PDA

View Full Version : Line Frequency Detection / Measurement



JonnyMac
03-29-2012, 10:06 PM
This should be easy -- I've done it before with the SX but with the Propeller I'm getting sporadic, unexpected results. What I would like to do is detect actual line frequency at start-up of my phase-dimmer object so that the same code will work world-wide. Again, should be easy.

Pics show the circuit (no-brainer, everyone uses it), output from collector/pull-up junction, and wacky sporadic results in test program.

All thoughts and suggestions are appreciated.


dat

org 0

entry mov outa, #0
mov dira, chdirs

loop waitpne zcmask, zcmask
waitpeq zcmask, zcmask
neg period, cnt
waitpne zcmask, zcmask
waitpeq zcmask, zcmask
add period, cnt
sub period, #4

wrlong period, par

rdlong period, #0 ' 1/4s delay
shr period, #2
add period, cnt
waitcnt period, #0
jmp #loop

Peter Jakacki
03-29-2012, 10:16 PM
I've only got a minute but I'd say that you have a combination of slow edges and very fast resolution. The Prop's counters are hooked up on caffeine and what it sees is the transition as going by like a river of molasses. It went by, no wait there it is, no wait?^*#$? What the circuit needs is either hysteresis or introduce a delay between pairs of waits. Just a quick thought. So wait till it goes high, wait 100us, wait till it goes low.

Bean
03-29-2012, 10:17 PM
Jon,
Seem like there may be some jitter on the signal when it crosses the threshold voltage. The SX may not have been fast enough to see the jitter, but the propeller is.

I'd try putting a couple millisecond delay between all the WAITPNE and WAITPEQ.

Just a guess...

Bean

P.S. Peter beat me to it...

Phil Pilgrim (PhiPi)
03-29-2012, 10:26 PM
Jon, there could be noise on the power line, in which case the Propeller's speed is your enemy. Try a Schmitt trigger to de-glitch your input.

-Phil

Beau Schwabe (Parallax)
03-29-2012, 11:08 PM
Jon,

I agree with the other posts... also if you have any X-10 devices connected anywhere in the house they could be causing this sort of behavior. Even some of the 'Smart Grid" equipment that communicates during the zero crossing might potentially cause this sort of behavior. Have you scoped your mains, to see just how clean it is?

JonnyMac
03-29-2012, 11:59 PM
I wasn't willing to hook my PC-based 'scope to the mains so I assumed my friends' supposition about noise was correct. I don't have an X-10 or other devices operating in my apartment but I do live in a small building and I have no idea what my neighbors are doing. Turns out a bit of debouncing does the trick.

Am I fooling myself?


org 0

entry mov outa, #0
mov dira, chdirs

waitlo1 mov t1, #10
:loop test zcmask, ina wc
if_c jmp #waitlo1
djnz t1, #:loop

waithi1 mov t1, #10
:loop test zcmask, ina wc
if_nc jmp #waithi1
djnz t1, #:loop

neg period, cnt

waitlo2 mov t1, #10
:loop test zcmask, ina wc
if_c jmp #waitlo2
djnz t1, #:loop

waithi2 mov t1, #10
:loop test zcmask, ina wc
if_nc jmp #waithi2
djnz t1, #:loop

add period, cnt
sub period, #4

wrlong period, par

rdlong period, #0
shr period, #2
add period, cnt
waitcnt period, #0
jmp #waitlo1


Update: Well... that seems to have solved the problem and having installed this into my 4-channel AC dimmer object everything seems to be working well. Of course, I will spend a lot of time wringing it out before going public. The attached image is the board I'm working with (prototype of the EFX-TEK FC-4+).

jmg
03-30-2012, 01:18 AM
All thoughts and suggestions are appreciated.


Getting reliable Zero crossing on mains, is non trivial.

If you just want to make a Frequency decision, you can do a voting system, and discard the 'silly answers'
- ie a form of simple digital filtering.

If you want a more reliable phase-zero, for phase control, then you should make a bandpass filter, that removes spikes but does not phase shift too much at mains frequencies.

eg a series RC gives a phase-lead, and then a shunt RC gives a phase lag.

The toughest test situation is with three phase power, where another phase switches real power, right where you hope zero is.



Turns out a bit of does the trick.

A bit of ... what ?

JonnyMac
03-30-2012, 01:52 AM
Turns out a bit DEBOUNCING of does the trick. Sometimes my fingers work faster than my mind. We are going to put a bit of filtering on the input to the Propeller but I'm hoping that the code trick is enough for bad situations. I'm happily running a set of incandescent lights up and down now.

jmg
03-30-2012, 02:14 AM
The attached image is the board I'm working with (prototype of the EFX-TEK FC-4+).

I'd be cautious about those metal tabs just waving up in the air... anyone can slip..

Can you not get the devices in the TO220F or equivalent package ?

JonnyMac
03-30-2012, 02:37 AM
We're using triacs with insulated tabs.

