Department of Electrical Engineering

Over the last decade, the usage of unmanned systems such as Unmanned Aerial Vehicles (UAVs), Unmanned Surface Vessels (USVs) and Unmanned Ground Vehicles (UGVs) has increased drastically, and there is still a rapid growth. Today, unmanned systems are being deployed in many daily operations, e.g. for deliveries in remote areas, to increase efficiency of agriculture, and for environmental monitoring at sea. For safety reasons, unmanned systems are often the preferred choice for surveillance missions in hazardous environments, e.g. for detection of nuclear radiation, and in disaster areas after earthquakes, hurricanes, or during forest fires. For safe navigation of the unmanned systems during their missions, continuous and accurate global localization and attitude estimation is mandatory.

Over the years, many vision-based methods for position estimation have been developed, primarily for urban areas. In contrast, this thesis is mainly focused on vision-based methods for accurate position and attitude estimates in natural environments, i.e. beyond the urban areas. Vision-based methods possess several characteristics that make them appealing as global position and attitude sensors. First, vision sensors can be realized and tailored for most unmanned vehicle applications. Second, geo-referenced terrain models can be generated worldwide from satellite imagery and can be stored onboard the vehicles. In natural environments, where the availability of geo-referenced images in general is low, registration of image information with terrain models is the natural choice for position and attitude estimation. This is the problem area that I addressed in the contributions of this thesis.

The first contribution is a method for full 6DoF (degrees of freedom) pose estimation from aerial images. A dense local height map is computed using structure from motion. The global pose is inferred from the 3D similarity transform between the local height map and a digital elevation model. Aligning height information is assumed to be more robust to season variations than feature-based matching.

The second contribution is a method for accurate attitude (pitch and roll angle) estimation via horizon detection. It is one of only a few methods that use an omnidirectional (fisheye) camera for horizon detection in aerial images. The method is based on edge detection and a probabilistic Hough voting scheme. The method allows prior knowledge of the attitude angles to be exploited to make the initial attitude estimates more robust. The estimates are then refined through registration with the geometrically expected horizon line from a digital elevation model. To the best of our knowledge, it is the first method where the ray refraction in the atmosphere is taken into account, which enables the highly accurate attitude estimates.

The third contribution is a method for position estimation based on horizon detection in an omnidirectional panoramic image around a surface vessel. Two convolutional neural networks (CNNs) are designed and trained to estimate the camera orientation and to segment the horizon line in the image. The MOSSE correlation filter, normally used in visual object tracking, is adapted to horizon line registration with geometric data from a digital elevation model. Comprehensive field trials conducted in the archipelago demonstrate the GPS-level accuracy of the method, and that the method can be trained on images from one region and then applied to images from a previously unvisited test area.

The CNNs in the third contribution apply the typical scheme of convolutions, activations, and pooling. The fourth contribution focuses on the activations and suggests a new formulation to tune and optimize a piecewise linear activation function during training of CNNs. Improved classification results from experiments when tuning the activation function led to the introduction of a new activation function, the Shifted Exponential Linear Unit (ShELU).

Electronic devices with wireless connectivity are fast becoming a part of daily life. According to some estimates, in the next five years, 10 billion new devices with internet connectivity would be produced. To lower the costs and extend the battery life of electronic circuits, there is an increased interest in using lowcost, low-power CMOS circuits. By taking advantage of the higher integration capabilities of modern CMOS, the analog, digital, and radio circuits can be integrated on a single die, typically called a radio-frequency system-on-chip (RF-SoC).

In an RF-SoC, most of the power is usually consumed by the radio circuits, especially the power amplifier (PA). Hence, to take advantage of the improved switching capability of transistors in modern CMOS, the use of switch-mode PAs (SMPAs) is becoming more popular. SMPAs exhibit a much higher efficiency as compared to their linear counterparts and can be easily integrated with the digital baseband circuits.

To satisfy the demand for higher data throughput, modern wireless standards like LTE and IEEE 802.11 generate envelope-varying signals using advanced modulation schemes like M-QAM and OFDM. Among several other techniques, pulse-width modulation (PWM) allows for the amplification of the envelopevarying signals using SMPAs.

