A device for VCSEL-SA multi-layer photon pulse neural network classification
By introducing multi-layer structure and supervised learning algorithms into the VCSEL-SA photon pulse neural network, the problem of small scale and lack of hidden layers in the prior art VCSEL-SA photon pulse neural network is solved, and the classification of multiple data sets and the recognition of high accuracy is realized, which improves the applicability and robustness of the system.
Patent Information
- Application Number
- CN202210265757.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-17
- Publication Date
- 2025-05-06
- Estimated Expiration
- 2042-03-17
AI Technical Summary
The existing VCSEL-SA photon pulse neural network is small in scale, difficult to cope with complex scenarios, and lack of hidden layers, resulting in insufficient in nonlinear classification, affecting the key space and security of laser chaotic communication systems.
A system based on VCSEL-SA multi-layer photon pulse neural network and supervised learning algorithm is adopted to generate signals through external optical input samples, inject them into the input layer, and process them through a multi-layer structure, and finally match the sample labels at the output layer to realize the classification of the data set.
It realizes the classification of multiple data sets, has a larger network structure, can handle richer data sets, has a wider range of applications, and maintains a high recognition accuracy when facing abnormal situations of delay jitter in input signals, which has good robustness.
Smart Images
Figure CN114595807B_ABST
Abstract
Description
Technical Field
[0001] The invention discloses a device for VCSEL-SA multi-layer photon pulse neural network classification, belonging to the field of communication technology, and specifically relates to a method for realizing data sample classification by a multi-layer photon pulse neural network. Background Art
[0002] Since light has many excellent properties, such as high speed, high bandwidth, low power consumption, and low crosstalk, it is very suitable for ultrafast information processing. VCSEL-SA photon pulse neural networks are introduced into the communication field to achieve high-speed and efficient data processing. However, ordinary VCSEL-SA photon pulse neural networks are small in scale and difficult to cope with complex scenarios, which reduces the availability of photon pulse neural networks. Therefore, VCSEL-SA multi-layer photon pulse neural networks have become an extremely attractive research hotspot.
[0003] As far as the current research progress is concerned, the VCSEL-SA photon pulse neural networks constructed in most theoretical or experimental studies do not contain hidden layers, and have great deficiencies in nonlinear classification. In addition, the parameter space for a single-type chaotic light source system to achieve delay-free chaotic laser output is limited, which seriously affects the key space of the laser chaotic communication system and threatens the security of the chaotic communication system. Summary of the invention
[0004] In view of the above-stated deficiencies in the prior art, the present invention aims to provide a system combining a VCSEL-SA multi-layer photon pulse neural network with a supervised learning algorithm to achieve classification of multiple data sets.
[0005] The object of the present invention is achieved by the following means.
[0006] The device for VCSEL-SA multi-layer photon pulse neural network classification is characterized in that after the input sample is generated into an input signal by external light, it is injected into the input layer so that the output signal of the output layer correctly matches the sample label. The experimental device includes: a group of laser diodes LD with a working wavelength of 850nm and their temperature and current controllers, a Mach-Zehnder modulation array MZM Array, a pulse pattern generator PPG, three groups of vertical cavity surface emitting lasers VCSEL-SA1, VCSEL-SA2, VCSEL-SA3 with saturated absorbers without built-in optical isolators and their temperature and current controllers, three groups of optical couplers OC1, OC2, OC3, a group of optical splitters, two groups of adjustable optical attenuators VOA1, VOA2, three groups of photodetectors PD1, PD2, PD3, and an electrical controller EC. The processing steps include: LD generates an optical signal, which is modulated by PPG and MZM to obtain an input signal, which is injected into the input layer VCSEL-SA1, and then is divided into two paths after passing through OC1. One path of light is injected into EC after passing through PD1, and the other path of light enters VOA1 after passing through the optical splitter, and is injected into the hidden layer VCSEL-SA2 after passing through OC2, and then enters OC3 after passing through VOA2, and is injected into the output layer VCSEL-SA3, and finally injected into PD3 to obtain the sample classification result.
[0007] After the above design, we only need to set the adjustable optical attenuator according to the electrical controller and inject the samples in the data set into the input layer after preprocessing, so that the output signal of the output layer can be correctly matched with the sample label.
[0008] Compared with the reported VCSEL-SA multi-layer photon pulse neural network, the device for realizing classification of multiple data sets based on the VCSEL-SA multi-layer photon pulse neural network of the present invention has the following advantages: it has a larger network structure, can process richer data sets, and has a wider range of applications; it can still maintain a high recognition accuracy rate in the face of abnormal conditions of input signal delay jitter, and has good robustness.
[0009] The accompanying drawings are as follows:
[0010] Figure 1 It is a system scheme diagram of the device of the present invention;
[0011] Figure 2 This is the inference result graph of the XOR dataset;
[0012] Figure 3 This is the inference result diagram of the WBC dataset;
[0013] Figure 4 Graph showing the impact of input signal delay jitter on the performance of the device of the present invention. DETAILED DESCRIPTION
[0014] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.
