Optoelectronic reserve computing device, method, apparatus, medium and program product

Abstract

The invention provides a photoelectric reserve pool computing device, method and equipment, a medium and a program product, relates to the technical field of informatization software systems, and is used for reducing the complexity of a hardware structure of a deep photoelectric reserve pool computing system and improving the overall computing performance. The device comprises a light source module, a multiplexer and a reservoir processing module which are connected in sequence, the light source module is used for generating at least two light signals with different characteristics; the multiplexer is used for multiplexing the at least two optical signals to the same optical transmission path to obtain a multiplexed optical signal; the reserve pool processing module is used for receiving an input electric signal to be processed, coupling the input electric signal to the multiplexed optical signal, performing photoelectric processing on the coupled optical signal to obtain and output a reserve pool state signal, and the reserve pool state signal comprises at least two state components; the at least two state components are associated with different characteristics of the at least two optical signals, respectively. The method is applied to the depth photoelectric reserve pool computing platform.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information software systems, and particularly relates to an optoelectronic reservoir computing device, method, equipment, medium and program product. BACKGROUND

[0002] Artificial intelligence technology has shown great potential in the field of signal processing and computing, but its training process usually requires complex algorithms and a large amount of computing resources. Reservoir computing simplifies neural network training by fixing internal connection weights and training only output weights. Optoelectronic reservoir combines reservoir computing and photonics technology, and has the advantages of simple training and high speed, low power consumption of optical computing. To improve performance, existing solutions introduce a deep architecture, that is, multiple physical reservoir layers or feedback loops are connected in series.

[0003] These existing deep optoelectronic reservoir systems usually construct independent or partially independent optoelectronic processing loops for each layer. This implementation mode leads to complex system hardware structure and numerous components. When signals are transmitted and processed between different physical loops, additional distortion and delay are easily introduced, and system noise is aggravated, resulting in decreased calculation accuracy. At the same time, the management of multiple physical layers also significantly increases the calculation and regulation complexity of the overall system. The above factors seriously restrict the improvement of the overall performance of the deep optoelectronic reservoir. Therefore, the existing deep optoelectronic reservoir computing system has the problems of complex hardware structure and poor overall calculation performance. SUMMARY

[0004] The present application provides an optoelectronic reservoir computing device, method, equipment, medium and program product, which is used to reduce the complexity of the hardware structure of the deep optoelectronic reservoir computing system and improve the overall calculation performance.

[0005] In a first aspect, the present application provides an optoelectronic reservoir computing device, comprising: an optical source module, a multiplexer and a reservoir processing module connected in sequence; the optical source module is used to generate at least two optical signals with different characteristics; the multiplexer is used to multiplex the at least two optical signals to the same optical transmission path to obtain a multiplexed optical signal; the reservoir processing module is used to receive an input electrical signal to be processed, and couple the input electrical signal to the multiplexed optical signal, and perform optoelectronic processing on the coupled optical signal to obtain and output a reservoir state signal, the reservoir state signal comprising at least two state components, the at least two state components being respectively associated with different characteristics of the at least two optical signals.

[0006] The technical scheme provided by the application brings at least the following beneficial effects: the optoelectronic reserve pool computing device is composed of a light source module, a multiplexer and a reserve pool processing module connected in sequence, a plurality of characteristic light signals are generated and multiplexed to the same transmission path, the physical hardware is highly simplified, a plurality of logical processing layers are mapped to light signals of different characteristics, and unified modulation, delay and conversion are completed in a shared physical loop, this architecture avoids the problems of hardware quantity explosion and structural complexity caused by traditional multi-physical layer stacking, all signals are processed in a shared optical path, the interlayer signal conversion interface is eliminated, additional noise, distortion and inconsistent delay introduced by multiple independent loops are reduced, and the fidelity and consistency of signal processing are improved. In this way, without increasing the complexity of physical hardware and signal degradation, the processing capacity and dynamic characteristics of the system can be effectively expanded by increasing the number of multiplexed light signal characteristics, thereby reducing the complexity and cost of the hardware structure of the deep optoelectronic reserve pool computing system and improving the overall computing performance.

[0007] In a possible implementation manner, the multiplexer includes any one of the following: a wavelength division multiplexer, a polarization multiplexer, and a space division multiplexer; the polarization multiplexer is configured to multiplex at least two light signals with orthogonal polarization states to the same optical transmission path; and the space division multiplexer is configured to multiplex at least two light signals from different spatial channels or different spatial modes to the same optical transmission path.

[0008] In another possible implementation manner, the reserve pool processing module includes a polarization controller, an optical modulator, a shared optical delay loop, a photodetector and a signal processing unit; the polarization controller is connected to the multiplexer and is configured to perform polarization adjustment on the multiplexed light signal; the optical modulator is connected to the polarization controller and is configured to receive an input electrical signal and couple the input electrical signal to the multiplexed light signal after polarization adjustment to output a modulated light signal; the shared optical delay loop is connected to the optical modulator and is configured to cyclically transmit the modulated light signal to increase the transmission delay of the modulated light signal; the photodetector is connected to the shared optical delay loop and is configured to convert the cyclically transmitted light signal into a first electrical signal; and the signal processing unit is connected to the photodetector and is configured to process the first electrical signal to generate and output a reserve pool state signal.

[0009] In still another possible implementation manner, the signal processing unit includes an electrical amplifier, an analog-to-digital converter and a digital signal processor; the electrical amplifier is connected to the photodetector and is configured to amplify the first electrical signal to obtain an analog electrical signal; the analog-to-digital converter is connected to the electrical amplifier and is configured to convert the analog electrical signal into a digital signal; and the digital signal processor is connected to the analog-to-digital converter and is configured to extract at least two state components from the digital signal to generate and output the reserve pool state signal.

[0010] In another possible implementation, an output terminal of the above-mentioned electric amplifier is connected to an electric input terminal of the optical modulator, for inputting part of the analog electric signals as a feedback electric signal to the optical modulator.

[0011] In a second aspect, the present application provides an optoelectronic reservoir computing method, comprising: obtaining at least two optical signals with different characteristics; multiplexing the at least two optical signals to the same optical transmission path to obtain a multiplexed optical signal; receiving an input electric signal to be processed, and coupling the input electric signal to the multiplexed optical signal, and performing optoelectronic processing on the coupled optical signal to obtain a reservoir state signal, the reservoir state signal comprising at least two state components, the at least two state components being respectively associated with the different characteristics of the at least two optical signals.

[0012] In a possible implementation, the above-mentioned optoelectronic processing on the coupled optical signal to obtain the reservoir state signal comprises: obtaining a transmission time delay of the coupled optical signal after cyclic transmission; performing optoelectronic conversion on the optical signal after the cyclic transmission based on the product of the response coefficient of the photodetector and the matching resistance to obtain a first electric signal; amplifying the first electric signal using a target gain, and filtering the amplified electric signal using a target lower cutoff frequency and a target upper cutoff frequency to obtain a second electric signal; calculating the at least two state components based on the transmission time delay, the second electric signal, the input electric signal, and a preset parameter group to generate the reservoir state signal, the preset parameter group comprising a half-wave voltage of the optical modulator, and a bias voltage, an input scaling factor, and an initial optical signal intensity corresponding to each optical signal respectively.

[0013] In another possible implementation, the above-mentioned calculation of the at least two state components based on the transmission time delay, the second electric signal, the input electric signal, and the preset parameter group comprises: for each optical signal, calculating a merged electric signal based on the input electric signal, and the bias voltage, the input scaling factor, and a historical state component of the optical signal before the transmission time delay; performing nonlinear modulation processing on the merged electric signal based on the half-wave voltage to obtain a modulation intensity coefficient; calculating a first optical intensity value based on the initial optical signal intensity corresponding to the optical signal and the modulation intensity coefficient; and calculating a state component corresponding to the optical signal based on the product of the first optical intensity value and the response coefficient of the photodetector and the matching resistance to obtain the at least two state components.

