Radar Signal Auto-Encoding for Loss-Aware Data Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current radar systems for automated vehicles face challenges in transferring detailed radar data due to sparse signal nature and existing data compression methods that introduce noise artifacts and are not suitable for advanced applications like pedestrian or landmark classification.
Innovation Solution
A radar system with an embedded auto-encoding device that applies non-linear functions to compress and encode raw radar signals, allowing for efficient data transfer without arbitrary noise thresholds, and a decoding device to reconstruct signals with minimal loss, using neural networks to optimize parameters for accurate data transfer.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If threshold-based compression is applied to radar raw signals, then data transfer rate is improved, but information loss and noise artifacts occur
Solution Approach 1:
The auto-encoding model is trained in advance on large datasets to learn optimal compression representations before actual radar signal processing. This preliminary training enables the model to understand the relationship between compressed and original signals, allowing it to perform lossless or near-lossless compression when deployed in the radar system.
Solution Approach 2:
The patent introduces an intermediary auto-encoding model that acts as a bridge between the radar sensor and the processing unit. This model transforms raw radar signals into a compressed intermediate representation that preserves all essential information while reducing data volume, and then reconstructs the original signals at the decoding end, avoiding direct threshold-based compression.
2Quantity of substance
If arbitrary noise thresholds are applied for compression, then data volume is reduced, but harmful artifacts are introduced
Solution Approach 1:
The patent changes the compression approach from fixed threshold-based parameter reduction to adaptive parameter transformation using learned models. The auto-encoding model dynamically adjusts compression parameters based on the input signal characteristics, preserving signal integrity while achieving compression through intelligent parameter transformation rather than arbitrary thresholding.
3Measurement precision
If detailed radar data is transferred for sophisticated applications, then classification accuracy is improved, but data transfer complexity increases
Solution Approach 1:
The patent segments the data processing function into two parts: compression at the sensor end using the auto-encoding model, and reconstruction at the processing end. This segmentation allows detailed radar data to be transferred in a compressed form that maintains classification accuracy while reducing the complexity of the data transfer infrastructure.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A radar system suitable for an automated vehicle comprises at least one receiving element configured to detect reflected radar signals reflected by an object in a field-of-view of the system. The receiving element is further configured to generate detected signals indicative of the reflected radar signals detected by the receiving element. Furthermore, an auto-encoding device performing an auto-encoding operation on the detected signals is embedded in the receiving element such that the receiving element outputs the detected signals in an encoded form.