Search

Radar-based gesture recognition can play a vital role in autonomous vehicles’ interaction with vulnerable road users (VRUs). However, in automotive scenarios the same gesture produces strongly differing radar responses owed to the wide range of variations such as position, orientation, or ego-motion. Since including all kinds of modifications in a measured dataset is laborious, gesture simulations alleviate the measurement effort and increase the robustness against edge and corner cases. Hence, this paper presents a flexible geometric human target model allowing the direct introduction of a wide range of modifications, while it facilitates the handling of shadowing effects and multiradar constellations. Using the proposed simulation model, a dataset recorded with a radar sensor network consisting of three chirp sequence (CS) radars is resimulated based on motion data simultaneously captured with a stereo video system. Completely substituting the measured by simulated data for training, a convolutional neural network (CNN) classifier still achieves 80.4% cross-validation accuracy on a challenging gesture set, compared to 89.4% for training on measured data. Moreover, using simulated data the classifier is shown to successfully generalize to new scenarios not observed in measurements., acceptedVersion

Verletzliche Verkehrsteilnehmer sind im Straßenverkehr besonders gefährdet, da sie nicht durch eine Fahrzeugkarosserie geschützt sind. Ein Großteil dieser Gruppe machen hierbei Fußgänger und Fahrradfahrer aus. Die Überwachung des direkten Fahrzeugumfeldes bietet hierbei die Möglichkeit die Anzahl der Unfälle zu reduzieren. Unter anderem können dabei Radarsensoren eingesetzt werden, die anhand der mikro-Doppler-Signatur verletzliche Verkehrsteilnehmer klassifizieren. Zum zeitsparenden Test des Radars kann ein Radarzielsimulator verwendet werden. In dieser Arbeit wird ein Radarzielsimulator vorgestellt, mit dem die charakteristische mikro-Doppler-Signatur von verletzlichen Verkehrsteilnehmern generiert werden kann. Das System arbeitet hierbei bei dem für den Automobilbereich relevanten Frequenzbereich von 77 GHz. Die Ziele können hierbei anhand von Simulationen generiert, oder aus realen Messungen extrahiert werden. Durch die Überlagerung von Einzelzielen lassen sich somit realistische Verkehrsszenarien generieren.

Airborne ground-penetrating synthetic aperture radar (GPSAR) using unmanned aerial vehicles (UAVs) offers advantages for many applications compared to established methods. In particular, these systems are useful for non-traversable terrain, where they could be used for the detection and localization of buried objects and structures. At distances of a few meters to the ground, refraction effects at the surface need to be considered in processing even for shallowly buried objects. Often the ground is assumed to be horizontal, which is not the case in many situations, such as uneven meadows or acres. This work investigates the influence of uneven ground on GPSAR images. It is shown that neglecting tilt in backprojection for shallowly buried targets leads to an apparent target shift. A model is derived to predict the shift for circular measurement trajectories and to estimate, if quality loss is to be expected. The model is verified using simulations, which are verified by measurement. It is concluded that tilt can be neglected in processing for shallowly buried targets. This is confirmed in a realistic measurement scenario., acceptedVersion

The improvement of the angular resolution of radar sensors is one of the crucial goals of current radar research. A promising approach to achieve this goal is inspired by the ears of a fly called Ormia ochracea. The working principle was adapted for antennas, and the so-called biomimetic antenna arrays (BMAAs) aroused the interest of several research groups. In this work, BMAAs are incorporated into multiple-input multiple-output (MIMO) arrays, a very common approach of improving the angular resolution, to gain more degrees of freedom in array design. The MIMO BMAAs are modeled utilizing the effective biomimetic antenna distance, a fundamentally new measure introduced in this article to translate the special biomimetic phase progression into a spatial quantity. We present straightforward antenna configurations but also describe how a genetic algorithm can be utilized to optimize both antenna positions and BMAA parameters. The proposed arrays show various beneficial effects such as a wider angular range for unambiguous angle estimation or a narrower beamwidth. The impact of MIMO BMAAs on the angular resolution is thoroughly analyzed both theoretically and by radar measurements in the range of 77 GHz. The measurements confirm the modeling method very well and show a significant increase in the angular separability., publishedVersion

In dieser Dissertation werden lernende Verfahren zur Gesten- und Kochklassifikation mittels Radarsensorik vorgestellt. Die Verfahren umfassen zum einen die Klassifikation der Radardaten mithilfe verschiedener tiefer neuronaler Netzwerke. Zum anderen wird ein kognitives Radar vorgestellt, bei dem das Radar adaptiv auf die Szene reagieren und die Messparameter entsprechend anpassen kann. Die Verfahren werden anhand der Anwendung Kochsensor evaluiert. Der Kochsensor ist ein 120GHz FMCW-Radar, das oberhalb eines Kochfelds angebracht ist und die aktive Kochposition, verschiedene Steuergesten und das Kochen selbst klassifiziert. Auf diese Weise wird ein Überkochschutz realisiert. Die in dieser Arbeit erforschten Verfahren zeigen verschiedene Klassifikationsmöglichkeiten anhand von Range-Doppler-Matrix-Daten auf. Zunächst erfolgt die Klassifikation mit einer einzelnen Messung (Einzelbild-Klassifikation). Dieses Verfahren zeigt speziell bei dynamischen Klassen, wie den Gesten oder dem Kochen, Schwächen. Deutlich bessere Ergebnisse zeigt die Sequenz-Klassifikation, bei der eine Sequenz von Messungen mithilfe netzwerkinterner Speicher ausgewertet wird. Ein ähnlich gutes Ergebnis wird erreicht, wenn bei der Einzelbild-Klassifikation zusätzliche Sensormodalitäten zum Trainieren des Klassifikationsnetzwerks herangezogen werden. Bei diesem sogenannten Cross Learning werden die zusätzlichen Sensoren, wie eine Wärmebildkamera, allerdings nur zum Trainieren und nicht zur Klassifikation in der Anwendung benötigt. Ein weiteres lernendes Verfahren ist das Reinforcement Learning, das dem Radar ein kognitives Verhalten erlaubt. Das Radar kann adaptiv auf die Szene reagieren und die adaptiven Messparameter optimal wählen.

