A multi-dimensional pulse reservoir pool computing method and system based on a semiconductor laser
By extracting multidimensional features from the peak pulse light signal of a semiconductor laser, an extended-dimensional reservoir state vector is constructed, resolving the contradiction between data processing accuracy, speed, cost, and power consumption in existing technologies, and achieving efficient and low-cost reservoir computation.
Patent Information
- Application Number
- CN202610243708.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-02
- Publication Date
- 2026-05-19
- Estimated Expiration
- 2046-03-02
AI Technical Summary
Existing reservoir calculation methods based on semiconductor lasers improve data processing accuracy but reduce data processing speed and increase power consumption, making it difficult to meet the needs of applications with high real-time requirements.
By performing photoelectric conversion on the spike pulse optical signal generated by the semiconductor laser under the excitation of the original optical signal, the amplitude feature, quantity feature and time position feature are extracted, and an extended dimension reservoir state vector is constructed. These feature virtual nodes are used to perform multidimensional pulse reservoir calculation, avoiding the increase of the number of semiconductor lasers and complex feedback structure.
It significantly improves the data processing accuracy and speed of the reservoir computing system, while maintaining low cost and low power consumption, enriches the reservoir state space, and enhances the ability to process complex signals.
Smart Images
Figure CN121765354B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of reservoir calculation technology, and in particular to a multidimensional pulse reservoir calculation method and system based on semiconductor lasers. Background Technology
[0002] In the context of rapid development in information technology, artificial intelligence (AI) is rapidly permeating all areas of social production and scientific research, demonstrating immense value, particularly in data processing, pattern recognition, and intelligent decision-making. However, as application scenarios become increasingly complex and data scale and task dimensions continue to grow, traditional computing systems often face limitations such as modeling difficulties, high computational costs, and insufficient real-time performance when dealing with high-dimensional, strongly nonlinear, and strongly time-series correlated problems. Therefore, researchers are constantly exploring new computing architectures to improve the efficiency and accuracy of complex information processing tasks. Neural networks, as information processing models that mimic biological nervous systems, have achieved significant results in tasks such as image recognition, speech processing, and time series prediction due to their powerful nonlinear expressive capabilities. However, deep learning neural networks typically require large amounts of labeled data and long-term iterative training, relying on high-performance computing resources. This not only incurs huge time and energy costs but also limits their further application in resource-constrained or high-real-time-requirement scenarios.
[0003] To address the aforementioned issues, Reservoir Computing (RC) has emerged as a new computing framework that is gaining attention. This framework typically utilizes physical systems with nonlinear dynamics to construct reservoirs. By keeping the internal connections and input mappings of the reservoirs fixed, the learning task can be completed by performing linear training only on the output layer. This avoids the complex backpropagation training process in traditional deep learning and offers advantages such as simple training, high efficiency, and suitability for hardware acceleration. In recent years, photonic computing has become an important direction for the realization of RC due to its advantages such as ultra-high bandwidth, low latency and high energy efficiency. Among them, semiconductor lasers are considered to be one of the ideal devices for building photonic reservoirs due to their small size, low power consumption, fast response speed and easy integration. Taking a reservoir system based on semiconductor lasers as an example, when classifying images, it is only necessary to encode the image to be classified into an input light signal and input it into the reservoir. After receiving the input light signal, the semiconductor laser will generate a response signal due to its nonlinear optical characteristics and optical feedback. These response signals contain the spatial and grayscale feature information of the image to be classified. The output module can obtain the output weight of the reservoir by training the output signal of the reservoir. Finally, the predicted category of the output image to be classified is obtained by using the output weight.
[0004] In semiconductor laser-based reservoir computing systems, the performance of the reservoir is closely related to the number of virtual nodes. A larger number of nodes allows the system to capture richer input signal features and node response states, thereby improving the system's ability to reconstruct states and its prediction accuracy. For example, in time series prediction, more nodes mean the ability to record historical information over a longer time range, thus more accurately capturing the dynamic changes in input data and improving prediction accuracy. Based on this theory, existing technologies expand node scale by increasing the number of physical devices, extending multiplexing links, or introducing delayed feedback structures. However, these methods significantly increase system cost, hardware size, and power consumption, and may also reduce system real-time performance, limiting their application in high-real-time-requirement scenarios such as autonomous driving and medical monitoring.
