End-to-end channel coding optimization method

Abstract

The invention discloses an end-to-end channel coding optimization method, which is applied to the technical field of high-speed optical communication. Comprising the following steps: establishing a direct modulation / direct detection (IM / DD) optical fiber channel model so as to simulate the transmission characteristics of a polarization code (Polar) coding signal in a single-mode optical fiber; establishing an end-to-end polarization code coding optimization system; and integrating the direct modulation and direct detection optical fiber channel model into an end-to-end polarization code coding optimization system to form a polarization code auto-encoder direct modulation and direct detection optical interconnection end-to-end system based on channel assistance, and performing parameter tuning by the system to determine an optimized polarization code coding polarization weight channel index. According to the method, the optical fiber channel is modeled through the RNN neural network, the sending end, the channel and the receiving end of the optical fiber system are combined and integrally optimized by using the coding and decoding structure of the auto-encoder, and the reliable channel suitable for the optical fiber system is found through training, so that the system performance of optimal coding and decoding is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of high-speed optical communication, and more particularly to an end-to-end channel coding optimization method. Background Art

[0002] Short-distance fiber-optic communication systems play an indispensable role in various practical applications such as local area networks, data centers, and in-building communication networks. They are characterized by limited transmission distances and simple modulation methods. Among the modulation schemes, the most commonly used for short-distance fiber-optic communication is intensity modulation direct detection (IM / DD), which has a simple structure and high cost performance. However, the inherent dispersion and nonlinear effects of optical fibers pose challenges to achieving high reliability and high performance transmission. At present, deep learning technology has been applied to the transmitter, receiver, and channel of optical fiber systems to improve system performance by optimizing specific modules. However, the results of this approach may be limited by the complexity of the system or device effects, making it difficult to achieve optimality. Therefore, it is necessary to integrate and optimize the entire system to achieve optimal end-to-end (E2E) performance.

[0003] Polar codes have attracted extensive attention in the field of coding theory due to their excellent performance close to the Shannon limit. It has become a promising error correction technology, but its effectiveness depends on the optimization of the code construction method. By irregularly pruning Polar codes, the decoding complexity and delay are reduced, while structural nesting or cascading improves the decoding performance. In optical fiber communication systems, the research on Polar codes is mainly focused on coherent optical communications. This research mainly combines Polar codes with polarization multiplexing or probability shaping to improve the signal-to-noise ratio, or uses chaotic encryption in optical fiber communication systems to improve the security of communication. However, in direct modulation and direct detection systems, the improvement of error correction performance is usually achieved by combining selective mapping with flexible Polar codes, and existing research has not included code construction optimization. In the encoding process of Polar codes, the input bits of the encoder are divided into two parts, namely, non-frozen bits carrying information bits and frozen bits carrying predefined values ​​known to the decoder. Since the input bits are divided into two groups, the problem of how to select the channel index of the non-frozen bits to provide the best error correction performance is called the construction problem of Polar codes. Traditional Polar code construction methods include polarization weight (PW), Bhattacharyya parameter construction, Gaussian approximation (GA) construction, etc. Polarization weight is calculated by calculating the β expansion of each polarization channel index, and then selecting K channels with the highest β expansion value for transmitting information bits. These technologies usually rank the reliability of fixed channels, and may not have the best error correction performance in the channels of direct modulation and direct detection optical fiber communication systems. Therefore, how to provide an end-to-end channel coding optimization method is a problem that technicians in this field urgently need to solve. Summary of the invention

[0004] In view of this, the present invention provides an end-to-end channel coding optimization method, which models the optical fiber channel through an RNN neural network, and then uses the encoding and decoding structure of the autoencoder to jointly optimize the transmitter, channel and receiver of the optical fiber system. Through training, a reliable channel index suitable for the optical fiber system is found, thereby achieving better performance.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] An end-to-end channel coding optimization method comprises the following steps:

[0007] S1. Establish a direct modulation and direct detection optical fiber channel model to simulate the transmission characteristics of polarization coded signals in single-mode optical fiber;

[0008] S2. Establish an end-to-end polar code optimization system;

[0009] S3, integrating the direct modulation and direct detection optical fiber channel model into the end-to-end polarization code encoding optimization system to form a channel-assisted polarization code autoencoder direct modulation and direct detection optical interconnection end-to-end system;

[0010] S4. Parameter tuning is performed on the channel-assisted polarization code autoencoder direct adjustment and direct detection optical interconnection system to determine the optimized polarization code encoding polarization weight channel index.

