Analysis of TDMP Algorithm of LDPC Codes Based on Density Evolution and Gaussian Approximation

Based on density evolution analysis of the existing belief propagation (BP) algorithm, the Turbo Decoding Message Passing (TDMP) algorithm was analyzed from the perspective of density evolution and Gaussian approximation, and the theoretical analysis process of TDMP algorithm was given. When calculating the prior message of each layer of the TDMP algorithm, the check message of the previous iteration should be subtracted. Therefore, the result will not be convergent, if the TDMP algorithm is directly analyzed based on density evolution and Gaussian approximation. We researched the TDMP algorithm based on the symmetry conditions to obtain the convergent result. When using density evolution (DE) and Gaussian approximation to analyze the decoding convergence of the TDMP algorithm, we can provide a theoretical basis for proving the superiority of the algorithm. Then, based on the DE theory, we calculated the probability density function (PDF) of the check-to-variable information of TDMP and its simplified algorithm, and then gave it a calculation based on the process of the normalization factor. Simulation results show that the decoding convergence speed of the TDMP algorithm was faster and the iterations were smaller compared to the BP algorithm under the same conditions.


Introduction
Low-density parity-check (LDPC) codes were first proposed by Gallager [1] in 1962. In 1996, MacKay and Neal [2] re-researched LDPC codes and discovered that irregular LDPC performs better than regular LDPC [3,4], the irregular LDPC codes can perform close to the Shannon limit error performance. Based on this, Vibha [5] proposed an extended method to convert the regular LDPC codes into the irregular LDPC codes. On the basis of the Slepian-Wolf coding problem, a coding theorem of finite-substitution linear coding was proposed. The research shows that there exist a series of linear coders for any set of finite-correlation discrete memoryless sources [6]. In Reference [7], rate-compatible LDPC coding was implemented by graphics expansion to ensure coding performance and to simplify the structure of the encoder. Based on the probabilistic iterative decoding algorithm, MacKay [8] proposed the belief propagation (BP) algorithm, which is currently of great academic interest. In BP algorithm decoding, information is passed along the edge of Tanner graph until the decoding end [9]. The BP decoding schedules are designed according to the reliability of check nodes, which can improve the error-correcting performance of BP decoding [10]. In addition, only one subset of check nodes is updated in each iteration, which reduces the resource consumption in the decoding process.

The Density Evolution of the BP Algorithm
A regular LDPC code can be represented by (dv, dc), where dv and dc is the degree of variable and check nodes, respectively. The iteration expression of the check messages and the variable messages are as follows [11] tanh u (l) j Entropy 2019, 21, 457 v (0) i represents the initial message of channel received by i-th variable node. Assuming that P (l) (v) and Q (l) (u) are the probability densities of v (l) i and u (l) j , respectively. We can obtain the result of density evolution from the check messages to the variable messages as [11] P (l) (v) = P (0) ⊗ Q (l)⊗(dv−1) , ⊗ represents the convolution operation. We can transform the convolution operation in Equation (3) to the multiplication operation based on the Fourier transform [11] The process of density evolution from the variable messages to the check messages was introduced by Richardson in detail in Reference [11]. Thus, we can calculate the bit error rate (BER) as ρ i x i−1 , respectively. We must consider the degree distribution to update probability density of the variable messages and the check messages [11]. The BER of irregular LDPC codes is also calculated by Equation (5).

The Gaussian Approximation of the BP Algorithm
For detailed analysis of BP algorithm one can refer to Reference [16]. According to Reference [16], we can get j , respectively, and m 0 is the mean of channel initial messages. Defining φ(x) in Reference [16] and take the expectation of Equation (1) to get Equation (7), where φ(x) is shown in Equation (8).
The BER of LDPC codes can be calculated as [21]