[0014] In a third aspect, the present application provides an electronic device, comprising the above-mentioned optoelectronic reservoir computing apparatus of the first aspect.

[0015] In a fourth aspect, the present application provides an electronic device, comprising: a processor and a memory; the memory stores instructions executable by the processor; and the processor is configured to execute the instructions to enable the electronic device to implement the method of the second aspect.

[0016] In a fifth aspect, the present application provides a computer readable storage medium, comprising: computer software instructions; when the computer software instructions are run in an electronic device, the electronic device implements the method of the second aspect.

[0017] In a sixth aspect, the present application provides a computer program product, comprising: a computer program; when the computer program is run in an electronic device, the electronic device implements the method of the second aspect.

[0018] The beneficial effects of the second aspect to the sixth aspect are described in the corresponding description of the first aspect, and will not be repeated. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A structural schematic diagram of a deep optoelectronic reserve pool computing system model provided by the related art; Figure 2 An application environment schematic diagram of an optoelectronic reserve pool computing method provided by the embodiment of the present application; Figure 3 A component schematic diagram of an optoelectronic reserve pool computing apparatus provided by the embodiment of the present application; Figure 4 A component schematic diagram of another optoelectronic reserve pool computing apparatus provided by the embodiment of the present application; Figure 5 A component schematic diagram of still another optoelectronic reserve pool computing apparatus provided by the embodiment of the present application; Figure 6 A flow schematic diagram of an optoelectronic reserve pool computing method provided by the embodiment of the present application; Figure 7 A schematic diagram of the overall implementation process of an optoelectronic reserve pool computing method provided by the embodiment of the present application; Figure 8 A component schematic diagram of an electronic device provided by the embodiment of the present application; Figure 9 A structural schematic diagram of an electronic device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0020] The optoelectronic reserve pool computing apparatus, method, device, medium and program product provided by the present application will be described in detail below with reference to the accompanying drawings.

[0021] The term "and / or" in this paper is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone.

[0022] The terms "first", "second", and the like in the description of the present application and in the claims of the present application are used to distinguish different objects, or to distinguish different treatments of the same object, and are not used to describe a specific order of the objects.

[0023] In addition, the terms "comprise" and "have" and any variations thereof in the description of the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but can optionally include other steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0024] It should be noted that in the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or advantageous than other embodiments or design solutions. Rather, the use of "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0025] In order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, "first", "second" and the like are used to distinguish the same or similar items with basically the same function and role. Those skilled in the art can understand that "first", "second" and the like are not limited in number and execution order.

[0026] In the description of the present application, unless otherwise specified, "a plurality of" means two or more.

[0027] The optical-electric reserve pool computing device, method, equipment, medium and program product provided by the embodiments of the present application can be applied to high-speed optical communication intelligent processing, complex timing prediction, real-time pattern recognition and low-power edge intelligent computing and other computing scenarios. Specifically, it can include but is not limited to the following scenarios: 1. High-speed optical communication and intelligent signal processing Optical performance monitoring: In a coherent optical communication system, the present scheme can be used for real-time and joint monitoring of modulation format identification and optical signal-to-noise ratio estimation. Its high bandwidth and optical domain parallel processing capability can directly analyze high-speed optical signals online, avoiding the processing delay and high power consumption problems caused by the sampling rate and computational complexity limitations of traditional electric domain digital signal processors.

[0028] Adaptive equalization and impairment compensation: suitable for nonlinear impairment compensation and channel equalization in optical fiber communication. The dynamic characteristics of the reserve pool can model complex channel distortion, and through training, realize real-time adaptive equalization, and improve the system transmission capacity and distance.

[0029] 2. Complex time series prediction and chaotic system modeling Network traffic prediction: In high-throughput transmission systems, it can be used for short-term and long-term prediction of network traffic waveform, realizing intelligent traffic scheduling and congestion control.

[0030] Nonlinear system simulation: It is suitable for classical nonlinear autoregressive moving average tasks. The rich state space and memory capacity of this scheme can accurately learn and predict high-order nonlinear time series dynamics, and can be applied to financial time series analysis, weather prediction, industrial process monitoring, etc.

[0031] 3. High-speed pattern recognition and classification Real-time image and speech processing: In optical correlator or optical computing architecture, this scheme can be used for high-speed image feature extraction and recognition, speech signal classification, etc. Its parallel optical processing capability can realize low-delay pattern recognition, and is suitable for real-time visual processing in security monitoring and automatic driving.

[0032] Radar and wireless signal classification: In electronic countermeasures and spectrum sensing, it can be used for real-time classification and identification of complex modulated radar signals or communication signals.

[0033] 4. Edge computing and intelligent sensing Distributed fiber sensing: Integrated in fiber sensing network, it realizes localized intelligent processing and event classification of sensing signals such as vibration, sound wave and temperature (such as pipeline leakage detection and perimeter security intrusion identification). The simplified structure of this scheme is conducive to realizing low-power and embedded sensing and calculation integrated nodes.

[0034] Intelligent terminal of Internet of Things: In Internet of Things devices with limited power consumption and volume, it can be used as a dedicated artificial intelligence acceleration unit to realize device state predictive maintenance, anomaly detection and other edge intelligent tasks.

[0035] This scheme has realized the joint monitoring of modulation format and optical signal-to-noise ratio in coherent optical communication system through numerical simulation. This scheme can be applied to the prediction module of high-throughput transmission system (such as network traffic waveform prediction). This application provides a hardware-simplified, high-performance deep optoelectronic reservoir computing general platform, which can complete complex calculations at the speed of light while significantly reducing system power consumption and hardware cost. This technology is suitable for next-generation optical communication, edge artificial intelligence, real-time prediction system and other fields.

[0036] Artificial intelligence technology requires a very complex training process. The connection weights between neurons in the reservoir computing system and the input connection weights are fixed, only the output connection weights need to be trained to determine, and the training process can be completed with a simple linear algorithm, so the training process is greatly simplified, overcoming the problem of neural network training difficulty and time-consuming. Optical reservoir not only has the advantages of reservoir computing, but also has the advantages of low power consumption and high energy efficiency of photonic neural networks. Therefore, optical reservoir computing can provide new solutions for signal performance monitoring and equalization schemes in optical fiber communication systems.

[0037] Optical reservoir is a method of implementing reservoir computing using photonic technology, which is currently mainly divided into all-optical reservoir and optoelectronic reservoir. All-optical reservoir has the advantages of simple structure, low noise, high performance, low energy consumption and high speed. Based on the structure of a single feedback loop, the nonlinear function of master-slave laser injection, semiconductor amplifier, optical fiber cavity, etc. is used to construct a feedback loop containing nonlinear nodes. The current optoelectronic reservoir computing is mainly based on the structure of a single feedback loop. With the in-depth study of optoelectronic reservoir computing, researchers have begun to focus on deep optoelectronic reservoir computing architecture and its implementation scheme, which facilitates the design and integration of subsequent deep optoelectronic reservoir chips.

[0038] Traditional electric domain neural networks have significant defects in signal processing and computing. Neural network training and inference require high-performance hardware, consume a large amount of computing resources, and have high energy consumption. Electrical signal transmission is not only susceptible to electromagnetic interference, but also has large delay in long-distance transmission and complex networks, making it difficult to adapt to high-frequency real-time processing scenarios. Photonic neural networks perform signal processing and computation in the optical domain, with high bandwidth, high interconnectivity, parallel processing, high computation speed and energy efficiency, etc. It can accelerate part of the operation of software and hardware, and the speed can reach the speed of light. It is considered to be a competitive solution to overcome the high power consumption and electronic bottleneck problems of traditional digital signal processing and artificial intelligence. Traditional reservoir computing relies on high-performance electronic hardware, and its signal transmission is limited by circuit delay, with processing speed usually below GHz, and high energy consumption in large-scale matrix operations. Electronic hardware is limited by physical wiring and cannot achieve high integration and high scalability in complex dynamic signal processing scenarios.

