This paper presents the design of an ultra-low energy, rakeness-based compressed sensing (CS) system that utilizes time-mode (TM) signal processing (TMSP). To realize TM CS operation, the presented implementation makes use of monostable multivibrator based analog-to-time converters, fixed-width pulse generators, basic digital gates and an asynchronous time-to-digital converter. The TM CS system was designed in a standard 0.18 µm IC process and operates from a supply voltage of 0.6V. The system is designed to accommodate data from 128 individual sensors and outputs 9-bit digital words with an average reconstruction SNR of 35.31 dB, a compression ratio of 3.2, with an energy dissipation per channel per measurement vector of 0.621 pJ at a rate of 2.23 k measurement vectors per second.@en

Due to its intrinsic sparsity both in time and space, event-based data is optimally suited for edge-computing applications that require low power and low latency. Time varying signals encoded with this data representation are best processed with Spiking Neural Networks (SNN). In particular, recurrent SNNs (RSNNs) can solve temporal tasks using a relatively low number of parameters, and therefore support their hardware implementation in resource-constrained computing architectures. These premises propel the need of exploring the properties of these kinds of structures on low-power processing systems to test their limits both in terms of computational accuracy and resource consumption, without having to resort to full-custom implementations. In this work, we implemented an RSNN model on a low-end, resource-constrained ARM-Cortex-M4-based Micro Controller Unit (MCU). We trained it on a down-sampled version of the N-MNIST event-based dataset for digit recognition as an example to assess its performance in the inference phase. With an accuracy of 97.2%, the implementation has an average energy consumption as low as 4.1μJ and a worst-case computational time of 150.4μs per time-step with an operating frequency of 180 MHz, so the deployment of RSNNs on MCU devices is a feasible option for small image vision real-time tasks.@en

Atrial electrograms (AEGs) acquired with a high spatio-temporal resolution are a promising approach for early detection of atrial fibrillation. Due to the high data rate, transmission of AEG signals is expensive in terms of power consumption and resources, making its adoption a challenge for low-power wireless devices. In this paper, we investigate the feasibility of using compressed sensing (CS) for the acquisition of AEGs while reducing redundant data without losing information. We apply two CS approaches, standard CS and rakeness-based CS (rak-CS) on a data set, composed of real medical recordings. We find that the AEGs are compressible in time, and, more interestingly, in the spatial domain. The performance of rak-CS is better than standard CS, especially at higher compression ratios (CR), both during sinus rhythm (SR) and atrial fibrillation (AF). The difference in the achieved average reconstruction signal-to-noise (ARSNR) in rak-CS and standard CS, for CR = 4.26, in the time domain is 7.7 dB and 2.6 dB for AF and SR, respectively. Multi-channel data is modeled as a multiple-measurement-vector problem and the mixed norm is used to exploit the group structure of the signals in the spatial domain to obtain improved reconstruction performance over $l_{1}$ norm minimization. Using the mixed-norm recovery approach, for CR = 4.26, the difference in achieved ARSNR performance between rak-CS and standard CS is 5 dB and 2 dB for AF and SR, respectively.@en

Atrial electrogram (AEG) acquired with a high spatio-temporal resolution is a promising approach for early detection of atrial fibrillation. Due to the high data rate, transmission of AEG signals requires considerable energy, making its adoption a challenge for low-power wireless devices. In this paper, we investigate the feasibility of using compressed sensing (CS) for the acquisition of AEGs while reducing redundant data without losing information. We apply two CS approaches, standard CS and rakeness-based CS (rak-CS) on real medical recordings. We find that the AEGs are compressible in time, and, more interestingly, in the spatial domain. The performance of rak-CS is better than standard CS, especially at higher compression ratios (CR), both during sinus rhythm (SR) and atrial fibrillation (AF). More specifically, the difference in the achieved average reconstruction signal-to-noise (ARSNR) in rak-CS and standard CS, for CR = 4.26, in the time domain is 7.7 dB and 2.6 dB for AF and SR, respectively. Multi-channel data is modeled as a multiple-measurement-vector problem and a suitable mixed norm is used to exploit the group structure of the signals in the spatial domain to obtain improved reconstruction performance over l1 norm minimization. Using the mixed-norm recovery approach, for CR = 4.26, the difference in achieved ARSNR performance between rak-CS and standard CS is 5 dB and 2 dB for AF and SR, respectively.@en