Optical fiber communication method and system based on low-complexity joint decision decoding, and storage medium

Abstract

The invention discloses an optical fiber communication method and system based on low-complexity joint decision decoding and a storage medium, and belongs to the technical field of optical fiber communication, and the method comprises the steps: at a transmitting end, superposing large and small power signals after QPSK modulation according to different mapping rules, and obtaining a 16QAM superposed signal; at a receiving end, obtaining symbol distribution corresponding to high-power and low-power signals in the 16QAM superposed signals; decoding a high-power signal overlapping area and a low-power signal symbol point by adopting a soft decision strategy, and decoding a high-power signal non-overlapping area symbol point by adopting a hard decision strategy, the method for carrying out soft decision on the symbol points in the overlapping area comprises the following steps: predicting LLR values of all bits of the symbol points through a pre-trained RBF neural network; the decoding calculation complexity is reduced, the decoding accuracy is improved at the same time, and high-speed transmission and reliability of optical fiber communication are guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber communication, and in particular relates to an optical fiber communication method, system and storage medium based on low-complexity joint decision decoding. Background Art

[0002] The increasing data traffic generated by network applications has laid the foundation for the next generation of optical networks with higher speed, flexibility and reliability. As one of the key enabling technologies of the physical layer of next generation optical communications, superposition coding technology has attracted widespread attention from academia and industry. This transmission method can combine multiple modulation schemes into one modulation scheme to provide higher data transmission rate and lower error rate. In this way, the system's demand for additional spectrum resources is reduced, and the decoding process is more flexible. At the same time, this coding modulation method is also one of the key technologies for realizing Orthogonal Frequency Division Multiplexing Non-Orthogonal Multiple Access (OFDM-NOMA). NOMA technology can be divided into power domain (PD-NOMA) and code domain (CD-NOMA). PD-NOMA is to give different powers to different users in the power domain and then superimpose them. By multiplexing the data of multiple users in the power domain and frequency domain, the PD-OFDM-NOMA system has the advantages of higher spectrum efficiency, more flexible access capability and low latency. This new non-orthogonal communication transmission method can greatly reduce unnecessary spectrum efficiency consumption and can be applied to large-scale access communication scenarios. As an important technology for fiber-optic communication, it effectively solves the problem of orthogonal multiple access technology limiting resource utilization and setting an upper limit for network users, allowing the system to reach a better working point with optimal spectrum efficiency and energy efficiency.

[0003] Receivers based on cascade interference cancellation (SIC) technology are now widely used in NOMA systems, but this method detects multiplexed signals in sequence, which can lead to error propagation. Since the decoding process of superposition coding is very flexible, multi-stage decoding algorithms and iterative decoding algorithms can be used. To ensure the accuracy of decoding, most studies use decoding methods based on forward error correction coding (FEC), which are further divided into hard decision schemes and soft decision schemes. In the hard decision decoding scheme, the received signal is judged as 0 or 1 according to the decision threshold. After receiving the decision output, the decoder completes the decoding process according to the output bits between sequences. Generally speaking, the hard decision decoding principle is simple and easy to implement in hardware, but due to the loss of a large amount of useful information, the bit error rate performance is relatively poor. Compared with the hard decision method, the soft decision method can usually achieve better decision performance in iterative decoding. By calculating the log-likelihood ratio (LLR) of each bit, the soft decision method effectively utilizes the useful information of the decoder and greatly improves the decoding performance. However, this method also greatly increases the computational complexity. In particular, in high-order modulation, each modulation symbol corresponds to multiple bits, which further increases the demodulation complexity of the soft decision scheme and brings challenges to practical applications. At present, there are many methods to reduce the complexity of receivers based on soft decision methods. The optimized soft decision method simplifies complex exponential, logarithmic and division operations into simple maximum and minimum selection, addition and subtraction operations, effectively simplifying the calculation process. However, the degree of complexity reduction of this method is still limited. Therefore, how to achieve further reduction in complexity while ensuring the accuracy of LLR calculation is an important issue facing the technical development of soft decision decoding schemes, which is of great significance to the high speed, flexibility and reliability of optical fiber communications.

[0004] In recent years, the rapid progress of machine learning algorithms such as neural networks has provided a new direction for the development of optical communication systems. Neural networks have powerful data fitting capabilities and can achieve high-precision function approximation through training data. They are regarded as an alternative method for accurately calculating LLR in optical fiber communication systems. However, in function fitting tasks, most neural networks have global approximation characteristics. The outputs of these globally approximated neural networks are affected by any adjustable parameters, weights, and thresholds. Since each parameter must be considered during the learning process, the training process is complicated and cannot meet the growing application needs. In addition, most neural networks use backpropagation algorithms to improve training efficiency, which introduces more parameters to the neural network, often causing the loss function to fall into a local minimum, thereby reducing the approximation accuracy of the model. In the LLR-based soft decision demodulation scheme, the problems of traditional neural networks make it difficult to accurately predict LLR. Summary of the invention

