Search

In this paper, we present a distributed cooperative surrounding control for mobile robots with uncertainties and aperiodic sampling. At the kinematic level, a cooperative circumnavigation guidance law with the capability of spatial-temporal decoupling, employing a line-of-sight (LOS) principle to accomplish closed orbit following and a path parameter synchronization to assign the desired speed, is devised to generate reference velocity and angular rate. At the kinetic level, to reduce the transmission burden in sensor-to-controller channel and compensate for the total uncertainties existing in velocity channels, aperiodic sampling based extended state observers (AS-ESOs), are developed to provide precise disturbance estimates with guaranteed convergence, while a nonnegative threshold-based event-triggering condition with a straightforward tuning procedure is designed to schedule the communication frequency without inducing Zeno behaviors. Then a robust anti-disturbance kinetic control protocol is synthesized that renders an equiangular distribution along the common circle. The salient merit is that a symmetric formation pattern over a closed curve defined by parameterized path, instead of time-related functions, can be obtained in a distributed manner with decreased sampling cost and disturbances. Moreover, all error variables in the closed-loop system are demonstrated to be bounded. Finally, the effectiveness of algorithm is verified by simulations. [ABSTRACT FROM AUTHOR]

This work investigates multiband spectrum sensing in a cognitive radio network where a multi-antenna secondary user attempts to detect several consecutive frequency bands occupied by multiple primary users. Specifically, we propose a multiband joint spectrum sensing approach based on the covariance matrix-aware convolutional neural network (CNN), in which the multiband sample covariance matrices are concatenated as the input of CNN. The proposed approach is free of model assumptions and the hidden correlation features between subbands can be learned to improve the spectrum sensing performance. Simulation results show that the proposed algorithm outperforms the state-of-the-art spectrum sensing methods in cases with and without the noise uncertainty. [ABSTRACT FROM AUTHOR]

In health assessment and disease prevention, vital signs such as breathing rate, heartbeat rate, blood pressure and posturography play a critical role. Not only the elderly, but also young people need to pay attention to vital signs monitoring because more and more young people die of sudden diseases, such as sudden death and myocardial infarction. Traditional vital signs monitoring methods are invasive and inconvenient to use, especially for the elderly because these methods usually require users to wear some dedicated sensors. In recent years, some non-invasive vital signs monitoring methods based on wireless sensing have been proposed. Usually, these methods need the user to remain still. When the user walks around, these methods cannot work well. One of the challenges of mobile people breathing monitoring is that the small body chest displacement caused by breathing is disturbed by people’s walking. In this paper, we propose RF-RES, which is an accurate continuous breathing monitoring method based on RFID technology and uses a pair of tags to eliminate the influence of the user’s body movement. We merge the data of the tag pairs to eliminate the influence of the user’s entire body movement and propose an algorithm based on doppler frequency shift to track the user’s displacement during exercise. We conduct practical experiments to evaluate the performance with Impinj R420 and passive tags. The results show that RF-RES achieves accurate respiratory rate monitoring when estimating the respiratory rate of different users, with an average error of 0.54 BPM. [ABSTRACT FROM AUTHOR]

In-memory computing (IMC) addresses the cost of accessing data from memory in a manner that introduces a tradeoff between energy/throughput and computation signal-to-noise ratio (SNR). However, low SNR posed a primary restriction to integrating IMC in larger, heterogeneous architectures required for practical workloads due to the challenges with creating robust abstractions necessary for the hardware and software stack. This work exploits recent progress in high-SNR IMC to achieve a programmable heterogeneous microprocessor architecture implemented in 65-nm CMOS and corresponding interfaces to the software that enables mapping of application workloads. The architecture consists of a 590-Kb IMC accelerator, configurable digital near-memory-computing (NMC) accelerator, RISC-V CPU, and other peripherals. To enable programmability, microarchitectural design of the IMC accelerator provides the integration in the standard processor memory space, area- and energy-efficient analog-to-digital conversion for interfacing to NMC, bit-scalable computation (1–8 b), and input-vector sparsity-proportional energy consumption. The IMC accelerator demonstrates excellent matching between computed outputs and idealized software-modeled outputs, at 1b TOPS/W of 192|400 and 1b-TOPS/mm2 of 0.60|0.24 for MAC hardware, at VDD of 1.2|0.85 V, both of which scale directly with the bit precision of the input vector and matrix elements. Software libraries developed for application mapping are used to demonstrate CIFAR-10 image classification with a ten-layer CNN, achieving accuracy, throughput, and energy of 89.3%|92.4%, 176|23 images/s, and 5.31|105.2 μJ/image, for 1|4 b quantization levels. [ABSTRACT FROM AUTHOR]

