CN111650803B - Parallel optical analog-to-digital conversion system and method based on convolution circulation automatic encoder - Google Patents
Parallel optical analog-to-digital conversion system and method based on convolution circulation automatic encoder Download PDFInfo
- Publication number
- CN111650803B CN111650803B CN202010673310.5A CN202010673310A CN111650803B CN 111650803 B CN111650803 B CN 111650803B CN 202010673310 A CN202010673310 A CN 202010673310A CN 111650803 B CN111650803 B CN 111650803B
- Authority
- CN
- China
- Prior art keywords
- mismatch
- module
- digital
- signals
- optical analog
- 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
-
- G—PHYSICS
- G02—OPTICS
- G02F—OPTICAL DEVICES OR ARRANGEMENTS FOR THE CONTROL OF LIGHT BY MODIFICATION OF THE OPTICAL PROPERTIES OF THE MEDIA OF THE ELEMENTS INVOLVED THEREIN; NON-LINEAR OPTICS; FREQUENCY-CHANGING OF LIGHT; OPTICAL LOGIC ELEMENTS; OPTICAL ANALOGUE/DIGITAL CONVERTERS
- G02F7/00—Optical analogue/digital converters
Abstract
The invention discloses a parallel optical analog-to-digital conversion system and a method based on a convolution cycle automatic encoder, wherein the system comprises a parallel optical analog-to-digital converter module, a digital signal processor module and a convolution cycle automatic encoder module; after training with one type of signals with two mismatch degrees, the convolution circulation automatic encoder module can learn system characteristics, namely time mismatch, of the parallel optical analog-to-digital converter module, and map the mismatch signals into high-quality digital signals in a system mismatch-free state; and the convolutional loop automatic encoder module after training of the signals of the two mismatch degrees can realize the mismatch compensation of various signals of various mismatch degrees. The invention can be widely applied to the improvement of the performance of the parallel optical analog-to-digital converter, and has important significance for the performance improvement and the capability expansion of the contemporary information processing system such as microwave photon radar, optical communication and the like which needs to realize high-frequency, large-bandwidth and high-precision sampling.
Description
Technical Field
The invention belongs to an optical signal processing technology and a deep learning technology, and particularly relates to a parallel optical analog-to-digital conversion system and a method based on a convolution cycle automatic encoder.
Background
With the development of information technology, electromagnetic signals used in the fields of modern information processing systems such as communication, radar, electronic countermeasure and the like generally have characteristics of high frequency band and large bandwidth. To convert these high frequency, large bandwidth analog signals to digital domain processing, an analog-to-digital converter is typically employed for analog-to-digital conversion. Although the electrical analog-digital converter has the advantages of high quantization precision, stable working state and the like, the electrical analog-digital converter is limited by time jitter and bandwidth of electrical devices, and the electrical analog-digital converter cannot digitize analog signals with high frequency and large bandwidth. The optical analog-to-digital converter can realize ultra-high sampling rate and ultra-wideband reception by utilizing the advantage of large bandwidth of light. An optical analog-to-digital converter for optical sampling electric quantization has both the advantages of large bandwidth of light and high precision of electric quantization, so that the optical analog-to-digital converter is widely researched and applied. To reduce the sampling rate of the electrical analog-to-digital converter and the bandwidth of the optical-to-electrical converter, the optical analog-to-digital converter usually adopts a parallel structure. However, since there is usually a hardware deviation between multiple parallel channels, for example, the physical lengths of the channels are not consistent, the signal obtained after the multi-output interleaving of the system contains a mismatch component with frequency fs/2-f, where fs is the sampling rate of the optical analog-to-digital conversion system, and f is the frequency of the sampled signal. Therefore, the performance of the optical analog-to-digital converter in a parallel structure is affected by hardware skew between the respective parallel channels. For example, in an all-optical coherent radar system, the spurious-free dynamic range of the system is limited by the different delays between the four channels of the photon receiving end (optical analog-to-digital converter) [ Ghelfi, p., Laghezza, f., scott, f.et al.a full photons-based coherent radiation system. nature vol.507, pp.341-345, 2014 ].
