AnsweredAssumed Answered

Improving Infiniium's phase measurement accuracy with EZJIT

Question asked by jtmccabe Employee on Aug 2, 2007
Sometimes users of real-time oscilloscopes want to measure the phase of two high frequency signals with high accuracy.  One common application is measuring the time that it takes two signals to come into phase lock.  Let's assume that we want to use a DSO80204B to confirm that two 1.5 GHz sine waves are phase locked within 1° within one second of some event. 

One potential problem with making phase measurements on high frequency signals is that the accuracy of a single measurement is relatively low.  The phase is calculated as: 

360°*[(time between edges)/(period of source 1)]

Both the numerator and denominator of this equation involve a delta-time measurement.  For the DSO80204B, the accuracy of a delta-time measurement is approximately ±7 ps peak.  The worst-case peak-to-peak accuracy would then be, in theory, as high as about ±4°.  That's quite a bit off from our requirement of ±1°. 

Although each measurement is not very accurate on its own, the scope has the opportunity to make many, many measurements and can average together sets of them to greatly reduce the effects of random noise, thereby improving the accuracy.  For example, you could capture the waveforms over a period of 200 us.  A 1.5 GHz signal would go through 300k cycles over that time period, and the scope could make 300k phase measurements. 

The screen shots below show the E2681A EZJIT software plotting a measurement trend of phase versus time.  A simple 1.5 GHz sine wave was fed into both channels.  One channel had a longer path than the other; hence the constant phase difference.  Notice that the "no smoothing" image shows approximately 8° of peak-to-peak "noise" on the phase measurement trend (third waveform from the top). 

However, the EZJIT software has an optional "smoothing" feature, which averages together a user-specified number of points.  The other screen shots illustrate how smoothing cleans up the measurement trend and can give you a pretty good idea of what the actual phase is over the 200-us interval.  As you can see from the "1k points" and "10k points" examples, there isn't much more than about 0.5° of uncertainty at any point in time.  You can trade off the number of points for response time of the plot. 

No smoothing:

Smoothing, 100 points:

Smoothing, 1k points:

Smoothing, 10k points:

Ideally, our scope would have enough acquisition memory to capture the entire time period over which we want to know the phase.  However, in this case, we would need (1 second)*(10 GSa/s) = 10 GSa or 10 GB of acquisition memory, which is well beyond the capability of all current oscilloscopes.  Instead, we can simply trigger on an edge that occurs at time zero and dial in one second of horizontal delay.  We will only be capturing the 200-us window that occurs between time 0.999900 and 1.000100 seconds, but we will be able to determine the phase with high accuracy where it really matters -- after one second. 

Note that it is also possible to improve phase measurement accuracy by using averaging.  You could average together multiple waveforms and then measure the phase.  Or you could turn on the phase measurement, let the scope run for a long period of time, and then look at the average value of the measurement.  These are useful techniques when the phase of the signals is relatively constant.  However, for applications such as measuring phase-lock time, EZJIT's measurement trend capability is far superior because there is no between-trigger "dead time" involved, and all cycles of the waveform contribute to the measured value.