UPS Transfer Time Measurement

Most offline/line interactive UPS‘ (Uninterrupted Power Supply) have a transfer time between 5 to 15 milliseconds. In this post, I will show you the results of the measured switch-over performance of my APC Back-UPS XS1500 and the technique I used.

XS 1500, according to its specifications, switches over to battery power in under 8 milliseconds in the event of a power failure. But what’s the real world performance like? We are about to find out here.

Intuitively, we wanted to observe the time interval between the mains power failure and the time when the UPS output waveform is established. For the most accurate measurement, we wanted to measure the waveform interruption duration at the output side of the UPS as the time interval measured here is what the connected equipment “sees”. So in theory, we should be able to hook up an oscilloscope to the output of the UPS and capture the switch-over event there.

But how do we capture the precise moment of this switch-over event? Without resorting to the internal circuitry of the UPS, we can use the top circuit in the schematics below to achieve this. The transformer used here serves two purposes. First, it provides isolation between the mains circuit and the oscilloscope. This is important as typically both the oscilloscope and the UPS are earth-grounded and accidentally connecting the ground clip to the live line can seriously damage the scope and/or the UPS. Second, lower voltage is much safer to work with regardless of the aforementioned grounding issue.

Measurement Setup

Measurement Setup

The lowered AC voltage is then converted to DC voltage via a bridge rectifier. The RC circuit connected after the bridge rectifier is the key here. For the given components values, the RC constant is roughly 1 millisecond. This means that during each cycle, the rectified waveform will not totally drop to zero before the next cycle begins. If we set the oscilloscope to trigger below this “residual” voltage on the falling edge we can capture the mains power event. In practice, the RC value can be chosen anywhere between 1 and 10 milliseconds.

In the bottom half of the schematics above, again a transformer is used for isolation and safety purpose. This transformer is connected to the output of the UPS. In my setup, the trigger waveform is connected to channel one and the UPS output is connected to channel two.

The following three oscilloscope captures are single-shot results from three different runs. In each image, you can clearly see the waveform changes from a sine wave (mains voltage) to a modified sine wave (UPS voltage). The transfer time is measured between the trigger cross-over point (the intersection between the orange trigger level line and the white cursor line) and the first major edge of the UPS output waveform.

As you can see, the measured transfer time is between roughly 5 and 6 millisecond which is well within the specified 8 ms figure.

Measurement 1

Measurement 1

Measurement 2

Measurement 2

Measurement 3

Measurement 3

Be Sociable, Share!


  1. Ross says:

    First of all, this is a great test! I had been curious how well my UPS performed, but never bothered to test it. Good to see that it only loses about half a period.

    However, why didn’t you align the beginning of your measurement(white cursor?) with the mains disconnection event? The triggering makes sense, but your trigger point is not the point at which mains power is lost, it is some settling time after. It looks to me like the transfer time is about double what you measured.

    • kwong says:

      Hi Ross,

      Good point! I think you are right, I should not have used the trigger point as the begin point. So the revised estimate suggests that the transfer time is at around 10ms (the spec’d figure is 8ms).

      • Debraj says:

        In India, we have inverters sold with a switch which selects — UPS mode/ inverter mode. While in UPS mode, the change over time is smaller (10mSec). I believe, all computers have capacitors which can sustain power for 20mSec and hence do not lose power during change over time.

  2. Interesting post! First, I definitely agree with Ross, I think the measurement start point is at the disconnect event, as defined by the point where the trigger waveform stops being repeated, namely the last (stunted) peak. I’d also add this, though: if you we’re asked to draw a sine wave, you’d probably start at the zero crossing. Now, this is a single steady state wave, so we don’t really care how it starts or what it’s phase is. But my point is, if you had some kind of power DAC outputting a sine wave, your lookup table would probably start at sin(0). Given that the modified sine output of the UPS is level at 0 for maybe 90 degrees, it stands to reason that you’d end the measurement at the beginning of this level 0 period rather than at the first rising or falling edge.

    • kwong says:

      Now I am really curious what the official definition of the cutover event is (e.g. what criteria APC or the industry uses for such measurements).

    • Ross says:

      I thought the same thing about the rising edge when I was looking at the first measurement, but the second and third measurements show a reduced time high pulse. This implies to me that the inverter doesn’t kick in until partway through the high pulse.

      Do you think that this delay time is a result of the power electronics having to charge a capacitor up to the high voltage output level, or do you imagine that they keep a capacitor charged all of the time?

  3. David says:

    Maybe a pair of mains-compatible fast opto-isolators would work better?

  4. Ross says:

    In the 3rd measurement, it looks like the mains was reconnected after a couple of periods. Now, this is probably just a bouncy switch, but it makes me wonder how long it takes for the UPS to switch back over to mains when it is reconnected. Also, even after just a few periods the two waves look like they might be out of phase. Is the inverter output (square wave) out of phase from the very beginning?

    I wonder if the UPS inserts some extra time at 0V before it reconnects the mains to make sure they the load doesn’t experience any really wide pulses or sudden changes in polarity. This wouldn’t be a problem for a switching power supply (most of them work fine off of DC anyway), but a 50 or 60Hz transformer might do something funky if it experienced an extra long time at a particular voltage.

    • kwong says:

      Yes, in each of the measurement I flipped connected power strip switch off and on quickly. On most UPS’ there is a setting on how long it has to wait after the mains is restored before switching back to mains power supply.

      I assume that the UPS would take this out-of-phase situation into consideration when switching back, possibly by adding a small delay so there wouldn’t be a spike in the RMS voltage.

      As I thought earlier, it would be nice to know exactly how this “transfer event” was defined at the first place.

  5. vb says:

    How did you connect this two signals to Rigol? Common ground of two probes must be short…

Leave a Reply