The convolutional neural network greatly improves the accuracy rate of picture recognition since 2012, people pay more attention to the deep neural network, deep learning becomes a research hotspot, and related technologies of the deep neural network are rapidly developed. In recent years, various novel deep learning network architectures such as a recurrent neural network, a generative countermeasure network and a time convolution network are proposed continuously, various optimization algorithms such as a stochastic gradient descent algorithm are improved continuously, and various activation functions such as Relu, Sigmoid and Tanh are explored continuously. This provides very good support for the ability and the application development of deep learning. A multilayer neural network can approximate any function according to the neural network approximation theory [ artemiga, C., Marreo, I.Application by neural networks with weights varying on a fine set of directionals.Neural Net., Vol.46, pp.299-305,2013 ]. Applications of deep learning have been extended from the initial fields of computer vision, speech recognition, etc. to many fields of medical diagnosis, games, optical system design, mobile communication, etc. Deep learning techniques are a potential method of compensating for mismatches between the individual parallel channels in a parallel optical analog-to-digital converter. The method is expected to realize mismatch compensation through function mapping between the digital signal output by the optical analog-to-digital conversion system and the ideal output signal of the system, so that a high-quality digital signal without mismatch components is obtained. Compared with the traditional method of firstly evaluating and then correcting the mismatch amount through a complex algorithm, the method is simple and efficient. Previous work has implemented compensation for time mismatch in parallel optical analog-to-digital conversion systems [ Xu, s.et al. deep-learning-powered analog-to-digital conversion. light sci. appl. vol.8, No.66, 2019] by using a residual neural network, however, the residual network is only valid for signals of trained degree of mismatch and not for signals of untrained degree of mismatch. According to the theory of 'free lunch in the world' in machine learning [ Wolpert, D.H., Macready, W.G.No free channels for optimization. IEEE T.Evolut.Compout.Vol.1, pp,67-82,1997], all the neural networks have a certain input effective domain, and different networks have different degrees of capabilities. This difference in capabilities depends on the architectural choice of the network and the hyper-parameters in the network such as learning rate, number of network layers, number of neurons per layer settings. Most networks can only learn some individual characteristics of each group of data by training, and cannot obtain hidden common characteristics and potential intrinsic rules in the data group. Therefore, most networks are ineffective for signals with features outside the training set. If the network is to be implemented to be valid for a certain degree of mismatch, the network needs to be trained with the corresponding degree of mismatch. Therefore, if the residual error network is used to compensate for the mismatch of signals with various degrees of mismatch, the residual error network needs to be trained by using signals with corresponding degrees of mismatch. In this case, the data set of the network would be very large and the collection of the data set would be very cumbersome and cumbersome.
Disclosure of Invention
To overcome the disadvantages of the prior art, it is an object of the present invention to provide a parallel optical analog-to-digital conversion system and method based on a convolution loop automatic encoder, which can solve the above problems.
The design principle is as follows: a novel deep neural network is based on characteristic learning of an optical analog-to-digital conversion system, can learn and compensate time mismatch in the system, and realizes a high-performance parallel optical analog-to-digital conversion system; specifically, a convolution cycle automatic encoder is trained to realize function mapping between channel mismatch and parallel optical analog-to-digital conversion system output without channel mismatch. After the mismatched signal output by the channel mismatched parallel optical analog-digital conversion system is input into the trained convolution cycle automatic encoder, the mismatched component in the signal is removed, so that the microwave signal is converted into a high-quality digital signal after passing through the parallel optical analog-digital conversion system based on the convolution cycle automatic encoder.
The technical scheme is as follows: the purpose of the invention is realized by adopting the following technical scheme.
