Teardown, Accuracy Testing: LP7663 Kill-a-Watt-clone Power Usage Meter

Sometimes when I mention an idea, I will quickly become fixated on it and try to flesh it out as best as I can. Just yesterday, I was mentioning about how low-cost power usage meters can mislead, and today, I’m looking at one which I have used in the past in my university days to see what’s inside and just how accurate it is.


20151223-2040-0832 20151223-2040-0833

The meter we are looking at today is a Solar Inverters branded LP7663 Power Usage Meter. This meter is pretty much a clone of the Kill-a-Watt meter and allows you to measure a range of parameters simply by plugging it in-line with your device under test. The LCD screen indicates the voltage, frequency, current, power, apparent power, power factor, electricity cost, running time and accumulated energy. Most parameters are given to a resolution of one decimal place.

The unit is held together with three small Phillips-head screws. Opening it up is as simple as removing the screws and carefully separating the shell.


The wires leading to the plug are soldered to the pins with assistance from a guide PCB. As they are “permanently” attached, it makes taking it apart a little less convenient.


The wires are soldered to the first board, which is the power supply and shunt board. It appears from the markings on the board that the unit was produced in Week 49 of 2009, and has a silkscreen text marking of P2H2820R-R.


Looking from the front side, we can see the brass spring-together contacts which form the socket at the front, as well as the shunt resistance (R2) made of a piece of brass wire link. Calibration at the factory has been achieved by grinding away some of the brass at the top to achieve the desired resistance. The shunt is installed on the active side of the plug, to keep the neutral and earth undisturbed.

Aside from that, a very basic half-rectified power supply appears to be in place, formed by D2, R2, charge stored in C1, all fused by F1. A further diode drop is added by D1 for the downstream board.


It’s good to see that the capacitor is an X1-rated capacitor for safety reasons.


The main board is a single-sided paper PCB type board, dated Week 48 of 2009. It’s got a product code of GF-01. On the rear, the buzzer that provides the button-push beeps, a timing crystal, and assorted wire-links, transistors, capacitors and resistors can be seen. The quality of the flex ribbon connecting the two boards and its voltage rating might be somewhat questionable given half-wave rectified mains is travelling on it.


The other side of the board has the buttons and LCD, hiding what is actually “powering” the whole meter. Carefully peeling them off reveals the following …


… namely a gob-topped main IC, which is unknown, along with a JRC 2902 general-purpose quad op-amp, and Atmel AT24C02B Two-wire Serial EEPROM (256 byte).


The LCD connects to the main board by an elastomeric connecters. For a nice touch, the button contacts are not elastomer rubber, but instead, are brass plates which offer better button lifetime.

I thought I’d give this one a standby power test to IEC standards as well, and it came out with a no-load consumption of 829.87mW (full report). In real life, when using it in-line with an appliance, there is a further power loss, namely in the shunt which consumes more energy, so if you’re looking to save energy, it’s not advisable to leave these meters connected in-line on a permanent or semi-permanent basis.

Why Testing Power Meters is Complicated

I hvoltage-accuracyad a thought about testing power meters for accuracy, and initially, I thought I would take the route similar to that of when I tested multimeters for accuracy.

This strategy would be fine for testing voltage indication where no load is connected, as in the diagram on the left, where basically a variac autotransformer is used to present both the PA1000 Power Analyzer (the reference meter) and the LP7663 (the unit under test) to a varying AC voltage. But as soon as we introduce a load to test for power accuracy, things get complicated.

Lets first consider testing for current alone. If I simply hooked it up in the manner above, but now added an incandescent globe load to the end, sure, the current through the entire system is one current as there is only one circuit, but because I am varying the load by varying the voltage with the variac, the voltage can get too low and the self-powered meter may shut down.

So lets reverse the situation and have the variac connected to the end with the globe load. This is now a better arrangement and would be fine for testing current alone, while ensuring the self-powered meter doesn’t see voltage variations that would affect its accuracy or cause it to shut down.

But AC is an interesting beast, and the phase relationship and waveform of the voltage and current is important to finding its power. Without knowing whether the meter displays Amps peak or Amps RMS, and how accurate its power factor readings are, it wouldn’t make sense to multiply the two together blindly, so instead we need to think about measuring power as well, as the meter’s outputs are not just the two figures multiplied together.