[0005] The present invention provides an optical fiber communication method, system and storage medium based on low-complexity joint decision decoding. At the transmitting end, large and small power signals are superimposed by different mapping rules to obtain an optimized 16QAM superimposed signal. At the receiving end, the demodulation process of the high-power signal is optimized by dividing the decision area. In the LLR-based soft decision process, an RBF neural network is used to simplify the calculation process and improve the prediction accuracy of the LLR value, thereby greatly reducing the complexity of the signal demodulation process, improving the transmission rate of optical fiber communication and reducing the transmission error rate.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides an optical fiber communication method based on low-complexity joint decision decoding, comprising:

[0008] On the sending side:

[0009] Pre-modulating each transmission signal, and allocating different powers to the output signals of the pre-modulation step;

[0010] Perform QPSK modulation on the transmitted signal after power distribution to obtain a high-power signal and a low-power signal;

[0011] Mapping the high-power signal and the low-power signal to each quadrant of the superimposed signal constellation diagram according to different mapping rules to obtain a 16QAM superimposed signal;

[0012] On the receiving end:

[0013] Receive the 16QAM superimposed signal transmitted by the optical fiber channel and perform pre-demodulation on the received signal;

[0014] According to different mapping positions of the high-power signal and the low-power signal in the 16QAM superposition signal constellation diagram, the symbol distribution of the high-power signal and the low-power signal in the output signal of the pre-demodulation step is obtained, wherein the symbol distribution of the high-power signal includes an overlapping area and a non-overlapping area;

[0015] A soft decision strategy is used to make regional decisions on symbol points in overlapping areas and symbol points of low-power signals, and a hard decision strategy is used to make regional decisions on symbol points in non-overlapping areas, wherein the method for making soft decisions on symbol points in overlapping areas includes: predicting the log-likelihood ratio (LLR) value of each bit of the symbol point by a pre-trained RBF radial basis neural network;

[0016] The judged signal is sent to the post-demodulation step to obtain the signal sent by each user.

[0017] Optionally, in the process of allocating different powers to the signals, the allocation is performed according to a ratio of 4 between the power of the high-power signal and the power of the low-power signal.

[0018] Optionally, the different mapping rules include:

[0019] Make the first, second, third and fourth quadrants of the high-power signal symbol point correspond to symbols 10, 00, 01 and 11 respectively;

[0020] Each quadrant of the low-power signal symbol point is divided into four areas to obtain sixteen symbol point areas, and the symbol point area close to the origin in each quadrant corresponds to symbol 11, the symbol point area far from the origin corresponds to symbol 00, the symbol point area distributed along the vertical axis corresponds to symbol 01, and the symbol point area distributed along the vertical axis corresponds to symbol 10.

[0021] Optionally, the method for dividing the overlapping area and the non-overlapping area includes: obtaining four circular decision areas with a radius of r according to the distribution of high-power signal symbols in the 16QAM superposition signal constellation diagram, and taking the intersection area between two adjacent circular decision areas as the overlapping area, and the other areas as the non-overlapping areas.

[0022] Optionally, the method for performing soft decision on low-power signal symbol points includes:

[0023] Calculate the logarithm of the ratio of the probability that each bit of the symbol point is 0 to the probability that it is 1, and get the LLR value. The calculation formula is as follows:

[0024] ,

[0025] in, Indicates the LLR value of each bit of the symbol calculated directly, Indicates the symbol point of two bits in the constellation diagram of the 16QAM superimposed signal received by the receiving end. express The i-th bit of , Indicates the case where the i-th bit is 0, Indicates the case where the i-th bit is 1, Indicates the symbol point of two bits in the constellation diagram of the 16QAM superposition signal sent by the transmitter. express and The offset between represents the variance of the channel noise.

[0026] Optionally, the method for acquiring the pre-trained RBF radial basis neural network includes:

[0027] Obtaining the LLR value of each bit of the historical high-power signal symbol point, and normalizing the obtained LLR value to obtain a training sample set;

[0028] Input the training sample set into a pre-built radial basis neural network, and calculate the Euclidean distance between each input sample of the input layer and the hidden layer point;

[0029] According to the calculated Euclidean distance, the center point, variance and weights between the hidden layer and the output layer of the radial basis function corresponding to each input sample are solved to obtain the trained RBF radial basis neural network. The formula for its output LLR value is:

[0030] ,

[0031] ,

[0032] in, It represents the LLR value of each symbol predicted by the RBF neural network. is the input sample, is the center point of the i-th basis function, for and The Euclidean distance between , is the weight corresponding to the i-th basis function, b represents the bias, represents the basis function variance.

[0033] Optionally, predicting the log-likelihood ratio (LLR) of each bit of the symbol point by using a pre-trained RBF radial basis neural network includes:

[0034] The two-bit symbol points in the overlapping area Input the trained RBF radial basis neural network;

[0035] Determine the symbol point with The two sample points with the closest Euclidean distance;

[0036] The ordinates of the two sample points are obtained according to the basis functions of the two sample points and the weights corresponding to the basis functions, and the symbol point is obtained by adding the two ordinates. The LLR value of .

