In this paper the Bit Error Rate (BER) performance of a Low-Density Parity-Check (LDPC) code operating over a Middleton’s Class A is improved by modifying the initialization step of the classic Sum-Product (SP) decoding algorithm for LDPC codes, implemented in its logarithmic form. The improvement is obtained by initializing the corresponding decoder with initialization values calculated using the pdf function of the Middleton’s Class impulsive noise channel, evaluated over channel samples for the two possible values ‘0’ and ‘1’. This procedure requires from the knowledge of channel parameters, which can be estimated by different methods. The best BER performance obtained corresponds to the initialization done with perfect knowledge of channel parameters.Estimation methods considered in this paper are the Kanemoto, Zabin - Poor and Expectation-Maximization (EM) methods. We have evaluated these estimation methods by using the estimated values of channel parameters needed to calculate the values of the Middleton’s Class A impulsive noise channel pdf, for the two possible values ‘0’ and ‘1’. We have found that none of the estimation methods utilized in its original form provided good enough channel parameters estimations, to make the corresponding decoder perform as in the case of perfect knowledge of the channel parameters.Averaging procedures have to be applied over the initial estimates obtained with the different methods to make the corresponding decoding algorithm obtain the best BER performance. Among the different methods, averaging over estimations provided by the Zabin-Poor method appears to be the best option if implementation complexity and performance are both taken into account.