A polarization-weighted based photonic convolutional neural network system and implementation method

Abstract

The application discloses a kind of polarization-weighted photonic convolutional neural network system and implementation method, its system is generated by tunable laser different wavelength light source, through polarization maintaining optical fiber by manual extrusion type polarization controller or electrically controlled polarization controller adjustment polarization state of optical signal, signal is combined again through wave division multiplexer, and input signal generated by high-speed digital-to-analog converter is loaded on optical intensity modulator.Light signal after modulation is amplified by optical amplifier, and by delay optical fiber, different wavelength channel one-bit information is misaligned on time domain, is collected by photodetector, and after conversion into digital signal by analog-digital converter, it is processed by microprocessor, and the convolution operation formed by multiplication and addition is realized.The method of the application is different from the existing convolution kernel loading mode, the angle of polarization controller can be directly adjusted using the pre-trained convolution kernel weight, the loading of convolution kernel is realized, the system cost is significantly reduced, and the system complexity is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photonic convolutional neural network, and particularly relates to a photonic convolutional neural network system based on polarization weighting and an implementation method. BACKGROUND

[0002] Convolutional neural network as an important model has shown great potential in image recognition, speech processing and natural language processing and other fields. However, the traditional electronic computing equipment faces the limitations of computing speed and energy consumption in processing large-scale data and complex operations, which prompts people to seek new computing models and technical means to improve the computing efficiency and processing capacity.

[0003] Optical computing as a new technology uses the high-speed parallel transmission characteristics and large bandwidth advantages of optical signals to provide new possibilities for solving the bottleneck problem of traditional computing equipment. As a frontier technology combining optical computing and deep learning, photonic convolutional neural network is attracting widespread attention and research in academia and industry. In the photonic convolutional neural network, the application of optical devices is crucial. For example, tunable lasers, polarization controllers, wavelength division multiplexers, optical intensity modulators and other devices, which not only can provide the generation, modulation and control functions of optical signals, but also can realize complex optical operations such as spectral processing, frequency selection and wavelength modulation.

[0004] Although optical computing has the advantages of high speed and parallel processing in theory, it faces many challenges in practical applications, such as the stability of optical signals, the integration performance of devices, and the complexity of the system. The existing technology proposes an optical vector convolution accelerator design based on integrated frequency comb light source to realize the simultaneous interleaving of time, wavelength and space dimensions (see [Xu X Y, Tan M X, Corcoran B, et al. 11TOPS photonic convolutional accelerator for optical neural networks[J]. Nature 2021, 589(7840): 44-51.]). There is also an optimization method for optical convolution kernel in optoelectronic hybrid convolutional neural network (OHCNN), which tests the accuracy of the network architecture with Fashion-MNIST dataset. The optimized optical convolution kernel improves the accuracy by 7.5%. The energy efficiency ratio (EER) of the dual-channel network is 46.7% higher than that of the single-channel network (see [Xu X F, Zhu LQ, Zhuang W, et al. Optimization of optical convolution kernel of optoelectronic hybrid convolution neural network[J]. Optoelectronics Letters, 2022, 18(3): 181-186.]).

[0005] With the progress of optical technology and the continuous optimization of optical devices, optical convolutional neural networks have great application potential in large-scale data processing, high-efficiency computing and real-time signal processing, and further reducing system complexity and cost have significant research significance. SUMMARY

[0006] The problem to be solved by the present application is to provide a polarization-weighted photonic convolutional neural network system and implementation method, which utilizes the high speed and parallel processing capability of optical systems to improve the computing efficiency of convolutional neural networks, is suitable for large-scale data processing and real-time signal processing scenarios, can significantly reduce system cost, and reduce system complexity.

[0007] The present application adopts the following technical scheme: a polarization-weighted photonic convolutional neural network system, comprising: a laser module, a polarization module, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a high-speed digital-to-analog converter, a delay optical fiber, a photodetector, an analog-to-digital converter, and a microprocessor.

