Communications Technology Laboratory

Energy efficiency is becoming increasingly important for small form factor mobile devices, as battery technology has not kept up with the growing requirements stemming from ubiquitous multimedia applications. This paper addresses link adaptive transmission for maximizing energy efficiency, as measured by the "throughput per Joule" metric. In contrast to the existing water-filling power allocation schemes that maximize throughput subject to a fixed overall transmit power constraint, our scheme maximizes energy efficiency by adapting both overall transmit power and its allocation, according to the channel states and the circuit power consumed. We demonstrate the existence of a unique globally optimal link adaptation solution and develop iterative algorithms to obtain it. We further consider the special case of flat-fading channels to develop an upper bound on energy efficiency and to characterize its variation with bandwidth, channel gain and circuit power. Our results for OFDM systems demonstrate improved energy savings with energy optimal link adaptation as well as illustrate the fundamental tradeoff between energy-efficient and spectrum-efficient transmission.

The vision of Industry 4.0, otherwise known as the fourth industrial revolution, is the integration of massively deployed smart computing and network technologies in industrial production and manufacturing settings for the purposes of automation, reliability, and control, implicating the development of an Industrial Internet of Things (I-IoT). Specifically, I-IoT is devoted to adopting the Internet of Things (IoT) to enable the interconnection of anything, anywhere, and at anytime in the manufacturing system context to improve the productivity, efficiency, safety and intelligence. As an emerging technology, I-IoT has distinct properties and requirements that distinguish it from consumer IoT, including the unique types of smart devices incorporated, network technologies and quality of service requirements, and strict needs of command and control. To more clearly understand the complexities of I-IoT and its distinct needs, and to present a unified assessment of the technology from a systems perspective, in this paper we comprehensively survey the body of existing research on I-IoT. Particularly, we first present the I-IoT architecture, I-IoT applications (i.e., factory automation (FA) and process automation (PA)) and their characteristics. We then consider existing research efforts from the three key systems aspects of control, networking and computing. Regarding control, we first categorize industrial control systems and then present recent and relevant research efforts. Next, considering networking, we propose a three-dimensional framework to explore the existing research space, and investigate the adoption of some representative networking technologies, including 5G, machine-to-machine (M2M) communication, and software defined networking (SDN). Similarly, concerning computing, we again propose a second three-dimensional framework that explores the problem space of computing in I-IoT, and investigate the cloud, edge, and hybrid cloud and edge computing platforms. Finally, we outline particular challenges and future research needs in control, networking, and computing systems, as well as for the adoption of machine learning, in an I-IoT context.

There is evidence that in addition to standard digital filter forms such as the direct, parallel, and cascade forms, digital lattice and ladder filters may play an important role in finite word length implementation problems. In this paper, techniques are developed in detail for efficiently synthesizing digital lattice and ladder filters from any stable direct form. In one form, a lattice filter canonic in terms of multiplies and delays is obtained. An internal scaling procedure is also introduced that will be of importance for optimizing one of the lattice forms for finite word length implementation.

A solution of the least-squares two-dimensional phase-unwrapping problem is presented that is simpler to understand and implement than previously published solutions. It extends the phase function to a periodic function using a mirror reflection, and the resulting equation is solved using the Fourier transform.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

