Photon reserve pool calculation system and method based on hierarchical pulses

Abstract

The invention belongs to the technical field of reservoir calculation, and relates to a photon reservoir calculation system and method based on hierarchical pulses, and the system comprises an input layer which is used for generating an input electric signal; the reserve pool layer comprises a multimode semiconductor laser which is connected with the input layer and is used for generating continuous and regular optical pulse signals under the excitation of the input electric signals and the incoherent optical feedback signals; the beam splitter prism is used for dividing the optical pulse signal into two paths, one path is used as a multiplexing optical signal to be transmitted to a feedback loop, and the other path is used as an output optical signal to be transmitted to an output layer; the feedback loop is used for generating an incoherent optical feedback signal based on the multiplexing optical signal and transmitting the incoherent optical feedback signal to the multimode semiconductor laser; and the output layer is used for obtaining an output signal of the photon reserve pool computing system based on the output optical signal. The photon reserve pool computing system of the novel coding mode designed by the scheme has good information processing performance and stability.

Description

Technical Field

[0001] The present invention relates to the technical field of reservoir computing, and in particular to a photon reservoir computing system and method based on hierarchical pulses. Background Art

[0002] Speech recognition, handwritten digit classification, and time series prediction are key tasks in the field of artificial intelligence and play a crucial role in application scenarios such as voice communication, voice intelligent driving, and power load prediction. With the continuous improvement of the complexity of artificial intelligence algorithms, the energy efficiency ratio and operation timeliness of traditional electronic computing systems gradually reveal physical limitations when completing these tasks. Photonic neuromorphic computing integrates micro-nano photonics and neuromorphic algorithms to construct a new architecture that breaks through the von Neumann bottleneck. Using photons as information carriers, their inherent ultra-high-speed transmission characteristics and multi-wavelength parallel processing capabilities open up a new path for building low-latency and high-throughput intelligent computing systems. When using a photonic neuromorphic computing system for speech recognition, digit classification, and data prediction, it not only has lower energy consumption but also higher processing rates.

[0003] Based on its unique dynamic characteristics, the reservoir computing framework can perform rich non-linear transformations and processing on input signals, thereby capturing the time series information and dynamic patterns in the input signals. Combining it with photonics technology can not only give full play to the advantages of photonics but also realize the functions of neuromorphic computing. Therefore, the reservoir computing framework has become the preferred architecture in photonic neuromorphic computing. This architecture constructs a high-dimensional feature space through a fixed random optical interconnection network, and only needs to optimize the parameters of the output layer to complete the training of the entire photonic neural network. This physically friendly training mechanism significantly reduces the tuning difficulty of the optical system. In recent years, photon neural network systems based on reservoir computing have achieved great success in speech recognition, handwritten digit recognition, etc. By converting the speech or digit signal to be recognized into an input optical signal and inputting it into the photon neural network, the reservoir layer is used to extract the input optical signal for non-linear transformation and mixing to extract the feature information of the input optical signal, and then the signal to be recognized is recognized and predicted.

[0004] In a reservoir-based photonic neural network system, the transmitted signals are divided into two types: continuous signals and discrete pulse signals. When the transmitted signal is a continuous signal, since the signal changes continuously, any tiny noise interference may change the value of the entire transmitted signal, thus affecting the accuracy of the transmitted signal. Moreover, the continuous transmission of the signal will cause the accumulation of optical energy in the system, making the optical power and temperature of the optical device exceed its working range, resulting in device damage or performance degradation, and further affecting the accuracy of information processing. When the transmitted signal is a discrete pulse signal, the simplification of the transmitted signal is achieved by converting the information to be transmitted into discrete binary information. Although the anti-interference performance of the signal is improved and the problem of optical energy accumulation during signal transmission is reduced to a certain extent, when the signal is converted into a binary discrete pulse signal, many detailed information will also be discarded. For example, for a continuously changing light intensity signal, when it is converted into a binary pulse signal, only whether the light intensity exceeds a certain threshold is retained. If it exceeds, it is represented by "1", and if it does not exceed, it is represented by "0". The specific light intensity value and its change information are all discarded, which affects the integrity and accuracy of the information. If the system performs data prediction or recognition based on inaccurate feature information, it will inevitably affect the accuracy of the data prediction or recognition result.

[0005] In summary, in the existing reservoir-based photonic neural network system, after encoding the data to be recognized or classified into the input data of the input layer, the accuracy of the feature information of the input data extracted by using the continuous or discrete pulse signal encoding method is relatively low, resulting in the photonic neural network system outputting incorrect classification results or recognition results based on inaccurate feature information, and further causing consequences such as incorrect command judgment and incorrect power dispatching when applied to fields such as voice communication, voice intelligent driving, power load prediction, or stock market prediction. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to overcome the problem that in the existing reservoir-based photonic neural network system, using the continuous or discrete pulse signal encoding method will reduce the accuracy of the feature information, thereby affecting the accuracy of the output result of the photonic neural network system.

