The computing power of the photonic Convolution Neural Network (CNN) have achieved Tera-level operations per second (OPS) for supporting machine vision algorithms. However, restrained by the nonlinear characteristics ...
详细信息
The computing power of the photonic Convolution Neural Network (CNN) have achieved Tera-level operations per second (OPS) for supporting machine vision algorithms. However, restrained by the nonlinear characteristics of weighting devices, it is hard to realize convergent training of photonic CNN under poor controlling precision. In this paper we proposed a Physical Aware Clustering training method where the Physical Aware Cluster Quantizer is embed with the straight through estimator (STE) algorithm for integrated photonic CNN with quantized and nonlinear distributed weights. Experiment shows that, by employing the upgraded STE methods (namely, STE plus), the photonic CNN using micro ring weighting bank and PAM4 controlling modules achieves the nearly 99.3% accuracy ratio for Fashion MNIST recognition task, whereas only 41.5% of that by using baseline STE training algorithms with the same physical devices.
Mobile edge computing (MEC) is a new paradigm that improves the quality of service compared with traditional cloud computing. In MEC, computational tasks are submitted by numerous end users and are partially offloaded...
详细信息
ISBN:
(数字)9781665410205
ISBN:
(纸本)9781665410212
Mobile edge computing (MEC) is a new paradigm that improves the quality of service compared with traditional cloud computing. In MEC, computational tasks are submitted by numerous end users and are partially offloaded to edge servers or a central cloud. However, the characteristics of tasks are different from each other, and the limited resources of computational nodes are also heterogeneous, which brings great challenges to computation offloading and resource allocation for MEC. This work establishes an end-edge-cloud collaborative computing network, which consists of end devices, edge servers, and a central cloud. Task execution location and CPU running frequency determine the execution time and energy consumption to finish the tasks. Considering the aforementioned factors, a multi-objective constrained optimization problem is formulated. To solve the problem, an improved Non-dominated Sorting Genetic Algorithm ii (NSGA-ii) with self-adaptive crossover and mutation rates is proposed, which is called Improved NSGA-ii with _Self-adaptive Crossover and Mutation (INSCM). The total execution time and energy consumption can be jointly minimized with our proposed INSCM. Numerous experiments are carried out to test the performance of INSCM. Simulation results show that INSCM effectively improves the performance of NSGA-ii and surpasses random offloading and NSGA-iiI, which shows practical use in real-life scenarios.
Direction of arrival estimation technology is the foundation for communications, radar, navigation, etc. However, conventional electronic devices for DOA estimation require expensive RF circuits, high-precision analog...
详细信息
ISBN:
(数字)9798350379266
ISBN:
(纸本)9798350379273
Direction of arrival estimation technology is the foundation for communications, radar, navigation, etc. However, conventional electronic devices for DOA estimation require expensive RF circuits, high-precision analogue-to-digital converters and complex digital signal processing algorithms. Here, we propose a super-resolution diffractive neural network (S-DNN) to directly process electromagnetic (EM) waves for super-resolution DOA estimation over a broadband frequency range. The multilayer meta-structures of the S-DNN produces super-oscillatory angular responses in local angular range, which can perform all-optical DOA estimation with angular resolution beyond the diffraction limit. Space-time multiplexing of passive and reconfigurable S-DNNs is used to achieve high-resolution DOA estimation over a wide field of view. Experiments show that the angular resolution of the S-DNN is four times higher than the diffraction limited resolution. The fabricated S-DNN can be used for superresolution DOA estimation of multiple coherent sources over a 5 GHz frequency bandwidth, with an estimation delay that is in principle two to four orders of magnitude lower than commercial devices. We also use the all-optical DOA estimation capability of the S-DNN to provide angular direction to reconfigurable intelligent surface, to achieve low-latency and low-power integrated sensing and communication. Our work is an important step towards enabling various wireless sensing and communication tasks using photoniccomputing processors that outperform electronic computing in both computational paradigm and performance.
The benefits introduced by novel network technologies such as 5G and beyond, including low latency and support for billions of devices, have the potential to transform the lives of people. However, the features promis...
详细信息
The widespread usage of wearables set the foundations for many new applications that process the wearable sensor data. Human Activity Recognition (HAR) is a well-studied application that targets to classify the data c...
详细信息
Current radio frequency (RF) sensors at the Edge lack the computational resources to support practical, in-situ training for intelligent spectrum monitoring. This is true for sensor data classification in general. We ...
详细信息
ISBN:
(纸本)9781665412520
Current radio frequency (RF) sensors at the Edge lack the computational resources to support practical, in-situ training for intelligent spectrum monitoring. This is true for sensor data classification in general. We propose a solution via Deep Delay Loop Reservoir computing (DLR), a processing architecture that supports general machine learning algorithms on compact mobile devices by leveraging delay-loop reservoir computing in combination with innovative electro-optical hardware. With both digital and photonic realizations of our design of the loops, DLR delivers reductions in form factor, hardware complexity and latency, compared to the State-of-the-Art (SoA). The main impact of the reservoir is to project the input data into a higher dimensional space of reservoir state vectors in order to linearly separate the input classes. Once the classes are well separated, traditionally complex, power-hungry classification models are no longer needed for the learning process. Yet, even with simple classifiers based on Ridge regression (RR), the complexity grows at least quadratically with the input size. Hence, the hardware reduction required for training on compact devices is in contradiction with the large dimension of state vectors. DLR employs a RR-based classifier to exceed the SoA accuracy, while further reducing power consumption by leveraging the architecture of parallel (split) loops. We present DLR architectures composed of multiple smaller loops whose state vectors are linearly combined to create a lower dimensional input into Ridge regression. We demonstrate the advantages of using DLR for two distinct applications: RF Specific Emitter Identification (SEI) for IoT authentication, and wireless protocol recognition for IoT situational awareness.
