CN113240104B - Time domain Talbot effect-based serial optical neural network system - Google Patents
Time domain Talbot effect-based serial optical neural network system Download PDFInfo
- Publication number
- CN113240104B CN113240104B CN202110587814.XA CN202110587814A CN113240104B CN 113240104 B CN113240104 B CN 113240104B CN 202110587814 A CN202110587814 A CN 202110587814A CN 113240104 B CN113240104 B CN 113240104B
- Authority
- CN
- China
- Prior art keywords
- optical
- ultra
- neural network
- narrow
- network system
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000003287 optical effect Effects 0.000 title claims abstract description 211
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 80
- 230000000694 effects Effects 0.000 title claims abstract description 23
- LFEUVBZXUFMACD-UHFFFAOYSA-H lead(2+);trioxido(oxo)-$l^{5}-arsane Chemical compound [Pb+2].[Pb+2].[Pb+2].[O-][As]([O-])([O-])=O.[O-][As]([O-])([O-])=O LFEUVBZXUFMACD-UHFFFAOYSA-H 0.000 title abstract description 12
- 210000002569 neuron Anatomy 0.000 claims abstract description 50
- 238000006243 chemical reaction Methods 0.000 claims abstract description 29
- 239000006185 dispersion Substances 0.000 claims abstract description 23
- 230000007274 generation of a signal involved in cell-cell signaling Effects 0.000 claims abstract description 4
- 230000010287 polarization Effects 0.000 claims description 23
- 239000000835 fiber Substances 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 9
- 239000002131 composite material Substances 0.000 claims description 6
- 239000013307 optical fiber Substances 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 5
- 230000001427 coherent effect Effects 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 9
- 238000000034 method Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000001093 holography Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
- G06N3/0675—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means using electro-optical, acousto-optical or opto-electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)
- Optical Communication System (AREA)
- Lasers (AREA)
Abstract
The present disclosure provides a serial optical neural network system based on a time domain taber effect, comprising: the signal generating device is used for generating ultra-narrow light pulses and binary events to be detected, and modulating the binary events to be detected onto the ultra-narrow light pulses to obtain the ultra-narrow light pulses to be detected; the conversion device comprises at least one optical neuron, is connected with the signal generation device and is used for receiving the ultra-narrow optical pulse to be detected, generating an electrical weight factor and modulating the electrical weight factor onto the ultra-narrow optical pulse to be detected to obtain the weight ultra-narrow optical pulse, converting the weight ultra-narrow optical pulse into an electric signal and introducing the electric signal to be consistent withTo each optical neuron to achieve a full connection of at least one optical neuron. According to the serial optical neural network system based on the time domain Talbot effect, the dispersion value conforming to the time domain Talbot effect is introduced, so that all optical neurons are fully connected, and therefore the serial data can be directly and efficiently judged without serial-parallel conversion.
Description
Technical Field
The disclosure belongs to the technical field of artificial intelligence, and particularly relates to a serial optical neural network system based on a time domain Talbot effect.
Background
Artificial neural networks are a technique that utilizes high performance computing to obtain features in mass data and may implement different functions, such as speech recognition, decision making, speech recognition, etc. Although the advent of many artificial intelligence-specific processing chips has now facilitated the development and application of artificial neural networks to some extent, electronic chip-based artificial neural networks still suffer from problems such as latency and energy loss.
If the artificial neural network wants to obtain a further breakthrough, the necessary requirement breaks through the existing framework to search for a low-delay, low-energy-consumption and high-rate computing framework. Photon techniques have the characteristics of large bandwidth (meaning high processing rate), low latency, low energy loss, and can implement advanced mathematical operations such as convolution, differentiation, integration, etc. So photon technology is very suitable for artificial neural networks, several deep learning solutions (optical neural network technologies) based on photon technology have been proposed and verified so far, for example, optical neural network technologies based on principles including holography, interference of light, multimode optical fibers, wavelength division multiplexing, and the like. Existing photonic neural network technologies are used for processing parallel data, and the used structure is completed on the basis of simulating a traditional deep learning network. The neural network architecture that processes parallel data can only process parallel data and does not have the capability to process direct serial data. But in the field of information today, serial data is widely used in a variety of contexts, such as communications, sensing, lidar, radar, ultra-fast imaging, and the like. Serial data generated by these application scenarios cannot be compatible with existing optical neural network systems. Although the serial-parallel conversion device can convert serial data into parallel data acceptable by the traditional optical neural network, the serial-parallel conversion device can cause great processing delay and reduce the efficiency of the whole system, thereby greatly limiting the application prospect of the optical neural network to the serial data.
