Optical communication signal optimization method and device based on optical neural network

Abstract

The invention discloses an optical communication signal optimization method and device based on an optical neural network, and relates to the technical field of optical communication. According to the method, an optical neural network chip is designed, and an optical transmission signal of an optical communication system is optimized on an optical path; the weight of the trained neural network is electrically loaded into the optical neural network, and the weight is continuously optimized through on-chip online training, so that the quality of optical signal optimization is improved. After training is completed, the corresponding micro-ring modulator can be designed according to the weight of each device so that all-optical processing can be achieved, photoelectric conversion is not needed, and noise caused by photoelectric devices of the system is reduced. According to the method, the optical communication signals are processed through the optical neural network, dependence on electronic computing resources is avoided, and the response speed and the extinction ratio of an optical communication system are increased.

Description

Technical Field

[0001] The present invention relates to the field of optical communication technology, and in particular to a method and device for optimizing optical communication signals based on an optical neural network. Background Art

[0002] With the rapid development of information technology, the demand for high-speed, large-capacity, and long-distance data transmission is growing. Traditional electrical communication systems can no longer meet the needs of modern communications in terms of transmission rate and distance. Therefore, optical communication technology has become a research hotspot in the future communication field due to its advantages such as high bandwidth, long distance, and low loss.

[0003] In optical communication systems, signal modulation and transmission are one of the core technologies. Return to Zero (RZ), Non-Return to Zero (NRZ) and 4-level Pulse Amplitude Modulation (PAM4) are common digital signal modulation methods. Especially in high-speed optical communication systems with a rate of more than 100Gbit / s, PAM4 is widely used due to its high efficiency and high transmission rate.

[0004] However, there are many noise sources in high-speed optical communication systems, such as current noise, thermal noise, and crosstalk noise. In addition, the signal will be affected by dispersion and nonlinear effects during transmission, resulting in distortion and inter-symbol interference. These factors will interfere with the signal transmission quality and lead to an increase in bit error rate.

[0005] To solve this problem, researchers are committed to optimizing signal processing methods, such as proposing some new signal equalization technologies to improve the transmission performance of optical communication systems. In the face of the shortcomings of existing high-speed optical communication systems, especially the increase in bit error rate and the decrease in transmission efficiency, it is urgent to develop a new type of signal optimization chip. Summary of the invention

[0006] The problem to be solved by the present invention is to provide a method and device for optimizing optical communication signals based on optical neural networks, aiming to provide an efficient, low-power and all-optical optical communication signal optimization method, which, combined with optical neural networks, can improve the transmission quality of optical communication systems, reduce bit error rates, achieve data transmission reliability, and significantly improve the overall performance and efficiency of the system.

[0007] The present invention adopts the following technical solution: including: receiving optical communication data, initializing the weight of the micro-ring modulator voltage signal, calculating the optical neural network gradient, updating the weight of the micro-ring modulator voltage signal and outputting the optical communication data, and the specific steps are as follows:

[0008] S1. Receive optical communication data:

[0009] S1.1, the optical communication system sends various optical communication data formats, the photoelectric detector detects the light intensity of different optical communication data formats, and constructs the eye diagram and data set I of the sent data;

[0010] S1.2, connecting the optical signal input module of the optical chip to the receiving end of the optical communication system through an optical fiber;

[0011] S2. Initialize the weight of the micro-ring modulator voltage signal:

[0012] S2.1, constructing an electrical neural network model;

[0013] S2.2, loading the weights of the electrical neural network into the optical neural network, training the electrical neural network, continuously updating the weights, determining the final weights, and converting the final weights of the electrical neural network into a voltage signal to drive the micro-ring modulator;

[0014] S3. Calculate the optical neural network gradient:

[0015] S3.1, forward propagation, completing the input and output of optical communication signals in the optical neural network;

[0016] S3.2, calculating the loss, calculating the loss between the output light intensity in the optical neural network and the ideal output light intensity;

[0017] S3.3, gradient descent, calculates the gradient between layers in the optical neural network;

[0018] S4. Update the weight of the micro-ring modulator voltage signal:

[0019] During the training process of the optical neural network system, if the actual communication signal index is not within the expected target signal index, the error is back-propagated to update the weight parameters of the optical neural network;

[0020] S5. Output optical communication data:

[0021] The optical signal output module of the optical chip is connected through optical fiber and directly connected to the receiving end of the optical communication system.