[0037] Optionally, the pre-modulation step includes: sequentially performing serial / parallel conversion, convolution coding, QPSK modulation, inverse fast Fourier transform, adding a guard gap, and parallel / serial conversion on the transmission signal;

[0038] The pre-demodulation step includes: performing serial / parallel conversion, guard interval removal, fast Fourier transformation and channel equalization on the 16QAM signal received by the receiving end in sequence;

[0039] The post-demodulation step includes: deconvolution and parallel / serial conversion of the decoded signal.

[0040] In a second aspect, the present invention provides an optical fiber communication system based on low-complexity joint decision decoding, comprising a transmitting end system and a receiving end system;

[0041] The sending end system comprises:

[0042] Power distribution module: used to pre-modulate each transmission signal and distribute different powers to the output signals of the pre-modulation step;

[0043] QPSK modulation module: used to perform QPSK modulation on the transmission signal after power distribution to obtain high-power signal and low-power signal;

[0044] Signal superposition module: used for mapping the high-power signal and the low-power signal to each quadrant of the superposition signal constellation diagram according to different mapping rules to obtain a 16QAM superposition signal;

[0045] The receiving end system comprises:

[0046] Pre-demodulation module: used to receive the 16QAM superimposed signal transmitted by the optical fiber channel and perform pre-demodulation on the received signal;

[0047] Symbol distribution extraction module: used to obtain the symbol distribution of the high-power signal and the low-power signal in the output signal of the pre-demodulation step according to the different mapping positions of the high-power signal and the low-power signal in the 16QAM superposition signal constellation diagram, wherein the symbol distribution of the high-power signal includes an overlapping area and a non-overlapping area;

[0048] The regional decision decoding module is used to make regional decisions on the overlapping region symbol points and the low-power signal symbol points by using a soft decision strategy, and to make regional decisions on the non-overlapping region symbol points by using a hard decision strategy, wherein the method for making soft decisions on the overlapping region symbol points includes: predicting the log-likelihood ratio (LLR) value of each bit of the symbol point by using a pre-trained RBF radial basis neural network;

[0049] Post-demodulation module: used to send the judged signal to the post-demodulation step to obtain the signal sent by each user.

[0050] In a third aspect, the present invention provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the optical fiber communication method based on low-complexity joint decision decoding as described in any step of the first aspect.

[0051] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: in order to explore a low-complexity demodulation scheme that is more suitable for the NOMA system, the present invention uses different mapping rules to superimpose high-power signals in the constellation diagram at the transmitting end to obtain a more optimized 16QAM superimposed signal, and formulates corresponding decision schemes at the receiving end according to the signal characteristics of different powers and the mapping characteristics of the constellation diagram, so that the optical fiber communication has higher transmission speed, flexibility and reliability, and then by adopting a joint decision scheme based on the circular area to flexibly divide the high-power signal decision area, a higher performance gain can be obtained compared with the hard decision scheme, and the system demodulation performance can be effectively improved. In order to cope with the large amount of computational complexity introduced by calculating LLR in the soft decision process, an RBF neural network is used to predict the LLR value. By using the excellent approximation characteristics of the radial basis function, not only the computational complexity of the soft decision is significantly reduced, but also the loss function of the traditional neural network can be avoided from falling into the local optimal solution, thereby reducing the approximation accuracy of the model, further ensuring the high accuracy of the soft decision based on the LLR value, while maintaining high decoding accuracy. Reduce the complexity of the receiving end. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The figure is a flow chart of an optical fiber communication method based on low-complexity joint decision decoding in one embodiment of the present invention;

[0053] Figure 2 The figure is a signal modulation flow chart based on superposition coding in one embodiment of the present invention;

[0054] Figure 3 The figure shows a schematic diagram of the superposition of the constellation diagrams of a high-power signal and a low-power signal in an embodiment of the present invention;

[0055] Figure 4 Shown is a signal demodulation flow chart based on a joint decision strategy in an embodiment of the present invention;

[0056] Figure 5 The figure shows a schematic diagram of the symbol distribution of a high-power QPSK signal constellation diagram in an embodiment of the present invention;

[0057] Figure 6 The figure shows a schematic diagram of symbol distribution of a low-power QPSK signal constellation diagram in an embodiment of the present invention;

[0058] Figure 7 The figure shows a schematic diagram of high-power signal decision area division based on circular area division in an embodiment of the present invention;

[0059] Figure 8 Shown is a schematic diagram of the distribution of received symbol points in an embodiment of the present invention;

[0060] Fig. 9 The figure shows a schematic diagram of the distribution of symbol points in the overlapping area after the circular area is divided in one embodiment of the present invention;

[0061] Fig.10 The figure shows a schematic diagram of the distribution of symbol points in non-overlapping areas after circular area division in one embodiment of the present invention;

[0062] Fig.11 The figure shows a schematic diagram of function approximation using a RBF neural network in one embodiment of the present invention;

[0063] Fig.12 Shown is a flow chart of using RBF neural network to predict two-bit LLR values ​​in one embodiment of the present invention;

[0064] Fig.13 The figure shows a structure diagram of an experimental system for conducting experiments using the technical solution of the present invention in one embodiment of the present invention;

