PWM PV Charger ... Need Help ... RFI / General good practice - fast switch

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
I have been mucking about with my solar setup, initally playing with buck modules in paralell but that is another thread...
Anyway I needed to put something in place that actually worked so decided to go with 'simple' pwm

I should have known better, simple.... Not

The PWM signal is from an arduono, no problem altering freequency or duty cycle.
I have 5 Fets in paralell, P-Type with an RDSOn of arround 0.06 Ohms.
The gate is level shifted with transistors to around S-0.4 / S-11.8, Off and On respectively.
Maximum current from the PV is about 32A right now althouh I am about to add another 600W , the system is 12V.

The gate drive works well, so far as I can tell and fets only require a modest heat sink to handle the heat dissipated.

To be honest I was chuffed when it all did what I had hoped but then I realised it was also causing my broadband to drop out...

I have tried freequencys between about 250Hz and 20kHz and the DSL drops out fairly consistantly.
The PWM duty cycle is limmited to between 10% and 90%, aproximatly

I suspect the issue is RFI as opposed to there being anything superimposed on the DC. The router is down streame of a boost buck combination which stabilises its 12V supply, allong with the satellite box and the phone. Neither converter is isolated so I suppose noise could conceevacly pass through on the ground line but I am struggling with the mechanisme.

Interestingly the DSL occasionally drops out when the genny powered charger is on... more PWM I suspect!

So the question is... How do I fix it without resorting to long slow pulses which is possible but hardly optimal.

I literally have no idea, apart from proxcimity, which isnt going to be easy to change.

Thoughts?

Cherrs,
Al
 

Papabravo

Joined Feb 24, 2006
21,159
The only way to deal with this problem, that I know of, is to hire an RF engineer, or develop the skills, and gather a small fortune to buy the test equipment you will need to become an RF engineer and solve the problem yourself. You might think this is a flippant answer, but this stuff is actually complicated enough that there are seldom any easy fixes. Especially since you have no control over the DSL equipment and we know it pushes high frequency signals over unshielded twisted pair which is notorious for EM susceptibility.

You might want to start by renting a Spectrum Analyzer for a month to figure out where the energy that disturbs the DSL is coming from exactly. Once you know that you can begin to develop an approach to mitigating the effects.
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
Mmmmmm.....
Not the reply I was hoping for but I take your point, and I have to say when it comes to understanding the issue, properly, I would have to agree, I think.

That said it is clear that the driver freequency isnt the problem so it must be harmonics which suggests ringing of some sort and I cant help thinking that there must be some basic things I could try... Surly not all devices get a complete RF design study, do they?
Perhaps I am missing the point but I was hoping to find some pointers with respect to what sort of solutions I might research.

Dont get me wrong I apriciate the input and your comments have certainly given me pause for thought, in fact since posting I have discovered a couple of comertial products that flok are complaining about...

I think I will hook up my scope to the phone line and have a looksee... It cant hurt.

I would be interested in some pointers with respect to what to measue and how... I have 4 channel storage scope, underutilised thus far but I am learning.

Anyway I am going to start with an earthed metal box, probably a food tin to start with.
I will also see if I can quantify the interfeeriace... Anyone know what DSL likes least and at what sort of amplitude?

Thanks,
Al
 

Lestraveled

Joined May 19, 2014
1,946
@Dyslexicbloke
Hey, I hear you. I had a UPS near my DSL modem and I was getting drop outs and my DSN was getting changed all the time. I read online the UPS can interfere with DSL modems and I removed the UPS. Huge difference, a DSN change once every few months now.
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
Mobving stuff... now thats my kind of solution.
I had actually thought of that and will try it but I was hoping to identify the speciffics of the problem because not fixing it will almost cirtainly see my struggling with other kit at some point.

The Router and batteries are in close proxcimity right now but bats charger and the like is due to take residence in a new shed ovr the next few weeks, perhaps sooner now.
I could also move the phone line, although its a bit of a pain but then so is no internet.

Al
 

Papabravo

Joined Feb 24, 2006
21,159
  1. Can you enclose everything inside a metal box, or inside a conductive screen -- as an experiment.
  2. Can you slow down the slew rate of your signals?
  3. Can you use shielded cables?
I'm grasping at straws here, and that is the best I can do unless you can tell me something I don't already know.
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
I doubt that, you not knowing I mean, I wasnt questioning the validity of your comments, they didnt sound like the sot of thing anyone would say without some experiance.