[0007] To solve the above technical problem, the present invention provides a photonic reservoir computing system based on hierarchical pulses, including: An input layer for generating an input electrical signal; A reservoir layer, which specifically includes: A multimode semiconductor laser, connected to the input layer, is used to output a continuous regular optical pulse signal containing initial feature information under the excitation of an input electrical signal; under the excitation of an incoherent optical feedback signal, it generates multimode sub-beams and outputs a continuous regular optical pulse signal containing depth feature information based on the non-linear interference effect between the multimode sub-beams and the incoherent optical feedback signal; A beam splitter prism, used to divide the continuous regular optical pulse signal into two paths, one path is transmitted as a multiplexed optical signal to the feedback loop, and the other path is transmitted as an output optical signal to the output layer; A feedback loop, used to generate an incoherent optical feedback signal based on the multiplexed optical signal and transmit it to the multimode semiconductor laser; An output layer, used to obtain the output signal of the photon reservoir computing system based on the output optical signal.

[0008] Preferably, the feedback loop is a short cavity feedback structure, and its feedback delay is less than the relaxation oscillation period of the multimode semiconductor laser.

[0009] Preferably, the feedback loop includes: A neutral density filter, arranged on the side of the beam splitter prism away from the multimode semiconductor laser, used to transmit the multiplexed optical signal, generate a transmitted optical signal and transmit it to the mirror; transmit the coherent optical feedback signal to the beam splitter prism so that the beam splitter prism transmits the coherent optical feedback signal to the Faraday rotator; A mirror, arranged on the side of the neutral density filter away from the beam splitter prism, used to reflect the transmitted optical signal, generate a reflected optical signal with the same frequency, the same vibration direction and a constant phase difference as the transmitted optical signal and transmit it to the neutral density filter, so that the transmitted optical signal and the reflected optical signal interfere to generate a coherent optical feedback signal; A Faraday rotator, arranged between the multimode semiconductor laser and the beam splitter prism, used to rotate the coherent optical feedback signal, change the polarization state of the coherent optical feedback signal, generate an incoherent optical feedback signal and transmit it to the multimode semiconductor laser.

[0010] Preferably, the Faraday rotator, the beam splitter prism, the neutral density filter and the mirror are discrete optical elements.

[0011] Preferably, the Faraday rotator, the beam splitter prism, the neutral density filter and the mirror are integrally arranged on the same substrate to form a photon integrated circuit.

[0012] Preferably, the input layer includes: A waveform generator, used to generate a mask signal to modulate the signal to be processed and output a modulated signal; An electrical amplifier, connected to the waveform generator, used to amplify the modulated signal; A bias circuit, with its input terminal connected to an electrical amplifier and its output terminal connected to a multimode semiconductor laser, is used to generate an input electrical signal based on a modulation signal; A voltage source, connected to the bias circuit, is used to supply voltage to the bias circuit.

[0013] Preferably, the output layer includes: An optical coupler, which is used to receive the output optical signal transmitted by a beam splitting prism, separate the output optical signal, and output multiple sub-beams; Multiple filters, which are used to filter each sub-beam; Multiple photodetectors, which are connected to the multiple filters one by one, and are used to convert the filtered sub-beams into electrical signals; An adder, connected to the multiple photodetectors, is used to superimpose the multiple electrical signals; A programmable gate array, connected to the adder, is used to train the output of the reservoir layer to obtain the output weights of multiple sub-beams, so as to obtain the output signal of the photonic reservoir computing system.

[0014] Preferably, the output layer further includes: An optical fiber collimator, which is used to collimate the output optical signal transmitted by the beam splitting prism; An optical amplifier, connected to the optical fiber collimator, is used to amplify the collimated output optical signal and transmit the amplified output optical signal to the optical coupler.

[0015] The present invention also provides a photonic reservoir computing method based on hierarchical pulses. The method is implemented by using the above-mentioned photonic reservoir computing system based on hierarchical pulses, and includes: Using the input layer to generate an input electrical signal and transmit it to the multimode semiconductor laser, so that the multimode semiconductor laser performs a laser response to the input electrical signal to generate a continuous regular optical pulse signal containing initial feature information; Using the beam splitting prism to divide the continuous regular optical pulse signal into two optical signals, one as a multiplexed optical signal is transmitted to the feedback loop, and the other as an output optical signal is transmitted to the output layer; Using the feedback loop to generate an incoherent optical feedback signal based on the multiplexed optical signal and transmit the incoherent optical feedback signal to the multimode semiconductor laser; The semiconductor laser receives the incoherent optical feedback signal, generates multiple mode sub-beams under the excitation of the incoherent optical feedback signal, and outputs a continuous regular optical pulse signal containing deep feature information based on the nonlinear interference effect between the multiple mode sub-beams and the incoherent optical feedback signal; Using the output layer to obtain the output signal of the photonic reservoir computing system based on the output optical signal.