[0065] Fig.14 The figure is a schematic diagram of the comparison of mean square error of BP and RBF neural network prediction results in one embodiment of the present invention;

[0066] Fig.15 The figure is a schematic diagram showing a comparison of the curves of the bit error rate of a high-power signal with the received optical power under the joint decision-making scheme and the hard decision-making scheme in an embodiment of the present invention;

[0067] Fig.16 The figure is a schematic diagram showing the reduction of the amount of calculation using the RBF neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0069] Example 1

[0070] like Figure 1 As shown, this embodiment provides an optical fiber communication method based on low-complexity joint decision decoding, including:

[0071] On the sending side:

[0072] Pre-modulating each transmission signal, and allocating different powers to the output signals of the pre-modulation step;

[0073] Perform QPSK modulation on the transmitted signal after power distribution to obtain a high-power signal and a low-power signal;

[0074] Mapping the high-power signal and the low-power signal to each quadrant of the superimposed signal constellation diagram according to different mapping rules to obtain a 16QAM superimposed signal;

[0075] On the receiving end:

[0076] Receive the 16QAM superimposed signal transmitted by the optical fiber channel and perform pre-demodulation on the received signal;

[0077] According to different mapping positions of the high-power signal and the low-power signal in the 16QAM superposition signal constellation diagram, the symbol distribution of the high-power signal and the low-power signal in the output signal of the pre-demodulation step is obtained, wherein the symbol distribution of the high-power signal includes an overlapping area and a non-overlapping area;

[0078] A soft decision strategy is used to make regional decisions on symbol points in overlapping areas and symbol points of low-power signals, and a hard decision strategy is used to make regional decisions on symbol points in non-overlapping areas, wherein the method for making soft decisions on symbol points in overlapping areas includes: predicting the log-likelihood ratio (LLR) value of each bit of the symbol point by a pre-trained RBF radial basis neural network;

[0079] The judged signal is sent to the post-demodulation step to obtain the signal sent by each user.

[0080] At the transmitting end, different mapping rules are used to superimpose and encode high and low power signals, and corresponding decoding strategies are developed for signals of different powers at the receiving end, thereby improving the transmission rate of optical fiber communication and reducing the bit error rate; at the receiving end, the computational complexity of the demodulation process of high-power signals is reduced by flexibly dividing the circular decision area, and at the same time, a higher performance gain can be obtained compared with the hard decision scheme; by introducing a demodulator architecture based on a low-complexity trainable RBF neural network to optimize the soft decision strategy of high-power signals, it is further guaranteed that the low-complexity soft decision scheme can obtain excellent decision accuracy.

[0081] Example 2

[0082] Based on Example 1, this example also makes the following design.

[0083] This embodiment is divided into the following modules: the first part is carried out by the transmitting system to perform superposition coding on the user signal; the second part demodulates the received signal according to the joint decision-making strategy formulated by the transmitting system; the third part uses the RBF neural network to reduce the computational complexity of the joint decision; and the fourth part verifies the feasibility of the scheme.

[0084] 1. Power Domain Non-Orthogonal Multiple Access PD-NOMA System Based on Superposition Coding

[0085] The schematic diagram of the non-orthogonal multiple access-orthogonal frequency division multiplexing NOMA-OFDM scheme based on superposition coding is as follows Figure 2 As shown. At the transmitting end, the different power levels of the OFDM signal are superposition coded and then linearly combined through power distribution. Superposition coding technology, as a physical layer technology, enables a transmitter to send signals from multiple independent users at the same time. The signal after power division multiplexing can be mathematically expressed as:

[0086] ,

[0087] In the formula, ,and and Respectively represent The present invention uses power division multiplexing technology to multiplex the signals and signal power of each user in the digital domain. and Overlay. Definition For high power signals, is a low-power signal. Since the transmitting end of the NOMA system superimposes the signals of different users, the receiving end distinguishes the signals of different users by the signal power. Therefore, the power allocation ratio PDR between different users at the transmitting end directly affects the demodulation performance of the receiver and also determines the transmission rate of the NOMA system. Definition , in order to achieve the best demodulation result, set PDR=4.

[0088] At the transmitting end, a convolutional code with a constraint length of 4 and a generating polynomial of [13, 17] is used, the code rate is set to 1 / 2, the number of states is 8, and the traceback depth is set to 10 clock cycles. For two signals with different powers, the transmitted data is first converted from serial to parallel. Subsequently, after convolution coding, they are mapped to the QPSK modulation format to obtain a frequency domain OFDM signal. Next, an inverse fast Fourier transform IFFT operation is performed on the two OFDM signals to convert them from the frequency domain to the time domain. To avoid inter-code interference, a cyclic prefix CP needs to be added to each OFDM signal in the time domain. After parallel-to-serial conversion and power allocation, the two QPSK signals are converted into NOMA-OFDM signals using constellation mapping. Finally, the superimposed signal is in the form of a 16QAM signal. In addition, before performing the IFFT operation, a training sequence TS is inserted into the high-power signal for channel estimation and equalization at the receiving end. The constellation diagrams of high-power, low-power and superimposed signals are shown in Figure 2. Figure 3 shown.

