When I used the 5244D MSO to test the G-pole waveform of the SMPS MOS-FET, I found that the square wave frequency with variable duty cycle displayed in the measurement toolbar is different from the result of the ruler test. Obviously, the value in the frequency measurement column is incorrect. What is the reason?
Is my measurement method wrong, or my understanding is wrong?
The Automatic Frequency Measurement from the Measurement statistics tool is not wrong, it's just not what you were expecting. The problem happens to be that you're applying the measurement to a corrupted waveform. Let me explain exactly what I mean by that.
The Automatic Frequency Measurement is calculated by defining 2 points (an upper and lower point) through which the edges of a waveform pass through, to be able to decide what represents the leading and trailing edges of the waveform. It then uses these edges to calculate the frequency of the waveform. Ordinarily, there should be no problem with doing that for your pulse waveform, if the edges are clearly able to be identified, but they are not. In the example below I've zoomed into the edge of the waveform right in the centre of the display.
You can see that there is a large fluctuation in signal level which isn't even noise, because it's much larger than the actual noise that you can see at the top of your waveform. So this burst of modulation in the signal is right at the point where the Frequency calculating algorithm is trying to identify pulse edges, so naturally it assumes that these are edges of a signal, as you can see in the example below, where I've placed rulers around the fluctuations so that the tool will only calculate the frequency of the fluctuations.
This happens to be the only edge that is affected in this way as the falling edges are relatively clean and the other rising edges have a slight bump that shouldn't cause too much of a problem (as you can see in the example below)
In effect then, what the Automatic Frequency Measurement tool is doing is finding different waveforms that it's calculating the frequencies for, and then displaying the average of those waveforms, because it can't know which values are going to be most important for you (remember that I had to zoom in to isolate the high frequency waveform using the rulers). Normally you could get around this problem by just placing the rulers around a minimum of 3 other edges (2 other rising or falling edges representing just over 1 period of the waveform) and do the measurement between the rulers. However, within the rulers, all of the edges need to be free from fluctuations that could corrupt the measurement. You happen to have the unfortunate situation whereby there are no 3 edges that you can pick that don't include that corrupted edge (as can be seen from using the falling edge measurement below)
Only you would know if these fluctuations are going to be a problem in your Switched Mode, or whatever it's powering but, from a perspective of getting an accurate measurement of frequency, you should choose a longer Timebase to reduce the likelihood of not having 3 other edges to pick for a 'between rulers' measurement, in future. (Note that you can sometimes, get away with creating a Math channel to filter out the fluctuations, but I had mixed results with your waveform because the fluctuations are so large.)