WiFi-based perception systems can realize various gesture recognition in theory, but they cannot realize large-scale applications in practice. Later, some work solved the problem of cross-domain identification of the ...
详细信息
ISBN:
(纸本)9783030953881;9783030953874
WiFi-based perception systems can realize various gesture recognition in theory, but they cannot realize large-scale applications in practice. Later, some work solved the problem of cross-domain identification of the WiFi system, and promoted the possibility of the practical application of WiFi perception. However, the existing cross-domain recognition work requires a large number of calculations to extract motion features and recognition through a complex network, which determines that it cannot be deployed directly on edge devices. In addition, some hardware limitations of edge devices (for example, the network card is a single antenna), the amount of data we obtain is far less than that of the general network card. If the original data is not calibrated, the error information carried by the data will have a huge impact on the recognition result. Therefore, in order to solve the above problems, we propose WiRD, a system that can accurately calibrate the amplitude and phase in the case of a single antenna, and can be deployed on edge devices to achieve real-time detection. Experimental results show that WiRD is comparable to existing methods for gesture and body recognition within the domain, and has 87% accuracy for gesture recognition cross the domain, but the overall system processing time is reduced by 9x and the model inference time is reduced by 50x.
Real-time monitoring systems in remote sensing could enable a variety of new applications concerning emergency and disaster management. Coupling remote sensing imaging devices with hardware accelerators for artificial...
详细信息
ISBN:
(数字)9781665485746
ISBN:
(纸本)9781665485746
Real-time monitoring systems in remote sensing could enable a variety of new applications concerning emergency and disaster management. Coupling remote sensing imaging devices with hardware accelerators for artificial intelligence (AI) post-processing algorithms can pave the way for future constellations of small satellites offering services such as detection of wildfires, volcanoes eruptions, landslides, floods, etc. This paper presents MONSTER, a robotic facility hosted in the ARCA laboratory (ARCAlab) at the School of Aerospace Engineering in Rome. The facility was born as a simulator of lunar landing operations, but is currently being updated to perform hardware-in-the-loop simulation campaigns of many different scenarios including wildfires and volcanic eruption observation from space or airborne platforms. Having the possibility to perform translation and rotation maneuvers of the target observation platform, properly simulating the terrain, and performing the AI-aided computation on hardware accelerators such as graphics processing units (GPUs), visual processing units (VPUs), and field programmable gate arrays (FPGAs), the MONSTER facility could be used to efficiently simulate new remote sensing paradigms for disaster monitoring through on-the-edge artificial intelligence.
A widespread challenge in the industrial domain is the modernization and digitization of assembly processes involving human workers to increase production efficiency and thus stay competitive with rival companies. Spe...
详细信息
ISBN:
(纸本)9781450396318
A widespread challenge in the industrial domain is the modernization and digitization of assembly processes involving human workers to increase production efficiency and thus stay competitive with rival companies. Specifically, in assembly processes involving low lot sizes, human workers are required to deal with variations to individual assembly work processes due to product customization. In case of complex tasks this leads to mistakes and further expensive dis- and reassembly steps. This paper investigates which are the quantifiably best data sources, pre-procession steps, features, and machine learning algorithms to determine the correct execution of a specific work process in the manufacturing environment. To answer this question, a wearable sensor system consisting of multiple heterogeneous sensor devices was developed. The data used for this work was specifically collected from the actual production environment in multiple recording sessions, and particular focus was given to achieve this in a realistic yet controlled way. An assistance provisioning pipeline for industrial workers consisting of (i) an activity recognition system, (ii) a work flow correlation engine, (iii) a wrench activity estimator and a (iv) feedback system was developed. These systems were designed and evaluated using authentic, task-specific expert knowledge and using a grid search study to determine the best selection of data sources, pre-procession steps, features, and machine learning algorithms. This study was able to answer the given research question and reifies the final results in the form of a guidance system to be deployed in an industrial manufacturing line.
The popularity of 4K videos is on the rise. However, streaming such high-quality videos over mm Wave to several users presents significant challenges due to directional communication, fluctuating channels, and high ba...
详细信息
ISBN:
(数字)9798350386059
ISBN:
(纸本)9798350386066
The popularity of 4K videos is on the rise. However, streaming such high-quality videos over mm Wave to several users presents significant challenges due to directional communication, fluctuating channels, and high bandwidth demands. To address these challenges, this paper introduces an innovative 4K layered video multicast streaming system. We (i) develop a video quality model tailored for layered video coding, (ii) optimize resource allocation, scheduling, and beamforming based on the channel conditions of different users, and (iii) design a streaming strategy that integrates fountain code to eliminate redundancy in multicast groups, coupled with a Leaky-Bucket approach for congestion control. We implement our system on Commodity-Off- The-Shelf (COTS) WiGig devices and demonstrate its effectiveness through comprehensive testbed and emulation experiments.
暂无评论