An optimized initialization for LDPC decoding over GF(q) in impulsive noise environments

Haoqiang Liu; Hongbo Zhao; Xiaowen Chen; Wenquan Feng

doi:10.1371/journal.pone.0250930

Abstract

Modern navigation satellite communication has the characteristic of high transmitting rate. To avoid bit errors in data transmission, low density parity check (LDPC) codes are widely recognized as efficient ways for navigation communication. Conventionally, the LDPC decoding is applied for additive white Gaussian noise (AWGN) channel and degrades severely while facing the impulsive noise. However, navigation communication often suffers from impulsive interference due to the occurrence of high amplitude “spikes”. At this time, the conventional Gaussian noise assumption is inadequate. The impulsive component of interference has been found to be significant which influences the reliability of transmitted information. Therefore the LDPC decoding algorithms for AWGN channel are not suitable for impulsive noise environments. Consider that LDPC codes over GF(q) perform better than binary LDPC in resisting burst errors for current navigation system, it is necessary to conduct research on LDPC codes over GF(q). In this paper, an optimized initialization by calculating posterior probabilities of received symbols is proposed for non-binary LDPC decoding on additive white Class A noise (AWAN) channel. To verify the performance of the proposed initialization, extensive experiments are performed in terms of convergence, validity, and robustness. Preliminary results demonstrate that the decoding algorithm with the optimized initialization for non-binary LDPC codes performs better than the competing methods and that of binary LDPC codes on AWAN channel.

Citation: Liu H, Zhao H, Chen X, Feng W (2021) An optimized initialization for LDPC decoding over GF(q) in impulsive noise environments. PLoS ONE 16(5): e0250930. https://doi.org/10.1371/journal.pone.0250930

Editor: Lisu Yu, Nanchang University, CHINA

Received: May 26, 2020; Accepted: April 17, 2021; Published: May 6, 2021

Copyright: © 2021 Liu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the manuscript and its Supporting Information files.

Funding: This paper was supported by the National Natural Science Foundation of China (No.91438116 and No.61901015). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

With the explosive development of communication technology, the new mobile communication systems, such as beyond fifth generation (B5G) and sixth generation (6G) systems, will suffer from severe challenges imposed by the requirement for heavy connection density and high efficiency [1]. Especially for 6G, satellite communications play an important role in providing high quality communication services to achieve the worldwide connectivity [2]. As one of the critical components, modern navigation satellite communication has the characteristic of high transmitting rate which is a real challenge to ensure the correctness of transmitted information in various channels [3,4]. To avoid bit errors in data transmission, it is an efficient way to employ low density parity check (LDPC) codes in navigation satellite communication channel. Nowadays, LDPC codes have been widely applied in the 2nd-generation digital video broadcast via satellite (DVB-S2) standard [5], 5G mobile communications [6], DNA barcoding [7], as well as the uplink of mobile satellite communication [8].

For modern navigation system, LDPC codes over GF(q) have been employed in encoding and decoding as important codes. With the rapid development of BeiDou Navigation Satellite System (BDS) 3, B1C signal and B2a signal have been utilized gradually. The B-CNAV1 navigation message is transmitted through B1C signal. Each frame of message consists of 3 sub frames, of which the second sub frame exploits LDPC codes (200,100) over GF(64) and the third sub frame leverages LDPC codes (88, 44) over GF(64) for encoding [9]. Meanwhile, the B-CNAV2 navigation message is transmitted through B2a signal with LDPC codes (96, 48) over GF(64). Non-binary LDPC codes have been recognized as a powerful technology to encode navigation information efficiently. With the continuous development of BDS 3 and the increasing demand for navigation, people hope to access precise navigation information in time. However, working in severe environments, such as underwater environment surrounded by acoustical noises or in the power industry, the decoder of non-binary LDPC codes is inevitably subject to impulsive interference, which causes poor reliability of transmitted information and even leads to communication failure. Therefore, nowadays, evaluation and analysis of LDPC codes over GF(q) for navigation satellite communication, especially in harsh environments, arouse attentions of researchers all over the world.

Conventionally, the LDPC decoding algorithms are designed for AWGN channel. However, navigation satellite communication often suffers from irregular noises due to the occurrence of high amplitude “spikes”. For navigation satellite communication, such spikes can be generated in atmosphere where lightning discharges in the vicinity of the receiver, or underwater environment where the ambient acoustical noises includes impulses due to noisy aquatic animals such as snapping shrimp [10–13]. At this time, the original assumption that noises are the Gaussian noises is inadequate. Unfortunately, few attention has been given to decoding algorithms for LDPC codes over GF(q) in impulsive noise environments.

Generally, compared with binary LDPC, LDPC codes over GF(q) show superior performance in resisting burst errors such as impulsive noises for the characteristic of inner interleaving [14,15], which makes non-binary LDPC codes more suitable for navigation communication. Furthermore, non-binary LDPC codes combined with q-ary modulation can increase transmission rate obviously [16,17]. Although there is an increment of computational complexity by adopting LDPC codes over GF(q), with the development of terminal computation, non-binary LDPC codes have become a hot research topic and would become more prevalent and applicable. Therefore, we focus on non-binary LDPC codes decoding on additive white Class A noise (AWAN) channel and present an optimized initialization for decoding in this paper. With simulation experiments, we demonstrate the efficiency of the proposed algorithm. The main contributions are summarized as follows:

We investigate the problem of LDPC decoding in impulsive noise environments for navigation communication and formalize the impulsive noise as the Class A noise model.
We propose an optimized initialization by calculating posterior probabilities of received symbols for non-binary LDPC decoding on AWAN channel, which makes use of series truncation for computing effectively.
Extensive experiments are conducted on convergence, validity, and robustness. The experimental results reveal that the optimized initialization has a significant effect on the decoding performance for non-binary LDPC codes on AWAN channel.

The rest of the paper is organized as follows. We illustrate the related work in Section 2. Section 3describes AWAN model based on the statistics and conventional initialization of decoder for LDPC codes over GF(q). Section 4illustrates the optimized initialization process on AWAN channel. Simulation results are shown and discussed in Section 5. Finally, the conclusion is given in Section 6.

