Investigation: How Accurate is that 3.5 digit Multimeter?

A few weeks ago, one of the commenters made a suggestion that I test some more inexpensive test equipment, after my stint at reverse engineering USB charger doctors/detectors and determining their accuracy. While I won’t go chasing low end test equipment, at the time, I reasoned that testing them would be a waste of time.

But in the end, curiosity got the better of me, so I decided I would give a pile of 3.5 digit multimeters a test and see just how accurate they were. These sorts of meters are the ones which are now very affordable and turn up almost everywhere – they can even be had for under $5 now, with probe leads and batteries, with more deluxe units featuring more than the “basic” 19-ranges (aka DT830 and variants) and claims of better accuracy.

Seeing as I had some test equipment, I decided to test them against the Keithley Model 2110 5.5 digit digital multimeter as a reference. After almost two weeks of work, I’ve managed to come up with some results! But first, a few notes and important points.

Getting a Reference

In general, it’s considered bad form to use a multimeter to calibrate another multimeter. Part of the reason for this is that a proper calibration normally uses sources with a very high level of precision and stability. Utilizing other sources and cross-referencing readings can be vulnerable to errors introduced by the reference multimeter, its resolution/accuracy, loading on the source, instability in the sources, etc.

Unfortunately, high accuracy calibration sources are not one thing I possess. That being said, if the reference multimeter is sufficiently accurate and in calibration, it could be possible to get a reasonable reference reading by which to judge your other multimeters on. In order for this to happen, you would expect that your reference multimeter needs to have more digits than the meters to be calibrated, and the error ranges to be less than the reading gradations on your meter to be tested. In order to understand this, you must consult your datasheet.

In our case, I have chosen my highest accuracy instrument – the Keithley Model 2110 to serve as the reference measurement. Its 5.5 digit display surpasses the 3.5 digit display on the meters to be tested by two digits. I also performed some comparisons between its stated error in reading versus the step sizes on the 3.5 digit multimeters we will be testing.

Reading-Error

This table itself is compiled from the errors as a % of reading + % of range. Only the values at the zero and full-scale readings are presented, however further computation can provide more insight. A careful examination of the table will show that the Keithley 2110’s readout error is almost small enough to be negligible at the 3.5 digit level. For example, for the 3.5 digit multimeter, it will read up to 2V in 1mV steps, but for a 2V signal, we will be in the 10V range with 0.44mV of error (i.e. less than one display unit). In fact, for voltage and resistance measurements, this appears to be the case.

However, this is not the case for current, where for 200mA of input, the readout error on the Keithley is 0.5mA which is 5 units. At 10A, the error at full scale reaches 3 units. Therefore, we must be a little careful in our interpretation of the test results.

In order to give this experiment the best chance of success, the multimeters were all kept in my room for days, temperature controlled by an air conditioner at 23 degrees C. The Keithley 2110 has not been switched off in weeks, and has been in thermal equilibrium, thus avoiding drift. It’s important to note that the above error margins are typically conservative, and actual measurement units may perform better than specified in the data sheet – so we might be able to ignore the source meter’s influence in the results after all.

Another complication is the granularity of readings. We are comparing a 5.5 digit result to a 3.5 digit result, and due to rounding, you will expect to see errors of up to one decimal place of the 3.5 digit result.

SIM-Del

As a result, we can sometimes see this kind of sawtooth pattern. This was generated by producing a sequence of numbers (1, 2.002, 3.004, 4.008, etc) and rounding that to 2 decimal places and finding the difference between the “actual” and rounded values. Basically, because the input value might sometimes lie right on the reported result (i.e. zero error), or it might lie on the boundary of the reported result (i.e. maximum error), the results have to be seen as closely-spaced spot checks. The actual maximum error would be the “furthest” distance from the zero difference (delta) result.

SIM-PC

However, if we plot this as a percentage error, we can see that we get a decaying sawtooth shape. This is because as the input value gets larger, the rounding error still remains the same size, thus the rounding error represents a smaller fraction of the entire value. This is partly why meter errors are rated with a % of reading (for linearity issues) and % of range (for offset/rounding issues).

At times, the meters struggle to keep stable display values, often “dropping” by 2-3 counts. I have chosen the value that the meter rests at after 15 seconds of measurement as the “final” value.

