Research On Codec Implementations And Their Performance Optimizations Of Fountain Codes

One of the most important technologies used in modern communication systems is channel coding. The basic idea behind channel coding is to send redundant information together with a message to make the transmission more error resistant. Fountain codes are a new type of sparse graph codes and it is characterized by natural rate adaptivity. Recently, fountain codes have got much attention not only in theoretic researches but also in practical applications.In this dissertation, the author focuses on the study of LT codes with short code-length, unequal error protected LT codes and Turbo fountain codes from the perspective of overhead, which is the key index to evaluate the performance of fountain codes. The main contents and contributions of this dissertation are as follows:1. Based on the particle swarm optimization(PSO) with gradient, a new design is proposed to optimize the degree distribution in the procedure of Luby transform(LT) encoding. The concept of importance sampling is employed to construct an objective function. The estimation for the gradient is obtained by the objective function. And then, the degree distribution can be optimized by PSO with gradient. At last, forms with sparse degree distributions and soliton distributions were optimized with the proposed method respectively. Simulations show that the degree distributions obtained by the proposed method are more efficient compared with that of Robust Soliton distributions and importance sampling approach, it can reduce the overhead packets in LT codes with short code-length.2. An improvement method of unequal error protected LT codes based on block duplication is given. The operation mechanism of the unequal error protected LT codes based on block duplication is analyzed, and an improved selection strategy of information symbols is used to improve the protection of more important bits(MIB) in the local area. Degree1encoded symbols are selected from MIB, and a certain number of degree2encoded symbols are chosen from MIB and less important bits(LIB) respectively. Simulation results show that, compared with existing method, the improved method can get better performance on decoding MIB without penalizing the decoding of LIB as well as unequal recovery time(URT) property. From another perspective, the improved method can reduce the overhead under the same situation.3. An implementation method based on PLCM(Piecewise Linear Chaotic Map) is put forward to improve the performance of the Turbo-fountain on noisy channels. The proposed method employs PLCM to design the interleaver in the precode unit and the same initial value and structure of PLCM to design branch sequence generators for infinite sequence, which make the scrambling degree of information sequence as large as possible. The methods of the Turbo-fountain with PLCM and random interleaver were simulated on both the AWGN channel and Rayleigh fading channel. The simulation results show that the performance improvements of the proposed algorithms are obvious compared with the method of random interleaver for the Turbo-fountain and the proposed method, flexible and controllable, enables the performance of the Turbo-fountain to be much closer to the Shannon limits.4. A novel class of fountain codes based on modulo named Chinese Transform(CT) codes is studied. CT codes require smaller transmission overhead than LT codes, however, it cause new problem of high decoding complexity. In view of this problem, a decoding method based on improved Hopfield neural network is proposed. The original packets can be recovered according to fast inverse operations on signal packets received by improved Hopfield neural network and the problem of high decoding complexity is solved. Simulations verify that the proposed scheme is feasible and effective.