Research On Blind Parameter Recognition Of Convolutional Codes

As an important part of communication system,channel coding is a key measure to improve the reliability of communication transmission.However,in some application scenarios,such as intelligent communication and communication countermeasure,coding parameters used by transmitters are often unknown to the receivers.Therefore,in order to recover the original information,it is necessary to identify the coding parameters from the received noisy data.The blind recognition of channel codes is defined as the process of identifying the coding type and its parameters without or only part of the coding prior information.At present,blind recognition of channel codes can be roughly divided into blind recognition based on a closed set and blind recognition based on an open set.The blind recognition problem based on a closed set is defined as the receiver selects the coding parameters that best fit the received data from a candidate set of channel coding parameters according to the received data.However,the blind identification problem based on an open set is defined as the receiver identifies the channel coding parameters only based on the received data without a candidate set.Convolutional codes is one of the most commonly used channel coding types,and its blind recognition has attracted extensive attention of researchers.At present,the commonly used recognition methods of convolutional codes,such as Gaussian elimination method,Euclidean recognition method,Walsh-Hadamard analysis method and so on,all of which only use the hard decision information of the transmitted data and have poor anti-noise performance.However,most of the existing communication systems use soft-decision rather than hard-decision for decoding,and there are relatively few related methods that use soft information of the transmission data for recognition.Therefore,this thesis studies the recognition method based on the soft information of the transmitted data for the problem of closed-set blind recognition and open-set blind recognition of convolutional codes.The main contents are as follows.For the problem of closed-set blind recognition of convolutional codes parameters,three methods based on the check equation’s cosine function,cosine Taylor expansion function and log-likelihood ratio function are proposed to detect and identify the convolutional codes in a given candidate set.Firstly,the convolutional codes with a given memory order are traversed to eliminate the malignant encoder.In addition,the free distance of convolutional codes is calculated according to the generator polynomials,the convolutional codes with small free distance will be eliminated,so as to construct candidate code sets with different memory orders.Then,for each convolutional code in the candidate set,the check equation’s cosine Taylor expansion function,cosine function and log likelihood ratio function are calculated by using the received data,and the convolutional code that maximizes the function value is selected.The recognition accuracy of the two recognition methods based on the check equation’s cosine function and cosine Taylor expansion function is analyzed theoretically.Simulation results demonstrate that the recognition performance of the three recognition methods based on the check equation’s cosine function,cosine Taylor expansion function and log-likelihood ratio function is better than the existing methods,and the method based on cosine Taylor expansion function has the best recognition performance.For the problem of open-set blind recognition of convolutional codes,an improved parameter estimation method of generator polynomials of convolutional code is proposed by using the adaptive moment estimation(ADAM)optimization algorithm.Firstly,a predicted probability description according to the check equation of convolutional codes is given to describe whether the estimation results are correct.Then,the cross entropy cost function is used to describe the difference between the predicted probability distribution and the real probability distribution,so that the estimation of the coefficients of the generator polynomials is transformed into an optimization problem.Afterwards,the ADAM update rule is adopted to minimize the cost function.In addition,data segmentation and multi-frame schemes are utilized to further improve the recognition performance when the memory order of the encoder increases.Simulation results demonstrate that compared with the existing method,the proposed method can improve the estimation accuracy significantly,especially for convolutional codes with larger memory order.