So do keep these points in mind when you look at the graphs.

The Contenders

The contenders in this round are a set of multimeters which have been lying around my room getting occasional use. These things just seem to collect, and are of various ages. Given that these units never came with calibration certificates (and you wouldn’t expect them to for the small money you pay for them), I’m not sure how reliable their values are. Many of the original spec sheets are lost, but some of them claimed 0.5% accuracy +/- 3-5 digits if I recall correctly. This is not really going to be an article about which meter to buy, as calibrations do drift over time and can be impacted by environment, transit and care. All meters have been treated with care, however. It’s more of a look at what you can expect if you “pick up” a random meter and make a measurement.

DSC_9505Dick Smith Q1459

The oldest meter in the bunch (over 8 years), this one comes with a red rubberized skin, data hold functionality and backlight. It doesn’t feature any transistor test options, but does have an audible beeper for continuity testing and a square wave output. It does differ from some other “19-range” meters by having a 200M ohm range, but sacrifices a 2M ohm range for it. How peculiar.

Further research seems to show that this is a rebranded Uni-Trend UT33D as confirmed by the imagery and rear serial number label format.

DSC_9507Unbranded DT-830B

Arguably one of the more popular meters on the market today for beginning hobbyists, these units have no branding whatsoever and can differ subtly in the quality of the casing and the colour. This one, with the B suffix, measures resistance to 2M ohm, and has a transistor tester but no continuity beeper. This one is a fairly recent purchase, from two years back.

 

 

DSC_9509Excel DT9205A

Another Chinese special, this one is a more deluxe meter of a different sort, featuring capacitance test (but with limited range), wider resistance scale, continuity with beeper, transistor test, DC and AC current with wider range and a data hold facility.

This one comes with a yellow rubberized skin, and was about twice the price of the DT830’s. It features a large display but still only the average 3.5 digits.

 

 

DSC_9508Unbranded DT-830D

Another variant of the same DT830 recipe, this one features square wave output and continuity beeper with the diode test moved to the 2k ohm range. Otherwise, it’s pretty similar to the yellow unit, just a bit older at around eight years old.

 

 

 

 

DSC_9506Jaycar Digitech QM1540

A more deluxe meter, this one is technically on loan to me, featuring capacitance and inductance test, AC and DC current and high resistance ranges with temperature and transistor gain available via external plug-in dongle. It has a data hold feature as well, but just the regular 3.5 digits. This unit is about 1 year old.

On further research, it was determined that this is an rebranded version of the Mastech MS8269.

 

 

The Results

As much as I would have wanted to test all the ranges and all the values – equipment limitations, time and patience are limited. In order to undertake these measurements, special purpose devices and improvisations were made. However, I have covered the most useful ranges on the meters, by hand-measuring 6976 measurements – a painstaking operation that took more than a whole week of spare time.

Resistance – Methodology

Testing for resistance can be a bit tricky. The first thing you need is a large set of resistances – now unfortunately I don’t own any precision decade boxes, so I sorted out all my 1/4W carbon film resistors (mostly) and soldered them to some Veroboard as a reference.

DSC_9592

They mostly go ascending in resistance although I do have one resistor out of place (see if you can spot it). Unfortunately, I don’t have every E12 value in there, but that’s just life. Knowing that carbon film resistors can be affected by temperature and humidity, the air conditioning took as best care of that as possible. Using just one set of test leads (so as to cancel out the lead resistance impact), I measured the set with the Keithley, and then measured the set with each of the meters. I then determined the absolute difference in readings (this includes rounding errors – i.e. sub 1-display unit errors) and also the error in percentage (absolute value – no negative values).

Resistance – 200 Ohm Range

R-Del-200R

At the 200 ohm range, measurements of the resistors showed resistance values which were within about 9-digits of the Keithley value, with the multimeters over-reporting the resistance slightly. This wasn’t the case with the DT9205A which was abysmal with accuracy being a whole 30-digits out near 150 ohms! Yikes!

R-PC-200R

As a percentage, we can see that the errors are astronomical for small resistance values, only dipping below 3% above 50 ohms or thereabouts. Most of the meters except the DT9205A managed to be below 1% of error by 56 ohms.