[0005] In summary, existing reservoir computing methods based on semiconductor lasers improve data processing accuracy by increasing the number of nodes in the reservoir computing system by increasing the number of components. However, while improving data processing accuracy, they not only reduce data processing speed but also increase data processing cost and power consumption. Summary of the Invention
[0006] Therefore, the technical problem to be solved by the present invention is to overcome the problem that the existing semiconductor laser-based reservoir calculation method cannot simultaneously take into account data processing accuracy, data processing speed, cost and operating power consumption.
[0007] To address the aforementioned technical problems, this invention provides a method for calculating a multidimensional pulse reservoir based on a semiconductor laser. The method is applied to a multidimensional pulse reservoir calculation system based on a semiconductor laser, and includes:
[0008] The signal to be processed is encoded to generate the original optical signal;
[0009] The original optical signal is injected into the semiconductor laser, and the spike pulse optical signal containing the characteristic information of the signal to be processed is collected by the semiconductor laser under the excitation of the original optical signal.
[0010] The peak pulse optical signal is photoelectrically converted to obtain a time-domain signal; the amplitude, quantity, and temporal position features of the peak pulses in the time-domain signal are extracted, and virtual nodes for amplitude, quantity, and temporal position features are constructed.
[0011] The virtual nodes for amplitude features, quantity features, and time and location features are concatenated to obtain an extended-dimensional reservoir state vector; the output signal of the multidimensional pulse reservoir computing system is obtained based on the extended-dimensional reservoir state vector.
[0012] Preferably, the amplitude features, quantity features, and temporal position features of the spike pulses in the time-domain signal are extracted, and virtual nodes for amplitude features, quantity features, and temporal position features are constructed, including:
[0013] Divide the time-domain signal into N time windows;
[0014] The amplitude of the first peak in each time window is taken as the amplitude feature of that time window, the number of peaks in each time window is taken as the quantity feature of that time window, and the time offset of the first peak in each time window is taken as the time position feature of that time window.
[0015] Amplitude feature virtual nodes are obtained by combining amplitude features within N time windows, quantity feature virtual nodes are obtained by combining quantity features within N time windows, and time position feature virtual nodes are obtained by combining time position features within N time windows.
[0016] Preferably, after acquiring the spiked optical signal containing the characteristic information of the signal to be processed generated by the semiconductor laser under the excitation of the original optical signal, the process further includes:
[0017] The spiked optical signal is coupled to the original optical signal, and the coupled optical signal is injected into a semiconductor laser. The spiked optical signal generated by the semiconductor laser under the excitation of the coupled optical signal contains the depth characteristics of the signal to be processed.
[0018] Preferably, after acquiring the spike pulse optical signal, the process further includes:
[0019] The spike pulse optical signal is divided into a first spike pulse optical signal and a second spike pulse optical signal;
[0020] The first peak pulse optical signal is photoelectrically converted to obtain a time-domain signal, thereby extracting the features of the peak pulse to construct an extended-dimensional reservoir state vector;
[0021] The spectrum of the second peak pulse signal is observed to characterize the operating state and dynamic characteristics of the semiconductor laser.
[0022] The present invention also provides a multidimensional pulse reservoir calculation system based on a semiconductor laser, comprising:
[0023] The input module is used to encode the signal to be processed to generate the original optical signal;
[0024] The pulse reservoir module specifically includes:
[0025] A semiconductor laser, connected to an input module, is used to generate a spiked optical signal containing characteristic information of the signal to be processed under the excitation of the original optical signal.
[0026] The output module, connected to a semiconductor laser, performs photoelectric conversion on the spike pulse optical signal to obtain a time-domain signal. It extracts the amplitude, quantity, and temporal position features of the spike pulses from the time-domain signal and constructs virtual nodes for amplitude, quantity, and temporal position features. These virtual nodes are then concatenated to obtain an extended-dimensional reservoir state vector. The output signal of the multi-dimensional pulse reservoir computing system is obtained based on this extended-dimensional reservoir state vector.
[0027] Preferably, the input module includes:
[0028] Drive the laser to generate laser signals;
[0029] A waveform generator is used to generate a mask signal based on the parameters of the signal to be processed.
[0030] Radio frequency amplifier, connected to waveform generator, is used to amplify mask signal;
[0031] The modulator, connected to the driving laser and the radio frequency amplifier, is used to modulate the laser signal based on the amplified mask signal to generate the original optical signal.
[0032] Preferably, the input module further includes:
[0033] A variable optical attenuator, connected to the modulator, is used to adjust the injection intensity of the original optical signal;
[0034] A polarization controller, connected to a modulator, is used to adjust the polarization state of the original optical signal.