The Decoding Procedure of the TDMP Algorithm
Before discussing density evolution and Gaussian approximation of the TDMP algorithm, firstly, we simply review the decoding procedure of the TDMP algorithm [26,27]. Check matrix H with m rows and n columns is composed of c check sub-matrices in series, and each sub-matrix has k rows, then m = c × k. Meanwhile, each sub-matrix can be regarded as a horizontal layer, and the main idea of the TDMP algorithm is to decode each layer successively.
In iteration decoding, we first update the check messages in the first layer, and then update the variable messages in this layer immediately. After that, we use the updated variable messages to calculate the prior messages which will be sent to the next layer for updating the check messages until the last layer. Then the updated prior messages are the outputs of iteration decoding. The TDMP algorithm directly uses the updated information of the previous layer when updating the current layer information, thereby speeding up the convergence speed of the decoding. The message passing scheme of the TDMP algorithm is shown in Figure 1. with m rows and n columns is composed of c check sub-matrices in series, and each sub-matrix has k rows, then m c k =. Meanwhile, each sub-matrix can be regarded as a horizontal layer, and the main idea of the TDMP algorithm is to decode each layer successively.
In iteration decoding, we first update the check messages in the first layer, and then update the variable messages in this layer immediately. After that, we use the updated variable messages to calculate the prior messages which will be sent to the next layer for updating the check messages until the last layer Decoding the convergence speed can be increased by utilizing the prior messages of the previous layer directly to update the check messages of the current layer. Meanwhile, the essence of layered decoding is optimizing the procedure of message passing but not changing the implementation of the check messages update in each layer. The BP algorithm can also use layered decoding, and the analysis of density evolution and Gaussian approximation of the TDMP algorithm is based on the BP algorithm. Thus, the procedure of the check messages update in each layer of the TDMP algorithm is the same as that of the BP algorithm which is introduced in Section 2. Therefore, in the next section, the major work is to illustrate the variable messages updated in each layer.

The Analysis of Density Evolution of the TDMP Algorithm
The variable messages updated [28,29] in each layer is shown in Equation (10), then prior messages updated in the next layer [30] is shown in Equation (11).
l represents the number of iterations, and j represents the current number of layers. c  is the channel initial message which initializes the variable message before decoding. According to the symmetry condition of density evolution [24], we can get Equation (12).
When updating to the last layer, the posterior message ( ) l  is an output of the l-th iteration decoding.
According to the knowledge related to probability, the probability density of the sum of the two statistically independent variables X and Y is convolution of probability density of X and Y. Since the variables in Equations (10) and (12) satisfy the condition of statistically independent distribution, we can obtain Decoding the convergence speed can be increased by utilizing the prior messages of the previous layer directly to update the check messages of the current layer. Meanwhile, the essence of layered decoding is optimizing the procedure of message passing but not changing the implementation of the check messages update in each layer. The BP algorithm can also use layered decoding, and the analysis of density evolution and Gaussian approximation of the TDMP algorithm is based on the BP algorithm. Thus, the procedure of the check messages update in each layer of the TDMP algorithm is the same as that of the BP algorithm which is introduced in Section 2. Therefore, in the next section, the major work is to illustrate the variable messages updated in each layer.

The Analysis of Density Evolution of the TDMP Algorithm
The variable messages updated [28,29] in each layer is shown in Equation (10), then prior messages updated in the next layer [30] is shown in Equation (11).
l represents the number of iterations, and j represents the current number of layers. λ c is the channel initial message which initializes the variable message before decoding.
According to the symmetry condition of density evolution [24], we can get Equation (12).
When updating to the last layer, the posterior message γ (l) is an output of the l-th iteration decoding.
According to the knowledge related to probability, the probability density of the sum of the two statistically independent variables X and Y is convolution of probability density of X and Y. Since the variables in Equations (10) and (12) satisfy the condition of statistically independent distribution, we can obtain Use the Fourier transform to convert the convolution operation in Equations (13) and (14) into a multiplication operation, as shown in Equations (15) and (16) P v When updating to the last layer, the probability density P γ (l) of the posterior message γ (l) is the final output probability density at the l-th iteration decoding. Thus, we can calculate BER based on Equation (5).

The Analysis of Gaussian Approximation of TDMP Algorithm
According to Section 3.2, we derived Equation (12). By taking expectation of Equations (10) and (12), we can obtain m We can calculate BER based on Equation (9).

Calculation of TDMP Normalization Factor Based on Density Evolution
When calculating the check node information, the TDMP algorithm uses the same nonlinear function as BP, which makes the hardware implementation difficult. The simplified algorithm in Reference [31] simplifies the nonlinear function and provides a possibility for hardware implementation. On the basis of the previous analysis, this paper presents the calculation process of the normalized factor of the simplified TDMP algorithm in Reference [31] based on the density evolution theory. The detailed process is as follows.