[0016] Preferably, generating an incoherent optical feedback signal based on the multiplexed optical signal using a feedback loop comprises: The multiplexed optical signal is transmitted by a neutral filter to generate a transmitted optical signal and transmit it to a reflector; The transmitted light signal is reflected by a reflector to generate a reflected light signal with the same frequency, the same vibration direction and a constant phase difference as the transmitted light signal and transmit it to a neutral filter, so that the transmitted light signal and the reflected light signal interfere with each other to generate a coherent light feedback signal; The neutral density filter transmits the coherent optical feedback signal to the beam splitter prism, so that the beam splitter prism transmits the coherent optical feedback signal to the Faraday rotator; The Faraday rotator rotates the coherent optical feedback signal, changes the polarization state of the coherent optical feedback signal, generates an incoherent optical feedback signal and transmits it to the multi-mode semiconductor laser.

[0017] The photon reserve pool computing system based on graded pulses provided in this application has the following beneficial effects: When an electrical signal is injected into a multi-mode semiconductor laser, the carrier concentration inside the multi-mode semiconductor laser will change accordingly, thereby changing the gain and refractive index of the multi-mode semiconductor laser, so that the relaxation oscillation frequency of the multi-mode semiconductor laser matches the frequency of the electrical signal, thereby generating a continuous regular optical pulse signal containing characteristic information. Therefore, the present application encodes the data to be processed into an input electrical signal to excite the multi-mode semiconductor laser, so that it generates a continuous regular optical pulse signal containing initial characteristic information during preliminary feature extraction. Furthermore, a beam splitter is used to divide the optical pulse signal into two paths, one of which is transmitted to the output layer as an output optical signal, and the other is transmitted to the feedback loop as a multiplexed optical signal. The feedback loop enables the multi-mode semiconductor laser to perform multiple depth feature extractions on the multiplexed optical signal. If the multiplexed optical signal is directly returned to the multi-mode semiconductor laser, the multi-mode semiconductor laser cannot output a continuous regular optical pulse signal under its excitation. Therefore, in order to ensure that the multi-mode semiconductor laser can output a continuous regular optical pulse signal during each depth feature extraction, the present application uses a feedback loop to process the multiplexed optical signal to generate an incoherent optical feedback signal. Since the multi-mode semiconductor laser can generate a multi-mode sub-beam under the excitation of an incoherent optical feedback signal, sub-beams of different modes have different phases. Based on this characteristic, an incoherent optical feedback signal is used to introduce a phase disturbance to destroy the stable relationship between the gain and phase of the multi-mode semiconductor laser, thereby triggering the self-excited oscillation of the multi-mode semiconductor laser. This nonlinear interference effect enables the multi-mode semiconductor laser to output a continuous regular optical pulse signal containing depth feature information, thereby ensuring that a continuous regular optical pulse signal is output during the entire feature extraction process.

[0018] This application uses a multimode semiconductor laser to form a reservoir layer, and at the same time uses an electrical signal and an incoherent optical feedback signal to enable it to output a continuous regular optical pulse signal containing feature information during each feature extraction. Compared with a continuous signal, this continuous regular optical pulse signal has a fixed period and clear time characteristics, thus effectively avoiding the interference of noise on the feature information, ensuring the accuracy of the feature information, and at the same time concentrating the energy in each pulse, solving the problem of continuous accumulation of energy during signal transmission caused by the continuous distribution of energy in time; compared with a discrete pulse signal, the pulses in the continuous regular optical pulse signal are continuous in time, so the amplitude change, phase parameter, etc. of adjacent pulses all contain the feature information of the data to be processed, thus avoiding the problem of detail loss during data processing. The photon reservoir computing system designed in this application uses a new information encoding method to overcome the defects in the transmission process of continuous signals and discrete pulse signals, improving the feature extraction accuracy of the system when extracting features from input data encoded based on data to be recognized or classified, so as to accurately identify, classify or predict the input data, and improve the command recognition accuracy and prediction of the system in fields such as voice intelligent driving and power load prediction. Description of the Drawings

