| Summary: | 碩士 === 國立交通大學 === 電信工程研究所 === 102 === Low density parity check convolutional codes (LDPC-CCs) are usually defined as the null space of their syndrome former matrices. With a certain constraint on the syndrome former matrices, LDPC-CCs can be fast encoded by syndrome former encoders. When encoding, zero padding and tail biting are two often-used schemes for trellis termination.
There are three topics for investigation in this thesis. Two methods used to improve the performance of the terminated LDPC-CCs and tail biting LDPCCCs are introduced in the first two topics. For the tail biting LDPC-CCs, masking technique is used to improve the bit-error-rate (BER) performance. From the viewpoint of trapping sets, the performance improvement is then discussed. The variations on encoder and decoder of the tail biting LDPC-CCs corresponding to the masking technique are also proposed. For the terminated
LDPC-CCs, by analyzing the structure of its syndrome former encoder, wesuggest to avoid sending some known bits when codewords are transmitted, thus alleviating the possible rate-loss. Simulation results show that the codes utilizing these two methods have much better performance. In the last topic, we show that syndrome former matrices violating the constraint, which make fast encoding feasible, are not necessarily with poor BER performance. Thus we relax the constraint on syndrome former matrices of LDPC-CCs. Under
the relaxed constraint, the encoding method is then presented.
|
|---|