Resistance – 2kohm Range

R-Del-2kR

Performance was more consistent in general for values in the 2k ohm range. The DT9205A and DT-830B under-reported by up to 5 and 6 digits respectively, whereas the black DT-830D over-reported by about 6 digits. The remaining two units were pretty much within 3 digits, which is a good result.

R-PC-2kR

As a percentage, it can be seen the majority of readings seem to lie within 0.5% of the Keithley value with the exception of a spike in both DT-830’s and the DT9205A overall.

Resistance – 20kohm Range

R-Del-20kR

In the 20kohm range, the DT9205A remains the worst, being up to 11 digits out. The other units seemed to be within 5 digits, increasing in error with higher resistance values.

R-PC-20kR

As a percentage, however, most of the meters were better than 0.5% of error, and mostly better than 0.4% as well, with the exception of the DT9205A.

Resistance – 200kohm Range

R-Del-200kR

In the 200kohm range, different trends seem to emerge. While the DT9205A has been the outlier so far, its performance is now more similar to the DT-830B, under-reporting resistances by anywhere from 1-8 digits. The QM1548 didn’t do too badly with the error increasing steadily as the resistance value increased to just over 4 digits. The other meters were within 3 digits.

R-PC-200kR

As a percentage, we can see all meters except the DT9205A and DT-830B managed to cling to or stay below the 0.5% error line.

Resistance – 2Mohm Range

R-Del-2MR

The trend seems to change for the 2M range with most meters under-reporting the resistance values. The DT9205A retains its habit of being least accurate for the majority of the range, however all other meters except the DSE Q1459 “tangle” together in that sort of trend pointing towards a maximum error of about 11 digits. The DSE Q1459 is a little opposite, peaking out at 7 digits over-reported.

R-PC-2MR

As for percentage error, the stats look a little worse now, with errors hovering slightly higher near 0.7%, with the exception of the DT9205A of course.

Resistance – 20Mohm Range

R-Del-20MR

The 20M ohm range is a bit of a “special” class which the DT-830’s aren’t a part of. In all cases, it seems the trend seems to be the same – with the meters being anywhere from 7-10 digits out at ~8Mohms, but otherwise being within ~3 digits for smaller resistances.

R-PC-20MR

In general, in terms of percentage error, the meters performed well for the smaller Mohm resistances, rapidly degrading at ~8Mohm. The DT9205A retains its characteristic problems with accuracy.

Resistance – Conclusion

From a quick examination, I’d have to say that the DSE Q1459 did a good job over all the ranges compared with the others. Its errors seemed more controlled throughout. The standout dud in this is the Excel DT9205A which consistently missed the mark almost every single time. The rest of the meters seemed to have varying trends depending on the range – some better, some worse. At times, errors at extremes of scale were noted at 10 digits, in other words, rendering the last digit on the screen technically meaningless.

DC Voltage – Methodology

To test the error in DC voltage, my Manson HCS-3102 switching mode power supply was used to supply power from 0.8 – 36.2V in 0.1V steps. The actual voltage was recorded with the Keithley 2110 in parallel with the reading shown on the multimeter – thus paired readings were recorded. This was due to the inherent voltage “drift” of the switch mode controlling circuitry. Sawtooth pattens due to the granularity in the steps of the power supply are evident, as explained in the earlier section.

DC Voltage – 2V Range

DCV-Del-2V

Testing below 0.8v was not possible as the power supply was not able to produce such low voltages. It seems that each meter has its own characteristics. The DT9205A was over-reporting by about 5-digits, with the DT-830B under-reporting by about 7-digits at the extreme. There was one spike in readings experienced by both the DT-830B and QM1548. Overall, the DSE Q1459 remained remarkably close to the actual reading (within 2 units).

DCV-PC-2V

The accuracy of the DSE is clearly shown in the percentage error graph with it near or below 0.1% – an exemplary result. Aside from the spike, all were below about 0.4% which is reasonably close to what specification sheets may tell you.

DC Voltage – 20V Range

DCV-Del-20V

This test, spanning the entire 2-20V range really makes for an interesting graph. The curves are due to the fact the power supply voltage isn’t incremented in exact multiples, whereas the “steps” up and down are due to the granularity of the 3.5 digit display.