[0019] In order to make the content of the present invention easier to be clearly understood, the following further details the present invention according to specific embodiments of the present invention in conjunction with the drawings, where: Figure 1 It is a schematic structural diagram of a photon reservoir computing system based on hierarchical pulses provided by this application; Figure 2 It is a flowchart of a photon reservoir computing method based on hierarchical pulses provided by this application; Figure 3 It is a schematic diagram of the dynamics of a multimode semiconductor laser based on an incoherent optical feedback signal provided by an embodiment of this application; Figure 4 It is a schematic diagram of the prediction result of data prediction using a photon reservoir computing system based on hierarchical pulses provided by an embodiment of this application; where, Figure 4 in (a) is a schematic diagram of the true result of the data to be predicted, Figure 4 in (b) is a schematic diagram of the prediction result output by the photon reservoir computing system based on hierarchical pulses; Description of the reference numerals in the drawings: 1. Input layer; 11. Waveform generator; 12. Electrical amplifier; 13. Bias circuit; 14. Voltage source; 2. Reservoir layer; 21. Multi-mode semiconductor laser; 22. Beam splitter prism; 23. Neutral density filter; 24. Mirror; 25. Faraday rotator; 3. Output layer; 31. Optical coupler; 32. Optical filter; 33. Photodetector; 34. Adder; 35. Programmable gate array; 36. Fiber optic collimator; 37. Optical amplifier. Detailed implementation manners

[0020] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the specific embodiments cited are not intended to limit the present invention.

[0021] Please refer to Figure 1 , Figure 1 which shows a schematic structural diagram of a photon reservoir computing system based on hierarchical pulses provided by the present application. The system includes an input layer 1, a reservoir layer 2, and an output layer 3.

[0022] The input layer 1 is used to generate an input electrical signal.

[0023] The reservoir layer 2 includes a multi-mode semiconductor laser 21, a beam splitter prism 22, and a feedback loop.

[0024] The multi-mode semiconductor laser 21 is connected to the input layer 1 and is used to output a continuous regular optical pulse signal containing initial feature information under the excitation of the input electrical signal; under the excitation of the incoherent optical feedback signal, it generates multi-mode sub-beams, and based on the non-linear interference effect between the multi-mode sub-beams and the incoherent optical feedback signal, it outputs a continuous regular optical pulse signal containing depth feature information.

[0025] Optionally, the multi-mode semiconductor laser 21 can be a multi-longitudinal mode semiconductor laser, a multi-transverse mode semiconductor laser, or a multi-polarization state semiconductor laser.

[0026] The beam splitter prism 22 is used to divide the optical pulse signal into two paths, one path is used as a multiplexed optical signal to be transmitted to the feedback loop, and the other path is used as an output optical signal to be transmitted to the output layer 3.

[0027] The feedback loop is used to generate an incoherent optical feedback signal based on the multiplexed optical signal.

[0028] Specifically, the use of the incoherent optical feedback signal can introduce mismatched phase perturbations and amplitude variations, thereby disturbing the stable state of the multi-mode semiconductor laser, triggering non-linear effects, causing the multi-mode semiconductor laser to enter a non-linear self-excited oscillation mode, and further generating a regular pulse output signal under specific parameter conditions.

[0029] The output layer 3 is used to obtain the output signal of the photonic reservoir computing system based on the output optical signal.

[0030] In this application, an input electrical signal is used to excite a multimode semiconductor laser. When an electrical signal is injected into the multimode semiconductor laser, the carrier concentration inside the multimode semiconductor laser changes accordingly, thereby changing the gain and refractive index of the multimode semiconductor laser, making the relaxation oscillation frequency of the multimode semiconductor laser match the frequency of the electrical signal, and thus generating a continuous and regular optical pulse signal containing initial characteristic information. Further, a beam splitter prism is used to divide the optical pulse signal into two paths, one path is transmitted to the output layer, and the other path is transmitted to the feedback loop, so that the multimode semiconductor laser continuously extracts deep characteristic information. In order to enable the multimode semiconductor laser to still output a continuous and regular optical pulse signal during deep feature extraction, this application uses the feedback loop to process the multiplexed optical signal to generate an incoherent optical feedback signal. Since the multimode semiconductor laser can generate multimode sub-beams, and different mode sub-beams have different frequencies and phases, therefore, the incoherent optical feedback signal is used to introduce phase perturbation, so that the phases of different mode sub-beams output by the multimode semiconductor laser do not match the phase of the incoherent optical feedback signal, thereby destroying the stable relationship between the gain and phase of the multimode semiconductor laser, triggering the self-oscillation of the multimode semiconductor laser, and enabling the multimode semiconductor laser to output a stable response signal with a fixed frequency interval, that is, a continuous and regular optical pulse signal.

[0031] This application uses a multimode semiconductor laser to form a reservoir layer, and at the same time uses an electrical signal and an incoherent optical feedback signal to excite it to continuously output a continuous and regular optical pulse signal containing characteristic information during the entire feature extraction process. Compared with continuous signals, this continuous and regular optical pulse signal has a fixed period and clear time characteristics, so it can effectively avoid the interference of noise on the overall information, ensure the accuracy of the overall information, and at the same time concentrate the energy in each pulse, solving the problem of continuous accumulation of energy during signal transmission caused by continuous energy distribution in time. Compared with discrete pulse signals, the individual pulses in the continuous and regular optical pulse signal are continuous in time. Therefore, the amplitude change, phase parameter, etc. of adjacent pulses all contain the characteristic information of the data to be processed, and the information is not only represented by the presence or absence of pulses, thus avoiding the problem of detail loss during data processing. The novel information coding method designed in this application overcomes the defects in the transmission process of continuous signals and discrete pulse signals, improves the information processing accuracy of the system, and further improves the accuracy of the output result when the system performs data recognition and prediction.

