AI-Aided Annealed Langevin Dynamics for Rapid Optimization of Programmable Channels

Emerging technologies such as Reconfigurable Intelligent Surfaces (RIS) make it possible to optimize some parameters of wireless channels. Conventional approaches require relating the channel and its programmable parameters via a simple model that supports rapid optimization, e.g., re-tuning the parameters each time the users move. However, in practice such models are often crude approximations of the channel, and a more faithful description can be obtained via complex simulators, or only by measurements. In this work, we introduce a novel approach for rapid optimization of programmable channels based on AI-aided Annealed Langevin Dynamics (ALD), which bypasses the need for explicit channel modeling. By framing the ALD algorithm using the MAP estimate, we design a deep unfolded ALD algorithm that leverages a Deep Neural Network (DNN) to estimate score gradients for optimizing channel parameters. We introduce a training method that overcomes the need for channel modeling using zero-order gradients, combined with active learning to enhance generalization, enabling optimization in complex and dynamically changing environments. We evaluate the proposed method in RIS-aided scenarios subject to rich-scattering effects. Our results demonstrate that our AI-aided ALD method enables rapid and reliable channel parameter tuning with limited latency.

Frozen differential scattering in reconfigurable complex media

The sensitivity of transmission to the input wavefront is a hallmark feature of complex media and the basis for wavefront shaping techniques. Yet, intriguing special cases exist in which the output wavefront is "frozen" (agnostic to the input wavefront). This happens when special structure in the complex medium collapses the rank of its transmission matrix to unity. Here, we unveil that an analogous phenomenon exists more universally for differential scattering (including reflection) in reconfigurable complex media. Specifically, for a localized perturbation, the differential scattering matrix of any complex medium has rank one. One consequence is that the differential output signal is perfectly coherent irrespective of the input wavefront's coherence. Moreover, the thermal noise emitted into the frozen differential output mode has a particular structure that can be exploited for thermal noise management. We experimentally evidence frozen differential scattering in a rich-scattering wireless link parametrized by a programmable meta-atom. Then, we demonstrate "customized freezing" by optimizing the configuration of additional programmable meta-atoms that parametrize the wireless link, as envisioned for 6G networks. We impose particular shapes of the frozen differential output mode, and maximize its signal-to-thermal-noise ratio. Potential applications include filtering and stabilization of differential wavefronts, as well as imaging, sensing, and communication in complex media.

Ambiguity-Aware Segmented Estimation of Mutual Coupling in Large RIS: Algorithm and Experimental Validation

Optimizing a real-life RIS-parametrized wireless channel with a physics-consistent multiport-network model necessitates prior remote estimation of the mutual coupling (MC) between RIS elements. The number of MC parameters grows quadratically with the number of RIS elements, posing scalability challenges. Because of inevitable ambiguities, independently estimated segments of the MC matrix cannot be easily stitched together. Here, by carefully handling the ambiguities, we achieve a separation of the full estimation problem into three sequentially treated sets of smaller problems. We partition the RIS elements into groups. First, we estimate the MC for one group as well as the characteristics of the available loads. Second, we separately estimate the MC for each of the remaining groups, in each case with partial overlap with an already characterized group. Third, we separately estimate the MC between each distinct pair of groups. Full parallelization is feasible within the second and third sets of problems, and the third set of problems can furthermore benefit from efficient initialization. We experimentally validate our algorithm for a 4x4 MIMO channel parametrized by a 100-element RIS inside a rich-scattering environment. Our experimentally calibrated 5867-parameter multiport-network model achieves an accuracy of 40.5 dB, whereas benchmark models with limited or no MC awareness only reach 17.0 dB and 13.8 dB, respectively. Based on the experimentally calibrated models, we optimize the RIS for five wireless performance indicators. Experimental measurements with the optimized RIS configurations demonstrate only moderate benefits of MC awareness in RIS optimization in terms of the achieved performance. However, we observe that limited or no MC awareness markedly erodes the reliability of model-based predictions of the expected performance.

Experimental Multiport-Network Parameter Estimation and Optimization for Multi-Bit RIS

Physics-consistent theoretical studies on RIS-parametrized wireless channels use models from multiport-network theory (MNT) to capture mutual-coupling (MC) effects. However, in practice, RIS design and radio environment are partially or completely unknown. We fill a research gap on how to estimate the MNT model parameters in such experimentally relevant scenarios. Our technique efficiently combines closed-form and gradient-descent steps, and it can be applied to multi-bit-programmable RIS elements. We discuss inevitable (but operationally irrelevant) parameter ambiguities. We experimentally validate our technique in an unknown rich-scattering environment parametrized by eight 8-bit-programmable RIS elements of unknown design. We experimentally evaluate the performance of RIS configurations optimized with the estimated MNT model and an MC-unaware cascaded model. While the models differ in accuracy by up to 17 dB, the end-to-end performance differences are small.

End-to-End Dynamic Metasurface Antenna Wireless System: Prototype, Opportunities, and Challenges

Dynamic metasurface antennas (DMAs) are a promising hybrid analog/digital beamforming technology to realize next-generation wireless systems with low cost, footprint, and power consumption. The research on DMA-empowered wireless systems is still at an early stage, mostly limited to theoretical studies under simplifying assumptions on the one hand and a few antenna-level experiments on the other hand. Substantial knowledge gaps arise from the lack of complete end-to-end DMA-empowered wireless system prototypes. In addition, recently unveiled benefits of strong inter-element mutual coupling (MC) in DMAs remain untapped. Here, we demonstrate a K-band prototype of an end-to-end wireless system based on a DMA with strong inter-element MC. To showcase the flexible control over the DMA's radiation pattern, we present an experimental case study of simultaneously steering a beam to a desired transmitter and a null to an undesired jammer, achieving up to 43~dB discrimination. Using software-defined radios, we transmit and receive QPSK OFDM waveforms to evaluate the bit error rate. We also discuss algorithmic and technological challenges associated with envisioned future evolutions of our end-to-end testbed and real-life DMA-based wireless systems.