Related work

LDPC codes were introduced by Gallager in 1962 for the first time, which are linear codes with sparse parity-check matrix [18]. In 1996, Mackay and Wiberg found that LDPC codes achieved excellent performance approximating the Shannon limit, and it soon became a research hotspot in the channel coding theory [19]. To improve the performance of error correction and transmission, Davey and MacKay extended the belief-propagation (BP) decoding algorithm for binary LDPC to non-binary LDPC firstly [20]. Furthermore, they proposed a fast Fourier Transform-based belief-propagation (FFT BP) decoding algorithm for reducing complexity [21]. Unfortunately, this method was only valid when the Galois field was a binary extension field with q = 2^p and inefficient to handle other situations. To further simplify the decoding procedure, Declercq et al. introduced an extended min-sum algorithm, but it led to degradation of performance [14]. The United States adopted LDPC codes with 1/2 code rate in the L1C signal of GPS for the first time. Hareedy Ahmed et al. proposed non-binary LDPC codes for magnetic recording channels, and provided a comprehensive analysis of the error floor along with codes optimization guidelines for structured and regular non-binary LDPC codes [22]. In 2017, Huang Qin proposed message-passing decoding algorithms that decoded non-binary LDPC codes including ultra-sparse ones efficiently [23]. In 2019, Rehman employed parallel architecture for LDPC codes decoding to achieve the higher data rate, which in turn raised the memory conflict issue [24]. In recent years, with the emergence of decoding algorithms with low complexity, LDPC codes show superiority in practice with their excellent capability, and have been replacing the conventional codes as main codes in future navigation satellite communication gradually.

Generally, receivers adopts parameters of the AWGN channel and conventional decoding methods for LDPC decoding, which results in serious degradation in impulsive noise environments. The impulsive component of interference has been found to be significant which influences the reliability of transmitted information.

Various attempts have been made to develop models of impulsive noises that can be divided into empirical models and physical models. Class A noise model proposed by Middleton is a typical kind of physical models [25]. The statistical feature of Class A noise is much different from that of Gaussian noise, therefore the LDPC decoding algorithms for AWGN channel are not suitable for Class A noise environments. Maad et al. analyzed the performance of LDPC codes in heavy-tailed, symmetric alpha stable noise (SαS) channels [26]. Further, Nakagawa et al. proposed the sum-product decoding method for binary LDPC codes in Class A noise environment [27]. However, few researches have ever explored decoding algorithms for LDPC codes over GF(q) in Class A noise environments.

This paper is focused on the design of an optimized initialization for non-binary LDPC codes on AWAN channel. In contrast to metaheuristics, our optimized initialization is dedicated to deal with decoding problems based on BP. Generally, as the famous optimization techniques, metaheuristics are widely recognized as efficient approaches for optimization problems, such as particle swarm optimization (PSO) [28,29] and differential evolution (DE) [30]. Various metaheuristic methods have reported advantages in image segmentation [31], tuning hyper-parameters of deep neural networks [32,33], and benzene prediction model [34]. However, as illustrated in [35], the successful application of metaheuristics requires to find a good initial parameter setting, which is a tedious and time consuming task. Moreover, the performance of metaheuristics deteriorates quickly as the dimensionality increases, nevertheless high-dimensional circumstances are extremely common in encoding and decoding. Therefore, decoding algorithms based on BP are considered for non-binary LDPC codes on AWAN channel, and an optimized initialization by calculating posterior probabilities of received symbols is proposed.

Theoretical background

AWAN model

When encountering lightning or water during transmission, navigation signals will be interfered by the impulse noise, which leads to the unexpected change of the amplitude of the transmitted signal. In this part, the Class A noise model devised by Middleton [25] is described as a statistical AWAN model with the impulsive noise environment, which is widely applicable by adjusting parameters and provides fine closeness to experimental values. According to this theory, the Class A noise model is composed of Gaussian noise G(t) and impulsive noise X(t), which can be expressed as (1) where G(t) is considered as background noise and X(t) can be expressed as [25] (2)

Here, U_j denotes the j-th received impulse noise waveform from an interfering source and ϑ represents the random parameters which describe the waveform scale and structure. Assume that there is only one type of waveform, and U is generated with appropriate variations in the individual wave form under the variation of the parameter ϑ. According to the Class A noise model, the probability density function (PDF) of the noise amplitude z can be defined as (3) where , A is the impulsive coefficient which is defined as average number of impulses on the receiver in unit time. is the Gaussian-to-Impulsive noise power ratio (GIR) with Gaussian noise power and impulsive noise power . Therefore the total noise power is .

The Class A noise in (3) consists of the impulsive noise with variance and the background Gaussian noise with variance . The number of impulsive noise is distributed with Poisson distribution (e^-A∙A^m)/m! and the amplitude of each impulsive noise is characterized by a Gaussian PDF with variance . Therefore, at a certain observation time, assume that the number of impulsive noise is m, which is characterized by a Poisson distribution with mean A, the noise of receiver is characterized by a Gaussian PDF with variance .

Consider the independence between the background Gaussian noise and the impulsive noise, as the impulse coefficient A increases, the impulsive noise becomes more intensive and continuous, which makes the Class A noise approximate the Gaussian noise. In particular, if A is close to 10, the statistical feature of the Class A noise is almost similar to that of the Gaussian noise [25]. In addition, as Γ grows, i.e., the proportion of Gaussian white noise in the total noise increases, the Class A noise gets close to the Gaussian noise. Otherwise, the smaller Γ is, the more impulsive the Class A noise would be.

Conventional initialization in decoding process

Non-binary LDPC codes can be considered as a kind of linear block codes which are the extension of binary LDPC codes over GF(q). The difference between non-binary LDPC codes and binary LDPC codes is that each non-zero element of sparse parity-check matrix needs to be obtained from GF(q). The Tanner graph of non-binary LDPC codes given by a sparse parity-check matrix over GF(q) is constructed in the same way as that of binary LDPC codes. Compared with binary LDPC codes, non-binary LDPC codes perform better in communication due to advantages in resisting burst error and high transmission rate [36–39].

A significant amount of research has been concentrated on the design, encoding, decoding and performance analysis of non-binary LDPC codes. Iterative decoding algorithm based on BP is an important soft decision decoding algorithm. According to this algorithm, messages are delivered between variable nodes and check nodes during iterations after initialization. Further, the received codes are updated until satisfying the parity-check equations or the upper limit of the iteration number is reached.