The ubiquitous adoption of systems specialized for AI requires bridging two seemingly conflicting challenges—the need to deliver extreme processing efficiencies while employing familiar programming interfaces, making them compelling even for non-expert users. We take a significant first step towards this goal and present an end-to-end software stack for the RaPiD AI accelerator developed by IBM Research. We present a set of software extensions, called Deeptools, that leverage and work within popular deep learning frameworks. DeepTools requires no additional user input and enables aggressive, accelerator-specific performance optimization akin to a full, custom framework. DeepTools has two key components: 1) a compiler runtime called DeepRT, which automatically identifies how best to execute a given DNN graph on RaPiD and constructs the requisite program binaries; and 2) an execution runtime called RaPiDLib, which triggers and manages the execution of compute and data-transfer operations on RaPiD. We integrate DeepTools with TensorFlow and map popular DNNs (AlexNet, VGG, ResNet, LSTM) to RaPiD. We demonstrate substantial improvement in performance over hand-tuned mappings. [ABSTRACT FROM AUTHOR]

In-memory computing is an emerging approach for overcoming memory-accessing bottlenecks, by eliminating the costs of explicitly moving data from point of storage to point of computation outside the array. However, computation increases the dynamic range of signals, such that performing it via the existing structure of dense memory substantially squeezes the signal-to-noise ratio (SNR). In this brief, we explore how computations can be scaled up, to jointly optimize energy/latency/bandwidth gains with SNR requirements. We employ algorithmic techniques to decompose computations so that they can be mapped to multiple parallel memory banks operating at chosen optimal points. Specifically focusing on in-memory classification, we consider a custom IC in 130-nm CMOS IC and demonstrate an algorithm combining error-adaptive classifier boosting and multi-armed bandits, to enable segmentation of a feature vector into multiple subsets. The measured performance of 10-way MNIST digit classification, using images downsampled to $16{\times }16$ pixels (mapped across four separate banks), is 91%, close to that simulated using full unsegmented feature vectors. The energy per classification is 879.7 pJ, $14.3{\times }$ lower than that of a system based on separated memory and digital accelerator. [ABSTRACT FROM AUTHOR]

For each brain–computer interface system, efficiency is a key issue that considers both accuracy and speed. The P300 spellers built upon oddball paradigm are usually less efficient due to the repetitive stimulation of multiple characters for reliable detection. In this paper, based on the online EEG signal, we propose an interactive paradigm for P300 speller to improve its efficiency, primarily focusing within the single characterP300 paradigm. Specifically, after each stimulation, we first evaluate the posterior probability of each character in the stimuli set to be the target. The lowprobability characters are then removed fromthe stimuli set in the subsequent round(s), as character flash continues until the probability of any character surpasses a predefined threshold. Then, the character is selected as the target and data collection for the trial terminates. By reducing stimulus sequence characters, the system efficiency can be substantially improved. The spelling accuracy is insignificantly affected as the characters being removed have low probability to be the target. The online experimental results from a total of eight subjects show that an average practical information transfer rate of 50.26 bits/min (i.e. 9.07 characters/min) has been achieved, with 91% average spelling accuracy rate. [ABSTRACT FROM AUTHOR]

This paper presents a machine-learning classifier where computations are performed in a standard 6T SRAM array, which stores the machine-learning model. Peripheral circuits implement mixed-signal weak classifiers via columns of the SRAM, and a training algorithm enables a strong classifier through boosting and also overcomes circuit nonidealities, by combining multiple columns. A prototype 128 $\times $ 128 SRAM array, implemented in a 130-nm CMOS process, demonstrates ten-way classification of MNIST images (using image-pixel features downsampled from 28 $\times $ 28 = 784 to 9 $\times $ 9 = 81, which yields a baseline accuracy of 90%). In SRAM mode (bit-cell read/write), the prototype operates up to 300 MHz, and in classify mode, it operates at 50 MHz, generating a classification every cycle. With accuracy equivalent to a discrete SRAM/digital-MAC system, the system achieves ten-way classification at an energy of 630 pJ per decision, 113 times lower than a discrete system with standard training algorithm and 13 times lower than a discrete system with the proposed training algorithm. [ABSTRACT FROM AUTHOR]