What I took from them was that the question was unlikely to have some quick fix all answer like decoupling a supply rail might, IE put 0.1uf caps close to everything and only worry it that doesnt work...
I was hoping that the answer would involve some sort of snubber, RC turned to my carrier or something simmilar but given that you havent said that, because it wont work I assume, I am left with only basic options, faraday cage / move the kit further apart...

I will hook up the scope, but because I am interested and may lean something not because I am u der any ilusuons with respect to my ability to engineer a solution.
I work with PLC,s so deal with analough signals a great deal and ocasionally noise on them, particularly where VFD,s are involved, 720V at 8kHz will muck with just about anything if you put it in the wrong place. A few years back we had one site with a small motor, about 8kW I think on a 350m cable and despite fitting the line reactors specked by the drove manufacturer it took out two seperate broardband lines in adjcent buildings every time it was switched on....

However we also have a 100kW turbine in a metal cabin allong with a standard bt line and router not 3 miles down the road and that has been solid for years... sam drive manufactufer same seriese drive same 8kHz carrier but the cable to the motor is only a few meters, all be only 1m from the BT master socket.

All that proves is that you are right on the money and despite having seen this stuff before I an none the wiser...

I just wasnt expecting to have issues with 26V, PV open circuit, @ 10A, in fact my target is 60A. I have had it up to 30, more panels to fit, but didnt notice the DSL issue untill late afternoon when the sun was giong down. I an not saying it wasnt ther just that I hadnt noticed as I was playing with the control loop.

Its all very irritating.... There arn't may time I wised I didnt live in the middle of mowhre but this is one.

There are so many PWM charge controllers out there, they cant all be noisy can they?

This one of those times whre finding I have overlooked something obvious and ending up looking silly would be a good thing.

Just looking acros the room, think corner in old barn - its a long story, and realising I have the buck converters, the ones that supply the router basicall next to it on the shelf.
These are cheap Chinese units which are essentally just the typical application circuit of the buck chip on a small board, no cells, not even a case. Since theae aee just PWM into an LC filter, essentally, whats the difference?

Might I be better with a classic buck circuit? would that be less noisy for any reason?

Given that I have 5 paralell FET's I could fit 5 seperate diodes and eitber 5 individual inductors or potentally wind one huge one...
I that kust flight of fancy or is it likely to be easier than 'simple' PWM to get right.?

Shielding cables ist practical... I mean I could probably put everything in trunking but my battery cables are 2X25mm in paralell to the main inverter terminals and 1X25mm to the DC bus bars.

Slow switching is probably the way to go but I was having problems at a few hundred Hz which suggests each pulse could be an issue, at least I thought it did.

I think I can go as low as 31Hz with the PWM library I am using and below that I could just use timers so there is plenty of scope to ply arround.
Not sure what the batterys will think or it though, ripple is supposed to be less than 5%!!

If anyone has an opinion on that feel free to wade in... Ore suggezt a completly different option :)

Sorry just realised you said slew rate not speed....Long day.
Yes I could but the switching losses asociated with that would be huge, probably huge enough to make it impractical. The FETs are disipating 100W now when they are on and I have done what I can to keep them as fast as possible... individual BJT totempoles on each gate, seperate darlington line driver stages to pull down a zenner feeding those. In fact the only thing I forgot was local decoupling caps which was planning to add on the supply rails next to the transistors and if anytbing they will speed things up.

Al
 
Last edited:

Papabravo

Joined Feb 24, 2006
21,159
I meant to say there are things I don't know because I can't see them, not because I have not studied or done similar things in the past. If I could see and poke around what you've got going something might come to me. Words are a poor substitute for being there, but they're all we have.

My advice is to go one step at a time to identify the source, and experiment with potential solutions. Trial and error is about all we have.
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
I know what you mean... at least I do now.
Looking at something is a whole different thing than trying to visualise it...

I find the same with work... Think all day get nowhere, watch the plant for 1/2 hr and oftrn the problem will jump out because you noticeome little think you never considdered might be connected.

Playing tomorow I think... Unfortunatly the forcast is poor so I will not be doing any high current testing.