In recent years there has been significant and increasing interest in ad hoc wireless networks. The design, analysis and deployment of such wireless networks necessitate a fundamental understanding of how much information transfer they can support, as well as what the appropriate architectures and protocols are for operating them. This monograph addresses these questions by presenting various models and results that quantify the information transport capability of wireless networks, as well as shed light on architecture design from a high level point of view. The models take into consideration important features such as the spatial distribution of nodes, strategies for sharing the wireless medium, the attenuation of signals with distance, and how information is to be transferred, whether it be by encoding, decoding, choice of power level, spatio-temporal scheduling of transmissions, choice of multi-hop routes, or other modalities of cooperation between nodes. An important aspect of the approach is to characterize how the information hauling capacity scales with the number of nodes in the network. The monograph begins by studying models of wireless networks based on current technology, which schedules concurrent transmissions to take account of interference, and then routes packets from their sources to destinations in a multi-hop fashion. An index of performance, called transport capacity, which is measured by the bit meters per second the network can convey in aggregate, is studied. For arbitrary networks, including those allowing for optimization of node locations, the scaling law for the transport capacity in terms of the number of nodes in the network is identified. For random networks, where nodes are randomly distributed, and source-destination pairs are randomly chosen, the scaling law for the maximum common throughput capacity that can be supported for all the source-destination pairs is characterized. The constructive procedure for obtaining the sharp lower bound gives insight into an order optimal architecture for wireless networks operating under a multi-hop strategy. To determine the ultimate limits on how much information wireless networks can carry requires an information theoretic treatment, and this is the subject of the second half of the monograph. Since wireless communication takes place over a shared medium, it allows more advanced operations in addition to multi-hop. To understand the limitations as well as possibilities for such information transfer, wireless networks are studied from a Shannon information-theoretic point of view, allowing any causal operation. Models that characterize how signals attenuate with distance, as well as multi-path fading, are introduced. Fundamental bounds on the transport capacity are established for both high and low attenuation regimes. The results show that the multi-hop transport scheme achieves the same order of scaling, though with a different pre-constant, as the information theoretically best possible, in the high attenuation regime. However, in the low attenuation regime, superlinear scaling may be possible through recourse to more advanced modes of cooperation between nodes. Techniques used in analyzing multi-antenna systems are also studied to characterize the scaling behavior of large wireless networks.

As a typical application of the Internet of Things (IoT), the Industrial IoT (IIoT) connects all the related IoT sensing and actuating devices ubiquitously so that the monitoring and control of numerous industrial systems can be realized. Deep learning, as one viable way to carry out big-data-driven modeling and analysis, could be integrated in IIoT systems to aid the automation and intelligence of IIoT systems. As deep learning requires large computation power, it is commonly deployed in cloud servers. Thus, the data collected by IoT devices must be transmitted to the cloud for training process, contributing to network congestion and affecting the IoT network performance as well as the supported applications. To address this issue, in this article, we leverage the fog/edge computing paradigm and propose an edge computing-based deep learning model, which utilizes edge computing to migrate the deep learning process from cloud servers to edge nodes, reducing data transmission demands in the IIoT network and mitigating network congestion. Since edge nodes have limited computation ability compared to servers, we design a mechanism to optimize the deep learning model so that its requirements for computational power can be reduced. To evaluate our proposed solution, we design a testbed implemented in the Google cloud and deploy the proposed convolutional neural network (CNN) model, utilizing a real-world IIoT data set to evaluate our approach. <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> Our experimental results confirm the effectiveness of our approach, which cannot only reduce the network traffic overhead for IIoT but also maintain the classification accuracy in comparison with several baseline schemes. <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> Certain commercial equipment, instruments, or materials are identified in this article in order to specify the experimental procedure adequately. Such identification is not intended to imply recommendation or endorsement by the National Institute of Standards and Technology, nor is it intended to imply that the materials or equipment identified are necessarily the best available for the purpose.

Speech and music are highly complex signals that have many shared acoustic features. Pitch, Timbre, and Timing can be used as overarching perceptual categories for describing these shared properties. The acoustic cues contributing to these percepts also have distinct subcortical representations which can be selectively enhanced or degraded in different populations. Musically trained subjects are found to have enhanced subcortical representations of pitch, timbre, and timing. The effects of musical experience on subcortical auditory processing are pervasive and extend beyond music to the domains of language and emotion. The sensory malleability of the neural encoding of pitch, timbre, and timing can be affected by lifelong experience and short-term training. This conceptual framework and supporting data can be applied to consider sensory learning of speech and music through a hearing aid or cochlear implant.