[0089] 2. Joint Judgment Strategy

[0090] At the receiving end, the present invention adopts a joint decision strategy to demodulate high-power QPSK signals. First, the received data is converted from serial to parallel, and then the fast Fourier transform FFT is performed after removing the guard interval. After channel equalization, different decision strategies are adopted according to the different characteristics of high-power and low-power signals. The joint decision strategy proposed by the present invention combines a low-complexity hard decision strategy with a high-precision LLR-based soft decision method. The specific flow chart is as follows: Figure 4 shown.

[0091] After passing through the optical fiber channel, the received signal can be expressed as:

[0092] ,

[0093] In the formula, is the signal received in the frequency domain, is the signal received in the time domain, is the channel response of the ith signal in the time domain, is the channel response of the ith signal in the frequency domain, For high power signals, For low power signals, is the high power signal power, is the power of low-power signal.

[0094] After receiving the signal, channel estimation and equalization are performed by comparing the sent and received training sequences. The specific equalization process can be expressed as:

[0095] ,

[0096] Where H represents the channel response, and Represent the sent training sequence and the received training sequence respectively, and Represent the signals before and after equalization, is the equilibrium function.

[0097] After channel equalization, the high-power signal is first divided into different regions, and a decision-making scheme based on coordinate division is adopted in the hard decision region. This method focuses on selecting symbol points with lower decoding difficulty and avoids introducing too much complexity as much as possible by adopting a relatively simple demodulation method. Figure 5 As shown, in the process of making hard decisions on the high-power QPSK signal, it is determined that the superimposed 16QAM signal corresponds to symbols "10", "00", "01" and "11" in the four quadrants (1, 2, 3, 4) of the constellation points, respectively.

[0098] The present invention adopts a soft decision method based on LLR calculation for low-power signals. The present invention first optimizes the symbol distribution using a new mapping rule to reduce the complexity of decision making. In the decision making process of low-power QPSK signals, Figure 6 As shown in the corresponding coordinate area, the outermost constellation point is determined as "00", the innermost constellation point is determined as "11", the points distributed along the vertical axis are determined as "01", and the points distributed along the horizontal axis are determined as "10", and the log-likelihood ratio LLR of the corresponding bit is calculated according to the mapping rule. In the decision-making process, the demodulation of high-power and low-power signals does not affect each other, thereby obtaining more accurate demodulation data.

[0099] In the joint decision strategy process for high-power signals, circular decision regions are used for all symbols. Figure 7 As shown in the figure, the decoding of high-power signals is divided into four circular decision areas with a radius of r. In the entire decision process, for those intersections of decision areas where symbol decision errors are prone to occur, a soft decision method based on LLR calculation is used. In addition, in the non-intersection part of the decision area, that is, the area with the lowest decision difficulty, the hard decision method introduced above is used to analyze and determine the coordinates of the constellation points. By adopting this joint decision strategy, it is necessary to perform soft decisions in four areas for the decoding of high-power signals.

[0100] Figure 8 Shown is the data point distribution of 25600 received symbol points when the SNR is set to 6dB, PDR = 4, r = 0.95. Fig. 9 The figure shows the distribution of signal data points in the overlapped area after the constellation diagram is divided into circular areas. Fig.10 The distribution of signal data points in the non-overlapping area is shown. It can be seen that most of the received symbols are concentrated around the origin of the constellation diagram, and a small number of symbols are distributed along the edge of the constellation diagram. The area division strategy proposed in the present invention effectively divides the received symbol points according to the decision complexity, and optimizes the decision strategy of high-power signals.

[0101] 3. Soft decision method based on RBF neural network

[0102] The soft decision demodulation method usually uses the LLR value of each bit to demodulate the signal. LLR is defined as the logarithm of the ratio of the probability that a bit is 0 to the probability that a bit is 1. Its mathematical expression is:

[0103] ,

[0104] in, Indicates the symbol point of two bits in the constellation diagram of the 16QAM superimposed signal received by the receiving end. express The i-th bit of . Assume that the variance of the channel noise is , then the exact calculation of the LLR expression is:

[0105] ,

[0106] in, Indicates the symbol point of two bits in the constellation diagram of the 16QAM superposition signal sent by the transmitter. Indicates the case where the i-th bit is 0, Indicates the case where the i-th bit is 1, express and The offset between .

[0107] Since the distribution of low-power signals in 16QAM is relatively complex, the decision of low-power signals becomes more difficult after superposition. In order to better ensure the accuracy of low-power decisions, the LLR value of each bit of the low-power signal symbol point is calculated through the above calculation method to make soft decisions on low-power signals.