A parallel optical analog-to-digital conversion system based on a convolution cycle automatic encoder comprises a parallel optical analog-to-digital converter module, a digital signal processor module and a convolution cycle automatic encoder module; the output end of the parallel optical analog-to-digital converter module is connected with the input end of the digital signal processor module, and the output end of the digital signal processor module is connected with the input end of the convolution circulation automatic encoder module.
The parallel optical analog-digital converter module comprises a laser for generating a light source, a microwave source for generating a microwave signal, an electro-optical converter for modulating the microwave signal to light, a serial-parallel converter for converting a single-path optical signal into a multi-path parallel signal, an adjustable delay line for adjusting delay of each channel, and an electric analog-digital converter for converting the optical signal into an electric signal and converting an analog electric signal into a digital signal; the optical analog-to-digital converter module converts the microwave signals of the laser source into a plurality of parallel digital signals through parallel channels.
The digital signal processor module carries out channel interleaving and data segmentation preprocessing operations on the digital signals output by the parallel optical analog-to-digital converter module and outputs mismatch signals to the convolution circulation automatic encoder module.
The convolution cycle automatic encoder comprises a convolution neural network, a cycle neural network and an automatic encoder; the convolutional neural network extracts high-dimensional and abstract features from the data, the cyclic neural network is used for processing the time sequence relation in the data, and the automatic encoder extracts key features in the data by utilizing an encoding and decoding structure; and the convolution circulation automatic encoder module receives the mismatch signal transmitted by the digital signal processor module to train to obtain a trained digital signal, so that the mismatch compensation of various signals with various mismatch degrees is realized.
Preferably, the parallel optical analog-to-digital converter module is an optical analog-to-digital converter with a plurality of multi-channel parallel channels with the same structure and different time delays
Preferably, after training with one type of signals with two mismatch degrees, the convolution cycle automatic encoder module can learn the system characteristics, namely time mismatch, of the parallel optical analog-to-digital converter module, and map the mismatch signals into high-quality digital signals in a system mismatch-free state; and the convolution circulation automatic encoder module trained by the signals of one class with two mismatch degrees can realize the mismatch compensation of the signals of multiple mismatch degrees.
The method for converting the optical modulus by adopting the conversion system comprises the following steps:
1) a training stage:
when the parallel optical analog-digital converter module is under the condition of mismatch of two different degrees, microwave signals are modulated onto light through electro-optical conversion in the parallel optical analog-digital converter module, and after multi-path parallel output of the parallel optical analog-digital converter module is subjected to interleaving and segmentation pretreatment by the digital signal processor module, two groups of signals with mismatch of different degrees are obtained; the obtained mismatch signal is used as a training set of the convolution cycle automatic encoder module, and the convolution cycle automatic encoder module is trained by utilizing an optimization algorithm, so that the convolution cycle automatic encoder module can learn the system characteristics of the parallel optical analog-to-digital converter module, namely time mismatch, and can compensate the mismatch signal to output a high-quality digital signal without mismatch components;
2) an application stage:
microwave signals are input into the parallel optical analog-to-digital converter module, then are subjected to interleaving and segmentation pretreatment by the digital signal processor module and then are converted into digital signals, the digital signals are input into the convolution cycle automatic encoder module trained in the training stage, and high-quality digital signals without mismatch components are output through a network.
Compared with the prior art, the invention has the beneficial effects that:
1. a convolution cycle automatic encoder is adopted to learn and correct system defects, namely time mismatch, of the parallel optical analog-to-digital converter. The parallel analog-to-digital conversion system based on the convolution circulation automatic encoder is not influenced by hardware deviation among all parallel channels, and high-quality digital signals are output.
2. The convolution circulation automatic encoder is an end-to-end, simple, fast, efficient and generalized mismatch compensation method based on system feature learning.
3. The convolution cycle automatic encoder is trained by using one type of signals with mismatch of two degrees, and the trained convolution cycle automatic encoder can compensate the mismatch of multiple types of signals with mismatch of multiple degrees.
