Research And Application On Decoding Algorithm Of Polar Codes In EMBB

Polar codes,recently introduced by Ankan,is the first channel code which can be strictly proved to achieve the channel capacity of binary-input discrete memoryless channels(B-DMCs).This capacity-achieving code family is based on a method called channel polarization,which combines and splits the copies of a given B-DMC W in a recursive manner,and the symmetric capacity tend towards 0 or 1 as N goes to infinity.Due to the capacity-achieving property and a low-complexity encoding and decoding process,polar codes have been chosen to be adopted in 5G by 3GPP.In this thesis,we investigate the blind detection scheme of polar codes,the feasibility of the application of machine learning to the decoding of polar codes and the near ML decoding algorithm of polar codes.The related research contents are summarized as follows.A low-complexity blind detection scheme for polar code based on segmented CRC is proposed.As CRC detectors are placed at several breaking points in the decoding process,UEs are able to perform early terminations for erroneous decoding attempts,thus reducing the decoding complexity.The rationale of the proposed scheme is that such decoding attempts cause errors in the stream of decoded bits and hence can be detected by the segmented CRCs during the decoding process.Besides,in our scheme,both the frozen bit-channels and the priori knowledge of UE radio network temporary identifier(RNTI)are utilized to identify the control messages between different UEs.We have analyzed the computational complexity reduction introduced by the proposed scheme.It shows a significant gain compared to the conventional LTE approach,which just simply insert a long CRC at the end of channel codes.Simulation results verifies the good performance of the proposed scheme in terms of miss detections,false alarms and early terminations.A decoding algorithm based on machine learning is developed.Decision tree is able to learn the decoding information and is simple to understand and interpret.Besides,decoding algorithm based on decision tree has the near-ML decoding performance and low decoding latency,is a candidate of new type of decoding algorithm for polar codes.A decoder based on random forest,gradient boost decision tree and extra tree is proposed in this thesis.Compared with neural network decoder,the proposed decision tree decoder has much lower time complexity with training the model and is much easier for adjusting the parameters.The simulation results show that the proposed decoder has near-ML decoding performance.A method of expanding the decoding algorithm to longer polar codes is also proposed.Finally,we put forward a Threshold-based CRC-Aided BMA decoding algorithm.The box and match decoding algorithm is an efficient most reliable basis(MRB)based soft decision decoding algorithm.The BMA algorithm reduce the computational cost of ordered static decoding(OSD)at the expense of much higher space complexity.But the BMA need to find the global lower bound(GLB)of ellipsoidal weight(EW),which is of high computational cost.We introduced the CRC-Aided BMA algorithm to avoid the computation of GLB of EW so as to reduce the complexity.Since the average number of candidate codewords is still high,we introduced the threshold-based CRC-Aided BMA to further reduce the complexity.The simulation results show that a well-designed threshold-based CRC-Aided BMA can reduce the complexity without significant decoding performance loss.It is also illustrated that the proposed threshold-based CRC-Aided BMA outperforms the OSD.