Neural Network Kernel Replaces DSP in Radio Receivers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing radio receivers face challenges in rapidly adapting to novel edge conditions, such as novel multi-tone jamming signals, due to the time-consuming and expensive process of redesigning artifact-suppressing Digital Signal Processing (DSP) algorithms, which are often inadequate for real-time adjustments during military operations.
Innovation Solution
Implementing a Neural Network (NN) kernel that replaces the DSP, trained using measured or theoretically generated examples, including physics-based models, and capable of re-optimization for new edge conditions without increasing computational size or latency, utilizing Dilated Causal Convolutions and dense layers for effective artifact suppression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional DSP algorithms are used for artifact suppression, then the radio receiver can effectively suppress known artifacts under expected operating conditions, but the system cannot rapidly adapt to novel edge conditions such as new jamming signals without time-consuming and expensive redesign
Solution Approach 1:
The patent replaces traditional Digital Signal Processing (DSP) algorithms with a Neural Network (NN) kernel. The NN kernel is trained on training examples that include various artifact conditions, enabling it to automatically adapt to novel edge conditions without requiring manual algorithm redesign. This substitution of mechanical/DSP-based artifact suppression with neural network-based suppression allows the system to learn and adapt to new jamming signals and artifacts dynamically, resolving the contradiction between adaptability and redesign time.
2Reliability
If DSP algorithms are redesigned to handle new edge conditions, then artifact suppression performance improves for those conditions, but the process is expensive and time-consuming
Solution Approach 1:
The Neural Network kernel is pre-trained on a comprehensive set of training examples that include various artifact types and edge conditions before deployment. This preliminary training action enables the NN kernel to handle novel edge conditions effectively without requiring costly and time-consuming redesign processes later. The pre-training encompasses diverse scenarios including different jamming signals, tone interference, and multipath conditions, allowing the system to maintain high artifact suppression performance across varying operational environments.
3Adaptability or versatility
If the NN kernel is made more complex to handle diverse edge conditions, then adaptability improves, but computational size and latency may increase
Solution Approach 1:
The Neural Network kernel is designed as a universal, fixed-size computational structure that can handle multiple types of artifacts and edge conditions through its training rather than through structural complexity. The NN kernel performs multiple functions including tone interference suppression, multipath artifact reduction, and adaptation to novel jamming signals, all within a consistent computational footprint. This multi-functionality is achieved through the neural network's ability to learn diverse patterns from training examples without requiring increases in computational size or complexity.
Data Source
AI summary
An artifact-suppressing neural network (NN) kernel comprising at least one neural network, implemented in replacement of a DSP, provides comparable or better performance under non-edge conditions, and superior performance under edge conditions, due to the ease of updating the NN kernel training without enlarging its computational footprint or latency to address a new edge condition. In embodiments, the NN kernel can be implemented in a field programmable gate array (FPGA) or application specific integrated circuit (ASIC), which can be configured as a direct DSP replacement. In various embodiments, the NN kernel training can be updated in near real time when a new edge condition is encountered in the field. The NN kernel can include DCC lower layers and dense upper layers. Initial NN kernel training can require fewer examples. Example embodiments include a noise suppression NN kernel and a modem NN kernel.