(article continued from previous page)
How do you test the accuracy of a multimeter? You measure high-quality, tight-tolerance parts!
The ultimate in high-accuracy parts are certified NIST standard calibrated components used in a temperature and humidity-controlled environment. Expect to pay between $1200 and $2000 per calibration.
Julie Research Laboratories 100 Ohm Precision Resistor Standard (#NB-102A, 0.0015% tolerance).
For the home hobbyist, we don’t need to go to those extremes. For less than $50, I put together a fairly decent multimeter test suite. By eliminating a couple of the most expensive parts, you can do so for about half the price.
I am making the assumption that the manufacturers’ quality control and specifications are good and correct. Also, I assume that the components haven’t been damaged by age, storage conditions, or shipping.
All of the parts are plugged into a solderless breadboard, mostly for convenience. Only the voltage references need to be wired together, with 1.0 µF metalized polyester and 2.2 µF tantalum capacitors for stability, and a 9V desktop power source.
Precision components and other parts for testing multimeter accuracy.
This board contains:
One of the first surprises I had during testing is that the solderless breadboard did not have any measurable effect on the values. That is, the value of each component (resistors and capacitors) stayed the same regardless of whether they were installed in a solderless breadboard, were measured in the air, or even measured on my pant leg. So, I did not find any truth to the rumor that solderless breadboard are unreliable -- at least not for these parts, at these frequencies, for these measurements, on a 3M breadboard.
For testing various frequencies and duty cycles, the following frequency divider circuit was created:
74HC590 binary counter circuit for frequency and duty cycle measurement.
The frequency divider circuit consists of:
The oscillator connects to the first 8-bit binary counter, whose output (pin 9) connects to the input of the second binary counter (pin 11). Each stage of the counter divides the signal by two. Thus, if we feed in 1228800 Hz, we get the following frequencies...
Duty cycle is how often a wave spends high versus low. A 50% duty cycle is when the wave spends as much time high as it does low.
Regardless of the duty cycle of the output from the oscillator (but assuming that the oscillator duty cycle is steady), all outputs of the 74HC590 binary counter will be exactly 50% (assuming rise time and fall time of the counter chip are equal). The more that the binary counter divides the original oscillator signal, the closer the next output gets to exactly 50% duty cycle even if the oscillator duty cycle isn’t steady! The errors start to average out, and the duty cycle variations in the original fast signal are a much smaller percentage of the lower frequency.
When we feed a 50% duty cycle clock and half of its frequency into a 74HC00 NAND chip, the NAND output will be a 75% duty cycle. That’s because the pulses will only be high 1 out of 4 times, and the 'N' (not) of the NAND then inverts that to 3/4 or 75%. Put the 75% duty cycle output into both inputs of the next NAND gate and it will invert it to 25% duty cycle.
Logic analyzer proving accurate duty cycle output.
The image above is from a digital logic analyzer that is monitoring the 9.6 kHz output (“Clock”) of the first 74HC590 chip, the 4.8 kHz output (“Divide 2”) and the resulting inverted and non-inverted results. Note that the 25% duty cycle line spends 3 periods low and 1 period high, whereas the 75% duty cycle spends 3/4 of its time high.
For the multimeter tests, we’re only concerned with the frequency of each output. However, the multiple 74HC590 chips are often used to keep a running total of pulses.
An issue that people have with a group of 74HC590 chips hooked together is that all of the bits of the internal counters are not guaranteed to change states at exactly the same time. Partly this is because the signal takes some time to pass through all of the chips connected together (in this example we only have two chips).
If one binary digit (bit) changes a little earlier than another bit, the total binary number may appear to jump around in value until all of the new bits have reached their final changed state. This can cause havoc for anything that reads the value at that exact moment.
To reduce this problem, the output pins will retain their existing values until the register pin is toggled. At that point, all of the output pins quickly change to their new values at the same time.
I’ve fed the original clock signal into both inputs of the gate on the top of the 74HC00 NAND chip. That inverts the signal and sends it into the register toggle pin. By doing so, all of the 74HC590 chips are provided with a half a cycle to get the correct value ready internally before having the outputs set. This also means that if another chip only reads the output pins when the original clock goes high, it will never read the pins when their state is changing.
Logic analyzer showing the 74HC590 register is a half cycle behind.
How long does it really take from the moment a high-speed CMOS chip is told to do something, until it actually occurs?
In the image above, notice that the Clock has gone low. That signal gets fed into the 74HC00 chip, which apparently takes 10 ns (nanoseconds) until it outputs high to the Register. The 74HC590 itself takes 10 ns until the Divide 2 output pin goes high with its new value. The datasheets for these parts specify delays of around 8-12 ns and 14 ns respectively, depending on voltage and load. So, the real world results are just about right.
Each multimeter was tested when the meter, components, and the test apparatus had been at room temperature for at least 24 hours. The test probes included with the meter were used. (For the fun of it, I tried switching probes around without any change in results.)
The meters were tested in random order, and all tests per meter were completed in a single session of about 5 minutes. The meters were allowed to settle to value.
Random portions of tests of random meters were then retested days later to determine if there was any change in value (repeatability of the tests). Also, all tests with relatively inaccurate results were retried to give the meter the benefit of the doubt and to eliminate human transcription errors. The results were nearly identical, with only the last digit being different by 1 or 2 in the worst cases.
All of the meters were purchased by me -- none were gifts, samples, or loans. Some of the meters I’ve had for years; others were new purchases. Because I only obtained a single copy of a model, it is possible that the meter I received was better or worse than the average meter, either due to manufacturing differences, age, storage, or shipping. That is, I could have gotten a lemon or a rare gem.
All of the meters appear undamaged. Fresh alkaline batteries were installed in all of the meters with replaceable standard consumer batteries. Meters with 12 V or coin cell batteries were used in the state in which they were received.
What do you predict will be the accuracy of cheap multimeters versus more expensive meters?