Phil Pilgrim (PhiPi)
03-30-2012, 02:41 AM
Jon,

Looking at your board, I'm concerned about your mains isolation. It's not enough just to use optoisolators. You also have to maintain wide separations between conductors on the isolated vs. non-isolated sides of the circuit. In particular, you have a groundplane that surrounds the mains signals that really should not be there. Also, there's a diagonal trace emanating from the H11AA1 chip that runs perilously close to the hot side of the isolator and one of the 15K power resistors. In extreme cases, I've even seen slots cut into PCBs between the two sides to eliminate arc paths on the surface. That's probably not necessary here, but I mention it just to underline the importance of conductor separation distances.

Here's an article that might help:


www.edn.com/contents/images/454635.pdf

-Phil

Duane C. Johnson
03-30-2012, 03:07 AM
Hi JonnyMac;

Filtering greatly helps, but then the rise-time is even slower.

The Schmidt trigger, along with filtering is the best solution.

Usually we use a hardware gate with Schmidt trigger gates.
However, with the Prop one can make a Schmidt trigger using 2 pins.
Forgive me but I don't remember what the obex is but it was describer here recently.

Basically one pin is the input and immediately the value is copied to a second output pin.
There is a resister from the output pin back to the input pin which changes the input
bias trip point. This greatly decreases the likelihood of false triggers.

Duane J

JonnyMac
03-30-2012, 03:09 AM
@Phil: John Barrowman (also a former Parallax employee) did the layout and we've done UL approved HV boards in the past for clients. Still, I appreciate the feedback and will forward to JB. This is our first prototype and we have already identified some adjustments. Gotta love GoldPhoenixPCB for quick-turn, high-quality, low-cost prototypes.

JonnyMac
03-30-2012, 03:11 AM
@Duane: I will have a look for that in ObEx; I have a spare pin and as this is a first prototype, we can make adjustments.

jmg
03-30-2012, 03:45 AM
Also, there's a diagonal trace emanating from the H11AA1 chip that runs perilously close to the hot side of the isolator and one of the 15K power resistors.

Yes that is certainly marginal, perhaps the Layout designer thought the resistors were isolation ?

8mm is the creepage distance some use, but certainly you should maximize within practical reason.

I'd also spread the 15K power resistors apart further; most are rated standing alone.

Duane C. Johnson
03-30-2012, 03:46 AM
The Prop Schmidt trigger is directly related to the Sigma-Delta AtoD.
Except, the driving pin output is in phase with the sensing pin deliberately deviating from the trigger point, as opposed to, in the D-S AtoD, which is always attempting to drive toward the trigger point. I.e, NON-INVERTING output as opposed to INVERTING output.

Duane J

Phil Pilgrim (PhiPi)
03-30-2012, 03:52 AM
Looking at the board again, I may have mistaken glare from the lighting for a groundplane. If that's the case, please forgive the misrepresentation. But that diagonal trace is still a concern.

-Phil

Peter Jakacki
03-30-2012, 04:40 AM
That diagonal trace is certainly a big concern, it seems to compromise all the isolation that you have elsewhere. This would need to be fixed for the production version at least, I cannot see how this would pass compliance testing as moisture and dust and aging have to be taken into account as well as indirect lightning effects. But I notice the rather large 15K resistors used in the zero-cross detection and it seems to me that this is an area that could be improved. The grade of opto could be improved and the resistors increased in value and decreased in size or you could use a darlington output version (http://search.digikey.com/us/en/products/PS2506-1-A/PS2506-1A-ND/770684) (no need for speed) and far less current. The resistor on the output could be easily increased and a capacitor added for pre-filtering. Just some thoughts but we all know that whatever we come up with somebody always spots something that could do with improvement :)

lmclaren
03-30-2012, 09:06 AM
I use a propeller to phase dim 40amps mains with 100 step resolution. It controlls a heating element and a universal motor. The trick with zero crossing is to use two opto couplers with the inputs reversed on one.

Connect both outputs together and to a pullup resistor. Now the either opto is turned on all the time except for the very short interval between -0.6 and + 0.6. Works like a charm and is not sensitive to load or change.
As you can imagine, I am generating a lot of motor noise and the dimmer is rock solid.

DONT FORGET THE CURRENT LIMITING ON THE INPUT! (I know you wont but need to say it)

Lee

Paul
03-30-2012, 01:34 PM
@lmclaren, Could you use a PC814 with dual (but opposite) LEDS for this purpose? Or do I have your circuitry backwards? We currently use the PC814 to detect a phase loss in a 3-phase system.

EDIT: Just noticed the PC814 and H11AA1 are similar. ...never mind.

jmg
03-30-2012, 08:45 PM
Certainly the 15K can be increased, and the Secondary load can also go up, it just needs to be safe relative to the leakage currents. Somewhere in 100K-1M region ?

Interesting is the CTR symmetry (skew) on these parts, and not all are equal. Some say 2:1, or even 3:1! and I see an Everlight 814A says 0.7<1< 1.3;
If this is only for AC frequency that does not matter, but for true zero cross as Phase Zero, you really need a known balance, and probably a Centre of Two edges design to take OptoGain out of eqn.
Least skew would be from a single LED driven from a bridge, but I've not found a vendor doing that.

