We present Neuro-centric Networks (Neuro-C), a neural network architecture we design to eliminate multiply-accumulate operations for efficient inference on ultra-low-power microcontrollers (MCUs). Although some MCUs include specialized hardware for neural acceleration, many ultra-low-power MCUs do not, requiring neural networks to align with limited compute and memory resources. Rather than compressing existing models or assuming dedicated hardware, Neuro-C integrates hardware constraints directly into the architecture, effectively shaping the network design around the limitations of the target platform. We shift the computational burden from connections to neurons and encode connectivity with a fixed ternary adjacency matrix, overcoming the bottleneck of matrix multiplications and large weight storage. This design enables a specialized inference kernel implementation that reduces memory usage and latency through pointer-based traversal and sparse dynamic memory allocation, complex control flows, and index decoding logic common in sparse or compressed models. Experimental results show that Neuro-C achieves accuracy comparable to or better than standard multilayer perceptrons across multiple datasets, while reducing inference latency and program memory usage by up to 90%. Compared to conventional ternary neural networks, Neuro-C provides improved convergence and accuracy under identical architectural settings, with negligible impact on inference latency. © 2026

In outdoor mobile robots, terrain perception can be degraded by noisy, corrupted, or missing sensor inputs. We investigate robust multimodal terrain recognition using proprioceptive audio and IMU sensing. We propose a lightweight multimodal pipeline and evaluate adaptive fusion strategies under controlled degradations, including additive noise and modality dropout. Experiments on a new terrain dataset collected with a robotic mower, and on an existing audio-IMU related dataset, show that adaptive gating degrades more gracefully than standard feature fusion while remaining suitable for real-time deployment on embedded hardware. © 2026

The need for generating radio signals with a high frequency stability and very low phase noise is a demanding requirement in communication systems. In low-power designs, such high-frequency signal generation is often the main source of power consumption, which necessitates low-power alternatives. Although tunnel diodes can generate high-frequency signals at low power, the generated signal is frequency unstable with high phase noise and unwanted harmonics. State-of-the-art address these limitations through injection locking, which requires an external signal generator that significantly increases the overall system power consumption. We present SharpPeak, a low-power long-range transmitter design that achieves high frequency stability using tunnel diodes without relying on external injection signals. This design lowers both system power and transmitter complexity, while improving frequency stability, phase noise, and frequency drifts. SharpPeak achieves high (370 kbps) data rates and long (1 km) communication range with under 175 μW power consumption at the RF front-end, advancing the state-of-the-art in energy-efficient communication systems. We believe that this work is a significant advancement in the development of a new generation of low-power communication systems. © 2026

Wireless IoT networks are becoming increasingly important as they support a diverse range of applications. However, due to their resource constraints and limited security capabilities, these networks are particularly susceptible to attacks. Among these, jamming attacks pose a significant threat by degrading packet delivery, disrupting communications, and depleting the limited networking and energy resources of IoT systems. Hence, effective jamming detection is essential for safeguarding wireless IoT networks. However, dynamic operating environments and diverse application scenarios make it challenging to design a robust jamming detection system. To address the challenge, this paper introduces AJDet, a lightweight, online, Adaptive Jamming attack Detection system. AJDet leverages an online adaptation approach that enables the detection of both proactive and reactive jamming attacks without relying on a deployment-specific configuration. Our experimental results demonstrate that AJDet achieves over 96.5% detection accuracy without the need for manual intervention after deployment. Moreover, the system exhibits an efficient memory footprint, making it well-suited for resource-constrained devices.

Analog backscatter tags are primarily used for lowpower sensing applications. These tags communicate sensor data by directly modulating a carrier signal, typically as frequency variations. The processing and computation overhead is delegated to a remote receiver, thus analog backscatter tags require very low power for communication compared to digital backscatter. However, when multiple tags are deployed, all tags are activated by the carrier simultaneously, synchronizing their transmissions. Overlapping frequencies and their harmonics make it difficult for the receiver to read the tags' sensor data and associate the frequency to the tag that provides the data. Although multiple carrier frequencies can be used to query tags, backscatter tags are 'frequency agnostic' and reflect carriers of any frequency. We propose a solution to desynchronize tags, i.e., prevent synchronous transmissions from multiple tags, by limiting the tag activation frequencies without modifying the backscatter modules. Towards this end, we integrate a band-pass filter into the tag's frontend circuitry and employ a capacitor to further narrow the frequency band. Our results demonstrate that integrating a capacitor improves the frequency selectivity of our chosen filter by five times compared to solely relying on a band-pass filter.

