Unsupervised Bit Error Rate Estimation Using Maximum Likelihood Kernel Methods

This paper addresses the problem of unsupervised Bit Error Rate (BER) estimation for communication systems where no prior information about transmitted bits is available at the receiver. We introduce a new technique using kernel probability density function (pdf) estimation together with Maximum Likelihood-based smoothing parameter computation. Performance is evaluated in the case of CDMA, Turbo and LDPC codes. Compared to Monte Carlo method, we show that the new technique provides good BER estimates even in the region of high SNR while it requires a few number of observations.