Frequency sensor cct for a power line...

There are approximately one bazzilion different ways to do this. Even when you throw out the 99.44% of them that are pointless, that still leaves you with a bazzilion different good ways to solve this problem.

Tell us what you really need, and maybe someone can help. High precision? Quick acquisition? Low cost? Easily understood? Uses no processor resources? Minimizes external components through clever use of processor resources? Sensitive to fast variations of frequency? Insensitive to short-term changes? What?

--
www.wescottdesign.com

What is the country that uses 300V 50Hz for power? Homeworkia?

VLV

t

"small fluctuation" being what ?

Something low cost, and out of the box ? : see

Absolute precision matters ? - Divide by 50, and phase compare with the 1pps from a GPS system ?

More interested in Phase or Energy deviations ?

If you told us what part of Homeworkia this had to work in, that would help....

-jg

line

Thanks for the reply .. Ya,, followings are the specifications :-

1) sense 49.5Hz 2) Will be used in a 230V Electric power line 3) Quick response- within abt 100mS. 4) low cost 5) It will be used in portable device 6) Simple If you knw something similar to our requirements, pls be kind to forword.. :)

Thanks..

--------------------------------------- This message was sent using the comp.arch.embedded web interface on

It

actually i put it cos's the device should be withstand higher voltages also.. dat's Y..

--------------------------------------- This message was sent using the comp.arch.embedded web interface on

I=

I want to track 49.5Hz.. from the fluctuation around 50Hz , i want the moment dat goes below 49.5Hz. it's actually used in 230/400V, 50Hz LV system.

--------------------------------------- This message was sent using the comp.arch.embedded web interface on

.

Err, but elsewhere you said :

3) Quick response- within abt 100mS.

and 100ms, is NOT the same as 'the moment it goes below 49.5Hz'

- but is actually ~5 cycles. Which one is right ?! You could, in theory, detect inside half a cycle, but that will be prone to false triggering on phase disturbances too...

Simplest/cheapest is a monostable wired as a Frequency detector. (ie senses 10.101010, 20.20202020ms, or multiple - you can divide the AC to get some disturbance filtering tolerance.)

Google the CSS555C for a precisely trimable timer. that can be very low power, or if very low power and Digital trim are less important, a 4528 monstable and some manual adjust will do.

-jg

The phase difference will increase by 180 degrees every second.

That is only 5 cycles at which time the phase shift would increase by

18 degrees or 3.6 degrees each cycle.

This would not be too hard, if the mains would be constant amplitude pure sine wave, unfortunately this is not true for the mains these days.

Typically the drop in frequency is associated with a too heavy load compared to production, which might also cause voltage drops, so the peak amplitude varies. making it hard to use any waveform matching techniques.

The noise and distortion on the mains line will make it hard to use zero crossing techniques, since exact time of the zero crossing varies due to the distortion.

In three phase systems the neutral wire does not carry any current if there are symmetrical resistive loads on each phase, since the neutral currents cancel each other. However with single phase rectifier loads on all phases, these will produce a lot of third harmonics which are all in phase and are summed in the neutral wire, causing voltage losses in the neutral wire.

Trying to measure the frequency between one phase and neutral will suffer from the 150 Hz voltage drop in the neutral wire. Thus, it might be a better idea to measure between two phases (400 V) instead between phase and neutral (230 V).

To increase the reliability to detect the abnormal low frequency, one might use three separate detectors in the delta configuration and accept the result only if all three measurements are the same. A single badly polluted phase would not cause false alarms.

Paul

The minimum hardware count way would be to put it through a zero crossing detector to square everything up, a comparator with hysteresis. Feed that into the timer input on your micro, measure 10 or so high and low periods, reject any outside a set tolerance, then average out what's left to to get period / 2. Increase or decrease the number of samples to get more or less smoothing...

Regards,

Chris

I don't think you're going to find what you want, if for no other reason than because it is both too easy and too hard. The too easy part is actually figuring out the frequency of a sine wave; the too hard part is going from your 220V line to a nice clean sine wave.