Disclosure of Invention
First, the technical problem to be solved
In view of the above-mentioned shortcomings of the prior art, a primary object of the present disclosure is to provide a serial optical neural network system based on the time domain taber effect, so as to at least partially solve at least one of the above-mentioned technical problems
(II) technical scheme
To achieve the above object, according to one aspect of the present disclosure, there is provided a serial optical neural network system based on a time domain taber effect, the system comprising:
the signal generating device is used for generating ultra-narrow light pulses and binary events to be detected, and modulating the binary events to be detected onto the ultra-narrow light pulses to obtain the ultra-narrow light pulses to be detected;
a conversion device comprising at least one optical neuron;
the conversion device is connected with the signal generation device and receives the ultra-narrow light pulse to be detected;
the conversion device is used for generating an electrical weight factor and modulating the electrical weight factor onto the ultra-narrow light pulse to be detected to obtain a weight ultra-narrow light pulse;
the conversion device is also used for converting the weight ultra-narrow light pulse into an electric signal;
the conversion device is also used for introducing coincidenceTo each optical neuron to achieve a full connection of said at least one optical neuron, wherein s is any natural number, T is a repetition period of said ultra-narrow optical pulse,for each kilometer of dispersion value, L is the propagation distance of the above-mentioned ultra-narrow optical pulse in the dispersive medium.
Preferably, the signal generating device includes:
the analog signal generator is used for generating a single-frequency microwave analog signal;
an active mode-locked laser connected with the analog signal generator and used for generating the ultra-narrow light pulse;
the band-pass optical filter is connected with the active mode-locking laser and is used for filtering the ultra-narrow optical pulse;
a first arbitrary waveform generator for generating the binary event under test;
the polarization controller is connected with the band-pass optical filter and is used for adjusting the polarization state of the ultra-narrow light pulse output by the band-pass optical filter to be polarized along the optical axis direction;
an intensity modulator connected to the polarization controller for modulating the binary event to be measured to the amplitude of the ultra-narrow light pulse output by the polarization controller;
and the erbium-doped optical fiber amplifier is connected with the intensity modulator and is used for amplifying the power of the ultra-narrow optical pulse output by the intensity modulator to obtain the ultra-narrow optical pulse to be detected.
Preferably, the conversion device includes: at least one optical neural network and one photodetector;
the optical neural network comprises at least one second arbitrary waveform generator and at least one optical neuron;
the optical neural network includes the same number of the second arbitrary waveform generators and the optical neurons;
the second arbitrary waveform generator is used for generating the electrical weight factors;
the photoelectric detector is connected with the optical neural network and is used for converting the weight ultra-narrow optical pulse into an electric signal.
Preferably, the optical neuron comprises at least one complex modulator and at least one dispersive device;
the composite modulator is connected with the second arbitrary waveform generator and the dispersion device;
the optical neurons in the optical neural network are connected in series, and the composite modulator included in the former optical neuron is connected with the dispersion device included in the latter optical neuron;
the optical neurons include the same number of the complex modulators and the dispersive devices.
Preferably, the composite modulator is connected with the erbium-doped fiber amplifier and is used for modulating the electrical weight factors onto the ultra-narrow light pulse to be measured;
the second arbitrary waveform generator is connected with the composite modulator and is used for generating the electrical weight factors;
the dispersion device is used for introducing coincidenceWherein s is an arbitrary natural number, T is the repetition period of the above ultra-narrow optical pulse, < ->For each kilometer of dispersion value, L is the propagation distance of the above-mentioned ultra-narrow optical pulse in the dispersive medium.
Preferably, the optical neural network includes at least one of a phase modulation signal, an intensity modulation signal, and a complex modulation signal in each of the optical neurons controlling the electrical weight factor;
the signal frequency of the electrical weight factor is the multiple of the repetition frequency of the ultra-narrow light pulse to be detected.
Preferably, the ultra-narrow light pulses comprise mode-locked coherent light pulses.
Preferably, the binary event to be measured includes at least one of an intensity modulation event, a phase modulation event and a complex event.
Preferably, the complex modulator includes at least one of a phase modulator, an intensity modulator, and a complex modulator.
Preferably, the binary event to be measured includes any one of a rectangular wave, an inverted triangular wave, a sawtooth wave and a sine wave.
(III) beneficial effects
According to the serial optical neural network system based on the time domain Talbot effect, the dispersion value conforming to the time domain Talbot effect is introduced, so that all optical neurons are fully connected, and therefore high-speed real-time serial data can be directly and effectively judged without serial-parallel conversion. Meanwhile, the type of the data which can be directly identified by the existing photon deep learning technology based on the parallel neural network technology is expanded, and the application scene of the photon neural network is expanded.
