Siamese Neural Network for TDOA Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for passive localization of radiofrequency emission sources, such as radar systems, face challenges in compressing data for transmission over limited data links while maintaining precision, particularly for signals with short or long pulse intervals, and are resource-intensive, especially when processing all signal samples in real time.
Innovation Solution
A method using a Siamese neural network architecture with two identical neural networks on each platform to encode and transmit signal parameters, trained on a dataset of signal models with introduced time shifts to calculate differential time of arrival (TDOA), allowing for efficient data compression and precise localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all signal samples are processed in real time to maintain localization precision, then measurement precision is improved, but device complexity and computational resource consumption increase significantly
Solution Approach 1:
The patent extracts only the essential information needed for TDOA calculation from the full signal samples. Instead of processing all samples, the system identifies and extracts key temporal markers (such as pulse edges or modulation features) that contain the necessary timing information, thereby reducing computational complexity while maintaining localization precision
Solution Approach 2:
The patent segments the signal processing task by dividing the continuous signal into discrete identifiable events or markers. Rather than processing the entire signal stream, the system segments and processes only the critical timing events (pulse arrivals, modulation transitions), which significantly reduces the computational burden while preserving the essential TDOA information
2Productivity
If data compression is applied to reduce transmission bandwidth, then data link throughput constraints are satisfied, but loss of information may occur
Solution Approach 1:
The patent extracts only the essential TDOA-related information from the signal data for transmission. By identifying and transmitting only the critical temporal marker parameters rather than all raw signal samples, the system achieves effective data compression while preserving the information necessary for accurate TDOA calculation and localization
Solution Approach 2:
The patent performs preliminary processing and compression of signal data at the receiving platforms before transmission to the central station. By pre-extracting and preparing only the essential TDOA parameters in advance, the system reduces the amount of data that needs to be transmitted while ensuring that no critical information is lost
3Adaptability or versatility
If Type 1 compression methods are used to process all signal samples, then generic applicability is improved, but computational resource consumption increases
Solution Approach 1:
The patent implements a dynamic signal processing approach that adapts to different signal types and characteristics. Rather than applying a fixed complex processing pipeline to all signals, the system dynamically adjusts its processing strategy based on the specific signal properties (pulse width, modulation type, frequency), thereby maintaining versatility while improving computational efficiency for each signal type
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The method uses an overall neural network (700, 73, 75) of Siamese type composed of two neural networks, called α networks, having the same weights and a third neural network, called the β network, receiving the data output from said α networks as input: - a first α network arranged on a first platform, encoding the signals from said transmitter via a set of parameters and transmitting said parameters to said β network; - the second α network being arranged on the second platform, encoding the signals from said transmitter via a set of parameters and transmitting said parameters to said β network; said β network being trained to calculate the TDOA on the basis of said parameters received from said α networks, said overall network being trained on the basis of a set of models of signals likely to be encountered.