[0108] It can be observed from the equation that although the joint decision reduces the number of symbols that need to make soft decisions to some extent, the computational complexity of each symbol is still very high. In order to avoid complex calculations, this embodiment uses an RBF neural network to sequentially predict the LLR values ​​of two-bit symbols in the overlapping area of ​​the high-power QPSK signal. By using a pre-trained RBF neural network model to predict the LLR of each bit of the received symbol, the substantial increase in complexity caused by the increase in the number of received symbols can be avoided. This method alleviates the need for complex calculations in the soft decision area of ​​the high-power QPSK signal, thereby further reducing the complexity of the entire demodulation process. RBF, as an activation function of a neural network, is usually defined as a monotonic function of the Euclidean distance between a point in space and a specified center. The Gaussian function of the present invention is selected as a specific form of the radial basis function, and its formula is as follows:

[0109] ,

[0110] In this embodiment, n represents an independent variable in the input RBF function, which refers to the LLR value of the symbol in this embodiment, and φ represents the width of the function, and its value determines the selectivity of the function. Fig.11 The specific process of using RBF to approximate functions is explained. First, each coordinate data of the training sample corresponds to a basis function , the center of the function is the sample point, and the value of this function represents the LLR. , The value of can be regarded as the corresponding ordinate, which is numerically equal to and The sum of the ordinates of the points. The vertical coordinate of the point is obtained by multiplying the basis function value of the first sample symbol by a small weight, The ordinate of the point is obtained by multiplying the basis function value of the second sample symbol by a larger weight, and the corresponding weights of other symbols are all set to 0. When calculating the LLR, only the first and second sample points are considered, and the distant sample points with small contributions are ignored. According to this principle, when a new symbol is received, the basis function is multiplied by the corresponding weight and added to perfectly fit the curve of the real LLR. Therefore, the RBF neural network only needs a few connection weights to have a significant impact in the local area of ​​the input space, thus avoiding the complex weight multiplication operations in the global approximation network. As a local approximation network, its unique optimal approximation characteristics also prevent training from falling into local minima.

[0111] In summary, the activation function of the RBF neural network can be expressed as:

[0112] ,

[0113] The activation function is defined as the input vector and the center point of the i-th basis function The Euclidean distance between , b represents the bias. Assume is the weight corresponding to the i-th basis function, and the output of the network for:

[0114] .

[0115] Fig.12 The specific process of the first and second bit log-likelihood ratios (LLRs) of the superimposed 16QAM-OFDM signal and the high-power signal predicted by the RBF neural network is demonstrated. According to the specific constellation mapping rules, the LLR values ​​of the first and second bits are predicted in turn. First, the historical LLR true values ​​of the two bits of the symbol point are obtained to construct the training data. After normalization, the Euclidean distance between each input vector and the center of the basis function is calculated. According to the calculated Euclidean distance, the center point, variance and weights between the hidden layer and the output layer of the radial basis function corresponding to each input sample are solved to obtain the RBF neural network with the best prediction performance. Subsequently, the newly received two-bit symbol is input into the RBF neural network for LLR value prediction.

[0116] 4. Experimental system and results analysis

[0117] The experimental setup of this scheme is as follows Fig.13As shown. Based on the principle of intensity modulation / direct detection IM / DD, a seven-core optical fiber is used in the experimental verification, and the crosstalk can reach -50dB. At the transmitting end, the NOMA-OFDM signal is first generated in the offline digital signal processing DSP, and the number of effective subcarriers is 256. The number of IFFT points of the system is 1024, and the length of the training sequence is 1 / 51 of the OFDM symbol. Then, a guard interval CP is added to each symbol, and the length is 1 / 4 of the OFDM symbol. After modulation by the arbitrary waveform generator, the modulated signal is converted into a radio frequency RF signal. Then, the RF signal is injected into the Mach-Zehnder modulator MZM for intensity modulation after passing through the electric power amplifier EA. The optical signal carrying information is divided into seven beams by the optical coupler OC and injected into the 2km long seven-core optical fiber through the focusing device. At the receiving end, the optical signal is adjusted for optical power by the variable optical attenuator VOA, and the photoelectric conversion is realized by the photodiode. Subsequently, a mixed signal oscilloscope is used for analog-to-digital conversion, and finally the signal demodulation and recovery are realized by the offline DSP.

[0118] In order to verify the effectiveness of the RBF neural network, the experiment used both the back propagation neural network BPnn and the RBF neural network RBFnn to predict the LLR of two bits of 200 received symbols, and calculated the mean square error MSE of each bit prediction. During the training process, MSE is defined as the error function of network training, and the minimum error function value is set to 1×10 -6 , the maximum iteration period is set to 100. MSE can be defined as:

[0119] ,

[0120] in, represents the LLR prediction value output by the neural network, Indicates the actual value of the calculated LLR.

[0121] The MSE comparison results are as follows: Fig.14 As shown in Figure 2, by observing the prediction error, it is concluded that compared with the BP neural network, the RBF neural network solves the limitations of the traditional global approximation neural network and achieves a more accurate prediction output. Therefore, by using the RBF neural network for LLR prediction, the challenge of high computational complexity can be effectively solved without sacrificing decision accuracy.