[0008] The laser module is constructed by several tunable lasers, N tunable lasers emit N wavelengths with equal wavelength interval as the light source part of the system;

[0009] The polarization module sets a corresponding number of polarization controllers based on the number of tunable lasers, and the specific number N is determined by the size of the convolution kernel; the N polarization controllers respectively receive the output signals of each tunable laser, and adjust the polarization state of the optical signal through the corresponding polarization controller;

[0010] The wavelength division multiplexer is connected to the polarization controller and combines the laser signals of different wavelengths emitted by the tunable laser and sends them to the optical intensity modulator;

[0011] The high-speed digital-to-analog converter is used to generate the waveform of the input signal, and after generating the waveform according to the one-dimensional signal processed by the microprocessor, it is loaded onto the optical intensity modulator;

[0012] The optical amplifier is arranged after the optical intensity modulator and is used to amplify the optical signal that needs to be calculated by optical convolution, and is output to the photodetector through a delay optical fiber; the delay optical fiber is used to produce a time delay of one bit of information between the time domain waveforms of each wavelength;

[0013] The photodetector is used to detect the intensity of the optical signal amplified by the optical amplifier and output the total power of all wavelengths of light in each time interval, and convert it into an electrical signal and input it to the analog-to-digital converter;

[0014] The analog-to-digital converter converts the analog signal obtained by the photodetector into a digital signal and outputs it to the microprocessor for processing;

[0015] The microprocessor processes the output signal to process the image data into a one-dimensional signal and sends it to the high-speed digital-to-analog converter.

[0016] Further, the polarization controller is connected to the output of the tunable laser through a polarization maintaining optical fiber, and the polarization controller outputs elliptical polarized light by adjusting the pressure on the polarization maintaining optical fiber, and changes the direction angle of the elliptical polarized light by adjusting the angle, to ensure that the power of the optical signals entering the different channels of the modulator is consistent and maximum;

[0017] The polarization controller and the wavelength division multiplexer are also connected through a polarization maintaining optical fiber to maintain the polarization state of the optical signal.

[0018] The wavelength division multiplexer, optical intensity modulator, optical amplifier, delay optical fiber, photodetector, analog-to-digital converter, microprocessor and high-speed digital-to-analog converter are connected in sequence through a single-mode optical fiber.

[0019] Further, the polarization controller is a manual extrusion type polarization controller or an electrically controlled polarization controller, when being the electrically controlled polarization controller, the polarization controller is further connected with a microprocessor, and the angle of the polarization controller is changed by a stepping motor controlled by an electric signal sent by the microprocessor.

[0020] Further, in the polarization module, the polarization controller is at least one group, when the polarization controller is multiple groups, the output signal of the first group of polarization controllers is further adjusted by subsequent groups of polarization controllers, and then enters a wavelength division multiplexer for signal merging, so as to ensure that the optical signal of each wavelength reaches the optimal polarization state before merging, and the extinction ratio is improved.

[0021] The technical scheme of the present application also provides an implementation method of a photon convolutional neural network based on polarization weighting, which is applied to the system and loads a convolution kernel to an optical wave, and includes the following steps:

[0022] S1, input light source:

[0023] S1.1, N tunable lasers respectively output N wavelengths of light signals with equal wavelength intervals as system input light sources;

[0024] S1.2, after the output of the tunable laser is connected to the corresponding polarization controller, the polarization control signals from different tunable lasers are merged by a wavelength division multiplexer, and the merged optical signal is connected to an optical intensity modulator;

[0025] S1.3, before optical convolution, the polarization controller is used to adjust the power of different channel optical signals to be consistent and as large as possible, and an optical spectrum analyzer is connected to the output of the optical intensity modulator, and the power of each wavelength of optical signal is obtained by observing the waveform;

[0026] S2, load convolution kernel:

[0027] S2.1, change the angle of the polarization controller, and record the optical signal power corresponding to each angle;

[0028] S2.2, convert the pre-trained weight value into a power value, calculate the angle of the corresponding polarization controller, and control each polarization controller to adjust to a predetermined angle, so that the polarization state of each optical signal matches the corresponding weight value;

[0029] S2.3, after the optical signal is adjusted by the polarization controller, it enters a wavelength division multiplexer in sequence for signal merging, and in the merging process, each different wavelength of optical signal carries the information of the corresponding convolution kernel weight;

[0030] S3, multiplication and addition implementation:

[0031] S3.1, the combined optical signal is modulated by an optical intensity modulator, and an input signal waveform generated by a high-speed digital-to-analog converter is loaded to the optical intensity modulator, so that the optical signal produces a corresponding modulation effect in the time domain, and multiplication of the convolution kernel and the image signal is realized;

[0032] S3.2, the modulated optical signal is input into an optical amplifier for amplification, and the optical intensity modulator loads a bit stream at a certain rate. After the optical signals of several wavelength channels pass through the modulator and are subjected to delay optical fibers, time delay differences are generated between them, so that the data information carried by them is staggered by one bit information time in the time domain.

