Efficient bit error rate estimation for high-speed link by Bayesian model fusion
High-speed I/O link is an important component in computer systems, and estimating its bit error rate (BER) is a critical task to guarantee its performance. In this paper, we propose an efficient method to estimate BER by Bayesian Model Fusion. Its key idea is to borrow conventional extrapolated BER value as prior knowledge, and combine it with additional measurement data to 'calibrate' the BER value. This method can be viewed as an application of Bayesian Model Fusion (BMF) technique. We further propose some novel methodologies to make BMF applicable in the BER estimation case. In this way, we can sufficiently decrease the number of bits needed to estimate BER value. Several experiments demonstrate that our proposed method achieves up to 8x speed-up over direct estimation method.