If we have the arrangement above, the PA1000 measures the power consumed by the meter as well as the load, which would not make the figures comparable. The worst part is that the meter power consumption varies with the load too because of the shunt resistor, whose value is unknown and can’t be measured unless I desolder the shunt from the board and then replace it, which would alter the calibration of the unit.

So lets reverse the arrangement and have the LP7663 measure the current first, followed by the PA1000 which draws power from another circuit. The PA1000’s 1A shunt is known to be 0.6 ohms, so we can theoretically compensate for it by using P=I^2*R and using the current read-out to add to the power read-out. However, this arrangement does have its drawback, namely that now we are dealing with two variables from the PA1000, and their errors can become additive.

Instead, because of a peculiarity of the PA1000’s connections, we can kill two birds with one stone by connecting the loads as follows:


As the PA1000 has Voltage and Current input pairs, and the break-out box has a Vlo-source and Vlo-load, if we use V-lo-source, we can measure the power of the load with the shunt resistance power as well. So in this arrangement, the LP7663 meter measures the power of the load + PA1000 shunt, and the PA1000 measures the same thus no compensation is required. It’s amazing what you can do with a little thought.


As a result, we can measure current and power accuracy metrics in the same run by reading the appropriate readings, as the current and power should be measured identically for both meters in this configuration.

It pays to remember that the AC current waveform is also important, and sadly, I don’t have a way to generate all different sorts of waveform “safely” and conveniently, so I settled on two main choices – a 20151224-1231-0863non-distorting and a distorting waveform to see how the meter reacts. The non-distorting waveform is made by varying the voltage supplied to a 150W halogen globe using a variac, which smoothly changes the voltage, resulting in a current through the globe which is sinusoidal and matches the input voltage.


The distorting waveform was supplied by varying the power through the 150W halogen globe using a triac dimmer, in this case, an IKEA Dimma lead.

20151224-1229-0858 20151224-1230-0860

20151224-1230-0862This is a highly convenient option, as I don’t have to construct anything. Instead, it comes pre-made with a slider on the top, and is rated for incandescent loads up to 300W. It comes with a Y-lead where the incoming power and dimmed device is connected, and a controller puck at the end of a twin-sheathed cable. Triac dimmers work by chopping up the mains waveform, and thus represent a distorting load with varying power factor.


20151224-1227-0856This might have been enough, if it were not for the fact that the variac itself has a 3W quiescent load, so for testing accuracy at lower loads, we needed something even less. As a result, I picked out an old Nokia linear phone charger that has an unloaded power of about 0.5W, and hooked up various resistors to it (and clipped them off one at a time, added a few in series) to achieve lower 20151224-1245-0864loads to assess the standby measurement abilities of the LP7663.





The results of the experiment will be split into sections addressing voltage, current, power and power factor accuracy. A final section addressing standby power accuracy will also be presented. As I didn’t have the equipment to test frequency accuracy, it was not tested, and likewise, the result is not likely to have been of great interest.

Voltage Accuracy


This was simple and relatively straightforward, and generally represents a good result for the meter. Below 80Vrms, the meter display became weak and the voltage measurement became “single value” even down to 55Vrms. Above 80Vrms, the error in voltage reading remained below 1% through to 250Vrms, however, the meter had a habit of underestimating the voltage by more than 1Vrms after about 210Vrms, severely degrading above 240Vrms.

As a result, the meter seems to be a good one to have if you’re travelling and you have doubts about the line voltage, as it will read it out within about 3Vrms in a majority of real-life cases.

Current Accuracy


Current accuracy was a little more complicated, and indeed, was the achilles heel of the meter. First looking at the non-distorting load, for which a continuous set of data was available, at very low currents, the error magnitude was 800%+. It was only about 350mA current that the error became much more controlled, reaching around 5%. The reason for this is that the meter indicates a current of 0.14A as its first step, so either the reading is 0.00A or 0.14A. This is very peculiar behaviour indeed. The meter itself reaches a cross-over point about 0.55A, where higher currents become under-reported, meaning high power consumption non-distorting devices above 125W are likely to see their power under-reported.

Distorting load current accuracy was limited by the minimum setting on the dimmer having either an off LED-indicator only 0.55W (which was reported as zero, hence the first error point of 100%), and minimum of about 380mA. The magnitude of the current error is positive with respect to the non-distorting load, confirming that the current readout of distorting loads generally suffers with the meter, resulting in over-reported current. As the distorting loads’ effect reduces with increasing current (as the power factor approaches 1), at full power, the error magnitudes are similar, as expected.

As a current measurement device, the LP7663 is a poor choice, especially below 350mA.

