New Construction and Performance Analysis of Polar Codes over AWGN Channels Bashar Tahir, and Markus Rupp Institute of Telecommunications Technische Universit¨ at (TU) Wien Vienna, Austria Email: {bashar.tahir, markus.rupp}@tuwien.ac.at Abstract—We propose a new construction algorithm for Polar codes operating over Additive White Gaussian Noise channels under Successive Cancellation decoding. Our approach is based on tracking the bit error probabilities of the bit channels as they evolve through the decoder, allowing us on the one hand to characterize the performance of these channels, and on the other hand, providing a solid construction algorithm. We then use our approach to derive a modification for the density evolution (with Gaussian approximation) based construction, providing better accuracy and implementation. I. I NTRODUCTION Polar codes attracted a lot of attention since they were introduced by Arıkan in 2008 [1]. They are the first practical codes that are proven to achieve the channel capacity at infinite length. The structure of polar codes is based on the channel polarization transform, where a specific dependency between the input bits is introduced. Such dependency can be exploited using a Successive Cancellation (SC) based decoder, which causes subsets of the bits to have improved reliability when the successively fedback bits are correct. In this case, the task of constructing polar codes involves finding the most reliable bits’ positions (bit channels) that gain the most from the successive decoding, and use them for the transmission of information bits. While foreknown bits (usually zeros) are transmitted over the other unreliable channels (Frozen set), ensuring correct feedback of the those bits since they are known by the receiver, thus producing the coding gain. Polar codes are non-universal, in the sense that their con- struction is dependent on the Signal-to-Noise Ratio (SNR) of the receiver. Such property is actually a drawback, since the optimum transmission scheme would then require an adaptive construction based on the SNR. Therefore, it’s no surprise that the SNR is a design parameter when it comes to the construction algorithms of polar codes. The first construction algorithm was introduced by Arıkan, where the Bhattacharyya parameters of the bit channels are evolved through the polarization structure. A construction based on Density Evolution with Gaussian Approximation (DEGA) is given in [2], which is based on tracking (evolving) the density functions of the Log-Likelihood Ratios (LLRs) The financial support by the Austrian Federal Ministry of Science, Research and Economy and the National Foundation for Research, Technology and Development is gratefully acknowledged. through the decoder. Some other constructions are given in [3]–[6], with varying performance and complexity. In this paper, we propose a new construction algorithm for polar codes operating over Additive White Gaussian Noise (AWGN) channels under SC decoding, based on tracking the bit error probabilities of the bit channels as they evolve through the decoder, providing also a performance characterization for those channels. Our results show that the new algorithm deliv- ers similar performance to that of DEGA, and because the two algorithms are connected through the bit error probabilities, we derive a modification for DEGA, improving its accuracy and implementation. We finally benchmark their performance. II. PROBABILITY ANALYSIS In this section, we perform the probability analysis for the successive cancellation decoder. Fig. 1 shows the SC decoder for a polar code of length 4. f g f g f g f g u 0 ^ u 1 ^ u 1 ^ u 2 ^ u 0 ^ u 0 ^ u 1 ^ u 2 ^ u 3 ^ L 0 (0) L 1 (0) L 2 (0) L 3 (0) L 0 (2) L 1 (2) L 2 (2) L 3 (2) L 0 (1) L 1 (1) L 2 (1) L 3 (1) Fig. 1. Polar decoder of length 4. In the LLR domain, the nodes f and g perform the following calculations for given input LLRs L a and L b f (L a ,L b ) = log e La+L b +1 e La + e L b , (1) g(L a ,L b ,u s )=(-1) us L a + L b , where u s is the Partial Sum, which is the sum of the previously decoded bits that are fedback to the current node g. 978-1-5386-0643-8/17/$31.00 ©2017 IEEE