Drawings
FIG. 1 is a schematic structural diagram of a parallel optical analog-to-digital conversion system based on a convolution loop automatic encoder according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a parallel optical analog-to-digital converter;
FIG. 3 is a schematic diagram of an embodiment of an automatic convolution cyclic encoder;
fig. 4 is a performance demonstration diagram of a parallel optical analog-to-digital converter and a parallel optical analog-to-digital conversion system based on a convolution-cycle auto-encoder.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
Referring to fig. 1-3, a parallel optical analog-to-digital conversion system based on a convolution loop auto-encoder is characterized in that: the system comprises a parallel optical analog-to-digital converter module 1, a digital signal processor module 2 and a convolution cycle automatic encoder module 3.
Connection relation: the output end of the parallel optical analog-to-digital converter module 1 is connected with the input end of the digital signal processor module 2, and the output end of the digital signal processor module 2 is connected with the input end of the convolution cycle automatic encoder module 3. The construction process is divided into a training phase and an application phase. In the training stage, the multiple groups of outputs of the parallel optical analog-to-digital converter module 1 are preprocessed by the digital signal processor module 2 and then train the convolution cycle automatic encoder module 3, so that the output mismatch compensation of the parallel optical analog-to-digital converter module 1 can be realized. In the application stage, the mismatch signal output by the parallel optical analog-to-digital converter module 1 is input into the trained convolution cycle automatic encoder module 3 in the training stage, and a high-quality digital signal without mismatch components is obtained. The structural schematic diagram of the parallel optical analog-to-digital converter module 1 is shown in fig. 2. In the embodiment, the parallel optical analog-to-digital converter module 1 adopts a dual-channel parallel optical analog-to-digital converter device with a parallel demultiplexing structure and a sampling rate of 20 GSa/s. In the system, a laser is used as a light source, a broadband microwave signal generated by a microwave source is modulated onto an optical pulse string after electro-optical conversion, then the optical pulse string is subjected to serial conversion and is divided into a plurality of sub-pulses to enter parallel channels, the sub-pulses of each parallel channel are converted into electric signals after photoelectric conversion, and then the electric signals are converted into digital signals through an electric analog-digital converter. Before the photoelectric conversion, the channel delay can be changed by adjusting the adjustable delay line of each parallel channel, so that the parallel optical analog-to-digital converter module 1 is in a channel mismatch state and a non-channel mismatch state.
The digital signal processor module 2 performs channel interleaving and data segmentation preprocessing operations on the digital signals output by the parallel optical analog-to-digital converter module 1. The output of the digital signal processor module 2 is used to train the convolution loop autoencoder module 3.
The parallel optical analog-to-digital converter module 1 comprises a laser for generating a light source, a microwave source for generating a microwave signal, an electro-optical converter for modulating the microwave signal to light, a serial-to-parallel converter for converting a single-path optical signal into a plurality of paths of parallel signals, an adjustable delay line for adjusting delay of each channel, and an electric analog-to-digital converter for realizing photoelectric conversion for converting the optical signal into an electric signal and converting an analog electric signal into a digital signal; the optical analog-to-digital converter module 1 converts the microwave signal of the laser source into a plurality of parallel digital signals through parallel channels.
The digital signal processor module 2 performs channel interleaving and data segmentation preprocessing operations on the digital signal output by the parallel optical analog-to-digital converter module 1, and outputs a mismatch signal to the convolution cycle automatic encoder module 3.
Wherein, the convolution cycle automatic encoder 3 comprises a convolution neural network, a cycle neural network and an automatic encoder; the convolutional neural network extracts high-dimensional and abstract features from the data, the cyclic neural network is used for processing the time sequence relation in the data, and the automatic encoder extracts key features in the data by utilizing an encoding and decoding structure; the convolution circulation automatic encoder module 3 receives the mismatch signal transmitted by the digital signal processor module 2 to train and obtain the trained digital signal, and realizes the mismatch compensation of various signals with various mismatch degrees.