[0039] Deep optoelectronic reservoir computing system is introduced by introducing multiple reservoir layers, designing the connection mode between multiple reservoir layers, expanding the scale of reservoir layers, and obtaining higher memory capacity and richer dynamic characteristics. When the number of reservoir nodes is constant, the introduction of multi-layer structure reduces the mutual connection between layers in the reservoir, and reduces the number of reservoir nodes in each layer. Therefore, the multi-layer reservoir computing system has the advantages of low hardware cost and low computational complexity.

[0040] AsFigure 1 As shown, a structural diagram of a deep optoelectronic reservoir computing system model provided by the related art is provided, which takes the state output of the previous layer reservoir as the input of the next layer reservoir, further increasing the state number of the reservoir layer. The deep optoelectronic reservoir computing system model adopts a multi-layer series independent physical reservoir structure, each layer corresponding to a complete processing loop, and the layers are sequentially transmitted through the state signal, including the following components: Input / input signal (S(t)): as the input of the whole model.

[0041] First layer reservoir (Layer 1): including a nonlinear node (Nonlinear node), usually an optical modulator, realizing electro-optical conversion and nonlinear modulation. The node output is connected to an optical delay loop, and the loop delay is marked as T1. The end of the loop is converted into an electrical signal through a photodetector. The output is the first layer state signal X1(t), and directly as the input of the second layer.

[0042] Figure 1 The bias θ is marked, which is used to set the operating point of the modulator.

[0043] Second layer reservoir (Layer 2): the structure is completely symmetrical with the first layer, including a nonlinear node, a delay loop (delay T2), and a photodetector. X1(t) is received as the input. The output is the second layer state signal X2(t), and is transmitted to the subsequent layer.

[0044] Subsequent layer (Layer N): can be expanded to the Nth layer, and each layer has the same structure. Here, it is not listed one by one.

[0045] Output layer (Output Layer): all layer state signals X1(t), X2(t), …, X N (t) are collected.

[0046] Linearly combined through a weight / output weight matrix (W).

[0047] Finally, the output / output signal (Output) of the system is generated.

[0048] In addition, each nonlinear node includes f(·) inside, representing a nonlinear activation function. The whole model clearly shows that the traditional deep optoelectronic reservoir needs to independently equip modulators, delay loops, detectors and corresponding bias and scaling parameters for each layer, resulting in many hardware components, complex structure, cumulative noise and inconsistent delay introduced by layer-by-layer signal processing.

[0049] The state of the node in the deep reservoir computing can be represented as shown in formula 1: Formula 1 Where, x n (t) represents a state node, i.e., the state of the nth layer at (continuous) time t; A n γ represents the random weights between virtual nodes, specifically the connection weight matrix between virtual nodes within the nth layer (typically fixed, sparse, and random); n is the input coefficient for each layer, i.e., the scaling factor of the input signal of the nth layer; f(·) is the nonlinear function / nonlinear activation function; S(t) is the input signal, i.e., the (continuous) time signal input to the system; n represents the number of layers; T represents the loop delay or system time constant of the nth layer reservoir.

[0050] In this embodiment of the application, to avoid the structural complexity caused by multiple physical layers, a simplified structure based on multiplexing technology is proposed. By sharing related optoelectronic devices, wavelength division multiplexing introduces coupling between different reservoir layers. This scheme can effectively reduce the operational complexity of the system and reduce the performance degradation caused by system noise. Specifically, it is a simplified deep optoelectronic reservoir computing system (i.e., optoelectronic reservoir computing device). The deep optoelectronic reservoir model based on wavelength division multiplexing technology expands the scale of the reservoir and breaks the constraint between the complexity and performance of traditional large-scale single-layer reservoir hardware implementation.

[0051] The embodiments provided in this application will now be described in detail with reference to the accompanying drawings.

[0052] The photoelectric reservoir calculation method provided in this application embodiment can be applied to, for example... Figure 2 The application environment shown. For example... Figure 2 As shown, the application environment includes a photoelectric reservoir computing device 101 and a front-end device 102. The photoelectric reservoir computing device 101 and the front-end device 102 are interconnected.

[0053] In some embodiments, the photoelectric storage pool computing device 101 may be a server cluster consisting of multiple servers, a single server, a computer, or a processor or processing chip in a server or computer, etc. This application does not limit the specific device form of the photoelectric storage pool computing device 101. Figure 2 The Sino-Israeli photoelectric storage pool computing device 101 is shown as an example of a single server.

[0054] In some embodiments, the front-end device 102 can be a device with wireless transceiver function, such as a mobile phone, a tablet computer, a wearable device, a vehicle-mounted device, an augmented reality (AR) / virtual reality (VR) device, a notebook computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc. The embodiments of the present application do not limit the specific device form of the front-end device 102. Figure 2 The front-end device 102 is taken as a mobile phone as an example.

[0055] In some embodiments, the optoelectronic reservoir computing device 101 generates a multi-feature optical signal through an internal light source module, combines the multi-feature optical signal into the same optical path through a multiplexer, and then couples the input electrical signal (such as real-time collected sensor data, to-be-predicted time series signal, or to-be-recognized image feature vector) from the front-end device 102 (such as a mobile phone) to the optical path through a reservoir processing module, and performs optoelectronic processing including cyclic delay, photoelectric conversion, amplification and filtering, and state solving, and finally generates and outputs a reservoir state signal containing multiple state components or a final result after further processing. The front-end device 102 can send the original input electrical signal to the optoelectronic reservoir computing device 101, receive and analyze the returned calculation result, and perform parameter adjustment or trigger subsequent application logic according to the feedback of the optoelectronic reservoir computing device 101, such as displaying the prediction result on the mobile phone, executing the control instruction, or updating the user interface.

[0056] It should be noted that the system architecture described in the embodiments of the present application is for more clearly illustrating the technical solutions of the embodiments of the present application, and does not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems as the system architecture evolves.

[0057] Referring to Figure 3 , a composition schematic diagram of an optoelectronic reservoir computing device provided by the embodiments of the present application is shown. As Figure 3 shown, the optoelectronic reservoir computing device 20 provided by the embodiments of the present application can include a light source module 21, a multiplexer 22, and a reservoir processing module 23 connected in sequence.

[0058] The light source module 21 is configured to generate at least two optical signals with different characteristics.

[0059] The multiplexer 22 is configured to multiplex the at least two optical signals into the same optical transmission path to obtain a multiplexed optical signal.

[0060] The reserve pool processing module 23 is configured to receive an input electrical signal to be processed, and couple the input electrical signal to a multiplexed optical signal, and perform optoelectronic processing on the coupled optical signal to obtain and output a reserve pool state signal, the reserve pool state signal comprising at least two state components, the at least two state components being associated with different characteristics of the at least two optical signals, respectively.

[0061] In some embodiments, the optoelectronic reserve pool computing device 20 physically combines optical devices and electronic devices to form an optoelectronic hybrid computing system, which works by using the high speed, parallelism and low power consumption of light to perform nonlinear transformation and delay memory of signals, and using mature electronic technology to perform signal amplification, digitization and intelligent algorithm processing.

[0062] In some embodiments, the light source module 21 comprises a coherent light source array or an optical frequency comb, which is mainly used to generate a plurality of laser signals as optical carriers.

[0063] In some embodiments, the different characteristics of the optical signals include at least one of the following: different wavelengths, different polarization states, different spatial modes.

[0064] In some embodiments, the light source module 21 is a light signal generation unit, which is configured to generate at least two optical signals with different characteristics. These different characteristics are the physical basis for constructing a logically multiple parallel processing channel (i.e. a deep reserve pool layer), and the characteristics of the optical signals are the physical dimensions that can be used to distinguish and independently modulate during propagation. Specific implementation methods include but are not limited to: Using a coherent light source array: for example, using a plurality of distributed feedback lasers or external cavity lasers, each laser outputs a specific and stable wavelength, and through precise temperature and current control, the interval and stability between wavelengths can be ensured to meet the requirements of wavelength division multiplexing; Using an optical frequency comb: a more integrated solution is to use an optical frequency comb source, which can generate a series of equally spaced and phase-coherent frequency lines in the frequency spectrum, each frequency line can be used as an independent optical carrier. This method has a more compact light source structure, is easy to generate a large number of optical signals with different characteristics (wavelengths), and is very suitable for constructing a large-scale deep reserve pool.