[0022] Preferably, step S2.1, constructing an electrical neural network model:

[0023] Define the neural network model input as X, the original RZ, NRZ and PAM4 data format signals sent in the optical communication system are defined as the ideal output Y' of the neural network model, the actual output is Y, and the electrical neural network model f is constructed, then:

[0024] Y=f(X;w0)

[0025] Where w0 is the initialization weight of the electrical neural network, the network loss is the mean square error between the ideal output Y' and the actual output Y, and the scale of the electrical neural network model f can be expanded or reduced with the mode of the input signal.

[0026] Preferably, step S2.2, loading the weights of the electrical neural network into the optical neural network:

[0027] The electrical neural network is trained, and the weights are continuously updated to determine the final weight w. The neural network weights are converted into a voltage signal v that drives the micro-ring modulator w , the steps are as follows:

[0028] Assume that the voltage corresponding to the maximum output of N micro-ring modulators is V max , the voltage corresponding to the minimum output is V min , normalize the final weight of the electrical neural network and map its value to between 0 and 1 to obtain w', where w ′ =w1,w2,…,w N , the weights corresponding to N voltages can be obtained by converting them through the following formula:

[0029] v w =Kw′(V max -V min )

[0030] Where K = k1, k2, ..., k N For each micro-ring modulator, the power factor, v w Contains a range of voltage values, namely V1, V2 to V N The obtained N voltages are loaded into the corresponding micro-ring modulator to realize the weight initialization of the voltage signal of the micro-ring modulator.

[0031] Preferably, step S3.1, forward propagation:

[0032] After initializing the weights of the voltage signals of N micro-ring modulators, the receiver of the optical communication system sends data format information such as RZ, NRZ and PAM4 to the optical neural network, and its power is recorded as P in , the data is input into the neural network and the final output is the prediction signal, that is, the optimized RZ, NRZ and PAM4 data format information, and its power is recorded as P out , the original RZ, NRZ and PAM4 data format information, its power is recorded as P label ,Due to the inherent errors of the optical neural network system, the effect is poor compared with the results of the electrical neural network.

[0033] Preferably, step S3.2, calculate the loss:

[0034] The output of the optical neural network is P out , labelled P label , compare whether the power output of the two is in the same order of magnitude, and scale their labels to make them in the same order of magnitude. Then calculate the output P out With label P label The loss function uses the mean square error:

[0035] L P =MSE P =(P out -P label ) 2

[0036] Preferably, step S3.3, gradient descent:

[0037] The calculation formula of the output layer gradient is:

[0038]

[0039] The output of the hidden layer is P Z =v w ·P in , then the gradient calculation formula of the hidden layer output to the weight is:

[0040]

[0041] The gradient of the interneuron is:

[0042] ΔP Z =ΔP out *P in

[0043] The calculation formula of hidden layer weight gradient is:

[0044] Δv w =ΔP Z *P in

[0045] According to the above formula, the gradient values ​​of each part in the optical neural network can be obtained.

[0046] Preferably, step S4, updating the weight of the voltage signal of the micro-ring modulator:

[0047] During the training process of the optical neural network system, if the actual communication signal index is not within the expected target signal index, the error is back propagated. That is, the output error is back propagated layer by layer through the hidden layer to the input layer, and the error is apportioned to the voltage controller of all micro-ring modulator voltages in each layer, and the error signal obtained from each layer is used as the basis for adjusting the weight of each unit.

[0048] The weight update formula of the optical neural network system is:

[0049]

[0050] Where θ is the learning rate of the optical neural network training. Then the updated micro-ring modulator voltage signal is loaded to the corresponding voltage controller.

