Estimation of trapezoidal-shaped overlapping nuclear pulse parameters based on a deep learning CNN-LSTM model

Area of Science:

• Nuclear physics instrumentation
• Signal processing
• Machine learning applications

Background:

• Nuclear pulse signals present challenges for parameter estimation, especially when overlapping.
• Traditional Long Short-Term Memory (LSTM) networks struggle with training efficiency due to large datasets of pulse sequences.
• Convolutional Neural Networks (CNNs) can reduce data complexity by extracting key features.

Purpose of the Study:

• To develop an efficient and accurate method for estimating parameters of overlapping, digitally shaped nuclear pulse signals.
• To leverage the feature extraction capabilities of CNNs with the sequence learning of LSTMs for improved performance.
• To overcome limitations of traditional methods, such as local convergence and reduced training efficiency.

Main Methods:

• A hybrid CNN-LSTM deep neural network was designed to process time-series data of trapezoidally shaped, overlapping nuclear pulses.
• CNNs were employed for initial feature extraction from pulse sequences, reducing the input complexity for the LSTM.
• The model was trained using a dataset of simulated pulse signals and their corresponding shaping parameters, optimized via gradient-based algorithms.

Main Results:

• The CNN-LSTM model demonstrated high accuracy in estimating parameters (e.g., amplitude, time constant) of overlapping nuclear pulses.
• Significant improvements in training efficiency were achieved compared to using LSTM alone.
• The method effectively overcame local convergence issues inherent in traditional estimation techniques.

Conclusions:

• The CNN-LSTM approach provides an effective solution for the optimal, global estimation of nuclear pulse parameters, particularly for complex overlapping signals.
• This hybrid deep learning model offers a substantial advancement in nuclear signal processing and parameter estimation accuracy and speed.
• The method's ability to handle wide flat-top pulses enhances its applicability in real-world nuclear measurements.