[0065] In some embodiments, the multiplexer 22 comprises any one of the following: a wavelength division multiplexer, a polarization multiplexer, a space division multiplexer.

[0066] The polarization multiplexer is used to multiplex at least two optical signals with orthogonal polarization states into the same optical transmission path. The space division multiplexer is used to multiplex at least two optical signals from different spatial channels or different spatial modes into the same optical transmission path.

[0067] In some embodiments, the above-mentioned wavelength division multiplexer is mainly to combine multiple signals into one optical signal through wavelength division multiplexing.

[0068] In some embodiments, the above-mentioned polarization multiplexer is mainly to combine at least two signals with orthogonal polarizations into one optical signal through polarization multiplexing, which can form a reserve pool system with two layers of depth.

[0069] In some embodiments, the above-mentioned space division multiplexer can be a coupler or an optical mode division multiplexer, which divides the optical signal into multiple paths, modulates the multiple optical signals through parallel modulators, uses multi-core or multi-mode optical fiber to realize time delay of the multiple signals, and uses an array detector to realize detection of the multiple signals.

[0070] It can be understood that the multiplexing technology used by the optoelectronic reserve pool computing device can include wavelength division multiplexing, polarization multiplexing, mode division multiplexing, space division multiplexing, etc.

[0071] In some embodiments, the above-mentioned multiplexer 22 multiplexes at least two optical signals with different characteristics into the same optical transmission path to obtain a composite multiplexed optical signal, which physically realizes the mapping of multiple parallel logical channels into a single physical transmission medium (such as a single-mode optical fiber). The specific implementation depends on the characteristics of the optical signal: If the characteristics are different wavelengths, the multiplexer 22 is a wavelength division multiplexer, which uses arrayed waveguide gratings, thin film filters or fiber gratings, etc. to guide optical signals of different wavelengths from different physical ports to the same output port. At the output, all wavelengths of light are completely overlapped in space and time, but can be distinguished in the frequency domain; If the characteristics are different polarization states (such as orthogonal TE and TM modes), the multiplexer 22 is a polarization multiplexer, which combines two optical signals with orthogonal polarization states into the same optical fiber through a polarization maintaining coupler or a polarization beam combiner. The two signals propagate independently in the optical fiber and can be perfectly separated through subsequent polarization diversity reception technology; If the characteristics are different spatial modes (such as LP01, LP11, etc. in optical fiber) or different spatial channels (such as different cores in multi-core optical fiber), the multiplexer 22 is a space division multiplexer or a mode division multiplexer, which uses a special fiber coupler to couple multiple spatially separated optical signals into different channels of a multi-core optical fiber or a multi-mode optical fiber.

[0072] In this way, the above-mentioned multiplexing mode reduces the number of optical devices required for the subsequent processing link. Related technologies require N sets of parallel modulators, delay lines and detectors to process N channels. This scheme only needs one set of common devices to process the multiplexed optical signal, thereby reducing the hardware complexity, volume and cost of the system.

[0073] Thus, wavelength division multiplexing utilizes the wavelength dimension, polarization multiplexing utilizes the polarization state orthogonality, and spatial division multiplexing utilizes the spatial channel or mode. Different multiplexing methods can adapt to different system requirements (such as bandwidth, integration, cost), for example, wavelength division multiplexing is easy to expand the number of wavelengths to achieve a deeper logical layer, while polarization multiplexing may have a simpler structure, and spatial division multiplexing points to the high-integration chip prospect.

[0074] In some embodiments, the reservoir processing module 23 is responsible for loading the input electrical information onto the optical path, utilizing optical nonlinearity and delay feedback to generate rich dynamic states, and finally converting back to electrical signals for processing.

[0075] In some embodiments, in combination with Figure 3 As shown in Figure 4 The above-mentioned reservoir processing module 23 includes a polarization controller 230, an optical modulator 231, a common optical delay loop 232, a photodetector 233, and a signal processing unit 234.

[0076] The polarization controller 230 is connected to the multiplexer 22 and is used to adjust the polarization of the multiplexed optical signal.

[0077] The optical modulator 231 is connected to the polarization controller 230 and is used to receive the input electrical signal and couple the input electrical signal to the polarization-adjusted multiplexed optical signal, outputting the modulated optical signal.

[0078] The common optical delay loop 232 is connected to the optical modulator 231 and is used to cyclically transmit the modulated optical signal to increase the transmission delay of the modulated optical signal.

[0079] The photodetector 233 is connected to the common optical delay loop 232 and is used to convert the cyclically transmitted optical signal into a first electrical signal.

[0080] The signal processing unit 234 is connected to the photodetector 233 and is used to process the first electrical signal to generate and output the reservoir state signal.

[0081] In some embodiments, the polarization controller 230 can adjust the polarization state of the optical signal to improve the modulation depth.

[0082] In some embodiments, the polarization controller 230 is connected after the output of the multiplexer 22, and its function is to adjust the polarization state of the multiplexed optical signal. Since the modulation efficiency of many optical modulators is related to the polarization state of the input light, optimizing the polarization state can improve the modulation depth, thereby enhancing the system nonlinear response strength. This can be achieved through manual or electrically controlled wave plates, polarization rotators.

[0083] In some embodiments, the optical modulator 231 is used to implement photoelectric conversion, i.e. modulating an electrical signal onto an optical carrier.

[0084] In some embodiments, the optical modulator 231 is a key device for implementing electro-optical coupling (i.e. coupling an input electrical signal onto a multiplexed optical signal) in the present scheme, which can be a broadband Mach-Zehnder intensity modulator. The implementation process is as follows: the multiplexed optical signal from the polarization controller 230 is used as the optical input of the modulator; the input electrical signal s(t) (possibly containing feedback electrical signals) to be processed is used as the RF driving signal of the modulator, which is applied to its electrodes. Through this nonlinear modulation process, the time-varying information of the input electrical signal is encoded into the intensity of the optical carrier of all wavelengths / polarization characteristics, and the modulated optical signal is output. It is worth noting that the optical carriers of different characteristics (such as wavelengths) are modulated by the same s(t) at the same time, but because they share the same nonlinear transfer function, their modulation depth and the resulting nonlinear dynamics are synchronized and correlated.

[0085] In some embodiments, the common optical delay loop 232 is connected after the modulator 231, which is usually a ring cavity composed of a section of single-mode fiber, and its main function is to introduce a fixed transmission time delay T. The value of T is determined by the length L of the fiber and the propagation speed c / n of light in the fiber (T=nL / c). The modulated optical signal propagates in this common optical delay loop 232, and each cycle introduces a time delay T. This makes the optical signal entering the loop at the current time interfere or nonlinearly interact with the historical optical signal that has been in the loop for T time and has experienced a cycle of the loop before reaching the modulator again (through feedback). The common optical delay loop 232 makes all different characteristic optical signals experience the same transmission time delay T in the same physical fiber loop, ensuring the consistency of the time scale of the dynamics of each channel.

[0086] In some embodiments, the photodetector 233 can convert the optical signal into an electrical signal.

[0087] In some embodiments, the photodetector 233 is arranged at an output of the optical loop (usually a coupler is used to couple part of the optical power out of the loop), and its function is to convert the optical signal after the cycle transmission (i.e. the multiplexed optical signal containing the delay and historical information) into a first electrical signal. This is a photoelectric conversion process, and the conversion efficiency is determined by the responsivity of the detector (the photocurrent generated per unit optical power) and the matching resistance of the subsequent transimpedance amplifier, and their product is the parameter σ. The detector outputs an analog current or voltage signal, which integrates the optical intensity information of all wavelength / polarization channels.