[0122] Furthermore, the effectiveness of the solution proposed in the present invention is demonstrated by analyzing the bit error rate performance of the system. Fig.15The relationship between the bit error rate BER curve of high-power signals and the received optical power ROP under different schemes is shown. In the experiment, PDR = 4 and r = 0.95 are set. The experimental results show that the joint decision scheme proposed in the present invention is superior to the hard decision method in terms of receiving sensitivity. Specifically, when the bit error rate BER is 1×10 -3 A gain of 0.45 dB can be obtained under this condition. In addition, compared with the joint decision method without RBF neural network optimization, the proposed scheme achieves low-complexity demodulation of high-power signals with almost no sacrifice in BER performance.

[0123] In order to further verify that the use of RBF neural network in the joint decision-making process of high-power signals can reduce the complexity, the reduced computational complexity was calculated when different area division radii were selected. During the test, the signal-to-noise ratio was set to 6dB, and 25,600 received symbols were tested. As r increases, the number of data points using the soft decision scheme gradually increases, and the system receiving sensitivity becomes better and better. At the same time, the calculation amount of the soft decision method based on LLR calculation also gradually increases. Fig.16 The reduction in computational complexity using the RBF neural network is shown. In the specific calculation, the degree of complexity optimization is measured based on the number of saved exponential and logarithmic, multiplication and division, and addition and subtraction operations. The results show that the scheme can effectively avoid the increase in complexity caused by the increase in soft decision data and reduce the computational burden of the system. The complexity optimization demonstrated by the test results is attributed to the use of RBF neural network prediction as a substitute for LLR calculation, which is the result of further reducing the complexity on the low-complexity design of the joint decision strategy itself.

[0124] Example 3

[0125] This embodiment provides an optical fiber communication system based on low-complexity joint decision decoding, including a transmitting end system and a receiving end system;

[0126] The sending end system comprises:

[0127] Power distribution module: used to pre-modulate each transmission signal and distribute different powers to the output signals of the pre-modulation step;

[0128] QPSK modulation module: used to perform QPSK modulation on the transmission signal after power distribution to obtain high-power signal and low-power signal;

[0129] Signal superposition module: used for mapping the high-power signal and the low-power signal to each quadrant of the superposition signal constellation diagram according to different mapping rules to obtain a 16QAM superposition signal;

[0130] The receiving end system comprises:

[0131] Pre-demodulation module: used to receive the 16QAM superimposed signal transmitted by the optical fiber channel and perform pre-demodulation on the received signal;

[0132] Symbol distribution extraction module: used to obtain the symbol distribution of the high-power signal and the low-power signal in the output signal of the pre-demodulation step according to the different mapping positions of the high-power signal and the low-power signal in the 16QAM superposition signal constellation diagram, wherein the symbol distribution of the high-power signal includes an overlapping area and a non-overlapping area;

[0133] The regional decision decoding module is used to make regional decisions on the overlapping region symbol points and the low-power signal symbol points by using a soft decision strategy, and to make regional decisions on the non-overlapping region symbol points by using a hard decision strategy, wherein the method for making soft decisions on the overlapping region symbol points includes: predicting the log-likelihood ratio (LLR) value of each bit of the symbol point by using a pre-trained RBF radial basis neural network;

[0134] Post-demodulation module: used to send the judged signal to the post-demodulation step to obtain the signal sent by each user.

[0135] Example 4

[0136] This embodiment provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the optical fiber communication method based on low-complexity joint decision decoding as described in any step of Embodiment 2 is implemented.

[0137] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0139] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0141] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A fiber optic communication method based on low-complexity joint decision decoding, characterized in that: include: On the sending side: Pre-modulating each transmission signal, and allocating different powers to the output signals of the pre-modulation step; Perform QPSK modulation on the transmitted signal after power distribution to obtain a high-power signal and a low-power signal; Mapping the high-power signal and the low-power signal to each quadrant of the superimposed signal constellation diagram according to different mapping rules to obtain a 16QAM superimposed signal; On the receiving end: Receive the 16QAM superimposed signal transmitted by the optical fiber channel and perform pre-demodulation on the received signal; According to different mapping positions of the high-power signal and the low-power signal in the 16QAM superposition signal constellation diagram, the symbol distribution of the high-power signal and the low-power signal in the output signal of the pre-demodulation step is obtained, wherein the symbol distribution of the high-power signal includes an overlapping area and a non-overlapping area; A soft decision strategy is used to make regional decisions on symbol points in overlapping areas and symbol points of low-power signals, and a hard decision strategy is used to make regional decisions on symbol points in non-overlapping areas, wherein the method for making soft decisions on symbol points in overlapping areas includes: predicting the log-likelihood ratio (LLR) value of each bit of the symbol point by a pre-trained RBF radial basis neural network; The judged signal is sent to the post-demodulation step to obtain the signal sent by each user.

2. The optical fiber communication method based on low-complexity joint decision decoding according to claim 1, characterized in that: In the process of allocating different powers of the signal, the allocation is performed according to the ratio of the power of the high-power signal to the power of the low-power signal being 4.

