Interpreting Error Statistics

Interpreting Error Statistics can be difficult. It is, in fact, the reason why there are so many other forms of error location analysis, as it is difficult to extract much understanding about your system from a few numbers.

That being said, it can be very informative to simply look at the separation between burst and non-burst errors, as well as the number of burst events, to be able to reduce the number of likely candidates for your error situation. Many systems de facto have certain types of errors. These are typically either random or systematic, either small or bursty. The Error Statistics view cannot distinguish between small systematic errors and random errors (for this, you will want to look at Error Free Interval analysis, too), but it can distinguish between small errors and larger errors.

For example, if working with a radio modem where occasional single bit error occurrence is normal behavior, observing a burst-related phenomenon certainly sends you looking into the abnormal error causes. With the Error Statistics analysis, you might see that you continue to have the same small errors and that an additional error source has been added, causing burst errors. If you know that your background small error rate is a result of signal-to-noise ratio quality, you can deduce that your signal-to-noise ratio is probably not responsible for this added burst error (or else the smaller-sized error rate would increase as well). By looking at this one view, you are able to decide not to spend time on the SNR, but to look elsewhere.

The density of burst errors can also be an indicator of the type of trouble you are having. For example, if you start to get an abnormal number of burst error events, you might check the settings for your burst criteria and the average number of errors found in a burst. It is common in some systems to see that one error may propagate into another error somewhere down the line. An example of this is a Viterbi decoder. Burst errors that are sparse (such as one error followed by 10 good bits, followed by another single error), would be measured as a large burst error rate (if your minimum burst length was set to 12 or smaller). You could quickly identify this, seeing that the number of burst errors was only two times the number of burst events.