A method is proposed, called l-formula, to design efficient successive cancellation (SC) decoding of polar codes with medium-dimensional binary kernels (dimensions up to 16). Our l-formula method obtains the simplified recursive formulas of the SC decoder in the likelihood ratio domain for any binary kernel.We confirm that the complexity of the SC decoder based on l-formulas achieves considerable advantages over the straightforward SC decoder for polar codes with medium-dimensional binary kernels.
As the largest target in remote sensing images, buildings have important application value in urban planning and old city reconstruction. However, most networks have poor recognition ability on high resolution images, resulting in blurred boundaries in the segmented building maps. Then, the similarity between buildings and backgrounds will lead to inter-class indistinction. Finally, the diversity of buildings brings difficulties to segmentation, which requires the network to have better generalization ability. To address these problems, we propose Fusion Multi-scale Attention Mechanism Network(FMAM-Net). Firstly, we design Feature Refine Compensation Module(FRCM) to improve the boundary ambiguity problem, including Feature Refinement Module(FRM) and Feature Compensation Module(FCM). FRM utilizes the densely connected architecture to refine features and increase recognition capabilities. FCM introduces low-level features to make up for the lack of boundary information in high-level features. Secondly, to handle inter-class indistinction, we design Tandem Attention Module(TAM) and Parallel Attention Module(PAM). TAM is designed to sequentially filter some features from channels and spaces for adaptive feature refinement. PAM combines context information and uses high-level features to guide low-level features to select more distinguishable features. Finally, based on the binary cross entropy loss function, we add an evaluation index to reduce the error caused by determining the optimization direction only through cross entropy. On the Inria Aerial Image Labeling Dataset, FMAM-Net achieves mean IoU of 85.34%, which is 5.58% higher than AMUNet and 3.77 % higher than our baseline(U-Net ResNet-34). On the WHU Dataset, IoU reached the maximum value of 91.06% on FMAM-Net, 1.67% higher than SARB-UNet and 0.2% higher than MAP-Net. The visualization results show that FMAM-Net improves the fuzzy boundary of building segmentation and reduces the inter-class indistinction.
Polar codes constructed by large kernels can attain better finite length performance than those originating from Arıkan’s 2 × 2 kernel. However, the successive cancellation (SC) decoding for these polar codes is impractical even for relatively small kernel size of m because complexity of the kernel computation grows exponentially with m . This research shows when m > 2, there exists a large amount of like terms in the kernel computation which yields a ground for facilitating the decoding. By transferring the kernel computation from the probability domain to the likelihood ratio domain ( l -domain), the so-called l -formula method provides an efficient way to combine the like terms in the kernel computation for kernels up to size 11. However, the l -formula method becomes intractable for kernel size beyond 11. To further reduce the computational complexity, this paper proposes a W -formula method which transforms the kernel computation into the probability pair domain ( W -domain). Advanced from the l -domain, the numerator and denominator of the likelihood ratio are considered separately, which eases the restrictions of combining like terms. The W -formula method can combine much more like terms resulting in a significant reduction on the number of sub-formulas for medium kernels ( m ≤ 16). Furthermore, in the W -domain, sub-formulas become regular and there exist many common sub-formulas whose computations can be shared. Being able to handle kernels of size up to 16, we show that the W -formula based SC decoding achieves a significant complexity reduction over the existing non-approximate SC decoding (the l -formula based SC decoding).
We propose a two-phase Monte Carlo (TPMC) method to accelerate the original Monte Carlo (MC) method for constructing polar codes with high-dimensional kernels. In the TPMC method, some of the most reliable and unreliable bits are obtained by Gaussian approximate-density evolution (GADE) method in the first phase; in the second phase, these most reliable and unreliable bits are viewed as frozen bits. Then, the MC method are used to evaluate the remaining bits and select some best bits from the remaining bits. Finally, these best bits and the most reliable bits are combined as the information bits of the constructed polar code. By our investigation, most bits can be fixed as frozen bits in the second phase without error performance loss of the constructed polar codes. Because computation of frozen bits can be saved and computation of the GA-DE method can be ignored in contrast to that of the MC method, the TPMC method substantially reduces the complexity of the MC method. Simulation results show that 1) For a G⊗315 polar code with block length 3375 and code rate 1=2, the TPMC method can fix 3200 bits as frozen bits in the second phase without error performance loss, which reduces the computation cost by approximately 92:6% over the MC method; 2) Polar codes with high-dimensional kernels constructed by the TPMC method outperforms polar codes with the G2 kernel constructed by the Tal-Vardy method in terms of error performance; 3) With the same computational cost, the TPMC method can construct better polar codes than the MC method.
This paper presents a new latency reduction method for successive-cancellation (SC) decoding of polar codes that performs a frozen-bit checking on the rate-other (R-other) nodes of the Fast Simplified SC (Fast-SSC) pruning tree. The proposed method integrates the Fast-SSC algorithm and the Improved SSC method (frozen-bit checking of the R-other nodes). We apply a recognition-based method to search for as many constituent codes as possible in the decoding tree offline. During decoding, the current node can be decoded directly, if it is a special constituent code; otherwise, the frozen-bit check is executed. If the frozen-bit check condition is satisfied, the operation of the R-other node is the same as that of the rate-one node. In this paper, we prove that the frame error rate (FER) performance of the proposed algorithm is consistent with that of the original SC algorithm. Simulation results show that the proportion of R-other nodes that satisfy the frozen-bit check condition increases with the signal-to-noise-ratio (SNR). Importantly, our proposed method yields a significant reduction in latency compared to those given by existing latency reduction methods. The proposed method solves the problem of high latency for the Improved-SSC method at a high code rate and low SNR, simultaneously.
Compared to Arikan’s kernel, large-kernel polar codes exhibit higher polarization rates and superior error correction performance. The critical steps of exact successive cancellation (SC) decoding for such codes can be implemented via trellis-based computations to reduce complexity. However, the complexity remains high for large kernels. This paper proposes a permutation-based trellis optimization scheme. The approach builds on the Massey minimal trellis and reorders its time axis to find a permutation that minimizes the number of trellis edges, thereby further reducing the exact SC decoding complexity. For smaller kernels (–), an exhaustive search is conducted to identify the optimal trellis. For larger kernels (–), where an exhaustive search becomes infeasible due to the factorial growth of the permutation space, an ant colony optimization (ACO)-based method is employed to find a near-optimal permutation. Simulation results show that the permutation-optimized trellis lowers the direct SC decoding complexity drastically. Furthermore, compared to the -expression, the -formula and original Massey trellis methods, it achieves multiplication operation reductions of up to 99.2%, 58.1%, and 56.5%, respectively. The improvement is particularly beneficial for large kernels, where traditional decoding methods become computationally prohibitive.