Multiple-input multiple-output (MIMO) wireless systems obtain large diversity and capacity gains by employing multielement antenna arrays at both the transmitter and receiver. The theoretical performance benefits of MIMO systems, however, are irrelevant unless low error rate, spectrally efficient signaling techniques are found. This paper proposes a new method for designing high data-rate spatial signals with low error rates. The basic idea is to use transmitter channel information to adaptively vary the transmission scheme for a fixed data rate. This adaptation is done by varying the number of substreams and the rate of each substream in a precoded spatial multiplexing system. We show that these substreams can be designed to obtain full diversity and full rate gain using feedback from the receiver to transmitter. We model the feedback using a limited feedback scenario where only finite sets, or codebooks, of possible precoding configurations are known to both the transmitter and receiver. Monte Carlo simulations show substantial performance gains over beamforming and spatial multiplexing.

The Video Quality Experts Group (VQEG) was formed in October 1997 to address video quality issues. The group is composed of experts from various backgrounds and affiliations, including participants from several internationally recognized organizations working int he field of video quality assessment. The first task undertaken by VQEG was to provide a validation of objective video quality measurement methods leading to recommendations in both the telecommunications and radiocommunication sectors of the International Telecommunications Union. To this end, VQEG designed and executed a test program to compare subjective video quality evaluations to the predictions of a number of proposed objective measurement methods for video quality in the bit rate range of 768 kb/s to 50 Mb/s. The results of this test show that there is no objective measurement system that is currently able to replace subjective testing. Depending on the metric used for evaluation, the performance of eight or nine models was found to be statistically equivalent, leading to the conclusion that no single model outperforms the others in all cases. The greatest achievement of this first validation effort is the unique data set assembled to help future development of objective models.

This study examined the effect of primary stimulus level on the ability of distortion product otoacoustic emission (DPOAE) measurements to separate normal-hearing from hearing-impaired ears. Complete I/O functions were obtained for nine f2 frequencies on 210 people approximately evenly divided between normal hearing and hearing impaired. Clinical decision theory was used to assess both DPOAE amplitudes and DPOAE threshold as diagnostic indicators of hearing status. Moderate level primary stimuli elicited responses that separated normal from impaired better than either lower level or higher level stimuli. The two populations were differentiated for all frequencies above 500 Hz by DPOAE amplitude, given primary levels, L1 and L2, of 65 and 55 dB SPL. DPOAE threshold performed equally well, but threshold ambiguity in noise and longer testing times make it a less suitable DPOAE measure to use diagnostically.

Standing waves can cause errors during in-the-ear calibration of sound pressure level (SPL), affecting both stimulus magnitude and distortion-product otoacoustic emission (DPOAE) level. Sound intensity level (SIL) and forward pressure level (FPL) are two measurements theoretically unaffected by standing waves. SPL, SIL, and FPL in situ calibrations were compared by determining sensitivity of DPOAE level to probe-insertion depth (deep and "shallow") for a range of stimulus frequencies (1-8 kHz) and levels (20-60 dB). Probe-insertion depth was manipulated with the intent to shift the frequencies with standing-wave minima at the emission probe, introducing variability during SPL calibration. The absolute difference in DPOAE level between insertions was evaluated after correcting for an incidental change caused by the effect of ear-canal impedance on the emission traveling from the cochlea. A three-way analysis of variance found significant main effects for stimulus level, stimulus frequency, and calibration method, as well as significant interactions involving calibration method. All calibration methods exhibited changes in DPOAE level due to the insertion depth, especially above 4 kHz. However, SPL demonstrated the greatest changes across all stimulus levels for frequencies above 2 kHz, suggesting that SIL and FPL provide more consistent measurements of DPOAEs for frequencies susceptible to standing-wave calibration errors.

