with sufficient bias current simply let's these spikes through in the speaker output.

so ideally I think I should eliminate these spikes and make the DC supply as clean as possible.as for the peak loads, from what I know the highest power consumption of an audio amplifier is at the lower side of the audio frequency range (say 20-200Hz) so what do you think, would this really interfere with the Pi filter given the filter only is "working" with frequencies that are above the audio limit and higher?

 

[Rive]

Snubbers are used to protect semiconductor switch components, they are NOT used to reduce switch fundamental output noise.

I wrote about excess EMI. And in this usage their role is exactly the limiting of the 'sharpness' of the switching, since that is usually one significant source of high frequency noise.

@artis :
- I would first try to find others who built this circuit and ask for reference. Right now nobody here (you included) has any idea if the measured spike is 'normal' for this circuit or not.
- I would also try to make a guess about the energy contained in that spike. I could not find the exact output voltage/current of your DCDC here, so I have nothing to compare that 0.8V to. At what current and voltage it is 0.8V? What's the value when it is running at full power? All that filtering might be just useless against a spike which has significant energy in it.
- I would try to google up some design guides for this kind of 'float' based inverters, where the coil actually has a cap in series (!) to a middle of a cap bank. I have a feeling that the frequency and the size of those caps has something to do with the inductivity of the transformer coils: and that is exactly one parameter you have just modified in your build with picking a bigger core.

 

[essenmein]

I am looking how to minimize switching interference on an smps that is powering an AB class amplifier.
Measuring at the amplifier output there is about 0.8 to 1v parasitic high frequency oscillation, then measuring on the DC supply rails there is also a parasitic ripple around 0.8v amplitude, I cannot read the waveform exactly (my older scope can't focus on very spikey waveforms) but they appear to be small very "sharp" spikes seemingly at or near the switching frequency of the smps (around 50 Khz)
I already have ceramics parallel to secondary filter caps , also snubbers across the secondary DC output.I tried wrapping the DC output wires that supply the amp around a ferrite choke, toroidal ferrite two turns per core. This seems to remedy the situation a little by "rounding" the spikes into a bit lower amplitude sine-like oscillations at the amp output, audibly it help by almost eliminating the sharp crackling noise but still the scope shows an ugly output.

I wonder how to design a choke that can deal with such spikes as best as possible?
Any advice is welcome.
thanks.

How much experience do you have with SMPS design?

You changed from FET to IGBT, did you adjust or consider dead time? (IGBT's do not switch like FET's)

You have a 1uF 400V blocking cap, this is not necessary with the half bridge topology, your split supply capacitors already do this for you.

Looking at your build I would bet you are dealing with common mode noise, your IGBT collectors are connected to the heat sink and I don't see a shield layer + Y caps to to provide a current return path back to 400V ground for the currents that want to flow through those capacitance's during switching edges.

Coils won't help, you'll need to shields and Y caps to start with.

Also if your scope cannot resolve the high frequency noise then you are basically working blind.

No regulation is fine, I assume you are running at or near 50% duty so you basically have a fixed ratio DC-DC and the output will sag a little with load, totally fine for audio.

However there appears to be zero protection in the SMPS section, short of some quick bench tests I would not use this as a long term power supply, esp not unsupervised!

Also FYI under no circumstances is it ever a good idea to put any capacitors in parallel with a switching device like IGBT or FET, unless you really know what you are doing (resonant supplies). Snubbers yes but this is only to deal with a specific problem, and generally in the nF not uF! and they need to be tuned, which you'll find very difficult if you cannot see the over voltage spikes clearly.

 

[artis]

well the voltage spike is about 1v max in amplitude, the DC output voltage is +-85 volts, or 170 VDC.
The IGBT's are shielded from the heatsink electrically if that is what you mean by insulators.
I have watched the gate drive waveform , there is enough deadtime, the switches aren't overlapping and if they would they would have been long destroyed which they are not.

what protection are you looking for? the gate drive IC monitors basic functions like overvoltage etc, the secondary has galvanic isolation due to the transformer and the outputs have fuses , what else do I need.

 

[essenmein]

well the voltage spike is about 1v max in amplitude, the DC output voltage is +-85 volts, or 170 VDC.
The IGBT's are shielded from the heatsink electrically if that is what you mean by insulators.
I have watched the gate drive waveform , there is enough deadtime, the switches aren't overlapping and if they would they would have been long destroyed which they are not.

what protection are you looking for? the gate drive IC monitors basic functions like overvoltage etc, the secondary has galvanic isolation due to the transformer and the outputs have fuses , what else do I need.