[0033] S3.3, the optical signal subjected to the delay processing enters a photodetector for conversion, the optical signals staggered with each other are received by the photodetector, and the values obtained by the multiplication in step S3.1 are accumulated to realize accumulation of the convolution operation; the photodetector converts the optical signal into an electrical signal, which is converted into a digital signal after an analog-to-digital converter, and is processed by a microprocessor to complete nonlinear, pooling, and full connection operations, and a complete photonic convolutional neural network is constructed.

[0034] S4, output and processing:

[0035] The microprocessor processes the collected electrical signal, processes the image data into a one-dimensional signal, obtains the feature map restored after the optical calculation, and sends it to the high-speed digital-to-analog converter.

[0036] Further, the polarization controller changes the direction angle of the elliptically polarized light by adjusting the angle. When the polarization controller is a manual extrusion type polarization controller, the angle is adjusted manually. When the polarization controller is an electrically controlled polarization controller, the angle is controlled by the electrical signal sent by the microprocessor to change the angle of the polarization controller.

[0037] Further, in step S3.3, the photonic convolutional neural network includes a multi-layer optical convolutional network, the multi-layer optical convolutional network includes a plurality of optical convolutional modules connected in series or a same group of optical convolutional modules used multiple times, each optical convolutional module includes a tunable laser, a polarization controller, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a delay optical fiber, a photodetector, and an analog-to-digital converter; the output of each optical convolutional module is used as the input of the next optical convolutional module, and multi-layer convolution is realized through the multi-layer structure.

[0038] Further, in step S3.4, the microprocessor processes the collected electrical signal, expands the one-dimensional data into a two-dimensional image, obtains the feature map restored after the optical calculation, stores the result through the storage device of the microprocessor, and / or directly uses the result for deep learning network training and inference process, displays through a visualization tool, and sends to the high-speed digital-to-analog converter.

[0039] Compared with the prior art, the application has the following technical effects:

[0040] 1. The photon convolutional neural network implementation method based on polarization weighting utilizes the high speed and parallel processing capability of the optical system to improve the calculation efficiency of the convolutional neural network, is suitable for large-scale data processing and real-time signal processing scenes, can significantly reduce the system cost, and reduces the system complexity.

[0041] 2. The photon convolutional neural network system and implementation method based on polarization weighting utilize polarization to load the convolution kernel, effectively realize optical convolution operation by introducing polarization control and optical coupling technology, and improve the performance ratio of the system. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is the photon convolutional neural network structure diagram of the manual polarization controller of the application;

[0043] Figure 2 is a graph of the relationship between the angle and power of the manual polarization controller used in the application;

[0044] Figure 3 is a schematic diagram of the convolution calculation of the application;

[0045] Figure 4 is the photon convolutional neural network structure diagram of the electric polarization controller of the application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the application clearer, the technical scheme of the application will be further described in detail below with reference to the drawings. The described embodiments are only a part of the embodiments involved in the application. All non-innovative embodiments of other researchers in the field on the basis of the embodiments belong to the protection scope of the application. At the same time, the step numbers in the embodiments are only set for the convenience of description and explanation, and the order between the steps is not limited in any way. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0047] The photon convolutional neural network system based on polarization weighting comprises a laser module, a polarization module, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a high-speed digital-to-analog converter, a delay optical fiber, a photodetector, an analog-to-digital converter and a microprocessor. Based on the system, the loading of the convolution kernel is further realized by polarization, which can significantly reduce the system cost and reduce the system complexity.

[0048] Embodiment one:

[0049] The photon convolutional neural network system based on polarization weighting comprises a laser module, a polarization module, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a high-speed digital-to-analog converter, a delay optical fiber, a photodetector, an analog-to-digital converter and a microprocessor. Based on the system, the loading of the convolution kernel is further realized by polarization, which can significantly reduce the system cost and reduce the system complexity. Figure 1As shown, including: laser module, polarization module, wavelength division multiplexer, optical intensity modulator, optical amplifier, high-speed digital-to-analog converter, delay optical fiber, photodetector, analog-to-digital converter.