Of all of them, the DT-830B is the worst, hitting a whopping 14 digits of error through the range. Everything else remained within 5 digits, however, interestingly, the DSE that performed so well in the last test is now the second worst. The DT9205A isn’t half bad.

DCV-PC-20V

As a percentage, we can see that at the low end of the range, the DT9205A struggles, and the DT-830B struggles across the board. The rest of the meters seem to meet their 0.5% spec and are under 0.3% for the top half of the scale.

DC Voltage – 200V Range

DCV-Del-200V

Unfortunately, we can’t get too far with voltage with the Manson power supply I have here, and I wasn’t going to bother jury rigging a rectifier and capacitor filter to a variac output to go further. Regardless, it shows that at the low end of the 200V scale, it seemed all meters were within 2 digits, and all seemed to under-report the voltage slightly.

DCV-PC-200V

The magnitude of the error starts off higher, as we’re in the low end of the scale, but settles down to sit under 0.5% as expected.

DC Voltage – Conclusion

It seems the meters themselves aren’t too bad at measuring DC voltage. That being said, the yellow DT-830B doesn’t seem to be in good calibration, and while the DSE Q1459 appeared good at the 2V scale, it wasn’t anywhere near as accurate on the 20V scale. The DT9205A didn’t fare too well, but I suppose it wasn’t too bad either. Jaycar’s Digitech QM1548 seems to have performed decently through most of the tests, although having a spike in the 2V range.

AC Voltage – Methodology

In order to measure AC voltage, a 230v 50hz true sine wave sourced from an HIP-300 inverter and fed through a 1A variac to produce voltages between 1-260V. This was measured by the Keithley and meter under test in parallel. The paired readings were taken at every nudge of the variac, with no regards for having a particular number of steps.

AC Voltage – 200V and 750V Scale Combined

ACV-Del

When reading the graph, please note that the 750V range has units which are 10 grid lines tall. It seems that the meters generally had trouble with accurate DC measurements – the DT9205A surprises again with 8 digits of deviation, and the DT-830B having about 11 digits. All meters had difficulty keeping the error within 5 digits on the 200V range. When switched to the 750V range, only the lower end was tested, but generally error didn’t exceed 4 digits within the tested area.

ACV-PC-Full

The impact of this can be seen as a whopping large value of error at low readings below about 50V. The DSE, DT-830D and DT-830B are all offenders when it comes to error.

ACV-PC-Enlarged

Zoomed in, we can see the other units were starting at lower values of error, but at higher values they were all able to meet about 0.5% error except for the DT-830B. At the higher 750V range, the DT9205A had a higher amount of errors than the others, and the DSE seemed to perform very well with errors below 0.3% (a complete reversal of its performance in the 200V range).

AC Voltage – Conclusion

At a glance, all meters weren’t terribly accurate when it came to measuring AC Voltage. However, some were worse than others – the DT830’s and the DSE being poor at low voltage in the 200V range, and the DT9205A being poor in the upper 200V range and 750V range. However, into the 750V range, the DSE managed to reverse its behaviour. On the whole, it seems Jaycar’s Digitech QM1548 was most consistent when it came to AC voltage.

DC Current – Methodology

Testing DC current was performed by having the Keithley and device under test in a series circuit with a load. For testing the 200mA range, a load comprised of 12×560 ohm resistors in parallel was used. For testing the 10A range, a 50W MR16 halogen downlight globe and the Manson HCS-3102 power supply was used. Unfortunately, this limited the readings to about 4-5A.

DSC_9503

DC Current – 20 and 200mA range

LDCI-Del

Due to minimum voltage and load construction, testing below about 10mA was not possible. In the brief moment tested in the 20mA range, it was noted that the DSE performed the worst of the lot, showing an over-reporting of about 4-digits while other units remained well within 2 units. The DT9205A surprisingly put in a good result.

On the 200mA range, the DSE managed to perform poorly again, but in the opposite direction, now under-reporting current by about 9 digits at full scale. The worst result was about 12 digits over-reported by the DT9205A, closely followed by the DT-830D with over-reporting by 8 digits.

Surprisingly, the cheap yellow DT-830B was most consistent throughout these tests despite performingly poorly in other tests (ACV, DCV) previously. A break is seen in the data points due to an automatic range change by the Keithley Model 2110 – this accompanies a change in burden resistance which changes the current flow step-wise.