[0011] Optionally, S1 is:

[0012] S11. Build an optical fiber system experimental platform, including arbitrary waveform generator AWG, electric amplifier EA, laser Laser, modulator MZM, single-mode fiber SMF, variable optical attenuator VOA, photodetector PD and oscilloscope OSC;

[0013] S12, repeatedly transmit the polar code encoded by PAM4 modulation, and record the data sent by the arbitrary waveform generator AWG and the data collected by the oscilloscope OSC;

[0014] S13, using a sliding window to process the received data, and dividing the recorded data into 80% training set and 20% test set in proportion;

[0015] S14, establish a direct modulation and direct inspection optical fiber channel model based on RNN, use the data received by the oscilloscope OSC in the training set as the RNN model input, and the data sent by the arbitrary waveform generator AWG as the label, debug the parameters and perform training, and finally obtain the direct modulation and direct inspection optical fiber channel model;

[0016] S15. Use the test set to verify the direct adjustment and direct inspection fiber channel model, and compare the output data of the direct adjustment and direct inspection fiber channel model with the labels in the test set to evaluate the accuracy of the direct adjustment and direct inspection fiber channel model.

[0017] Optionally, S2 is specifically:

[0018] S21. A three-layer neural network is introduced in polar code encoding, with a constant t determined by experimental experience as input and a channel index in polar code encoding as output;

[0019] S22. Perform polarization code encoding on the information sequence u according to the channel index output by the neural network to obtain an encoded sequence x;

[0020] S23, the encoder and decoder of the autoencoder correspond to the transmitting end and the receiving end in the optical fiber system respectively, the autoencoder is composed of two neural networks with the same structure, wherein the encoder comprises an input layer, two hidden layers and an output layer, the input layer comprises M neurons, the two hidden layers each comprise 2M neurons, the output layer comprises n neurons, and the decoder has the same structure as the encoder;

[0021] S24. Polar decoding is performed through multiple iterations of the belief propagation algorithm to calculate the bit error rate of the system.

[0022] Optionally, the three-layer neural network in S21 is specifically:

[0023] channel index=Softmax(W2′((W1′×t+b1′))+b2′)

[0024] Where W1′ and W2′ are the weights of the fully connected layer of the three-layer neural network, b1′ and b2′ are the biases of the fully connected layer of the three-layer neural network, and Softmax represents the activation function.

[0025] Optionally, the encoder neural network is specifically:

[0026] z=(3″((2″(L(1″×y+1″))+2″))+3″)

[0027] Where W3″, W2″, W1″ are the weights of the fully connected layer of the encoder neural network, b1″, b2″, b3″ are the biases of the fully connected layer of the encoder neural network, Sigmoid represents the activation function, y is the input of the encoder neural network, and z is the output of the encoder neural network.

[0028] Optionally, the decoder neural network is specifically:

[0029] b=(3″′((2″′(U(1″′×a+1″′))+2″′))+3″′)

[0030] Where W3″′, W2″′, and W1″′ are the weights of the fully connected layer of the decoder neural network, b1″′, b2″′, and b3″′ are the biases of the fully connected layer of the decoder neural network, Softmax represents the activation function, a is the input of the decoder neural network, and b is the output of the decoder neural network.

[0031] Optionally, S4 and later also include:

[0032] S5. Experimentally test the optimized polarization weight channel index, compare its performance with that of the traditional polarization code construction method, and calculate the improvement of the optimization method in receiving sensitivity under the same bit error rate conditions.