Drawings
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort to a person skilled in the art.
Fig. 1 is a block diagram of a serial optical neural network system based on a time domain taber effect according to an embodiment of the present disclosure;
fig. 2 is a flowchart of a training method of a serial optical neural network system based on a time domain taber effect according to an embodiment of the present disclosure;
FIG. 3 is a graph showing the recognition result of a waveform trained by a time domain Talbot effect-based serial optical neural network system according to an embodiment of the present disclosure;
FIG. 4 is a graph showing the recognition result of a waveform trained by a time domain Talbot effect-based serial optical neural network system according to an embodiment of the present disclosure;
FIG. 5 is a graph showing the recognition result of a waveform trained by a time domain Talbot effect-based serial optical neural network system according to an embodiment of the present disclosure;
fig. 6 is a recognition result of a waveform trained by a serial optical neural network system based on a time domain taber effect according to an embodiment of the present disclosure.
Detailed Description
For a better understanding of the objects, features, aspects and advantages of the present disclosure, reference is made to the following detailed description of specific embodiments, which is to be taken in conjunction with the accompanying drawings, it being apparent that the embodiments described are only some, but not all, of the embodiments of the disclosure. All other embodiments, which can be made by those skilled in the art without the inventive effort, are intended to be within the scope of the present disclosure, based on the embodiments herein.
Fig. 1 is a block diagram of a serial optical neural network system based on a time domain taber effect according to an embodiment of the present disclosure, as shown in fig. 1, in an embodiment of the present disclosure, the system includes:
the signal generating device 100 is configured to generate an ultra-narrow optical pulse and a binary event to be measured, and modulate the binary event to be measured onto the ultra-narrow optical pulse to obtain the ultra-narrow optical pulse to be measured; the conversion device 200 comprises at least one optical neuron and is connected with the signal generation device 100, the conversion device 200 receives the ultra-narrow optical pulse to be detected, and the conversion device 200 is used for generating an electrical weight factor and modulating the electrical weight factor onto the ultra-narrow optical pulse to be detected to obtain a weight ultra-narrow optical pulse; the conversion device 200 is further configured to convert the weight ultra-narrow light pulse into an electrical signal; the conversion device 200 is also used to introduce compliance To each optical neuron to achieve full connection of several layers of optical neurons, where s is any natural number, T is the repetition period of ultra-narrow optical pulses,/>For each kilometer of dispersion value, L is the propagation distance of the ultra-narrow optical pulse in the dispersive medium.
In this embodiment, the signal generating device 100 generates an ultra-narrow optical pulse and a binary event to be measured, and modulates the binary event to be measured onto the ultra-narrow optical pulse to obtain the ultra-narrow optical pulse to be measured. The signal generating device 100 is connected with the converting device 200, the converting device 200 receives the ultra-narrow light pulse to be detected transmitted by the signal generating device 100, generates an electrical weight factor, modulates the electrical weight factor onto the received ultra-narrow light pulse to be detected, obtains a weight ultra-narrow light pulse, and converts the weight ultra-narrow light pulse into an electrical signal. The conversion device 200 includes at least one optical neuron therein, and the conversion device 200 introduces coincidenceTo each optical neuron to achieve the full population of several optical neuronsAnd (5) connection.
In one embodiment of the present disclosure, the signal generating device 100 includes: an analog signal generator 1 for generating a single frequency microwave analog signal; the active mode-locked laser 2 is connected with the analog signal generator 1 and is used for generating ultra-narrow light pulses; the band-pass optical filter 3 is connected with the active mode-locked laser 2 and is used for filtering ultra-narrow optical pulses; a first arbitrary waveform generator 4 for generating binary events to be measured; a polarization controller 5 connected to the band-pass optical filter 3 for adjusting the polarization state of the ultra-narrow light pulse output from the band-pass optical filter 3 to be polarized along the optical axis direction; an intensity modulator 6, connected to the polarization controller 5, for modulating the binary event to be measured onto the amplitude of the ultra-narrow light pulse output by the polarization controller 5; and the erbium-doped optical fiber amplifier 7 is connected with the intensity modulator 6 and is used for amplifying the power of the ultra-narrow optical pulse output by the intensity modulator 6 to obtain the ultra-narrow optical pulse to be detected.