I _suspect_ that after you figure out a good way to sample the voltage of the line across whatever galvanic barriers are between the line and your device (this would be fairly easy if you are using a power transformer, even an AC-out wall-wart) you should be thinking along the lines of a low- pass filter to clean up the worst transients, a comparator, and the built- in timer circuitry in the microprocessor that I hope you're using.

--
www.wescottdesign.com

I don't understand why he would need a nice clean sine wave, other than samplings more than 100Hz. But his accuracy and response requirements would be difficult to achieve.

The input signal (i.e. mains voltage) can be for instance a sine wave (the ideal case), a square wave (e.g. a sine wave through a limiter), a triangle wave or even a sawtooth wave and the waveform can change several times each second. The frequency detector must work reliably with any of these waveforms.

However, if you are close to a big hydroelectric or gas/steam turbine generator, most of the distortion is compensated by the large inertia of the turbine/generator. For example, for the Scandinavian network (Nordel: Danish Isles, Finland, Norway and Sweden) the actual frequency and the mains derived clock error can be found at

However, if you are on a weak grid and need to determine when to kick start some local production to compensate for the frequency fall (production deficiency) it might be a better idea to use some form of communication link to increase the production, before the frequency drops too much.

Paul

The shape of the waveform does not affect the frequency, as long as he sample at the same level repeatedly. So, sine wave or not, it doesn't matter.

Yes it does. He wants to get a reading in 100ms, or 5 cycles. There can easily be enough variation in the waveform over 5 cycles to give a false reading.

You are right in the steady state, but not the dynamic state. If you count cycles over a second or ten, you'll get a very accurate value for the average frequency over that period. At the opposite extreme, if you only count one half cycle, if it's a pure sine wave you'll get a very accurate reading again, but if it's something else you'll get a wildly inaccurate reading. The accuracy depends on the sample length.

Nobby

If you are using zero crossing or 50 V crossing or some peak detection above 300 V, you may experience false crossings due to the distortion.

My initial reaction to this question was to use FFT, but the requirement for less than 0.5 Hz bins and 100 ms reaction times is not practical.

If the real need is to increase fuel injection to a prime mover (diesel or turbine) of a generator, I would simply monitor the rotation speed of the generator axle, which can be measured with a large number of pulses each rotation.

Paul

er

tic

=A0Even

ill

blem.

d? =A0Uses no

ver use

?

within

eason

is

rt is

That depends on the type of load on the system.

Perhaps the OP can clarify on the type of work load so sensitive to frequency and response time.

An FFT may not be practical, but a DFT at 50 Hz, with a large oversampling, is.

Sample at, say 800 Hz = 16*50 Hz. Every 16 samples (one nominal cycle), do a discrete Fourier transform for 50 Hz: float re=0,im=0; for (i=0 ; i < 16 ; i++) { re += adc[i] * cos(i * 360.0 degrees/16); m += adc[i] * sin(i * 360.0 degrees/16); } phi=atan2(re,im); and then repeat with the next set of 16 samples, and so on.

And check to see how fast phi is varying from cycle to cycle. If it varies in the appropriate direction by more than 72 degrees in 5 cycles (0.5 Hz * 100 ms = 0.05 cycles = 72 degrees) then you are past 49.5 Hz.

Of course, the library routines will use radians instead of degrees, you have to watch out for the wrap (suddenly going from -179 to +179 is only a 2 degree change), and you will want to make a table of 16 values for cos and sin rather than calculate the same values 800 times per second each. You will also want to throw out wild values unless they persist for several cycles, and monitor the mean ADC value for baseline offset and re^2+im^2 to see if the AC voltage is in spec. Use an unrectified wall wart to get a cleanish low voltage image of the power line. But those are details.

There are plenty more speed-ups available to cut down on the computational load, but once it is fast enough to run on a 53-cent processor, it is probably not worth your time to optimize further to fit it on a 37-cent processor.

--
David M. Palmer  dmpalmer@email.com (formerly @clark.net, @ematic.com)

These shapes are all caused by harmonics. If you apply a good low pass filter, you'll end up with a fairly clean 50Hz sine wave again.

Meindert

What is the delay of such filter :-).

Sure, you can add a steep filter, but does it produce a usable response in the specified time ?

Paul