[0033] It can be seen from the above technical solution that, compared with the prior art, the present invention provides an end-to-end channel coding optimization method, which has the following beneficial effects:

[0034] 1. The present invention discloses an overall optimization learning strategy for a direct adjustment and direct inspection optical fiber system, which achieves optimal system performance by uniformly modeling the transmitter, channel, and receiver as an end-to-end system and performing global training in combination with an autoencoder algorithm;

[0035] 2. The present invention takes into account the influence of dispersion and nonlinear effects in the modeling of the direct modulation and direct detection optical fiber system, optimizes the polarization weight channel index in the polarization code encoding, achieves performance improvement for the direct modulation and direct detection optical fiber system, and significantly improves the receiving sensitivity;

[0036] 3. The present invention can obtain polarization weight channel indexes for different channel conditions by adjusting different optical fiber lengths, has higher adaptability and versatility, and effectively reduces costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0038] Figure 1 It is a flow chart of the end-to-end channel coding optimization method of the present invention;

[0039] Figure 2 It is a schematic diagram of the end-to-end channel coding optimization method of the present invention;

[0040] Figure 3 Schematic diagram of the measured code distance of the experimental results in the embodiment of the present invention;

[0041] Figure 4 Schematic diagram of bit error rate performance of experimental results in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] The embodiment of the present invention discloses an end-to-end channel coding optimization method, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0044] S1. Establish a direct modulation and direct detection optical fiber channel model to simulate the transmission characteristics of polarization coded signals in single-mode optical fiber;

[0045] S2. Establish an end-to-end polar code optimization system;

[0046] S3, integrating the direct modulation and direct detection optical fiber channel model into the end-to-end polarization code encoding optimization system to form a channel-assisted polarization code autoencoder direct modulation and direct detection optical interconnection end-to-end system;

[0047] S4. Parameter tuning is performed on the channel-assisted polarization code autoencoder direct adjustment and direct detection optical interconnection system to determine the optimized polarization code encoding polarization weight channel index.

[0048] Furthermore, S1 is specifically:

[0049] S11. Build an optical fiber system experimental platform, including arbitrary waveform generator AWG, electric amplifier EA, laser Laser, modulator MZM, single-mode fiber SMF, variable optical attenuator VOA, photodetector PD and oscilloscope OSC;

[0050] S12, repeatedly transmit the polar code encoded by PAM4 modulation, and record the data sent by the arbitrary waveform generator AWG and the data collected by the oscilloscope OSC;

[0051] S13, using a sliding window to process the received data, and dividing the recorded data into 80% training set and 20% test set in proportion;

[0052] S14, establish a direct modulation and direct inspection optical fiber channel model based on RNN, use the data received by the oscilloscope OSC in the training set as the RNN model input, and the data sent by the arbitrary waveform generator AWG as the label, debug the parameters and perform training, and finally obtain the direct modulation and direct inspection optical fiber channel model;

[0053] S15. Use the test set to verify the direct adjustment and direct inspection fiber channel model, and compare the output data of the direct adjustment and direct inspection fiber channel model with the labels in the test set to evaluate the accuracy of the direct adjustment and direct inspection fiber channel model.

[0054] Furthermore, S2 is specifically:

[0055] S21. A three-layer neural network is introduced in polar code encoding, with a constant t determined by experimental experience as input and a channel index in polar code encoding as output;

[0056] In the embodiment of the present invention, the value of t is selected according to the specific requirements of the system and the nature of the processed data to effectively optimize the performance of the neural network. In order to achieve the goal of outputting the channel index, it is necessary to constrain the output of the three-layer neural network to ensure that its number is N and the size is an integer between 1 and N, and these values ​​as channel indexes are not repeated. At the same time, it is also necessary to maintain the propagability of the gradient in the three-layer neural network to ensure that the parameters of the encoder and decoder neural networks can be effectively updated during the training process.

[0057] S22. Perform polarization code encoding on the information sequence u according to the channel index output by the neural network to obtain an encoded sequence x;

[0058] In the embodiment of the present invention, since the transmission sequence x consists of only 01 bits, a onehot encoding method is used to encode 2 bits of x into 4 bits and send them to the input end of the autoencoder.

[0059] S23, the encoder and decoder of the autoencoder correspond to the transmitting end and the receiving end in the optical fiber system respectively, the autoencoder is composed of two neural networks with the same structure, wherein the encoder comprises an input layer, two hidden layers and an output layer, the input layer comprises M neurons, the two hidden layers each comprise 2M neurons, the output layer comprises n neurons, and the decoder has the same structure as the encoder;

[0060] S24. Polar decoding is performed through multiple iterations of the belief propagation algorithm to calculate the bit error rate of the system.