Re IGBT, the back of the device is the collector, this tab has a decent area and in conjunction with the electric insulation, creates a collector to chassis capacitance. Since the low side collector is flapping up and down at switching freq, this capacitance creates a common mode noise issue. One of the ways of handling this is put a metal layer in there, ie:

Collector tab
insulation
metal shield layer
insulation
heatsink

The shield is connected to the converter power ground, ie the IGBT emitter in this case, this provides a low imp return for this.

Re protections that are missing that I'd add before leaving it on while not watching it:

- thermal
- over current

 

[artis]

well over current is basically fuse , and thermal is at the AC mains side , I have a thermal safety switch which opens the incoming phase if the heatsink reaches over 90 degrees C. By the way I ahve tested it for a while and it seems pretty reliable, if anything heat is not an issue the aluminum heatsink I provided has plenty of reserve in it and is attached to the whole chassis which I also made from aluminum as to serve as one giant heatsink.

Ok I understand your point about the IGBT issue although from what I have measured this seems like a switching issue that runs through the wires instead of being induced into the wires from outside because putting other wires around the smps doesn't produce similar effects, I once even tried a few turn wire connected to scope and there was nothing similar.

 

[artis]

@Baluncore I would be interested in some good Pi filter schematics that I'm sure you know of that would best suit my application here given the 50 Khz frequency and DC voltage that I gave which is around 85-0-85v DC a split supply.
thanks

 

[jim hardy]

Layout of smps circuit is important.
One must minimize the area encircled by his high frequency switched current.
He accomplishes that by running the conductors close to one another
else it'll radiate EMI at switching frequency that gets into everything nearby.

That's why pictures of your device would help.

Mysef i'd spend the money for some giant toroidal line frequency transformers and use a linear supply.

old jim

 

[Baluncore]

I would be interested in some good Pi filter schematics that I'm sure you know of that would best suit my application here given the 50 Khz frequency and DC voltage that I gave which is around 85-0-85v DC a split supply.

Get a copy of LTspice. C1 and C2 are your existing supply reservoir capacitors.
L1, L2 and C3 are low pass filter. C4 and L3 make a trap, tuned to 50 kHz.

 

[artis]

Thanks, I'll try this one out, first have to get hold of the inductors and capacitors,
the psu filter caps are electrolytics but the filter caps should be bipolar , I wonder which ones better, polypropylene, polyester or other or there is no difference?
Probably a lower ESR is better?

PS. I have some ferrite rings I guess I'll have to find a calculator to know how many turns I should wind to get 10mH as well as 3.18uH. problem is those rings have no markings so I don't know the ferrite material specifically

 

[Baluncore]

first have to get hold of the inductors and capacitors,

You missed my first point. FIRST get a copy of LTspice.

There are simple modifications to that circuit that increase rejection. You cannot quickly test sensitivity to value in hardware. What if you split C3 into two 0u1F ceramic caps and ground the centre? What if you change the trap to 31u8F with 318nH, then how does the Q of the notch change, for 50kHz, but also for harmonics. What if you wind L1 and L2 on the same core to cancel the supply current induced field?

the psu filter caps are electrolytics but the filter caps should be bipolar.

Why? They are polarised by the power supply voltage. What do you actually mean by filter caps. You must be more specific.

I wonder which ones better, polypropylene, polyester or other or there is no difference?Probably a lower ESR is better?

You can test ESR and capacitor loss with spice. You waste time asking silly questions when spice can answer questions quickly, and put real numbers on the answers. There is no “better” unless you specify a numerical requirement.

PS. I have some ferrite rings I guess I'll have to find a calculator to know how many turns I should wind to get 10mH as well as 3.18uH. problem is those rings have no markings so I don't know the ferrite material specifically

Why waste time on unknown parts. Design a solution and specify a core, or measure the core and model it in the circuit with spice. When you build the inductor, if the core gets hotter than the wire then you have the wrong core material. If the wire gets hotter than the core then use thicker wire with a more expensive core. If it all gets too hot, you don't know what you are doing because you have not modeled the circuit. If it stays cold you have invested too much buying big cores and winding them with over-thick wire.

Either you engineer minimum circuits that work, or you cut and paste circuit blocks without understanding the costs and implications.

 

[artis]

Now I had a few months away for other things that I'm still doing but I started learning LT spice recently so here is my own version of the filter @Baluncore posted here, so may I ask what the author himself thinks of it?
I played around with the values and did an AC analysis of 1volt amplitude from 1Khz to 70Khz.

 

[Baluncore]

First the praise; well done getting into LTspice.

Now the constructive criticism.

You increased C3 from 1uF to 550uF. The advantage of the 1uF cap was that it could be ceramic with much lower inductance and ESR than an electrolytic. C3 was needed to attenuate ultrasonic spikes with HF harmonics. Only the 50 kHz fundamental remained a problem for the L3 & C4 trap.

