CN112615677B - Non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning - Google Patents
Non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning Download PDFInfo
- Publication number
- CN112615677B CN112615677B CN202011375971.6A CN202011375971A CN112615677B CN 112615677 B CN112615677 B CN 112615677B CN 202011375971 A CN202011375971 A CN 202011375971A CN 112615677 B CN112615677 B CN 112615677B
- Authority
- CN
- China
- Prior art keywords
- time domain
- optical
- neural network
- domain signal
- time delay
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/50—Transmitters
- H04B10/516—Details of coding or modulation
- H04B10/5165—Carrier suppressed; Single sideband; Double sideband or vestigial
-
- 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
-
- 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
- G06N3/048—Activation functions
-
- 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
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/50—Transmitters
- H04B10/572—Wavelength control
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Optical Communication System (AREA)
Abstract
The invention discloses a non-orthogonal wavelength division multiplexing time domain ghost imaging method and a non-orthogonal wavelength division multiplexing time domain ghost imaging system based on deep learning, and belongs to the field of photoelectric signal processing. The invention adopts the neural network to recover the time domain signal to be detected. The neural network is actually a function fitting, and the neural network model learns the mapping relation between the power integral vector measured from the receiving end and the time domain signal to be measured in the training process. The neural network is trained by reasonably generating and selecting the training set, and the many-to-one mapping in non-orthogonal time is realized, so that the time delay of the light source mode is compensated, the light source mode is not sensitive to the time delay, and the robustness to the time delay is ensured. The invention can transmit and recover high-quality signals at high speed under the non-orthogonal condition by using incomplete orthogonal devices, the signals can reach 0.78ns, and 9 orders of magnitude are improved compared with the prior art. When the time delay is less than 0.5bit, the accuracy rate can reach 99.8 percent, so that the quality of the signal is ensured under the condition of high-speed transmission.
Description
Technical Field
The invention belongs to the field of photoelectric signal processing, and particularly relates to a non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning.
Background
The time domain ghost imaging technology is a high-speed time domain signal measurement technology and is widely applied to the fields of signal detection, transmission encryption and the like at present. The technology modulates the time domain signal to be detected to the pre-coded optical carrier wave and transmits the pre-coded optical carrier wave, and the time domain signal can be recovered without time resolution at a receiving end. The performance requirement of the photoelectric detector at the receiving end is extremely low, and a high-speed photoelectric detector with large bandwidth is not needed. In addition, the imaging mode is insensitive to the loss of the channel, and is beneficial to high-speed time domain signal detection and recovery.
In general, in order to reduce the number of channels in measurement, the light source code pattern used in the wdm-tdm (wavelength division multiplexing-time domain ghost imaging) is an orthogonal code pattern. However, the time delay between different channels caused by device differences often destroys the orthogonality of the code patterns, resulting in poor quality of recovered signals and high error rate.
In order to solve the above problems, the conference paper "Single Shot Time Domain Ghost Imaging using wavelet Imaging" of pitter ryzkowski et al proposes a wavelength-multiplexed calculation Time Domain Ghost Imaging method, which mainly comprises the following steps: the transmitting end adopts a light source (comprising a plurality of wavelengths) of Wavelength Division Multiplexing (WDM), and modulates a Hadamard code pattern for the light source. Light sources with different wavelengths are combined in a link to form an optical carrier, and a signal to be measured is modulated onto the carrier and transmitted. And the receiving end demultiplexes the optical carrier to recover information of different wavelength channels, and recovers the signal to be detected by using a conventional ghost imaging correlation algorithm.