Recently, Boege and Janssen [J. Acoust. Soc. Am. 111, 1810-1818 (2002)] fit linear equations to distortion product otoacoustic emission (DPOAE) input/output (UO) functions after the DPOAE level (in dB SPL) was converted into pressure (in microPa). Significant correlations were observed between these DPOAE thresholds and audiometric thresholds. The present study extends their work by (1) evaluating the effect of frequency, (2) determining the behavioral thresholds in those conditions that did not meet inclusion criteria, and (3) including a wider range of stimulus levels. DPOAE I/O functions were measured in as many as 278 ears of subjects with normal and impaired hearing. Nine f2 frequencies (500 to 8000 Hz in 1/2-octave steps) were used, L2 ranged from 10 to 85 dB SPL (5-dB steps), and L1 was set according to the equation L1 = 0.4L2 + 39 dB [Kummer et al., J. Acoust. Soc. Am. 103, 3431-3444 (1998)] for L2 levels up to 65 dB SPL, beyond which L1 = L2. For the same conditions as those used by Boege and Janssen, we observed a frequency effect such that correlations were higher for mid-frequency threshold comparisons. In addition, a larger proportion of conditions not meeting inclusion criteria at mid and high frequencies had hearing losses exceeding 30 dB HL, compared to lower frequencies. These results suggest that DPOAE I/O functions can be used to predict audiometric thresholds with greater accuracy at mid and high frequencies, but only when certain inclusion criteria are met. When the SNR inclusion criterion is not met, the expected amount of hearing loss increases. Increasing the range of input levels from 20-65 dB SPL to 10-85 dB SPL increased the number of functions meeting inclusion criteria and increased the overall correlation between DPOAE and behavioral thresholds.

Designing mobile edge computing (MEC) systems by jointly optimizing communication and computation resources, which can help increase mobile batteries' lifetime and improve quality of experience for computation-intensive and latency-sensitive applications, has received significant interest. In this paper, we consider energy-efficient resource allocation schemes for a multi-user mobile edge computing system with inelastic computation tasks and non-negligible task execution durations. First, we establish a mathematical model to characterize the offloading of a computation task from a mobile to the base station (BS) equipped with MEC servers. This computation-offloading model consists of three stages, i.e., task uploading, task executing, and computation result downloading, and allows parallel transmissions and executions for different tasks. Then, we formulate the weighted sum energy consumption minimization problem to optimally allocate the task operation sequence, the uploading and downloading time durations as well as the starting times for uploading, executing and downloading, which is a challenging mixed discrete- continuous optimization problem and is NP-hard in general. We propose a method to obtain an optimal solution and develop a low-complexity algorithm to obtain a suboptimal solution, by connecting the optimization problem to a three-stage flow-shop scheduling problem and utilizing Johnson's algorithm as well as convex optimization techniques. Finally, numerical results show that the proposed sub-optimal solution outperforms existing comparison schemes.

The amount of information transmissible through a communications channel is determined by the noise characteristics of the channel and by the quantities of available transmission resources. In classical information theory, the amount of transmissible information can be increased twice at most when the transmission resource is doubled for fixed noise characteristics. In quantum information theory, however, the amount of information transmitted can increase even more than twice. We present a proof-of-principle demonstration of this superadditivity of classical capacity of a quantum channel by using the ternary symmetric states of a single photon, and by event selection from a weak coherent light source. We also show how the superadditive coding gain, even in a small code length, can boost the communication performance of the conventional coding technique.

Singlet oxygen production efficiency of several laser-excited fullerene derivatives is compared by measuring the 1268 nm emissions; the efficiency was independent of the kind of addends, but it decreased with an increase in the number of addends, and addends at adjacent positions caused a greater decrease than remote addends.

The paper measures the ability of face recognition algorithms to distinguish between identical twin siblings. The experimental dataset consists of images taken of 126 pairs of identical twins (252 people) collected on the same day and 24 pairs of identical twins (48 people) with images collected one year apart. In terms of both the number of paris of twins and lapsed time between acquisitions, this is the most extensive investigation of face recognition performance on twins to date. Recognition experiments are conducted using three of the top submissions to the Multiple Biometric Evaluation (MBE) 2010 Still Face Track [1]. Performance results are reported for both same day and cross year matching. Performance results are broken out by lighting conditions (studio and outside); expression (neutral and smiling); gender and age. Confidence intervals were generated by a bootstrap method. This is the most detailed covariate analysis of face recognition of twins to date.