[0088] In some embodiments, the signal processing unit 234 is connected to the photodetector 233 and is responsible for further processing the analog first electrical signal to ultimately generate and output a digital reservoir status signal.

[0089] Thus, the reservoir processing module includes a polarization controller, an optical modulator, a shared optical delay loop, a photodetector, and a signal processing unit. The polarization controller optimizes the modulation depth, the optical modulator realizes the loading (coupling) of electrical signals to optical carriers, the shared optical delay loop is the core of generating the dynamic characteristics (memory and recursion) required for the reservoir, the photodetector completes the conversion of optical signals to electrical signals, and the signal processing unit is responsible for the final extraction of information, so as to realize the photoelectric processing function through these modules.

[0090] In some embodiments, combined with Figure 4 ,like Figure 5 As shown, the signal processing unit 234 includes an electric amplifier 2340, an analog-to-digital converter 2341, and a digital signal processor 2342.

[0091] The aforementioned electrical amplifier 2340 is connected to the photodetector 233 to amplify the first electrical signal and obtain an analog electrical signal.

[0092] The analog-to-digital converter 2341 is connected to the electrical amplifier 2340 and is used to convert analog electrical signals into digital signals.

[0093] The aforementioned digital signal processor 2342 is connected to the analog-to-digital converter 2341 and is used to extract at least two state components from the digital signal to generate and output the reservoir state signal.

[0094] In some embodiments, the above-described electrical amplifier 2340 is used to amplify electrical signals and adjust the gain.

[0095] In some embodiments, the aforementioned electrical amplifier 2340 is used to amplify the first electrical signal, and its target gain G is an adjustable parameter used to adjust the signal amplitude to a range suitable for subsequent analog-to-digital conversion. Meanwhile, the amplifier itself or the subsequent independent filter will have a specific frequency response, typically exhibiting a bandpass characteristic, defined by the lower cutoff frequency f. L and the upper cutoff frequency f H The definition is used to suppress low-frequency drift and high-frequency noise, and to shape the effective bandwidth of the system.

[0096] In some embodiments, the analog-to-digital converter 2341 described above can convert analog signals into digital signals, facilitating subsequent digital signal processing.

[0097] In some embodiments, the analog-to-digital converter 2341 described above converts the amplified and filtered analog electrical signal (second electrical signal) into a digital signal to facilitate precise digital signal processing.

[0098] In some embodiments, the digital signal processor 2342 calculates the output weights using algorithms such as ridge regression, linear regression, or adaptive gradient.

[0099] In some embodiments, the digital signal processor 2342 described above can extract at least two state components from the digital signal. Since the photodetector 233 outputs a mixed signal from all channels, the digital signal processor 2342 needs to use a digital signal processing algorithm (e.g., if wavelength division multiplexing is used, digital spectrum analysis or matched filtering techniques) to extract the signal corresponding to each original optical signal feature (such as wavelength λ). n The signal components x1(t), x2(t), ..., x2(t) are separated out. N (t) represents the reservoir state signals, which are associated with different characteristics of the original optical signals that generated them. State components x n The generation of (t) is not a simple separation; its amplitude is determined by a formula (such as Formula 2 in the following embodiment) that includes all physical parameters of the system. Another key task of the digital signal processor 2342 is to execute training and inference algorithms, such as ridge regression, linear regression, or adaptive gradient algorithms, to calculate the output layer weights W using the state component sequence. i And finally, the system's final output for a specific task (such as prediction) is obtained through linear combination (as shown in Formula 3 in the following embodiments).

[0100] In some embodiments, such as Figure 5 As shown, the output terminal of the aforementioned electrical amplifier 2340 is connected to the electrical input terminal of the optical modulator 231, and is used to input a portion of the electrical signal in the analog electrical signal as a feedback electrical signal to the optical modulator 231.

[0101] In some embodiments, a portion of the analog electrical signal output from the electrical amplifier 2340 is fed back to another electrical input of the optical modulator 231 via a physical connection. This introduces optoelectronic hybrid feedback, where the feedback signal is an amplified representation of the system state from time T prior, which is superimposed on the new input signal s(t) at the modulator. This mechanism ensures that the system's current dynamics depend not only on the current input but also strictly on its own historical state, thus constituting a nonlinear self-recursive dynamical system. This enriches the system's dynamic characteristics and enables reservoir computing to handle complex time-series tasks.

[0102] It should be noted that the optical output end of the multiplexer 22 is connected to the optical input end of the polarization controller 230, the optical output end of the polarization controller 230 is connected to the optical input end of the optical modulator 231; the optical output end of the optical modulator 231 is connected to the input end of the common optical delay loop 232, the output end of the common optical delay loop 232 is connected to the optical input end of the photodetector 233; the electrical output end of the photodetector 233 is connected to the input end of the electrical amplifier 2340, the output end of the electrical amplifier 2340 is connected to the analog input end of the analog-to-digital converter 2341 in one way, and connected to the electrical input end of the optical modulator 231 in another way through a feedback path; the digital output end of the analog-to-digital converter 2341 is connected to the data input interface of the digital signal processor 2342, and the final state and result signals are output by the digital signal processor 2342.

[0103] In this way, the electrical amplifier is used to adjust the signal amplitude and match the dynamic range of the subsequent circuit, the analog-to-digital converter digitizes the analog signal, which is convenient for accurate and flexible digital signal processing, and the digital signal processor is responsible for executing complex algorithms to accurately extract multiple state components corresponding to different optical signal characteristics from the digital signal. This structure realizes reliable conversion and intelligent processing of analog optical signals into high-fidelity digital information, ensuring the accuracy of the reserve pool state signal generation and the overall stability of the system.

[0104] In addition, an optoelectronic hybrid feedback mechanism is introduced, the feedback signal is superimposed with the original input signal at the modulator, so that the system state depends not only on the current input, but also on its own historical state (feedback after loop delay), thereby forming a dynamic recursive system. This structure enhances the intrinsic ability of the system to process time sequence signals and nonlinear problems, and is the way in which the reservoir computing is superior to the simple feedforward network. The setting of the feedback path further optimizes and perfects the function of the reservoir processing module.

[0105] The optoelectronic reservoir computing device provided by the application is composed of a light source module, a multiplexer and a reservoir processing module connected in sequence. By generating a plurality of characteristic light signals and multiplexing them into the same transmission path, the physical hardware is highly simplified, so that a plurality of logical processing layers are mapped onto light signals of different characteristics, and unified modulation, delay and conversion are completed in a common physical loop. This architecture avoids the problem of hardware quantity explosion and structural complexity caused by traditional multi-physical layer stacking, and all signals are processed in a shared optical path, eliminating the interlayer signal conversion interface, reducing the additional noise, distortion and inconsistent delay introduced by multiple independent loops, thereby improving the fidelity and consistency of signal processing. In this way, without increasing the complexity of physical hardware and signal degradation, the processing capacity and dynamic characteristics of the system can be effectively expanded by increasing the number of multiplexed light signal characteristics, thereby reducing the complexity and cost of the hardware structure of the deep optoelectronic reservoir computing system and improving the overall computing performance.

[0106] Referring to Figure 6 A flowchart of an optoelectronic reservoir computing method provided by an embodiment of the application is shown in FIG. 2. As shown in FIG. 2, the optoelectronic reservoir computing method provided by the embodiment of the application can be implemented by the above-mentioned optoelectronic reservoir computing device, and specifically includes the following steps 201 to 203. Figure 6

[0107] Step 201: The optoelectronic reservoir computing device acquires at least two light signals with different characteristics.

[0108] Step 202: The optoelectronic reservoir computing device multiplexes the at least two light signals into the same optical transmission path to obtain a multiplexed light signal.

[0109] Step 203: The optoelectronic reservoir computing device receives an input electrical signal to be processed, and couples the input electrical signal to the multiplexed light signal. The optoelectronic reservoir computing device performs optoelectronic processing on the coupled light signal to obtain a reservoir state signal.