Reliable communication requires error detection and correction mechanisms. As the error characteristics depend on the physical layer, we study them for two recent physical layers: power-plane communications (PPC) in fabric and spacecraft multi-layer insulation (MLI). As a first step towards this goal, we investigate the dependency between the medium parameters (such as temperature, wetness, ambient radio, bending and crumpling) and the occurrence of errors in data transfer between sensor nodes in conductive fabric and MLI. Our results indicate that some medium parameters lead to error generation and loss of data transfer while some have no effect on the occurrence of errors. While our earlier work has assumed that it is only capacitance that attenuates the modulated carrier signal in PPC, our results show that other medium parameters such as electrolysis affect data transfer. Electrolysis has two effects: it leads to loss of data transfer in the case of conductive fabric with metallic mesh. In the case of fabric with silver coated threads or copper, electrolysis destroys the channel. In the case of MLI, bending and crumpling lead to error generation with a bit error probability of almost 90%. We also demonstrate that data signals can be used as an indicator of fabric wetness. They can also be used to indicate a micrometeorite impact with a spacecraft. We further show that the use of Manchester encoding and implementation of the data slicer for PPC provides a robust way to restore the transmitted data when the medium parameters change

In-body communication is a key enabler for next-generation healthcare applications, allowing seamless networking of implants. Fat tissue, with its lower water content and reduced signal attenuation compared to other body tissues at microwave frequencies, has emerged as a promising medium for radio-based in-body networks. Despite this advantage, signal leakage through the body can compromise privacy, exposing sensitive data and the mere presence of implants to external adversaries. This paper investigates the feasibility of covert communication in fat tissue-based in-body networks by leveraging the previously unexplored signal attenuation properties of human tissue to transmit data undetectable to adversaries, ensuring privacy beyond encryption. We develop a system in which an implanted transmitter communicates discreetly with an implanted receiver, shielded from external passive eavesdroppers. Our theoretical analysis and experimental results demonstrate that the attenuation properties of human tissues enable covert communication at reduced transmit power levels without requiring friendly jamming, unlike over-the-air systems. To further enhance covertness, we explore the use of an external friendly jammer and show its significant benefits. Experimental results show a 500% increase in the maximum channel capacity of covert communication, from 2.86 bps/Hz at -56 dBm transmit power without jamming, to 17 bps/Hz with no bit errors at 0 dBm transmit power with a friendly jammer, using the IEEE 802.15.4 standard for communication in the 2.45 GHz frequency band. These findings highlight that covert communication is achievable in fat tissue-based in-body networks at low data rates without additional infrastructure such as an external jammer. For applications requiring higher data rates, a friendly jammer offers a scalable solution, making this approach practical for a wide range of implant communication scenarios.

We report on our experience deploying a CubeSat to study fault and error distributions against different fault tolerance schemes when using Common Off-The-Shelf (COTS) hardware in Low-Earth Orbit (LEO). Space radiation commonly causes faults in COTS hardware, such as bit flips in memory, which can lead to errors in a program’s execution. Fault tolerance techniques can prevent faults from turning into errors. Accurately quantifying the fault and error distributions is vital for choosing an appropriate fault tolerance scheme. We equip the CubeSat with heterogeneous hardware combining a regular System on a Chip (SoC) with programmable logic resources. Based on in-orbit experiments and post-processing of logs, we check the validity of two fault models. We find the single fault model to be valid, encouraging the use of techniques such as triple modular redundancy. We also demonstrate, however, that the single-bit error fault model is not valid, which means that common techniques such as Hamming (7,4) codes should not be used. We observe that most errors are short-lived, allowing simple reexecutions to correct them. Contrary to intuition, we also conclude that floating-point encodings are more appropriate to build faulttolerance schemes in our setting, due to faults being easier to detect than in their integer counterparts. Our insights confirm existing findings in the literature while also providing new ones, while providing a foundation to conceive fault tolerance schemes for COTS hardware deployments in space

The rapid expansion of Internet of Things (IoT) networks has increased the demand for low-power communication technologies, with Zero-Energy Devices (ZEDs) being particularly important as they overcome battery limitations by harvesting energy from the environment. Backscatter communication has emerged as a promising solution for enabling such devices. Traditional backscattering suffers from self-interference, which often requires complex interference mitigation techniques. Harmonic backscattering addresses this challenge by using a harmonic frequency for the backscattered signal, thereby eliminating interference issues and simplifying system design. In this paper, we develop a mathematical model for harmonic backscattering and compare its performance with regular backscattering. Our results show that harmonic backscattering significantly outperforms regular backscattering.