[0035] Preferably, the pulse reservoir module further includes:
[0036] The first optical coupling module, connected to the input module and the semiconductor laser, is used to couple the original optical signal and the spiked optical signal, and inject the coupled optical signal into the semiconductor laser, so that the semiconductor laser generates a spiked optical signal containing the depth characteristics of the signal to be processed under the excitation of the coupled optical signal.
[0037] Preferably, the output module includes:
[0038] An optical signal acquisition unit, connected to a semiconductor laser, is used to perform photoelectric conversion on peak pulse optical signals to obtain time-domain signals;
[0039] An oscilloscope, connected to an optical signal acquisition unit, is used to extract the amplitude, quantity, and temporal location characteristics of spike pulses in a time-domain signal.
[0040] A digital signal processor, connected to an oscilloscope, is used to construct virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of spike pulses;
[0041] The adder, connected to the digital signal processor, is used to concatenate the virtual nodes of amplitude feature, quantity feature, and time position feature of the spike pulse to obtain the extended dimension reservoir state vector.
[0042] A programmable gated array, connected to an adder, is used to train the output of the pulse reservoir module to obtain the output weights of the virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse, thereby obtaining the output signal of the multidimensional pulse reservoir computing system.
[0043] Preferably, the output module further includes:
[0044] The second optical coupling module has its input end connected to the output end of the semiconductor laser and its output end connected to the optical signal collector and the spectrometer. It is used to divide the spike pulse optical signal into a first spike pulse optical signal and a second spike pulse optical signal, and transmit the first spike pulse optical signal to the optical signal collector and the second spike pulse optical signal to the spectrometer.
[0045] A spectrometer is used to perform spectral analysis on the second peak pulse light signal, thereby characterizing the operating state and dynamic characteristics of the semiconductor laser.
[0046] The multidimensional pulse reservoir calculation method based on semiconductor lasers provided in this application has the following advantages:
[0047] 1. After acquiring the spiked optical signal containing the characteristic information of the signal to be processed, the signal was not directly used to train the weights of the output module. Instead, it was first converted into a time-domain signal, and the amplitude, quantity, and temporal position characteristics of the spikes in the time-domain signal were acquired. The amplitude characteristics were used to reflect the intensity of the internal transient nonlinear response of the semiconductor laser under the excitation of the original optical signal, thus reflecting the intensity information of the original optical signal. The quantity characteristics were used to reflect the activity of spike emission of the semiconductor laser under the excitation of the original optical signal, characterizing the overall response intensity of the semiconductor laser under excitation, thus reflecting the temporal excitation information of the original optical signal. The temporal position characteristics were used to reflect the dynamic response time structure and delay characteristics of the semiconductor laser under the excitation of the original optical signal, thus reflecting the time-coded information of the original optical signal. The relationship between information and the internal dynamics of semiconductor lasers is investigated. Finally, the amplitude, quantity, and temporal position features containing different characteristic information of the original optical signal are mapped into virtual nodes for amplitude, quantity, and temporal position, respectively, and then spliced together to form the reservoir state vector. Without increasing the number of semiconductor lasers, introducing complex feedback structures, or significantly extending the time multiplexing link, by constructing three times the number of virtual nodes, the multidimensional feature information related to the signal to be processed contained in the spike pulse optical signal is fully explored. This not only significantly improves the dimensionality and expressive power of the reservoir state and enhances the data processing accuracy of the reservoir computing system, but also ensures high data processing speed, low data processing cost, and low operating power consumption without increasing the number of components in the reservoir computing system.
[0048] 2. The spiked light signal output by the semiconductor laser is coupled with the original light signal and then injected back into the semiconductor laser. Deeper features are extracted from the coupled light signal to achieve multi-layer feature extraction of the signal to be processed. Through the coupling and re-excitation design, the deep features of the signal to be processed can be extracted during the reservoir calculation process, enriching the reservoir state space and further expanding the number of virtual nodes, thereby improving the processing accuracy of complex signals to be processed. Attached Figure Description
[0049] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein:
[0050] Figure 1 Flowchart of the multidimensional pulse reservoir calculation method based on semiconductor laser provided in this application;
[0051] Figure 2 A schematic diagram of the multidimensional pulse reservoir computing system based on a semiconductor laser provided in this application;
[0052] Figure 3A schematic diagram illustrating the dynamic response characteristics of the semiconductor laser provided in this application under external optical signal injection; wherein... Figure 3 (a) in the diagram is a schematic diagram of the external optical signal pulse sequence. Figure 3 (b) in the figure is the carrier density response curve of the active region inside the semiconductor laser. Figure 3 (c) in the figure represents the dynamic change curve of the number of laser output photons. Figure 3 (d) in the figure represents the carrier density variation curve in the passive region;
[0053] Figure 4 This is a schematic diagram illustrating the classification results when using the existing reservoir computing system and the multidimensional pulse reservoir computing system of this application to perform an iris classification task; wherein, Figure 4 (a) in the diagram is a schematic of the confusion matrix when using the amplitude features of spike pulses for classification. Figure 4 (b) in the figure is a schematic diagram of the confusion matrix when using the amplitude features, quantity features and temporal location features of spike pulses for classification;
[0054] Explanation of reference numerals in the accompanying drawings: 1. Input module; 11. Driver laser; 12. Waveform generator; 13. RF amplifier; 14. Modulator; 15. Variable optical attenuator; 16. Polarization controller; 2. Pulse reservoir module; 21. Semiconductor laser; 22. First optical coupling module; 3. Output module; 31. Optical signal acquisition unit; 311. Optical isolator; 312. Photodetector; 32. Oscilloscope; 33. Digital signal processor; 34. Adder; 35. Programmable gate array; 36. Second optical coupling module; 37. Spectrum analyzer. Detailed Implementation
[0055] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.