[0050] The laser module includes N tunable lasers, and the polarization module includes N manual squeeze fiber polarization controllers, and the size of N is determined by the size of the convolution kernel.

[0051] Based on the above-mentioned polarization-weighted photonic convolutional neural network implementation method, the following parts are included:

[0052] 1. Input light source part:

[0053] Step 1.1: Construct a laser module composed of N tunable lasers,

[0054] In particular, N is four in this embodiment, four tunable lasers are constructed, four wavelengths with equal wavelength interval are used, and a 2x2 convolution kernel is used as the light source part of the entire system;

[0055] Step 1.2: The laser output is connected to the manual squeeze fiber polarization controller, and the wavelength division multiplexer is used to combine the polarization control signals from different lasers, and the combined optical signal is connected to the modulator;

[0056] Step 1.3: Since the modulator is sensitive to polarization, before multiplication and addition of optical convolution, the power of optical signals in different channels is adjusted to be consistent and as large as possible by using the manual squeeze fiber polarization controller, which is mainly used to adjust the elliptical polarization. Different powers will affect the results of subsequent addition and multiplication. The power of each wavelength can be observed on the spectrum analyzer connected to the modulator output.

[0057] 2. After adjusting the input light source part, load the convolution kernel, the specific steps are as follows:

[0058] Step 2.1: Change the angle of the polarization controller and record the optical signal power corresponding to each angle;

[0059] Step 2.2: Convert the pre-trained weight value to power value, and calculate the angle of the manual squeeze fiber polarization controller.

[0060] The angle and power relationship of the manual squeeze fiber polarization controller used in this embodiment is as shown in the figure: Figure 2 When the angle of the manual squeeze fiber polarization controller is 48° and 88°, the minimum power -14.2dbm can be obtained, and when the angle of the manual squeeze fiber polarization controller is 67°, the maximum power -1.7dbm can be obtained. Adjust each manual squeeze fiber polarization controller to the predetermined angle so that the polarization state of each optical signal matches the corresponding weight value.

[0061] Step 2.3: The manual squeezing fiber polarization controller can use two or more groups, repeat step 2.1 and step 2.2, for example, the output signal of the first group of manual squeezing fiber polarization controller is further adjusted by the second group of manual squeezing fiber polarization controller, and finally enters the wavelength division multiplexer for signal merging;

[0062] Step 2.4: After the optical signal is adjusted by the manual squeezing fiber polarization controller, it enters the wavelength division multiplexer for signal merging. In the merging process, each different wavelength of optical signal will carry the corresponding convolution kernel weight information;

[0063] 3, after completing the convolution kernel loading, the multiplication and addition operations are performed, as shown in Figure 3 , the specific steps are as follows:

[0064] Step 3.1: The merged optical signal is modulated by the optical intensity modulator, and the input signal waveform generated by the high-speed digital-to-analog converter is loaded to the modulator.

[0065] In this embodiment, the speed of the high-speed digital-to-analog converter is 12.5Gbps, the vertical resolution is 10bit, the bandwidth is 5.3GHz, and the time corresponding to the emission of one bit is 80ps, so that the optical signal produces a corresponding modulation effect in the time domain, and finally realizes the multiplication of the convolution kernel and the one-dimensional image signal.

[0066] Wherein, the method of expanding two-dimensional image signal into one-dimensional signal is as follows:

[0067] First, the input image is divided into blocks, and each convolution kernel is slid according to the step window, and the covered area is extracted, and each area is expanded into a row vector; Finally, all the expanded column vectors are arranged in order to form a large row vector; The convolution kernel is expanded into a row vector;

[0068] Step 3.2: The modulated optical signal enters the optical bit information amplifier for amplification to ensure that the signal intensity is sufficient for subsequent processing. The modulator loads the bit stream at a certain rate, and the light of the four wavelength channels is loaded with the same data after passing through the modulator. After passing through the delay fiber, they produce a time delay difference, which makes the data information they carry be offset by one bit in the time domain;

[0069] Step 3.3: The optical signal after delay processing enters the photodetector for conversion.