In this embodiment, the signal generating device 100 includes an analog signal generator 1, an active mode-locked laser 2, a band-pass optical filter 3, a polarization controller 5, an intensity modulator 6, and an erbium-doped fiber amplifier 7, which are sequentially connected, and the first arbitrary waveform generator 4 is connected to the intensity modulator 6, where the analog signal generator 1 is configured to generate a single-frequency microwave analog signal, and transmit the single-frequency microwave analog signal to the active mode-locked laser 2; an active mode-locked laser 2 for generating an ultra-narrow optical pulse and transmitting the ultra-narrow optical pulse to a band-pass optical filter 3; a band-pass optical filter 3 for filtering the ultra-narrow optical pulse and transmitting the filtered ultra-narrow optical pulse to the polarization controller 5; a polarization controller 5 for adjusting the polarization state of the ultra-narrow light pulse output from the band-pass optical filter 3 to be polarized in the optical axis direction and transmitting the adjusted ultra-narrow light pulse to the intensity modulator 6; a first arbitrary waveform generator 4 for generating a binary event to be measured and transmitting the binary event to be measured to an intensity modulator 6; an intensity modulator 6 for modulating the binary event to be measured onto the amplitude of the ultra-narrow optical pulse output by the polarization controller 5, and transmitting the modulated ultra-narrow optical pulse to an erbium-doped fiber amplifier 7; and the erbium-doped optical fiber amplifier 7 is used for amplifying the power of the modulated ultra-narrow optical pulse output by the intensity modulator 6 to obtain the ultra-narrow optical pulse to be detected.
In an embodiment of the present disclosure, the conversion device 200 includes: at least one optical neural network 13 and one photodetector 12; the optical neural network 13 comprises at least one second arbitrary waveform generator 9 and at least one optical neuron 11; a second arbitrary waveform generator 9 for generating an electrical weight factor; the photodetector 12 is connected with the optical neural network 13 and is used for converting the ultra-narrow light pulse with the weight into an electric signal.
In this embodiment, the conversion device 200 includes at least one optical neural network 13 and one photodetector 12, one end of the optical neural network 13 is connected to the erbium-doped fiber amplifier 7 of the signal generating device 100, the optical neural network 13 is used for receiving the ultra-narrow optical pulse to be tested transmitted by the erbium-doped fiber amplifier 7, the other end of the optical neural network 13 is connected to the photodetector 12, and the photodetector 12 is used for converting the weight ultra-narrow optical pulse into an electrical signal. The optical neural network 13 comprises at least one second arbitrary waveform generator 9 and at least one optical neuron 11, the optical neuron 11 is connected with the erbium-doped fiber amplifier 7 and the second arbitrary waveform generator 9, the second arbitrary waveform generator 9 is used for generating electric weight factors, and the number of the second arbitrary waveform generators 9 and the optical neuron 11 included in the optical neural network 13 is the same.
In an embodiment of the present disclosure, the optical neuron 11 comprises at least one complex modulator 8 and at least one dispersive device 10; the complex modulator 8 is connected with the second arbitrary waveform generator 9 and the dispersion device 10; the optical neurons 11 in the optical neural network 13 are connected in series, and the composite modulator 8 included in the former optical neuron 11 is connected with the dispersion device 10 included in the latter optical neuron 11; the optical neuron 11 includes the same number of complex modulators 8 and dispersion devices 10.
In this embodiment, the optical neuron 11 includes at least one complex modulator 8 and at least one dispersion device 10, the number of the complex modulator 8 and the number of the dispersion devices 10 are the same, the complex modulator 8 is connected with the erbium-doped fiber amplifier 7, the second arbitrary waveform generator 9 and the dispersion device 10, and the complex modulator 8 receives the ultra-narrow optical pulse to be measured transmitted by the erbium-doped fiber amplifier 7 and the electrical weight factor transmitted by the second arbitrary waveform generator 9. The optical neurons 11 in the optical neural network 13 are connected in series, that is to say the complex modulator 8 comprised by the preceding optical neuron 11 and the dispersive device 10 comprised by the following optical neuron 11 are connected.
In an embodiment of the disclosure, a complex modulator 8 is connected to the erbium-doped fiber amplifier 7, and is used for modulating the electrical weight factor onto the ultra-narrow light pulse to be measured output by the erbium-doped fiber amplifier 7, and a second arbitrary waveform generator 9 is connected to the complex modulator 8, and is used for generating the electrical weight factor; a dispersive device 10 for introducing complianceWherein s is an arbitrary natural number, T is the repetition period of the ultra-narrow optical pulse,/->For each kilometer of dispersion value, L is the propagation distance of the ultra-narrow optical pulse in the dispersive medium.
