Connected cyber-physical systems perform inference based on real-time inputs from multiple data streams. Uncertain communication delays across data streams challenge the temporal flow of the inference process. State-of-the-art (SotA) non-blocking inference methods rely on a reference-modality paradigm, requiring one modality input to be fully received before processing, while depending on costly offline profiling. We propose a novel, neuro-inspired non-blocking inference paradigm that primarily employs adaptive temporal windows of integration (TWIs) to dynamically adjust to stochastic delay patterns across heterogeneous streams while relaxing the reference-modality requirement. Our communication-delay-aware framework achieves robust real-time inference with finer-grained control over the accuracy-latency tradeoff. Experiments on the audio-visual event localization (AVEL) task demonstrate superior adaptability to network dynamics compared to SotA approaches.

Autonomous reconfigurable intelligent surfaces (RISs) offer the potential to simplify deployment by reducing the need for real-time remote control between a base station (BS) and an RIS. However, we highlight two major challenges posed by autonomy. The first is implementation complexity, as autonomy requires hybrid RISs (HRISs) equipped with additional onboard hardware to monitor the propagation environment and perform local channel estimation (CHEST), a process known as probing. The second challenge, termed probe distortion, reflects a form of the observer effect: during probing, an HRIS can inadvertently alter the propagation environment, potentially disrupting the operations of other communicating devices sharing the environment. Although implementation complexity has been extensively studied, probe distortion remains largely unexplored. To further assess the potential of autonomous RISs, this paper comprehensively and pragmatically studies the fundamental trade-offs posed by these challenges collectively. In particular, we examine the robustness of an HRIS-assisted massive multiple-input multipleoutput (mMIMO) system by considering its critical components and stringent conditions. The latter include: 1) two extremes of implementation complexity, represented by minimalist operation designs of two distinct HRIS hardware architectures, and 2) an oblivious BS that fully embraces probe distortion. To make our analysis possible, we propose a physical-layer orchestration framework that aligns HRIS and mMIMO operations. We present empirical evidence that autonomous RISs remain promising under stringent conditions and outline research directions to deepen probe distortion understanding.

This paper introduces EcoPull, a sustainable Internet of Things (IoT) framework powered by Tiny Machine Learning (TinyML) models for efficient image retrieval from multiple devices. The devices are equipped with two types of TinyML models: i) a behavior model and ii) an image compressor model. The behavior model filters out irrelevant images based on the current task, minimizing unnecessary data transmission and reducing communication resource competition among devices. The image compressor model enables devices to communicate with the edge server (ES) using latent representations of images, thereby reducing communication bandwidth usage. While integrating TinyML models into IoT devices does increase energy consumption due to the inference process, this cost is carefully accounted for in our design. Numerical results show that the proposed framework can achieve over 77% and 43% energy savings compared to the simple offloading and a state-of-the-art baseline while still maintaining the quality of the retrieved images at the ES.

The widespread adoption of Reconfigurable Intelligent Surfaces (RISs) in future practical wireless systems is critically dependent on the integration of the RIS into higher-layer protocols beyond the physical (PHY) one, an issue that has received minimal attention in the research literature. In light of this, we consider a classical random access (RA) problem, where uncoordinated users’ equipment (UEs) transmit sporadically to an access point (AP). Differently from previous works, we ponder how a RIS can be integrated into the design of new medium access control (MAC) layer protocols to solve such a problem. We consider that the AP is able to control a RIS to change how its reflective elements are configured, namely, the RIS configurations. Thus, the RIS can be opportunistically controlled to favor the transmission of some of the UEs without the need to explicitly perform channel estimation (CHEST). We embrace this observation and propose a RIS-assisted RA protocol comprised of two modules: Channel Oracle and Access. During channel oracle, the UEs learn how the RIS configurations affect their channel conditions. During the access, the UEs tailor their access policies using the channel oracle knowledge. Our proposed RIS-assisted protocol is able to increase the expected throughput by approximately 60% in comparison to the slotted ALOHA (S-ALOHA) protocol.

In a cell-free massive multiple-input–multiple-output (CF-mMIMO) network, multiple access points (APs) actively cooperate to serve users’ equipment (UEs). We consider how the random access (RA) problem can be addressed by such a network under the occurrence of pilot collisions. To find a solution, we embrace the user-centric perspective, which basically dictates that only a preferred set of APs needs to serve a UE. Due to the success of the strongest-user collision resolution (SUCRe) protocol for cellular (Ce) mMIMO, we extend it by considering the new setting. Besides, we establish that the user-centric perspective naturally equips a CF network with robust fundamentals for resolving collisions. We refer to this foundation as spatial separability, which enables multiple colliding UEs to access the network simultaneously. We then propose two novel RA protocols for CF-mMIMO: 1) the baseline cell-free (BCF) that resolves collisions with the concept of spatial separability alone and 2) the cell-free sucre (CF-SUCRe) that combines SUCRe and the spatial separability principle to resolve collisions. We evaluate our proposed RA protocols against the Ce-SUCRe. Respectively, the BCF and CF-SUCRe can support$7\times $and$4\times $more UEs’ access on average compared to the Ce-SUCRe with an average energy efficiency gain based on the total power consumed (TPC) by the network per access attempt of$52\times $and$340\times $. Among our procedures, even with a higher overhead, the CF-SUCRe is superior to BCF regarding TPC per access attempt. This is because the combination of methods for collision resolution allows many APs to be disconnected from the RA process without sacrificing much the performance. Finally, our numerical results can be reproduced using the code package available on:github.com/victorcroisfelt/cf-ra-spatial-separability.

In order to allow the operation of cheap and simple computational nodes, the reduction of the complexity related to the computation of signal processing techniques in massive multiple-input multiple-output (M-MIMO) is a desirable property to be obtained. With this in mind, several methods have been proposed to reduce the complexity of the classical combining/precoding schemes, being one of these based on the Kaczmarz algorithm (KA). Recent works demonstrated that KA-based approaches can suitably estimate the signals received and transmitted in the uplink and downlink phases; simultaneously, the procedure seemed to result in a very low computational complexity. Motivated by such context, this paper proposes a modification in the primary KA-based scheme for M-MIMO, striving to improve its rate of convergence, reducing complexity while holds the performance. Such improvement underlies the effects of pathloss and shadowing over the KA's rate of convergence, as well as in the spectral efficiency. Numerical results are given supporting that the proposed method outclasses the original proposal.