Invention Reference Number
This invention utilizes new techniques in machine learning (ML) to accelerate the training of ML-based communication receivers.
Description
Self-supervised learning (SSL) is a type of machine learning that does not require labeled data to identify patterns. It represents a newer approach, effective for processing large amount of raw unlabeled information. This invention applies SSL to solve a problem in communication systems known as the channel autoencoder. By training the system using SSL with a specific technique called a contrastive loss, it was discovered that the overall training time for a neural receiver, a key component in proposed next-generation communication systems, could be significantly reduced. This invention demonstrates potential for improving the performance of neural receivers across a wide range of communication environments.
Benefits
- Improved interference handling: The technology has a potential to provide improved mitigation for co-channel interference in crowded spectral environments such as dense urban areas.
- Facilitation of autonomous factories: In spectrally and physically crowded factory environments, the ability to “cut the cord” and reliably communicate wirelessly is necessary for full autonomy. This technology has the potential to reduce wireless dead spots and support higher data rates in a given wireless channel.
Applications and Industries:
- Smart manufacturing: This invention will be of value to wireless equipment manufacturers working in this industry who are looking to provide better indoor coverage.
- 6G cellular (commercial): This invention will improve the performance and reduce the training time of ML-based gNodeB receivers.
- Internet of Things (IoT): Pre-trained neural receivers can be designed into IoT devices to provide improved coverage and link reliability.
Contact
To learn more about this technology, email partnerships@ornl.gov or call 865-574-1051.