[0032] Further, in some embodiments of this application, the feedback loop is a short cavity feedback structure, and its feedback delay is less than the relaxation oscillation period of the multimode semiconductor laser 21.

[0033] Specifically, the relaxation oscillation period is a periodic oscillation phenomenon that appears in the output power of the multi-mode semiconductor laser 21 after it is turned on. When the feedback delay is short, the incoherent optical feedback signal can return within a short time of the photon lifetime in the multi-mode semiconductor laser 21, enabling the multi-mode semiconductor laser 21 to respond more quickly to the feedback signal, improving the stability of its output power, reducing power fluctuations and noise, and thus outputting a more stable optical pulse signal.

[0034] Further, as Figure 1 shown, the feedback loop includes a neutral filter 23, a mirror 24, and a Faraday rotator 25.

[0035] The neutral filter 23 is disposed on the side of the beam splitter prism 22 away from the multi-mode semiconductor laser 21, and is used to transmit the multiplexed optical signal to generate a transmitted optical signal; the coherent optical feedback signal is transmitted to the beam splitter prism 22 so that the beam splitter prism 22 transmits the coherent optical feedback signal to the Faraday rotator 25.

[0036] The mirror 24 is disposed on the side of the neutral filter 23 away from the beam splitter prism 22, and is used to reflect the transmitted optical signal to generate a reflected optical signal with the same frequency, the same vibration direction, and a constant phase difference as the transmitted optical signal and transmit it to the neutral filter 23, so that the reflected optical signal and the transmitted optical signal interfere to generate a coherent optical feedback signal.

[0037] The Faraday rotator 25 is disposed between the multi-mode semiconductor laser 21 and the beam splitter prism 22, and is used to rotate the coherent optical feedback signal, change the polarization state of the coherent optical feedback signal, generate an incoherent optical feedback signal and transmit it to the multi-mode semiconductor laser 21.

[0038] Specifically, by setting the Faraday rotator 25 to rotate the polarization state of the coherent optical feedback signal, the polarization state of the generated incoherent optical feedback signal can be better matched with the optical signal of a specific mode in the multi-mode semiconductor laser 21, so as to more effectively excite the optical signal of this mode to generate a continuous and regular optical pulse signal, and improve the quality and stability of the optical pulse signal output by the multi-mode semiconductor laser.

[0039] Optionally, in some embodiments of the present application, the Faraday rotator 25, the beam splitter prism 22, the neutral filter 23, and the mirror 24 can be discrete optical elements, and these optical elements are coaxially arranged in sequence along the optical path propagation direction to form a coaxial optical system.

[0040] In other embodiments, the Faraday rotator 25, the beam splitter prism 22, the neutral filter 23, and the mirror 24 can also be integrally disposed on the same substrate and arranged in the form of a photonic integrated circuit in the photonic reservoir computing system.

[0041] Further, as Figure 1 shown, the input layer 1 includes a waveform generator 11, an electrical amplifier 12, a bias circuit 13, and a voltage source 14.

[0042] The waveform generator 11 is configured to generate a mask signal to modulate the signal to be processed and output a modulated signal.

[0043] Specifically, the signal to be processed is the signal to be recognized or predicted, and the waveform generator 11 generates equally spaced square wave signals to modulate the signal to be processed.

[0044] The electrical amplifier 12 is connected to the waveform generator 11 and is configured to amplify the modulated signal.

[0045] The input end of the bias circuit 13 is connected to the electrical amplifier 12, and the output end is connected to the multimode semiconductor laser 21, and is configured to generate an input electrical signal based on the modulated signal.

[0046] The voltage source 14 is connected to the bias circuit 13 and is configured to provide a voltage for the bias circuit 13.

[0047] In this application, using an electrical signal can not only stimulate the multimode semiconductor laser 21 to output continuous and regular optical pulse signals. At the same time, the generation, modulation, and control of the electrical signal are relatively simple. By changing parameters such as the current or voltage of the input electrical signal, precise control over characteristics such as the output intensity and frequency of the multimode semiconductor laser 21 can be achieved, without the need to use a complex optical system to generate a pump light source, reducing the volume of the photonic reservoir computing system.

