Reporting a Zero Error PRBS Test with a Meaningful BER Bound

A serial link that reports zero errors has not demonstrated a true bit error ratio of zero. It has completed a finite observation without a counted error. A useful report states how many valid bits were checked, under which conditions, and what statistical upper bound follows from the chosen model.

Define the counter’s observation

Record line rate, PRBS polynomial, generator seed or synchronization method, polarity, and checker configuration. Confirm that transmitter and receiver use compatible patterns. Distinguish raw serial errors from errors after decoding or forward error correction.

Check the measurement chain by using a supported deliberate error injection or another controlled checker test. A counter that never increments because it is disconnected from the active lane can produce an impressive but meaningless result.

Count only qualified bits

Identify whether the instrument stops counting during loss of lock, excludes synchronization intervals, or resets automatically. Preserve lock-loss events separately. Do not count wall-clock time as valid observation time when part of the interval contains no checked data.

For a continuous hypothetical 10 Gb/s test, 300 seconds contains three trillion transmitted bits. That calculation applies only if the checker actually evaluates that stream throughout the interval. Packet idle time, lane gating, or counter qualification can change the relevant denominator.

State the confidence model

The Keysight explanation of bit error ratio measurement describes the relationship between zero observed errors, tested bits, and confidence. Under an independent, constant error-probability model, the one-sided upper bound after zero errors is 1 minus alpha raised to the power of 1/N, where N is the checked bit count and confidence is 1 minus alpha.

At 95 percent confidence and large N, the bound is approximately 3/N. Three trillion qualified bits with zero errors therefore support an approximate upper bound of one part in a trillion under that model. They do not establish that errors are impossible.

Keep operating conditions attached

Record channel, cable or fixture, equalization, temperature, supply, and traffic pattern. Repeat meaningful corners rather than merging unlike conditions into one unexplained total. Bursty interference and changing channel conditions can violate the simple stationary model.

If errors occur, retain their timing and associated link events. The zero-error approximation no longer applies directly; use an appropriate interval calculation and investigate whether failures cluster.

Save checked bits, error count, lock history, confidence method, and conditions together. Present the result as a finite test with a stated bound. This lets a reviewer distinguish statistical observation from assumptions about future operation and decide whether additional patterns, corners, or longer testing are warranted.