[0110] In some embodiments, the reservoir state signal includes at least two state components, which are respectively associated with different characteristics of the at least two light signals.

[0111] In some embodiments, the step 203 of "performing optoelectronic processing on the coupled light signal to obtain a reservoir state signal" can be implemented as the following steps 203a to 203d.

[0112] Step 203a: The optoelectronic reservoir computing device acquires the transmission delay of the coupled light signal after cyclic transmission.

[0113] ​Step 203b, the optoelectronic reservoir computing device performs photoelectric conversion on the optical signal after the circulation transmission based on the product of the response coefficient of the photodetector and the matching resistance, to obtain a first electrical signal.

[0114] Step 203c, the optoelectronic reservoir computing device amplifies the first electrical signal using a target gain, and filters the amplified electrical signal using a target lower cutoff frequency and a target upper cutoff frequency, to obtain a second electrical signal.

[0115] Step 203d, the optoelectronic reservoir computing device calculates at least two state components based on the transmission time delay, the second electrical signal, the input electrical signal, and a preset parameter group, to generate a reservoir state signal.

[0116] In some embodiments, the preset parameter group includes a half-wave voltage of the optical modulator, and a bias voltage, an input scaling factor, and an initial optical signal intensity corresponding to each optical signal, respectively.

[0117] In this way, photoelectric conversion based on a specific coefficient defines a deterministic relationship for signal conversion from the optical domain to the electrical domain, ensuring the fidelity and repeatability of signal conversion, and using a target gain and a specific cutoff frequency for amplification and filtering to accurately control the signal amplitude and bandwidth, effectively amplifying the useful signal and suppressing the out-of-band noise. Thus, based on a complete set of preset parameters, i.e., a set of measurable and preset physical parameters (half-wave voltage, bias, scaling factor, light intensity, etc.), each state component is calculated, making the calculation more reliable and stable.

[0118] In some embodiments, the step 203d of "the optoelectronic reservoir computing device calculating at least two state components based on the transmission time delay, the second electrical signal, the input electrical signal, and the preset parameter group" can be implemented as steps 203d1 to 203d4 as follows.

[0119] Step 203d1, for each optical signal, the optoelectronic reservoir computing device calculates a merged electrical signal based on the input electrical signal, and the bias voltage, the input scaling factor, and the historical state component before the transmission time delay corresponding to one optical signal.

[0120] In some embodiments, the optoelectronic reservoir computing device can retrieve two parameters dedicated to the channel from the preset parameter group: a bias voltage φ n and an input scaling factor γ n , φ n is used to set the optical modulator at the optimal linear or nonlinear operating point of the channel, and γ n is used to adjust the intensity of the feedback signal and control the strength of the recursive dynamics of the channel. At the same time, the optoelectronic reservoir computing device reads the state value of the channel at T time ago, i.e., the historical state component xn (t-T). T is the transmission delay / physical loop delay. Then, the opto-electronic reservoir computing device performs a calculation: merged electrical signal = s(t) + φ n + γ n x n (t-T). Where s(t) is the input electrical signal at the current time instant. The calculation is implemented in a digital signal processor as a numerical operation, or in the analog domain by a summing circuit.

[0121] Step 203d2, the opto-electronic reservoir computing device performs a nonlinear modulation process on the merged electrical signal based on the half-wave voltage to obtain a modulation intensity coefficient.

[0122] In some embodiments, the opto-electronic reservoir computing device simulates the physical nonlinear effect of the optical modulator, and the half-wave voltage V π is an inherent physical parameter of the modulator, which can be obtained from a preset parameter set. The processing process of the opto-electronic reservoir computing device is: substituting the value of the above-mentioned merged electrical signal into a function describing the nonlinear transmission characteristics of the MZM (Mach-Zehnder modulator), and calculating the modulation intensity coefficient based on V π and the merged electrical signal. The nonlinear function maps the linear electrical signal merged value to the nonlinear optical intensity modulation. The modulation intensity coefficient is a dimensionless value between 0 and 1, which represents the intensity modulation depth of the optical modulator on the optical carrier of the channel under the given merged electrical signal drive. The modulation intensity coefficient of 1 indicates that the light is completely passed (maximum light intensity), and the modulation intensity coefficient of 0 indicates that the light is completely blocked.

[0123] Step 203d3, the opto-electronic reservoir computing device calculates a first optical intensity value based on the initial optical signal intensity corresponding to the optical signal and the modulation intensity coefficient.

[0124] In some embodiments, the initial optical signal intensity I0 is the original optical power (or amplitude) of the channel optical carrier before entering the modulator, which is a preset and calibratable parameter. The first optical intensity value = I0 modulation intensity coefficient, which is completed in a digital signal processor. The first optical intensity value represents the instantaneous optical power that the optical signal of the specific channel should have at the output end of the modulator in an ideal case (without considering transmission loss and detector response), which reflects the energy encoding result of the input information (after nonlinear modulation) on the wavelength / polarization channel.

[0125] Step 203d4, the opto-electronic reservoir computing device calculates a state component corresponding to the optical signal based on the product of the first optical intensity value and the response coefficient and the matching resistance, to obtain at least two state components.

[0126] In some embodiments, the optoelectronic reservoir computing device can multiply the first light intensity value by the product of the responsivity of the photodetector and the matching resistance σ, which is a fixed conversion factor from optical power (watt) to the voltage at the output of the detector (volt) determined by the physical properties of the detector. This multiplication of the first light intensity value σ models the conversion process of the photodetector and results in a theoretical initial voltage signal. This signal then needs to be processed by the backend electronics of the system, i.e. multiplied by the target gain G (analog electrical amplifier) and passed through a band-limited filter operation defined by the target lower cutoff frequency f L and the target upper cutoff frequency f H . These operations are implemented in the digital signal processor through digital filters and multiplications. The resulting value is the state component x n (t) of the optical signal channel corresponding to the current time t. Repeating the above steps 203d1 to 203d4 for all optical signal channels (n = 1, 2,..., N) results in the complete set of at least two state components {x1(t), x2(t),..., x N (t)} which together constitute the multi-dimensional reservoir state of the system.

[0127] In some embodiments, the optoelectronic reservoir computing device can load the input signal s(t) = s in (t) + s m (t) into the optoelectronic reservoir computing device, where s in (t) is the original input signal and s m (t) is a mask signal. The output state signal of the reservoir can be obtained as shown in Equation 2 below: Equation 2 where G is the gain of the electrical amplifier; σ is the product of the responsivity of the photodetector and the matching resistance; x n (t) is the optical signal in the loop at the state node; s(t) is the input electrical signal to be processed; φ n is the bias voltage; f L is the lower cutoff frequency; f H is the upper cutoff frequency; n is the number of layers; I0is the intensity of the optical signal; T is the loop / transmission delay; V π is the half-wave voltage of the modulator; γ n is the scaling factor of the input signal.

[0128] In some embodiments, the optoelectronic reservoir computing device can sum the output states of the reservoir to obtain the output signal y(n), which is shown in Equation 3 below: Equation 3 where W iis the weight of the output layer.

[0129] An embodiment, by the above device, can constitute an optoelectronic reservoir computing device. Taking the Nonlinear Auto Regressive Moving Average model of order 10 (NARMA10) task as an example, the reservoir computing device can perform nonlinear task prediction. The 10th order moving average is shown in the following formula 4: Formula 4 Wherein, the NARMA10 is a strong nonlinear, long-time dependent system identification task. The target output depends on the history values of the previous 10 time points and the input u(n). The input signal u(n) is usually a uniformly distributed random sequence in the interval [0, 0.5], that is, the input signal u(n) is uniformly distributed in [0, 0.5]. This task is usually used to judge the nonlinear and chaotic state of the reservoir. According to the above formula 4, the output of n=1-50 is calculated. The 50 inputs and 50 outputs are used as the input and output of the reservoir computing device for training and solving the output weight W i . Then, by inputting the 51st input, the reservoir can obtain the 51st output through the above formula 2 and formula 3, realizing the learning and prediction of the 10th order moving average model.