The Initialization
Under the BIAWGN channel and Binary Phase Shift Keying (BPSK) modulation, code words c i are mapped to x i = (−1) c i ,i = 1, 2, · · ·n. The received code words are y i = x i + n i , n i is independent and identically distributed Gaussian random variable and bilateral power spectral density is n 0 2 . Hence, y i is a Gauss variable with a mean value of 1 and variance of σ 2 [18].
Assuming that the average energy of the signal bit transmitted by the channel is Eb, N 0 is the power spectral density of noise and the code rate is r. The relationship between the signal-noise ratio (SNR) and noise variance is σ 2 = 1 2r(Eb/N0) . Suppose that the information is an equal probability distribution [21], where Pr(x i = +1) = Pr(x i = −1) = 1 2 . The initial information is: The log-likelihood ratio message is: At this time, the initial information of the LLR is Gaussian random variable [21], whose mean value is 2 σ 2 and variance is 4 σ 2 , that is, it satisfies LLR ∼ N(µ, 2µ), where µ = 2 σ 2 . According to the Gaussian distribution of LLR and relevant knowledge of probability theory, its is the probability density function of the standard Gaussian distribution, the relationship between q(t) and f (x) is:

Normalization Factor Calculation
A TDMP simplification algorithm without non-linear functions is given in Reference [30], which can be expressed as The update of the check node information can be simplified to In order to improve the decoding performance of the TDMP simplification algorithm, the normalization factor α is used for improving the calculation of the check node information. Its calculation formula is as shown in Equation (23). The following is the calculation of the normalization factor α based on density evolution [24].
To facilitate calculation and illustration, variables L 1 and L 2 are introduced to represent min j=1,2,...dc−1 |v j |, According to the knowledge of probability theory, the probability distribution of |L 1 | is known [24].
Thus, its expectation is which using integral properties can be written as And then the calculation of E(L 3 ). Extend tanh −1 into a series form According to the series expansion [18] According to mathematical knowledge and relevant literature, such as [21,24,25], the high-order expansion terms of the Taylor series are very small (negligible), so the first few terms of the series are usually taken for approximate calculation. In this paper, the first five terms were selected as approximate results by referring to Reference [25].

Simulation Results
The IEEE 802.16e standard defines six code rates (1/2, 2/3A, 2/3B, 3/4A, 3/4B, and 5/6) of the LDPC codes. In this paper, density evolution and Gaussian approximation simulation of LDPC codes with six code rates under 802.16 standard were carried out under the BIAWGN channel. Then, from the perspective of density evolution and Gaussian approximation, the convergence of BP algorithm and TDMP algorithm for 1/2 code rate LDPC codes under 802.16 and DTMB standards was compared. In the simulation, the maximum number of iterations of BP algorithm and TDMP algorithm were set as 20 and 10, respectively. Finally, the calculation of the normalization factor and the performance comparison of the TDMP algorithm were simulated.

The Simulation Results of Density Evolution of the BP Algorithm and the TDMP Algorithm
The performance curves of the BP algorithm and the TDMP algorithm at six code rates are shown in Figure 2a,b respectively. The values of the signal-noise ratio (SNR) of rate 1/2, rate 2/3A (or 2/3B), rate 3/4A (or 3/4B) and rate 5/6 are 0 to 2.5(dB), 0.5 to 3.0(dB), 1.0 to 3.5(dB), and 1.5 to 4.0(dB), respectively.

