Jitter remains one of the most important specifications when measuring the goodness of a device. Compliance packages specify random jitter, deterministic jitter, time interval error, and total jitter. Some technologies specify such measurements as bounded uncorrelated jitter and J2/J9 jitter specification. A low jitter measurement can be the difference between having to do a redesign versus shipping the product.
When it comes to jitter an oscilloscope has become a powerful tool for measuring it. Most oscilloscopes which have the Windows operating system installed on them provide many tools to measure jitter. However, oscilloscopes are inherently noisy and have a jitter measurement floor (meaning you can’t measure jitter that is lower than the jitter measurement floor of the oscilloscope).
As a result, anyone who purchases an oscilloscope wants to know that they are getting the best oscilloscope for their money, especially when spending hundreds of thousands of dollars. Thus, it is important to understand the jitter specifications of the oscilloscope. The most important specification for jitter is the jitter measurement floor , which combines both oscilloscope sample clock jitter and noise floor . Oscilloscope vendors specify jitter measurement floor in different ways, therefore it is important to understand what each specification means. It's especially critical for looking at fast signals such as the 10 Gbase recovered clocks with intersymbol interference, Figure 1 .
Figure 1: 10 Gbase KR signal with significant ISI
Intrinsic jitter or sample clock jitter
Real-time oscilloscopes sample data very fast, at up to 120 GSa/s, which means keeping data points in alignment is extremely important. All sample points must be aligned in time by the oscilloscope.
To align the sample points, an oscilloscope could use a chip or a system known as the time base, which provides the necessary tight time correlation needed between the sampled input signals delivered to analog to digital converter and the internal clock. The clock itself has a jitter specification, along with how well it is able to align the points of the oscilloscope through its time base. The jitter specification of the entire oscilloscope time base is known as the sample clock jitter, or also can be called the oscilloscope’s intrinsic jitter .
Oscilloscope vendors differ in how they specify intrinsic jitter. For instance, the Agilent 90000 X-Series oscilloscope (Figure 2 ) specifies its intrinsic jitter at 150 femtoseconds (fs) as its sample clock jitter. Other vendors call this the jitter measurement floor (which will be discussed later). Intrinsic jitter is the absolute best an oscilloscope can do for its jitter measurement of time interval error. In other words, intrinsic jitter shows an absolute best-case jitter measurement for an oscilloscope.
Figure 2: Infiniium 90000 X Series oscilloscope
Jitter measurement floor
While intrinsic jitter explains what is the theoretical best-case jitter of an oscilloscope (in other words, what the jitter would be, if there were no other variables present), the specification by itself will not allow the user to know how much jitter an oscilloscope will contribute to a measurement. One of the largest contributors to oscilloscope jitter is oscilloscope noise, which is unaccounted for using only the sample clock jitter specification. Jitter measurement floor (JMF) takes into account the oscilloscope noise floor.
An example of the oscilloscope jitter measurement floor can be taken from the Agilent 90000 X-Series data sheet:
Notice that the specification has three components, the aforementioned sample clock jitter, noise, and slew rate. It is important to have all three components in the specification. Oscilloscope data sheets which only give one number, such as 200 fs, are only specifying their best-case jitter or sample clock jitter, but are not properly representing their jitter measurement floor in their data sheet.
So why are the three specifications used into a single jitter measurement floor specification ? Again, noise influences oscilloscope jitter. The slower the rise time or slew rate, the more oscilloscope noise influences the jitter contribution of the oscilloscope. While oscilloscope noise is not typically considered a horizontal specification, a slow rise time allows for noise to effect the horizontal measurement.
This phenomenon can be seen by looking at the time interval error of different sine wave frequencies. Notice that for a 20 GHz sine wave, the oscilloscope has a time interval error of 190 fs (Figure 3a ). Now the sine wave is slowed to 5 GHz and the time interval error is 400 fs (Figure 3b ). The third mage (Figure 3c) shows an entire jitter versus frequency curve of the 90000 X-Series.
Figure 3: Notice how the jitter measurement increases
as the frequency decreases;
images show frequencies of
a) 20 GHz, b) 10 GHz, and c) 2 GHz, top to bottom.
Notice that as the sine wave frequency increases to above 30 GHz, the jitter measurement floor approaches the sample clock jitter specification of 150 fs. Sine wave frequencies’ slew rates become faster as the frequency increase. At 30 GHz, the slew rate is over ten times faster than a sine wave running at 5 GHz, which means significantly less oscilloscope noise contributing to the time interval error of the measurement.
Long term jitter
Another specification which affects jitter is the long-term jitter of the oscilloscope. As an oscilloscope’s memory increases the time base must align even more sample points at extremely fast rates. This problem has been exasperated in recent years by oscilloscopes offering deeper memory.