Although this approach avoids the occurrence of time delays through tightly synchronized low speed modulation devices, it suffers from the following drawbacks and deficiencies: for a light source modulation Hadamard code pattern, because a traditional correlation algorithm is very sensitive to the orthogonality of the code pattern (the orthogonality of the code pattern can be ensured by synchronous non-time delay), the paper adopts a waveform rectifier (Waveshape) to ensure the orthogonality. Although Waveshaper can ensure that the wavelengths are completely synchronized, the modulation rate of the device is very slow, and the time resolution (the minimum time interval of signals which can be resolved) is only 0.5 s. If it is desired to increase the signal rate, the use of conventional modulators is necessary, and the problem of the individual wavelengths of the optical source signal being out of synchronization is necessarily encountered, which can result in significant distortion of the signal recovered by conventional correlation algorithms. It is difficult to ensure high-speed signal transmission and high quality at the same time.
Disclosure of Invention
Aiming at the defects and improvement requirements of the prior art, the invention provides a non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning, and aims to compensate the time delay of a light source mode through a fully-connected convolutional neural network, so that the light source mode is insensitive to the time delay, and the robustness to the time delay is ensured.
To achieve the above object, according to a first aspect of the present invention, there is provided a deep learning based non-orthogonal wavelength division multiplexing time domain ghost imaging method, comprising:
a training stage:
(S1) generating a training set under the maximum time delay in a simulation manner, where the training sample is a power integration vector of the demultiplexed optical carriers obtained in the simulation — a real time domain signal to be measured, the maximum time delay is a maximum time delay difference between N optical carriers and the time domain signal to be measured, and a code pattern order N of the optical carriers is an integer power of 2 (for example, but not limited to, N is 8, the number of channels may be extended, and a communication system with more channels is generated);
(S2) the power integral vector of the optical carrier obtained by simulation after demultiplexing is used as the input of the neural network, the real time domain signal to be tested is used as the output of the neural network, and the neural network is trained by using a training set;
an application stage:
(T1) the transmitting terminal generates N wavelength optical carriers precoded as orthogonal code patterns, and the N optical carriers with different wavelengths are multiplexed and then start to be transmitted, and a certain time delay is allowed among the wavelengths without complete synchronization;
(T2) modulating the time domain signal to be measured to the multiplexed optical carrier in the transmission link;
(T3) the receiving end de-multiplexes the optical carrier after receiving the optical carrier modulated by the time domain signal to be detected, to obtain a power integral vector, inputs the power integral vector to the trained neural network, and outputs the recovered time domain signal to be detected.
Preferably, the maximum delay is less than 0.5 bit.
Has the advantages that: the invention eliminates relevant problem data by limiting the maximum delay to be less than 0.5 bit.
Preferably, the neural network is a fully-connected neural network.
Has the advantages that: all information is required to be trained when time domain signals are processed in time domain ghost imaging, all dimension characteristics can be completely reserved by the fully-connected neural network when all the characteristics are extracted, and the effect of the fully-connected neural network is stable when a gradient descent method is used for training. Compared with other neural networks (for example, the convolutional neural network only retains local characteristics).
To achieve the above object, according to a second aspect of the present invention, there is provided a deep learning based non-orthogonal wavelength division multiplexing time domain ghost imaging system, comprising: the system comprises a transmitting end (1), a transmission link (2) and a receiving end (3);
the transmitting terminal (1) is used for generating N-wavelength optical carriers precoded into orthogonal code patterns, multiplexing the N optical carriers with different wavelengths and then starting transmission, and allowing a certain time delay among the wavelengths without complete synchronization;
the transmission link (2) is used for modulating the time domain signal to be detected onto the multiplexed optical carrier;
and the receiving end (3) is used for demultiplexing the optical carrier after receiving the optical carrier modulated by the time domain signal to be detected to obtain a power integral vector, inputting the power integral vector to the trained neural network and outputting the recovered time domain signal to be detected.
Preferably, the neural network is obtained by training using a training set under the maximum time delay, the training sample is a power integration vector of the demultiplexed optical carriers obtained by simulation-a real time domain signal to be detected, the maximum time delay is a maximum time delay difference between N optical carriers and the time domain signal to be detected, and the code pattern order N of the optical carriers is an integer power of 2.