[0048] Further, since the multimode semiconductor laser 21 can generate multimode sub-beams, and the sub-beams of different modes have different characteristics such as frequency, phase, and spatial distribution, etc., therefore, in this application, the output optical signal is first separated in the output layer, and the multimode characteristics of the multimode semiconductor laser 21 are used to parallelly distribute information to different sub-beams for processing. By assigning different weights to different sub-beams, the information carried by each sub-beam is amplified, attenuated, and combined to different degrees, thereby improving the information processing ability and efficiency of the photonic reservoir computing system and improving the accuracy of its output results.

[0049] Specifically, as Figure 1 shown, the output layer 3 includes an optical coupler 31, a plurality of filters 32, a plurality of photodetectors 33, an adder 34, and a programmable gate array 35.

[0050] The optical coupler 31 is configured to receive the output optical signal transmitted by the beam splitting prism 22 and separate the output optical signal to output multiple sub-beams.

[0051] The plurality of filters 32 are configured to filter each sub-beam.

[0052] A plurality of photodetectors 33 are connected to a plurality of filters 32 in a one-to-one correspondence, and are configured to convert the filtered sub-beams into electrical signals.

[0053] An adder 34 is connected to the plurality of photodetectors 33, and is configured to superimpose the plurality of electrical signals.

[0054] A programmable gate array 35 is connected to the adder 34, and is configured to train the output of the reservoir layer 2 to obtain the output weights of the multi-path sub-beams, so as to obtain the output signal of the photonic reservoir computing system.

[0055] Specifically, in a specific example of the present application, a linear regression algorithm and a ridge regression algorithm are used to train the output of the reservoir layer 2.

[0056] Optionally, the output layer 3 further includes an optical fiber collimator 36 and an optical amplifier 37.

[0057] The optical fiber collimator 36 is configured to collimate the output optical signal transmitted by the beam splitter prism 22.

[0058] The optical amplifier 37 is connected to the optical fiber collimator 36, and is configured to amplify the collimated output optical signal and transmit the amplified output optical signal to the optical coupler 31.

[0059] Exemplarily, the output of the reservoir layer 2 can be expressed as: X = X 1 (t) + X 2 (t) + …… X n (t), where X 1 (t) represents the electrical signal obtained after the conversion of the first path of sub-beam, and X 2 (t) represents the electrical signal obtained after the conversion of the second path of sub-beam, and X n (t) represents the electrical signal obtained after the conversion of the nth path of sub-beam, and n represents the number of sub-beams.

[0060] Based on the photon reservoir computing system based on hierarchical pulses provided in the above embodiments, the embodiments of the present application further provide a method for photon reservoir computing based on hierarchical pulses, as Figure 2 shown, the method includes: S10: Using the input layer to generate an input electrical signal and transmit it to the multimode semiconductor laser, so that the multimode semiconductor laser performs a laser response to the input electrical signal to generate a continuous regular optical pulse signal containing initial feature information.

[0061] S20: Using the beam splitter prism to divide the continuous regular optical pulse signal into two optical signals, one as a multiplexed optical signal is transmitted to the feedback loop, and the other as an output optical signal is transmitted to the output layer.

[0062] S30: Generate an incoherent optical feedback signal based on the multiplexed optical signal using a feedback loop, and transmit the incoherent optical feedback signal to a multimode semiconductor laser.

[0063] S40: The semiconductor laser receives the incoherent optical feedback signal, generates multiple mode sub-beams under the excitation of the incoherent optical feedback signal, and outputs a continuous regular optical pulse signal containing depth feature information based on the non-linear interference effect between the multiple mode sub-beams and the incoherent optical feedback signal.

[0064] Specifically, the incoherent optical feedback signal is used to introduce phase perturbation, so that the signal phases of the incoherent optical feedback signal and the multiple mode sub-beams output by the multimode semiconductor laser do not match. This perturbation destroys the stable relationship between the gain and phase of the multimode semiconductor laser, resulting in a rapid change in gain. Under certain conditions, this gain fluctuation will trigger the self-oscillation of the multimode semiconductor laser, thus triggering periodic pulse output.

[0065] S50: Use the output layer to obtain the output signal of the photonic reservoir computing system based on the output optical signal.

[0066] Further, generating the incoherent optical feedback signal based on the multiplexed optical signal using the feedback loop includes: Transmit the multiplexed optical signal using a neutral density filter to generate a transmitted optical signal and transmit it to a mirror; Reflect the transmitted optical signal using the mirror to generate a reflected optical signal with the same frequency, the same vibration direction, and a constant phase difference as the transmitted optical signal, and transmit it to the neutral density filter, so that the transmitted optical signal and the reflected optical signal interfere to generate a coherent optical feedback signal; The neutral density filter transmits the coherent optical feedback signal to a beam splitter prism, so that the beam splitter prism transmits the coherent optical feedback signal to a Faraday rotator; The Faraday rotator rotates the coherent optical feedback signal, changes the polarization state of the coherent optical feedback signal, generates an incoherent optical feedback signal and transmits it to the multimode semiconductor laser.