[0070] In this embodiment, the bandwidth of the photodetector is 40GHz, and the photodetector converts the optical signal into an electrical signal. The collected electrical signal is processed by the microprocessor to complete the nonlinear, pooling, fully connected and other operations, and a complete photonic convolutional neural network is constructed.

[0071] After completing the addition and multiplication in the convolution operation, the result is finally output and processed:

[0072] The microprocessor processes the collected electrical signals, re-expanding the one-dimensional data into a two-dimensional image, thus obtaining the feature map restored after optical computation. The processing results can be stored in the microprocessor's storage device or directly used in subsequent deep learning network training and inference processes. The calculation results can be displayed using visualization tools to verify the implementation and performance of the convolutional neural network.

[0073] In addition, it can also be extended to the implementation of multi-layer convolution. The specific steps are as follows:

[0074] Repeating the above steps creates a multi-layer optical convolutional network structure, which consists of multiple optical convolutional modules connected in series. Each module includes a tunable laser, a polarization controller, a wavelength division multiplexer, an optical intensity modulator, a delay fiber, a photodetector, and an analog-to-digital converter. The output of each optical convolutional module serves as the input to the next module, enabling deeper convolution operations through the multi-layer structure. Furthermore, the same set of modules can be reused multiple times to achieve multi-layer convolution.

[0075] Example 2:

[0076] The first embodiment above is based on a weighted optical convolutional neural network implementation method using a traditional manually squeezed fiber polarization controller. However, manual angle adjustment is relatively rough. Therefore, the manually squeezed fiber polarization controller can be replaced by an electrically controlled polarization controller to achieve weighting.

[0077] In this embodiment, the electrically controlled polarization controller has a fast response speed and high precision, and can achieve more flexible polarization state control. The electrically controlled polarization controller implements a weighted alternative solution, such as Figure 4 The specific steps are as follows:

[0078] Step 1: Construct a laser module consisting of N tunable lasers. For example, construct four tunable lasers, each using four wavelengths with equal wavelength intervals, corresponding to a 2×2 convolution kernel, as the light source part of the entire system;

[0079] Step 2: The laser output is connected to an electrically controlled polarization controller. A wavelength division multiplexer is used to combine the polarization control signals from different lasers, and the combined optical signal is connected to a modulator.

[0080] Step 3: Before performing the multiplication and addition of the optical convolution, the microprocessor sends an electrical signal to control the stepper motor to change the angle of the polarization controller. The electrically controlled polarization controller is used to adjust the optical signal power of different channels to be consistent and maximized as much as possible. The specific power of each wavelength can be observed by connecting the spectrometer to the modulator output waveform.

[0081] Step four: change the angle of the polarization controller, record the corresponding optical signal power of each angle;

[0082] Step five: through the pre-trained weight value, it is converted into power value, the corresponding angle of the polarization controller is calculated, the electric signal of the microprocessor is used to control the step motor to change the angle of the polarization controller, and the angle is adjusted to the predetermined angle to match the polarization state of each optical signal with the corresponding weight value;

[0083] Step six: the polarization controller can use two groups or more, and the above steps four and five are repeated, for example, the output signal of the first group of polarization controllers is further adjusted through the second group of polarization controllers, and finally enters the wavelength division multiplexer for signal merging. Through this multi-stage adjustment, it can be ensured that the optical signal of each wavelength reaches the best polarization state before merging, and the extinction ratio is improved.

[0084] Step seven: the optical signal adjusted by the electrically controlled polarization controller enters the wavelength division multiplexer for signal merging, and the merged optical signal carries the information of the corresponding convolution kernel weight.

[0085] Step eight: the merged optical signal is modulated by the optical intensity modulator, and the input signal waveform generated by the high-speed digital-to-analog converter is loaded to the modulator, so that the optical signal produces a corresponding modulation effect in the time domain.

[0086] Step nine: the modulated optical signal enters the optical amplifier for amplification to ensure that the signal intensity is sufficient for subsequent processing. The amplified optical signal passes through the delay optical fiber, and a 1-bit delay is generated between each wavelength channel to realize the addition operation.

[0087] Step ten: the optical signal after delay processing enters the photodetector for conversion, and the photodetector converts the optical signal into an electric signal. The collected electric signal is processed by the microprocessor, and a feature map can be obtained by reducing the one-dimensional signal on the microprocessor, so as to perform subsequent neural network training.

