Single-Snapshot Amplitude Direction Finding Using Cascaded Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional angle-of-arrival estimation systems require multiple snapshots and complex hardware for phase information retrieval, which is suboptimal for fast-moving targets and increases computational time and cost, especially in ultrawideband and low-latency applications.
Innovation Solution
A machine-learning based, single-snapshot amplitude-only direction finding system using two cascaded neural networks to identify the subregion and direction of a signal, reducing system size and complexity by reusing the angle of arrival estimator across observation sectors and leveraging geometrical symmetry for circular arrays.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional phase-based direction finding methods are used, then measurement precision is improved, but device complexity and computational resources increase
Solution Approach 1:
The patent extracts and utilizes only the amplitude component of the signal, discarding the phase component. This is achieved by taking the absolute value of the received signal at each antenna element, thereby simplifying the system while maintaining direction finding capability through amplitude-only processing
Solution Approach 2:
The patent replaces complex phase-based signal processing with machine learning-based amplitude processing. Neural networks are trained to infer direction of arrival from amplitude patterns alone, substituting traditional phase measurement mechanisms with an intelligent system that processes simplified amplitude data
2Measurement precision
If multiple snapshots are used for direction finding, then measurement precision is improved, but productivity and response time deteriorate
Solution Approach 1:
The patent performs preliminary action by pre-training machine learning models offline with extensive simulation data that encompasses multiple snapshots and various signal conditions. Once trained, the model can make rapid predictions on single-snapshot data without requiring multiple actual measurements, thus achieving both accuracy and speed
Solution Approach 2:
The patent uses a single snapshot (partial action) instead of multiple snapshots, relying on the pre-trained model's ability to extract sufficient directional information from limited data. The model has learned to make accurate predictions even with incomplete information, trading off some robustness for significant speed improvement
3Measurement precision
If large sensor arrays are deployed, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent designs a universal direction finding system using a uniform circular array that can handle multiple signal sources and various incident angles through a single machine learning model. The model is trained to recognize amplitude patterns from any direction, making the system universally applicable without requiring complex beamforming or multiple specialized processing chains
Solution Approach 2:
The patent changes the processing parameter from phase to amplitude, and from multiple snapshots to single snapshot. This parameter transformation allows the system to achieve comparable performance with simpler hardware, as the machine learning model compensates for the reduced data dimensionality through learned patterns
Data Source
AI summary
A computer system for angle of arrival estimation receives one or more snapshots from a circular array of antennas. The computer system processes the one or more snapshots for amplitude and/or phase-based direction finding using two cascaded algorithms. The first algorithm of the two cascaded algorithms is configured to identify a target subregion from which a signal arrives. The second algorithm of the two cascaded algorithms is configured to identify a direction of the signal within the target subregion.