 |
|||||||
 |
 |
 |
 |
 |
 |
 |
 |
BACK
Interpreting a Burst Length Histogram
 |
A Burst Length profile will show the distribution of error events of different lengths. The length of an error event is determined by the distance between the first error and the last error in a burst. |
The characterization of a burst varies depending upon your application. The analysis parameters can be adjusted to match your need. In particular, the Minimum Burst Length should be set to the smallest number of errors occurring in close proximity to one another that you want to be defined as a burst. The "close proximity" is then defined by choosing a Burst Error Free Threshold; when this number of good bits is exceeded, one error event is concluded and counting begins for the next.
For example, the mechanisms involved in errors on a fiber system might be expected to be random; therefore, anything with a burst length greater than those predicted by white noise is not random—and the definition of a burst in this case should be any two or more errors in very close proximity to one another.
- Fiber Optics: Set Minimum Burst Length to two (2). You should expect no burst errors to occur because random errors are totally independent, and the probability of getting two 10^-N independent errors right next to each other is 10^-2N.
- ECC: Set Minimum Burst Length to a length beyond which errors cannot be corrected.
 |
1. All one-bit errors. You may have random errors or systematic short errors. You cannot distinguish yet; use Error Free Interval analysis to verify. Also, see Example 4. |
 |
2. Same as Example 1. If errors are random, there is a probability of getting two-bit and three-bit errors; it is just very small. You can check the probability difference between the number of one-bit errors and the number of two-bit errors (see the note above). You may have two error sources occurring simultaneously. |
 |
3. Testing prior to ECC. Bursts less than the ECC threshold can be corrected, while those greater than the threshold cannot be. This view can give you a rough idea of how well a corrector will work. |
 |
4. Specific peak. This may indicate that you are losing a whole byte or block. Is your processing done in bytes? You will want to check for a digital processing problem. If the peak is at a recognizable number (i.e., 1632—an MPEG frame, 16—a 16:1 MUX/DeMUX word boundary), check for internal processing that could be associated with that number, like a format boundary or block size. |
 |
5. Curve. This is probably an underlying random "bursty" channel. Examples stem from truly random phenomena such as random bursty interferences from rain, snow and thin film sputtering. Such a signal is a good candidate for ECC coding. |
 |
6. Hump. The probability of errors starts to rise with the burst length. This indicates a physical problem that is random, but has a characteristic size. For example, rain in a microwave link is random, but the error burst may relate to raindrop size. Sputtering of a magnetic surface is similar; each will have a characteristic profile. You now know it is a characteristic duration causing bursts of a characteristic size. The next step is to go to Error Free Interval analysis to isolate random from systematic errors. |
 |
7. Interference. The probability of an error burst is related to time rather than to bits. Look for an outside cause—shots of interference over so many microseconds. For example, switching power supply breakthrough, electromechanical vibration, etc. Try EFI next—do the bursts happen repetitively (e.g., power supply breakthrough) or randomly (like wind gusts)? |
- If a Burst Length analysis looks interesting, use Error Free Interval analysis to isolate repetitive errors or error bursts from random ones.