The first part of this thesis explores techniques to improve the spectral performance of PWM-based transmitters. The proposed transmitters are fully digital, and the entire signal chain up to the PA can be implemented using the digital design flow, which is especially beneficial in sub-micron CMOS processes with low voltage headroom. A new transmitter is proposed that compensates for the aliasing distortion in polar PWM transmitters by using outphasing. The transmitter exhibits an improvement of up to 9 dB in dynamic range for a 1.4 MHz LTE uplink signal. The idea is extended to compensate for both image and aliasing distortions in all-digital implementations of polar PWM transmitters. By using a field programmable gate array (FPGA) and Class-D SMPAs, the proposed transmitter shows an improvement of up to 6.9 dBc in the adjacent channel leakage ratio (ACLR) and 10% in the error vector magnitude (EVM) for a 20 MHz LTE uplink signal. The proposed transmitter is fully programmable and can be easily adapted for multi-band and multi-standard transmission.

To enhance the phase linearity of all-digital PWM transmitters, a new transmitter architecture based on outphasing is presented. The proposed transmitter uses outphasing to improve the phase resolution and exhibits an improvement of 2.8 dBc and 3.3% in ACLR and EVM, respectively.

The difference between the polar and quadrature implementations of RFPWM based transmitters is explored. By using mathematical derivations and simulations, it is shown that the polar implementation outperforms the quadrature implementation due to the lower quantization noise. An RF-PWM based transmitter that eliminates both image and aliasing distortions is presented. The proposed transmitter has an all-digital implementation, uses a single SMPA, and eliminates the need for a power combiner resulting in a more compact design. For a 1.4 MHz LTE uplink signal, the proposed transmitter exhibits an improvement of up to 11.3 dBc in ACLR.

The second part of this work focuses on the design of all-digital area-efficient architectures of time-to-digital converters (TDCs). A TDC is essentially a stopwatch with a pico-second resolution and can be used to accurately quantify the pulse width and position of PWM signals.

A Vernier delay line-based TDC is presented that replaces the conventionally used sampling D flip-flops by a single transistor. This resulting implementation does not suffer from blackout time associated with D flip-flops allowing for a more compact design. The proposed TDC achieves a time resolution of 5.7 ps, and consumes 1.85 mW of power while operating at 50 MS/s.

A modified switching scheme to reduce the power consumed by the thermometerto- binary encoder used in the TDCs is presented. By taking advantage of the operating nature of the TDCs, the proposed switching scheme reduces the power consumption by up to 40% for a 256-bit encoder.

Automatic decision making and pattern recognition under uncertainty are difficult tasks that are ubiquitous in our everyday life. The systems we design, and technology we develop, requires us to coherently represent and work with uncertainty in data. Probabilistic models and probabilistic inference gives us a powerful framework for solving this problem. Using this framework, while enticing, results in difficult-to-compute integrals and probabilities when conditioning on the observed data. This means we have a need for approximate inference, methods that solves the problem approximately using a systematic approach. In this thesis we develop new methods for efficient approximate inference in probabilistic models.

There are generally two approaches to approximate inference, variational methods and Monte Carlo methods. In Monte Carlo methods we use a large number of random samples to approximate the integral of interest. With variational methods, on the other hand, we turn the integration problem into that of an optimization problem. We develop algorithms of both types and bridge the gap between them.

First, we present a self-contained tutorial to the popular sequential Monte Carlo (SMC) class of methods. Next, we propose new algorithms and applications based on SMC for approximate inference in probabilistic graphical models. We derive nested sequential Monte Carlo, a new algorithm particularly well suited for inference in a large class of high-dimensional probabilistic models. Then, inspired by similar ideas we derive interacting particle Markov chain Monte Carlo to make use of parallelization to speed up approximate inference for universal probabilistic programming languages. After that, we show how we can make use of the rejection sampling process when generating gamma distributed random variables to speed up variational inference. Finally, we bridge the gap between SMC and variational methods by developing variational sequential Monte Carlo, a new flexible family of variational approximations.