Power Accuracy


I suppose this is the most interesting graph, because most people buy these meters to measure power and not any other variable. With that respect, the power measurement accuracy can be quite variable.

For the non-distorting load, the meter’s reported value was quite accurate at the 5W end, however, the delta became increasingly negative, indicating under-reporting of power, to about 14W under-reported at 180W load. The resulting error percentage showed a relatively flat line between 6-8% error, which is not particularly good.

But what is very surprising is that when faced with a distorting load, the meter tended to over-report power. This results in a delta curve that curves upwards, and then later, curves back down as the distortion becomes less as the dimmer approaches full power. As a result, with the distorting load, we saw about 14W over-reported at a power of 60W, and a generally quite poor accuracy throughout the range with unpredictable amounts of error.

As a power meter, for non-distorting loads, it’s barely passable although the error is quite significant. When distorting loads are introduced, the power error is much more unpredictable, and this makes it a poor device to properly understand power usage. Worst of all, just with the halogen lamp, a variac and a triac dimmer, I was able to show both under-reporting and over-reporting by 14W. Other loads could be even more distorting and provoke such errors even more.

Power Factor Accuracy


This is a very odd plot, so please bare with me. As only the distorting load was able to present a wide range of power factors, its lines dominate the plot. For the distorting load, as the power factor varied between 0.4 to 0.98, the error in power factor was only about 0.1 either side, which would seem good.

However, the accuracy of the power factor determination also depends on load, so the non-distorting tests which only saw a small range of power factors close to 1 actually saw gross errors of 0.9, indicating that the reported power factor was completely bogus and not correlated with the actual power factor at all.


This is made clearer in a power factor error vs load graph, where we can see that at small loads, the power factor error is generally higher. Strangely, the power factor error with the distorting load is lower than that of the power factor error with the non-distorting load, although the curves converge from different sides as the dimmer is switched to its full range. This is similar to the power accuracy trend, and may be somehow related.

In short, the LP7663 is a horrible device for assessing the power factor of small appliances, and has a very undefined behaviour depending on the current waveform of the load.

Standby Power Accuracy


Another thing people would like to do is measure standby currents with their power meters. In brief testing in the sub-5W range, it seems that this meter’s accuracy is quite passable, although its zero blanking cuts in at about 1.8W approximately, so a 1.5672W load toggles between the actual value and zero, whereas a 1.3854 and 0.52394W load both read zero.

As standby power mandates have been in place for numerous years, most devices have standby powers of <1W and newer devices even less than 0.5W. As a result, the Lp7663 is not a meter capable of measuring standby power for most devices.


While Kill-a-Watt style power meter devices have been relatively common as part of the “green” craze, I could not help but be a little dubious about the reported values. From this systematic testing of a unit that I owned, it has been discovered that the meter has very strange tendencies and significant reading errors under certain circumstances that should be cause for concern if you are using the meter for actual power measurements and comparing small watt differences between configurations, as some other reviewers tend to do.

As a voltage meter, it’s fine. But when current is involved, gross errors start to appear. Power factor is almost next to meaningless, given the actual value ranges from 0 to 1, and deviations of 0.9 were recorded. The power value for non-distorting devices were acceptable with a nearly constant percentage error, but distorting devices create an unpredictable error which makes it hard to be confident in any reading provided by the device.

This review only tested one device, and it’s quite possible other samples of the device will exhibit different behaviour due to different calibration. Other products with different design are likely to show different performance as well, however, similar errors may indeed be present in other meters. As a result, just because the meter gives you a number, it doesn’t mean you can be confident that the number is accurate. The reference power analyzer in this case was a Tektronix PA1000, an AU$5,500 power analyzer with specified error boundaries of 0.04% reading + 0.04% range, and thus is likely to be at least an order of magnitude better than any low-cost meter, making its errors negligible in comparison and justifying the use of the analyzer figures as reference.

Bonus: Iron-Core Transformer Saturation Demonstration

I made a mention of the iron core transformer saturation problem in the previous posting about standby power, so I thought I’d just grab an example iron core transformer and subject it to some common voltages to see how the no-load power evolves over voltage.