Preferably, the maximum delay is less than 0.5 bit.
Preferably, the neural network is a fully-connected neural network.
Preferably, the transmitting end includes:
the field programmable gate array FPGA is used for generating N paths of electric signals of an Hadamard code type as a code type of a light source, and the code type of '1' in the Hadamard code type is changed into '0';
the optical module transmitter is used for receiving an electric signal of a Hadamard code type and generating N paths of optical signals;
the arrayed waveguide grating AWG1 is used for multiplexing N optical signals into one path and transmitting the path as an optical carrier in one optical fiber.
Has the advantages that: compared with the prior art which is an analog signal light source, the light source of the digital code type is generated, and the time domain signal to be detected generated by the transmitting end is a digital signal.
Preferably, the transmission link comprises:
the erbium-doped fiber amplifier EDFA is used for amplifying signal light power so as to compensate power loss caused by the optical fiber;
the polarization controller PC is used for adjusting the polarization state of the optical carrier to be consistent with that of the Mach-Zehnder modulator MZM;
the radio frequency amplifier RFA is used for amplifying a time domain signal to be detected generated by the FPGA;
and the Mach-Zehnder modulator MZM is used for modulating the amplified time domain signal to be measured generated by the FPGA onto the optical carrier wave after the polarization state is adjusted.
Preferably, the receiving end includes:
the arrayed waveguide grating AWG2 is used for demultiplexing the optical carrier to generate optical signals of N different wavelength channels;
the photoelectric detector PD is used for converting the N paths of optical signals into N paths of electric signals;
the Oscilloscope is used for detecting and integrating to generate power integral of the N paths of electric signals to form a power integral vector;
and the prediction module is used for inputting the power integral vector into the trained neural network model and outputting the predicted time domain signal to be detected.
Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained:
compared with the prior art, the method adopts the neural network to recover the time domain signal to be detected. The neural network is actually a function fitting, and the neural network model learns the mapping relation between the power integral vector measured from the receiving end and the time domain signal to be measured in the training process. One-to-one mapping in the case of orthogonality, and many-to-one mapping in the case of non-orthogonality. The neural network is trained by reasonably generating a training set and preferably selecting a proper training set, and the many-to-one mapping is realized, so that the time delay of the light source mode is compensated, the light source mode is insensitive to the time delay, and the robustness to the time delay is ensured. Therefore, a high-quality signal can be transmitted and recovered at a high speed under the non-orthogonal condition by using an incomplete orthogonal device (such as an XFP optical module), the signal can reach 0.78ns, 9 orders of magnitude is improved compared with the prior art, and the accuracy can reach 99.8% when the time delay is less than 0.5bit, so that the quality of the signal can be ensured under the high-speed transmission condition.
Drawings
FIG. 1 is a diagram of a deep learning based non-orthogonal WDM time-domain ghost imaging system;
FIG. 2 is a schematic diagram of an orthogonal code pattern of a precoded Hadamard matrix provided by the present invention;
FIG. 3 is a block diagram of a fully-connected neural network provided by the present invention;
fig. 4 is a comparison graph of results of neural network prediction and conventional correlation algorithm, which includes three different code pattern recovery situations, where a gray curve is a signal waveform actually measured, a black curve is a theoretical waveform, a circle is a code pattern result predicted by the scheme of the present invention, a plus sign is a signal result recovered by conventional time domain ghost imaging correlation, and a specific code pattern is: (a)00111100, (b)010001001, and (c) 01110100.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention provides a non-orthogonal wavelength division multiplexing time domain ghost imaging method and a non-orthogonal wavelength division multiplexing time domain ghost imaging system based on deep learning. The time domain signal to be detected can be restored by single transmission and measurement through wavelength division multiplexing time domain ghost imaging, but the problem that all wavelengths are difficult to completely synchronize exists. Therefore, high quality signals can be transmitted and recovered at high speed without being orthogonal by using a device (e.g., an XFP optical module) with incomplete orthogonality.