In fact, now oscilloscopes offer up to 2 Gpts of data, which puts extreme stress on the oscilloscope’s time base. In addition to deeper memory, chip manufacturers are separating jitter on longer patterns, such as PRBS23 or PRBS31. To properly separate random jitter from deterministic jitter on a PRBS23 pattern can require up to and beyond 500 Mpts of data. At these deep memory depths an oscilloscope’s long-term jitter becoming vitally important.
Long-term jitter takes into account the effects of time-base drift that an oscilloscope will experience. A time base that was not designed for deep memory will experience significant drift and will result in very large jitter measurements.
What could be even worse is the fact that a jitter measurement taken with 2Mpts of data may show a completely different answer than one taken with 100Mpts of data–not because the device under test has any more jitter, but rather because of oscilloscope drift in the time base.
How to verify jtter
Verifying the jitter specification is actually relatively easy to do; however, each oscilloscope has some caveats that you need to consider as well.
Steps to Measure Jitter :
1: Find a very-low-jitter sine-wave source with as much, or more, bandwidth as the oscilloscope; for instance the Agilent’s E8267D has greater than 30 GHz of bandwidth and very low jitter
2: Connect the sine-wave generator to one of the inputs of the oscilloscope.
3: Start with 1 GHz on the sine-wave generator; input the sine wave into the oscilloscope
4: Turn on the time interval error measurement of the oscilloscope, and note the measurement (clock recovery can be set to the oscilloscope’s constant clock recovery setting).
This is the first measurement of the jitter measurement floor curve; we now repeat Steps 2 to 4 in either 1 GHz or 500 MHz steps to the bandwidth of the oscilloscope.
Notice that the jitter measurement floor will decrease as the bandwidth of the sine wave increases. This is because the faster the rise time, the less noise that contributes to the jitter measurement floor, Figure 4 .
Figure 4: The rise times also increase as the frequencies decrease,
which increases the oscilloscope noise contribution
in the jitter measurement; images show frequencies of
a) 20 GHz, b) 10 GHz, and c) 2 GHz, top to bottom.
If you are measuring rise times slower than 30 ps, than noise is actually the greater contributor to the jitter measurement floor than the sample clock jitter of the oscilloscope. Also, as the amplitude and offset of the sine wave are changed, the oscilloscopes will give different results. For instance, one brand of oscilloscope is very sensitive to offset changes and its jitters get significantly worse by adding offset to it. Another brand of oscilloscope is optimized with 75% of scale input, as you increase above 90% of scale the jitter measurement floor increases.
If time allows, it is a useful exercise to change variables such as amplitude, offset, and percentage of screen of input, and measure the same jitter measurement floor curve. Figure 4 also shows the jitter measurement floor of the 90000 X-Series. Notice that at sine wave frequencies of >20 GHz, it nears its intrinsic jitter of 150 fs.
Similar to the rise-time specification of an oscilloscope, if the jitter of the scope is close to the jitter of the measurement, the oscilloscope will actually contribute more jitter to your measurements. The only way to see the theoretical lowest jitter of the oscilloscope is to have significantly lower jitter on the device. For instance, if the jitter of the device is 150 fs and the scope’s lowest jitter measurement floor is 150 fs, you can expect to have 30-40% error added to the jitter measurement. This means that, at best, you would see a jitter measurement of 200 fs from the real time oscilloscope.
The jitter specification is one of the most important specifications of the oscilloscope. Simply reading a data-sheet jitter measurement of 200 fs is not enough. The oscilloscope user needs to understand what the oscilloscope vendor is specifying.
Is the vendor only specifying sample clock jitter? In most cases, vendors will specify sample-clock jitter (intrinsic jitter), as opposed to the jitter measurement floor. Ultimately it is the jitter measurement floor that determines what the true jitter of the oscilloscope. Jitter measurement floor combines sample clock jitter and noise floor, by adding a slew rate component. The jitter measurement floor varies by slew rate, so it is important to find the needed slew rate and test the jitter measurement floor at that point.
While it takes additional to time to evaluate an oscilloscope in this way, the payoff is more accurate measurements with a lower total jitter measurement, which means faster time to market on designs.
About the author
Brig Asay is Product Manager, High Performance Oscilloscopes, in the Digital Test Division – Scopes of the Electronic Measurements Group at Agilent Technologies, where he manages product planning and strategic marketing for Agilent’s high performance oscilloscope business. Brig joined Agilent Technologies in 2005 as a Technical Support Engineer. During his 5 years with Agilent, he has been a Marketing Operations Manager, where he oversaw the marketing budget and managed the technical support and learning products teams, and a Technical Support Engineer, which he helped solve numerous customer problems. Previous to Agilent, Brig worked at Micron Technologies, Inc. as a Test Engineer. Brig graduated with an MBA from Northwest Nazarene University and BS Electrical Engineering from the University of Wyoming. He is a published technical author.