The Simulation Results of Density Evolution of the BP Algorithm and the TDMP Algorithm
The performance curves of the BP algorithm and the TDMP algorithm at six code rates are shown in Figure 2a,b respectively. The values of the signal-noise ratio (SNR) of rate 1/2, rate 2/3A (or 2/3B), rate 3/4A (or 3/4B) and rate 5/6 are 0 to 2.5(dB), 0.5 to 3.0(dB), 1.0 to 3.5(dB), and 1.5 to 4.0(dB), respectively. As can be seen from Figure 2, as the SNR increases, the BER curve of the TDMP algorithm decreased faster than the BP algorithm. The following is a comparison of the convergence of the two algorithms under a single code rate. The code rate 2/3B and SNR 2.0(dB) were chosen in discussion. The relationship between BER and the iterations is shown in Figure 3. Obviously, the BER convergence speed of the TDMP algorithm was faster than the BP algorithm. When the two algorithms achieved the same BER, such as 10 −5 , the number of iterations of the BP algorithm was 13, but the number of the TDMP algorithm was only 7. As can be seen from Figure 2, as the SNR increases, the BER curve of the TDMP algorithm decreased faster than the BP algorithm. The following is a comparison of the convergence of the two algorithms under a single code rate. The code rate 2/3B and SNR 2.0(dB) were chosen in discussion. The relationship between BER and the iterations is shown in Figure 3. Obviously, the BER convergence speed of the TDMP algorithm was faster than the BP algorithm. When the two algorithms achieved the same BER, such as 10 −5 , the number of iterations of the BP algorithm was 13, but the number of the TDMP algorithm was only 7. When expected BER reaches to 10 −5 at code rate 2/3B under different SNRs, iterations needed for the BP algorithm and the TDMP algorithm are shown in Figure 4. The BP algorithm and the TDMP algorithm cannot reach to expected BER within the maximum iterations, when SNR are 0, 0.5, 1.0 and 1.5(dB), respectively. When SNR is greater than 1.5(dB), the TDMP algorithm needs fewer iterations compared with the BP algorithm.   When expected BER reaches to 10 −5 at code rate 2/3B under different SNRs, iterations needed for the BP algorithm and the TDMP algorithm are shown in Figure 4. The BP algorithm and the TDMP algorithm cannot reach to expected BER within the maximum iterations, when SNR are 0, 0.5, 1.0 and 1.5 (dB), respectively. When SNR is greater than 1.5 (dB), the TDMP algorithm needs fewer iterations compared with the BP algorithm.  From the simulation results of Figures 3 and 4, we can prove that decoding convergence speed of the TDMP algorithm was faster than the BP algorithm from the perspective of density evolution theory.
In order to verify the results, we simulated and analyzed the LDPC codes with 1/2 code rate under the 802. 16    From the simulation results of Figures 3 and 4, we can prove that decoding convergence speed of the TDMP algorithm was faster than the BP algorithm from the perspective of density evolution theory.
In order to verify the results, we simulated and analyzed the LDPC codes with 1/2 code rate under the 802. 16 Under the DTMB standard, the degree distribution of check nodes and variable nodes are shown in Equations (36) and (37) λ(x) = 0.3382x 2 + 0.2618x 3 + 0.4x 10 (36) ρ(x) = 0.1273x 6 + 0.8727x 7 (37) Figures 5 and 6 compare the BP and TDMP algorithms under the two standards. When the expected BER reaches to 10 −5 , the number of iterations required under different SNRs is shown in Figure 5. When the SNR is 2 dB, the relationship between BER and iteration is shown in Figure 6.  Figure 5. When the SNR is 2 dB, the relationship between BER and iteration is shown in Figure 6.    By comparing the simulation results under the two standards in Figures 5 and 6, it can be seen that when the SNR is 2 dB, the convergence speed of TDMP algorithm was faster than that of BP algorithm under the two standards. In addition, as can be seen from Figure 5, when the was greater than 1.5 dB, the number of iterations required by the TDMP algorithm is about 1/2 of the BP algorithm.

The Simulation Results of the Gaussian Approximation of the BP Algorithm and the TDMP Algorithm
Gaussian approximation algorithm is a simplified density evolution algorithm. Next, the correctness of the TDMP algorithm was verified from the point of view of Gauss approximation. Through Gaussian approximation analysis, the performance curves of the BP algorithm and the TDMP algorithm at six code rates are shown in Figure 7a,b, respectively. It can be seen from Figure  7 that the BER curve of the TDMP algorithm converges faster. When the code rate is 2/3B and the SNR is 2.0 (dB), the relationship between the BER and the number of iterations of the two algorithms is shown in Figure 8. When the expected BER reaches to 10 −5 at the code rate 2/3B under different SNRs, iterations needed for the BP algorithm and the TDMP algorithm are shown in Figure 9. By comparing the simulation results under the two standards in Figures 5 and 6, it can be seen that when the SNR is 2 dB, the convergence speed of TDMP algorithm was faster than that of BP algorithm under the two standards. In addition, as can be seen from Figure 5, when the was greater than 1.5 dB, the number of iterations required by the TDMP algorithm is about 1/2 of the BP algorithm.

