Single-frame measurement method of ultra-wideband arbitrary optical waveform based on neural network

Abstract

The neural network-based ultra-wideband arbitrary optical waveform single-frame measurement method belongs to the field of ultrafast laser technology. The specific steps are: generating the optical waveform to be measured, linear spectrum slicing processing, multi-channel photoelectric detection and data acquisition, building a neural network model and data set, training the neural network model, and reconstructing the amplitude and phase of the arbitrary optical waveform to be measured. The present invention realizes spectrum slicing and multi-channel parallel processing of the optical waveform to be measured through linear spectrum slicing processing, which can effectively reduce the complexity of the system and reduce the demand for analog bandwidth of photoelectric detection and data acquisition; realizes rapid reconstruction measurement of the amplitude and phase of the optical waveform to be measured through neural network model calculation, realizes rapid measurement of a single frame of arbitrary optical waveform and significantly improves measurement sensitivity and flexibility; based on commonly used optical filters, ultra-wideband spectrum slicing parallel output can be realized, which has the advantages of simple method, high efficiency and good stability.

Description

Technical Field

[0001] The invention belongs to the technical field of ultrafast lasers and relates to a single-frame measurement method for ultra-wideband arbitrary optical waveforms based on a neural network. Background Art

[0002] Measuring the amplitude and phase of arbitrary ultra-wideband optical waveforms plays a vital role in ultrafast optics, quantum information processing, spectral measurements, microwave photonics, optical communications, and other fields. Existing electronic devices such as photodetectors and oscilloscopes have analog bandwidths of only tens of GHz, insufficient for measuring picosecond and femtosecond ultrafast laser pulses. It is also difficult to directly and simultaneously obtain the amplitude and phase information of ultra-wideband optical waveforms. Commonly used optical waveform measurement methods, such as frequency-resolved optical gating (FROG), utilize nonlinear effects and phase recovery algorithms to measure ultrashort pulses.

[0003] Chinese invention patent CN112595425B discloses an ultrashort laser pulse measurement method and measurement system, which solves the technical problem that existing ultrashort laser pulse measurement methods have high requirements for nonlinear crystal processing. However, the system structure is complex and still requires the use of phase extraction algorithms such as iterative calculations to achieve pulse phase reconstruction. Chinese invention patent CN114216575B proposes a FROG ultrashort pulse reconstruction system and method based on a BP neural network, which uses a neural network instead of a traditional algorithm to calculate amplitude and phase data. However, the above method still relies on nonlinear effects to achieve pulse measurement, the system structure is complex and the pulse to be measured needs to be amplified, which limits the actual application scenarios.

[0004] In summary, it is of great significance to study a single-frame measurement method for ultra-wideband arbitrary optical waveforms that does not require nonlinear effects and has a simple structure and method. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention proposes a single-frame measurement method for ultra-wideband arbitrary optical waveforms based on neural networks, which realizes rapid reconstruction and measurement of the amplitude and phase of arbitrary optical waveforms through linear spectrum slicing processing, multi-channel photoelectric detection and deep learning.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The neural network-based single-frame measurement method for ultra-wideband arbitrary optical waveforms includes the following steps:

[0008] The first step is to generate the optical waveform to be measured.

[0009] Laser pulses are generated by a femtosecond pulse laser, and the laser pulse spectrum is amplitude- and phase-modulated by a programmable optical waveform shaper to produce a measured optical waveform. The measured optical waveform refers to a time-domain optical waveform. The amplitude and phase modulation are designed to impart arbitrary amplitude and phase to the measured optical waveform in the time domain. The amplitude and phase modulation signals in the measured optical waveform can be randomly generated Gaussian white noise signals or other randomly generated arbitrary waveform signals.

[0010] The second step is linear spectrum slicing processing.

[0011] The optical waveform to be measured passes through a 1×N splitter and N optical filters with different central wavelengths, or passes through a 1×N multiplexer for linear spectrum slicing processing, and then outputs N-channel spectrum slice signals in parallel.