3. The optical fiber communication method based on low-complexity joint decision decoding according to claim 2, characterized in that: The different mapping rules include: Make the first, second, third and fourth quadrants of the high-power signal symbol point correspond to symbols 10, 00, 01 and 11 respectively; Each quadrant of the low-power signal symbol point is divided into four areas to obtain sixteen symbol point areas, and the symbol point area close to the origin in each quadrant corresponds to symbol 11, the symbol point area far from the origin corresponds to symbol 00, the symbol point area distributed along the vertical axis corresponds to symbol 01, and the symbol point area distributed along the vertical axis corresponds to symbol 10.

4. The optical fiber communication method based on low-complexity joint decision decoding according to claim 3, characterized in that: The method for dividing the overlapping area and the non-overlapping area includes: obtaining four circular decision areas with a radius of r according to the distribution of high-power signal symbols in the 16QAM superposition signal constellation diagram, taking the intersection area between two adjacent circular decision areas as the overlapping area, and taking the other areas as the non-overlapping area.

5. The optical fiber communication method based on low-complexity joint decision decoding according to claim 3, characterized in that: The method for making soft decisions on low-power signal symbol points includes: Calculate the logarithm of the ratio of the probability that each bit of the symbol point is 0 to the probability that it is 1, and get the LLR value. The calculation formula is as follows: ， in, Indicates the LLR value of each bit of the symbol calculated directly, Indicates the symbol point of two bits in the constellation diagram of the 16QAM superposition signal received by the receiving end. express The i-th bit of , Indicates the case where the i-th bit is 0, Indicates the case where the i-th bit is 1, Indicates the symbol point of two bits in the constellation diagram of the 16QAM superposition signal sent by the transmitter. express and The offset between represents the variance of the channel noise.

6. The optical fiber communication method based on low-complexity joint decision decoding according to claim 3, characterized in that: The method for obtaining the pre-trained RBF radial basis neural network includes: Obtaining the LLR value of each bit of the historical high-power signal symbol point, and normalizing the obtained LLR value to obtain a training sample set; Input the training sample set into a pre-built radial basis neural network, and calculate the Euclidean distance between each input sample of the input layer and the hidden layer point; According to the calculated Euclidean distance, the center point, variance and weights between the hidden layer and the output layer of the radial basis function corresponding to each input sample are solved to obtain the trained RBF radial basis neural network. The formula for its output LLR value is: ， ， in, It represents the LLR value of each symbol predicted by the RBF neural network. is the input sample, is the center point of the i-th basis function, for and The Euclidean distance between , is the weight corresponding to the i-th basis function, b represents the bias, represents the basis function variance.

7. The optical fiber communication method based on low-complexity joint decision decoding according to claim 6, characterized in that: The method of predicting the log-likelihood ratio (LLR) of each bit of the symbol point by using a pre-trained RBF radial basis neural network includes: The two-bit symbol points in the overlapping area Input the trained RBF radial basis neural network; Determine the symbol point with The two sample points with the closest Euclidean distance; The ordinates of the two sample points are obtained according to the basis functions of the two sample points and the weights corresponding to the basis functions, and the symbol point is obtained by adding the two ordinates. The LLR value of .

8. The optical fiber communication method based on low-complexity joint decision decoding according to claim 1, characterized in that: The pre-modulation step includes: sequentially performing serial / parallel conversion, convolution coding, QPSK modulation, inverse fast Fourier transform, adding a guard interval, and parallel / serial conversion on the transmission signal; The pre-demodulation step includes: performing serial / parallel conversion, guard interval removal, fast Fourier transformation and channel equalization on the 16QAM signal received by the receiving end in sequence; The post-demodulation step includes: deconvolution and parallel / serial conversion of the decoded signal.

9. An optical fiber communication system based on low-complexity joint decision decoding, characterized in that: Includes a sending end system and a receiving end system; The sending end system comprises: Power distribution module: used to pre-modulate each transmission signal and distribute different powers to the output signals of the pre-modulation step; QPSK modulation module: used to perform QPSK modulation on the transmission signal after power distribution to obtain high-power signal and low-power signal; Signal superposition module: used for mapping the high-power signal and the low-power signal to each quadrant of the superposition signal constellation diagram according to different mapping rules to obtain a 16QAM superposition signal; The receiving end system comprises: Pre-demodulation module: used to receive the 16QAM superimposed signal transmitted by the optical fiber channel and perform pre-demodulation on the received signal; Symbol distribution extraction module: used to obtain the symbol distribution of the high-power signal and the low-power signal in the output signal of the pre-demodulation step according to the different mapping positions of the high-power signal and the low-power signal in the 16QAM superposition signal constellation diagram, wherein the symbol distribution of the high-power signal includes an overlapping area and a non-overlapping area; The regional decision decoding module is used to make regional decisions on the overlapping region symbol points and the low-power signal symbol points by using a soft decision strategy, and to make regional decisions on the non-overlapping region symbol points by using a hard decision strategy, wherein the method for making soft decisions on the overlapping region symbol points includes: predicting the log-likelihood ratio (LLR) value of each bit of the symbol point by using a pre-trained RBF radial basis neural network; Post-demodulation module: used to send the judged signal to the post-demodulation step to obtain the signal sent by each user.

10. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the optical fiber communication method based on low-complexity joint decision decoding as claimed in any one of claims 1 to 8 is implemented.

Images