In the embodiment of the invention, a non-orthogonal wavelength division multiplexing time domain ghost imaging system based on deep learning is applied to an optical fiber communication system, the structure is shown in figure 1, firstly, a field programmable gate array FPGA and eight XFP optical modules with the speed of 10.31Gb/s are used for generating an optical carrier wave with Hadamard code type pre-coding. The eight optical signals are non-orthogonal code type light sources because of the fact that channel devices are different and have certain time delay, complete synchronization cannot be achieved, and time delay is allowed to exist among channels. Eight optical signals are multiplexed into one path by an Arrayed Waveguide Grating (AWG), and the optical signals are transmitted as optical carriers in one optical fiber, wherein the channel interval of the arrayed waveguide grating is 100 GHz. The optical carrier is subjected to power amplification by using an erbium-doped fiber amplifier EDFA to compensate power loss caused by the optical fiber, and the polarization controller PC is used for adjusting the polarization state of the optical carrier to be consistent with that of a Mach-Zehnder modulator MZM (the bandwidth is 3.2 GHz). The bandwidth of the time domain signal to be detected and the bandwidth of the optical carrier are both 1.29 GHz. The time domain signal is an 8-bit OOK signal, is generated by the FPGA and modulated onto an optical carrier by the Mach-Zehnder modulator MZM, wherein the radio frequency amplifier RFA is used for amplifying the time domain signal to be measured. The arrayed waveguide grating AWG2 is used for demultiplexing the optical carrier to generate optical signals of different wavelength channels at a receiving end. At the receiving end, the optical signal is converted into an electrical signal using a photodetector (with a bandwidth of 10GHz), and detected and integrated by an Oscilloscope (Tektronix DSA 72504D, with a bandwidth of 25GHz) to generate a power integration vector. And finally, inputting the power integration vector into the trained neural network model by using a computer, and outputting the predicted time domain signal to be detected.
The FPGA and the eight optical modules Transceiver are integrally used as a light source of a wavelength division multiplexing optical fiber communication system to bear a Hadamard code type light intensity sequence. The code pattern of '1' in the Hadamard code pattern becomes '0'. Under the influence of time delay, the synchronization among channels can be destroyed, the orthogonality of Hadamard code types is influenced, and the system is a non-orthogonal wavelength division multiplexing system. Meanwhile, the communication system is not limited to eight channels, the number of the channels can be expanded, and more channels can be generated.
The photodetector PD and the Oscilloscope are used as a whole to receive power integration vectors of different channels, and do not acquire correlation waveforms to process signals.
The computer PC predicts the power integral vector by using a neural network model, and needs to simulate in advance to generate a training set with time delay for training.
In an embodiment of the present invention, a deep learning-based non-orthogonal wavelength division multiplexing time domain ghost imaging method is further provided, including:
step one, generating a data set and training a network. And (4) generating data sets under different maximum time delays in a simulation mode, wherein the maximum time delay is defined as the maximum time delay difference between the eight signals and the time domain signal to be detected. The data set inputs power integral vectors obtained by simulation and outputs real time domain signals (digital signals) to be measured. The data set is divided into a training set and a testing set, a model structure is defined, a model is trained and optimized, and performance analysis is carried out by using the testing set.
And step two, acquiring a power integral vector actually measured. The code pattern of the Hadamard code pattern which is '-1' is first changed to '0'. The code pattern is then generated to each wavelength of the wavelength division multiplexing light source to produce wavelength division multiplexed optical carriers. The optical carrier is modulated by the time domain signal to be measured, each wavelength signal is demultiplexed at a receiving end, and the power integral of each channel is detected. The power integrals of the individual channels form a power integral vector as a whole.
And step three, predicting the time domain signal to be measured, generating a power integral vector actually measured by using the device, inputting the power integral vector into the neural network model for prediction, and recovering the signal to be measured.