[0012] In order to achieve gapless spectrum slicing, the optical filter or multiplexer should have different central wavelengths in different channels, and the passband frequency coverage ranges of channels with adjacent central wavelengths overlap with each other.

[0013] The number N of channels of the optical filter or multiplexer is determined by the spectrum bandwidth corresponding to the optical waveform to be measured and the operating bandwidth of the optical filter or multiplexer.

[0014] The third step is multi-channel photoelectric detection and data acquisition.

[0015] The N-channel spectrum slice signals output in parallel in the second step are subjected to N-channel photoelectric detection and data acquisition to obtain time domain intensity waveform data of the N-channel spectrum slice signals.

[0016] Alternatively, the spectrum slice signals of each channel output in parallel in the second step are sequentially introduced with time delays and synthesized into a single output. Through single-channel photoelectric detection and data acquisition, serial detection and acquisition of multi-channel spectrum slice signals are realized to obtain time domain intensity waveform data, which can effectively reduce the number of required photoelectric detection channels.

[0017] The fourth step is to build a neural network model and data set.

[0018] The neural network model can include a fully connected neural network model or a convolutional neural network model. For a specific optical waveform to be measured, generated in the first step, its amplitude and phase are used as output labels, and the time-domain intensity waveform data obtained in the third step is used as input to the neural network model. The neural network model outputs the predicted optical waveform amplitude and phase, which serves as the reconstruction result.

[0019] By randomly generating a large number of Gaussian white noise signals or arbitrary waveforms in the first step, a large number of optical waveforms to be measured are generated, and a data set is constructed for subsequent neural network model training.

[0020] The fifth step is to train the neural network model.

[0021] The neural network model constructed in step 4 is trained using the dataset. Backpropagation and an optimizer are used to train the neural network model, minimizing the error between the reconstruction results and the labels and adjusting the neural network model parameters. Through multiple iterations of training, the neural network model effectively learns the mapping relationship between the time-domain intensity waveform data and the measured optical waveform, ultimately achieving accurate reconstruction of the optical waveform amplitude and phase, resulting in a fully trained neural network model.

[0022] Step 6: Reconstruct and measure the amplitude and phase of any optical waveform to be measured.

[0023] For any optical waveform to be measured, the second and third steps are used to process the time-domain intensity waveform data, which is then input into the neural network model trained in the fifth step. The predicted optical waveform amplitude and phase are output, ultimately achieving the reconstruction and measurement of the amplitude and phase of any optical waveform to be measured.

[0024] The beneficial effects of the present invention are:

[0025] (1) The present invention realizes spectrum slicing and multi-channel parallel processing of the optical waveform to be measured through linear spectrum slicing processing, avoiding the use of nonlinear effects, effectively reducing system complexity, and effectively reducing the demand for analog bandwidth of photoelectric detection and data acquisition.

[0026] (2) The present invention realizes rapid reconstruction measurement of the amplitude and phase of the optical waveform to be measured through neural network model calculation, avoiding the use of complex coherent receiving systems and phase recovery algorithms. It can realize rapid measurement of a single frame of any optical waveform and significantly improve measurement sensitivity and flexibility.

[0027] (3) The present invention can realize the parallel output of ultra-wideband spectrum slices based on commonly used optical filters, and does not require other prior information such as filters and photodetector spectrum responses. Compared with traditional methods, the present invention has the advantages of simple implementation method, high efficiency and good stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart for implementing the ultra-wideband arbitrary optical waveform single-frame measurement method provided by the present invention.

[0029] Figure 2 This is a reconstructed measurement simulation result of an optical waveform to be measured in Example 1 of the present invention.