Any thoughts with respect to using very slow pulses? From a battery perspective I mean as that is someting else I probably should know more about.

Al
 

Papabravo

Joined Feb 24, 2006
21,159
If you suspect harmonics from your PWM system, then the experiment is to leave the frequency of everything the same but slow down or shape the edges. For example if you have a period of 50 μsec, and rise/fall times of 1 μsec you will have a certain harmonic content. Now slow the edges down so that the rise/fall times are 5 μsec and the higher order harmonics will be substantially attenuated. This is why a spectrum analyzer is an essential tool for this kind of investigation. How else can you measure the effect of small changes? How do you know what done looks like.
 

ronv

Joined Nov 12, 2008
3,770
I have been mucking about with my solar setup, initally playing with buck modules in paralell but that is another thread...
Anyway I needed to put something in place that actually worked so decided to go with 'simple' pwm

I should have known better, simple.... Not

The PWM signal is from an arduono, no problem altering freequency or duty cycle.
I have 5 Fets in paralell, P-Type with an RDSOn of arround 0.06 Ohms.
The gate is level shifted with transistors to around S-0.4 / S-11.8, Off and On respectively.
Maximum current from the PV is about 32A right now althouh I am about to add another 600W , the system is 12V.

The gate drive works well, so far as I can tell and fets only require a modest heat sink to handle the heat dissipated.

To be honest I was chuffed when it all did what I had hoped but then I realised it was also causing my broadband to drop out...

I have tried freequencys between about 250Hz and 20kHz and the DSL drops out fairly consistantly.
The PWM duty cycle is limmited to between 10% and 90%, aproximatly

I suspect the issue is RFI as opposed to there being anything superimposed on the DC. The router is down streame of a boost buck combination which stabilises its 12V supply, allong with the satellite box and the phone. Neither converter is isolated so I suppose noise could conceevacly pass through on the ground line but I am struggling with the mechanisme.

Interestingly the DSL occasionally drops out when the genny powered charger is on... more PWM I suspect!

So the question is... How do I fix it without resorting to long slow pulses which is possible but hardly optimal.

I literally have no idea, apart from proxcimity, which isnt going to be easy to change.

Thoughts?

Cherrs,
Al
Maybe a schematic??
Is there an inductor between the FETs and the batteries?
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
Just the one instrument I am afraid ... Rygol DS1054Z but it has FFT which I had never used...
Anyway, quite a bit or reading later...
I know now that a square wave, when considdered as a wave rathr than a digital signal, is a fundamental sine wave with a whole host of well defined harmonics that eminate from it or for that matter would sum to make it...

The point is that my noise is a fundamental part of the process of switching thinks on and off so eliminating it is a lost cause.
I also did some reading on broardband and now apriciate that it is actually designed to notch out anything it cant modulate so intermittent or variable noise is far worae than a clean signal.

Having actually looked at my gate voltage in detail, along with the uP's signal, my gate drive and the output to the load, it was clear that there were some significant oscillations allong with a comparativly slow slew rate. This seemed to be spawning a huge range of harmonics, many 2 or 3 orders of magnitude greatr than the fundamental carrier.

I played arround with decoupling caps, acros the totempole, and found that a 4.7uf electrolytic got rid of almost al thr high frequency oscillations and also increesed the slew rate by at least 60%. I was expecting to have to add a smaller ceramic in paralell with the electrolytic but I tried a few values and there didnt seem to be much of an effect.

Had I thought about it before soldering in the caps I would have got a screen capture from the scope to dmonstrate the improuvement but you know how that go's

Its running now, all be it only at a couple of amps on a test load and the noise seems better, subjectively.
There are stll plenty of harmonics showing up on the scope but thag are far better defined than thwy were, which should make them easier for the broardband to deal with.

I am about to go and play with a snubber circuit, just a basic RC acros the switch, to see if I can get rid of any of the remaining junk
I will try and get some screenhots and will also post a schematichen I get the chance.

Anyone know what kit I need to hook up to the scope to actually measure RFI? I have see. plenty of folk on the web using Am radios but given that I actually have a peice of kit that will work as a rudimentarry spcrum aniliser, all be it not real time, using it with something feeding its front end seems like a reasonable way to go.... would a simple coil of wire work?

