Randomized Space-Time Stacked Intelligent Metasurfaces for Massive Multiuser Downlink Connectivity

TL;DR

A novel beamforming strategy for massive downlink connectivity using a randomized space-time (ST) coded SIM that leverages randomized steering vectors and limited user-side feedback based on signal quality measurements, thereby enabling scalable downlink connectivity in dense networks.

Abstract

Stacked intelligent metasurfaces (SIMs) represent a key enabler for next-generation wireless networks, offering beamforming gains while significantly reducing radio-frequency chain requirements. In conventional space-only SIM architectures, the rate of reconfigurability of the SIM is equal to the inverse of the channel coherence time. This paper investigates a novel beamforming strategy for massive downlink connectivity using a randomized space-time (ST) coded SIM. In addition to conventional space-only metasurface layers, the proposed design integrates a ST metasurface layer at the input stage of the SIM that introduces random time variations over each channel coherence time interval. These artificial time variations enable opportunistic user scheduling and exploitation of multiuser diversity under slow channel dynamics. To mitigate the prohibitive overhead associated with full channel state information at the transmitter (CSIT), we propose a partial-CSIT-based beamforming scheme that leverages randomized steering vectors and limited user-side feedback based on signal quality measurements. Numerical results demonstrate that the proposed ST-SIM architecture achieves satisfactory sum-rate performance while significantly reducing CSIT acquisition and feedback overhead, thereby enabling scalable downlink connectivity in dense networks.

Figures (10)

• Figure 1: ST-SIM-aided multiuser downlink system serving $N$ out of $U$ users. The ST-SIM consists of $L$ metasurface layers. The first layer acts as a ST-DAL, comprising both absorbing (in red color) and transmitting (in green color) meta-atoms. The last $L-1$ layers are S-only ones and consist of transmitting meta-atoms. S-only layers do not vary over each channel coherence interval of duration $T$, while the ST initial layer is reconfigured at a rate $M/T$.
• Figure 2: Time-slot structure within a coherence block. Each slot of duration $T_{\text{s}}$ begins with a downlink training/CSI acquisition phase of length $T_{\text{train}}$, followed by $P$ data-symbol intervals of duration $T_{\text{b}}$ indexed by $p \in \{0,\ldots,P-1\}$. This pattern repeats $M$ times so that $M T_{\text{s}}$ spans the channel coherence time $T$.
• Figure 3: Objective function \ref{['eq:obj']} versus the number of passive layers $L_{\text{pc}}$ for $Q \in \{ 25, 36 , 49 , 64\}$ meta-atoms. Parameters are set to $Z = 9$, $V = 25$, and $L_{\text{ac}} = 4$. The case $Q=V=25$ corresponds to a baseline SIM architecture without absorbing meta-atoms.. All metasurface layers are square. The signal from the RF chain first passes through the AC layers and, then, through the PC layers.
• Figure 4: Convergence rate of the PGD algorithm for $Q \in \{ 25, 36 , 49 , 64\}$ meta-atoms. Parameters are set to $Z = 9$, $V = 25$, $L_{\text{ac}} = 4$, and $L_{\text{pc}} = 8$. The case $Q=V=25$ corresponds to the baseline SIM without DAL. All metasurface layers are square. The signal from the RF chain first passes through the AC layers and, then, through the PC layers.
• Figure 5: Power gain of a DAL-aided and conventional SIM employing only PC layers for $V \in \{ 9, 16 , 25\}$ meta-atoms. Parameters are set to $Z = 9$ and $Q = 25$. The case $Q=V=25$ corresponds to a baseline SIM architecture without absorbing meta-atoms. All metasurface layers are square.