The neural network model sequentially comprises an input layer, a hidden layer and an output layer. The input is a measured power integration vector which comprises a measured value of the number of channels. The number of neurons in the output layer is the same as that in the input layer. The hidden layer needs to be optimized in number of layers and neurons to improve performance.
Preferably, the maximum delay in step one is high quality data at less than 0.5 bit. If the delay time is more than 0.5bit, the non-uniqueness problem exists, namely the same measured value can be generated under the condition of different delay times, the model prediction effect is influenced, and related problem data needs to be removed.
Preferably, in the model training process in step three, the data sets at different maximum delays need to be optimized. The division of the data set needs to divide different intervals according to the time delay interval, and the increase of the number of the samples of the data set is beneficial to the prediction result of the model.
As shown in fig. 2, changing the '1' in the Hadamard code pattern to '0' is the orthogonal code pattern for precoding. There are eight wavelength channels, and the length of the signal transmitted in each channel is 8 bits. Wherein the black code pattern is '1' and the white code pattern is '0'.
As shown in fig. 3, is a trained neural network. Wherein, all are full-connection neural networks, and input layer and output layer all contain 8 neurons, and the centre contains three hidden layers, has 100, 300, 100 neurons respectively. The activation function of the output layer is a sigmoid layer.
As shown in fig. 4, the final experimental results include three different code pattern recovery cases, (a) corresponding to code pattern 00111100, (b) corresponding to code pattern 010001001, and (c) corresponding to code pattern 01110100. The gray curve is a signal waveform actually measured, the black curve is a theoretical waveform, the circle is a code pattern result predicted by the scheme of the invention, and the plus sign is a signal result recovered by the conventional time domain ghost imaging correlation. The result shows that the result of the invention is identical with the actual code pattern, and the code pattern recovered by the traditional ghost imaging correlation algorithm has a certain bit error rate. The invention has better signal recovery quality and effect.
The non-orthogonal wavelength division multiplexing time domain ghost imaging device and method based on deep learning provided by the invention have the advantages that the neural network model is used for compensating the signal degradation problem caused by carrier asynchronization, the specific channel delay condition does not need to be measured, and the channel delay robustness is stronger. Aiming at the time delay between channels, software is used without a complex physical time delay compensation network, the structure is simple, and the system complexity is low. The quality and speed of the detected time domain signal are improved.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A deep learning-based non-orthogonal wavelength division multiplexing time domain ghost imaging method is characterized by comprising the following steps:
a training stage:
(S1) generating a training set under the maximum time delay in a simulation mode, wherein a training sample is a power integral vector of the demultiplexed optical carriers obtained in the simulation mode-a real time domain signal to be detected, the maximum time delay is the maximum time delay difference between the N optical carriers and the time domain signal to be detected, and the code pattern order N of the optical carriers is an integer power of 2;
(S2) the power integral vector of the optical carrier obtained by simulation after demultiplexing is used as the input of the neural network, the real time domain signal to be tested is used as the output of the neural network, and the neural network is trained by using a training set;
an application stage:
(T1) the transmitting terminal generates N optical carriers with different wavelengths precoded as orthogonal code patterns, and the N optical carriers with different wavelengths are multiplexed and then start to be transmitted, and the wavelengths are allowed to have time delay without complete synchronization;
(T2) modulating the time domain signal to be measured to the multiplexed optical carrier in the transmission link;
(T3) the receiving end de-multiplexes the optical carrier after receiving the optical carrier modulated by the time domain signal to be detected, to obtain a power integral vector, inputs the power integral vector to the trained neural network, and outputs the recovered time domain signal to be detected.
