An all-optical reservoir-pool parallel computing device

Abstract

The application discloses a kind of all-optical reservoir pool parallel computing device, by optical injection module, optical fiber ring laser module and signal detection module composition.Optical injection module includes the signal injection of through broadband laser, intensity modulator and coupler 1 in optical fiber ring and the signal injection of through phase modulator in optical fiber ring;Optical fiber ring laser module is by coupler, polarization controller, phase modulator, optical isolator, semiconductor optical amplifier, delay optical fiber and adjustable attenuator composition;Signal detection module is by coupler, polarimeter and photodetector composition, the optical signal of optical fiber ring laser is exported by coupler 2, again be divided into two ways by coupler 3, one way enters photodetector 1, another way enters photodetector 2 by polarimeter.The application is based on intensity dynamics and polarization dynamics and realizes the parallel processing of all-optical reservoir pool to two-way input signal, with simple structure, parallel processing performance good and the like characteristics.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of all-optical reserve pool parallel computing device, belong to optical information processing technical field. BACKGROUND

[0002] In recent years, inspired by the method of information processing in the brain, artificial neural networks have received extensive attention from scholars. Artificial neural network is an algorithm that simulates the neural behavior characteristics of human brain to process information, which relies on the complexity of the system itself to adjust the interconnection between the neurons (nodes) to achieve high-speed processing of data and information. Reservoir computing is a new type of recurrent neural network model. The initial hardware implementation of reservoir computing requires hundreds or even thousands of physical nodes to form a reservoir, which consumes a lot of energy and is difficult to implement. Using a structure of a nonlinear node plus a delay feedback loop to build a reservoir greatly saves hardware resources. Today, neural network algorithms implemented in the electrical domain have become mature, but there are certain limitations in processing speed. In the optical domain, research has focused on optoelectronic reservoir computing and all-optical reservoir computing. Optical reservoir computing has the advantages of fast information processing speed, strong processing capacity, and parallel processing.

[0003] Current research shows that most optical reservoir computing is based on intensity dynamics. This type of reservoir computing system is simple in structure and easy to implement, but its performance is easily affected by system intensity noise; a few researchers are committed to the exploration of wavelength dynamics optoelectronic reservoir computing, but it is not easy to implement. So far, there has been no application research on optical reservoir computing based on polarization dynamics. Optical reservoir computing based on polarization dynamics delay feedback nonlinear system has rich dynamic response, is not sensitive to intensity noise, has good network performance, and has broad application prospects. On the other hand, most of the current all-optical reservoir computing can only process one task, and a few all-optical reservoir computing structures that can achieve parallel computing are complex, generally requiring two physical nodes or two feedback loops, or using more control devices. Therefore, a new type of all-optical reservoir pool parallel computing device needs to be designed to improve the parallel processing performance of the network, reduce interference, and reduce energy consumption to meet the technical needs of optical reservoir pool computing. SUMMARY

[0004] The present application aims to overcome the shortcomings of the prior art and provide an all-optical reservoir pool parallel computing device. The present application simultaneously utilizes the intensity dynamics and polarization dynamics of a semiconductor optical amplifier fiber ring laser to achieve parallel processing of input signals.

[0005] To achieve the above-mentioned invention, the present application adopts the following technical solutions:

[0006] An all-optical reservoir parallel computing device includes three modules: an optical injection module, a fiber ring laser module, and a signal detection module;

[0007] The optical injection module consists of two parts: intensity injection, in which the broadband laser passes through an intensity modulator and coupler 1 in the fiber ring laser module to complete the optical injection; and phase injection, in which the optical injection is completed through a phase modulator in the fiber ring laser module.

[0008] The fiber ring laser module consists of coupler 1, polarization controller, phase modulator, optical isolator, semiconductor optical amplifier, coupler 2, delay fiber and adjustable attenuator. The output of the previous device is connected to the input of the next device in sequence to form a closed loop.

[0009] The signal detection module consists of coupler 2, coupler 3, analyzer, photodetector 1, and photodetector 2. The optical signal of the fiber ring laser is output through coupler 2, and then split into two paths through coupler 3. One path directly enters photodetector 1 for photoelectric conversion to obtain the intensity dynamic state output signal of the fiber ring laser. The other path enters photodetector 2 after passing through the analyzer for photoelectric conversion to obtain the polarization dynamic state output signal of the fiber ring laser.

[0010] Preferably, the output terminal of the polarization controller is connected to the input terminal of the phase modulator.

[0011] Preferably, the semiconductor optical amplifier operates in the linear region.