In this embodiment, the complex modulator 8 modulates the electrical weight factor generated by the second arbitrary waveform generator 9 onto the ultra-narrow optical pulse to be measured transmitted from the erbium-doped fiber amplifier 7, and the dispersion device 10 is used for introducing the coincidenceWherein s is an arbitrary natural number, T is the repetition period of the ultra-narrow optical pulse,/->For each kilometer of dispersion value, L is the propagation distance of the ultra-narrow optical pulse in the dispersive medium. When the dispersion value is in accordance with +.>When the optical nerve cells 11 in the optical nerve network 13 are connected, the performance of the serial optical nerve network system is improved.
In an embodiment of the present disclosure, the signal for controlling the electrical weight factor in each optical neuron 11 included in the optical neural network 13 includes at least one of a phase modulation signal, an intensity modulation signal, and a complex modulation signal; the signal frequency of the electrical weight factor is 2 times the repetition frequency of the ultra-narrow optical pulses output by the erbium-doped fiber amplifier 7.
In this embodiment, the optical neural network 13 includes at least one optical neuron 11 and at least one second arbitrary waveform generator 9, each optical neuron 11 includes a complex modulator 8, the complex modulator 8 modulates an electrical weight factor generated by the second arbitrary waveform generator 9 onto the ultra-narrow optical pulse to be measured transmitted by the erbium-doped fiber amplifier 7, a signal for controlling the electrical weight factor by the complex modulator 8 includes at least one of a phase modulation signal, an intensity modulation signal and a complex modulation signal, and a signal frequency of 1 electrical weight factor generated by the second arbitrary waveform generator 9 is 2 times of a repetition frequency of the ultra-narrow optical pulse to be measured output by the erbium-doped fiber amplifier 7.
In one embodiment of the present disclosure, the ultra-narrow light pulses comprise mode-locked coherent light pulses.
In this embodiment, the ultra-narrow light pulse generated by the active mode-locked laser 2 is a mode-locked coherent light pulse.
In one embodiment of the present disclosure, the binary test event includes at least one of an intensity modulation event, a phase modulation event, and a complex event.
In this embodiment, the binary test event generated by the first arbitrary waveform generator 4 includes at least one of an intensity modulation event, a phase modulation event, and a complex event.
In an embodiment of the present disclosure, the complex modulator 8 comprises at least one of a phase modulator, an intensity modulator, and a complex modulator.
In the present embodiment, the type of the complex modulator 8 includes at least one of a phase modulator, an intensity modulator, and a complex modulator.
In one embodiment of the present disclosure, the binary event under test includes any one of a rectangular wave, an inverted triangular wave, a sawtooth wave, and a sine wave.
In the present embodiment, the binary event to be measured generated by the first arbitrary waveform generator 4 includes any one of a rectangular wave, an inverted triangular wave, a sawtooth wave, and a sine wave.
It should be understood that the above examples of the type of ultra-narrow light pulses, the type of binary event under test, and the type of complex modulator 8 are merely exemplary to aid one skilled in the art in understanding the teachings of the present invention, but are not meant to limit the practice of the present invention thereto. In some embodiments, other suitable types may be used for the type of ultra-narrow light pulse, the type of binary event under test, and the type of complex modulator 8, without limitation.
Fig. 2 is a flowchart of a training method of a serial optical neural network system based on a time domain taber effect according to an embodiment of the present disclosure, as shown in fig. 2, in an embodiment of the present disclosure, the training method includes:
s201, inputting waveforms.
S202, the analog signal generator 1 generates a single-frequency microwave analog signal to drive the active mode-locked laser 2.
S203, the active mode-locked laser 2 generates ultra-narrow light pulses.
S204, the light-passing filter 3 filters the ultra-narrow light pulse.
S205, the polarization controller 5 adjusts the polarization state of the ultra-narrow light pulse after filtering to be polarized along the optical axis direction.
S206, the intensity modulator 6 modulates the input waveform onto the ultra-narrow light pulse with the polarization state adjusted.
S207, amplifying the power of the modulated ultra-narrow light pulse by the erbium-doped fiber amplifier 7.
S208, the optical neural network 13 adds an electrical weight factor to the ultra-narrow optical pulse of the amplified power.
S209, the photodetector 12 converts the ultra-narrow light pulse to which the electrical weight factor is added into an electrical signal.
And S210, displaying the electric signal to obtain an output waveform.
S211, obtaining a loss function according to the output waveform and a preset target waveform.
S212, feeding back the loss function to the serial optical neural network system, and adjusting parameters by the serial optical neural network system according to the loss function.