[0015] like Figure 1 As shown, the scheme of the present invention comprises a group of laser diodes LD with an operating wavelength of 850nm and temperature and current controllers thereof, a Mach-Zehnder modulation array MZM Array, a pulse pattern generator PPG, three groups of vertical cavity surface emitting lasers VCSEL-SA1, VCSEL-SA2, VCSEL-SA3 with saturated absorbers without built-in optical isolators and temperature and current controllers thereof, three groups of optical couplers OC1, OC2, OC3, a group of optical splitters, two groups of adjustable optical attenuators VOA1, VOA2, three groups of photodetectors PD1, PD2, PD3, and an electrical controller EC. The processing steps include: LD generates an optical signal, which is modulated by PPG and MZM to obtain an input signal, which is injected into the input layer VCSEL-SA1, and then is divided into two paths after passing through OC1. One path of light is injected into EC after passing through PD1, and the other path of light enters VOA1 after passing through the optical splitter, and is injected into the hidden layer VCSEL-SA2 after passing through OC2, and then enters OC3 after passing through VOA2, and is injected into the output layer VCSEL-SA3, and finally injected into PD3 to obtain the sample classification result.
[0016] In this example, the specific implementation steps of the method are:
[0017] Step 1: In Figure 1 In the scheme shown, the operating wavelength of LD is set to 850nm, the operating wavelength of VCSEL-SA1, VCSEL-SA2, and VCSEL-SA3 is set to 850nm, and the samples of the XOR data set are modulated by MZM Array and PPG to obtain the input signal. The signal width of the light injected into VCSEL-SA1 is 2ns, and the signal intensity of the light injected into VCSEL-SA1 is 10μW. Under the above conditions, the timing diagram of the VCSEL-SA3 output pulse signal is shown in Figure 2 As shown. Figure 2 We can judge that the output optical signal of VCSEL-SA3 matches the sample label, which shows that this system can achieve XOR classification. During the experiment, the laser kept the above wavelength unchanged, and the laser output wavelength can be precisely controlled by using current and temperature controllers.
[0018] Step 2: Adjust the attenuation values of VOA1 and VOA2, and study the classification of the WBC data set by this system. First, set the working wavelength of LD to 850nm, the working wavelength of VCSEL-SA1, VCSEL-SA2, and VCSEL-SA3 to 850nm, and obtain the input signal after the sample of the WBC data set is modulated by MZM Array and PPG. The signal width of the light injected into VCSEL-SA1 is 2ns, and the signal intensity of the light injected into VCSEL-SA1 is 10μW. The results are as follows: Figure 3 shown. Figure 3 The value of the vertical axis corresponding to the black dotted line in is 9.3. From the figure, we can see that the system can achieve the classification of WBC with an accuracy rate higher than 96%.
[0019] Step 3: Adjust the input signal injection time and study the output of VCSEL-SA3 under the influence of input signal delay of different strengths. The results are as follows: Figure 4 As shown in the figure, when the input signal delay is between 0 and 200 ps, the classification accuracy of WBC is higher than 90%, which indicates that within the input signal delay range of up to 200 ps, the system has a high input signal delay.
[0020] In summary, the present invention has the following features: 1) It has a larger network structure, can process rich data sets, and has a wide range of applications; 2) It can still maintain a high recognition accuracy in the face of abnormal conditions of input signal delay jitter and has good robustness.
[0021] In summary, the above-described implementation scheme is only an example of the present invention and is not intended to limit the scope of protection of the present invention. It should be pointed out that for ordinary technicians in this technical field, several equivalent deformations and substitutions can be made on the content disclosed in the present invention (such as appropriately changing the size of the working current, changing the frequency detuning), which should also be included in the scope of protection of the present invention.
Claims
1. A device for VCSEL-SA multi-layer photon pulse neural network classification, characterized in that: After the input signal is generated by the input sample using external light, it is injected into the input layer so that the output signal of the output layer can be correctly matched with the sample label. The experimental device includes: a group of laser diodes LD with an operating wavelength of 850nm and their temperature and current controllers, a Mach-Zehnder modulation array MZM Array, a pulse pattern generator PPG, three groups of vertical cavity surface emitting lasers VCSEL-SA1, VCSEL-SA2, VCSEL-SA3 with saturated absorbers without built-in optical isolators and their temperature and current controllers, three groups of optical couplers OC1, OC2, OC3, a group of optical splitters, two groups of adjustable optical attenuators VOA1, VOA2, three groups of photodetectors PD1, PD2, PD3, and an electrical controller EC. The processing steps include: LD generates an optical signal, which is modulated by PPG and MZM to obtain an input signal, which is injected into the input layer VCSEL-SA1, and then is divided into two paths after passing through OC1. One path of light is injected into EC after passing through PD1, and the other path of light enters VOA1 after passing through the optical splitter, and is injected into the hidden layer VCSEL-SA2 after passing through OC2, and then enters OC3 after passing through VOA2, and is injected into the output layer VCSEL-SA3, and finally injected into PD3 to obtain the sample classification result.
2. The device for VCSEL-SA multi-layer photon pulse neural network classification according to claim 1, characterized in that: After the input sample is generated into an input signal using external light, the output signal of the output layer is correctly matched with the sample label.
3. The device for VCSEL-SA multi-layer photon pulse neural network classification according to claim 1, characterized in that: For different data sets, the connection strength between VOA neurons can be controlled, and a high recognition accuracy can still be maintained in the face of abnormal conditions of input signal delay jitter, showing good robustness.