Qoherent Uses MATLAB to Accelerate Research on Next-Generation AI for Wireless
One of the key benefits of the MATLAB experience for wireless researchers is the breadth of hardware support—including USRP™, PlutoSDR, and Keysight®—and ease of use in the process from installation to visualization of IQ data.
Key Outcomes
- Reduced learning curve on LTE and 5G standards by several months. Prior to the MATLAB example release, the most challenging aspect of this project was researching each wireless standard to create a synthetic data set to compare with the one recorded from the testbed.
- Quick-started SDR and installation-to-spectrogram experiences in under 30 minutes. With wireless being its core, Qoherent invests significant effort in SDR to ramp up teams’ onboarding processes. MATLAB provides the quickest installation-to-spectrogram experience out of the various SDR frameworks.
- Saved significant development time using 5G Toolbox and LTE Toolbox to generate standard-compliant waveforms. This time would have otherwise been spent on manually generating waveforms, creating data sets, and optimizing and troubleshooting a training workflow from scratch.
Workflow of 5G and LTE signal classification.
Qoherent is a Toronto-based startup focused on driving the creation of intelligent radio technology, which uses AI and machine learning techniques to build more robust, aware, and adaptive RF communications and sensing systems.
To help develop this technology, Qoherent is building an open-source platform that will support research on next-generation wireless AI. Researchers who are required to innovate with a combination of open source and proprietary tools often face interoperability challenges when working with multiple platforms.
Qoherent engineers used the MATLAB® Spectrum Sensing with Deep Learning to Identify 5G and LTE Signals example to study the tooling interoperability challenges. With 5G Toolbox™, LTE Toolbox™, and Deep Learning Toolbox™, they were able to quickly compare and validate the MATLAB synthetic data set with a data set recorded from the commercial 5G radio access network equipment, as well as different models from frameworks implemented in-house. Additionally, the open-source AI network achieved the same classification results as the MATLAB based AI network.