My prize (by far) goes to Everlight EL814SATA 16c/500 at Mouser, and the best symmetry value.

JonnyMac
03-30-2012, 10:28 PM
Bear in mind that I want this to work in Europe and Australia where our customers could not use the previous version of the FC-4. We set the circuitry to work at 220v, and the AC dimmer code detects line frequency on start-up of the object.

jmg
03-30-2012, 10:52 PM
Bear in mind that I want this to work in Europe and Australia where our customers could not use the previous version of the FC-4. We set the circuitry to work at 220v, and the AC dimmer code detects line frequency on start-up of the object.

I was assuming that, which is why the 15K is a little on the low side, and also why a better spec Opto coupler matters.
I'd flip to SMD couplers as they remove voltage from more of the board area, and likely have better long term availability.

JonnyMac
03-31-2012, 12:21 AM
All the optos on the board are SMD -- "gull wing" versions of the DIP packages the manufacturers offer.

The board shown is a prototype (we try to do those nicely, saving work later) and have identified some changes (in addition to the comments in this thread). I will try the EL814SATA for the next run.

waltsailing
03-31-2012, 02:00 AM
Hi

Why not do it a different way, using the following, make an array, 1024 points, make an a/d out of a pin, sample the signal into the array, for say 1024 points... Set the sample rate so that you can easily resolve 50 or 60 hz, Maybe a 1000 Hz Sample rate... Then do an FFT, go and take the magnitude, Set a threshold, or search the array for the peak in the bins near 50 or 60 hz, and decide which is largest. The harmonics, and other parts of the signal are not that important, because you will effectively filter out the junk and will be able to decide what the line frequency is by looking at the fundemental and knowing about where to search. Right selection of sample rate, makes the index = the frequency. The fundemental of the line frequency will be clearly visible. You dont need to do this all the time, just once in the beginning. Perhaps you could verify that the line freq does not change much, either by running this more than once. I have not written this in Spin, but there are objects that do most of it. and the prop should be able to do this pretty easy. There is no need to really worry about real time, provided the sampling is done at a known fixed frequency. I dont think it matters much that that the signal is not a nice sine wave, the fundemental will show up just fine.

Just a thought, *** this would eliminate all the timing issues.

This is an interesting thing to read. Good luck with your project,

Regards,
Walt,

jmg
03-31-2012, 02:16 AM
FFT is possible, but an over-kill - here you have only one primary frequency, for which you already know the two main possible candidates.
A design usually wants a (stable) Zero Cross time reference anyway to base the Phase calcs on.
So it is easier to work from this base.

pjv
03-31-2012, 06:49 PM
Hi Jon;

I make a product that is targetted for 50/60 Hz and 110/240 volt operation around the world. It (still) happens to use an SX chip, but the same approach could be used for probably any micro.

The mains powers the micro directly through a "line rated" 1 uF capacitor into a power limiting 12 volt zenering transzorb, from which point it is series dioded to a 100 uF bulk storage capacitor that feeds a 5 volt regulator.

Beauty of this scheme is that the 1uF capacitor is a reactive limiter, and hence no heat is generated in dropping the 110/220 line voltage to the 12 volt zener.

Then I simply connect the hot side of the line through a couple of 1 meg resistors to a micro input pin as my phase reference. The input threshold is 2.5 volts, and the slew rate of the mains at that point is very fast, yielding wonderfully reliable results. On bootup, I first count and average the number of clocks between transitions on the phase input, and then adjust my master processor timing delay to get 128 time slots per half cycle, and sequence everything from that.

Works great.... thousands sold!

Cheers,

Peter (pjv)

JonnyMac
03-31-2012, 07:55 PM
My [timing] code is doing the same, thing, Peter; though I'm dividing each half cycle into 256 slots.

jmg
03-31-2012, 10:18 PM
Beauty of this scheme is that the 1uF capacitor is a reactive limiter, and hence no heat is generated in dropping the 110/220 line voltage to the 12 volt zener.


No series surge resistor ? It may avoid any heat, but the power factor is lousy, and it will cost more to operate this over its life.
I guess if it is in kW dimmers, the control cost is in the noise floor, but for other applications, this does need care.

You also should provide a discharge path for the 1uF, if it is plug-able, otherwise it can sit hot waiting to bite someone...



Then I simply connect the hot side of the line through a couple of 1 meg resistors to a micro input pin as my phase reference. The input threshold is 2.5 volts, and the slew rate of the mains at that point is very fast, yielding wonderfully reliable results.

We have used the VR37 series of high voltage resistors for this, but usually no less than 4M7, as you need to keep well clear of Earth-leakage tests of 250uA. Digikey shows 10M and 3M3 and 4M7 as stocked in tens of thousands.
Some designs use two VR37 in series, for better spike (lightning) protection.

I see the Prop specs +/- 1uA MAX Leakage, and Clamp max of 500uA in IO pins.