2. The method of claim 1, wherein the maximum delay is less than 0.5 bit.
3. The method of claim 1 or 2, wherein the neural network is a fully-connected neural network.
4. A deep learning based non-orthogonal wavelength division multiplexing time domain ghost imaging system, comprising: the system comprises a transmitting end (1), a transmission link (2) and a receiving end (3);
the transmitting terminal (1) is used for generating N optical carriers with different wavelengths precoded into orthogonal code patterns, multiplexing the N optical carriers with different wavelengths and then starting transmission, and allowing time delay among the wavelengths without complete synchronization;
the transmission link (2) is used for modulating the time domain signal to be detected onto the multiplexed optical carrier;
and the receiving end (3) is used for demultiplexing the optical carrier after receiving the optical carrier modulated by the time domain signal to be detected to obtain a power integral vector, inputting the power integral vector to the trained neural network and outputting the recovered time domain signal to be detected.
5. The system of claim 4, wherein the neural network is obtained by training using a training set at a maximum time delay, the training sample is a power integration vector of the demultiplexed optical carriers obtained by simulation-a real time domain signal to be measured, the maximum time delay is a maximum time delay difference between the N optical carriers and the time domain signal to be measured, and the code pattern order N of the optical carriers is an integer power of 2.
6. The system of claim 5, wherein the maximum delay is less than 0.5 bit.
7. The system of any one of claims 4 to 6, wherein the neural network is a fully-connected neural network.
8. The system of claim 7, wherein the transmitting end comprises:
the field programmable gate array FPGA is used for generating N paths of electric signals of an Hadamard code type as a code type of a light source, and the code type of '1' in the Hadamard code type is changed into '0';
the optical module transmitter is used for receiving an electric signal of a Hadamard code type and generating N paths of optical signals;
the arrayed waveguide grating AWG1 is used for multiplexing N optical signals into one path and transmitting the path as an optical carrier in one optical fiber.
9. The system of claim 7, wherein the transmission link comprises:
the erbium-doped fiber amplifier EDFA is used for amplifying signal light power so as to compensate power loss caused by the optical fiber;
the polarization controller PC is used for adjusting the polarization state of the optical carrier to be consistent with that of the Mach-Zehnder modulator MZM;
the radio frequency amplifier RFA is used for amplifying a time domain signal to be detected generated by the FPGA;
and the Mach-Zehnder modulator MZM is used for modulating the amplified time domain signal to be measured generated by the FPGA onto the optical carrier wave after the polarization state is adjusted.
10. The system of claim 7, wherein the receiving end comprises:
the arrayed waveguide grating AWG2 is used for demultiplexing the optical carrier to generate optical signals of N different wavelength channels;
the photoelectric detector PD is used for converting the N paths of optical signals into N paths of electric signals;
the Oscilloscope is used for detecting and integrating to generate power integral of the N paths of electric signals to form a power integral vector;
and the recovery module is used for inputting the power integral vector into the trained neural network model and outputting a recovered time domain signal to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011375971.6A CN112615677B (en) | 2020-11-30 | 2020-11-30 | Non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011375971.6A CN112615677B (en) | 2020-11-30 | 2020-11-30 | Non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112615677A CN112615677A (en) | 2021-04-06 |
| CN112615677B true CN112615677B (en) | 2022-02-15 |
Family
ID=75228223
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011375971.6A Active CN112615677B (en) | 2020-11-30 | 2020-11-30 | Non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112615677B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1992011549A1 (en) * | 1990-12-21 | 1992-07-09 | Minnesota Mining And Manufacturing Company | Light control film with reduced ghost images |
| CN108683458A (en) * | 2018-05-22 | 2018-10-19 | 华中科技大学 | A kind of time domain ghost imaging device and method based on orthogonal frequency division multiplexing |
