Neural Network ToF Correction for Multipath Interference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multipath interference (MPI) in time-of-flight (ToF) modules affects the accuracy of distance measurements by causing phase and amplitude distortions in the received light signals, leading to measurement errors.
Innovation Solution
A time of flight correction method using a neural network that incorporates structured light measurements to provide ground truths for training, where raw phases and amplitudes from the ToF module are inputted into a convolutional neural network (CNN) to generate phase corrections, which are then converted to corrected ranges, thereby mitigating the effects of MPI.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a ToF module receives all reflected light signals to calculate ToF from various directions, then the coverage and field of view are improved, but multipath interference occurs which deteriorates measurement accuracy
Solution Approach 1:
The patent segments the received light signals by direction, using an array of photodetectors to spatially separate signals arriving from different angles. This allows the system to identify and filter out multipath signals (which arrive from different directions than the direct path) while maintaining comprehensive field of view coverage.
Solution Approach 2:
The patent introduces a neural network as an intermediary between the raw ToF measurements and the final distance calculation. The neural network is trained to recognize patterns of multipath interference and correct the distorted measurements, effectively mediating between the contaminated input signals and the accurate output distances.
2Measurement precision
If the ToF module calculates ToF by analyzing round trip time of light signals, then distance measurement capability is achieved, but multipath waves with two or more reflections are received which affects accuracy
Solution Approach 1:
The patent converts the harmful multipath interference into a useful signal characteristic for training the neural network. By using structured light projections with known geometric patterns, the system creates ground truth data where multipath effects are predictable and can be used to train the neural network to recognize and correct similar interference patterns in real measurements.
Solution Approach 2:
The patent changes the parameter space by transitioning from direct ToF measurement to a neural network-based correction approach. The neural network learns to adjust the measured ToF values by predicting the magnitude and direction of multipath errors based on input features including the measured ToF, signal intensity, and spatial distribution across the photodetector array.
3Measurement precision
If structured light measurements are used to provide ground truths for training, then measurement accuracy is improved, but device complexity increases due to additional components
Solution Approach 1:
The patent makes the structured light projection system multi-functional: it serves both as the measurement signal source for normal ToF operation and as a calibration tool for generating training data. The same hardware components used for regular operation are leveraged to create ground truth datasets, eliminating the need for separate calibration equipment and reducing overall system complexity.
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
The method improves the accuracy of ToF module measurements by correcting for multipath interference, ensuring precise distance calculations.
Implementation Method 1
time of flight correction method by neural network... calculating the round trip time of an artificial light signal (typically emitted by a laser or a light emitting diode(LED)) as it travels to a target and back
Implementation Method 2
The emitted light pulse reflects off the object, and the camera measures the time it takes for the signal to complete this round trip
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
A time of flight correction method includes measuring a plurality of phases (402) of electromagnetic waves and a plurality of amplitudes (404) of the electromagnetic waves using a time of flight module, inputting the plurality of phases (402) and the plurality of amplitudes (404) into a pre-trained neural network model (406) to output a plurality of phase corrections, adding the plurality of phases (402) and the plurality of phase corrections to output a plurality of corrected phases (408), and converting the plurality of corrected phases (408) to output a plurality of corrected ranges (410) between the time of flight module and surrounding targets.