[0012] Compared with the prior art, the present invention has the following obvious and prominent substantive features and significant advantages:

[0013] 1. The device of this invention uses the polarization dynamics state of a delayed nonlinear system. Delayed nonlinear systems based on polarization dynamics are less susceptible to system intensity noise and have good network performance.

[0014] 2. The device of the present invention has a simple structure, low power consumption, is easy to implement, and has good parallel processing performance;

[0015] 3. When the device of the present invention runs two tasks simultaneously, it shares a single fiber optic ring as a reserve pool, which greatly saves hardware resources. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of the present invention.

[0017] Figure 2 The output waveforms of the classification task using intensity dynamics and the prediction task using polarization dynamics are shown in the preferred embodiment of the present invention.

[0018] Figure 3 The prediction task, which is a preferred embodiment of the present invention, uses intensity dynamics, and the classification task uses polarization dynamics, as shown in the output waveforms. Detailed Implementation

[0019] The above solution will be further described below with reference to specific embodiments. The preferred embodiments of the present invention are described in detail below:

[0020] Example 1:

[0021] In this embodiment, see Figure 1 A parallel computing device with an all-optical reservoir includes three modules: an optical injection module, a fiber ring laser module, and a signal detection module.

[0022] The optical injection module consists of two parts: intensity injection, in which the broadband laser passes through an intensity modulator and coupler 1 in the fiber ring laser module to complete the optical injection; and phase injection, in which the optical injection is completed through a phase modulator in the fiber ring laser module.

[0023] The fiber ring laser module consists of coupler 1, polarization controller, phase modulator, optical isolator, semiconductor optical amplifier, coupler 2, delay fiber and adjustable attenuator. The output of the previous device is connected to the input of the next device in sequence to form a closed loop.

[0024] The signal detection module consists of coupler 2, coupler 3, analyzer, photodetector 1, and photodetector 2. The optical signal of the fiber ring laser is output through coupler 2, and then split into two paths through coupler 3. One path directly enters photodetector 1 for photoelectric conversion to obtain the intensity dynamic state output signal of the fiber ring laser. The other path enters photodetector 2 after passing through the analyzer for photoelectric conversion to obtain the polarization dynamic state output signal of the fiber ring laser.

[0025] This embodiment of the all-optical reservoir parallel computing device utilizes the polarization dynamics of a delay nonlinear system. Delay nonlinear systems based on polarization dynamics are less susceptible to system intensity noise and exhibit good network performance. This embodiment also leverages the intensity and polarization dynamics of a semiconductor optical amplifier and fiber ring laser, enabling better parallel processing of input signals. When running two tasks simultaneously, this embodiment shares a single fiber ring as a reservoir, significantly saving hardware resources.

[0026] Example 2:

[0027] This embodiment is basically the same as Embodiment 1, except that:

[0028] In this embodiment, the output terminal of the polarization controller is connected to the input terminal of the phase modulator.

[0029] In this embodiment, the semiconductor optical amplifier operates in the linear region.

[0030] Example 3:

[0031] This embodiment is basically the same as the above embodiments, except that:

[0032] In this embodiment, to achieve simultaneous processing of two input signals, intensity modulation and phase modulation are used to load the input signals, corresponding to intensity dynamics state output and polarization dynamics state output, respectively. The laser generated by the fiber ring laser has rich polarization states, which are related to parameters such as the bias current of the semiconductor optical amplifier and the attenuation intensity of the adjustable attenuator. By adjusting the parameters of each component in the system, the two signals can be balanced, reducing their mutual influence.

[0033] The original input signals of the two tasks are masked on the computer and then simultaneously loaded onto the RF input terminals of the intensity modulator and phase modulator, respectively. The intensity dynamics state output signal and polarization dynamics state output signal of the fiber ring laser are collected by two photodetectors, respectively, as the state response of the two input signals.