Since the channel transition probability is only utilized in the initialization process, we focus on optimizing initialization of LDPC decoding on AWAN channel. In this subsection, traditional initialization method for non-binary decoding is introduced. Consider that the codeword u = (u₀,u₁,…,u_n-1) is obtained by encoding information sequence e = (e₀,e₁,…,e_k-1). Q-ary sequence u can be expanded into binary sequence ((u_0,1,u_0,2,…,u_0,p),…,(u_n-1,1,u_n-1,2,…,u_n-1,p)) and modulated by binary phase shift keying (BPSK) (using the mapping 0 to 1, 1 to -1) to obtain the transmitted sequence x = ((x_0,1,x_0,2,…,x_0,p),…,(x_n-1,1,x_n-1,2,…,x_n-1,p)), where p = log₂q. And y = ((y_0,1,y_0,2,…,y_0,p),…,(y_n-1,1,y_n-1,2,…,y_n-1,p)) is the received sequence transmitted on AWGN channel with variance σ² and mean 0.

The initial messages sent from variable node v_j to the check node c_i is the probabilities of the j-th code symbol u_j equal to {a₁,a₂,…,a_q} respectively, given received sequence y. And it can be expressed as (4) where 1≤k≤q and a_k denotes the element of GF(q). We expand a_k to binary sequence {a_k1,a_k2,…,a_kp} and the Eq (4) can be rewritten as: (5)

There is a mapping between expanded codeword u and transmitted sequence x due to BPSK modulation. Thus, P(u_j,1 = a_k1|y_j,1) equals to P(x_j,1 = 1−2a_k1|y_j,1). According to the Bayes formula, the posterior probabilities of received bits can be acquired by (6) where a= ±1, P(x_j,1 = ±a) = 0.5 and P_G(∙) is the Gaussian PDF with mean 0 and variance . Due to (7) thus (8)

And the initial messages of BP algorithm for LDPC codes over GF(q) could be obtained by adding (8) to (5).

An optimized initialization for LDPC decoding over GF(q) on AWAN channel

The conventional decoding methods degrade severely while facing the impulsive noise since those methods acquire the posterior probability of received symbols by making use of the transition probability on AWGN channel. To tackle this problem, based on the BP decoding algorithm, we present an optimized initialization for LDPC decoding over GF(q) on AWAN channel with series truncation.

As mentioned above, the initialization is the crucial part of the BP decoding over GF(q) algorithm. The posterior probability of received symbols is considered as the initial message transmitted from the variable nodes to the check nodes. Thus, to evaluate the posterior probability of received symbols correctly is vital for decoding.

During the initialization of conventional decoding process, we obtain the posterior probability of the received bits from the transition probability of the AWGN channel as Eq (9), which suffers from degradation of the bit error rate (BER) in impulsive noise environments. Therefore, for LDPC decoding over GF(q) on AWAN channel, since the channel parameters are only leveraged in the decoding initialization, we need to improve the initialization of the iterative decoding process without changing the information transmission mode between variable nodes and check nodes. Under this circumstance, the optimized initialization can be applied to various traditional LDPC decoding algorithms over GF(q), such as FFT BP and EMS algorithm. Generally, the BP decoding algorithm with optimized initialization for LDPC codes over GF(q) on AWAN channel can be illustrated by the flow chart in Fig 1.

Download:

Fig 1. The flow chart of BP decoding for LDPC codes over GF(q) on AWAN channel.

https://doi.org/10.1371/journal.pone.0250930.g001

In the following part, we elaborate the optimized initialization in FFT BP decoding algorithm and series truncation is proposed to calculate the PDF of the Middleton Class A noise effectively. The LLR of the received bit can be defined as (9) where 0≤j<n and 1≤k≤p. The posterior probability in Eq (9) can be expressed as prior probability by (10) where P(x_j,k = +1) = P(x_j,k = −1) = 0.5. Hence, LLR can be rewritten as (11) in which P_A(∙) denotes the PDF of the Class A noise mentioned in (3). Thus, this PDF is injected in LLR of the incoming BPSK modulated bits.

FFT BP decoding with the proposed initialization (i.e. the optimized FFT BP algorithm) for non-binary LDPC codes on AWAN channel is illustrated in Algorithm 1, where t denotes the t-th iteration, E_n(u) is the layered message being u (u∈GF(q)) and initialized with the channel message P_n(u) in Eq (5), represents the message from a check node (CN_m) to a variable node (VN_n), indicates the message from VN_n to CN_m, and k_v is a normalizing constant to ensure is the set of variable nodes connected to CN_m, and φ(m)/n is the set of variable nodes connected to CN_m except VN_n. φ(n) is the set of check nodes connected to VN_n, and φ(n)/m is the set of check nodes connected to VN_n except CN_m. P(∙) and P^-1(∙) denote the permutation function and inverse permutation function [40]. F(∙) and F^-1(∙) indicate the FFT function and IFFT function. denotes the permutation message of . is the Fourier domain message of . is the check node updated message in the probability domain. indicates the check node updated message in the Fourier domain. also indicates the inverse permutation message of .

In practice, we need to estimate the impulsive coefficient A, GIR Γ and the background Gaussian noise power properly. Fortunately, these parameters can be estimated from the second, fourth and sixth epochs of the received envelopes [41].

Furthermore, since the PDF of the Class A noise P_A(z) in (3) consists of infinite series, the calculation of LLR is extremely complex practically. Thus, the series truncation can be regarded as a valid alternative to compute effectively by (14)

To obtain an appropriate L, we need to analyze the PDF of the Class A noise. Given the noise amplitude z, the value of each term in (14) is mainly influenced by impulsive coefficient A and equal to the product of Poisson component and Gaussian component. Fig 2 illustrates the relationship between Poisson component and term number m with different A. From Fig 2, we can observe that for the smaller A, the term m corresponding to the maximum value of Poisson probability becomes smaller. If A = 1, Poisson probability is close to 0 under the condition of m≥4, which leads to a negligible value for the m-th term of P_A(z). Thus, L is considered as the minimal integer whose Poisson cumulative distribution with mean A is more than 99%: (15)

Download:

Fig 2. Poisson probability vs. the term number m with different mean A.

https://doi.org/10.1371/journal.pone.0250930.g002

According to (15), L equals to 3 with A = 1. P_A(z) with different L is shown in Fig 3 and we can find that the curves are overlapped with L>3, which indicates formula (15) is effective.

Download:

Fig 3. PDF of noise amplitude z for A = 1 and L = 0, 1, 2, ….

https://doi.org/10.1371/journal.pone.0250930.g003

After introducing the optimized BP decoding of non-binary LDPC codes on AWAN channel, we investigate the characteristics of LLR on AWGN channel and AWAN channel with A = 0.1, Γ = 0.1. Suppose that the code rate is 1/2 and the signal-to-noise ratio E_b/N₀ is 0 dB for the received signal, where E_b denotes the energy per information bit and N₀ is the one-sided Gaussian noise power spectral density. According to the relationship between E_b/N₀ and the background Gaussian noise power , the following formulas can be acquired: (16) (17)

Fig 4 depicts the characteristics of LLR with different amplitudes of the received signal on AWGN and AWAN channels. From Fig 4, we can draw the conclusion that, the LLR is proportional to the amplitude of the received signal for AWGN channel; while for AWAN channel, it has a nonlinear relationship with the amplitude of the received signal. Specifically, for AWAN channel, the LLR is proportional to the amplitude of the received signal and similar to that of AWGN channel when the signal amplitude is less than 1. On the other hand, as the received signal amplitude increases, the corresponding LLR degrades due to that the received signal may contain impulsive noise with high probability.

Download:

Fig 4. LLR for AWGN and AWAN channels.

https://doi.org/10.1371/journal.pone.0250930.g004

Results and discussion

In this section, series of experiments are conducted to demonstrate the efficient performance of the proposed optimized initialization for LDPC decoding over GF(q) on AWAN channel. Irregular quasi-cyclic (QC) LDPC codes (88, 44) and (200, 100) over GF(64) with bit rate of 1/2 and a binary irregular QC-LDPC code (528, 264) with bit rate of 1/2 were utilized in our experiments. And the decoder was set to perform at most 50 iterations. Different decoding methods are evaluated and compared with respect to convergence, validity, and robustness. Experimental results indicate the superiority of the proposed optimized initialization.

Convergence comparison

Firstly, convergence issues have been considered as in [42] and performance of different initializations was evaluated utilizing the BER. To be specific, we compared the BER of the FFT BP decoding algorithm with the optimized initialization (i.e. the optimized FFT BP algorithm) with that of the conventional FFT BP algorithm (designed for AWGN channel) on AWAN channel. We should note that the difference between the two algorithms is the initialization. Therefore, the statistical evaluation was performed using the Wilcoxon rank test metric of both algorithms for QC LDPC codes (88, 44) with the statistical significance value α = 0.05. The null hypothesis H0 is ‘The difference between E_b/N₀ obtained by the optimized FFT BP algorithm and the conventional FFT BP algorithm is identical with the same BER’. Meanwhile, the alternative hypothesis is set as ‘the optimized FFT BP algorithm is validated’.

Table 1 presents the convergence comparisons of different initializations using the Wilcoxon Signed-Rank Test metric at different BER, where the ‘+’ indicates the cases when the algorithm acquires better coding gain. It clearly shows that the FFT BP algorithm with the proposed optimized initialization is statistically more superior.

Download:

Table 1. Results of convergence comparisons at different BER by utilizing Wilcoxon rank test (α = 0.05).

https://doi.org/10.1371/journal.pone.0250930.t001

More specifically, as shown in Fig 5, the BER performance of the conventional FFT BP algorithm for the QC LDPC code (88, 44) suffers from large degradation on AWAN channel with A = 0.1 and Γ = 0.1. Actually, in contrast to the BER on AWGN channel, the conventional FFT BP on AWAN channel suffers about 14 dB degradation at BER = 10⁻⁵ because of the emergence of the impulse noise. In particular, assume that the GIR , then the total noise power σ² of AWAN channel is 11 times of . If the total noise power on AWAN channel is the same as that on AWGN channel, the performance of the conventional FFT BP on AWAN channel suffers about 3.6 dB degradation than that on AWGN channel at BER = 10⁻⁵.

Download:

Fig 5. BER performance of the optimized FFT BP algorithm on AWAN channel.

https://doi.org/10.1371/journal.pone.0250930.g005

Further, from Fig 5, we can find that the proposed optimized FFT BP decoding algorithm with series truncation performs efficiently on AWAN channel. More precisely, it achieves about 12.2 dB coding gain at BER = 10⁻⁵ compared to the conventional one.

Validity analysis

Secondly, the BER performance of LDPC code (88, 44) over GF(64) was compared with the binary LDPC code (528,264) on AWAN channel with A = 0.1, Γ = 0.1 in Fig 6. The decoding algorithm of binary LDPC codes on AWAN channel was introduced by Nakagawa et al. in [27]. Fig 6 demonstrates that the BER performance of binary LDPC codes on AWAN channel suffers from about 1.7 dB degradation than LDPC codes over GF(64), which reveals that LDPC codes over GF(64) outperform binary LDPC codes on AWAN channel. Although there is an increase in computational complexity for decoding of non-binary LDPC codes, this could be appropriately addressed with the rapid development of hardware in terminals.

Download:

Fig 6. Performance comparison of LDPC codes over GF(2) and GF(64) on AWAN channel.

https://doi.org/10.1371/journal.pone.0250930.g006

Robustness analysis

Moreover, the robustness of the proposed initialization by considering LDPC codes with different lengths is analyzed. Fig 7 shows the performance comparison of the proposed approach with QC LDPC codes (88, 44) and (200, 100) over GF(64) on AWAN channel with A = 0.1 and Γ = 0.1. We can observe that, for both long and short codes, the decoding algorithm with the optimized initialization can be performed accurately and QC LDPC codes (200, 100) has 1 dB coding gain over QC LDPC codes (88, 44) at BER = 10⁻⁵. It can be concluded that, the proposed initialization shows an excellent robustness.

Download:

Fig 7. Performance comparison of QC LDPC codes on AWAN channel with A = 0.1 and Γ = 0.1.

https://doi.org/10.1371/journal.pone.0250930.g007

Effect of channel parameters

We also investigate the effect of different AWAN channel parameters on performance of the proposed FFT BP decoding algorithm. Table 1 summarizes the signal-to-noise ratio E_b/N₀ corresponding to BER = 10⁻⁵ under the variation of the impulsive coefficient A and the GIR Γ respectively. From this table, it is clear that E_b/N₀ becomes larger as the impulsive coefficient A increases. The reason is that, the larger A is, the more continuous impulsive noises would be, i.e., the statistical characteristics of the Class A noise approximate these of the Gaussian noise. Thus, the impulsive noise would not be suppressed well which leads to a worse performance of the optimized decoding algorithm. Particularly, the proposed algorithm with A = 10 suffers from 10.5 dB degradation compared to that with A = 0.01 at BER = 10⁻⁵.

Additionally, Table 2 also shows that E_b/N₀ becomes larger as the GIR Γ increases from 0.01 to 1. It is because that, with a larger Γ, the Class A noise shows fewer impulse characteristics and gets similar to the Gaussian noise. Therefore, the impulsive noise would not be suppressed well, which results in a worse performance. In contrast, E_b/N₀ becomes smaller as the GIR increases to 10. The performance is improved due to the fact that the impulse noise power becomes much small as the GIR increases further. Accordingly, the optimized decoding algorithm with Γ = 10 performs about 1.4 dB coding gain over that with Γ = 0.1 at BER = 10⁻⁵.

Download:

Table 2. Effect of AWAN channel parameters on performance of the optimized FFT BP.

https://doi.org/10.1371/journal.pone.0250930.t002

Simplification

Series truncation of the PDF of the Class A noise is introduced to simplify the calculation in section 4. Further, we compare it with another approach mentioned in [43] by piecewise fitting the optimal LLR in (11). Consider the LLR is related to the channel parameters, two piecewise functions were chosen to fit the optimal LLR with Γ = 0.1 and A = 0.1, which could be expressed as , where z denotes the amplitude of the received signal and is the Gaussian noise power of the Class A noise.

From Fig 8, it can be observed that, the BER performance of LLR with series truncation achieves 0.4 dB coding gain over that of LLR with piecewise fitting. Similar to the previous conclusion, the decoding algorithm with the optimized initialization shows superiority over the competing methods. Although the fitting approach makes the calculation easier, unfortunately, it is difficult to derive the relationship between the fitting function and channel parameters. Hence, we have to recalculate the fitting function under the variation of the channel parameters A and Γ, which complicates the decoding procedure.

Download:

Fig 8. Performance comparison of the proposed method with piecewise fitting.

https://doi.org/10.1371/journal.pone.0250930.g008

Conclusion

In this paper, we investigate LDPC decoding over GF(q) in impulsive noise environments for modern navigation satellite communication. By jointly considering the Class A noise model and the series truncation, we propose an optimized initialization for LDPC decoding over GF(q) on AWAN channel, which can be employed in BP-based iterative decoding algorithm. In addition, convergence, validity, and robustness of the proposed initialization are analyzed and discussed with extensive experiments. Simulation results demonstrate that, the decoding algorithm with the optimized initialization achieves 12.2 dB coding gain at BER = 10⁻⁵ compared to conventional methods on the assigned AWAN channel. Moreover, LDPC codes over GF(q) acquire 1.7 dB coding gain over binary LDPC codes at BER = 10⁻⁵ on AWAN channel. Robustness and the effect of channel parameters are confirmed by considering LDPC codes with different lengths and AWAN channel with parameters. Furthermore, by comparing the proposed method and the piecewise fitting method, experimental result verifies the feasibility of our method in practical applications. With the continues development of BDS-3, to employ the optimized initialization for LDPC decoding over GF(q) can achieve superior performance significantly in impulsive noise environments. The optimized initialization proposed in this paper can be also extended to decoding process of other 5G or 6G applications. In the future work, we will focus on studying LDPC decoding over GF(q) in a more complex noise environment for B5G and 6G systems.

Supporting information

S1 Data.

https://doi.org/10.1371/journal.pone.0250930.s001

(ZIP)

Acknowledgments

The authors wish to thank Prof. Qin Huang for the provision of valuable datasets. Thanks also go to the authors of the papers mentioned in REFERENCES.

References

1. Yu L, Wu J, Zhou A, et al. Massively Distributed Antenna Systems With Nonideal Optical Fiber Fronthauls: A Promising Technology for 6G Wireless Communication Systems. IEEE Vehicular Technology Magazine. 2020; 15(4): 43–51.
- View Article
- Google Scholar
2. Dang S, Amin O, Shihada B, et al. What should 6G be?. Nature Electronics. 2020; 3(1): 20–29.
- View Article
- Google Scholar
3. Xue R, Yu H, Cheng Q. Adaptive coded modulation based on continuous phase modulation for inter-satellite links of global navigation satellite systems. IEEE Access. 2018;6:20652–20662.
- View Article
- Google Scholar
4. Yan Z, Fraire JA, Zhao K, Yan H, Madoery PG, Li W. Distributed Contact Plan Design for GNSSs. IEEE Transactions on Aerospace and Electronic Systems. 2019.
5. Chen S, Peng K, Song J, Zhang Y. Performance analysis of practical QC-LDPC codes: From DVB-S2 to ATSC 3.0. IEEE Transactions on Broadcasting. 2018;65(1):172–178.
- View Article
- Google Scholar
6. Arora K, Singh J, Randhawa YS. A survey on channel coding techniques for 5G wireless networks. Telecommunication Systems. 2019:1–27.
- View Article
- Google Scholar
7. Tapia E, Spetale F, Krsticevic F, Angelone L, Bulacio P. DNA barcoding through quaternary LDPC codes. PloS one. 2015;10(10). pmid:26492348
- View Article
- PubMed/NCBI
- Google Scholar
8. Yang Y, Wang W, Gao X. AMP Dual-turbo iterative detection and decoding for LDPC coded multibeam MSC uplink. China Communications. 2018;15(6):178–186.
- View Article
- Google Scholar
9. Xiao X, Cheng F, Liu F, Lu X, editors. Design and Implementation of BDS RTCM SSR Message. IOP Conference Series: Materials Science and Engineering; 2017: IOP Publishing.
10. Liu L, Amin M, editors. Performance analysis of GPS receivers in impulsive noise. Digital Wireless Communications VII and Space Communication Technologies; 2005: International Society for Optics and Photonics.
11. Mei Z, Johnston M, Le Goff S, Chen L. Performance analysis of LDPC-coded diversity combining on Rayleigh fading channels with impulsive noise. IEEE Transactions on Communications. 2017;65(6):2345–2356.
- View Article
- Google Scholar
12. Wang P, Yin L, Lu J. Spatially Coupled Generalized Low-Density Parity-Check Codes Over Class-A Impulsive Noise Channels. IEEE Access. 2019;7:55906–55915.
- View Article
- Google Scholar
13. Ali I, Lee H, Kil YS, Hussain A, Kim S-H. Spatially coupled LDPC codes for power line communications with impulsive noise. IEICE Electronics Express. 2018:15.20180486.
- View Article
- Google Scholar
14. Declercq D, Fossorier M. Decoding algorithms for nonbinary LDPC codes over GF $(q) $. IEEE transactions on communications. 2007;55(4):633–643.
- View Article
- Google Scholar
15. Melo P, Pimentel C, Alajaji F. LDPC decoding over nonbinary queue-based burst noise channels. IEEE Transactions on Vehicular Technology. 2015;65(1):452–457.
- View Article
- Google Scholar
16. Wijekoon V, Viterbo E, Hong Y, Micheloni R, Marelli A. A Novel Graph Expansion and a Decoding Algorithm for NB-LDPC Codes. IEEE Transactions on Communications. 2019.
- View Article
- Google Scholar
17. Liu Y, Chen W, Wang A, Han C. Combination of High-Order Modulation and Non-Binary LDPC Codes over GF (7) for Non-Linear Satellite Channels. Electronics. 2019;8(12):1396.
- View Article
- Google Scholar
18. Gallager R. Low-density parity-check codes. IRE Transactions on information theory. 1962;8(1):21–28.
- View Article
- Google Scholar
19. MacKay DJ, Neal RM. Near Shannon limit performance of low density parity check codes. Electronics letters. 1996;32(18):1645–1646.
- View Article
- Google Scholar
20. Davey MC, MacKay DJ, editors. Low density parity check codes over GF (q). 1998 Information Theory Workshop (Cat No 98EX131); 1998: IEEE.
21. MacKay DJ, Davey MC. Evaluation of Gallager codes for short block length and high rate applications. Codes, Systems, and Graphical Models: Springer; 2001. p. 113–130.
- View Article
- Google Scholar
22. Hareedy A, Amiri B, Galbraith R, Dolecek L. Non-binary LDPC codes for magnetic recording channels: Error floor analysis and optimized code design. IEEE Transactions on Communications. 2016;64(8):3194–3207.
- View Article
- Google Scholar
23. Huang Q, Song L, Wang Z. Set message-passing decoding algorithms for regular non-binary LDPC codes. IEEE Transactions on Communications. 2017;65(12):5110–5122.
- View Article
- Google Scholar
24. Rehman S, Awan SE, Mumtaz FR, Raja MAZ. A novel application of breadth first algorithm for achieving collision free memory mapping. PloS one. 2019;14(8). pmid:31415583
- View Article
- PubMed/NCBI
- Google Scholar
25. Middleton D. Statistical-physical models of electromagnetic interference. IEEE transactions on Electromagnetic Compatibility. 1977;(3):106–127.
- View Article
- Google Scholar
26. Mâad HB, Goupil A, Clavier L, Gelle G, editors. Asymptotic performance of LDPC codes in impulsive non-Gaussian channel. 2010 IEEE 11th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC); 2010: IEEE.
27. Nakagawa H, Umehara D, Denno S, Morihiro Y, editors. A decoding for low density parity check codes over impulsive noise channels. International Symposium on Power Line Communications and Its Applications, 2005; 2005: IEEE.
28. Silva-Juarez A, Rodriguez-Gomez G, de la Fraga LG, Guillen-Fernandez O, Tlelo-Cuautle E. Optimizing the kaplan–yorke dimension of chaotic oscillators applying de and pso. Technologies. 2019;7(2):38.
- View Article
- Google Scholar
29. Pannu HS, Singh D, Malhi AK. Multi-objective particle swarm optimization-based adaptive neuro-fuzzy inference system for benzene monitoring. Neural computing and applications. 2019;(1):1–11.
- View Article
- Google Scholar
30. Singh D, Kumar V, Kaur M. Classification of COVID-19 patients from chest CT images using multi-objective differential evolution–based convolutional neural networks. European Journal of Clinical Microbiology & Infectious Diseases. 2020;(27):1–11. pmid:32337662
- View Article
- PubMed/NCBI
- Google Scholar
31. Pham TX, Siarry P, Oulhadj H. Segmentation of MR Brain Images Through Hidden Markov Random Field and Hybrid Metaheuristic Algorithm. IEEE Transactions on Image Processing. 2020. pmid:32365028
- View Article
- PubMed/NCBI
- Google Scholar
32. Pathak Y, Shukla PK, Tiwari A, Stalin S, Singh S, Shukla PK. Deep Transfer Learning based Classification Model for COVID-19 Disease. IRBM. 2020.
33. Kaur M, Singh D. Fusion of medical images using deep belief networks. Cluster Computing. 2019; (4):1–5.
- View Article
- Google Scholar
34. Pannu HS, Singh D, Malhi AK. Improved Particle Swarm Optimization Based Adaptive Neuro‐Fuzzy Inference System for Benzene Detection. CLEAN–Soil, Air, Water. 2018;46(5):1700162.
- View Article
- Google Scholar
35. BoussaïD I, Lepagnot J, Siarry P. A survey on optimization metaheuristics. Information sciences. 2013;(237):82–117.
- View Article
- Google Scholar
36. Xu H, Chen C, Zhu M, Bai B, Zhang B. Nonbinary LDPC cycle codes: efficient search, design, and code optimization. Science China Information Sciences. 2018;61(8):089303.
- View Article
- Google Scholar
37. Zhao S, Ma X. Extended construction of array-based non-binary LDPC codes. Electronics Letters. 2019;55(4):196–198.
- View Article
- Google Scholar
38. Chen C, Bai B, Shi G, Wang X, Jiao X. Nonbinary LDPC codes on cages: Structural property and code optimization. IEEE Transactions on Communications. 2015;63(2):364–375.
- View Article
- Google Scholar
39. Wang L, Wang D, Ni Y, Chen X, Cui M, Yang F. Design of irregular QC-LDPC code based multi-level coded modulation scheme for high speed optical communication systems. China Communications. 2019;16(5):106–120.
- View Article
- Google Scholar
40. Carrasco R A, Johnston M. Non-binary error control coding for wireless communication and data storage. John Wiley & Sons. 2008.
41. Middleton D. Procedures for determining the parameters of the first-order canonical models of Class A and Class B electromagnetic interference [10]. IEEE Transactions on electromagnetic compatibility. 1979;(3):190–208.
- View Article
- Google Scholar
42. Drira N, Kotti M, Fakhfakh M, et al. Convergence rates of the efficient global optimization algorithm for improving the design of analog circuits. Analog Integrated Circuits and Signal Processing. 2020; 1–20.
- View Article
- Google Scholar
43. Dimanche V, Goupil A, Clavier L, Gelle G. On detection method for soft iterative decoding in the presence of impulsive interference. IEEE communications letters. 2014;18(6):945–948.
- View Article
- Google Scholar

[ref1] 1. Yu L, Wu J, Zhou A, et al. Massively Distributed Antenna Systems With Nonideal Optical Fiber Fronthauls: A Promising Technology for 6G Wireless Communication Systems. IEEE Vehicular Technology Magazine. 2020; 15(4): 43–51.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Dang S, Amin O, Shihada B, et al. What should 6G be?. Nature Electronics. 2020; 3(1): 20–29.
View Article
Google Scholar

[5] View Article

[6] Google Scholar

[ref3] 3. Xue R, Yu H, Cheng Q. Adaptive coded modulation based on continuous phase modulation for inter-satellite links of global navigation satellite systems. IEEE Access. 2018;6:20652–20662.
View Article
Google Scholar

[8] View Article

[9] Google Scholar

[ref4] 4. Yan Z, Fraire JA, Zhao K, Yan H, Madoery PG, Li W. Distributed Contact Plan Design for GNSSs. IEEE Transactions on Aerospace and Electronic Systems. 2019.

[ref5] 5. Chen S, Peng K, Song J, Zhang Y. Performance analysis of practical QC-LDPC codes: From DVB-S2 to ATSC 3.0. IEEE Transactions on Broadcasting. 2018;65(1):172–178.
View Article
Google Scholar

[12] View Article

[13] Google Scholar

[ref6] 6. Arora K, Singh J, Randhawa YS. A survey on channel coding techniques for 5G wireless networks. Telecommunication Systems. 2019:1–27.
View Article
Google Scholar

[15] View Article

[16] Google Scholar

[ref7] 7. Tapia E, Spetale F, Krsticevic F, Angelone L, Bulacio P. DNA barcoding through quaternary LDPC codes. PloS one. 2015;10(10). pmid:26492348
View Article
PubMed/NCBI
Google Scholar

[18] View Article

[19] PubMed/NCBI

[20] Google Scholar

[ref8] 8. Yang Y, Wang W, Gao X. AMP Dual-turbo iterative detection and decoding for LDPC coded multibeam MSC uplink. China Communications. 2018;15(6):178–186.
View Article
Google Scholar

[22] View Article

[23] Google Scholar

[ref9] 9. Xiao X, Cheng F, Liu F, Lu X, editors. Design and Implementation of BDS RTCM SSR Message. IOP Conference Series: Materials Science and Engineering; 2017: IOP Publishing.

[ref10] 10. Liu L, Amin M, editors. Performance analysis of GPS receivers in impulsive noise. Digital Wireless Communications VII and Space Communication Technologies; 2005: International Society for Optics and Photonics.

[ref11] 11. Mei Z, Johnston M, Le Goff S, Chen L. Performance analysis of LDPC-coded diversity combining on Rayleigh fading channels with impulsive noise. IEEE Transactions on Communications. 2017;65(6):2345–2356.
View Article
Google Scholar

[27] View Article

[28] Google Scholar

[ref12] 12. Wang P, Yin L, Lu J. Spatially Coupled Generalized Low-Density Parity-Check Codes Over Class-A Impulsive Noise Channels. IEEE Access. 2019;7:55906–55915.
View Article
Google Scholar

[30] View Article

[31] Google Scholar

[ref13] 13. Ali I, Lee H, Kil YS, Hussain A, Kim S-H. Spatially coupled LDPC codes for power line communications with impulsive noise. IEICE Electronics Express. 2018:15.20180486.
View Article
Google Scholar

[33] View Article

[34] Google Scholar

[ref14] 14. Declercq D, Fossorier M. Decoding algorithms for nonbinary LDPC codes over GF $(q) $. IEEE transactions on communications. 2007;55(4):633–643.
View Article
Google Scholar

[36] View Article

[37] Google Scholar

[ref15] 15. Melo P, Pimentel C, Alajaji F. LDPC decoding over nonbinary queue-based burst noise channels. IEEE Transactions on Vehicular Technology. 2015;65(1):452–457.
View Article
Google Scholar

[39] View Article

[40] Google Scholar

[ref16] 16. Wijekoon V, Viterbo E, Hong Y, Micheloni R, Marelli A. A Novel Graph Expansion and a Decoding Algorithm for NB-LDPC Codes. IEEE Transactions on Communications. 2019.
View Article
Google Scholar

[42] View Article

[43] Google Scholar

[ref17] 17. Liu Y, Chen W, Wang A, Han C. Combination of High-Order Modulation and Non-Binary LDPC Codes over GF (7) for Non-Linear Satellite Channels. Electronics. 2019;8(12):1396.
View Article
Google Scholar

[45] View Article

[46] Google Scholar

[ref18] 18. Gallager R. Low-density parity-check codes. IRE Transactions on information theory. 1962;8(1):21–28.
View Article
Google Scholar

[48] View Article

[49] Google Scholar

[ref19] 19. MacKay DJ, Neal RM. Near Shannon limit performance of low density parity check codes. Electronics letters. 1996;32(18):1645–1646.
View Article
Google Scholar

[51] View Article

[52] Google Scholar

[ref20] 20. Davey MC, MacKay DJ, editors. Low density parity check codes over GF (q). 1998 Information Theory Workshop (Cat No 98EX131); 1998: IEEE.

[ref21] 21. MacKay DJ, Davey MC. Evaluation of Gallager codes for short block length and high rate applications. Codes, Systems, and Graphical Models: Springer; 2001. p. 113–130.
View Article
Google Scholar

[55] View Article

[56] Google Scholar

[ref22] 22. Hareedy A, Amiri B, Galbraith R, Dolecek L. Non-binary LDPC codes for magnetic recording channels: Error floor analysis and optimized code design. IEEE Transactions on Communications. 2016;64(8):3194–3207.
View Article
Google Scholar

[58] View Article

[59] Google Scholar

[ref23] 23. Huang Q, Song L, Wang Z. Set message-passing decoding algorithms for regular non-binary LDPC codes. IEEE Transactions on Communications. 2017;65(12):5110–5122.
View Article
Google Scholar

[61] View Article

[62] Google Scholar

[ref24] 24. Rehman S, Awan SE, Mumtaz FR, Raja MAZ. A novel application of breadth first algorithm for achieving collision free memory mapping. PloS one. 2019;14(8). pmid:31415583
View Article
PubMed/NCBI
Google Scholar

[64] View Article

[65] PubMed/NCBI

[66] Google Scholar

[ref25] 25. Middleton D. Statistical-physical models of electromagnetic interference. IEEE transactions on Electromagnetic Compatibility. 1977;(3):106–127.
View Article
Google Scholar

[68] View Article

[69] Google Scholar

[ref26] 26. Mâad HB, Goupil A, Clavier L, Gelle G, editors. Asymptotic performance of LDPC codes in impulsive non-Gaussian channel. 2010 IEEE 11th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC); 2010: IEEE.

[ref27] 27. Nakagawa H, Umehara D, Denno S, Morihiro Y, editors. A decoding for low density parity check codes over impulsive noise channels. International Symposium on Power Line Communications and Its Applications, 2005; 2005: IEEE.

[ref28] 28. Silva-Juarez A, Rodriguez-Gomez G, de la Fraga LG, Guillen-Fernandez O, Tlelo-Cuautle E. Optimizing the kaplan–yorke dimension of chaotic oscillators applying de and pso. Technologies. 2019;7(2):38.
View Article
Google Scholar

[73] View Article

[74] Google Scholar

[ref29] 29. Pannu HS, Singh D, Malhi AK. Multi-objective particle swarm optimization-based adaptive neuro-fuzzy inference system for benzene monitoring. Neural computing and applications. 2019;(1):1–11.
View Article
Google Scholar

[76] View Article

[77] Google Scholar

[ref30] 30. Singh D, Kumar V, Kaur M. Classification of COVID-19 patients from chest CT images using multi-objective differential evolution–based convolutional neural networks. European Journal of Clinical Microbiology & Infectious Diseases. 2020;(27):1–11. pmid:32337662
View Article
PubMed/NCBI
Google Scholar

[79] View Article

[80] PubMed/NCBI

[81] Google Scholar

[ref31] 31. Pham TX, Siarry P, Oulhadj H. Segmentation of MR Brain Images Through Hidden Markov Random Field and Hybrid Metaheuristic Algorithm. IEEE Transactions on Image Processing. 2020. pmid:32365028
View Article
PubMed/NCBI
Google Scholar

[83] View Article

[84] PubMed/NCBI

[85] Google Scholar

[ref32] 32. Pathak Y, Shukla PK, Tiwari A, Stalin S, Singh S, Shukla PK. Deep Transfer Learning based Classification Model for COVID-19 Disease. IRBM. 2020.

[ref33] 33. Kaur M, Singh D. Fusion of medical images using deep belief networks. Cluster Computing. 2019; (4):1–5.
View Article
Google Scholar

[88] View Article

[89] Google Scholar

[ref34] 34. Pannu HS, Singh D, Malhi AK. Improved Particle Swarm Optimization Based Adaptive Neuro‐Fuzzy Inference System for Benzene Detection. CLEAN–Soil, Air, Water. 2018;46(5):1700162.
View Article
Google Scholar

[91] View Article

[92] Google Scholar

[ref35] 35. BoussaïD I, Lepagnot J, Siarry P. A survey on optimization metaheuristics. Information sciences. 2013;(237):82–117.
View Article
Google Scholar

[94] View Article

[95] Google Scholar

[ref36] 36. Xu H, Chen C, Zhu M, Bai B, Zhang B. Nonbinary LDPC cycle codes: efficient search, design, and code optimization. Science China Information Sciences. 2018;61(8):089303.
View Article
Google Scholar

[97] View Article

[98] Google Scholar

[ref37] 37. Zhao S, Ma X. Extended construction of array-based non-binary LDPC codes. Electronics Letters. 2019;55(4):196–198.
View Article
Google Scholar

[100] View Article

[101] Google Scholar

[ref38] 38. Chen C, Bai B, Shi G, Wang X, Jiao X. Nonbinary LDPC codes on cages: Structural property and code optimization. IEEE Transactions on Communications. 2015;63(2):364–375.
View Article
Google Scholar

[103] View Article

[104] Google Scholar

[ref39] 39. Wang L, Wang D, Ni Y, Chen X, Cui M, Yang F. Design of irregular QC-LDPC code based multi-level coded modulation scheme for high speed optical communication systems. China Communications. 2019;16(5):106–120.
View Article
Google Scholar

[106] View Article

[107] Google Scholar

[ref40] 40. Carrasco R A, Johnston M. Non-binary error control coding for wireless communication and data storage. John Wiley & Sons. 2008.

[ref41] 41. Middleton D. Procedures for determining the parameters of the first-order canonical models of Class A and Class B electromagnetic interference [10]. IEEE Transactions on electromagnetic compatibility. 1979;(3):190–208.
View Article
Google Scholar

[110] View Article

[111] Google Scholar

[ref42] 42. Drira N, Kotti M, Fakhfakh M, et al. Convergence rates of the efficient global optimization algorithm for improving the design of analog circuits. Analog Integrated Circuits and Signal Processing. 2020; 1–20.
View Article
Google Scholar

[113] View Article

[114] Google Scholar

[ref43] 43. Dimanche V, Goupil A, Clavier L, Gelle G. On detection method for soft iterative decoding in the presence of impulsive interference. IEEE communications letters. 2014;18(6):945–948.
View Article
Google Scholar

[116] View Article

[117] Google Scholar

Figures

Abstract

Introduction

Related work

Theoretical background

AWAN model

Conventional initialization in decoding process

An optimized initialization for LDPC decoding over GF(q) on AWAN channel

Results and discussion

Convergence comparison

Validity analysis

Robustness analysis

Effect of channel parameters

Simplification

Conclusion

Supporting information

S1 Data.

Acknowledgments

References