Student: Guannan Mu
Prof. Dr.-Ing. Werner Henkel
Middleton’s Class A model of impulse noise is a Gaussian mixture model with a background noise state and an impulsive one, where both are seen as Guassians with different standard deviations. The memory representing inter-arrival times and impulse durations is represented as a 2-state Markov model. The variances corresponding to the two states can be estimated using the trellis structure of the Markov model and using the Viterbi or BCJR algorithms. The state estimate in the Markov trellis is then used inside an LDPC decoder to correctly compute the intrinsic LLRs used in the decoding process. Iteratively, the LDPC decoder delivers estimates for the actual noise-free information, thereby allowing to improve the state estimate in a further Viterbi or BCJR, thereby realizing a Turbo-like iteration between state estimation and LDPC decoding.
We will implement the different components and do first performance evaluations.