[0051] The technical solution of the present invention also provides: an optical communication signal optimization device based on an optical neural network, used to implement any of the optical communication signal optimization methods described above, comprising: an optical signal input module, an optical neural network module, a first voltage source module, a second voltage source module and an optical signal output module;

[0052] An optical signal input module, used to output an optical signal in an optical communication system to an optical neural network module;

[0053] The optical neural network module includes a plurality of micro-ring modulators for encoding and decoding input optical signals, wherein the encoding and decoding weights are controlled by the micro-ring modulators, and an external first voltage source module and a second voltage source module jointly drive the micro-ring modulators to work, adjust the weights of the optical neural network, and output the encoded optical signals;

[0054] A first voltage source module outputs a plurality of regulated power supplies for controlling the output optical power of the micro-ring modulator in the upper part of the optical neural network module;

[0055] A second voltage source module outputs a plurality of regulated power supplies for controlling the output light intensity of the micro-ring modulator in the lower part of the optical neural network module;

[0056] The optical signal output module is used to receive the optical signal output by the optical neural network module and output it through the optical fiber.

[0057] Preferably, the modulation voltage range of the micro-ring modulator is 0V to 3V, and the resonance wavelength modulation range is 0nm to 2nm.

[0058] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0059] 1. The optical communication signal optimization method of the present invention adopts an optical neural network chip and a supporting algorithm, focusing on optimizing data format signals such as RZ, NRZ and PAM4 of optical communication transmission, improving the extinction ratio of the transmission signal, and reducing the bit error rate. The outstanding feature of the present invention is that it can optimize the signal at the terminal of the optical communication system by light-in and light-out, and process the optical communication signal in the optical domain. And when the weight of the optical neural network is fixed, all-optical computing can be realized without involving optoelectronic conversion equipment, and can be widely used in optical communication, optical computing and other fields, with good practicality.

[0060] 2. The optical communication signal optimization method of the present invention adopts an online training algorithm to implement training on the optical neural network chip, and performs end-to-end optimization on the actual communication signal, which can further reduce the error of directly loading the electrical neural network weights into the optical chip and improve the accuracy of the system. The system is adaptable to various complex electromagnetic environments. Whether it is a strong signal within the line of sight range or a weak signal strength at a long distance beyond the line of sight, the optical communication signal can be optimized by deploying the chip, which has good practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The schematic diagram of the optical neural network chip for optimizing optical communication signals of the present invention;

[0062] Figure 2 A flow chart of the method for optimizing optical communication signals of the present invention;

[0063] Figure 3 Schematic diagram of online training of the optical neural network chip of the present invention. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all 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.

[0065] The purpose of the present invention is to provide a method and device for optimizing optical communication signals based on optical neural networks, which can realize light-to-light optical signal optimization at the receiving end of the optical communication system, improve the stability of the optical communication system, and reduce the bit error rate.

[0066] In one embodiment of the present invention, an optical communication signal optimization device based on an optical neural network, such as Figure 1 As shown, it specifically includes: an optical signal input module, an optical neural network module, a first voltage source module, a second voltage source module and an optical signal output module.

[0067] An optical signal input module, used to output an optical signal in an optical communication system to an optical neural network module;

[0068] The optical neural network module includes several micro-ring modulators for encoding and decoding input optical signals. The encoding and decoding weights are controlled by the internal micro-ring modulators. The micro-ring modulators are driven by external voltages to output encoded optical signals.

[0069] A first voltage source module outputs a plurality of regulated power supplies for controlling the output light intensity of the micro-ring modulator in the upper part of the optical neural network module, that is, the weight of the optical neural network;

[0070] The second voltage source module outputs a plurality of regulated power supplies for controlling the output light intensity of the micro-ring modulator in the lower part of the optical neural network module, that is, the weight of the optical neural network;

[0071] The optical signal output module is used to receive the optical signal from the optical neural network module and output it through the optical fiber.

[0072] In this embodiment, there are 10 micro-ring modulators, and their initial resonant wavelengths are λ1, λ2, ..., λ 10 The first voltage source module outputs 5 regulated power supplies, which respectively control the micro-ring modulators with initial resonant wavelengths of λ1, λ2, ..., λ5, and the second voltage source module outputs 5 regulated power supplies, which respectively control the micro-ring modulators with initial resonant wavelengths of λ6, λ7, ..., λ 10 The input optical signal is input from the micro-ring modulator with an initial resonance wavelength of λ1. After passing through 10 micro-ring modulators, the optical signal is input from the micro-ring modulator with an initial resonance wavelength of λ1. 10 The micro-ring modulator outputs the optical signal.