The Simulation Results of the Gaussian Approximation of the BP Algorithm and the TDMP Algorithm
Gaussian approximation algorithm is a simplified density evolution algorithm. Next, the correctness of the TDMP algorithm was verified from the point of view of Gauss approximation. Through Gaussian approximation analysis, the performance curves of the BP algorithm and the TDMP algorithm at six code rates are shown in Figure 7a,b, respectively. It can be seen from Figure 7 that the BER curve of the TDMP algorithm converges faster. When the code rate is 2/3B and the SNR is 2.0 (dB), the relationship between the BER and the number of iterations of the two algorithms is shown in Figure 8. When the expected BER reaches to 10 −5 at the code rate 2/3B under different SNRs, iterations needed for the BP algorithm and the TDMP algorithm are shown in Figure 9.
The analysis of Figures 7-9 is similar as that of Figures 2-4 in Section 5.1. Thus, the performance of TDMP algorithm for multi-dimensional density evolution and one-dimensional Gaussian approximation process was similar. The analysis of Figures 7 to 9 is similar as that of Figures 2 to 4 in Section 5.1. Thus, the performance of TDMP algorithm for multi-dimensional density evolution and one-dimensional Gaussian approximation process was similar.        Based on Gaussian approximation theory, the BP and TDMP algorithms of LDPC codes with 1/2 code rate under two standards are simulated. Figure 10 shows the number of iterations required for the two algorithms to achieve the desired BER at different SNRs, where the expected BER is 10 −5 . When the code rate is 1/2 and the SNR is 2.0 dB, the relationship between the BER and the number of iterations is as shown in Figure 11. The standard adopted and the corresponding degree distribution function are the same as Section 5.1. Based on Gaussian approximation theory, the BP and TDMP algorithms of LDPC codes with 1/2 code rate under two standards are simulated. Figure 10 shows the number of iterations required for the two algorithms to achieve the desired BER at different SNRs, where the expected BER is 10 −5 . When the code rate is 1/2 and the SNR is 2.0 dB, the relationship between the BER and the number of iterations is as shown in Figure 11. The standard adopted and the corresponding degree distribution function are the same as Section 5.1.    The analysis of Figures 10 and 11 was similar to that of Figures 5 and 6 in Section 5.1. Combined with the simulation results of Sections 5.1 and 5.2, we can get the following conclusions. For LDPC codes under different standards, we only need to know the degree distribution of LDPC codes under the current standard. Therefore, for other standard LDPC codes, the density evolution theory can be used to analyze according to the degree distribution function. In terms of LDPC code structure, the TDMP algorithm is applicable to structured layered LDPC codes [32,33]. The currently known standard LDPC codes conform to this feature, so we can conclude that the density evolution and Gaussian approximation of the proposed TDMP algorithm are applicable to different standards. The analysis of Figures 10 and 11 was similar to that of Figures 5 and 6 in Section 5.1. Combined with the simulation results of Sections 5.1 and 5.2, we can get the following conclusions. For LDPC codes under different standards, we only need to know the degree distribution of LDPC codes under the current standard. Therefore, for other standard LDPC codes, the density evolution theory can be used to analyze according to the degree distribution function. In terms of LDPC code structure, the TDMP algorithm is applicable to structured layered LDPC codes [32,33]. The currently known standard LDPC codes conform to this feature, so we can conclude that the density evolution and Gaussian approximation of the proposed TDMP algorithm are applicable to different standards.

The Actual Algorithm Simulation of the BP Algorithm and the TDMP Algorithm
In this section, we simulate the LDPC codes with the code length 2304. With the code rates 2/3B and the SNR 3.0 (dB), the relationship between BER and iterations is shown in Figure 12. Here, SNR was not the same as that in Sections 5.1 and 5.2, because in the actual algorithm simulation, it cannot reach to the expected BER when SNR is 2.0 (dB). Though SNR was different, it did not influence the analysis of the convergence trend of the BP algorithm and the TDMP algorithm. As shown in Figure 12, the BER convergence speed of the TDMP algorithm in the actual algorithm simulation was also faster than the BP algorithm. When the actual algorithm can decode correctly at the code rate 2/3B under different SNRs, the iterations needed for the BP algorithm and the TDMP algorithm are shown in Figure 13. In this section, we simulate the LDPC codes with the code length 2304. With the code rates 2/3B and the SNR 3.0 (dB), the relationship between BER and iterations is shown in Figure 12. Here, SNR was not the same as that in Sections 5.1 and 5.2, because in the actual algorithm simulation, it cannot reach to the expected BER when SNR is 2.0 (dB). Though SNR was different, it did not influence the analysis of the convergence trend of the BP algorithm and the TDMP algorithm. As shown in Figure  12, the BER convergence speed of the TDMP algorithm in the actual algorithm simulation was also faster than the BP algorithm. When the actual algorithm can decode correctly at the code rate 2/3B under different SNRs, the iterations needed for the BP algorithm and the TDMP algorithm are shown in Figure 13.   The analysis of Figure 13 was similar to that of Figure 4 in Section 5.1. The analysis of the results from the actual algorithm simulation can prove the correctness of the density evolution and Gaussian approximation analysis of the TDMP algorithm. Table 1 shows the number of iterations required for the BP and TDMP algorithms in DE, GA, and actual simulation when the code rate was 2/3 under the 802.16 standard.  The analysis of Figure 13 was similar to that of Figure 4 in Section 5.1. The analysis of the results from the actual algorithm simulation can prove the correctness of the density evolution and Gaussian approximation analysis of the TDMP algorithm. Table 1 shows the number of iterations required for the BP and TDMP algorithms in DE, GA, and actual simulation when the code rate was 2/3 under the 802.16 standard. The relationship between the number of iterations required by BP and TDMP algorithm can be seen in Table 1 when the desired BER was reached under three simulations. When the SNR was greater than 2 dB, the number of iterations of the BP algorithm was about twice that of the TDMP algorithm.