A sensing system for simultaneously measuring strain and temperature is proposed by employing a polarimetric multilongitudinal mode fiber laser sensor and beat frequency demodulation device. The proposed fiber laser is built with two fiber Bragg gratings and a piece of Erbium-doped-fiber, which act as cavity reflector and gain medium, respectively. Owing to the intrinsic birefringence of fiber, two orthogonal polarization modes experience slight splitting, thus many longitudinal mode beat signals and polarization mode beat signals are generated in the laser cavity. Strain and temperature can be measured simultaneously by monitoring the frequency shift of longitudinal mode beat signal and polarization mode beat signal due to their different responses to strain and temperature. The selection of beat frequency signal pair for improving the laser sensitivity has also been analyzed and investigated experimentally. In addition, the repeatability and stability performances of the proposed laser were also investigated. The proposed sensing system just uses one polarimetric fiber laser sensor and one beat frequency demodulation equipment, and it shows distinctive advantages of simple structure, portability, high sensitivity, and low cost due to the avoidance of expensive optical detection device for wavelength monitoring. [ABSTRACT FROM PUBLISHER]

For a class of nonspherical-wrist manipulators, i.e., redundant or nonredundant, we propose unified singularity analysis and computation-effective avoidance methods. First, we construct a unified model to describe this class of manipulators and derive the kinematics equation and its modified form. Second, the Jacobian matrix of an $n$-degree-of-freedom (DOF) manipulator is partitioned into block triangle form, and the singularity conditions are isolated and collected in a $3\times(n-3)$ submatrix. By analyzing the rank degeneracy conditions of the submatrix, singularity configurations are identified. Third, based on the partitioned Jacobian matrix, the kinematics equation is decomposed into two smaller dimension subequations, only one of which (called singular subequation, as determined by the $3\times(n-3)$ submatrix) contains singularities. Finally, reduced-order approaches and the singularity parameter optimization (SPO) method are presented to plan the singularity-free trajectories by handling the singular subequation. Theoretical analysis shows that the computation costs of the reduced-order approaches are only 1/3–1/2 of the traditional methods. The SPO method is even more efficient than the reduced-order methods. Simulation results verify the proposed methods. [ABSTRACT FROM AUTHOR]

In wearable and implantable medical-sensor applications, low-energy classification systems are of importance for deriving high-quality inferences locally within the device. Given that sensor instrumentation is typically followed by A-D conversion, this paper presents a system implementation wherein the majority of the computations required for classification are implemented within the ADC. To achieve this, first an algorithmic formulation is presented that combines linear feature extraction and classification into a single matrix transformation. Second, a matrix-multiplying ADC (MMADC) is presented that enables multiplication between an analog input sample and a digital multiplier, with negligible additional energy beyond that required for A-D conversion. Two systems mapped to the MMADC are demonstrated: 13\times and 29\times, respectively, while achieving similar level of performance. [ABSTRACT FROM PUBLISHER]

A low-cost and real-time frequency division multiplexing system is proposed based on multi-longitudinal mode fiber laser sensor array and beat signal demodulation technology. The lasers have staggered cavity lengths with corresponding mode spacing. Beat signals of the longitudinal modes of each laser in the array are detected by a common detector. A relationship is worked out between the maximum multiplexing number of the sensors and maximum detectable frequency. A multiplexing array with four laser sensors is investigated experimentally and its potential application for strain measurement is also demonstrated. In the proposed multiplexing system, only one common photodetector and a commercial frequency analyzer are used to monitor all sensors simultaneously, which greatly decreases the cost of the system and makes it more competitive in large-scale monitoring field. [ABSTRACT FROM PUBLISHER]

A multi-sensor addressing method for fiber Bragg grating aided fiber loop ringdown (FBG-FLRD) sensor array is proposed. It is capable of simultaneous measurement of temperature and force. Light from a wavelength-swept fiber laser (WSFL) is modulated into pulsed light to illuminate FBG-FLRD sensor array. Based on the time sequence of pulsed lights reflected by fiber Bragg grating array, each FBG-FLRD sensor can be distinguished. The time interval between the trigger signal of the WSFL and the pulsed light reflected by FBG offers temperature variation information. By measuring the ringdown time of each FBG-FLRD, the loss of fiber ring induced by temperature and force can be obtained. And then simultaneous measurement can be implemented in real time. To prove the validity of the proposed system, a six elements array is experimentally validated with interrogation frequency of 27.5 Hz. [ABSTRACT FROM PUBLISHER]