Specifically, referring to fig. 3, the schematic structural diagram of the convolutional automatic encoder module 3 is that the convolutional automatic encoder module 3 includes an input layer, four convolutional neural network layers, two pooling layers, a cyclic neural network layer, three deconvolution neural network layers, and an output layer, where the fourth convolutional neural network layer and the cyclic neural network layer, and the cyclic neural network layer and the first deconvolution neural network layer respectively implement encoding and decoding functions to form an automatic encoder structure. In the embodiment, the parallel optical analog-to-digital converter module 1 is a serial-to-parallel double-balanced output modulator, the adjustable delay line is a digital adjustable delay line, and the electric analog-to-digital converter is an oscilloscope. The digital signal processor module 2 and the convolution loop autoencoder module 3 used in the embodiment described are implemented in a personal computer.
Further, the parallel optical analog-to-digital converter module 1 is an optical analog-to-digital converter having a plurality of multi-channel parallel channels with the same structure and different time delays.
Further, the serial conversion in the parallel optical analog-to-digital converter module 1 is a wavelength division multiplexer or a double balanced output modulator.
Further, an electrical analog-to-digital converter in the parallel optical analog-to-digital converter module 1 is a data acquisition board card or an oscilloscope.
Further, the digital signal processor module 2 is an FPGA or a DSP.
Further, after training with one type of signal of two mismatch degrees, the convolution cycle automatic encoder module 3 can learn the system characteristics, namely time mismatch, of the parallel optical analog-to-digital converter module 1, and map the digital signal into a high-quality digital signal in a system mismatch-free state; and the convolution circulation automatic encoder module 3 trained by the signals of one class with two mismatch degrees can realize the mismatch compensation of the signals of multiple mismatch degrees.
The conversion method using the above system, i.e., the module connection and the module function at two stages in the construction process, is described as follows.
A training stage:
firstly, a microwave source generates a group of broadband linear frequency modulation signals arbitrarily, and the signals are modulated onto light after being subjected to photoelectric conversion in the parallel optical analog-to-digital converter module 1; adjusting digital adjustable delay lines of each parallel channel in the parallel optical analog-to-digital converter module 1, and when a plurality of paths of parallel signals output by the system are interleaved in the digital signal processor module 2 and then the frequency spectrum does not contain mismatch components, it can be considered that there is no hardware deviation between the channels of the parallel optical analog-to-digital converter module 1, that is, the system is in a 0ps mismatch state. Secondly, 3000 groups of broadband linear frequency modulation signals with the frequency of 2.0-3.3GHz and 3000 groups of broadband linear frequency modulation signals with the frequency of 7.0-8.3GHz generated by the microwave source are input into the parallel optical analog-to-digital converter module 1 in a 0ps mismatch state, and after multi-path parallel output is interleaved and segmented in the digital signal processor module 2, the linear frequency modulation signals with 0ps mismatch are obtained. The segmentation operation in the digital signal processor module 2 is to divide the signal with the length of 20000 points into 200 segments after interleaving, and each segment has 100 points. Thirdly, the delay amount of a certain channel is increased by 35ps and 57ps by adjusting the digital adjustable delay line of the certain channel in the parallel optical analog-to-digital converter module 1. Fourthly, the microwave source generates the same 3000 groups of broadband linear frequency modulation signals with the frequency of 2.0-3.3GHz and 3000 groups of broadband linear frequency modulation signals with the frequency of 7.0-8.3GHz, the signals are input into the optical analog-to-digital converter module 1 with the mismatch degree of 35ps and 57ps and then are modulated onto light after being subjected to optical-to-electrical conversion, and the multipath parallel output of the optical analog-to-digital converter module 1 is interleaved and segmented in the digital signal processor module 2 to obtain the linear frequency modulation signals with the mismatch of 35ps and 57 ps. Fifthly, the 0ps mismatched linear frequency modulation signal is used as a network reference of the 35ps and 57ps mismatched linear frequency modulation signals. And sixthly, linear frequency modulation signals with mismatched 35ps and 57ps are input into the convolution cycle automatic encoder module 3, an Adam optimization algorithm is selected to reduce the absolute error between the network input and the network reference, and the learning rate of the optimization algorithm is 0.001. In the 20000 iterations, the network parameters are continuously adjusted, and the error finally converges to 0.003, so that the convolution loop automatic encoder module 3 learns the system characteristics, i.e. the time mismatch, in the parallel optical analog-to-digital converter module 1, and can compensate the mismatch and output a high-quality digital signal without mismatch components.