The Simulation Results of TDMP Normalization Factor
In this section, we calculated the normalization factor when the SNR was 1.7, 1.8, and 1.9 dB based on density evolution. The simulation results show that the normalization factor was stable at around 0.5, as shown in Figure 14, which can be used in the following simulations. Based on the density evolution process, the relationship between BER and the number of iterations of the BP algorithm, TDMP algorithm, and normalization factor simplification algorithm (TDMP-N) was simulated, the results are shown in Figure 15. Under the AWGN channel, the BER performance of LDPC codes with a code length of 2304 at different SNRs was simulated, as shown in Figure 16. The average number of iterations at different SNRs was simulated as shown in Figure 17. The convergence speed was analyzed from the perspective of density evolution. The simulation results in Figure 15 show that TDMP-N converged slightly faster than TDMP. The simulation results show that the TDMP-N convergence speed was slightly faster than TDMP, and the average iteration number was similar to TDMP. It indicates that the calculation process of TDMP-N was simpler and more convenient for hardware implementation in the case of less loss of BER performance.
LDPC codes with a code length of 2304 at different SNRs was simulated, as shown in Figure 16. The average number of iterations at different SNRs was simulated as shown in Figure 17. The convergence speed was analyzed from the perspective of density evolution. The simulation results in Figure 15 show that TDMP-N converged slightly faster than TDMP. The simulation results show that the TDMP-N convergence speed was slightly faster than TDMP, and the average iteration number was similar to TDMP. It indicates that the calculation process of TDMP-N was simpler and more convenient for hardware implementation in the case of less loss of BER performance.

Conclusion
This paper used the density evolution theory and the Gaussian approximation theory to analyze the TDMP algorithm. The TDMP algorithm optimizes information transmission in the decoding process, that is, the update of the current layer information directly uses the updated information of the previous layer, thereby accelerating the convergence speed of the decoding. Based on the density

Conclusions
This paper used the density evolution theory and the Gaussian approximation theory to analyze the TDMP algorithm. The TDMP algorithm optimizes information transmission in the decoding process, that is, the update of the current layer information directly uses the updated information of the previous layer, thereby accelerating the convergence speed of the decoding. Based on the density evolution analysis of the existing BP algorithm, this paper analyzed the TDMP algorithm from the perspective of density evolution, giving the theoretical analysis process of the TDMP algorithm, and verifying the analysis results of the TDMP algorithm through simulation comparison with the BP algorithm in this paper. In addition, a simplified TDMP algorithm based on a normalized factor was proposed. The simulation results show that the proposed optimization algorithm had similar performance with the TDMP algorithm. Moreover, it did not involve the operation of non-linear functions and was easier to implement in the hardware. We further simulated six code rates of LDPC codes under the IEEE 802.16e standard and obtained the performance curves of the BP algorithm and the TDMP algorithm based on density evolution and Gaussian approximation, respectively. Thus, we can prove that the decoding convergence speed of the TDMP algorithm was faster than the BP algorithm; meanwhile average iterations were fewer from the perspective of density evolution theory and Gaussian approximation, respectively. According to the structure of LDPC codes under the known standard and the simulation results in the text, we can conclude that the density evolution and Gaussian approximation of the proposed TDMP algorithm can be applied to different standards.