A Deep Learning Machine Modem for PHY Layer Communications

Traditional PHY layer communications consist of numerous signal processing blocks that help convey information (bits) across a, potentially unknown, channel with unknown noise. These blocks include such processing as: channel estimation, carrier-frequency offset correction, phase-locked loop, (de)modulation, error correction, and so on. The purpose of this invention is to replace the entirety of the traditional PHY layer communications processing chain with a single deep learning graph. This graph will learn to communicate in whatever channel it is placed in and attempt to create optimal communications similar to traditional approaches; however, unlike traditional digital communications, the invention can adapt all aspects of the processing chain in order to achieve some quality of service metric.

---

---