An application stage:
the microwave source generates a broadband linear frequency modulated signal or other type of signal such as Costas frequency modulated signal, which is input to the parallel optical analog-to-digital converter module 1 under other mismatch conditions. After being interleaved and segmented in the digital signal processor module 2, the multi-channel digital signals are input into the trained convolution cycle automatic encoder module 3, and high-quality digital signals without mismatch components are output through a network.
Fig. 4 is a performance demonstration of a parallel optical analog-to-digital converter and a convolution-cycle auto-encoder based parallel optical analog-to-digital converter. A and b are respectively frequency spectrums output after a linear frequency modulation signal is input into a parallel optical analog-to-digital converter and a parallel optical analog-to-digital conversion system based on a convolution cycle automatic encoder when the mismatch amount between all parallel channels is 92 ps; and c and d are spectrograms output after a Costas frequency modulation signal is input into the parallel optical analog-to-digital converter and the parallel optical analog-to-digital conversion system based on the convolution cycle automatic encoder when the mismatch amount between the parallel channels is 127 ps. Therefore, the convolution circulation automatic encoder can effectively compensate time mismatch, and a parallel optical analog-to-digital conversion system based on the convolution circulation automatic encoder outputs high-quality digital signals.
The invention can be widely applied to the improvement of the performance of the parallel optical analog-to-digital converter, and has important significance for the performance improvement and the capability expansion of the contemporary information processing system such as microwave photon radar, optical communication and the like which needs to realize high-frequency, large-bandwidth and high-precision sampling.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (7)
1. A parallel optical modulus conversion system based on a convolution cycle automatic encoder is characterized in that: the system comprises a parallel optical analog-to-digital converter module (1), a digital signal processor module (2) and a convolution cycle automatic encoder module (3); the output end of the parallel optical analog-to-digital converter module (1) is connected with the input end of the digital signal processor module (2), and the output end of the digital signal processor module (2) is connected with the input end of the convolution cycle automatic encoder module (3);
the parallel optical analog-digital converter module (1) comprises a laser used for generating a light source, a microwave source used for generating a microwave signal, an electro-optical converter used for modulating the microwave signal to light, a serial-parallel converter used for converting a single-path optical signal into a multi-path parallel signal, an adjustable delay line used for adjusting delay of each channel, and an electric analog-digital converter used for realizing photoelectric conversion used for converting the optical signal into an electric signal and converting an analog electric signal into a digital signal; the parallel optical analog-digital converter module (1) converts and outputs microwave signals of the laser source into a plurality of parallel digital signals through parallel channels;
the digital signal processor module (2) carries out channel interleaving and data segmentation preprocessing operations on the digital signals output by the parallel optical analog-to-digital converter module (1), and outputs mismatch signals to the convolution circulation automatic encoder module (3);
the convolution cycle automatic encoder module (3) comprises a convolution neural network, a cycle neural network and an automatic encoder; the convolutional neural network extracts high-dimensional and abstract features from the data, the cyclic neural network is used for processing the time sequence relation in the data, and the automatic encoder extracts key features in the data by utilizing an encoding and decoding structure; and the convolution cycle automatic encoder module (3) receives the mismatch signal transmitted by the digital signal processor module (2) to train to obtain a trained convolution cycle automatic encoder, so that the mismatch compensation of various signals with various mismatch degrees is realized.