In this embodiment, a waveform is input to the serial optical neural network system, in the signal generating device 100, the analog signal generator 1 generates a single-frequency microwave analog signal, and transmits the single-frequency microwave analog signal to the active mode-locked laser 2, for driving the active mode-locked laser 2, the active mode-locked laser 2 generates an ultra-narrow optical pulse, the ultra-narrow optical pulse is transmitted to the band-pass optical filter 3, the band-pass optical filter 3 filters the received ultra-narrow optical pulse, the filtered ultra-narrow optical pulse is transmitted to the polarization controller 5, the polarization controller 5 adjusts the polarization state of the filtered ultra-narrow optical pulse to be polarized along the optical axis direction, and transmits the adjusted ultra-narrow optical pulse to the intensity modulator 6, the intensity modulator 6 receives the adjusted ultra-narrow optical pulse and the waveform input to the serial optical neural network system, the erbium-doped optical fiber amplifier 7 receives the modulated ultra-narrow optical pulse, amplifies the power of the ultra-narrow optical pulse to be measured, and obtains the ultra-narrow optical pulse to be measured, and the ultra-narrow optical pulse to be measured is transmitted to the neural network 13.
In this embodiment, in the conversion device 200, the optical neural network 13 receives the ultra-narrow optical pulse to be detected, adds an electrical weight factor to the ultra-narrow optical pulse to be detected, and transmits the electrical weight factor to the photodetector 12, the photodetector 12 converts the ultra-narrow optical pulse to which the electrical weight factor is added into an electrical signal, the electrical signal is displayed by the display device to obtain an output waveform, a corresponding mean square error function can be obtained by calculation according to the output waveform and a preset target waveform, the mean square error function is used as a loss function of the serial optical neural network, the loss function is fed back to the serial optical neural network system, and the serial optical neural network system adjusts parameters according to the loss function.
Fig. 3-6 are identification results of a series optical neural network system based on a time domain taber effect after training some waveforms provided in an embodiment of the present disclosure, as shown in fig. 3, in an embodiment of the present disclosure, the upper, middle and lower three graphs respectively represent an input waveform, a target waveform and an output waveform, the input waveform of fig. 3 is a sine wave, as shown in fig. 4, in an embodiment of the present disclosure, the upper, middle and lower three graphs respectively represent an input waveform, a target waveform and an output waveform, the input waveform of fig. 4 is a rectangular wave, as shown in fig. 5, in an embodiment of the present disclosure, the upper, middle and lower three graphs respectively represent an input waveform, a target waveform and an output waveform, and the input waveform of fig. 5 is an inverted triangle wave, as shown in fig. 6, in an embodiment of the present disclosure, the upper, middle and lower three graphs respectively represent an input waveform, a target waveform and an output waveform, and the input waveform of fig. 6 are saw-tooth waveforms.
Through fig. 3-6, after training of the serial optical neural network system based on the time domain taber effect provided by the present disclosure, the matching degree between the output waveform and the target waveform is higher.
According to the serial optical neural network system based on the time domain Talbot effect, the dispersion value conforming to the time domain Talbot effect is introduced, so that all optical neurons are fully connected, and therefore high-speed real-time serial data can be directly and effectively judged without serial-parallel conversion. Meanwhile, the type of the data which can be directly identified by the existing photon deep learning technology based on the parallel neural network technology is expanded, and the application scene of the photon neural network is expanded.
While the foregoing embodiments have been described in some detail to illustrate the purposes, aspects and advantages of the present disclosure, it should be understood that the foregoing embodiments are merely illustrative of the present disclosure and are not limiting, and that various combinations and/or modifications of the various embodiments and/or features set forth in the claims, even though not explicitly recited in the disclosure, are intended to be within the spirit and principles of the disclosure.
Claims (10)
1. A time domain taber effect based serial optical neural network system, comprising:
the signal generating device (100) is used for generating ultra-narrow light pulses and binary events to be detected, and modulating the binary events to be detected onto the ultra-narrow light pulses to obtain ultra-narrow light pulses to be detected;
a conversion device (200) comprising at least one optical neuron;
the conversion device (200) is connected with the signal generation device (100), and the conversion device (200) receives the ultra-narrow light pulse to be detected;
the conversion device (200) is used for generating an electrical weight factor and modulating the electrical weight factor onto the ultra-narrow light pulse to be detected to obtain a weight ultra-narrow light pulse;
the conversion device (200) is further configured to convert the weight ultra-narrow light pulse into an electrical signal;
the conversion device (200) is also used for introducing coincidenceTo each optical neuron to achieve a full connection of said at least one optical neuron, where s is any natural number, T is the repetition period of said ultra-narrow optical pulse,for each kilometer of dispersion value, L is the propagation distance of the ultra-narrow optical pulse in the dispersive medium.
