Curve Fitting for Error Rate Data

摘要

The problem of fitting experimental or simulated error rate data with analytical curves is analyzed. The comparison is made between the maximum likelihood (ML) fitting and widely used minimization of squared error in log domain (Log-MSQ). It is shown that ML fitting demonstrates lower sensitivity to a particular sample realization of error events, and also allows to take into account different reliability of different BER measurements that is often the practical situation. Computer simulations confirmed superiority of ML fitting over Log-MSQ for various error rate measurement scenarios.