In the United States, the Federal Communications Commission has adopted rules permitting commercial wireless networks to share spectrum with federal incumbents in the 3.5 GHz Citizens Broadband Radio Service band. These rules require commercial systems to vacate the band when sensors detect radars operated by the U.S. military; a key example being the SPN-43 air traffic control radar. Such sensors require highly-accurate detection algorithms to meet their operating requirements. In this paper, using a library of over 14,000 3.5 GHz band spectrograms collected by a recent measurement campaign, we investigate the performance of thirteen methods for SPN-43 radar detection. Namely, we compare classical methods from signal detection theory and machine learning to several deep learning architectures. We demonstrate that machine learning algorithms appreciably outperform classical signal detection methods. Specifically, we find that a three-layer convolutional neural network offers a superior tradeoff between accuracy and computational complexity. Last, we apply this three-layer network to generate descriptive statistics for the full 3.5 GHz spectrogram library. Our findings highlight potential weaknesses of classical methods and strengths of modern machine learning algorithms for radar detection in the 3.5 GHz band.

In an automatic repeat request (ARQ) scheme, a packet is retransmitted if it gets corrupted due to transmission errors caused by the channel. Here we describe a ready-to-implement ARQ scheme with packet combining. An analytical description of the scheme in random error channel shows excellent agreement with simulation results. An upper bound for type-II schemes is defined. For smaller packet sizes, throughput of the proposed scheme is sufficiently close to the upper bound till a very high bit error rate.

Distortion product otoacoustic emission (DPOAE) input/output functions were measured at nine f2 frequencies ranging from 500 to 8000 Hz in 210 normal-hearing and hearing-impaired subjects. In a companion paper [Stover et al., J. Acoust. Soc. Am. 100, 956-967 (1996)], L1-L2 was held constant at 10 dB, and L2 was varied from 65 to 10 dB SPL in 5-dB steps. Based upon analyses using clinical decision theory, it was demonstrated that DPOAE amplitudes for 65/55 dB SPL primaries (L1/L2) and DPOAE thresholds resulted in the greatest separation between normal and impaired ears. In this paper, the data for these two conditions were recast as cumulative distributions, which not only describe the extent of overlap between normal and impaired distributions, but also provide the measured value (i.e., the specific DPOAE amplitude or threshold) for any combination of hit and false alarm rates. From these distributions, confidence limits were constructed for both DPOAE amplitude and threshold to determine the degree of certainty with which any measured response could be assigned to either the normal or impaired population. For these analyses, DPOAE measurements were divided into three categories (a) response properties that would be unlikely to come from normal ears, (b) response properties that would be unlikely to come from impaired ears, and (c) response properties for which hearing status was uncertain. Based upon DPOAE amplitude measurements, the region of uncertainty, defined between the 95 percentile for impaired ears and the 5 percentile for normal ears, was relatively narrow for f2 frequencies ranging from 707 to 4000 Hz. For DPOAE thresholds, this region was relatively narrow for F2 frequencies ranging from 1414 to 4000 Hz.

A model of cochlear mechanics is described in which force-producing outer hair cells (OHC) are embedded in a passive cochlear partition. The OHC mechanoelectrical transduction current is nonlinearly modulated by reticular-lamina (RL) motion, and the resulting change in OHC membrane voltage produces contraction between the RL and the basilar membrane (BM). Model parameters were chosen to produce a tonotopic map typical of a human cochlea. Time-domain simulations showed compressive BM displacement responses typical of mammalian cochleae. Distortion product (DP) otoacoustic emissions at 2f(1)-f(2) are plotted as isolevel contours against primary levels (L(1),L(2)) for various primary frequencies f(1) and f(2) (f(1)<f(2)). The L(1) at which the DP reaches its maximum level increases as L(2) increases, and the slope of the "optimal" linear path decreases as f(2)/f(1) increases. When primary levels and f(2) are fixed, DP level is band passed against f(1). In the presence of a suppressor, DP level generally decreases as suppressor level increases and as suppressor frequency gets closer to f(2); however, there are exceptions. These results, being similar to data from human ears, suggest that the model could be used for testing hypotheses regarding DP generation and propagation in human cochleae.