[0030] Figure 3 This is a reconstructed measurement simulation result of another optical waveform to be measured in embodiment 2 of the present invention. DETAILED DESCRIPTION

[0031] In order to make the method problems solved by the present invention, the method solutions adopted and the method effects achieved more clear, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0032] like Figure 1 As shown, the neural network-based ultra-wideband arbitrary optical waveform single-frame measurement method provided by the present invention includes the following steps:

[0033] The first step is to generate the optical waveform to be measured.

[0034] In this example, a femtosecond pulsed laser generates laser pulses with a 3-dB bandwidth of 400 GHz. These pulses are then fed into a programmable optical waveform shaper to generate the optical waveform to be measured. A MATLAB Gaussian white noise function generator is used to randomly generate amplitude- and phase-modulated signals. The programmable optical waveform shaper then modulates the laser pulse spectrum, producing an output with random amplitude and phase.

[0035] The second step is linear spectrum slicing processing.

[0036] This embodiment passes the measured optical waveform through a 1×3 splitter and three optical filters with different center wavelengths, then outputs three channels of spectrum sliced signals in parallel. To achieve seamless spectrum slicing, the optical filters have a 3-dB bandwidth of 150 GHz, and the center frequencies of filters in adjacent channels differ by 125 GHz. By slicing and processing the broadband spectrum of the measured optical waveform in parallel, spectral information loss due to analog bandwidth limitations of the photodetector and data acquisition is avoided, significantly reducing the bandwidth requirements for the photodetector and data acquisition.

[0037] The third step is multi-channel photoelectric detection and data acquisition.

[0038] In this embodiment, the three-channel spectrum slice signals output in parallel in the second step are subjected to three-channel photoelectric detection and data acquisition to obtain time-domain intensity waveform data of the three-channel spectrum slice signals. The analog bandwidth used for photoelectric detection and data acquisition is 30 GHz.

[0039] The fourth step is to build a neural network model and data set.

[0040] A fully connected neural network model is constructed, which includes an input layer, three fully connected layers and an output layer. Each fully connected layer contains 256 neurons, and each neuron uses a linear rectification function as the activation function.

[0041] The amplitude and phase of the measured optical waveform generated in the first step are used as output labels, and the time-domain intensity waveform data obtained in the third step is used as input to the neural network model. The neural network model outputs the predicted optical waveform amplitude and phase, which are used as the reconstruction result.

[0042] By randomly generating 5000 time-domain optical waveforms to be measured in the first step, the time-domain intensity waveform data of the corresponding 3-channel spectrum slice signals are obtained through the second and third steps to construct the data set.

[0043] The fifth step is to train the neural network model.

[0044] The neural network model constructed in step 4 is trained using the dataset. Backpropagation and the Adam optimizer are used to train the model, minimizing the mean squared error (MSE) between the reconstruction results and the labels and adjusting the model parameters. Through multiple iterations of training, the MSE is reduced to an acceptable level, ultimately achieving accurate optical waveform amplitude and phase reconstruction results and completing the trained neural network model.

[0045] Step 6: Reconstruct and measure the amplitude and phase of any optical waveform to be measured.

[0046] This embodiment generates an arbitrary optical waveform to be measured through the first step method, obtains time-domain intensity waveform data through the second and third steps, inputs it into the neural network model trained in the fifth step, and outputs the predicted optical waveform amplitude and phase, ultimately achieving the reconstruction measurement of the amplitude and phase of the arbitrary optical waveform to be measured.

[0047] Example 1

[0048] In this embodiment, a simulation result of an optical waveform to be measured is generated by the first step method. Figure 2 (dashed line), the reconstructed measurement simulation results for the optical waveform to be measured are as follows Figure 2 (solid line).

[0049] Example 2

[0050] In this embodiment, another optical waveform simulation result to be measured is generated by the first step method. Figure 3 (dashed line), the reconstructed measurement simulation results for the optical waveform to be measured are as follows Figure 3 (solid line).