2. The serial optical neural network system according to claim 1, wherein the signal generating device (100) comprises:
an analog signal generator (1) for generating a single frequency microwave analog signal;
an active mode-locked laser (2) connected with the analog signal generator (1) and used for generating the ultra-narrow light pulse;
the band-pass optical filter (3) is connected with the active mode-locking laser (2) and is used for filtering the ultra-narrow optical pulse;
a first arbitrary waveform generator (4) for generating the binary event under test;
a polarization controller (5) connected with the band-pass optical filter (3) and used for adjusting the polarization state of the ultra-narrow light pulse output by the band-pass optical filter (3) to be polarized along the optical axis direction;
an intensity modulator (6) connected with the polarization controller (5) and used for modulating the binary event to be tested to the amplitude of the ultra-narrow light pulse output by the polarization controller (5);
and the erbium-doped optical fiber amplifier (7) is connected with the intensity modulator (6) and is used for amplifying the power of the ultra-narrow light pulse output by the intensity modulator (6) to obtain the ultra-narrow light pulse to be detected.
3. The serial optical neural network system according to claim 2, wherein the conversion device (200) includes: at least one optical neural network (13) and one photodetector (12);
the optical neural network (13) comprises at least one second arbitrary waveform generator (9) and at least one optical neuron (11);
-the optical neural network (13) comprises the same number of second arbitrary waveform generators (9) and optical neurons (11);
-said second arbitrary waveform generator (9) for generating said electrical weight factor;
the photoelectric detector (12) is connected with the optical neural network (13) and is used for converting the weight ultra-narrow optical pulse into an electric signal.
4. A serial optical neural network system according to claim 3, characterized in that the optical neurons (11) comprise at least one complex modulator (8) and at least one dispersive device (10);
the composite modulator (8) is connected with the second arbitrary waveform generator (9) and the dispersion device (10);
-said optical neurons (11) in said optical neural network (13) are connected in series, said complex modulator (8) comprised by a preceding said optical neuron (11) being connected to said dispersive device (10) comprised by a following said optical neuron (11);
the optical neurons (11) comprise the same number of complex modulators (8) and dispersion devices (10).
5. The serial optical neural network system according to claim 4, characterized in that the complex modulator (8) is connected to the erbium-doped fiber amplifier (7) for modulating the electrical weight factors onto the ultra-narrow optical pulses to be measured;
-said second arbitrary waveform generator (9) connected to said complex modulator (8) for generating said electrical weight factors;
the dispersive device (10) is used for introducing complianceWherein s is an arbitrary natural number, T is the repetition period of the ultra-narrow optical pulse,/->For each kilometer of dispersion value, L is the propagation distance of the ultra-narrow optical pulse in the dispersive medium.
6. The serial optical neural network system of claim 3, wherein,
the signal controlling the electrical weight factor in each of the optical neurons (11) included in the optical neural network (13) includes at least one of a phase modulation signal, an intensity modulation signal, and a complex modulation signal;
the signal frequency of the electrical weight factor is 2 times of the repetition frequency of the ultra-narrow light pulse to be detected.
7. The serial optical neural network system of claim 2, wherein the ultra-narrow optical pulses comprise mode-locked coherent optical pulses.
8. The serial optical neural network system of claim 2, wherein the binary test event includes at least one of an intensity modulation event, a phase modulation event, and a complex event.
9. The serial optical neural network system of claim 4, wherein the complex modulator (8) comprises at least one of a phase modulator, an intensity modulator, and a complex modulator.
