Honeywell Technology Solutions Lab, India / Weather is a major contributing factor in aviation accidents, incidents and delays. Doppler weather radar has emerged as a potent tool to observe weather. Aircraft carry onboard radars but their range and angular resolution are limited. Networks of ground-based weather radars provide extensive coverage of weather over large geographic regions. It would be helpful if these data can be transmitted to the pilot. However, these data are highly voluminous and the bandwidth of the ground-air communication links is limited and expensive. Hence, these data have to be compressed to an extent where they are suitable for transmission over low-bandwidth links. Several methods have been developed to compress pictorial data. General-purpose schemes do not take into account the nature of data and hence do not yield high compression ratios. A scheme for extreme compression of weather radar data is developed in this thesis that does not significantly degrade the meteorological information contained in these data. The method is based on contour encoding. It approximates a contour by a set of systematically chosen ‘control points’ that preserve its fine structure up to a certain level. The contours may be obtained using a thresholding process based on NWS or custom reflectivity levels. This process may result in region and hole contours, enclosing `high' or `low' areas, which may be nested. A tag bit is used to label region and hole contours. The control point extraction method first obtains a smoothed reference contour by averaging the original contour. Then the points on the original contour with maximum deviation from the smoothed contour between the crossings of these contours are identified and are designated as control points. Additional control points are added midway between the control point and the crossing points on either side of it, if the length of the segment between the crossing points exceeds a certain length. The control points, referenced with respect to the top-left corner of each contour for compact quantification, are transmitted to the receiving end. The contour is retrieved from the control points at the receiving end using spline interpolation. The region and hole contours are identified using the tag bit. The pixels between the region and hole contours at a given threshold level are filled using the color corresponding to it. This method is repeated till all the contours for a given threshold level are exhausted, and the process is carried out for all other thresholds, thereby resulting in a composite picture of the reconstructed field. Extensive studies have been conducted by using metrics such as compression ratio, fidelity of reconstruction and visual perception. In particular the effect of the smoothing factor, the choice of the degree of spline interpolation and the choice of thresholds are studied. It has been shown that a smoothing percentage of about 10% is optimal for most data. A degree 2 of spline interpolation is found to be best suited for smooth contour reconstruction. Augmenting NWS thresholds has resulted in improved visual perception, but at the expense of a decrease in the compression ratio. Two enhancements to the basic method that include adjustments to the control points to achieve better reconstruction and bit manipulations on the control points to obtain higher compression are proposed. The spline interpolation inherently tends to move the reconstructed contour away from the control points. This has been somewhat compensated by stretching the control points away from the smoothed reference contour. The amount and direction of stretch are optimized with respect to actual data fields to yield better reconstruction. In the bit manipulation study, the effects of discarding the least significant bits of the control point addresses are analyzed in detail. Simple bit truncation introduces a bias in the contour description and reconstruction, which is removed to a great extent by employing a bias compensation mechanism. The results obtained are compared with other methods devised for encoding weather radar contours.

Cellular network traffic prediction is a critical challenge for communication providers, which is important for use cases such as traffic steering and base station resources management. Traditional prediction methods mostly rely on historical time-series data to predict traffic load, which often fail to model the real world and capture surrounding environment conditions. In this work, we propose a multi-modal deep learning model for 4G/5G Cellular Network Traffic prediction by considering external data sources such as satellite imagery and weather data. Specifically, our proposed model consists of three components (1) temporal component (modeling correlations between traffic load values with historical data points via LSTM) (2) computer vision component (using embeddings to capture correlations between geographic regions that share similar landscape patterns using satellite imagery data and state of the art CNN models), and (3) weather component (modeling correlations between weather measurements and traffic patterns). Furthermore, we study the effects and limitations of using such contextual datasets on time series learning process. Our experiments show that such hybrid models do not always lead to better performance, and LSTM model is capable of modeling complex sequential interactions. However, there is a potential for classifying or labelling regions by their urban landscape and the network traffic. / Förutsägelse av mobilnätstrafik är en kritisk utmaning för kommunikation leverantörer, där användningsområden inkluderar trafikstyrning och hantering av basstationsresurser. Traditionella förutsägelsesmetoder förlitar sig främst på historisk tidsseriedata för att förutsäga trafikbelastning, detta misslyckas ofta med att modellera den verkliga världen och fånga omgivande miljö. Det här arbetet föreslår en multimodal modell med djupinlärning förutsägelse av 4G/5G nätverkstrafik genom att beakta externa datakällor som satellitbilder och väderdata. Specifikt består vår föreslagna modell av tre komponenter (1) temporal komponent (korrelationsmodellering mellan trafikbelastningsvärden med historiska datapunkter via LSTM) (2) datorseende komponent (med inbäddningar för att fånga korrelationer mellan geografiska regioner som delar liknande landskapsmönster med hjälp av satelitbilddata och state-of-the-art CNN modeller), och (3) väderkomponent (modellerande korrelationer mellan vädermätningar och trafikmönster). Dessutom studerar vi effekterna och begränsningarna av att använda sådana kontextuella datamängder på tidsserieinlärningsprocessen. Våra experiment visar att hybridmodeller inte alltid leder till bättre prestanda och att LSTM-modellen är kapabel att modellera komplexa sekventiella interaktioner. Det finns dock en potential att klassificera eller märka regioner efter deras stadslandskap och nättrafiken. / La prévision du trafic sur les réseaux cellulaires est un défi crucial pour les fournisseurs de communication, ce qui est important pour les cas d’utilisation tels que la direction du trafic et la gestion des ressources des stations de base. Les méthodes de prédiction traditionnelles reposent principalement sur des données historiques de séries chronologiques pour prédire la charge de trafic, qui échouent souvent à modéliser le monde réel et à capturer les conditions de l’environnement environnant. Dans ce travail, nous proposons un modèle d’apprentissage profond multimodal pour la prédiction du trafic des réseaux cellulaires 4G/5G en considérant des sources de données externes telles que l’imagerie satellitaire et les données météorologiques. Plus précisément, notre modèle proposé se compose de trois composants (1) composant temporel (modélisation des corrélations entre les valeurs de charge de trafic avec des points de données historiques via LSTM) (2) composant de vision par ordinateur (utilisant des incorporations pour capturer les corrélations entre les régions géographiques qui partagent des modèles de paysage similaires à l’aide de données d’imagerie satellitaire et de modèles CNN de pointe) et (3) composante météorologique (modélisation des corrélations entre les mesures météorologiques et les modèles de trafic). De plus, nous étudions les effets et les limites de l’utilisation de tels ensembles de données contextuelles sur le processus d’apprentissage des séries chronologiques. Nos expériences montrent que de tels modèles hybrides ne conduisent pas toujours à de meilleures performances, et le modèle LSTM est capable de modéliser des interactions séquentielles complexes. Cependant, il est possible de classer ou d’étiqueter les régions en fonction de leur paysage urbain et du trafic du réseau.