Archived Forum

It is likely nothing is wrong. The standard CT and burden values give you a scale of 0 - 100 A. You don't say what your system voltage is, but assuming it is 230 V then your 10 W represents a current of 43 mA rms.   Even assuming all components are perfect, theory alone predicts massive errors at very low currents and the situation only becomes tolerable at around 0.25 A (even there, the theoretical error alone is about 20% of reading).

It's all down to the way that the input to the analog to digital converter is biased to the mid-point of the input range (512 counts), and then that value is removed by the software. If zero current accurately sat half-way between steps of the converter, you would correctly read zero with no current, and any electrical noise picked up would have to be more than half a bit to register. But if the bias network is such that it is just on the transition between two steps, then the slightest noise will cause the converter to flip between two values and it will see this as a current.

That is probably what you are seeing.

Ok i see.

So even without a CT attached it could read something other then 0 ?

I've noticed there are also 30A versions of the CT sensor  (SCT-013-030).

Since i wont measure anything more then 20A these would fit my setup an minimise these errors ?

According to the data sheet, the SCT-013-030 might indeed be more suitable. However, I have not tested a sample to verify the published parameters. You would need to remove the burden resistor (15 or 18 Ohms) on the emonTx circuit board because the burden is integral in this model. And of course the calibration constants in the sketch are different too.

Ok, just ordered some of them to see if they work 'better'  for me.

Another thing i was thinking: when using a 100A CT sensor and looping the live wire 3 or 4 times through the CT sensor would that give a more accurate reading (3 of 4 times higher of course but)

That is exactly how I tested the !00 A CT - up to 250 A ! 

[ Actually, 250 Ampere turns].

A cautionary note: The problem you will have is a physical one of getting the wire through the core without putting any appreciable strain on the ferrite (which is brittle and will break without too much persuasion). And make sure the primary wire is generously rated, because you now have 3 or 4 wires bunched together, albeit for only a short distance.

I did some tests today and indeed by looping a live wire two times through the sensor it doubled the output. So instead of the actual 10A its measuring 20A. I  had to divide the correction factor in two from 111.1 to 55.55 to get the output right.

The emonTX still says its measuring 8-10Watts with no load on the live wire. I also activated the CT2 and CT3 sensor in the sketch (emonTx_CT123_Voltage) but without actually sensors attached and these are displayed as 0 Watt.

"I  had to divide the correction factor in two from 111.1 to 55.55 to get the output right."

That is correct.

"The emonTX still says...  "

What does CT 1 channel display when the CT is unclipped from the wire?

And with the CT unplugged from the emonTx?

Even when no CT is connected it still displays about 8W-10W.

So i made workaround in the emonTX sketch not to transmit any power <20

The resolution of the A-D converter is 3.2 mV; 10 W with the standard CT and burden resistor gives 0.4 mV rms, or about 1 mV peak-peak.  It seems likely that the input voltage with no CT lies almost exactly on the decision voltage from one step to the next of the A-D converter, and then the slightest induced noise voltage will cause the ADC output to flip between two adjacent numbers (511 - 512 maybe) and this will be seen as a current.

Your solution is probably the most pragmatic.

I did some tests with the  SCT 30A version, removed the burden resistor from the emonTX and used 30.3 as calibration value in the sketch.

It does give better precision then the 100A CT.

I also did some tests with a 20A CT (TA17L-04) and these are pretty precise. Used a 220R as burden and 0.09 value for calibration.

Measuring with these three CT's on the same source (60W lamp and 2200W watercooker) shows the 100A was way off all the time.

If you study the theory of operation of an analog to digital converter in the presence of low amplitude signals, you will find that there is a systematic error that depends on both the signal amplitude and its quiescent level in relation to the signal zero of the converter. For any ADC, the relative error increases as the amplitude falls, and can exceed 5% below 20 counts, and 10% below 6 counts amplitude.

The maximum amplitude is of course 512 counts for our 10-bit converter, but depending on the choice of burden resistor and the allowance made for overload and component tolerances, the rated current of the CT will be less than this. (With the standard components at the lowest value within tolerance, 100 A represents an amplitude of 391 counts).

Therefore, on this theoretical basis alone (which ignores all other errors, both in the ADC and elsewhere), and all other things being equal, I would expect the 30 A CT to be one and a half times worse, and the 100 A CT to be 5 times worse, than the 20 A CT on the same load.

I could not find any published literature relating to this, I have written it up and I think Trystan Lea might incorporate it into the website at some point.