Com (channel working margin) is a quality factor composed of multiple measurement parameters, which is similar to the signal-to-noise ratio. Just like ENOB (significant digits), it is used to characterize the analog-to-digital converter. For com, the larger the margin, the better the channel. Because com is created according to different measurement parameters and includes model results, there are many factors that can cause it to go wrong.
Before we introduce this new observable variable, let’s review how we dealt with jitter about 15 years ago.
Compared with com, jitter seems to be very simple, that is, the difference between the signal change timing and their ideal values. It’s easy to think of this distribution as a histogram of these temporal changes.
Figure 1 shows how oscilloscopes use eye charts to show jitter. Now oscilloscopes also use histogram to show jitter.
Figure 1: oscilloscope can provide jitter distribution (source: Teledyne LeCroy).
Total jitter error
The generation of peak to peak jitters in the clock data book proved to be insufficient. The jitter from random process, mainly the phase noise from the reference clock of SerDes, will change with time; The longer the peak to peak jitter is measured, the larger the jitter is. At this time, the people who promote the standard of high-speed serial data technology follow yogi Berra’s advice: “when you come across a good opportunity, please grasp it.”
In order to avoid the peak to peak jitter which is not clearly defined and reproducible, we make a reasonable choice. We include BER (bit error rate) in a new number of definitions, which is TJ (BER) (the total jitter defined at a certain bit error rate). TJ measures the eye closure of a given BER. That is to say, if TJ (BER) is less than the bit period of the specified BER, then you have a certain jitter margin, which is no problem – this is the peak to peak measurement characteristic that people want.
That sounds cool. Can we celebrate?
Don’t worry. Because the BER we care about is very low, about 1e-12 to 1e-18, it takes a long time to measure TJ (BER), and the only equipment that can measure BER, Bert (bit error rate tester), is really expensive, and it is not so useful for other problems in the diagnostic laboratory. So we have developed many techniques that can quickly estimate TJ (BER) from the measurement results, but this leads to a big problem: you can call it group jitter. These extrapolation techniques rely on oscilloscope functions to measure multiple independent components of jitter: RJ, DJ, ISI, PJ, DCD and several other abbreviations that you may know or may have forgotten.
Different test and measurement companies have developed different methods, and their measurement results are quite different. From 2000 to 2006, the results measured by the equipment from different testing and measuring companies were at least 30% worse, and often more than 100%. These companies are the companies you know, trust and manufacture excellent equipment. Before 2004, no one could assemble a system that could tell exactly which results were right and which were wrong.
The problem comes from how we choose to integrate the results of RJ, DJ, ISI and PJ into a “package”, and then estimate TJ (BER) on this basis. By constructing TJ (BER) from the interdependent numerical packages (Figure 2), it is difficult to judge what is wrong with the measurement data entering the packages. You will find that if you change the ISI value, you will also change the RJ value. If crosstalk is added, all the measurements are in vain.
Figure 2: packages made up of interdependent variables are hard to debug, and com is such a package.
With our development from several GB / s to 10 + GB / s, ISI (inter symbol interference) has become the biggest problem. ISI is caused by the frequency response of the channel; It will change the amplitude and timing of the signal, and the degree of change depends on the order of the transmitted symbols. This problem brings us com and the possibility of history repeating itself.
Back in 2003-2004, some of us built a precision jitter transmitter at Agilent Technologies at that time, and put all the jitter analyzers provided by all the test and measurement companies together. We combined the measured precise RJ, PJ, ISI and DCD (random jitter, periodic jitter, ISI and duty cycle distortion) and made hundreds of measurements. Then… I sat quietly in the living room and began to judge the best technology. My biggest problem at that time was that there were too many ingredients in the set meal. Even if I know exactly how much of each jitter goes into each signal, there’s no way to tell why different technologies fail (and they all fail!).
In order to understand the significance of measurement, I have to do some basic scientific experiments: start with the cleanest and lowest jitter system we can build, then add one jitter at a time, compare the results, then add a second jitter, compare the results, and so on. It took me six weeks to analyze the data and figure out which technologies were accurate and why. More than two years later, the results of the test and measurement industry began to converge to within 10% – 15% of each other – the whole industry has been confused for more than five years, and at least one company has closed down because it can’t produce accurate results.
Back to com, having a single quality factor based on multiple reasons can achieve design flexibility without sacrificing interoperability. This combination of flexibility and interoperability is the pinnacle of a technical standard. No one wants a standard similar to “technological socialism,” but everyone wants it to ensure interoperability. succeed! COM。
Unless we have the same problem with jitter.
COM is equal to the ratio of signal amplitude to signal damage. This damage includes damage from signal channels, damage from all crosstalk attack channels, and damage from all other damage sources that the standard committee may think of.
Like TJ (BER), the accuracy of COM measurement depends on the accuracy of its components. Unlike TJ (BER), com measurement also depends on the model. Here is a list of COM package components:
1. Channel s parameters of the channel in question and all crosstalk attack lines
2. Package model of SerDes
3. 3dB bandwidth of receiver
4. Transmitter equalizer coefficient max / min
5. Continuous time linear equalizer (CTLE) gain
6. Peak to peak differential output voltage of victim line, near end and far end attack line
7. Level separation mismatch (for pam4 application)
8. Signal to noise ratio of transmitter
9. DFE length and coefficient limitation
11. Peak to peak DJ related to two-way Dirac model
12. Spectral density of unilateral noise
Editor in charge: GT