Machine Learning Photon Arrival Time Determination

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional time-correlated single-photon counting (TCSPC) systems face challenges in accurately determining photon arrival times at high photon count rates due to pulse pile-up, which limits measurement speed and requires low count rates, making it unsuitable for fast applications like fluorescence lifetime imaging and live cell imaging.

Innovation Solution

A method using a machine learning approach to process electric output signals from photon-number resolving detectors, sampling at high rates, and filtering to prolong signal fluctuations, allowing for real-time estimation of photon arrival times even in cases of overlapping signals, without requiring knowledge of detector response or filter characteristics.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional TCSPC electronics are used to detect photon arrival times, then the system is simple and reliable, but it cannot distinguish overlapping signal fluctuations from multiple photons, leading to pulse pile-up and inaccurate measurements at high count rates

Engineering Contradiction:
Improvephoton arrival time determination accuracyVSAvoidsignal processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces conventional electronic signal processing methods with a machine learning-based approach. The trained neural network model processes the analog electric output signal to distinguish overlapping photon detection events, substituting traditional electronics with an intelligent system that can resolve pulse pile-up effects and accurately determine photon arrival times even at high count rates

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the analog electric output signal into a format suitable for machine learning processing by applying filtering and sampling operations. The signal is converted from continuous analog form to discrete digital samples, changing its representation parameters to enable application of neural network algorithms for photon arrival time determination

Inventive Principle:
Principle #35Parameter changes

2Productivity

If low count rates are used to avoid pulse pile-up effects, then measurement accuracy is maintained, but measurement time increases significantly, making it unsuitable for fast applications

Engineering Contradiction:
Improvemeasurement speedVSAvoidphoton arrival time determination accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent replaces the conventional approach of reducing count rates to maintain accuracy with a machine learning system that can process high count rates accurately. The trained neural network model enables the system to operate at high productivity by resolving overlapping signals that would otherwise cause pulse pile-up, thus achieving both fast measurement speeds and accurate photon arrival time determination

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent converts the harmful pulse pile-up effect into a useful signal characteristic. Instead of treating overlapping signal fluctuations as noise to be avoided, the machine learning model learns to recognize and exploit these overlapping patterns to determine multiple photon arrival times simultaneously, turning the previously problematic high count rate condition into an advantageous operating state

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

3Measurement precision

If photon-number-resolving detectors are used to capture true number of detected photons, then measurement accuracy improves, but conventional TCSPC electronics cannot process the amplitude information quantitatively

Engineering Contradiction:
Improvephoton number resolution accuracyVSAvoidelectronics compatibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent replaces incompatible conventional TCSPC electronics with a machine learning-based processing system. The neural network model is specifically trained to handle the amplitude information from photon-number-resolving detectors, enabling quantitative interpretation of signal fluctuations that correspond to multiple photon detections while maintaining compatibility with the detector output

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Enables real-time processing of photon arrival times and number of detection events, reducing measurement time and overcoming limitations of conventional TCSPC systems, particularly in high-count-rate applications.

Implementation Method 1

providing the sequence of the discrete signal values to a machine learning method that is configured and particularly trained to determine a number of photon detection events and their associated photon arrival times from the sequence of the discrete signal values

Methodology Applied
Scientific EffectMachine learning pattern recognition:

Data Source

PatentEP3546926B1Method and device for determining photon arrival times by means of machine learning
Publication Date: 2022.05.11 PICOQUANT INNOVATIONS
  • EP3546926B1 patent drawingFigure 1
  • EP3546926B1 patent drawingFigure 2A~2D
  • EP3546926B1 patent drawingFigure 3

AI summary

The invention relates a method and a computer program for determining photon arrival times (t1, t2) from an electric output signal (11) of a photon detector (10), the method comprising the steps of: a) Acquiring (100) an analog electric output signal (11) from a photon detector (10), wherein the electric output signal (11) comprises signal fluctuations (12) originating from photon detection events (13); b) Sampling (180) the electric output signal (15), such that a plurality of discrete signal values (17) is generated from the electric output signal (11, 15); c) Determining a number (N) of photon detection events (13) and a photon arrival time (t1, t2) for each photon detection event (13) for a sequence (18a, 18b, 18c) of the discrete signal values (17) by providing the sequence (18a, 18b, 18c) of the discrete signal values (17) to a machine learning method (19) that is configured to determine the number (N) of photon detection events (13) and their associated photon arrival times (t1, t2) from the sequence (18a, 18b, 18c) of the discrete signal values (17). The invention further relates to a system configured to execute the method according the invention.