[0088] Step eleven: repeating the above steps can realize a multi-layer optical convolution network structure, and the multi-layer optical convolution network realizes a plurality of series-connected optical convolution modules, each module including a tunable laser, a polarization controller, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a delay optical fiber, a photodetector, and an analog-to-digital converter. The output of each optical convolution module is used as the input of the next module, and a deeper convolution operation is realized through the multi-layer structure. In addition, the same group of modules can be used multiple times to play a role in multi-layer convolution.

[0089] In summary, the application aims to provide a polarization-weighted photonic convolutional neural network implementation method, which significantly improves the calculation efficiency of the convolutional neural network by using the high speed and parallel processing capability of the optical system, and is suitable for large-scale data processing and real-time signal processing scenarios. By introducing polarization control and optical coupling technology, the application can effectively implement optical convolution operation and improve the performance ratio of the system, providing a new idea for the implementation of photonic convolutional neural network.

[0090] The above is a further detailed description of the application in combination with specific preferred embodiments, and the specific implementation of the application cannot be limited to these descriptions. For ordinary skilled persons in the technical field of the application, some simple deductions or substitutions can be made without departing from the concept of the application, and all of them should be considered as falling within the protection scope of the application.

Claims

1. A polarization-weighted photonic convolutional neural network implementation method, loading a convolution kernel to a light wave, characterized in that, Comprising the following steps: S1, input light source: S1.1, N tunable lasers output N wavelengths of light signals with equal wavelength interval as system input light source; S1.2, the output of the tunable laser is connected to the corresponding polarization controller, and then the polarization control signals from different tunable lasers are combined through a wavelength division multiplexer, and the combined light signal is connected to an optical intensity modulator; S1.3, before optical convolution, the power of different channel light signals is adjusted to be consistent by using the polarization controller, and the power is taken to be the maximum value as possible, and the output of the optical intensity modulator is connected to an optical spectrum analyzer, and the power of each wavelength of light signal is obtained by observing the waveform; S2, load convolution kernel: S2.1, change the angle of the polarization controller, and record the light signal power corresponding to each angle; S2.2, convert the pre-trained weight value into power value, calculate the angle of the corresponding polarization controller, and control each polarization controller to adjust to the predetermined angle, so that the polarization state of each light signal matches the corresponding weight value; S2.3, after the light signal is adjusted by the polarization controller, it enters the wavelength division multiplexer in turn for signal combination, and in the process of combination, each light signal of different wavelengths carries the information of the corresponding convolution kernel weight; S3, realization of multiplication and addition: S3.1, the combined light signal is modulated by the optical intensity modulator, the input signal waveform generated by the high-speed digital-to-analog converter is loaded onto the optical intensity modulator, so that the light signal produces corresponding modulation effect in time domain, realizing the multiplication of convolution kernel and image signal; S3.2, the modulated light signal is input into an optical amplifier for amplification, the optical intensity modulator loads the bit stream at a certain rate, and the light of several wavelength channels is loaded with the same data after passing through the modulator, and after passing through the delay optical fiber, a time delay difference is generated between them, so that the data information carried by the light of each wavelength channel is staggered by one bit information time in time domain; S3.3, the light signal after delay processing enters a photodetector for conversion, the staggered light signals are received by the photodetector, the values obtained by the multiplication in step S3.1 are added up, and the convolution operation is added up; the photodetector converts the light signal into an electrical signal, which is converted into a digital signal after an analog-to-digital converter, and is processed by a microprocessor to complete nonlinear, pooling and full connection operations, and a complete photonic convolutional neural network is constructed; The photonic convolutional neural network comprises a plurality of light convolutional modules connected in series or a same set of light convolutional modules used multiple times, each light convolutional module comprising a tunable laser, a polarization controller, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a delay optical fiber, a photodetector and an analog-to-digital converter; the output of each light convolutional module is used as the input of the next light convolutional module, and multi-layer convolution is realized through the multi-layer structure; S4, output and processing: The microprocessor processes the collected electrical signal to process the image data into a one-dimensional signal, obtains the feature map restored after optical calculation, and sends it to a high-speed digital-to-analog converter.