You defeated the series L3&C4 trap that was tuned in the circuit to the 50 kHz fundamental. If L3&C4 are not tuned correctly you may as well leave those components out of the circuit.

Making components in circuits bigger because they appear to work better is bad practice. Components only have to work well enough to meet the circuit specs. Beyond that, they cost space and money.

Spice can lead you to think all components are perfect. For example electrolytics are also resistive, you must think in parallel about component value, availability and cost.

 

[artis]

thanks for answering, yes I'm getting into spice slowly at first it seemed a strange program the user interface is meant for learners more than those who seek ready plug n play solutions I guess.

Well I do agree with your criticism and about the component choices, this was not meant as the final version of what I would do but more of a playing around way of learning spice.

So here are a few questions, leaving out the cost and space of bigger/overdesigned parts, why is the larger capacitance in C3 worse than the smaller capacitance apart from that ceramics would have lower ESR? I assume that the goal is to get the noise line as low as possible and for a wider frequency range? Also I don't know for sure whether the switching frequency is perfectly 50khz , depending on load etc it can vary a little I think.
Well just physics wise would i be still better off using that low capacitance ceramic in c3 or can I put multiple smaller electrolytics in parallel? The thing is in SPICE simulation it seems the larger capacitance is better , well maybe I should write in the ESR on cap that I'm using and see what changes, since spice allows such fine tuning.L3 and C4 should be left as is in your original schematic ?

 

[Baluncore]

why is the larger capacitance in C3 worse than the smaller capacitance apart from that ceramics would have lower ESR?

Because the fundamental is at 50 kHz with harmonics above that. It is unnecessary to use more capacitance since there is no noise below 50 kHz. Electrolytics are not much use above the audio range.

Well just physics wise would i be still better off using that low capacitance ceramic in c3 or can I put multiple smaller electrolytics in parallel?

Why waste time and money installing components that will not perform?
Simulate it with LTspice, with realistic ESR specified for the capacitors.

 

[artis]

Ok, I understand your points, let me take a few days to get used to spice more, i will try to find some relevant ESR data for the various caps I have and see what changes what.

 

[Baluncore]

Avoid generating noise you will have to attenuate later.
Where does the noise you are trying to attenuate come from?
Go back to the start and calculate the values again.

The current at full power is 1 kW / 170V = 5.9 A.
At 50 kHz the reservoir caps are charged every 20usec.
Allow maximum reservoir ripple at full power to be 100 mV.
C = i * dt / dv = 5.9 * 20u / 0.1 = 1180 uF.

Power supply wiring that passes reservoir caps has current flow towards load.
But reservoir ESR drops opposite voltages during the charge and discharge cycles.

Assume that the recharge pulse lasts 20% of the cycle 20 usec = 4 usec.
During charge, current i, flows to the load, while 4 * i flows to the reservoir capacitor.
That drops a positive voltage across the ESR, Rs. V = +Rs * i * 4.

During discharge, current i flows from reservoir to load.
That drops a negative voltage across Rs; V = –Rs * i.

Peak-peak ripple voltage due to ESR is therefore V = Rs * i * 5;
Which will be 100 mV when ESR = Rs = dV / ( i * 5) = 0.1 / ( 5.9 * 5 ) = 3.4 milliohm.
When added to the reservoir ripple voltage that gives 0.2 V at ESR = 3m4R.

The problem starts with capacitors in series, they have half the capacitance and double the ESR of a single capacitor.

Electrolytic caps dry out which reduces capacitance when running hot due to ESR * ripple current squared. Poly and ceramic have a longer lifetime. Electrolytic capacitor construction also has higher inductance that allows the start of the charge cycle to produce a significant voltage spike. Parallel ceramic and poly pulse reservoir capacitors can reduce it to a bandwidth that can be handled with the electrolytic cap series inductance.

Look at the ESR and series inductance of different types of cap construction.
Use spice to generate a 4 usec pulse at 50 kHz, so you can study the best combination for your charge reservoir. Start with Vsource as a PULSE(0 170 1u 1u 1u 3u 20u). Then steepen the edges.

 

[artis]

as for my smps secondary filter/reservoir caps see the attache image, there are 12 caps in total each cap is 160v/1000uF.

as for the spice filter, I have some wima mks4 0.1/630 on hand but I tried googling etc and can't find ESR data , even in the original datasheet.

 

[Baluncore]

see the attache image

Where?

 

[artis]

ups sorry forgot to attach.

 

[Baluncore]

Firstly, avoid having the reservoir caps on a separate wiring spur. Run DC power and ground rails from rectifier to load, via the reservoir terminals.