10. The serial optical neural network system of claim 2, wherein the binary test event includes any one of a rectangular wave, an inverted triangle wave, a sawtooth wave, and a sine wave.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110587814.XA CN113240104B (en) | 2021-05-27 | 2021-05-27 | Time domain Talbot effect-based serial optical neural network system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110587814.XA CN113240104B (en) | 2021-05-27 | 2021-05-27 | Time domain Talbot effect-based serial optical neural network system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113240104A CN113240104A (en) | 2021-08-10 |
| CN113240104B true CN113240104B (en) | 2023-11-14 |
Family
ID=77139316
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110587814.XA Active CN113240104B (en) | 2021-05-27 | 2021-05-27 | Time domain Talbot effect-based serial optical neural network system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113240104B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710248A (en) * | 2018-07-25 | 2018-10-26 | 中国科学院半导体研究所 | The stealthy system of time domain based on time domain Tabo effect |
| JP2018200391A (en) * | 2017-05-26 | 2018-12-20 | 日本電信電話株式会社 | Optical signal processing circuit |
| CN110148879A (en) * | 2019-04-03 | 2019-08-20 | 北京大学 | It is a kind of to manipulate the method and system for realizing light pulse frequency multiplication by frequency spectrum |
| CN111093123A (en) * | 2019-12-09 | 2020-05-01 | 华中科技大学 | Flexible optical network time domain equalization method and system based on composite neural network |
| CN111769875A (en) * | 2020-06-05 | 2020-10-13 | 杭州电子科技大学 | Arbitrary waveform generating device and method based on integer-order time domain Talbot effect |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200327403A1 (en) * | 2019-04-15 | 2020-10-15 | The Hong Kong University Of Science And Technology | All optical neural network |
-
2021
- 2021-05-27 CN CN202110587814.XA patent/CN113240104B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018200391A (en) * | 2017-05-26 | 2018-12-20 | 日本電信電話株式会社 | Optical signal processing circuit |
| CN108710248A (en) * | 2018-07-25 | 2018-10-26 | 中国科学院半导体研究所 | The stealthy system of time domain based on time domain Tabo effect |
| CN110148879A (en) * | 2019-04-03 | 2019-08-20 | 北京大学 | It is a kind of to manipulate the method and system for realizing light pulse frequency multiplication by frequency spectrum |
| CN111093123A (en) * | 2019-12-09 | 2020-05-01 | 华中科技大学 | Flexible optical network time domain equalization method and system based on composite neural network |
| CN111769875A (en) * | 2020-06-05 | 2020-10-13 | 杭州电子科技大学 | Arbitrary waveform generating device and method based on integer-order time domain Talbot effect |
Non-Patent Citations (2)
| Title |
|---|
| 应用光纤分数塔尔博特效应产生重复频率倍频的光脉冲序列;吴波;于晋龙;王争;韩丙辰;罗俊;郭精忠;王菊;杨恩泽;;光学学报(第05期);第41-45页 * |
| 应用泰伯效应进行光学测量的研究进展;刘婷婷;孙海滨;;激光杂志(第11期);第38-42页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113240104A (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110224764A (en) | The method for generating vector terahertz signal using PM and IM based on ISB and multicarrier | |
| CN102916807A (en) | Polarization compensation implementation method of continuous variable quantum key distribution system | |
| CN114815959B (en) | Photon tensor calculation acceleration method and device based on wavelength division multiplexing | |
| CN114819132B (en) | Photon two-dimensional convolution acceleration method and system based on time-wavelength interleaving | |
| CN109687259A (en) | Chaos optical-electronic oscillator and its chaotic signal producing method | |
| CN108259166A (en) | Continuous variable quantum key distribution system and its implementation based on SVM processing | |
| CN107014478A (en) | Modulation-demodulation device for optical fiber vector hydrophone | |
| CN111565075A (en) | Broadband microwave photon phase coding signal generation device and method | |
| CN113240104B (en) | Time domain Talbot effect-based serial optical neural network system | |
| CN112882310B (en) | Kerr optical comb-based arbitrary high-order modulation format signal phase regeneration method | |
| Ren et al. | Transfer Learning Aided Optical Nonlinear Equalization Based Feature Engineering Neural Network | |
| Ye et al. | Photonic-assisted modulation format identification for RF signals under low sampling rate | |
| CN113938211B (en) | Photon full-dimension intelligent identification method and device | |
| CN101741004B (en) | SBS technology-based long-service-life broadband optimal pulse memory | |
| CN206114112U (en) | Automatically controlled light sampling system and terahertz be time domain spectrum appearance now | |
| CN206340819U (en) | System occurs for the Terahertz based on unidirectional carrier transport photodetector | |
| CN111442851B (en) | Time lens measuring system based on Raman soliton self-frequency shift | |
| Deng et al. | Deep learning approaches for photonic-assisted modulation format recognition | |
| CN114355697A (en) | Time domain stealth method and device based on photo-induced amplifier | |
| CN115459855B (en) | Digital pulse shaping method based on linear superposition and optical fiber communication system | |
| CN114338097B (en) | Transparent self-adaptive line type optical time domain stealth device and stealth method | |
| CN217877992U (en) | Distributed optical fiber vibration monitoring equipment | |
| CN117031500B (en) | Light source system and method for long-distance all-fiber laser Doppler wind-finding radar | |
| CN115481723A (en) | Optical convolution accelerating device and method based on time domain Talbot effect | |
| CN108390729B (en) | Bandwidth-controllable optical random signal source generation scheme |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |