A joint uncertainty compression method and system for nuclear pulse multi-parameter measurement

Abstract

This invention provides a joint uncertainty compression method for multi-parameter measurement of nuclear pulses, comprising the following steps: S1, data acquisition and preprocessing: calculating the initial energy estimate and initial rise time estimate for each nuclear pulse waveform; S2, constructing a joint denoising model; S3, designing a joint loss function L. total S4, train the joint denoising model using nuclear pulse waveform data; S5, input the initial two-dimensional observation data into the trained joint denoising model, which outputs the accurate energy estimate and accurate rise time estimate after joint denoising. This invention also provides a system for performing this method, including a data input and preprocessing layer, an S-CVAE core model layer, a joint optimization and physical constraint layer, an output and evaluation layer, and a control and monitoring layer. This invention achieves a simultaneous breakthrough improvement in both the energy resolution and time resolution of nuclear pulses by jointly compressing the uncertainties in energy and rise time measurements.

Description

Technical Field

[0001] This invention relates to the field of nuclear electronics and signal processing technology, and in particular to a joint uncertainty compression method and system for multi-parameter measurement of nuclear pulses. Background Technology

[0002] In nuclear physics experiments, radiation detection, and energy spectrum analysis, scintillators or semiconductor detectors convert the energy of incident particles into electrical pulse signals. These pulse waveforms are typically sampled and digitized using a Flash ADC system. The digital integral and / or a function thereof under the pulse waveform are used to calculate the incident particle energy E; the time difference between the leading edge of the pulse waveform reaching a specific threshold is used to calculate the rise time T. Energy resolution ΔE / E and time resolution ΔT are core indicators for evaluating the performance of the detection system.

[0003] However, there are many uncertainties in the actual measurement process, including detector noise, electronic noise, pulse stacking effect and baseline fluctuation. These factors can affect the measurement accuracy of energy and time at the same time, resulting in a decrease in resolution.

[0004] Traditional signal processing methods typically optimize the energy and time channels independently. For energy E, analog or digital shaping techniques such as trapezoidal filtering and CR-RC4 filtering are usually used to suppress noise and optimize the signal-to-noise ratio (SNR). However, these designs are primarily geared towards energy resolution and may sacrifice time information. For rise time T, discrimination techniques such as constant ratio timing CFD and zero-crossing timing ZD are typically used to reduce time shift, but their designs are independent of the energy channel.

[0005] While this strategy achieves some success within each channel, it overlooks the inherent physical and statistical correlation between energy E and rise time T. For example, noise and baseline fluctuations simultaneously affect both the energy integral value and the timing point determination. Processing them separately fragments this joint uncertainty, leading to suboptimal filtering results and failing to extract the deepest information from the data, thus limiting further improvements in the overall system performance.

[0006] Therefore, existing technologies urgently need a novel signal processing scheme that can effectively utilize the joint statistical characteristics between energy E and rise time T to achieve joint uncertainty compression of the two key parameters, thereby simultaneously breaking through the limits of energy and time resolution. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a joint uncertainty compression method and system for multi-parameter measurement of nuclear pulses. This method solves the problem that existing technologies, by independently processing energy and time parameters and ignoring their inherent statistical correlation, restrict the simultaneous improvement of energy and time resolution.

[0008] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0009] A joint uncertainty compression method for multi-parameter measurement of nuclear pulses includes the following steps:

[0010] S1, Data Acquisition and Preprocessing: A multi-channel Flash ADC system is used to continuously sample the nuclear pulse signal output from the radiation detector to acquire nuclear pulse waveform data. Based on this data, the initial energy estimate and initial rise time estimate of each nuclear pulse waveform are calculated to form an initial two-dimensional observation data pair. S2, Construction of a Joint Denoising Model: The joint denoising model is a structured variational autoencoder, whose latent variable vector z is structured into: shared latent variable Zi shared Energy-specific latent variable Z E and time-specific latent variable Z T ; wherein, the shared latent variable Z shared The energy-specific latent variable Z is used to characterize a common noise source that simultaneously affects both energy and rise time measurements. E The time-specific latent variable Z is used to characterize noise sources that only affect energy measurements. T S3, used to characterize noise sources that only affect rise time measurements; design the joint loss function L for physical information constraints. total S4, Use the nuclear pulse waveform data obtained in step S1 to train the joint denoising model, and optimize the model parameters through the backpropagation algorithm until the joint loss function L is obtained. total Convergence; S5, the initial two-dimensional observation data pair is input into the trained joint denoising model, and the model outputs the accurate energy estimate and accurate rise time estimate after joint denoising.

[0011] The present invention also provides a system for performing a joint uncertainty compression method for nuclear pulse multi-parameter measurement, comprising a data input and preprocessing layer, an S-CVAE core model layer, a joint optimization and physical constraint layer, an output and evaluation layer, and a control and monitoring layer.

[0012] The present invention has the following beneficial effects:

[0013] 1. A joint uncertainty compression framework for nuclear pulse energy and time parameters is proposed for the first time: This invention breaks through the traditional paradigm of independently processing energy and time channels, treating energy and time as a coupled two-dimensional observation system. Through joint denoising, the model can utilize the statistical correlation between the two parameters to more effectively distinguish between real signals and random noise, thereby achieving synergistic compression of the uncertainties of both.

[0014] 2. A structured latent space was designed to achieve uncertainty decomposition and source tracing: By decomposing the latent variable vector z into z_shared, z_E, and z_T, this invention not only achieves denoising but also provides interpretable analysis of noise sources. This provides valuable diagnostic information for subsequent optimization of detector design and electronic systems.

[0015] 3. A loss function with physical information constraints is introduced: By embedding physical constraints into a purely data-driven loss function, this invention ensures that the denoised output strictly conforms to the physical laws of kernel pulse formation, avoids generating physically unreliable results, and significantly improves the robustness of the method and the reliability of the output.

[0016] 4. Achieved simultaneous improvement in energy resolution and time resolution: After processing by the method of this invention, the distribution variance of energy value and the distribution variance of rise time value are significantly compressed, thereby fundamentally and simultaneously improving the energy resolution and time resolution of the detection system. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an embodiment of the present invention; Detailed Implementation

[0018] The technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] A joint uncertainty compression method for multi-parameter measurement of nuclear pulses, such as Figure 1 As shown, it includes the following steps:

[0020] S1. Data Acquisition and Preprocessing: The nuclear pulse signal output by the radiation detector is continuously sampled using a multi-channel Flash ADC system to acquire a large amount of nuclear pulse waveform data. Based on this data, the initial energy estimate and initial rise time estimate of each nuclear pulse waveform are calculated to form an initial two-dimensional observation data pair. In this embodiment, the initial energy estimate can be obtained by digital integration, and the initial rise time estimate can be obtained by constant fraction discrimination.

[0021] S2. Constructing a Joint Denoising Model: The joint denoising model is a structured variational autoencoder (S-VAE). The joint denoising model includes: an encoder network: used to receive the input initial two-dimensional observation data pairs and output a latent variable vector z, which is structuredly decomposed into: a shared latent variable Z shared Energy-specific latent variable Z E and time-specific latent variable Z T Among them, the shared latent variable Z shared The energy-specific latent variable Z is used to characterize a common noise source that simultaneously affects both energy and rise time measurements.E The time-specific latent variable Z is used to characterize noise sources that only affect energy measurements. T Used to characterize noise sources that only affect rise time measurements. Decoder network: receives signals from a shared latent variable Z. shared Energy-specific latent variable Z E and time-specific latent variable Z T The concatenated latent variable vector z is used to reconstruct the denoised energy estimate and rise time estimate. When reconstructing the energy estimate, the decoder can simultaneously access the latent variable vector z. shared Provided common background information that is relevant to both energy and rise time, and by Z E It provides energy-specific information only. Similarly, when reconstructing the rise time estimate, it can also simultaneously access information derived from Z. shared Provided common background information that is relevant to both energy and rise time, and by Z T Provided is specific information that is only relevant to time.

[0022] S3. Design the joint loss function L for physical information constraints. total The training objective of the model is to minimize the following joint loss function L. total .

[0023] L total =L reconstruction +β*L KL +λ*L physics

[0024] in:

[0025] L reconstruction The reconstruction loss term is used to calculate the error between the denoised energy estimate, rise time estimate, and the initial two-dimensional observation data pair.

[0026] L KL The divergence loss term is used to constrain the distribution of the latent variable vector z to be close to the standard normal prior, preventing overfitting and normalizing the latent space.

[0027] L physics The physical constraint loss term applies physical regularity constraints to the denoised output based on the physical principles of pulse formation.

[0028] β is the weight hyperparameter of the KL divergence loss term, used to control the strength of latent space regularization; λ is the weight hyperparameter of the physical constraint loss term, used to control the strength of physical constraints.

[0029] (1) The reconstruction loss term is set as follows:

[0030]

[0031] Reconstruction loss measures the difference between the reconstructed data and the original input data.

[0032] The mean squared error is used as an approximation for the reconstruction loss;

[0033] Expected items This represents the log-likelihood expectation under the variational posterior distribution;

[0034] Specifically, regarding energy and rise time:

[0035]

[0036] in,

[0037] The variational posterior distribution defined for the encoder;

[0038] The likelihood distribution defined for the decoder;

[0039] Batch size;

[0040] For the input observation vector;

[0041] To reconstruct the output vector;

[0042] Let be the measured energy value of the i-th sample;

[0043] Let be the rise time of the measurement for the i-th sample;

[0044] Let be the reconstruction energy value of the i-th sample;

[0045] Let be the reconstruction rise time of the i-th sample;

[0046] For encoder parameters;

[0047] These are decoder parameters;

[0048] (2) The KL divergence loss term is set as follows:

[0049]

[0050] The KL divergence loss term is used to measure the variational posterior distribution. With prior distribution The differences between them;

[0051] Furthermore, as a regularization term, it prevents model overfitting and promotes the structuring of the latent space;

[0052] Variational posterior distribution Prior distribution ；

[0053] but,

[0054]

[0055] in,

[0056] The mean vector of the latent variables

[0057] Let the standard deviation vector of the latent variables be denoted as .

[0058] Let be the variance vector of the latent variables;

[0059] The total dimension of the potential space;

[0060] For encoder parameters;

[0061] It is the identity matrix;

[0062] It is a zero vector;

[0063] It follows a multivariate Gaussian distribution.

[0064] (3) The physical constraint loss term is set as follows:

[0065]

[0066] The physical constraint loss term embeds domain knowledge into the model and ensures the physical rationality of the output by weighted combination of multiple physical constraints.

[0067] in,

[0068] The number of physical constraints;

[0069] The weight of the k-th constraint;

[0070] For physical constraint functions;

[0071] To reconstruct the energy vector;

[0072] To reconstruct the rising time vector;

[0073] It is a pulse waveform matrix;

[0074] The waveform sampling sequence of the i-th pulse;

[0075] The sampling points for each pulse.

[0076] Physical constraint loss term Includes at least one of the following physical constraints:

[0077] 1. Energy nonnegativity constraint:

[0078]

[0079] The energy nonnegativity constraint is used to ensure that the reconstructed energy value is nonnegative, consistent with physical reality; when A penalty is imposed when <0. No penalty when ≥0.

[0080] 2. Rise time-waveform slope consistency constraint:

[0081]

[0082] in,

[0083] Waveform gradient vector ;

[0084] A proportionality constant specific to the detector;

[0085] The absolute value of the waveform gradient vector;

[0086] It represents the maximum absolute value of the waveform gradient vector.

[0087] 3. Energy-time empirical relationship constraint:

[0088]

[0089] in, For the empirical relation intercept term; This is the slope term of the empirical relationship; These are empirical rise time predictions based on energy. Based on statistical relationships from experimental observations, higher energy corresponds to shorter rise times.

[0090] 4. Waveform area-energy consistency constraint:

[0091]

[0092] in,

[0093] Integral area of ​​pulse waveform:

[0094] This is the system's energy calibration coefficient;

[0095] This represents the sampling time interval.

[0096] This constraint is based on the principle of energy conservation, where the pulse integral area is proportional to the deposited energy; this constraint is used to ensure that the reconstructed energy is consistent with the integral area of ​​the observed waveform.