[0130] It can be understood that the training stage is: inputting the sequence u(1) to u(50) as the input electrical signal s(t) into the optoelectronic reservoir computing device. The optoelectronic reservoir computing device works according to the above process, and the digital signal processor records the state component sequence x n (1),...,x n (50) generated at each time point. At the same time, the corresponding target output sequence y(1),...,y(50) is calculated according to formula 4. Then, the digital signal processor uses a linear regression algorithm (such as ridge regression) to solve a set of optimal output weights W i with the state sequence as the input and the target sequence as the expected output. This process is fitting and training.

[0131] The prediction stage: after training, the weight W i is fixed. When the 51st input u(51) arrives, the optoelectronic reservoir computing device runs again, generates a new state component x n (51) according to formula 2, and then calculates the predicted output y(51) through formula 3. By comparing the predicted output y(51) with the true y(51), the learning and prediction ability of the optoelectronic reservoir computing device for the complex nonlinear task can be evaluated.

[0132] It should be noted that the specific description of the optical-electric reserve pool calculation device, the implementation principle of the optical-electric reserve pool calculation device, and each module / unit in the optical-electric reserve pool calculation device involved in the optical-electric reserve pool calculation method can be referred to the description in the above embodiments, which will not be repeated here.

[0133] In the optical-electric reserve pool calculation method provided by the application, the multi-feature optical signal is obtained and multiplexed to the same path, avoiding the structural complexity and cost brought by constructing independent hardware for each channel. The input electrical signal is coupled to the multiplexed optical path and subjected to optical-electric processing, ensuring that the signals of all channels experience uniform high-fidelity processing in the shared physical medium. This not only eliminates the distortion and inconsistent delay introduced by repeated conversion and cross-connection in traditional multi-channel systems, but also guarantees the high efficiency of the processing process. Finally, the reserve pool state signal containing multiple feature-related state components is output. In this way, without increasing the physical hardware complexity and signal degradation, the processing capacity and dynamic characteristics of the system can be effectively expanded by increasing the number of multiplexed optical signal features, thereby reducing the complexity and cost of the hardware structure of the deep optical-electric reserve pool calculation system and improving the overall computing performance.

[0134] The optical-electric reserve pool calculation method of the embodiments of the application will be introduced below with reference to a specific embodiment. As shown in Figure 7 , it is a schematic diagram of the overall implementation process of the optical-electric reserve pool calculation method provided by the embodiments of the application. The specific process is as follows S1 to S7: S1, initialization and signal preparation: the optical-electric reserve pool calculation device is powered on and initialized, the light source module (such as a coherent light source array or an optical frequency comb) is started, and at least two optical signals with different features (such as different wavelengths λ1, λ2, …, λn or different polarization states) are generated; at the same time, a preset parameter group (including the half-wave voltage Vπ of the optical modulator, the bias voltage φ of each optical signal, the input scaling factor γ, the initial optical signal intensity I0, and the preset transmission time delay T, the target gain G of the electrical amplifier, the product σ of the response coefficient and the matching resistance of the photoelectric detector, the target cutoff frequency f1 and f2, etc.) is loaded into the storage or configuration unit of the optical-electric reserve pool calculation. N π n n L H

[0135] S2, optical signal multiplexing: the multiplexer multiplexes the generated multiple optical signals to the same optical transmission path (such as the same single-mode optical fiber), forming a multiplexed optical signal containing multiple feature components.

[0136] ​​​​​​S3, input signal receiving and polarization adjustment: the optoelectronic reservoir computing device receives the input electrical signal s(t) from outside; meanwhile, the multiplexed optical signal is first sent to a polarization controller to adjust the polarization state, so as to optimize the modulation efficiency of the subsequent optical modulator.

[0137] S4, optoelectronic coupling and nonlinear modulation: the multiplexed optical signal after polarization adjustment is input to the optical input end of the optical modulator (MZM), and the input electrical signal s(t) is applied to the electrical input end of the MZM. In the MZM, s(t) changes the phase of the optical wave through the electro-optic effect, and is converted into modulation of the optical intensity through the interference structure, which adopts a nonlinear transfer function characterized by a half-wave voltage V π ; the information of the input electrical signal is coupled and nonlinearly modulated onto the intensity of all the characteristic optical carriers, and the output is the modulated optical signal.

[0138] S5, cyclic delay and optoelectronic conversion: the modulated optical signal is injected into a common optical delay loop (such as an optical fiber ring), and the optical signal is transmitted in the common optical delay loop, and a fixed transmission delay T is introduced each time. A part of the optical power in the common optical delay loop is coupled out to the photodetector; the photodetector converts the optical signal after delay processing into an analog first electrical signal, and the conversion efficiency is determined by the coefficient σ.

[0139] S6, electrical signal conditioning and digitization: the first electrical signal is amplified by an electrical amplifier with a target gain G, and is band-limited filtered by a filter with a cutoff frequency of f L and f H , to obtain a second electrical signal; then an analog-to-digital converter converts the second electrical signal into a digital signal, which is sent to a digital signal processor.

[0140] S7, state component calculation: the digital signal processor performs the following parallel or serial calculation for each optical signal characteristic (index n) to calculate the state component x n (t).

[0141] Wherein, S7 is implemented as follows: Data preparation: obtain the current input s(t), the preset φ n , γ n , I0 and historical state x n (t-T) of the channel; Merging calculation: calculate the merged electrical signal V merge : V merge =s(t)+φ n +γ n x n (t-T); Nonlinear mapping: based on V π, calculate the modulation intensity coefficient Mod Index ; Theoretical light intensity calculation: calculate the first light intensity value I theory : theory = I0 Mod Index ; State component generation: calculate the current state: x n (t) = G σ H(f L ,f H )[I theory ], where H represents the filtering operation. Calculate for all n to obtain the state component set {x1(t), x2(t), …, x n (t)}, i.e. the reservoir state signal.

[0142] Part of the analog signal output by the electrical amplifier is sent back to the electrical input of the optical modulator as a feedback electrical signal, combined with the input at the next time to form a recursion.

[0143] Therefore, after performing the above S1 to S7, the optoelectronic reservoir computing device can output the final signal, specifically as follows: Training phase: input a training signal sequence to the optoelectronic reservoir computing device, and repeatedly obtain the corresponding state sequence by S3-S7. Take the state sequence as input, and take the expected output sequence calculated according to the task model as target, and train a set of optimal output weights W i by linear regression algorithm (such as ridge regression).

[0144] Inference / prediction phase: for a new input signal, the system performs S3-S7 to obtain the current state component, and then calculates the final output signal (such as prediction value) of the system by linear combination. The result is output to complete the specific application task (such as signal prediction, classification, etc.).

[0145] It should be noted that the description of each step S1 to S7 of the present embodiment can refer to the description in the above embodiment, which will not be repeated here.

[0146] It can be seen that the above mainly introduces the scheme provided by the embodiments of the present application from the perspective of method and device. In order to realize the above functions, the embodiments of the present application provide corresponding hardware structure and / or software module for executing each function. Those skilled in the art should easily realize that, in combination with the modules and algorithm steps of the examples described in the embodiments disclosed in the present application, the embodiments of the present application can be realized in the form of hardware or the combination of hardware and computer software. Whether a certain function is realized in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0147] The embodiments of the present application provide an electronic device, as shown in Figure 8 The electronic device 80 provided by the embodiments of the present application includes the above-mentioned optoelectronic reservoir computing device 20.

[0148] It should be noted that the specific description of the optoelectronic reservoir computing device can be referred to the description in the above embodiments, which will not be repeated here.

[0149] In the case of realizing the functions of the above-mentioned integrated modules in the form of hardware, the embodiments of the present application provide a possible structure schematic diagram of the electronic device involved in the above-mentioned embodiments. As shown in Figure 9 The electronic device 90 includes a processor 92, a communication interface 93 and a bus 94. Optionally, the electronic device 90 can further include a memory 91.

