Li-Yu Yu, supervised by Professor Sixian You in the Computational Biophotonics Group, focuses his research on enhancing the speed and fidelity of light manipulation in multimode fibers, an important factor in the development of next-generation high-speed, low-distortion communication systems. Multimode fibers are valued in communications for their ability to transmit multiple light modes simultaneously, effectively boosting bandwidth and data transfer capacity. Yet, achieving high-speed, high-fidelity light manipulation for efficient data transmission remains a challenge due to inherent tradeoffs between speed and accuracy in current light modulation techniques.
Li-Yu’s new approach addresses this problem by leveraging two physical limitations. Firstly, random mode coupling in multimode fibers, which converts a single input point to a complex speckle, affecting image quality and communication efficiency. Secondly, the constraints of high-speed wavefront shaping devices, which can only produce sparse patterns with high fidelity due to limited degrees of freedom. Using a strategy that combines the inherent point-to-speckle (or sparse-to-dense) transformation of multimode fibers with the sparsity preference of wavefront shaping devices, Li-Yu’s work surmounts the speed-fidelity tradeoff. The simple but effective method not only overcomes the two physical constraints but also mitigates issues related to the lack of global minimum guarantee and overfitting, thus enabling high-speed, high-fidelity light manipulation.
This novel optimization paradigm can be easily adopted by the wavefront shaping and multimode fiber community. Its simplicity will facilitate more widespread use of multimode fiber and wavefront shaping tools, benefiting both the research and industry communities in optical communications and imaging.
Ronald Davis III
Ronald Davis III’s research is motivated by the increasing congestion and complexity of signals present in the wireless spectrum. In particular, urban environments contain an unprecedented amount of information due to a plethora of devices ranging from cellphones and laptops to radios and cell towers. Traditional signal processing approaches requires hand-crafting signal processing systems, which is a solution that is becoming less viable in an ever-more crowded spectral environment.
With a background in electrical engineering and physics, Ronald sought to synthesize the two fields to find a creative solution to achieve high-throughput and low-latency signal processing. Working in the Quantum Photonics group led by Dirk Englund, Ronald designed an RF-photonic computing architecture. This architecture, called MAFT-ONN (Multiplicative Analog Frequency Transform Optical Neural Network) computes various signal processing functions and machine learning inference on frequency-encoded signals at the speed of light. After design and simulation, Ronald constructed an experimental demonstration of the MAFT-ONN architecture to implement a deep neural network. Ronald is currently pursuing further signal processing applications of the MAFT-ONN.