LDCI-PC

In terms of error percentage, in the 200mA range, close to all readings were under 0.7% with the better meters staying mostly under 0.5%. The 20mA range saw mixed performance from most, but the DT9205A produced an exemplary under 0.2% error result.

DC Current – 10A Range

HDCI-Del

Testing for high 10A DC current range produced interesting results. It seems the DSE Q1459 is not suited to continuous current measurement as written on the front panel of the meter – the results clearly indicate why. The meter heated up significantly, which is likely to have affected the resistance of the hunt causing the error to show a very non-linear shape.

The other meters seemed to perform better – although the DT9205A retains its signature of being an inaccurate meter. It managed to deviate by a whole 24 digits by 4.5A. The other meters were able to keep it within about 4 digits.

HDCI-PC

Still, the numbers reflect the difficulty with accurate current measurements, with the better meters struggling to keep it under 1.5% of error. The DT9205A and DSE Q1459 need no consideration at all.

DC Current – Conclusion

Surprisingly, the stand out performer is the yellow DT-830B again, which failed abymsally at voltage but seems great at current accuracy. The Jaycar QM1548 and DT-830D both performed fairly well as well. Unfortunately, the DSE was inconsistent, and performed poorly at high currents, whereas the DT9205A was poor all round.

What are you Calibrating?

So maybe those results weren’t what you hoped for, and you’re thinking there must be some way to improve that multimeter! Well, maybe you’re in luck, because depending on your meter, there maybe one or (sometimes) many trimpots to play with.

DSC_9510 DSC_9512

DSC_9511Unfortunately, as it’s not labelled, who knows what setting you’re really adjusting. Worse still, the trimpots aren’t exactly precision units either, and it’s unlikely that even with the best handling and care, that they will retain their precise values for very long.

Other times, you might be unlucky and come across something like my black DT830 which has no trimpot installed at all. I wonder how that works – calibration free, fixed calibration courtesy of surface mount resistor?

Whatever it is, I don’t really think it’s a great idea to touch it unless you have some decent equipment, patience and time to check just how well it is performing. From the data presented above, it seems very unlikely that changing the calibration will fix reading errors for all ranges, as the errors don’t seem to even be in the same direction for all ranges. I wonder whether these units were even properly calibrated at the factory, or whether they just did a one-voltage spot check on the calibration and fine tuned it for that.

Conclusion

Well, it was a fairly involved investigation, which took a fair amount of time, but rather surprisingly I suppose it proves some of the things I’ve anecdotally heard about the 3.5 digit multimeters – namely that the last digit is pretty useless on many of them. Judging from the amount of error, this could well be true depending on which unit you have. It’s also interesting to see the errors in play – showing “curved” delta plots which suggest a linearity problem, others showing offset problems, and without consistency between ranges.

One must really wonder what they are calibrating if they choose to change the values of the trimpot inside their meter – it could well be point calibrated for one particular voltage, but the other ranges could still be incorrect. This probably comes down to the choice of components (e.g. resistors) used inside, not being of high precision. As a result, it’s probably pointless to even try calibrating one of these units.

I suppose this does tell you just how indicative the readings are – it’s fine for telling the difference between 9V and 12V, but it’s not able to tell whether the absolute value of a voltage is 5.00V or 5.05V. At the low end of the range, the errors can be quite high, and thus readings should be avoided there (as well as for small resistances <50 ohms).

It’s all hit and miss really – some meters are good at certain ranges or types of readings, and terrible at others. However, in general, it seems the DT9205A takes the wooden spoon. I suppose that’s why quality multimeters exist, and are more expensive while being also better specified.

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 Investigation: How Accurate is that 3.5 digit Multimeter?

  1. Pingback: Teardown, Accuracy Testing: LP7663 Kill-a-Watt-clone Power Usage Meter | Gough's Tech Zone

  2. Andy Holmes says:

    Thanks so much for this article, many of these of low-cost multimeters are still available and your extensive (and no doubt gruelling) tests and charts were exactly the sort of review I was hoping for: tentatively conclusive with lots of data to make my own decision based on my needs. I’ll certainly be spending more time on your site as a result.

Error: Comment is Missing!