More later

Al
 

Papabravo

Joined Feb 24, 2006
21,159
...

Anyone know what kit I need to hook up to the scope to actually measure RFI? I have see. plenty of folk on the web using Am radios but given that I actually have a peice of kit that will work as a rudimentarry spcrum aniliser, all be it not real time, using it with something feeding its front end seems like a reasonable way to go.... would a simple coil of wire work?

...
There is no substitute for the right tool. How hard could it be to rent one for a month?
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
Probably not that hard at all.... The problem. the hard bit is understanding what the kit is actually telling me...

I have got my head around the FFT Math function which appears to work well. I suppose I might think differently if I actually had a spectrum analizer to compare it to but I don't.

Anyway that aside I have more besic questions... when FET switching what is fast, how much ringing is too much.
How should I test the drive circuit, I am driving a halogen bulb as a test load right now which I amuse has far more inductance than a resistor but then I am also amusing a battery would have enough inductance to need to be considered.

I have the tools to measure down to 100pS if necessarry. I don't have, and cant afford to buy. a fast current probe but I assume I can look at a shunt if necessary as the scope has good dynamic range and offset capabilities.

Its a Rigol DS1054Z upgraded to DS1154Z 100MHz

Bandwidth 100 MHz
Analog Channels 4
Sample Rate 1GSa/s (Single-channel), 500MSa/s(Dual-channel), 250MSa/s(Fulll-channel)
Max Memory Depth 12Mpts(standard) ,24Mpts(option)
Max. Waveform Capture rate Up to 30,000 wfms/s
Real Time waveform Record and Replay Up to 60, 000 Frames(Opt.)
Std. Probes RP2200 150MHz BW Passive Probe:4 sets

There are a raft of math and analysis functions I need to learn how to use.

'O' I played with RC networks on across the S&D but everything either made the aberrations worse or did very little. I ended up with a 504 ceramic right on the pins of the FET which got rid if the big, as in a couple of voltd tall but very skinny, undershoot and overshoot spikes which sems to have helped with the noise a little more
The first harmonic is now approximately. 50% in DB, of the fundamental with a classic descending curve beyond that. I am still seeing quite a messy signal but then it is messy with the circuit shut down so I suspect it is noise from elsewhere anyway.
I am probably probing in the wrong place and in the wrong way... Pointers?

Work is getting in the way at the moment. quite unreasonably my boss expected me to stop playing and go fix something so no mucking likely about until the end of the week, at least its not sunny.
 
Last edited:

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
This is what I have now... Thoughts?
upload_2016-4-26_13-33-22.png
The fact that it looks too simple is of concern, as I suspect I haven't included something I should have...

I plan to increase the number of switches when they are correct and may also look at common drain FET's so I can remove the reverse protection diodes. Turning it into a buck as opposed to just a PWM might be a good plan too.

upload_2016-4-26_13-54-19.png
(6K8 ... Oops! )
Thoughts?
 
Last edited:

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
This is my FET board as it is now ...
upload_2016-4-28_1-16-44.png



upload_2016-4-28_1-17-48.png

I know the heat sink is too small and when I add apropper one I will include a temp sensor and a variable speed fan.
All the tracks under the FETs are as short as possible, as in they do not extend along the board anywhere they are not needed.

Can anyone tell me what rise and fall time I should be able to achieve and how to get there if the circuit I have is not geed enough.
In fact I would appreciate a general critique, irrespective of the noise issues I mentioned earlier.
The new less noisy pulses have not been tested with the DSL yet, no time and no sun!
Feel free to be frank :)

Cheers,
Al
 

ronv

Joined Nov 12, 2008
3,770
This is what I have now... Thoughts?
View attachment 104987
The fact that it looks too simple is of concern, as I suspect I haven't included something I should have...
Thoughts?
Let's look at this one:

I like your idea of turning it more into a buck. Even a little inductance between the circuit and the battery would slow down the current rise time. Right now there is a big voltage and current pulse when it switches. I'm not sure what the solar panel thinks about that, but it's probably happy enough.
Noise wise you could twist the wires from the panel and to the batteries. The inductance there is probably doing some strange things. A big cao at the input from the panel might also help.
Some scope pictures of the input, output and gate voltage would be useful.
 

