Our approach to signal processing design uses machine learning to learn optimized models directly from data, rather than manually designing specialized algorithms under simplified toy models. Our learned algorithms benefit from complexity and data, improving with more experience and observations and adapting to real world effects. We optimize for the performance of entire systems, inclusive of hardware and channel impairments, rather than stitching together separately optimized components.
Our process for creating learned signal processing systems enables us to use a common core software architecture for a range of different applications while still customizing and optimizing model performance for specific application and deployment requirements.
DeepSig is pioneering the use of deep learning to realize state of the art signal processing and radio systems by developing fundamentally new methodologies and software systems for the design and optimization of wireless communications. By creating new tools, algorithms, and approaches for signal processing systems, DeepSig is able to achieve unparalleled results in system performance.
The results of DeepSig's research are already having a significant impact on the fields of communications and signal processing. DeepSig's engineers have published many of the seminal scientific papers in this area (see our Publications page), and are the technical leaders in building real-world practical systems with this technology. DeepSig is applying its products to existing wireless systems and standards (e.g., WiFi, 4G LTE), and our goal is to see our work used in the next generation of communications standards.
We are an early-stage startup located in Arlington, Virginia.