Research On Technology Of Error-Tolerant Recognition Of Channel Coding Parameters

Channel coding parameter recognition means to recover the coding parameters by inversely analyzing the received coding sequences.The purpose is to provide necessary parameters for the decoding procedure and thus to recover the information,under the situations where the coding parameters are not known to the receiver.This technology is essential in applications including cognitive radio and signal intercepting.Because of the never-stopping developments in communication schemes and signal processing techniques,the signal transmitting power necessary for reliable communication is becoming lower and lower,which leads to the result that the received signals are becoming weaker and weaker;and in signal intercepting,the intercepted signals are often weak ones.Therefore,the low signal-to-noise ratio property of the received sequences demands that the recognition methods have strong error-tolerant capability.So this thesis is devoted to design recognition methods with better error-tolerant performances,or to reduce the recognition computation complexity while keeping the error-tolerant ability not degraded.The thesis introduces basic knowledge in the introduction chapter.Backgrounds of the study are introduced,including application prospects and practical demands.Some necessary channel coding theory is recalled for use in subsequent chapters.State-of-the-art of this topic is presented in detail respectively for different classes of channel codes.Shortcomings of existing research works are then analyzed and summarized.Main research contents are decided according to a deep look into the research status and practical demands,which also yields the basic research idea of the thesis,that is,to use the soft outputs of the demodulator and to exploit the parity-check relations and soft-input soft-output(SISO)decoding algorithms of channel codes.For recognition of channel codes within a candidate set,both the error-tolerant ability and the complexity performances are important in practice.The existing method is based on the average log-likelihood ratio(LLR)of parity-check relations,for which the setting of certain thresholds remains to be an unsolved problem and the complexity is still reducible.The concept of likelihood difference(LD)is developed in this thesis to replace the LLR.The new method based on average LD of parity-check relations has notably lower complexity.Using Gaussian distribution theory,the theoretical probability of correct recognition is derived,and the thresholds are provided explicitly.Besides,for convolutional codes,a new method based on the average absolute value of the outputs of BCJR decoding is proposed.Simulation results show that this new method improves the error-tolerant capability by 1~3dB compared to the average LLR method,but this is at the cost of higher computational complexity for convolutional code recognition.For convolutional code recognition without a candidate set,although existing methods have already achieved rather good error-tolerant performances,there is still higher request in practice since convolutional codes are very widely used and are used as sub-codes of Shannon-limitapproaching turbo codes.Several important existing methods are introduced and the advantage of exhaustive searching methods is pointed out.Through the derivation of theoretical probability of correct recognition of the exhaustive methods,the problem existing in soft decision data based exhaustive methods is found out and a solution based on the least square(LS)cost function is thus proposed.Theoretical analysis shows that,while keeping the complexity to be of the same order to existing methods,the LS method improves the error-tolerant capability by about 1d B.And simulation experiments strongly prove the theoretical results.Turbo codes,consisting of convolutional codes and block random interleavers,are of high theoretical and practical importance.For the interleaver recognition problem,the theoretical lower bound is derived for the received data quantity in order to achieve a given probability of correct recognition.The existing optimum method suffers from the fact that once the recognition process fails,lots of wrong results and waste computation will be produced subsequently.To solve this,a simple and efficient failure-detecting method is designed and the setting of its two thresholds is discussed.Based on the failure-detecting,an automatic failure-correcting method is then proposed to make the recognition process return to the correct state.Further,a “combined” recognition method is developed,which combines turbo iterative decoding with the failure-detecting based recognition.Simulation results show that the combined method enhances error-tolerant performances effectively,with the needed data quantity reduced by about 1/3 and thus to be much closer to the theoretical lower bounds,compared to the existing optimum method.LDPC codes form another class of Shannon-limit-approaching codes.Recognition of LDPC codes without a candidate set is one of the toughest tasks.In this thesis,dual space of the linear space spanned by the received code words is constructed.Through theoretical derivation,a lower bound for the number of received words is decided,in order to make sure that sparse vectors in the dual space are all sparse parity-check vectors of the LDPC code.These sparse vectors are then found out by an existing fast search algorithm.Under noiseless conditions,number of iterations is analyzed using the exponential distribution theory,and a stop criterion is thus obtained.This new method overcomes the problems of the existing method that it is only applicable to LDPC codes whose sparse parity-check matrices have “diagonal structure”,and that a large number of received words are needed.Under noisy conditions,the combined recognition method is proposed,which combines the belief-propagation(BP)iterative decoding with the search of sparse parity-check vectors.Simulation results show that the new method is able to complete the recognition of practical LDCP codes within reasonable time scopes,under channel conditions of real noise level.This solves the problem of existing methods that they are actually not practical since their computation complexity is too high or the error-tolerant capability is too weak.Convolutional interleavers are widely used in practice.The existing recognition method based on the existence of frame sync codes has good performances except for its high computational complexity.The main computation is in recognizing the period of frame sync codes through a partitioning-and-summing method.It is made clear in the thesis that this method involves lots of repeated computation.A low-complexity method is proposed which avoids such repeated computation.Both theoretical and simulation results show that,the proposed method is able to reduce the computational complexity by a percentage ranging from around 50% to more than 90%.Further,a new method called cyclic convolutional de-interleaving is developed to recover the interleaver parameters following the recognition of the period.Simulation results show that the new method improves the error-tolerant capability by more than 2d B and is able to recognize the interleaving depth starting point,which the existing method is not able to recognize.