The accelerating demands on spectrum resources are pushing radio operations into performance regimes that require more efficient spectrum utilization and more reliable coexistence solutions. Emerging spectrum sharing techniques are demanding new processing approaches to enable blind source separation (BSS) for effective interference cancellation and information assurance. RF integrated circuits have dominated the frontend analog processing in microwave electronic systems for decades; however, these technology platforms are inherently frequency dependent, resulting in severely limited reconfigurability in terms of both functional programmability and frequency agility.
We propose photonic BSS to provide an orthogonal signal processing solution with a completely different set of physical properties. By effectively upconverting the RF signal to a 193 THz intermediate frequency, photonic processors are nearly frequency independent—even GHz signal is considered narrowband because optical waveguides have a flat frequency response over a 5 THz window. Our photonic processor employs silicon microring resonator (MRR) bank to implement linear analog computation, leveraging support from integrated photonic technology and enormous information density made possible with wavelength-division multiplexing (WDM). We also utilize statistical learning algorithms solely based on the moment observations from photonic frontend to achieve source separation in real-life scenarios where complete waveform information of incoming mixtures is not readily available.