The ability to estimate the internal weather conditions of shipping containers globally without using sensors will enable new monitoring and risk assessment solutions. We tackle this problem by developing both linear and nonlinear regression models to predict the internal temperatures and relative humidity of containers from meteorological data. Our training data consists of sensor measurements from 85 shipments across 3 continents over 4 months - the largest data ever collected in an academic publication. We extract features from 761 gigabytes of weather data and incorporate physics (psychrometry) models to engineer new features that boost the performance of our models. Our best models make temperature and relative humidity predictions with mean absolute errors of $1.8 ^{\circ}\mathrm{C}$ and 5.0% - measurements well within the uncertainty range of the sensors - and thereby demonstrate accurate sensorless monitoring.

The advent of Industry 4.0 has led a wide variety of engineering fields to incorporate more automation into their existing work processes. Various engineering sectors intend to imbibe aspects of Industry 4.0 technologies by leveraging Internet of Things coupled with Machine Learning and Artificial Intelligence for process optimization. This, in turn, has led to the surge of cross-domain data integration strategies which when enriched with domain specific knowledge creates dynamic models, termed as Digital Shadows. In this paper, we present the adaptation of the Digital Shadow modeling approach to die casting processes. We propose a generic pipeline for the creation of the model and test the efficacy of such an approach by transforming a predictive analytics model into a digital shadow model. For the predictive modeling, we present a novel approach of image based pixel classification which accurately predicts the occurrence as well as the location of damages on the cast object surfaces.

S.241-248

We propose a quantizer design method for practical half-duplex estimate-and-forward (EF) relaying. First, we identify the regime in which EF relaying yields substantial gains where the SNR is low and the relay-destination link is strong. Then we discover design simplifications that reduce complexity with little loss in the above regime. For relay quantizer design, we first consider mean-squared distortion minimization. To illustrate the unsuitability of the approach, we present an example with AWGN links and a BPSK source where the quantizer with worst mean squared distortion in a given set maximizes achievable rate. A distortion-minimizing quantizer attempts to preserve the received signal at the relay. The quantizer should instead preserve source information. In information theoretical terms, the quantizer should maximize the mutual information between the source transmission and the quantizer output conditioned on the side information at the destination subject to a rate constraint. The above conclusion, derived from information theoretical principles, is then translated to a quantizer design method for the low SNR regime. Using LDPC codes of blocklength 100000, BPSK modulation, and quantizers designed using the proposed criterion, we observe performance less than a decibel away from the achievable rate at a BER of 10-4.

This paper studies the different uses of repeaters and remote radioheads (fiber repeaters) in cellular networks. Traditional uses include coverage extension and eliminating points of pilot pollution (three-way handoff). In addition, we find that careful placement of repeaters can improve capacity substantially. We also find that in layouts where some sectors are much more heavily loaded than others, remote radioheads can be used to transfer load from heavily loaded to lightly loaded sectors, thereby improving network capacity. Therefore, repeaters can enhance network performance without changing either the software or the hardware of existing layouts.

We propose Low Density Parity Check (LDPC) code designs for the half-duplex relay channel. Our designs are based on the information theoretic random coding scheme for decode-and-forward relaying. The source transmission is decoded with the help of side information in the form of additional parity bits from the relay. We derive the exact relationships that the component LDPC code profiles in the relay coding scheme must satisfy. These relationships act as constraints for the density evolution algorithm which is used to search for good relay code profiles. To speed up optimization, we outline a Gaussian approximation of density evolution for the relay channel. The asymptotic noise thresholds of the discovered relay code profiles are a fraction of a decibel away from the achievable lower bound for decode-and-forward relaying. With random component LDPC codes, the overall relay coding scheme performs within 1.2 dB of the theoretical limit.

We propose a practical coding scheme for half-duplex estimate-and-forward relaying. The proposed construction is guided by the information theoretic coding scheme for the estimate-and-forward relay protocol. Our construction incorporates several design features to reduce receiver complexity without compromising performance. Observing that the relaying gain is significant only at low SNRs, we use binary LDPC codes in the source broadcast phase of relaying. Estimation is performed by entropy constrained scalar quantization of the received signal at the relay. Finally, a procedure similar to maximal ratio combining is used to aggregate direct and relayed signals at the destination. An important practical advantage of our scheme is that it does not require source-relay symbol synchronization. The codes outperform direct and two-hop channel capacities, as well as decode-and-forward relaying when the relay-destination link is strong

We present a procedure for communication power optimization in a network of randomly distributed sensors with an observer (data collector) moving on a fixed path. The key challenge in using a mobile observer is that it remains within communication range of any sensor for a brief duration, and inability to transfer data in this duration leads to data loss. We establish that the process of data collection can be modeled by a queue with deadlines, where arrivals correspond to the observer entering the range of a sensor and a missed deadline means data loss. The queuing model is then used to identify the combination of system parameters that ensures adequate data collection with minimum power. The results obtained from the queuing analogy take a simple form in the asymptotic regime of dense sensor networks. Additionally, for sensor networks that cannot tolerate data loss, we derive a tight bound on minimum sensor separation that ensures that no data will be lost on account of mobility. We present two examples to illustrate our results, from which it is seen that power reduction by two orders of magnitude or more is typical relative to a static sensor network. The scenarios chosen for power comparisons also provide guidelines on the choice of path, if such a choice is available.

The main goal of this paper is to show that multi-hop single-user communication achieves the per node transport capacity of Θ(ln N N) in homogeneous sensor networks, making it order-optimal. Our contributions in this paper are three-fold. First, we construct a route-discovery and scheduling scheme based on spatial TDMA for sensor networks. Second, we show that our schedule achieves a per node transport capacity of Θ(lnN N), the same as that achievable by beamforming. Third, we compare multi-hop communication and beamforming based methods in terms of the network power consumption required to attain a fixed throughput. Based on our power calculations, we conclude that if the channel attenuation is above a certain threshold (which we calculate), then multi-hop communication performs better, whereas below the threshold, beamforming is preferable.