The powers recorded were:

  • 208V – 1246mW (23.464% saving, lowest expected voltage on 230V system)
  • 220V – 1428mW (12.285% saving)
  • 230V – 1628mW (nominal voltage)
  • 240V – 1879mW (15.418% increase, Australian nominal voltage)
  • 245V – 2080mW (27.764% increase, average voltage at my house)
  • 250V – 2295mW (40.971% increase, near highest expected on 230V system)
  • 264V – 3288mW (101.966% increase, overvoltage swell)
  • 277V – 4800mW (194.840% increase, maximum overvoltage my equipment allows)

As a result, it’s clear that the supply voltage has a significant bearing on the standby power consumption of these transformers as they have very little margin against magnetic saturation. Utilities providing more voltage than is absolutely necessary actually can cause consumers to waste more power and thus consume more energy, although the amount of energy difference across a mixture of loads is still probably small, as any switchmode device will not be vulnerable to this.

Aside: A Milestone?


It’s been a long time coming, but today as I was soldering up the resistors, I realized that I was down to the last wrap on my 250g reel of Multicore 362 1.0mm 60/40 tin/lead solder, and what a journey it has been. In all, it took 11 years to get to the end of that reel, although for a while, I was using Dick Smith’s “Super Solder” instead, a 0.8mm 60/40 solder.

It’s an interesting story, as when I was younger, the Multicore reel was the first one I had bought. Unfortunately, the 1.0mm wire diameter with the thick tips on cheap irons made it hard to control, and using too much solder was a bad habit of mine. The rosin flux also left brown-yellow deposits which I was convinced by friends to be unsightly. As a result, I switched over to Dick Smith’s “Super Solder” where the 0.8mm wire made things a little easier, and the clear flux residue made things look a lot cleaner and neater.

Fast forward a few years of using the 0.8mm reel, and I rediscover the Multicore stuff. Lately, even when doing the fine work, I prefer the Multicore stuff, and the flux residue doesn’t bother me anymore. In fact, I’m appreciative that the flux works so well, and now that I use a finer tip iron and have developed better control of my “soldering hand”, the 1.0mm wire isn’t a big deal anymore even for some finer work. In fact, it saves me from having to feed a bunch of wire in at rapid pace to get a larger joint done. Funny how opinions change …

Ultimately, based on putting my used and partially used reels onto the scales, my solder usage works out to be about 49 grams per year, or a 250g reel in 5-years if I didn’t chop and change reels. Regardless, it is a milestone to see the end of a reel of solder, as it’s not a common sight to see, and it’s a form of evidence of my dedication to my hobby and the development of my skills.

About lui_gough

I'm a bit of a nut for electronics, computing, photography, radio, satellite and other technical hobbies. Click for more about me!
This entry was posted in Electronics and tagged , , , , . Bookmark the permalink.

2 Responses to Teardown, Accuracy Testing: LP7663 Kill-a-Watt-clone Power Usage Meter

  1. matson says:

    Grumble: distorting load below 380mA is distracting. I would have excluded it from Current Accuracy graph.

    When I bought my Kill A Watt, it had only golden “fours and fives” reviews in Amazon.com and Newegg.com. No perceivable quality problem. After approximately year, when I want to use it one day, it became useless. All of its numbers were wrong, such as 170 V, 4 A, 66 W with NO load attached (these are made-up illustrative numbers, I did not record what it actually reported when failed). Then, a few one-star reviews began to appear in Amazon.com, by long-term owners. My specimen was never abused: no impact, no drop of moisture, no brownouts or surges. I used it a few minutes or hours at a time (at most). I always put it away when done.

    • lui_gough says:

      Interesting. Well, for all I know, this one is just a “clone”, it’s not actually a Prodigit-made product (the guys behind the original Kill-a-Watt) as far as I know as the LCD has a slightly different layout and the internals don’t look quite the same. I think it’s likely that there might just have been bad components in the meters which don’t age gracefully (e.g. capacitors), or whatever forms the reference had degraded. At least, superficially, mine doesn’t seem to be doing anything different to when I first got it – it was never very good, and finally, I’ve been able to quantify *how* not-good.

      However, it doesn’t seem that there are any dedicated references inside the unit, and when we get mains off-peak ripple signalling, the meter does erroneously report 0.0W/0.0VA until the signal is taken off, suggesting the accuracy may also be tied in with the shape of the mains waveform. This makes the meter even less useful in my opinion, but I suppose it is interesting how they could do with so few components. Also, given the current “lumpiness”, I wonder whether the power factor indication is merely a victim in trying to ensure the power values are more sane.

      Agreed with you on the graph point, but I’m not going to change it now. I figured the readers would be able to work it out … :).

      – Gough

Error: Comment is Missing!