[0073] Based on the above-mentioned optical communication signal optimization device based on optical neural network, the corresponding optimization method of this embodiment includes receiving optical communication data part, initializing the weight part of the micro-ring modulator voltage signal, calculating the optical neural network gradient part, updating the weight part of the micro-ring modulator voltage signal and outputting the optical communication data part. The process is as follows Figure 2 shown.

[0074] The part receiving optical communication data includes:

[0075] (1) The optical communication system sends the optical communication data format, and the photodetector detects the light intensity of different data formats to construct the eye diagram and data set I of the sent data;

[0076] (2) connecting the optical signal input module of the optical chip to the receiving end of the optical communication system through an optical fiber;

[0077] Specifically, in the process of constructing data set I, it includes three types of data formats: RZ, NRZ and PAM4. The input of each type of data is the light intensity received by the photodetector after passing through the optical communication system, and the label of each type of data is the light intensity directly received by the photodetector when the signal transmitted by the optical communication system.

[0078] In this embodiment, the number of samples of each type of data is 10,000, of which the number of samples of the test set is 2,000.

[0079] Initializing the weight part of the micro-ring modulator voltage signal includes:

[0080] (1) Constructing an electrical neural network model:

[0081] Defined as the neural network model input X, the original RZ, NRZ and PAM4 data format signals sent in the optical communication system are defined as the ideal output Y' of the neural network model, and the actual output is Y. Constructing the electrical neural network model f, we have:

[0082] Y=f(X;w0)

[0083] Where w0 is the initialization weight of the electrical neural network, and the network loss is the mean square error between the ideal output Y' and the actual output Y. The scale of the electrical neural network model f can be expanded or reduced with the mode of the input signal.

[0084] (2) Loading the weights of the electrical neural network into the optical neural network:

[0085] The electrical neural network is trained, and the weights are continuously updated to determine the final weight w. The neural network weights are converted into a voltage signal v that drives the micro-ring modulator w , the steps are as follows:

[0086] In this embodiment, the voltage corresponding to the maximum output of the 10 micro-ring modulators in the system is assumed to be V max , the voltage corresponding to the minimum output is V min , normalize the final weight of the electrical neural network and map its value to between 0 and 1 to obtain w', where w ′ =w1,w2,…,w 10 , the weights corresponding to the 10 voltages can be obtained by converting them through the following formula:

[0087] v w =Kw′(V max -V min )

[0088] Where K = k1, k2, …, k 10 For each micro-ring modulator, the power factor, v w Contains a range of voltage values, namely V1, V2 to V 10 The adjustable voltage.

[0089] The obtained 10 voltages are loaded into the corresponding micro-ring modulator to realize the weight initialization of the voltage signal of the micro-ring modulator.

[0090] The calculation of the optical neural network gradient part includes:

[0091] (1) Forward propagation:

[0092] After initializing the weights of the voltage signals of 10 micro-ring modulators, the receiver of the optical communication system sends data format information such as RZ, NRZ and PAM4 to the optical neural network, and its power is recorded as P in , the data is input into the neural network and the final output is the prediction signal, that is, the optimized RZ, NRZ and PAM4 data format information, and its power is recorded as P out , the original RZ, NRZ and PAM4 data format information, its power is recorded as P label ,Due to the inherent errors of the optical neural network system, the effect is poor compared with the results of the electrical neural network;

[0093] (2) Calculate the loss:

[0094] The output of the optical neural network is P out , labelled P label , compare whether the power output of the two is in the same order of magnitude, and scale their labels to make them in the same order of magnitude. Then calculate the output P out With label P label The loss function uses the mean square error:

[0095] L P =MSE P =(P out -P label ) 2

[0096] (3) Gradient descent:

[0097] The calculation formula of the output layer gradient is:

[0098]

[0099] The output of the hidden layer is P Z =v w ·P in , then the gradient calculation formula of the hidden layer output to the weight is:

[0100]

[0101] The gradient of the interneuron is:

[0102] ΔP Z =ΔP out *P in

[0103] The calculation formula of hidden layer weight gradient is:

[0104] Δv w =ΔP Z *P in

[0105] According to the above formula, the gradient values ​​of each part in the optical neural network can be obtained.

