Research Highlights

Wireless Communication in Metasurface-Programmable Environments

Traditionally, the propagation environment is regarded as an uncontrollable element in the wireless communication architecture. Adding a programmable metasurface as "reconfigurable intelligent surface" (RIS) to the environment constitutes a paradigm shift, allowing us to control how waves propagate between transmitter and receiver.

In Nature Electronics 2, 36 (2019), we have shown that by tweaking the disorder of an indoor propagation environment with a programmable metasurface, optimal channel diversity can be achieved for multi-channel wireless links. By creating an environment with perfectly orthogonal channels, the number of available independent channels reaches its optimal value.

In Nature Communications 11, 3926 (2020), we introduced the concept of massive backscatter wireless communication in which it is the programmable metasurface itself that encodes digital information into stray ambient waves. In contrast to traditional backscatter techniques, our scheme offers a drastically increased number of degrees of freedom to control the waves, paving the path to secure and high-speed wireless communication without active signal generation.

In Advanced Functional Materials 30, 2005310 (2020), we demonstrated a novel route to achieve perfect absorption of waves in a metasurface-programmble scattering enclosure. Perfect absorption is a very special condition that is extremely sensitive. We leveraged our technique's unique features to propose a physically secure scheme for receiver-powered wireless communication in which only the intended receiver can obtain the encoded information by determining if a perfect-absorption condition occurs on their port or not.

Learned Sensing with Programmable Metasurfaces

In state-of-the-art sensing architectures (e.g., to recognize a hand gesture), the measurement process and data analysis are optimized separately. Therefore, the physical layer lacks "intelligence" (task-awareness) and indiscriminately acquires all information. By interpreting reconfigurable meta-atoms in the acquisition hardware as physical trainable weights, we introduced a "learned sensing" paradigm to jointly learn optimal settings for both physical and digital weights in an integrated neural network. With learned measurement settings, only task-relevant information is acquired, significantly reducing the number of necessary measurements and thereby latency, computational burden, etc.

We introduced the idea in Advanced Science 7, 1901913 (2020) and demonstrated it experimentally for gesture recognition in Patterns 1, 100020 (2020).

Information Processing with Waves in Complex Media

As traditional electronic all-purpose processors face the end of Moore's law, the need for computing continues to increase exponentially with the advent of artificial intelligence, resparking great interest in analog specific-purpose computing. One enticing route is to swap the electronic circuits for materials with carefully tailored scattering properties such that the computational operation is performed on an incident wave front as it interacts with the material. Unfortunately, current approaches rely on intricate material designs which require extremely accurate fabrication techniques and often lack reconfigurability. In Physical Review X 8, 041037 (2018), we proposed that in fact a random scattering material could be used to perform a desired operation with waves – provided that the incident wave front can be shaped appropriately. Since our proposed concept is applicable to any type of wave and random medium, we were able to provide an experimental demonstration mimicking nothing but a standard Wi-Fi system operating in a metasurface-programmable indoor environment. In Optica 6, 465 (2019) we transposed the idea to the optical domain where we used a digital micromirror device to shape light before it propagates through a multimode fiber.

Precise Localization in Complex Environments by Wave Fingerprinting

Conventional approaches to localize objects are based on ray-tracing, the simplest example being triangulation. However, many crucial applications in context-awareness (e.g., remote health care) take place in complex propagation environments that resemble chaotic cavities. Rather than fighting the resulting multipath effects, we made them a virtue: each object position can be associated with a unique "wave fingerprint". To ink the wave fingerprint, rather than using a cumbersome array of antennas or a broadband measurement, we leveraged the programmability of a RIS-equipped environment in Physical Review Letters 121, 063901 (2018). In Physical Review Research 2, 043224 (2020) we took an information-theoretic perspective on wave fingerprinting to reveal that it can be successfully employed even in dynamically evolving complex propagation environments.