[0056] Please see Figure 1 , Figure 1 The diagram shows the flowchart of the multidimensional pulse reservoir calculation method based on semiconductor lasers provided in this application. The method specifically includes steps S10 to S40:
[0057] S10: Encode the signal to be processed to generate the original optical signal.
[0058] Specifically, a mask signal is generated based on the signal to be processed, and the original optical signal is generated by encoding the laser signal using the mask signal. For example, the signal to be processed can be a time series signal to be predicted, an image or text to be classified, etc.
[0059] S20: Inject the original optical signal into the semiconductor laser and collect the spike pulse optical signal containing the characteristic information of the signal to be processed generated by the semiconductor laser under the excitation of the original optical signal.
[0060] S30: Perform photoelectric conversion on the spike pulse optical signal to obtain a time-domain signal; extract the amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse in the time-domain signal, and construct virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics.
[0061] S40: Concatenate the virtual nodes for amplitude features, quantity features, and time and position features to obtain an extended-dimensional reservoir state vector; based on the extended-dimensional reservoir state vector, obtain the output signal of the multi-dimensional pulse reservoir computing system.
[0062] Since the original optical signals generated after encoding different signals differ in intensity, time modulation form, and spectral characteristics, when the original optical signal is injected into a semiconductor laser, it will excite different nonlinear dynamic responses inside the semiconductor laser, thus outputting optical response signals with significant differences in time structure and amplitude characteristics. Therefore, in the prior art, the response optical signal of the semiconductor laser is usually directly collected to characterize the processing state of the reservoir computing system on the original optical signal. That is, virtual nodes are constructed based on the peak characteristics of the response optical signal to reflect the response characteristics of the semiconductor laser under different original optical signal excitations. Then, by training the weights of the output module of the reservoir computing system, the final output signal of the reservoir computing system after processing the original optical signal is obtained.
[0063] This application further discovers that for semiconductor lasers operating in an excitable or self-pulsating state, the response generated by the disturbance of the original optical signal is not a continuously changing light intensity signal, but rather manifested as a pulse sequence. Therefore, the spike pulse optical signal output by the semiconductor laser not only contains spike pulse amplitude information reflecting the nonlinear physical processes such as gain recovery, saturation absorption, and carrier-photon coupling within the semiconductor laser, but also contains spike pulse quantity information reflecting the excitation activity level of the semiconductor laser, as well as spike appearance time position information reflecting the dynamic response delay and time coupling relationship. These information changes with the input optical signal. Based on this, after acquiring the spike pulse optical signal containing the characteristic information of the signal to be processed, this application does not directly use this signal to train the weights of the output module. Instead, it first converts it into a time-domain signal, acquires the amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulses in the time-domain signal, and uses the amplitude characteristics to reflect the internal dynamic response delay and time coupling relationship of the semiconductor laser under the excitation of the original optical signal. The transient nonlinear response intensity is used to reflect the intensity information of the original optical signal. Quantitative features are used to reflect the peak emission activity of the semiconductor laser under the excitation of the original optical signal, characterizing the overall response intensity of the semiconductor laser under excitation, thus reflecting the temporal excitation information of the original optical signal. Temporal position features are used to reflect the dynamic response time structure and delay characteristics of the semiconductor laser under the excitation of the original optical signal, thus reflecting the temporal coding information of the original optical signal and the temporal relationship of the internal dynamics of the semiconductor laser. Finally, the amplitude features, quantitative features, and temporal position features containing different characteristic information of the original optical signal are mapped into amplitude feature virtual nodes, quantitative feature virtual nodes, and temporal position feature virtual nodes, and spliced to form the reservoir state vector. Compared with the prior art, which directly calculates the reservoir based on the amplitude features of the optical signal output by the semiconductor laser, this application constructs three times the number of virtual nodes, and the constructed virtual nodes contain richer characteristic information of the original optical signal, thereby improving the accuracy of reservoir calculation.