[0150] The processor 92 can be various exemplary logical blocks, modules and circuits described in combination with the disclosure of the present application. The processor 92 can be a central processing unit, a general purpose processor, a digital signal processor, an application specific integrated circuit, a field programmable gate array or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It can realize or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure of the present application. The processor 92 can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0151] The communication interface 93 is used to connect with other devices through a communication network. The communication network can be Ethernet, wireless access network, wireless local area network (WLAN) and the like.

[0152] The memory 91 can be a Read-Only Memory (ROM) or other type of static storage device that can store static information and instructions, a Random Access Memory (RAM) or other type of dynamic storage device that can store information and instructions, an Electrically Erasable Programmable Read-Only Memory (EEPROM), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to this.

[0153] As a possible implementation, the memory 91 can exist independently of the processor 92, and the memory 91 can be connected to the processor 92 through the bus 94 for storing instructions or program codes. When the processor 92 invokes and executes the instructions or program codes stored in the memory 91, the optoelectronic reserve pool calculation method provided by the embodiments of the present application can be implemented.

[0154] In another possible implementation, the memory 91 can also be integrated with the processor 92.

[0155] The bus 94 can be an Extended Industry Standard Architecture (EISA) bus or the like. The bus 94 can be divided into an address bus, a data bus, a control bus, and the like. For the sake of convenience and brevity, Figure 9 Only one thick line is used in the figure to represent the bus, but it does not mean that there is only one bus or only one type of bus.

[0156] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the above division of functional modules is taken as an example for illustration, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the service calling device is divided into different functional modules to complete all or part of the functions described above.

[0157] The embodiments of the present application further provide a computer readable storage medium. All or part of the processes in the method embodiments above can be instructed by computer instructions to complete relevant hardware, and the program can be stored in the computer readable storage medium. When the program is executed, the program can include the processes of the method embodiments above. The computer readable storage medium can be the memory of any of the foregoing embodiments. The computer readable storage medium can also be an external storage device of the service calling device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the computer readable storage medium can include both the internal storage unit of the service calling device and the external storage device. The computer readable storage medium is used to store the computer program and other programs and data required by the service calling device. The computer readable storage medium can also be used to temporarily store data that has been output or will be output.

[0158] The embodiments of the present application further provide a computer program product, which contains a computer program, and when the computer program product runs on a computer, the computer executes any one of the optoelectronic reservoir computing methods provided in the embodiments above.

[0159] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this. Any change or replacement within the technical scope disclosed in the present application should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A photoelectric storage tank computing device, characterized in that, include: The light source module, multiplexer, and reservoir processing module are connected in sequence. The light source module is used to generate at least two optical signals with different characteristics; The multiplexer is used to multiplex the at least two optical signals to the same optical transmission path to obtain multiplexed optical signals. The reservoir processing module is used to receive the input electrical signal to be processed, couple the input electrical signal to the multiplexed optical signal, perform photoelectric processing on the coupled optical signal, obtain and output the reservoir status signal, the reservoir status signal includes at least two status components, and the at least two status components are respectively associated with different characteristics of the at least two optical signals.

2. The photoelectric storage pool computing device according to claim 1, characterized in that, The multiplexer includes any one of the following: wavelength division multiplexer, polarization multiplexer, and space division multiplexer; The polarization multiplexer is used to multiplex at least two optical signals with orthogonal polarization states to the same optical transmission path; the spatial multiplexer is used to multiplex at least two optical signals from different spatial channels or different spatial modes to the same optical transmission path.

3. The photoelectric storage tank computing device according to claim 1, characterized in that, The reservoir processing module includes: a polarization controller, an optical modulator, a shared optical delay loop, a photodetector, and a signal processing unit; The polarization controller is connected to the multiplexer and is used to adjust the polarization of the multiplexed optical signal; The optical modulator is connected to the polarization controller and is used to receive the input electrical signal, couple the input electrical signal to the polarization-adjusted multiplexed optical signal, and output the modulated optical signal. The shared optical delay loop is connected to the optical modulator and is used to cyclically transmit the modulated optical signal to increase the transmission delay of the modulated optical signal. The photodetector is connected to the shared optical delay loop and is used to convert the optical signal after cyclic transmission into a first electrical signal. The signal processing unit is connected to the photodetector and is used to process the first electrical signal to generate and output the state signal of the storage pool.

4. The photoelectric storage tank computing device according to claim 3, characterized in that, The signal processing unit includes: an electrical amplifier, an analog-to-digital converter, and a digital signal processor; The electrical amplifier is connected to the photodetector and is used to amplify the first electrical signal to obtain an analog electrical signal. The analog-to-digital converter is connected to the electrical amplifier and is used to convert the analog electrical signal into a digital signal; The digital signal processor is connected to the analog-to-digital converter and is used to extract the at least two state components from the digital signal to generate and output the reservoir state signal.

5. The photoelectric storage tank computing device according to claim 4, characterized in that, The output terminal of the electrical amplifier is connected to the electrical input terminal of the optical modulator, and is used to input a portion of the electrical signal in the analog electrical signal as a feedback electrical signal to the optical modulator.

6. A method for calculating a photoelectric storage tank, characterized in that, include: Acquire at least two optical signals with different characteristics; The at least two optical signals are multiplexed to the same optical transmission path to obtain multiplexed optical signals; The system receives an input electrical signal to be processed and couples the input electrical signal to the multiplexed optical signal. The coupled optical signal is then subjected to photoelectric processing to obtain a reservoir state signal. The reservoir state signal includes at least two state components, and the at least two state components are respectively associated with different characteristics of the at least two optical signals.

7. The method for calculating a photoelectric storage tank according to claim 6, characterized in that, The process of photoelectric processing of the coupled optical signal to obtain the reservoir status signal includes: The transmission delay of the coupled optical signal after cyclic transmission is obtained; Based on the product of the response coefficient of the photodetector and the matching resistance, the optical signal after the cyclic transmission is photoelectrically converted to obtain the first electrical signal. The first electrical signal is amplified using the target gain, and the amplified electrical signal is filtered using the target lower cutoff frequency and the target upper cutoff frequency to obtain the second electrical signal. Based on the transmission delay, the second electrical signal, the input electrical signal, and a preset parameter set, the at least two state components are calculated to generate the reservoir state signal. The preset parameter set includes the half-wave voltage of the optical modulator, and the bias voltage, input scaling factor, and initial optical signal intensity corresponding to each optical signal.

8. The method for calculating the photoelectric storage tank according to claim 7, characterized in that, The calculation of the at least two state components based on the transmission delay, the second electrical signal, the input electrical signal, and a preset parameter set includes: For each optical signal, a combined electrical signal is calculated based on the input electrical signal, the bias voltage corresponding to the optical signal, the input scaling factor, and the historical state components before the transmission delay. Based on the half-wave voltage, the combined electrical signal is subjected to nonlinear modulation processing to obtain the modulation intensity coefficient; Based on the initial optical signal intensity corresponding to the optical signal and the modulation intensity coefficient, the first optical intensity value is calculated; Based on the product of the first light intensity value and the response coefficient with the matching resistance, the state component corresponding to the one optical signal is calculated to obtain the at least two state components.

9. An electronic device, characterized in that, It includes the photoelectric reservoir computing device as described in any one of claims 1 to 5.

10. An electronic device, characterized in that, It includes a processor and a memory, the processor being coupled to the memory; the memory is used to store computer instructions, which are loaded and executed by the processor to enable the computer device to implement the photoelectric reservoir calculation method as described in any one of claims 6 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes computer-executable instructions that, when executed on a computer, cause the computer to perform the photoelectric reservoir calculation method as described in any one of claims 6 to 8.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when run on an electronic device, causes the electronic device to perform the photoelectric reservoir calculation method as described in any one of claims 6 to 8.

Images