[0061] Furthermore, the three-layer neural network in S21 is specifically:

[0062] channel index=Softmax(W2′((W1′×t+b1′))+b2′)

[0063] Where W1′ and W2′ are the weights of the fully connected layer of the three-layer neural network, b1′ and b2′ are the biases of the fully connected layer of the three-layer neural network, and Softmax represents the activation function.

[0064] Furthermore, the encoder neural network is specifically:

[0065] z=(3″((2″(L(1″×y+1″))+2″))+3″)

[0066] Where W3″, W2″, W1″ are the weights of the fully connected layer of the encoder neural network, b1″, b2″, b3″ are the biases of the fully connected layer of the encoder neural network, Sigmoid represents the activation function, y is the input of the encoder neural network, and z is the output of the encoder neural network.

[0067] Furthermore, the decoder neural network is specifically:

[0068] b=(3″′((2″′(U(1″′×a+1″′))+2″′))+3″′)

[0069] Where W3″′, W2″′, and W1″′ are the weights of the fully connected layer of the decoder neural network, b1″′, b2″′, and b3″′ are the biases of the fully connected layer of the decoder neural network, Softmax represents the activation function, a is the input of the decoder neural network, and b is the output of the decoder neural network.

[0070] In an embodiment of the present invention, the loss function is calculated by the input y of the encoder neural network and the output b of the decoder neural network. During training, the cross entropy loss function is used as the standard for training the neural network. The optimizer uses the Adam algorithm, sets the learning rate to 0.001, and uses stochastic gradient descent to optimize the parameters of the encoder neural network and the decoder neural network.

[0071] Since the input data of the direct modulation and direct detection optical fiber channel model is determined by a sliding window, the output data of the corresponding encoder needs to be processed by the sliding window before being sent to the direct modulation and direct detection optical fiber channel model.

[0072] Furthermore, S4 and later also include:

[0073] S5. Experimentally test the optimized polarization weight channel index, compare its performance with that of the traditional polarization code construction method, and calculate the improvement of the optimization method in receiving sensitivity under the same bit error rate conditions.

[0074] In an embodiment of the present invention, an optical fiber system experimental platform is built, and polarization code coded signals modulated by PAM4 are transmitted multiple times, and 2.4 million sets of data are collected, of which 1.92 million are used for training and 480,000 are used for testing. The data is input into the established direct-adjustment and direct-detection optical fiber channel model, and the direct-adjustment and direct-detection optical fiber channel model can be obtained after training. In the experiment, the polarization weight channel index after end-to-end training optimization is transmitted, and the signal after polarization code encoding and PAM4 modulation has a transmission rate of 50GBaud and a distance of 5 kilometers. By calculating the bit error rate performance of the signal and comparing it with the traditional Polar construction method, the experimental result diagram is drawn, and it can be seen that the proposed end-to-end polarization code coding optimization method has a certain improvement in the receiving sensitivity performance.

[0075] Figure 3 The code length distribution after end-to-end optimization and without end-to-end optimization is shown. It can be seen that when the code length is 4 and 6, the code length after end-to-end optimization exceeds that without end-to-end optimization. After calculation, the total code length after end-to-end optimization and without end-to-end optimization is 1294 and 1259.5 respectively. Therefore, the total code length after end-to-end optimization is higher, which improves the fault tolerance of the optical fiber communication system and achieves better BER performance. Figure 4 The experimental results show that the optimized polarization weights in Polar code transmission have superior bit error rate performance. -3 The received power sensitivity of the back-to-back transmission (BTB) link is improved by 0.9 dB at a bit error rate of 1.3 %. In a 5 km single-mode fiber transmission link, considering the influence of the frequency attenuation effect caused by dispersion on the signal, the proposed end-to-end coding scheme can improve the optical power loss by an additional 0.4 dB compared with the traditional Polar code encoder.