[0067] The following further explains the above embodiments through a specific example. In this example, the above-mentioned photonic reservoir computing system based on hierarchical pulses is used to predict time series data. Specifically, first, the photonic reservoir computing system based on hierarchical pulses is simulated based on numerical simulation, as Figure 3 shown is the dynamic schematic diagram of the multimode semiconductor laser based on the incoherent optical feedback signal in this system. The rate equations of this system are as follows: ,

[0068] , , Among them, represents the photon density of the multimode semiconductor laser; represents the carrier density of the multimode semiconductor laser; represents the photon lifetime; represents the carrier lifetime; represents the feedback delay time; represents the feedback intensity; represents the confinement factor; represents the bias current; represents the active region volume; represents the gain coefficient; represents the charge number; represents Gaussian white noise; represents the optical gain of the laser cavity; represents the differential gain coefficient; represents the transparent carrier coefficient.

[0069] Specifically, the specific steps for using this system to predict time series data are as follows: Step 1: The waveform generator generates a mask signal to modulate the time series data to be predicted, generating a modulated signal. The bias circuit multiplies the modulated signal by the output voltage of the voltage source to obtain an input electrical signal.

[0070] Step 2: Input the input electrical signal into the reservoir layer, and use the multimode semiconductor laser to extract the initial feature information of the input electrical signal, and output a continuous regular optical pulse signal containing the initial feature information.

[0071] Step 3: Use the beam splitter prism to divide the optical pulse signal into two paths. One path is used as the output optical signal and transmitted to the output layer, and the other path is used as the multiplexed optical signal and transmitted to the feedback loop.

[0072] Step 4: The feedback loop generates an incoherent optical feedback signal based on the multiplexed optical signal, and returns the incoherent optical feedback signal to the multimode semiconductor laser, so that the multimode semiconductor laser generates multimode sub-beams under the excitation of the incoherent optical feedback signal, and outputs a continuous regular optical pulse signal containing deep feature information.

[0073] Step 5: The output layer divides the output optical signal into n sub-beams, filters and performs optoelectronic conversion on each sub-beam to obtain n electrical signals as the output of the reservoir layer; trains the output to obtain the output weights of each sub-beam, and obtains the predicted time series data based on each sub-beam and its output weights.

[0074] Such as Figure 4 is a schematic diagram of the prediction result of using the photon reservoir computing system based on hierarchical pulses provided by the embodiment of the present application for data prediction; among them,Figure 4 Among them, (a) is a schematic diagram of the true result of the data to be predicted, Figure 4 and (b) is a schematic diagram of the prediction result output by the photon reservoir computing system based on hierarchical pulses. As can be seen from Figure 4 it, the predicted time series data obtained in this embodiment is very close to the actual time series data, and the value of the calculated normalized mean square error NMSE is only 0.003, which fully proves that the photon reservoir computing system based on hierarchical pulses designed in this application can improve the accuracy of data processing through a new information encoding method, thereby improving its accuracy in tasks such as data prediction and data recognition.

[0075] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented 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.

[0076] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0077] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide for implementing in the processFigure 1 One process or multiple processes and / or boxes Figure 1 Steps of the functions specified in one box or multiple boxes.

[0079] Obviously, the above embodiments are only examples given for clear illustration and are not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all implementation manners here. And the obvious changes or modifications derived therefrom are still within the protection scope of the present invention.

Claims

1. A photon reserve pool computing system based on graded pulses, characterized in that: include: An input layer, used to generate an input electrical signal; The reserve pool layer includes: A multi-mode semiconductor laser is connected to the input layer and is used to output a continuous regular optical pulse signal containing initial characteristic information under the stimulation of an input electrical signal; generate a multi-mode sub-beam under the stimulation of an incoherent optical feedback signal, and output a continuous regular optical pulse signal containing depth characteristic information based on the nonlinear interference between the multi-mode sub-beam and the incoherent optical feedback signal; A beam splitter prism is used to split the continuous regular optical pulse signal into two paths, one of which is transmitted to the feedback loop as a multiplexed optical signal, and the other is transmitted to the output layer as an output optical signal; A feedback loop, used for generating an incoherent optical feedback signal based on the multiplexed optical signal and transmitting the incoherent optical feedback signal to the multimode semiconductor laser; The output layer is used to obtain the output signal of the photon reservoir computing system based on the output light signal.

2. The photon reserve pool computing system based on graded pulses according to claim 1, characterized in that: The feedback loop is a short cavity feedback structure, and its feedback delay is less than the relaxation oscillation period of the multi-mode semiconductor laser.