2. The polarization-weighted based photonic convolutional neural network implementation method according to claim 1, wherein, The polarization controller changes the direction angle of the elliptical polarized light by adjusting the angle, when the polarization controller is a manual extrusion type polarization controller, the angle is adjusted manually, when the polarization controller is an electrically controlled polarization controller, the angle is changed by the electric signal sent by the microprocessor to control the step motor.

3. The polarization-weighted based photonic convolutional neural network implementation method according to claim 1, wherein, In step S3.4, the microprocessor processes the collected electric signal, expands the one-dimensional data into a two-dimensional image, obtains the feature map after the light on calculation and restoration, stores the result through the storage device of the microprocessor, and / or directly uses the result for deep learning network training and inference process, displays through a visualization tool, and sends to a high-speed digital-to-analog converter.

4. A polarization-weighted based photonic convolutional neural network system for performing the method of any one of claims 1 to 3, characterized in that, It comprises: a laser module, a polarization module, a wavelength division multiplexer, an optical intensity modulator, an optical amplifier, a high-speed digital-to-analog converter, a delay optical fiber, a photodetector, an analog-to-digital converter, a microprocessor, The laser module is composed of a plurality of tunable lasers, serving as the light source part of the system. The polarization module sets a corresponding number of polarization controllers based on the number of tunable lasers, respectively receives the output signal of each tunable laser, and adjusts the polarization state of the optical signal through the corresponding polarization controller. The wavelength division multiplexer connects the polarization controller and sends the combined laser signals of different wavelengths emitted by the tunable lasers to the optical intensity modulator. The high-speed digital-to-analog converter is used to generate the waveform of the input signal, and after generating the waveform according to the one-dimensional signal processed by the microprocessor, it is loaded onto the optical intensity modulator. The optical amplifier is arranged after the optical intensity modulator and is used to amplify the optical signal that needs to be calculated by optical convolution, and is output to the photodetector through the delay optical fiber. The delay optical fiber is used to produce a time delay of one bit of information between the time domain waveforms of each wavelength. The photodetector is used to detect the intensity of the optical signal amplified by the optical amplifier and output the total power of all wavelengths of light in each time interval, which is converted into an electric signal and input to the analog-to-digital converter. The analog-to-digital converter converts the analog signal obtained by the photodetector into a digital signal and outputs it to the microprocessor for processing. The microprocessor processes the output signal, processes the image data into one-dimensional signals, and sends them to the high-speed digital-to-analog converter.

5. The polarization-weighting-based photon convolutional neural network system of claim 4, wherein, In the laser module, the number N of tunable lasers is determined by the size of the convolution kernel, and N tunable lasers emit N wavelengths with equal wavelength intervals, which are respectively input to each polarization controller in the polarization module.

6. The polarization-weighting-based photon convolutional neural network system of claim 5, wherein, The polarization controller is connected to the output of the tunable laser through a polarization maintaining optical fiber, and the polarization controller outputs elliptical polarized light by adjusting the pressure on the polarization maintaining optical fiber, changes the direction angle of the elliptical polarized light by adjusting the angle, and ensures that the optical signal power of different channels entering the modulator is consistent and maximum. The polarization controller is also connected to the wavelength division multiplexer through a polarization maintaining optical fiber to maintain the polarization state of the optical signal.

7. The polarization-weighting-based photon convolutional neural network system of claim 4, wherein, The polarization controller is a manual extrusion type polarization controller or an electrically controlled polarization controller, and when it is an electrically controlled polarization controller, the polarization controller is also connected to the microprocessor, and the angle of the polarization controller is changed by the electric signal sent by the microprocessor to control the step motor.

8. The polarization-weighting-based photon convolutional neural network system of claim 4, wherein, The polarization module has at least one group of polarization controllers, when the polarization controllers are multiple groups, the output signal of the first group of polarization controllers is further adjusted by the subsequent groups of polarization controllers, and then enters the wavelength division multiplexer for signal merging, so as to ensure that the optical signal of each wavelength reaches the optimal polarization state before merging, and the extinction ratio is improved.

9. The polarization-weighting-based photon convolutional neural network system of claim 4, wherein, The wavelength division multiplexer, the optical intensity modulator, the optical amplifier, the delay optical fiber, the photodetector, the analog-to-digital converter, the microprocessor and the high-speed digital-to-analog converter are sequentially connected through a single-mode optical fiber.

Images