[0064] Furthermore, without increasing the number of semiconductor lasers, introducing complex feedback structures, or significantly extending the time multiplexing link, this application significantly improves the dimensionality and expressive power of the reservoir state by fully exploiting the multidimensional feature information related to the signal to be processed contained in the spike pulse optical signal. This enhances the nonlinear mapping capability and short-time memory characteristics of the reservoir computing system, thereby improving the data processing accuracy of the reservoir computing system.
[0065] Furthermore, the amplitude, quantity, and temporal location features of the spike pulses in the time-domain signal are extracted, and virtual nodes for amplitude, quantity, and temporal location features are constructed, including steps 1-1 to 1-3:
[0066] Step 1-1: Divide the time-domain signal into N time windows.
[0067] Step 1-2: Take the amplitude of the first peak in each time window as the amplitude feature of that time window, take the number of peaks in each time window as the quantity feature of that time window, and take the time offset of the first peak in each time window as the time position feature of that time window.
[0068] Steps 1-3: Combine the amplitude features within N time windows to obtain virtual nodes for amplitude features, combine the quantity features within N time windows to obtain virtual nodes for quantity features, and combine the time position features within N time windows to obtain virtual nodes for time position features.
[0069] Furthermore, after acquiring the spike pulse optical signal containing the characteristic information of the signal to be processed generated by the semiconductor laser under the excitation of the original optical signal, the process further includes: coupling the spike pulse optical signal with the original optical signal, and injecting the coupled optical signal into the semiconductor laser, thereby acquiring the spike pulse optical signal containing the depth characteristics of the signal to be processed generated by the semiconductor laser under the excitation of the coupled optical signal.
[0070] Specifically, the spiked optical signal output by the semiconductor laser under the excitation of the original optical signal is the primary nonlinear response of the semiconductor laser to the original optical signal. It only carries the surface and basic features of the signal to be processed. After coupling it with the original optical signal and re-injecting it into the semiconductor laser, it is equivalent to mining deeper features from the coupled optical signal, realizing multi-layer feature extraction of the signal to be processed. Through this coupling and re-excitation design, the reservoir computing system can extract deep features of the signal to be processed, enrich the reservoir state space, and further expand the number of virtual nodes, thereby improving the processing accuracy of the reservoir computing system for complex signals to be processed.
[0071] Furthermore, after acquiring the spike pulse optical signal, steps 2-1 to 2-3 are also included:
[0072] Step 2-1: Divide the spike pulse light signal into a first spike pulse light signal and a second spike pulse light signal.
[0073] Step 2-2: Perform photoelectric conversion on the first peak pulse optical signal to obtain a time-domain signal, thereby extracting the features of the peak pulse to construct an extended-dimensional reservoir state vector.
[0074] Steps 2-3: Observe the spectrum of the second peak pulse signal to characterize the working state and dynamic characteristics of the semiconductor laser.
[0075] By characterizing the operating state and dynamic characteristics of semiconductor lasers, it is possible to confirm whether the semiconductor laser is in an excitable or self-pulsating state, and to evaluate the operating performance of the reservoir computing system.
[0076] This application also provides a multidimensional reservoir computing system based on a semiconductor laser, such as... Figure 2 As shown, it specifically includes an input module 1, a pulse storage pool module 2, and an output module 3.
[0077] Input module 1 is used to encode the signal to be processed to generate the original optical signal.
[0078] The pulse reservoir module 2 includes a semiconductor laser 21.
[0079] The semiconductor laser 21 is connected to the input module 1 and is used to generate a spike pulse optical signal containing the characteristic information of the signal to be processed under the excitation of the original optical signal.
[0080] Output module 3 is connected to semiconductor laser 21 and is used to perform photoelectric conversion on the spike pulse optical signal to obtain a time-domain signal; extract the amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse in the time-domain signal, and construct virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse; splice the virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse to obtain an extended-dimensional reservoir state vector; and obtain the output signal of the multi-dimensional pulse reservoir calculation system based on the extended-dimensional reservoir state vector.