Beyond-Diagonal RIS Prototype and Performance Evaluation

We present the first experimental prototype of a reflective beyond-diagonal reconfigurable intelligent surface (BD-RIS), i.e., a RIS with reconfigurable inter-element connections. Our BD-RIS consists of an antenna array whose ports are terminated by a tunable load network. The latter can terminate each antenna port with three distinct individual loads or connect it to an adjacent antenna port. Extensive performance evaluations in a rich-scattering environment validate that inter-element connections are beneficial. Moreover, we observe that our tunable load network's mentioned hardware constraints significantly influence, first, the achievable performance, second, the benefits of having inter-element connections, and, third, the importance of mutual-coupling awareness during optimization.

Programmable metasurfaces for future photonic artificial intelligence

Photonic neural networks (PNNs), which share the inherent benefits of photonic systems, such as high parallelism and low power consumption, could challenge traditional digital neural networks in terms of energy efficiency, latency, and throughput. However, producing scalable photonic artificial intelligence (AI) solutions remains challenging. To make photonic AI models viable, the scalability problem needs to be solved. Large optical AI models implemented on PNNs are only commercially feasible if the advantages of optical computation outweigh the cost of their input-output overhead. In this Perspective, we discuss how field-programmable metasurface technology may become a key hardware ingredient in achieving scalable photonic AI accelerators and how it can compete with current digital electronic technologies. Programmability or reconfigurability is a pivotal component for PNN hardware, enabling in situ training and accommodating non-stationary use cases that require fine-tuning or transfer learning. Co-integration with electronics, 3D stacking, and large-scale manufacturing of metasurfaces would significantly improve PNN scalability and functionalities. Programmable metasurfaces could address some of the current challenges that PNNs face and enable next-generation photonic AI technology.

Beyond-Diagonal Dynamic Metasurface Antenna

Dynamic metasurface antennas (DMAs) are an emerging technology for next-generation wireless base stations, distinguished by hybrid analog/digital beamforming capabilities with low hardware complexity. However, the coupling between meta-atoms is fixed in existing DMAs, which fundamentally constrains the achievable performance. Here, we introduce reconfigurable coupling mechanisms between meta-atoms, yielding finer control over the DMA's analog signal processing capabilities. This novel hardware is coined "beyond-diagonal DMA" (BD-DMA), in line with established BD-RIS terminology. We derive a physics-consistent system model revealing (correlated) "beyond-diagonal" programmability in a reduced basis. We also present an equivalent formulation in a non-reduced basis with (uncorrelated) "diagonal" programmability. Based on the diagonal representation, we propose a general and efficient mutual-coupling-aware optimization algorithm. Physics-consistent simulations validate the performance enhancement enabled by reconfigurable coupling mechanisms in BD-DMAs. The BD-DMA benefits grow with the mutual coupling strength.

Virtual VNA 3.1: Non-Coherent-Detection-Based Non-Reciprocal Scattering Matrix Estimation Leveraging a Tunable Load Network

We refine the recently introduced "Virtual VNA 3.0" technique to remove the need for coherent detection. The resulting "Virtual VNA 3.1" technique can unambiguously estimate the full scattering matrix of a non-reciprocal, linear, passive, time-invariant device under test (DUT) with $N$ monomodal ports using an $N_\mathrm{A}$-channel coherent wavefront generator and an $N_\mathrm{A}$-channel non-coherent detector, where $N_\mathrm{A}<N$. Waves are injected and received only via a fixed set of $N_\mathrm{A}$ "accessible" DUT ports while the remaining $N_\mathrm{S}$ "not-directly-accessible" DUT ports are terminated by a specific tunable load network. To resolve all ambiguities, an additional modified setup is required in which waves are injected and received via a known $2N_\mathrm{A}$-port system connected to the DUT's accessible ports. We experimentally validate our method for $N_\mathrm{A}=N_\mathrm{S}=4$ considering a non-reciprocal eight-port circuit as DUT. By eliminating the need for coherent detection, our work reduces the hardware complexity which may facilitate applications to large-scale or higher-frequency systems. Additionally, our work provides fundamental insights into the minimal requirements to fully and unambiguously characterize a non-reciprocal DUT.

Scalable Multiport Antenna Array Characterization with PCB-Realized Tunable Load Network Providing Additional "Virtual" VNA Ports

We prototype a PCB-realized tunable load network whose ports serve as additional "virtual" VNA ports in a "Virtual VNA" measurement setup. The latter enables the estimation of a many-port antenna array's scattering matrix with a few-port VNA, without any reconnections. We experimentally validate the approach for various eight-element antenna arrays in an anechoic chamber in the 700-900 MHz regime. We also improve the noise robustness of a step of the "Virtual VNA" post-processing algorithms by leveraging spectral correlations. Altogether, our PCB-realized VNA Extension Kit offers a scalable solution to characterize very large antenna arrays because of its low cost, small footprint, fully automated operation, and modular nature.