3. The photon reserve pool computing system based on graded pulses according to claim 1, characterized in that: The feedback loop includes: A neutral filter is arranged on the side of the beam splitter prism away from the multi-mode semiconductor laser, and is used to transmit the multiplexed optical signal, generate a transmitted optical signal and transmit it to the reflector; transmit the coherent optical feedback signal to the beam splitter prism, so that the beam splitter prism transmits the coherent optical feedback signal to the Faraday rotator; A reflector is arranged on the side of the neutral filter away from the beam splitter prism, and is used to reflect the transmitted light signal, generate a reflected light signal with the same frequency, the same vibration direction and a constant phase difference as the transmitted light signal, and transmit the reflected light signal to the neutral filter, so that the transmitted light signal and the reflected light signal interfere with each other to generate a coherent light feedback signal; The Faraday rotator is arranged between the multi-mode semiconductor laser and the beam splitter prism, and is used to rotate the coherent light feedback signal, change the polarization state of the coherent light feedback signal, generate an incoherent light feedback signal and transmit it to the multi-mode semiconductor laser.

4. The photon reserve pool computing system based on graded pulses according to claim 3, characterized in that: The Faraday rotator, the beam splitter, the neutral density filter and the reflector are optical components which are arranged separately.

5. The photon reserve pool computing system based on graded pulses according to claim 3, characterized in that: A Faraday rotator, a beam splitter, a neutral filter and a reflector are integrated on the same substrate to form a photonic integrated circuit.

6. The photon reserve pool computing system based on graded pulses according to claim 1, characterized in that: The input layer consists of: A waveform generator is used to generate a mask signal to modulate the signal to be processed and output a modulated signal; an electrical amplifier, connected to the waveform generator, for amplifying the modulated signal; a bias circuit, the input end of which is connected to the electric amplifier and the output end of which is connected to the multi-mode semiconductor laser, for generating an input electric signal based on the modulation signal; The voltage source is connected to the bias circuit and is used to provide voltage to the bias circuit.

7. The photon reserve pool computing system based on graded pulses according to claim 1, characterized in that: The output layer consists of: An optical coupler is used to receive the output optical signal transmitted by the beam splitter prism, separate the output optical signal, and output multiple sub-beams; Multiple filters, used for filtering each sub-beam; A plurality of photodetectors are connected to the plurality of filters in a one-to-one correspondence, and are used to convert the filtered sub-beams into electrical signals; an adder, connected to the plurality of photodetectors, for superimposing the plurality of electrical signals; The programmable gate array is connected to the adder and is used to train the output of the reservoir layer to obtain the output weights of multiple sub-beams, thereby obtaining the output signal of the photon reservoir computing system.

8. The photon reserve pool computing system based on graded pulses according to claim 7, characterized in that: The output layer also includes: A fiber collimator, used to collimate the output optical signal transmitted by the beam splitter prism; The optical amplifier is connected to the optical fiber collimator and is used to amplify the collimated output optical signal and transmit the amplified output optical signal to the optical coupler.

9. A photon reserve pool calculation method based on graded pulses, characterized in that: The method is implemented using the hierarchical pulse-based photon reserve pool computing system according to any one of claims 1 to 8, comprising: An input electrical signal is generated by the input layer and transmitted to the multi-mode semiconductor laser, so that the multi-mode semiconductor laser performs a laser response to the input electrical signal and generates a continuous regular optical pulse signal containing initial characteristic information; The continuous regular optical pulse signal is divided into two optical signals by using a beam splitter, one of which is transmitted to the feedback loop as a multiplexed optical signal, and the other is transmitted to the output layer as an output optical signal; generating an incoherent optical feedback signal based on the multiplexed optical signal using a feedback loop, and transmitting the incoherent optical feedback signal to a multimode semiconductor laser; The semiconductor laser receives an incoherent optical feedback signal, generates a plurality of mode sub-beams under the stimulation of the incoherent optical feedback signal, and outputs a continuous regular optical pulse signal containing depth feature information based on the nonlinear interference between the multi-mode sub-beams and the incoherent optical feedback signal; The output layer is used to obtain the output signal of the photon reservoir computing system based on the output light signal.

10. The photon reserve pool calculation method based on graded pulses according to claim 9, characterized in that: Generating an incoherent optical feedback signal based on the multiplexed optical signal using a feedback loop includes: The multiplexed optical signal is transmitted by a neutral filter to generate a transmitted optical signal and transmit it to a reflector; The transmitted light signal is reflected by a reflector to generate a reflected light signal with the same frequency, the same vibration direction and a constant phase difference as the transmitted light signal and transmit it to a neutral filter, so that the transmitted light signal and the reflected light signal interfere with each other to generate a coherent light feedback signal; The neutral density filter transmits the coherent optical feedback signal to the beam splitter prism, so that the beam splitter prism transmits the coherent optical feedback signal to the Faraday rotator; The Faraday rotator rotates the coherent optical feedback signal, changes the polarization state of the coherent optical feedback signal, generates an incoherent optical feedback signal and transmits it to the multi-mode semiconductor laser.

Images