[0106] The weight part of updating the voltage signal of the micro-ring modulator includes:

[0107] During the training process of the optical neural network system, if the actual communication signal index is not within the expected target signal index, the error is back propagated. That is, the output error is back propagated layer by layer through the hidden layer to the input layer, and the error is apportioned to the voltage controller of all micro-ring modulator voltages in each layer, and the error signal obtained from each layer is used as the basis for adjusting the weight of each unit.

[0108] The weight update formula of the optical neural network system is:

[0109]

[0110] Where θ is the learning rate of the optical neural network training. Then the updated micro-ring modulator voltage signal is loaded to the corresponding voltage controller.

[0111] The output optical communication data part includes:

[0112] The optical signal output module of the optical chip is connected through optical fiber and can be directly connected to the receiving end of the optical communication system;

[0113] A complete online training solution for optical neural network chips, such as Figure 3 shown.

[0114] In summary, the present invention avoids dependence on electronic computing resources through light-to-light computing, improves the anti-interference capability of the optical communication system, improves the stability of the optical communication system through optimization of the communication light, realizes online training in the deployment of the optical neural network, and reduces the bit error rate.

[0115] 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.

[0116] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those of ordinary skill in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as limiting the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principles of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for optimizing optical communication signals based on optical neural networks, characterized in that: include: The specific steps of receiving optical communication data, initializing the weight of the micro-ring modulator voltage signal, calculating the optical neural network gradient, updating the weight of the micro-ring modulator voltage signal and outputting optical communication data are as follows: S1. Receive optical communication data: S1.1, the optical communication system sends various optical communication data formats, the photoelectric detector detects the light intensity of different optical communication data formats, and constructs the eye diagram and data set I of the sent data; S1.2, connecting the optical signal input module of the optical chip to the receiving end of the optical communication system through an optical fiber; S2. Initialize the weight of the micro-ring modulator voltage signal: S2.1, constructing an electrical neural network model; S2.2, loading the weights of the electrical neural network into the optical neural network, training the electrical neural network, continuously updating the weights, determining the final weights, and converting the final weights of the electrical neural network into a voltage signal to drive the micro-ring modulator; S3. Calculate the optical neural network gradient: S3.1, forward propagation, completing the input and output of optical communication signals in the optical neural network; S3.2, calculating the loss, calculating the loss between the output light intensity in the optical neural network and the ideal output light intensity; S3.3, gradient descent, calculates the gradient between layers in the optical neural network; S4. Update the weight of the micro-ring modulator voltage signal: During the training process of the optical neural network system, if the actual communication signal index is not within the expected target signal index, the error is back-propagated to update the weight parameters of the optical neural network; S5. Output optical communication data: The optical signal output module of the optical chip is connected through optical fiber and directly connected to the receiving end of the optical communication system.

2. The method for optimizing optical communication signals based on optical neural network according to claim 1, characterized in that: In step S1.1, the data set I includes three data formats: RZ, NRZ, and PAM4. The input of each type of data is the light intensity received by the photodetector after passing through the optical communication system, and the label of each type of data is the light intensity directly received by the photodetector when the signal transmitted by the optical communication system is transmitted. A test set is constructed, and the number of samples in the test set is not less than 20% of the total samples of data set I.

3. The method for optimizing optical communication signals based on optical neural network according to claim 1, characterized in that: In step S2.1, an electrical neural network model is constructed, specifically including: S2.1.

1. Define the neural network model input as X, the actual output as Y, and the original RZ, NRZ and PAM4 data format signals sent in the optical communication system as the ideal output Y' of the neural network model; S2.1.2, construct the electrical neural network model f, with the following formula: Y=f(X;w0) Where w0 is the initialization weight of the electrical neural network, and the network loss is the mean square error between the ideal output Y' and the actual output Y; S2.1.