[0051] It can be seen that the present invention can well complete the measurement of arbitrary optical waveforms, and based on the photoelectric detection with a bandwidth of tens of GHz, it can achieve accurate reconstruction of the amplitude and phase of optical waveforms with a bandwidth of hundreds of GHz.

[0052] The above-described embodiments merely express the implementation methods of the present invention, but should not be understood as limiting the scope of the patent of the present invention. It should be pointed out that for those skilled in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A neural network-based single-frame measurement method for ultra-wideband arbitrary optical waveforms, characterized in that: The ultra-wideband arbitrary optical waveform single-frame measurement method comprises the following steps: The first step is to generate the optical waveform to be measured; A laser pulse is generated by a femtosecond pulse laser, and the laser pulse spectrum is amplitude modulated and phase modulated by a programmable optical waveform shaper to obtain an optical waveform to be measured; the amplitude modulation signal and the phase modulation signal obtained by the amplitude modulation and the phase modulation are randomly generated Gaussian white noise signals or other randomly generated arbitrary waveform signals; The amplitude modulation and phase modulation are intended to make the optical waveform to be measured have arbitrary amplitude and phase in the time domain; The second step is linear spectrum slicing processing; The optical waveform to be measured is processed by passing it through a 1×N splitter and N optical filters with different central wavelengths, or through a 1×N multiplexer for linear spectrum slicing, and then outputting N-channel spectrum slice signals in parallel; The third step is multi-channel photoelectric detection and data acquisition; The N-channel spectrum slice signals output in parallel in the second step are subjected to N-channel photoelectric detection and data acquisition to obtain the time domain intensity waveform data of the N-channel spectrum slice signals; Alternatively, the spectrum slice signals of each channel output in parallel in the second step are sequentially introduced with time delay and synthesized into a single output. Through single-channel photoelectric detection and data acquisition, serial detection and acquisition of multi-channel spectrum slice signals are achieved to obtain time domain intensity waveform data. The fourth step is to build a neural network model and data set; The neural network model includes a fully connected neural network model or a convolutional neural network model; for a specific optical waveform to be measured generated in the first step, its amplitude and phase are used as output labels, and the time domain intensity waveform data obtained in the third step is used as input to the neural network model; the output of the neural network model is the predicted optical waveform amplitude and phase, which is used as the reconstruction result; The optical waveform to be measured is generated by randomly generating a Gaussian white noise signal or an arbitrary waveform signal in the first step, and a data set is constructed for subsequent neural network model training; Step 5: Train the neural network model; Using the data set constructed in the fourth step, the neural network model constructed in the fourth step is trained to obtain a trained neural network model; Step 6: Reconstruct and measure the amplitude and phase of any optical waveform to be measured; For any optical waveform to be measured, the second and third steps are used to process the time-domain intensity waveform data, which is then input into the neural network model trained in the fifth step. The predicted optical waveform amplitude and phase are output, ultimately achieving the reconstruction and measurement of the amplitude and phase of any optical waveform to be measured.

2. The neural network-based ultra-wideband arbitrary optical waveform single-frame measurement method according to claim 1, characterized in that: In the second step, the optical filter or multiplexer should have different central wavelengths in different channels, and the passband frequency coverage ranges of channels with adjacent central wavelengths overlap with each other to achieve gapless spectrum slicing.

3. The neural network-based ultra-wideband arbitrary optical waveform single-frame measurement method according to claim 1, characterized in that: In the second step, the number N of channels of the optical filter or multiplexer is determined by the spectrum bandwidth corresponding to the optical waveform to be measured and the operating bandwidth of the optical filter or multiplexer.

4. The neural network-based ultra-wideband arbitrary optical waveform single-frame measurement method according to claim 1, characterized in that: In the fifth step, the neural network model is trained using a backpropagation algorithm and an optimizer to reduce the error between the reconstruction result and the label and adjust the neural network model parameters; through multiple iterative training, accurate optical waveform amplitude and phase reconstruction results are obtained, and finally a trained neural network model is obtained.

Images