[0081] Furthermore, the input module 1 includes a driving laser 11, a waveform generator 12, an RF amplifier 13, and a modulator 14.
[0082] The laser 11 is used to generate laser signals.
[0083] Waveform generator 12 is used to generate a mask signal based on the parameters of the signal to be processed.
[0084] The radio frequency amplifier 13 is connected to the waveform generator 12 and is used to amplify the mask signal.
[0085] The modulator 14 is connected to the driver laser 11 and the radio frequency amplifier 13, and is used to modulate the laser signal based on the amplified mask signal to generate the original optical signal.
[0086] Optionally, the modulator 14 is a phase modulator or an intensity modulator, which realizes time-division mapping encoding of the signal to be processed by loading the mask signal onto the laser signal.
[0087] Furthermore, the input module 1 also includes a variable optical attenuator 15 and a polarization controller 16.
[0088] The variable optical attenuator 15 is connected to the modulator 14 and is used to adjust the injection intensity of the original optical signal.
[0089] The polarization controller 16 is connected to the modulator 14 and is used to adjust the polarization state of the original optical signal.
[0090] Specifically, by adjusting the injection intensity and polarization state of the original optical signal, precise control of the operating state of the semiconductor laser 21 can be achieved.
[0091] Furthermore, the pulse reservoir module 2 also includes a first optical coupling module 22.
[0092] The first optical coupling module 22 is connected to the input module 1 and the semiconductor laser 21. It is used to couple the original optical signal and the spike pulse optical signal, and inject the coupled optical signal into the semiconductor laser 21, so that the semiconductor laser 21 generates a spike pulse optical signal containing the depth characteristics of the signal to be processed under the excitation of the coupled optical signal.
[0093] Furthermore, the output module 3 includes an optical signal acquisition unit 31, an oscilloscope 32, a digital signal processor 33, an adder 34, and a programmable gate array 35.
[0094] The optical signal collector 31 is connected to the semiconductor laser 21 and is used to perform photoelectric conversion on the spike pulse optical signal to obtain a time domain signal.
[0095] Specifically, the optical signal collector 31 includes an optical isolator 311 and a photodetector 312.
[0096] Optical isolator 311 is connected to semiconductor laser 21 for transmitting spike pulse optical signals; photodetector 312 is connected to optical isolator 311 for photoelectric conversion of spike pulse optical signals.
[0097] The oscilloscope 32 is connected to the optical signal acquisition unit 31 and is used to extract the amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulses in the time domain signal.
[0098] The digital signal processor 33 is connected to the oscilloscope 32 and is used to construct virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of spike pulses.
[0099] Adder 34 is connected to digital signal processor 33 and is used to concatenate virtual nodes of amplitude feature, quantity feature, and time position feature of peak pulses to obtain an extended dimension reservoir state vector.
[0100] The programmable gate array 35 is connected to the adder 34 and is used to train the output of the pulse reservoir module to obtain the output weights of the virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse, thereby obtaining the output signal of the multidimensional pulse reservoir computing system.
[0101] Furthermore, the output module 3 also includes a second optical coupling module 36 and a spectrometer 37.
[0102] The input terminal of the second optical coupling module 36 is connected to the output terminal of the semiconductor laser 21, and the output terminal is connected to the optical signal collector 31 and the spectrum analyzer 37. It is used to divide the peak pulse optical signal into a first peak pulse optical signal and a second peak pulse optical signal, and transmit the first peak pulse optical signal to the optical signal collector 31 and the second peak pulse optical signal to the spectrum analyzer 37.
[0103] The spectrometer 37 is used to perform spectral analysis on the second peak pulse light signal, thereby characterizing the operating state and dynamic characteristics of the semiconductor laser 21.
[0104] This application embodiment also applies the multidimensional pulse reservoir calculation system based on semiconductor lasers provided in the above embodiments to the iris flower classification task. Specifically, the multidimensional pulse reservoir calculation system based on semiconductor lasers was first simulated using numerical simulation, and the resulting rate equation is as follows:
[0105] ,
[0106] ,
[0107] ,
[0108] in, This represents the carrier density in the active region; This represents the carrier density in the saturable absorption region; Indicates the number of photons in a laser beam; Indicates the number of externally injected photons; Indicates the self-emission coefficient; This represents the bias current in the active region; Indicates the volume of the active region cavity; This indicates the volume of the saturable absorption region cavity; Indicates the number of charges; Indicates carrier lifetime; Indicates photon lifetime; Indicates the optical limiting factor in the active region; Indicates the light limiting factor in the saturable absorption region; This represents the gain coefficient of the active region; This represents the gain coefficient in the saturable absorption region; This represents the transparent carrier density in the active region; This represents the transparent carrier density in the saturable absorption region.