3. The scale of the electrical neural network model f expands and shrinks with the pattern of the input signal.

4. The method for optimizing optical communication signals based on optical neural network according to claim 3, characterized in that: In step S2.2, the weights of the electrical neural network are loaded into the optical neural network, specifically including: S2.2.1, the optical neural network includes N micro-ring modulators, the voltage corresponding to the maximum output of the micro-ring modulator is V max , the voltage corresponding to the minimum output is V min , normalize the final weight w of the electrical neural network and map its value to between 0 and 1, and get w', w'=w1,w2,…,w N , where w i represents the final weight of the output voltage of the i-th micro-ring modulator, i=1,2,...,N; S2.2.

2. The final weights of the output voltages of the N micro-ring modulators are calculated and converted into voltage signals for driving the micro-ring modulators. The conversion formula is as follows: in w =Kw′(V max -V min ) Where K = k1, k2, ..., k N , k i is the power coefficient of the i-th micro-ring modulator, v w is the driving voltage signal, v w Contains N tunable voltage values; S2.2.

3. Load the obtained N tunable voltage values ​​into the corresponding micro-ring modulators to achieve weight initialization of the voltage signals of the N micro-ring modulators.

5. The method for optimizing optical communication signals based on optical neural network according to claim 4, characterized in that: In step S3.1, the forward propagation initializes the weights of the voltage signals of the N micro-ring modulators, and the receiver of the optical communication system sends RZ, NRZ and PAM4 data format information to the optical neural network. The input power is P in ; The data is transmitted to the optical neural network and the predicted signal is output to obtain the optimized RZ, NRZ and PAM4 data format information. The output power is P out ; Original RZ, NRZ and PAM4 data format information, original power is P label .

6. The method for optimizing optical communication signals based on optical neural network according to claim 5, characterized in that: In step S3.2, the loss is calculated and the output power P is calculated. out With the original power P label The loss of L P , the loss function uses mean square error, which is expressed as follows: L P =MSE P =(P out -P label ) 2 Among them, MSE is the mean square error function of power.

7. The method for optimizing optical communication signals based on optical neural network according to claim 6, characterized in that: In step S3.3, the gradient descent includes the following sub-steps: S3.3.

1. Calculate the output layer gradient ΔP out : Where d is the gradient calculation function; S3.3.

2. Calculate the gradient P of the hidden layer output with respect to the weight in : Among them, P Z is the output of the hidden layer, P Z =v w ·P in ; S3.3.

3. Calculate the gradient ΔP of the interneuron Z : ΔP Z =ΔP out *P in S3.3.

4. Calculate the hidden layer weight gradient Δv w : Δv w =ΔP Z *P in 。 8. The method for optimizing optical communication signals based on optical neural network according to claim 7, characterized in that: In step S4, the weight of the voltage signal of the micro-ring modulator is updated, the error of the output voltage is transmitted back to the input layer layer by layer through the hidden layer, and the error is apportioned to the voltage controllers of all the micro-ring modulator voltages in each layer, and the error signal is obtained from each layer as the basis for adjusting the weight of the voltage signal of each micro-ring modulator of the optical neural network; The weight update formula of the optical neural network is: Where θ is the learning rate of the optical neural network training, and the updated micro-ring modulator voltage signal v w Loaded to the corresponding voltage controller.

9. A device for optimizing optical communication signals based on optical neural network, used to implement the method according to any one of claims 1 to 8, characterized in that: include: An optical signal input module, an optical neural network module, a first voltage source module, a second voltage source module and an optical signal output module; The optical signal input module is used to output the optical signal in the optical communication system to the optical neural network module; The optical neural network module includes a plurality of micro-ring modulators for encoding and decoding input optical signals, the encoding and decoding weights are controlled by the micro-ring modulators, and the micro-ring modulators are driven externally by a first voltage source module and a second voltage source module to adjust the weights of the optical neural network and output the encoded optical signal; The first voltage source module outputs a plurality of regulated power supplies for controlling the output optical power of the micro-ring modulator in the upper part of the optical neural network module; The second voltage source module outputs a plurality of regulated power supplies for controlling the output light intensity of the micro-ring modulator in the lower half of the optical neural network module; The optical signal output module is used to receive the optical signal output by the optical neural network module and output it through optical fiber.

10. The device for optimizing optical communication signals based on optical neural network according to claim 9, characterized in that: The modulation voltage range of the micro-ring modulator is 0V to 3V, and the resonance wavelength modulation range is 0nm to 2nm.

Images