2. The parallel optical analog-to-digital conversion system of claim 1, wherein: the parallel optical analog-to-digital converter module (1) is an optical analog-to-digital converter with a plurality of multi-channel parallel channels which have the same structure and different time delays.
3. The parallel optical analog-to-digital conversion system of claim 1, wherein: and the serial conversion in the parallel optical analog-to-digital converter module (1) is a wavelength division multiplexer or a double-balanced output modulator.
4. The parallel optical analog-to-digital conversion system of claim 1, wherein: the electric analog-digital converter in the parallel optical analog-digital converter module (1) is a data acquisition board card or an oscilloscope.
5. The parallel optical analog-to-digital conversion system of claim 1, wherein: the digital signal processor module (2) is an FPGA or a DSP.
6. The parallel optical analog-to-digital conversion system of claim 1, wherein: after training with one type of signals with two mismatch degrees, the convolution cycle automatic encoder module (3) can learn the system characteristics, namely time mismatch, of the parallel optical analog-to-digital converter module (1), and map the mismatch signals into high-quality digital signals in a system mismatch-free state; and the convolution circulation automatic encoder module (3) trained by one type of signals with two mismatch degrees can realize the mismatch compensation of multiple types of signals with multiple mismatch degrees.
7. A method of optical analog-to-digital conversion using a conversion system as claimed in any one of claims 1 to 6, characterized in that the method comprises the steps of:
1) a training stage:
when the parallel optical analog-to-digital converter module (1) is under the condition of mismatch of two different degrees, microwave signals are modulated onto light through electro-optical conversion in the parallel optical analog-to-digital converter module (1), and after multi-path parallel output of the parallel optical analog-to-digital converter module (1) is subjected to interleaving and segmentation preprocessing by the digital signal processor module (2), two groups of signals with mismatch of different degrees are obtained; the obtained mismatch signal is used as a training set of the convolution cycle automatic encoder module (3), and the convolution cycle automatic encoder module (3) is trained by using an optimization algorithm, so that the convolution cycle automatic encoder module (3) can learn the system characteristics, namely time mismatch, of the parallel optical analog-to-digital converter module (1), and the mismatch signal can be compensated to output a high-quality digital signal without mismatch components;
2) an application stage:
microwave signals are input into the parallel optical analog-to-digital converter module (1), then are subjected to interleaving and segmentation pretreatment by the digital signal processor module (2) and then are converted into mismatch signals, the mismatch signals are input into the convolution cycle automatic encoder module (3) trained in the training stage, and high-quality digital signals without mismatch components are output through a network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010673310.5A CN111650803B (en) | 2020-07-14 | 2020-07-14 | Parallel optical analog-to-digital conversion system and method based on convolution circulation automatic encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010673310.5A CN111650803B (en) | 2020-07-14 | 2020-07-14 | Parallel optical analog-to-digital conversion system and method based on convolution circulation automatic encoder |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111650803A CN111650803A (en) | 2020-09-11 |
| CN111650803B true CN111650803B (en) | 2021-07-27 |
Family
ID=72346321
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010673310.5A Active CN111650803B (en) | 2020-07-14 | 2020-07-14 | Parallel optical analog-to-digital conversion system and method based on convolution circulation automatic encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111650803B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH02156232A (en) * | 1988-12-08 | 1990-06-15 | Nec Corp | Optical digital-to-analog converter |
| JP2008185867A (en) * | 2007-01-31 | 2008-08-14 | Japan Science & Technology Agency | Optical waveform digitizer |
| CN104242933A (en) * | 2013-08-22 | 2014-12-24 | 西安电子科技大学 | Digital background calibration method for high-speed analog-digital converter |