| CN109035282A (en) * | 2018-08-02 | 2018-12-18 | 吉林工程技术师范学院 | The thresholding method of Hadamard coded modulation relevance imaging |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3620989A1 (en) * | 2018-09-07 | 2020-03-11 | Panasonic Intellectual Property Corporation of America | Information processing method, information processing apparatus, and program |
| CN111652059B (en) * | 2020-04-27 | 2023-03-24 | 西北大学 | Target identification model construction and identification method and device based on computational ghost imaging |
-
2020
- 2020-11-30 CN CN202011375971.6A patent/CN112615677B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1992011549A1 (en) * | 1990-12-21 | 1992-07-09 | Minnesota Mining And Manufacturing Company | Light control film with reduced ghost images |
| CN108683458A (en) * | 2018-05-22 | 2018-10-19 | 华中科技大学 | A kind of time domain ghost imaging device and method based on orthogonal frequency division multiplexing |
| CN109035282A (en) * | 2018-08-02 | 2018-12-18 | 吉林工程技术师范学院 | The thresholding method of Hadamard coded modulation relevance imaging |
Non-Patent Citations (4)
| Title |
|---|
| Computational Temporal Ghost Imaging Using Intensity-Only Detection Over a Single Optical Fiber;Jiang Tang等;《IEEE Photonics Journal》;20180314;第10卷(第2期);第1555-1558页 * |
| Performance Evaluation of Ghost Imaging with Orthogonal/Non-orthogonal Quantum States in Terms of Image Quality;Takahashi, Y等;《2020 International Symposium on Information Theory and Its Applications (ISITA)》;20201027;111-15 * |
| Single-shot thermal ghost imaging using wavelength-division multiplexing;Chao Deng等;《Applied Physics Letters》;20180129;第112卷(第5期);全文 * |
| 基于卷积神经网络的计算鬼成像方法研究;冯维 等;《光子学报》;20200622(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112615677A (en) | 2021-04-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102396170A (en) | Method of and system for detecting skew between parallel signals | |
| CN1883144B (en) | Optical sub-carrier multiplexed transmission | |
| KR100961819B1 (en) | All optical forward/inverse discrete fourier transform apparatus and method for optical ofdm transmission and transceiver having the same | |
| Khan et al. | Non-data-aided joint bit-rate and modulation format identification for next-generation heterogeneous optical networks | |
| Zhao et al. | Synchronized random bit sequences generation based on analog-digital hybrid electro-optic chaotic sources | |
| RU2636391C1 (en) | System and method for tone modulation of pilot signal by means of data offset | |
| US5440415A (en) | Optical successive modulation-muliplexing for fiber optical network | |
| Zhang et al. | Transformer-based long distance fiber channel modeling for optical OFDM systems | |
| Ullah et al. | Pulsed laser-based optical frequency comb generator for high capacity wavelength division multiplexed passive optical network supporting 1.2 Tbps | |
| EP2779497B1 (en) | All-Optical Regeneration System for Optical Wavelength Division Multiplexed Communication Systems | |
| De Lima et al. | Scalable wideband principal component analysis via microwave photonics | |
| CN112615677B (en) | Non-orthogonal wavelength division multiplexing time domain ghost imaging method and system based on deep learning | |
| US10298323B2 (en) | Generating a pilot tone for an optical telecommunications system | |
| EP3497825B1 (en) | Encoding for optical transmission | |
| JP2012049801A (en) | Signal modulation device and method | |
| Peng et al. | A photonics-inspired compact network: toward real-time AI processing in communication systems | |
| JP4426605B2 (en) | Crosstalk compensation receiver for WDM channels | |
| Rosas-Fernandez et al. | 18 Gchips/s electronic CDMA for low-cost optical access networks | |
| Sygletos et al. | All-optical add-drop multiplexer for OFDM signals | |
| Salgals et al. | Research of M-PAM and duobinary modulation formats for use in high-speed WDM-PON systems | |
| Nakamura et al. | Experimental evaluation of indoor optical wireless MIMO systems with square and linear array constellations | |
| JP6074823B2 (en) | Optical time division multiplexing transmission system | |
| Ahmed et al. | Performance analysis of diagonal permutation shifting (DPS) codes for SAC-OCDMA systems | |
| Bajaj et al. | Application of soft computing technique for optical add/drop multiplexer in terms of OSNR and optical noise power | |
| Meijerink | Coherence multiplexing for optical communication systems |
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 |