In this section, simulation results are presented to evaluate the BER performance and complexities of the proposed algorithms. The following simulation is performed based on binary phase shift keying (BPSK) modulation and additive white Gaussian noise (AWGN) channel. The code construction is carried out via the method proposed in [
31]. In [
31], the Gaussian Approximation method is proposed to construct polar codes, aimed at minimizing the exact block error ratio (BLER) rather than the upper bound of the BLER, which is suitable for AWGN channel and has lower complexity than the Density Evolution method [
31]. The code rate is 1/2, and the maximum number of iterations is set at 40. In those simulations, we employ the BP algorithm in [
15], the MS algorithm in [
17], the SMS algorithm with the scale parameter
in [
18], and the low complexity BP (LCBP) decoding algorithm in [
32].
4.1. Performance Comparison
Figure 3,
Figure 4,
Figure 5 and
Figure 6 show the BER performance of different algorithms with different code lengths. Firstly, the BER performance of the proposed Exp-BP algorithm is analyzed in detail. As shown in
Figure 3, when the code length is 1024 and
≤ 1.3 dB, the BER performance of the proposed Exp-BP algorithm basically coincides with that of the BP algorithm. However, with the increasing of the SNR, i.e.,
, the BER performance of the proposed Exp-BP algorithm is inferior to that of the BP algorithm.
In
Figure 3, the BER performance of the proposed Exp-BP algorithm outperforms that of the SMS algorithm in the low SNR region, but it is not as good as that of the SMS algorithm in the high SNR region. In addition, the BER performance of the BP algorithm, the SMS algorithm, and the proposed Exp-BP algorithm are better than that of the MS algorithm.
According to the simulation results in
Figure 3 and
Figure 4, when the code block length is 2048, the BER curve gap between the Exp-BP algorithm and the BP algorithm is smaller than that when the code block length is 1024. Meanwhile, in the low SNR region, the gap between the BER curves of the proposed Exp-BP algorithm and the SMS algorithm increases with the increasing of the code length. As shown in
Figure 5, when the code length is 4096, the proposed Exp-BP algorithm can obtain BER performance approximating that of the BP algorithm.
From
Figure 3,
Figure 4 and
Figure 5, we can observe that the BER performance of the proposed Exp-BP algorithm and the BP algorithm gradually approach with the increasing of the code block length. The proposed Exp-BP algorithm has better BER performance than the SMS algorithm in the low SNR region. Similar BER performance can be obtained in the high SNR region by increasing the code length. Meanwhile, under different code block lengths, the BER performance of the proposed Exp-BP algorithm is better than that of the MS algorithm.
Secondly, the following part analyzes the BER performance of the proposed QF-BP algorithm.
According to
Figure 3, when the code length is 1024, the BER performance of the proposed QF-BP algorithm is lower than that of the BP algorithm. The simulation results in
Figure 4 show that, compared with the BP algorithm, the BER performance of the proposed QF-BP algorithm decreases in the low SNR region, while maintaining a similar BER performance at high SNR. According to
Figure 3 and
Figure 4, the BER performance of the proposed QF-BP algorithm is better than that of the MS algorithm under different code lengths. Meanwhile, the BER performance of the proposed QF-BP algorithm in the low SNR region outperforms that of the SMS algorithm. With the increasing of the code length, the proposed QF-BP algorithm can obtain similar BER performance with the SMS algorithm in the high SNR region.
As shown in
Figure 6, the BER performance of the proposed Exp-BP algorithm is slightly better than that of the LCBP algorithm with
. Similarly, when
≤ 1.7 dB, the BER performance of the proposed Exp-BP algorithm is similar to that of the LCBP algorithm with
, but is slightly lower than that of the LCBP algorithm with
in the high SNR region. In addition, compared with the LCBP algorithm with
, the proposed QF-BP algorithm has similar BER performance in the high SNR region.
Collectively, combining with the above simulation graphics and analysis, it can be concluded that for long code length, the BER performance of the proposed Exp-BP algorithm is basically consistent with that of the original BP algorithm. The BER performance of the proposed QF-BP algorithm in the high SNR region approach matches that of the original BP algorithm for long code block length.
4.2. Complexities Analysis
This part compares the computational complexities of the proposed Exp-BP algorithm and the proposed QF-BP algorithm with other related algorithms.
Firstly, we analyze the operation types contained in the iterative updating formulas of different algorithms. Secondly, the computation complexities of different algorithms in each iteration are explored in detail. Finally, this part provides the percentage of decline in computational complexities of different algorithms compared with the BP algorithm.
When comparing the computation complexities, it is necessary to deform the iterative updating formulas of different algorithms. Especially, in this paper, the iterative updating formula of the LCBP algorithm is the same as that of the BP algorithm. Specific conversion formulas are as follows.
The updating formula of the BP algorithm [
15] is
The updating formula of the MS algorithm [
17] is
The updating formula of the SMS algorithm [
18] is
The updating formula of the proposed Exp-BP algorithm is
The updating formula of the proposed QF-BP algorithm is
Table 4 represents the number of operation types contained in the iterative updating formula of different algorithms.
Firstly, special operations in the updating formula are analyzed. As can be seen from
Table 4, there are two special operations in the BP algorithm, and the proposed Exp-BP decoding algorithm contains a non-linear operation according to (
32). Compared with the BP algorithm, the most significant feature of the proposed QF-BP algorithm is that no special operations (ln and exp) are contained in the updating formula. According to
Table 4, there are also no special operations in the iterative updating formulas of the MS algorithm and the SMS algorithm.
Secondly, the number of additions and multiplications of the proposed Exp-BP algorithm is larger than that of the BP algorithm. Compared with the MS algorithm and the SMS algorithm, the BER performance of the proposed Exp-BP algorithm is improved at the cost of increasing computational complexity.
In addition, this paper analyzes the overall operations of different algorithms in each iteration. In the LCBP decoding algorithm, the number of operations required in each iteration is influenced by the code length, the SNR, the threshold, the code rate, etc. Therefore, in the following comparisons, we consider the approximate average complexity of the LCBP algorithm with
in [
32].
Table 5 summarizes the number of operation types contained in different algorithms in each iteration. As shown in
Table 5, compared with the BP algorithm, the number of special operations of the proposed Exp-BP algorithm is
in each iteration, which is less than
in the BP algorithm, only at cost of some additions and multiplications. In
Table 5, compared with the BP algorithm, the number of special operations reduced by the proposed QF-BP algorithm is
, and the additions and multiplications are also reduced by
and
, respectively. Compared with the LCBP algorithm, the proposed Exp-BP algorithm reduces the types of non-linear operations, and the proposed QF-BP algorithm eliminates non-linear operations in the iterative updating formula at the cost of some multiplications and table look-up operations.
According to the analysis in
Table 4 and
Table 5, the proposed Exp-BP algorithm reduces the types of non-linear operations and converts non-linear operations into additions and multiplications compared with the BP algorithm. The most significant advantage of the proposed QF-BP algorithm is that the updating formula does not contain non-linear operations, which effectively reduces the non-linear complexity. Compared with the MS algorithm and the SMS algorithm, the proposed QF-BP algorithm has no comparison operations in the iterative updating formula of the message. The number of table look-up operations of the proposed QF-BP algorithm is larger than the number of comparison operations of the MS algorithm and the SMS algorithm, but the corresponding BER performance is improved.
To clarify the decoding complexities of the proposed algorithms,
Table 6 shows the percentage of decline in different operations of different algorithms compared with the BP algorithm. In
Table 6, ’+’ represents the proportion of incremental operations of the decoding algorithm compared with the BP decoding algorithm, and ’−’ represents the proportion of reduced operations. As can be seen from
Table 6, compared with the BP algorithm, the proposed Exp-BP algorithm increases additions by
and multiplications by
, respectively, but reduces the ln operation by
. The additions and multiplications of the proposed QF-BP algorithm are
and
less than that of the BP algorithm, respectively, and special operations are eliminated in the iterative updating formula at cost of some table look-up operations. The reduced proportion of non-linear operations of the proposed QF-BP algorithm is higher than that of the LCBP decoding algorithm. In the FPGA hardware implementation, the numbers of required clocks (CLK) for addition, multiplication, sign, absolute, compare, table Look-up, ln, and exp operations are generally set as 1, 3, 1, 2, 1, 2, 18, and 14, respectively. Based on this, the percentages of decline in the number of required CLK in the hardware implementation for different decoding algorithms compared with the BP algorithm are presented in
Table 6. As can be seen from
Table 6, the percentage of decline in the number of required CLK of the proposed Exp-BP algorithm is lower than that of the LCBP algorithm, while having slightly better BER performance. Moreover, compared with the BP algorithm, both the proposed Exp-BP algorithm and the proposed QF-BP algorithm reduce the number of required CLK, which facilitates the hardware implementation.
Compared with the MS algorithm and the SMS algorithm, the proposed Exp-BP algorithm increases the computational complexity, but the BER performance is improved. Meanwhile, the proposed QF-BP algorithm replaces the comparison operation with the table look-up operation, and the overall reduction rate is slightly lower than that of the SMS algorithm. According to the simulation results, the BER performance of the proposed QF-BP algorithm is better than that of the SMS algorithm in the low SNR region.
In conclusion, compared with the BP algorithm, the proposed Exp-BP algorithm reduces the types of non-linear functions in the iterative updating formula and reduces the complexity of non-linear operations. In the case with long code length, the BER performance of the proposed Exp-BP algorithm is similar to that of the BP algorithm. Compared with the BP algorithm, the updating formula of the proposed QF-BP algorithm does not contain non-linear operations. Moreover, in the high SNR region, by increasing of the code length, the BER performance of the proposed QF-BP algorithm approximates that of the BP algorithm. The proposed QF-BP algorithm reduces the computational complexity and eases the implementation of polar codes in practical hardware. The BER performance of the proposed Exp-BP algorithm is better than that of the proposed QF-BP algorithm in the low SNR region.