Massive MIMO (Multiple-Input--Multiple-Output) is a cellular-network technology in which the base station is equipped with a large number of antennas and aims to serve several different users simultaneously, on the same frequency resource through spatial multiplexing. This is made possible by employing efficient beamforming, based on channel estimates acquired from uplink reference signals, where the base station can transmit the signals in such a way that they add up constructively at the users and destructively elsewhere. The multiplexing together with the array gain from the beamforming can increase the spectral efficiency over contemporary systems.

One challenge of practical importance is how to transmit data in the downlink when no channel state information is available. When a user initially joins the network, prior to transmitting uplink reference signals that enable beamforming, it needs system information---instructions on how to properly function within the network. It is transmission of system information that is the main focus of this thesis. In particular, the thesis analyzes how the reliability of the transmission of system information depends on the available amount of diversity. It is shown how downlink reference signals, space-time block codes, and power allocation can be used to improve the reliability of this transmission.

In order to estimate the uplink and downlink channels from uplink reference signals, which is imperative to ensure scalability in the number of base station antennas, massive MIMO relies on channel reciprocity. This thesis shows that the principles of channel reciprocity can also be exploited by a jammer, a malicious transmitter, aiming to disrupt legitimate communication between two single-antenna devices. A heuristic scheme is proposed in which the jammer estimates the channel to a target device blindly, without any knowledge of the transmitted legitimate signals, and subsequently beamforms noise towards the target. Under the same power constraint, the proposed jammer can disrupt the legitimate link more effectively than a conventional omnidirectional jammer in many cases.

Visual tracking is one of the fundamental problems in computer vision. Its numerous applications include robotics, autonomous driving, augmented reality and 3D reconstruction. In essence, visual tracking can be described as the problem of estimating the trajectory of a target in a sequence of images. The target can be any image region or object of interest. While humans excel at this task, requiring little effort to perform accurate and robust visual tracking, it has proven difficult to automate. It has therefore remained one of the most active research topics in computer vision.

In its most general form, no prior knowledge about the object of interest or environment is given, except for the initial target location. This general form of tracking is known as generic visual tracking. The unconstrained nature of this problem makes it particularly difficult, yet applicable to a wider range of scenarios. As no prior knowledge is given, the tracker must learn an appearance model of the target on-the-fly. Cast as a machine learning problem, it imposes several major challenges which are addressed in this thesis.

The main purpose of this thesis is the study and advancement of the, so called, Discriminative Correlation Filter (DCF) framework, as it has shown to be particularly suitable for the tracking application. By utilizing properties of the Fourier transform, a correlation filter is discriminatively learned by efficiently minimizing a least-squares objective. The resulting filter is then applied to a new image in order to estimate the target location.

This thesis contributes to the advancement of the DCF methodology in several aspects. The main contribution regards the learning of the appearance model: First, the problem of updating the appearance model with new training samples is covered. Efficient update rules and numerical solvers are investigated for this task. Second, the periodic assumption induced by the circular convolution in DCF is countered by proposing a spatial regularization component. Third, an adaptive model of the training set is proposed to alleviate the impact of corrupted or mislabeled training samples. Fourth, a continuous-space formulation of the DCF is introduced, enabling the fusion of multiresolution features and sub-pixel accurate predictions. Finally, the problems of computational complexity and overfitting are addressed by investigating dimensionality reduction techniques.

As a second contribution, different feature representations for tracking are investigated. A particular focus is put on the analysis of color features, which had been largely overlooked in prior tracking research. This thesis also studies the use of deep features in DCF-based tracking. While many vision problems have greatly benefited from the advent of deep learning, it has proven difficult to harvest the power of such representations for tracking. In this thesis it is shown that both shallow and deep layers contribute positively. Furthermore, the problem of fusing their complementary properties is investigated.

The final major contribution of this thesis regards the prediction of the target scale. In many applications, it is essential to track the scale, or size, of the target since it is strongly related to the relative distance. A thorough analysis of how to integrate scale estimation into the DCF framework is performed. A one-dimensional scale filter is proposed, enabling efficient and accurate scale estimation.