[0076] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0077] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An end-to-end channel coding optimization method, characterized in that: The following steps are involved: S1. Establish a direct modulation and direct detection optical fiber channel model to simulate the transmission characteristics of polarization coded signals in single-mode optical fiber; S2. Establish an end-to-end polar code optimization system; S3, integrating the direct modulation and direct detection optical fiber channel model into the end-to-end polarization code encoding optimization system to form a channel-assisted polarization code autoencoder direct modulation and direct detection optical interconnection end-to-end system; S4. Parameter tuning is performed on the channel-assisted polarization code autoencoder direct adjustment and direct detection optical interconnection system to determine the optimized polarization code encoding polarization weight channel index.

2. The end-to-end channel coding optimization method according to claim 1, characterized in that: S1 is specifically: S11. Build an optical fiber system experimental platform, including arbitrary waveform generator AWG, electric amplifier EA, laser Laser, modulator MZM, single-mode fiber SMF, variable optical attenuator VOA, photodetector PD and oscilloscope OSC; S12, repeatedly transmitting the polar code encoded by PAM4 modulation, and recording the data sent by the arbitrary waveform generator AWG and the data collected by the oscilloscope OSC; S13, using a sliding window to process the received data, and dividing the recorded data into 80% training set and 20% test set in proportion; S14, establish a direct modulation and direct inspection optical fiber channel model based on RNN, use the data received by the oscilloscope OSC in the training set as the RNN model input, and the data sent by the arbitrary waveform generator AWG as the label, debug the parameters and perform training, and finally obtain the direct modulation and direct inspection optical fiber channel model; S15. Use the test set to verify the direct adjustment and direct inspection fiber channel model, and compare the output data of the direct adjustment and direct inspection fiber channel model with the labels in the test set to evaluate the accuracy of the direct adjustment and direct inspection fiber channel model.

3. The end-to-end channel coding optimization method according to claim 1, characterized in that: S2 is specifically: S21. A three-layer neural network is introduced in polar code encoding, with a constant t determined by experimental experience as input and a channel index in polar code encoding as output; S22. Perform polarization code encoding on the information sequence u according to the channel index output by the neural network to obtain an encoded sequence x; S23, the encoder and decoder of the autoencoder correspond to the transmitting end and the receiving end in the optical fiber system respectively, the autoencoder is composed of two neural networks with the same structure, wherein the encoder comprises an input layer, two hidden layers and an output layer, the input layer comprises M neurons, the two hidden layers each comprise 2M neurons, the output layer comprises n neurons, and the decoder has the same structure as the encoder; S24. Polar decoding is performed through multiple iterations of the belief propagation algorithm to calculate the bit error rate of the system.

4. The end-to-end channel coding optimization method according to claim 3, characterized in that: The three-layer neural network in S21 is as follows: channel index=Softmax(W2′((W1′×t+b1′))+b2′) Where W1′ and W2′ are the weights of the fully connected layer of the three-layer neural network, b1′ and b2′ are the biases of the fully connected layer of the three-layer neural network, and Softmax represents the activation function.

5. The end-to-end channel coding optimization method according to claim 3, characterized in that: The encoder neural network is specifically: z=(3″((2″(L(1″×y+1″))+2″))+3″) Where W3″, W2″, W1″ are the weights of the fully connected layer of the encoder neural network, b1″, b2″, b3″ are the biases of the fully connected layer of the encoder neural network, Sigmoid and ReLU represent activation functions, y is the input of the encoder neural network, and z is the output of the encoder neural network.

6. The end-to-end channel coding optimization method according to claim 5, characterized in that: The decoder neural network is specifically: b=(3″′((2″′(U(1″′×a+1″′))+2″′))+3″′) Where W3″′, W2″′, and W1″′ are the weights of the fully connected layer of the decoder neural network, b1″′, b2″′, and b3″′ are the biases of the fully connected layer of the decoder neural network, Softmax represents the activation function, a is the input of the decoder neural network, and b is the output of the decoder neural network.

7. The end-to-end channel coding optimization method according to claim 1, characterized in that: S4 and later also include: S5. Experimentally test the optimized polarization weight channel index, compare its performance with that of the traditional polarization code construction method, and calculate the improvement of the optimization method in receiving sensitivity under the same bit error rate conditions.

Images