[0109] like Figure 3 The diagram shown illustrates the dynamic response characteristics of the semiconductor laser provided in this embodiment when an external optical signal is injected; wherein, Figure 3 (a) in the diagram is a schematic diagram of the external optical signal pulse sequence, showing eight excitation pulses with equal time intervals and increasing amplitude. Figure 3 (b) in the figure is the carrier density response curve of the active region in the semiconductor laser, which reflects the rapid fluctuation of carrier concentration over time caused by pulse injection. Figure 3 (c) in the figure shows the dynamic change curve of the number of laser output photons, presenting three characteristics of the output pulse in terms of amplitude, number, and time position. Figure 3 (d) in the figure represents the change curve of carrier density in the passive region, which helps to illustrate the overall response process of the internal carriers of the semiconductor laser to the external optical signal pulse sequence.
[0110] like Figure 4 The diagram shows the classification results of iris flower classification tasks using existing reservoir computing systems and the multidimensional pulse reservoir computing system of this application; where, Figure 4 (a) in the diagram is a schematic of the confusion matrix when using the amplitude features of spike pulses for classification. Figure 4 (b) in the figure is a schematic diagram of the confusion matrix when using the amplitude, quantity and temporal location features of the spike pulse for classification.
[0111] Specifically, the iris dataset in this embodiment contains 150 samples from three species (Iris tectorum, Iris variegata, and Iris virginiana), with 50 samples for each species. The raw light signals are obtained by encoding each image sample in the iris dataset, and then input into the semiconductor laser in the multidimensional reservoir module. Finally, the classification results are obtained based on the output module.
[0112] from Figure 4As can be seen, when using a single peak amplitude feature (directly using the peak pulse light signal output by the semiconductor laser to train the weights of the output layer) for classification tasks, as in existing technologies, the classification accuracy is 88.33%. However, when this application combines three features (peak amplitude, peak quantity, and peak time position) for classification tasks, the classification accuracy is improved to 96.67%. This comparative result shows that the multidimensional pulse reservoir computing system based on semiconductor lasers provided in this application, compared with traditional reservoir computing systems, breaks through the bottleneck of traditional systems in expanding the number of virtual nodes by extracting and expanding the multidimensional features in the peak pulse light signal, improving data processing accuracy and speed, while maintaining high computing performance and low latency.
[0113] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0114] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0117] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A method for calculating a multidimensional pulse reservoir based on a semiconductor laser, characterized in that, The method is applied to a multidimensional pulse reservoir computing system based on semiconductor lasers, including: The signal to be processed is encoded to generate the original optical signal; The original optical signal is injected into the semiconductor laser, and the spike pulse optical signal containing the characteristic information of the signal to be processed is collected by the semiconductor laser under the excitation of the original optical signal. The peak pulse optical signal is photoelectrically converted to obtain a time-domain signal; the amplitude, quantity, and temporal position features of the peak pulses in the time-domain signal are extracted, and virtual nodes for amplitude feature, quantity feature, and temporal position feature are constructed, specifically including: Divide the time-domain signal into N time windows; The amplitude of the first peak in each time window is taken as the amplitude feature of that time window, the number of peaks in each time window is taken as the quantity feature of that time window, and the time offset of the first peak in each time window is taken as the time position feature of that time window. Amplitude feature virtual nodes are obtained by combining amplitude features within N time windows, quantity feature virtual nodes are obtained by combining quantity features within N time windows, and time position feature virtual nodes are obtained by combining time position features within N time windows. The virtual nodes for amplitude features, quantity features, and time and location features are concatenated to obtain an extended-dimensional reservoir state vector; the output signal of the multidimensional pulse reservoir computing system is obtained based on the extended-dimensional reservoir state vector.
2. The method for calculating a multidimensional pulse reservoir based on a semiconductor laser according to claim 1, characterized in that, After acquiring the spiked optical signal containing the characteristic information of the signal to be processed generated by the semiconductor laser under the excitation of the original optical signal, the process also includes: The spiked optical signal is coupled to the original optical signal, and the coupled optical signal is injected into a semiconductor laser. The spiked optical signal generated by the semiconductor laser under the excitation of the coupled optical signal contains the depth characteristics of the signal to be processed.