Electrolytic capacitor series inductance tends to be proportional to capacitance. Lower the capacitance value until reservoir-ripple ramp-amplitude at full power is reduced to 100 mV, no further. The capacitor self-inductance spike, and ESR voltage step will appear added to that ramp. Too much capacitance will increase both the dominant inductive spike and ESR step.

 

[artis]

My psu layout is already such that the caps are soldered in the rails which are thick and run from traffo secondary to output where wires are soldered that go to amplifier boards.

So if I got right , what you are saying is that too much capacitance on the secondary side instead of filtering out the spikes and ripples to smooth DC tend to add itself a parasitic spike which is the result of the capacitor inductance working together with the switching spikes from the transformer output which is the capacitor input?PS. can you or anyone for that matter give me any reference for the ESR value of typical 1n to 100n polyester capacitor? I feel weird as I am searching parts suppliers and various brands but seemingly none offer this value ?

 

[Baluncore]

So if I got right , what you are saying is that too much capacitance on the secondary side instead of filtering out the spikes and ripples to smooth DC tend to add itself a parasitic spike which is the result of the capacitor inductance working together with the switching spikes from the transformer output which is the capacitor input?

If you have more reservoir capacitance than needed, then you also have more inductance than necessary. The terminal and foil inductance of electrolytic caps can be a big problem with switching supplies as it produces a voltage spike at the start of conduction. I am saying use two small caps in parallel rather than one larger one.

I feel weird as I am searching parts suppliers and various brands but seemingly none offer this value ?

Search for "pulse capacitor".
Pulse caps are used as snubbers in power switching applications, but they can also be used in parallel with electrolytics to reduce the initial inductive spike. You will not usually find ESR quoted for pulse caps, only for electrolytics.

[edit]
Here is a data sheet.
https://au.mouser.com/datasheet/2/88/PPB_series-553044.pdf
Notice that dV/usec is specified for different capacitor values.
From that you can calculate pulse current from; C = Q / V; i = C * V / t;

 

[artis]

after searching in my parts catalog it seems that these polypropylene film capacitor for ac pulse are rather expensive, the values I would need for the filter which are 1uF and 3 or 3.3 uF run about 10 usd a piece.

So I am also considering the chokes, now would you advise me a little on what current ratings I should look for in the chokes, in the series LC filter part of the 3 or 3.3uH choke (depending on which one I will be able to get in my store) I assume I would do fine with mA of rated capacity but for the main chokes on the DC rails I would assume I need to calculate my maximum operating current of the amplifier at full load and then choose accordingly correct? Otherwise the filter will choke the amp at or near full power operation?

 

[Baluncore]

The power rail chokes will saturate if not rated for peak DC supply current. Any notch filter choke must be rated for ripple current. You can measure that by plotting current in the time domain simulation with LTspice. Do you really need a trap? Does optimising electrolytic selection not simply resolve the problem?

You should design and build chokes optimised sufficiently for the power of the supply. That is where understanding and experimentation comes in. Streaming words will not resolve your fears. Your fear may be real, or procrastination, but if you don't experiment you will never find a pragmatic solution that actually works.

For a production design, optimisation for minimum construction cost is important. For a one-off, cost is not so important and you can trade cost for design time.

... it seems that these polypropylene film capacitor for ac pulse are rather expensive, ...

Pulse capacitors cost money because they work. If you design and select reservoir components for minimum inductance you will probably not need pulse caps. You might also design your amplifier to better reject power supply noise. If the inductive spike is too big, because you can hear it, then do something about it. Buying a more expensive pulse cap is the penalty for lack of flexibility and experimentation in reservoir design and layout. Sticking with the cheap electrolytics you first obtained may result in the most expensive solution.

 

[artis]

I agree with you , although I must say my smps secondary side reservoir caps were not exactly cheap, I bought some rather expensive low ESR nippon-chemicon or nichinon forgot which ones.
I have done some experimentation in the past with the samps to find out the problems, I have also probed my secondary side and I am afraid that the capacitors are not the whole blame here, there is some small ringing present in the primary side of the switches so I think it all comes together to form this spike behavior in the secondary.
This was my first large power smps so for me it would be much cheaper to simply build a small choke filter for it than to redesign all of the smps. I also have some other projects going on where I'm involved so time is also a constraint.

On the flip side I have never made a filter so why not , I see this as an opportunity to learn more and see in real time how a spice model performs in a known real life situation where I know the problem.
I am also thankful to you @Baluncore for giving me advice here.

Ok , so just as I thought the DC rail chokes should be rated for my full work current and with some reserve , how about the smaller series LC choke, given it's a series LC the current through it will only be as large as the AC parasitic waveform riding on top of the DC rails so I assume a small current, but I guess I should go to spice to see what the value could be.