Interpreting a Block Error Histogram

Block Error Analysis is helpful for studying system behavior in systems that will process data in blocks. Examples include MPEG packet transport, SONET frames and disk sectors. It is common to set a defined limit to the number of errors that can exist in a block for that block to be considered good. At other stages in the development, it is important just to understand the expected number of errors in each block.

The user defines a block size (e.g., 1024, MPEG packet size, satellite MPEG plus ECC, etc.) and the analyzer divides up time into these units. It counts the number of errors occurring within that time, independent of their location, and displays the number as occurrences in bins.

This distribution shows the typical number of errors per block. This often creates a signature, which can be monitored day-to-day in manufacturing to look for deviations.

The shaded region has more than the acceptable number of errors per block. The threshold here is arbitrary and depends on your system. Block analysis like this is applicable in manufacturing and Quality Assurance as a "go/no-go" test. Below a certain threshold of errors per packet, the packet is still usable.

ECC definition can also benefit from this type of analysis. An error correction architecture might include random error correction strategies that improve the general background error rate, some burst error correction capability to be able to withstand rare larger error occurrences, and some network protocol to support error checking and retransmission, in the ultimate case where errors do get through. In each of these cases, the codewords are designed to protect blocks of various sizes and these approaches can fail depending on the number of errors that actually do occur inside these blocks.

Block error analysis is often coupled with symbol-mode processing to allow looking for the number of symbol errors that occur in blocks. For example, Reed-Solomon error correctors work on eight-bit symbols. It is not important how many bits are wrong in a symbol. If any bit is wrong in a symbol, the entire symbol will need to be corrected. For this type of system, symbol error statistics with symbol sizes set to eight bits are appropriate.

As in the other forms of BitAlyzer error location analysis, the shape of the distribution can give you important clues as to what is happening deep inside your system that is causing errors. For example, a Block Error distribution might be smooth or might have some discrete peaks. A smooth error distribution may be the result of general random error; however, discrete entries cannot be explained this way. When you have discrete spikes, look into your system and isolate what components might cause the number of errors where the spike occurs. You can also couple this analysis with a correlation analysis, where the correlation length is set to the same block length. In this way, you might find that common error occurrence also happens at one particular location within the block.