3. The method for calculating a multidimensional pulse reservoir based on a semiconductor laser according to claim 1 or 2, characterized in that, After acquiring the spike pulse light signal, the following is also included: The spike pulse optical signal is divided into a first spike pulse optical signal and a second spike pulse optical signal; The first peak pulse optical signal is photoelectrically converted to obtain a time-domain signal, thereby extracting the features of the peak pulse to construct an extended-dimensional reservoir state vector; The spectrum of the second peak pulse signal is observed to characterize the operating state and dynamic characteristics of the semiconductor laser.
4. A multidimensional pulse reservoir calculation system based on a semiconductor laser, characterized in that, include: The input module is used to encode the signal to be processed to generate the original optical signal; The pulse reservoir module specifically includes: A semiconductor laser, connected to an input module, is used to generate a spiked optical signal containing characteristic information of the signal to be processed under the excitation of the original optical signal. The output module, connected to a semiconductor laser, performs photoelectric conversion on the spike pulse optical signal to obtain a time-domain signal. It extracts the amplitude, quantity, and temporal position features of the spike pulses from the time-domain signal and constructs virtual nodes for amplitude, quantity, and temporal position features. These virtual nodes are then concatenated to obtain an extended-dimensional reservoir state vector. The output signal of the multi-dimensional pulse reservoir computing system is obtained based on this extended-dimensional reservoir state vector. Specifically, the amplitude, quantity, and temporal location features of spike pulses in the time-domain signal are extracted, and virtual nodes for amplitude, quantity, and temporal location features are constructed. Divide the time-domain signal into N time windows; The amplitude of the first peak in each time window is taken as the amplitude feature of that time window, the number of peaks in each time window is taken as the quantity feature of that time window, and the time offset of the first peak in each time window is taken as the time position feature of that time window. Amplitude feature virtual nodes are obtained by combining amplitude features within N time windows, quantity feature virtual nodes are obtained by combining quantity features within N time windows, and time position feature virtual nodes are obtained by combining time position features within N time windows.
5. The multidimensional pulse reservoir calculation system based on a semiconductor laser according to claim 4, characterized in that, The input module includes: Drive the laser to generate laser signals; A waveform generator is used to generate a mask signal based on the parameters of the signal to be processed. Radio frequency amplifier, connected to waveform generator, is used to amplify mask signal; The modulator, connected to the driving laser and the radio frequency amplifier, is used to modulate the laser signal based on the amplified mask signal to generate the original optical signal.
6. The multidimensional pulse reservoir calculation system based on a semiconductor laser according to claim 5, characterized in that, The input module also includes: A variable optical attenuator, connected to the modulator, is used to adjust the injection intensity of the original optical signal; A polarization controller, connected to a modulator, is used to adjust the polarization state of the original optical signal.
7. The multidimensional pulse reservoir calculation system based on a semiconductor laser according to claim 4, characterized in that, The pulse reservoir module also includes: The first optical coupling module, connected to the input module and the semiconductor laser, is used to couple the original optical signal and the spiked optical signal, and inject the coupled optical signal into the semiconductor laser, so that the semiconductor laser generates a spiked optical signal containing the depth characteristics of the signal to be processed under the excitation of the coupled optical signal.
8. The multidimensional pulse reservoir calculation system based on a semiconductor laser according to claim 7, characterized in that, The output module includes: An optical signal acquisition unit, connected to a semiconductor laser, is used to perform photoelectric conversion on spike pulse optical signals to obtain time-domain signals; An oscilloscope, connected to an optical signal acquisition unit, is used to extract the amplitude, quantity, and temporal location characteristics of spike pulses in a time-domain signal. A digital signal processor, connected to an oscilloscope, is used to construct virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of spike pulses; The adder, connected to the digital signal processor, is used to concatenate the virtual nodes of amplitude feature, quantity feature, and time position feature of the spike pulse to obtain the extended dimension reservoir state vector. A programmable gated array, connected to an adder, is used to train the output of the pulse reservoir module to obtain the output weights of the virtual nodes for amplitude characteristics, quantity characteristics, and time position characteristics of the spike pulse, thereby obtaining the output signal of the multidimensional pulse reservoir computing system.
9. The multidimensional pulse reservoir calculation system based on a semiconductor laser according to claim 8, characterized in that, The output module also includes: The second optical coupling module has its input end connected to the output end of the semiconductor laser and its output end connected to the optical signal collector and the spectrometer. It is used to divide the spike pulse optical signal into a first spike pulse optical signal and a second spike pulse optical signal, and transmit the first spike pulse optical signal to the optical signal collector and the second spike pulse optical signal to the spectrometer. A spectrometer is used to perform spectral analysis on the second peak pulse light signal, thereby characterizing the operating state and dynamic characteristics of the semiconductor laser.