[0097] S4. Model Training: Use the kernel pulse waveform data obtained in step S1 to train the joint denoising model. Optimize the model parameters using the backpropagation algorithm until the joint loss function L is reached. total convergence;

[0098] S5. Joint Uncertainty Compression and Inference: The initial two-dimensional observation data of the nuclear pulse to be processed is input into the trained joint denoising model. The model output is the accurate energy estimate and accurate rise time estimate after joint denoising, with significantly reduced uncertainty.

[0099] An experiment was conducted using a bismuth germanate (BGO) scintillation detector paired with a FLASH ADC system with a sampling rate of 1 GSps as an example.

[0100] 1. Data acquisition: 100,000 nuclear pulse signals generated by the Co-60 source were acquired.

[0101] 2. Initial parameter extraction: For each nuclear pulse signal, the initial energy estimate is obtained by digital integration; the initial rise time estimate is obtained by digital constant ratio timing method.

[0102] 3. Model Construction: Both the encoder and decoder use 3-layer fully connected neural networks, with the latent space dimension set to Z. shared 5-dimensional, Z E 3D, Z T 3D.

[0103] 4. Design a physical constraint loss term to penalize outputs with energy estimates less than 0 after reconstruction and denoising, since energy should be positive.

[0104] 5. Training: The model is trained using 90,000 data points with a batch size of 256 and 1,000 training cycles.

[0105] 6. Testing: The remaining 10,000 data points were used for testing. The results show improvements in both energy resolution and temporal resolution. Simultaneously, the two-dimensional joint distribution map exhibits a significant tightening effect.

[0106] The present invention also provides a system for performing a joint uncertainty compression method for nuclear pulse multi-parameter measurements, comprising the following parts:

[0107] Data input and preprocessing layer: used to receive raw nuclear pulse signals, calculate and output the initial energy estimate and rise time estimate of each pulse, forming the initial observation data pair;

[0108] The S-VAE core model layer includes an encoder, which receives initial observation data pairs and outputs shared latent variables, energy-specific latent variables, and time-specific latent variables; it also includes a decoder, which receives the spliced ​​shared latent variables, energy-specific latent variables, and time-specific latent variables and reconstructs the denoised energy estimate and rise time estimate.

[0109] Joint Optimization and Physical Constraint Layer: Used to calculate the joint loss function to update the S-VAE core model layer;

[0110] Output and Evaluation Layer: This layer receives and outputs the precise energy estimate and precise rise time estimate, and evaluates the improvement in energy resolution and time resolution of the output results based on the test dataset.

[0111] Control and monitoring layer: configured to perform at least one of the following operations: control system training process, including initializing model parameters, executing backpropagation algorithm and determining loss function convergence; monitor and record changes in loss value during training and evaluate the performance indicators of the layer output; schedule the preprocessing layer, feature extraction layer and S-VAE core processing layer to work together to process the pulse data to be tested.

[0112] In summary, this invention provides a systematic solution that can extract deeper information from data, break through the traditional barriers to energy and time resolution optimization, and provide core technologies for the design of next-generation high-precision radiation measurement systems.

Claims

1. A joint uncertainty compression method for multi-parameter measurement of nuclear pulses, characterized in that, Includes the following steps: S1, Data Acquisition and Preprocessing: The nuclear pulse signal output by the radiation detector is continuously sampled using a multi-channel Flash ADC system to acquire nuclear pulse waveform data. Based on this data, the initial energy estimate and initial rise time estimate of each nuclear pulse waveform are calculated to form an initial two-dimensional observation data pair. S2, Constructing a joint denoising model: The joint denoising model is a structured variational autoencoder, whose latent variable vector z is structuredly decomposed into: shared latent variable Z shared Energy-specific latent variable Z E and time-specific latent variable Z T ; wherein, the shared latent variable Z shared The energy-specific latent variable Z is used to characterize a common noise source that simultaneously affects both energy and rise time measurements. E The time-specific latent variable Z is used to characterize noise sources that only affect energy measurements. T Used to characterize noise sources that only affect rise time measurements; S3, Design the joint loss function L for physical information constraints. total ; S4, use the nuclear pulse waveform data obtained in step S1 to train the joint denoising model, and optimize the model parameters through the backpropagation algorithm until the joint loss function L is obtained. total convergence; S5, the initial two-dimensional observation data is input into the trained joint denoising model, and the model outputs the accurate energy estimate and accurate rise time estimate after joint denoising.

2. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 1, characterized in that, The structured variational autoencoder includes: An encoder network is used to receive the input initial two-dimensional observation data pairs and output a latent variable vector z to generate a shared latent variable Z. shared Energy-specific latent variable Z E and time-specific latent variable Z T ; Decoder network for receiving the shared latent variable Z shared Energy-specific latent variable Z E and time-specific latent variable Z T The concatenated latent variable vector z is used to reconstruct the denoised energy estimate and rise time estimate.

3. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 2, characterized in that, The joint denoising model minimizes the joint loss function L. total During training, the joint loss function includes at least: Reconstruction loss term L reconstruction This is used to constrain the error between the denoised energy estimate, the denoised rise time estimate, and the initial two-dimensional observation data pair in the reconstructed output. KL divergence loss term L KL This is used to constrain the distribution of the latent variable vector z to approximate a standard normal prior. Physical constraint loss term L physics Based on the physical principles of pulse formation, physical regularity constraints are applied to the denoised output; That is, L total = L reconstruction + β * L KL + λ * L physics Where β is the weight hyperparameter of the KL divergence loss term, and λ is the weight hyperparameter of the physical constraint loss term.

4. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 3, characterized in that, The physical constraint loss term L physics It is configured to penalize negative energy estimates after denoising and / or constrain the rise time estimates after denoising to be within a physically reasonable range.

5. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 3, characterized in that, The reconstruction loss term is: Using mean squared error as an approximation of the reconstruction loss term, the expected term... This represents the log-likelihood expectation under the variational posterior distribution; The variational posterior distribution defined for the encoder; The likelihood distribution defined for the decoder; Batch size; For the input observation vector; To reconstruct the output vector; Let be the measured energy value of the i-th sample; Let be the rise time of the measurement for the i-th sample; Let be the reconstruction energy value of the i-th sample; Let be the reconstruction rise time of the i-th sample; For encoder parameters; These are decoder parameters; Right now, 6. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 3, characterized in that, The KL divergence loss term is: Variational posterior distribution ; Prior distribution ; For the latent variable vector; is the mean vector of the latent variables; Let the standard deviation vector of the latent variables be denoted as . Let be the variance vector of the latent variables; The total dimension of the potential space; For encoder parameters; It is the identity matrix; It is a zero vector; It follows a multivariate Gaussian distribution.

7. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 3, characterized in that, The physical constraint loss term is: The number of physical constraints; The weight of the k-th constraint; For physical constraint functions; To reconstruct the energy vector; To reconstruct the rising time vector; It is a pulse waveform matrix.

8. A joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 7, characterized in that, The physical constraint loss term Includes at least one of the following physical constraints: Energy nonnegativity constraint, rise time-waveform slope consistency constraint, energy-time empirical relationship constraint, and waveform area-energy consistency constraint.

9. The joint uncertainty compression method for multi-parameter measurement of nuclear pulses according to claim 1, characterized in that, The initial energy estimate is obtained by numerical integration, and the initial rise time estimate is obtained by constant fraction discrimination.

10. A system for performing the method as described in any one of claims 1-9, characterized in that, Includes the following parts: Data input and preprocessing layer: used to receive raw nuclear pulse signals, calculate and output the initial energy estimate and rise time estimate of each pulse, forming the initial observation data pair; The S-VAE core model layer includes an encoder, which receives initial observation data pairs and outputs shared latent variables, energy-specific latent variables, and time-specific latent variables; it also includes a decoder, which receives the spliced ​​shared latent variables, energy-specific latent variables, and time-specific latent variables and reconstructs the denoised energy estimate and rise time estimate. Joint Optimization and Physical Constraint Layer: Used to calculate the joint loss function to update the S-VAE core model layer; Output and Evaluation Layer: This layer receives and outputs the precise energy estimate and precise rise time estimate, and evaluates the improvement in energy resolution and time resolution of the output results based on the test dataset. Control and monitoring layer: configured to perform at least one of the following operations: control system training process, including initializing model parameters, executing backpropagation algorithm and determining loss function convergence; monitor and record changes in loss value during training and evaluate the performance indicators of the layer output; schedule the preprocessing layer, feature extraction layer and S-VAE core processing layer to work together to process the pulse data to be tested.

Images