Thread Starter

Dyslexicbloke

Joined Sep 4, 2010
566
Good point RE the cable, but twisting it isn't an option... Its a 10mm 4 core SWA with two parallel conductors.
However I haven't but absolutely could ground the armor which would probably go some way to helping with any radiation.

I wondered about a cap on the input but I think that could potentially make things worse because at the point of switch on the effective impedance would be lower thus creating a higher current. I appreciate that there is capacitance in the cable but there is also inductance and the limiting factor right now is the PV as far as I know.
Of course if I go buck-esque this will change and an input capacitor would be required.
It is probably worth doing some tests, starting with small values and keeping an eye on the current spike.
I will try a shunt along with A-B on the scope to give me a pseudo differential and see if it is still quick enough to be useful.
If I install the shunt with one end grounded first I will have a single ended capture to compare a mathematical differential capture to and see what is lost, something will be.

Scope pictures ... Yes, probably over the weekend if we actually get any sun. I will try and find something to use as a shunt, even if its just a thin wire. I picked up a couple of cut ends from a 35mm earlier today so I now have some copper that I could probably make a shunt from, I may even be able to do a rough calibration by crimping it a little, like you find in cheap multimeters.

With respect to buck converters, or a buck converter, would I be better with several conductors, one on each switch, or one big one?
I have a nasty suspicion that is a whole other ball of wax but given the need for a diode/switch to carry current when the field is collapsing I suspect many small switches is more practical with readily available components.

I have absolutely no experience and little understanding of magnetics and how to use them. I dont even have a feel for how big things are or how that relates to value. I do know that the core geometry and permeability will affect the value and current limits because saturation will occur at different for components that look similar but I couldn't quantify this and dont understand the spec sheets ot the impact of the values on a real world circuit.

This ignorance is what drove me to try to parallel small buck boards in the first place. There was another thread for that which is dead now and it would be bad form to rehash it here so I will stop there, for now at least.

Anyone spot any issues with the existing circuit, board. or proposal to use two FETs with common drains?
 

ronv

Joined Nov 12, 2008
3,770
Good point RE the cable, but twisting it isn't an option... Its a 10mm 4 core SWA with two parallel conductors.
However I haven't but absolutely could ground the armor which would probably go some way to helping with any radiation.

I wondered about a cap on the input but I think that could potentially make things worse because at the point of switch on the effective impedance would be lower thus creating a higher current. I appreciate that there is capacitance in the cable but there is also inductance and the limiting factor right now is the PV as far as I know.
Of course if I go buck-esque this will change and an input capacitor would be required.
It is probably worth doing some tests, starting with small values and keeping an eye on the current spike.
I will try a shunt along with A-B on the scope to give me a pseudo differential and see if it is still quick enough to be useful.
If I install the shunt with one end grounded first I will have a single ended capture to compare a mathematical differential capture to and see what is lost, something will be.

Scope pictures ... Yes, probably over the weekend if we actually get any sun. I will try and find something to use as a shunt, even if its just a thin wire. I picked up a couple of cut ends from a 35mm earlier today so I now have some copper that I could probably make a shunt from, I may even be able to do a rough calibration by crimping it a little, like you find in cheap multimeters.

With respect to buck converters, or a buck converter, would I be better with several conductors, one on each switch, or one big one?
I have a nasty suspicion that is a whole other ball of wax but given the need for a diode/switch to carry current when the field is collapsing I suspect many small switches is more practical with readily available components.

I have absolutely no experience and little understanding of magnetics and how to use them. I dont even have a feel for how big things are or how that relates to value. I do know that the core geometry and permeability will affect the value and current limits because saturation will occur at different for components that look similar but I couldn't quantify this and dont understand the spec sheets ot the impact of the values on a real world circuit.

This ignorance is what drove me to try to parallel small buck boards in the first place. There was another thread for that which is dead now and it would be bad form to rehash it here so I will stop there, for now at least.

Anyone spot any issues with the existing circuit, board. or proposal to use two FETs with common drains?
Yea, I don't think you need a true buck converter, just some inductance in the output (more than just the cable) to slow down the current and voltage rise times.
Actually I think you could just wind a small air core inductor.
I think the scope picture will scare you when you see them. :eek:
 
Top