| CN106444216A (en) * | 2016-08-31 | 2017-02-22 | 上海交通大学 | Broadband signal acquisition channel mismatch correction method in multichannel analog-to-digital conversion system |
| US9634679B1 (en) * | 2016-08-19 | 2017-04-25 | Guzik Technical Enterprises | Digital down converter with equalization |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008156400A1 (en) * | 2007-06-21 | 2008-12-24 | Signal Processing Devices Sweden Ab | Compensation of mismatch errors in a time-interleaved analog-to-digital converter |
-
2020
- 2020-07-14 CN CN202010673310.5A patent/CN111650803B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH02156232A (en) * | 1988-12-08 | 1990-06-15 | Nec Corp | Optical digital-to-analog converter |
| JP2008185867A (en) * | 2007-01-31 | 2008-08-14 | Japan Science & Technology Agency | Optical waveform digitizer |
| CN104242933A (en) * | 2013-08-22 | 2014-12-24 | 西安电子科技大学 | Digital background calibration method for high-speed analog-digital converter |
| US9634679B1 (en) * | 2016-08-19 | 2017-04-25 | Guzik Technical Enterprises | Digital down converter with equalization |
| CN106444216A (en) * | 2016-08-31 | 2017-02-22 | 上海交通大学 | Broadband signal acquisition channel mismatch correction method in multichannel analog-to-digital conversion system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111650803A (en) | 2020-09-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109828421B (en) | Photon analog-to-digital conversion method and system based on intensity adjustment and differential coding technology | |
| CN108375861B (en) | High-speed high-precision optical analog-to-digital conversion device and method capable of realizing intelligent signal processing | |
| CN106990642B (en) | Optical analog to digital conversion device based on modulator multichannel demultiplexing | |
| CN111884727B (en) | High-speed photon digital-to-analog conversion method and system based on digital mapping | |
| CN105319798B (en) | Sample rate presses the optics analog-digital commutator of 2 any power restructural | |
| CN109254471B (en) | Photon analog-to-digital conversion method and system with improved bit precision | |
| CN105372902A (en) | High speed reconstructible optical analog-to-digital conversion apparatus | |
| Xu et al. | Photonic-assisted time-interleaved ADC based on optical delay line | |
| CN111650803B (en) | Parallel optical analog-to-digital conversion system and method based on convolution circulation automatic encoder | |
| CN113359370B (en) | Optical digital-to-analog conversion method and device | |
| CN110995270A (en) | Sectional type photon digital-to-analog converter and waveform generation method thereof | |
| CN112684650B (en) | Photon analog-to-digital conversion method and system based on weighted modulation curve | |
| CN108259090B (en) | Radio frequency arbitrary waveform light generation method and system based on digital logic operation | |
| CN107135005B (en) | Ultra-wideband signal multi-path parallel compression sampling method based on photoelectric combination | |
| Zou et al. | Photonic analog-to-digital converter powered by a generalized and robust convolutional recurrent autoencoder | |
| CN113238428B (en) | High-speed photon digital-to-analog conversion method based on dual-drive electro-optical modulator array | |
| CN106506088A (en) | A kind of carrier aggregation based on single sideband modulation and the method and system for depolymerizing | |
| Yang et al. | Photonic quantization and encoding scheme with improved bit resolution based on waveform folding | |
| CN111679530B (en) | Photon time delay stretching analog-to-digital conversion method and system based on radio frequency signal | |
| CN111835366A (en) | Parallel signal processing device and method based on convolution circulation automatic encoder | |
| CN111045275A (en) | Photon analog-to-digital conversion system and method based on hierarchical quantization principle | |
| CN109884839B (en) | Photon analog-to-digital conversion system and method based on asymmetric digital coding scheme | |
| CN114265261B (en) | High-speed photon analog-to-digital conversion method and system based on pulse processing | |
| CN113794547B (en) | Multipath signal synchronization method, system, electronic equipment and computer readable storage medium | |
| CN111478729B (en) | Method for testing performance of demultiplexing module in optical analog-to-digital conversion system |
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 |