BER estimation from synchronous sample statistics

Summary form only given. Bit-by-bit simulation requires the capture of at least 10-100 errors to obtain statistically meaningful BER readings. Several Gigabits of user data are needed when the BER is low. By analyzing statistics of synchronous samples, a few sectors of data (or 5-50 kbits) are sufficient to predict BER with no range limit. In a PRML (partial response maximum likelihood) channel, statistics of sequence metrics at the Viterbi detection point can be derived from synchronous samples and directly correlated to the BER. We found the BER is usually over-estimated if a Gaussian shaped error histogram is assumed.