[0034] The broadband laser is from Shanghai Hanyu Fiber Optic Communication Technology Co., Ltd. The intensity modulator and phase modulator are a 2.5 Gbit / s intensity modulator from JDSU and a LiNbO3 electro-optic phase modulator (10053) from Covega, with the drive signals generated by an arbitrary waveform generator (AWG520) from Tektronix. The polarization controller uses a fiber optic squeezer (PLC-001) from General Photonics to adjust the polarization state of the light entering the phase modulator. The semiconductor optical amplifier is a semiconductor optical amplifier (SOA-SC-14-FCA) from CIP (UK), serving as a nonlinear node to generate a rich dynamic response. The delay fiber is a 1 km long G.652 standard single-mode fiber, which determines the feedback delay of the delay nonlinear system. Coupler 1 has a coupling ratio of 80:20, while couplers 2 and 3 both have coupling ratios of 50:50. The adjustable attenuator is a Thorlabs adjustable fiber optic attenuator (M-VC / 000 19875) used to control the power of light in the fiber optic loop, thereby controlling the system feedback intensity. The optical isolator is used to control the unidirectional propagation of light in the fiber optic loop. The photodetector is a PIN-TIA detector manufactured by Shenzhen Feitong Co., Ltd. Electrical signals were acquired using a PicoScope 5203 digital oscilloscope from PICO Corporation. Signal processing was implemented using MATLAB software on a general-purpose microcomputer with an Intel(R) Core(TM) i5-6200U CPU@2.30GHz, 12GB RAM, and Windows 10 system.

[0035] Parallel processing of a binary classification task using sinusoidal / square wave signals and a SantaFe chaotic time series prediction task was attempted using different signal input methods or dynamic states. Both tasks used uniformly distributed random signals within the range [0,1] as mask signals, but the number of virtual nodes differed, being set to 200 and 50 respectively. The output waveforms for the classification task using intensity dynamics and the prediction task using polarization dynamics are shown in [reference needed]. Figure 2 The output waveforms for the prediction task using intensity dynamics and the classification task using polarization dynamics can be found in [the original text]. Figure 3It can be seen that regardless of the input method or the corresponding dynamic state, both tasks yield relatively good output waveforms. For the classification task, 360 data points were used for training and 120 data points for testing; for the time series prediction task, 1000 data points were used for training and 300 data points for testing. When using intensity dynamics for the classification task and polarization dynamics for the prediction task, the obtained test classification accuracy and test normalized mean square error were 93.06% and 0.082, respectively; when using intensity dynamics for the prediction task and polarization dynamics for the classification task, the obtained test classification accuracy and test normalized mean square error were 95.83% and 0.096, respectively.

[0036] The all-optical reservoir parallel computing device described in the above embodiment consists of an optical injection module, a fiber ring laser module, and a signal detection module. The optical injection module includes signal injection via a broadband laser, an intensity modulator, and coupler 1 in the fiber ring, as well as signal injection via a phase modulator in the fiber ring. The fiber ring laser module comprises a coupler, a polarization controller, a phase modulator, an optical isolator, a semiconductor optical amplifier, a delay fiber, and an adjustable attenuator. The signal detection module comprises a coupler, an analyzer, and a photodetector. The optical signal from the fiber ring laser is output through coupler 2, and then split into two paths by coupler 3: one path enters photodetector 1, and the other path enters photodetector 2 via the analyzer. This invention achieves parallel processing of two input signals in an all-optical reservoir based on intensity dynamics and polarization dynamics, featuring simple structure and good parallel processing performance.

[0037] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made according to the purpose of the invention. Any changes, modifications, substitutions, combinations or simplifications made based on the spirit and principle of the technical solution of the present invention shall be equivalent substitutions. As long as they meet the purpose of the invention and do not deviate from the technical principle and inventive concept of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A parallel computing device with an all-optical reservoir, characterized in that, It includes three modules: an optical injection module, a fiber ring laser module, and a signal detection module; The optical injection module consists of two parts: intensity injection, in which the broadband laser passes through an intensity modulator and coupler 1 in the fiber ring laser module to complete the optical injection; and phase injection, in which the optical injection is completed through a phase modulator in the fiber ring laser module. The fiber ring laser module consists of coupler 1, polarization controller, phase modulator, optical isolator, semiconductor optical amplifier, coupler 2, delay fiber and adjustable attenuator. The output of the previous device is connected to the input of the next device in sequence to form a closed loop. The signal detection module consists of coupler 2, coupler 3, analyzer, photodetector 1, and photodetector 2. The optical signal of the fiber ring laser is output through coupler 2, and then split into two paths through coupler 3. One path directly enters photodetector 1 for photoelectric conversion to obtain the intensity dynamic state output signal of the fiber ring laser. The other path enters photodetector 2 after passing through the analyzer for photoelectric conversion to obtain the polarization dynamic state output signal of the fiber ring laser.

2. The all-optical storage pool parallel computing device according to claim 1, characterized in that, The output of the polarization controller is connected to the input of the phase modulator.

3. The all-optical storage pool parallel computing device according to claim 1, characterized in that, The semiconductor optical amplifier operates in the linear region.

Images