Modern vehicles are equipped with many semi-automated and even highly automated driving functions, which enormously increase safety and comfort while traveling. A basic prerequisite for these features is an accurate and reliable perception of the vehicle’s environment, which is modeled by suitable algorithms on the basis of various sensor measurements. One appropriate sensor for this purpose is a radar sensor, which is able to directly measure the position, the orientation as well as the radial velocity of objects in the vehicle’s surroundings. Contemporary automotive radar sensors provide high-resolution data, such that multiple radar reflections are generated for a single object. Subsequently, the object detection task is to recognize all existing objects in the measured radar target list which means to determine their localization and class. This thesis proposes a novel technique for object detection in high-resolution radar data, which is then further used to realize environment perception at an object level. In summary, this covers the complete processing pipeline from a radar target list to the object recognition and multi-object tracking. The radar object detection system designed in this thesis is based on a Deep Learning technique, where the entire processing chain basically consisting of three main components. The first module divides the radar target list into several small regions and determines whether there is an object located in it, and if given, then its associated class. The second component segments all radar targets belonging to the actual object of interest. Finally, the third module determines the parameters for a 2D bounding box modeling the respective object and outputs all predicted hypotheses after the object extraction. The training of all neural networks exclusively utilizes real radar data from typical traffic situations in rural and urban areas. The evaluation validates the radar object detector based on three different datasets, and scores the intermediate results of each module as well as the final object detection result. Two of the radar datasets are publicly available ensuring that all results are reproducible and verifiable. The third dataset is self-generated and includes high-resolution radar data from real sensors. The evaluation reveals that the radar object detector provides convincing results on all datasets, where the best performance is obtained with the self-generated dataset. Further, a successful implementation and a parallel operation of three radar object detection systems on three radar sensors installed in a real system for autonomous driving demonstrates that this technology is real-time capable and appropriate for real-world applications in the automotive field. In the case of multiple sensors with an overlapping field of view, the respective radar object detectors provide different hypotheses for the same object, which may be contradictory. In order to obtain a consistent environment model, it is necessary to fuse and track all generated object detections over time. A multi-object tracking algorithm allows to combine detections from independent sources into one overall result. A tracking algorithm consists of two main stages: the prediction of existing objects according to a motion model and the correction of all object states based on measurements. This thesis presents a measurement model to process all hypotheses provided by the developed radar object detection system using an established multi-object tracking algorithm. The tracker processes all radar object detections and generates a robust environment model. Therefore, the algorithm optimizes the perception of objects such that the resulting tracks correspond accurately to reality. The proposed radar object detector and the subsequent multi-object tracking perform precisely and reliably in real-world applications, which makes this advanced technique an essential component for the environment perception in a real system for automated driving., Moderne Fahrzeuge verfügen über viele teil- und sogar hochautomatisierte Fahrfunktionen, welche die Sicherheit und den Komfort während der Fahrt enorm erhöhen. Als Grundvoraussetzung erfordern solche Systeme eine genaue und zuverlässige Wahrnehmung der Umgebung um das Fahrzeug, welche durch dafür vorgesehene Algorithmen auf Basis von verschiedenen Sensormessungen modelliert wird. Ein für diesen Zweck geeigneter Sensor stellt das Radar dar, welches in der Lage ist Position, Orientierung und Radialgeschwindigkeit von Objekten in der Fahrzeugumgebung direkt zu messen. Heutige Automobilradarsensoren liefern bereits hochauflösende Daten, sodass für ein Objekt mehrere Radarreflektionen erzeugt werden. Anschließend ist es die Aufgabe der Objektdetektion in der gemessenen Radartargetliste alle vorhandenen Objekte zu erkennen, das heißt deren Lokalisierung sowie deren zugehörigen Klasse zu bestimmen. Diese Arbeit stellt eine neuartige Technik für die Objektdetektion in hochauflösenden Radardaten vor, welche anschließend weiterverarbeitet wird um eine Wahrnehmung der Umgebung auf Objektebene zu realisieren. Zusammenfassend handelt es sich um ein vollständiges System von der Radartargetliste über die Objekterkennung bis hin zur zeitlichen Verfolgung all dieser detektierten Objekten. Das in dieser Arbeit entworfene Radar-Objektdetektionssystem basiert auf der Methode des Deep Learning, wobei der Detektor aus drei Hauptkomponenten besteht. Das erste Modul teilt die Radartargetliste in mehrere kleine Regionen auf und bestimmt, ob sich darin ein Objekt befindet und gegeben des Falls auch dessen Klassen. Die zweite Komponente segmentiert alle Radarreflektionen, die zum Zielobjekt gehören. Das dritte Modul bestimmt schließlich die Parameter für die zugehörige 2D-Bounding-Box, welche das Objekt modelliert. Dieser Prozess wird für alle Teiltargetlisten durchgeführt und letztendlich fasst die Objektextraktion alle prädizierten Hypothesen zu einem gesamtheitlichen Detektionsergebnis zusammen. Das Training der neuronalen Netzwerke verwendet ausschließlich reale Radardaten, welche aus typischen Verkehrssituationen in ländlichen und städtischen Gebieten stammen. Die Auswertung evaluiert den Radar-Objektdetektor auf drei verschiedenen Datensätze und bewertet die Zwischenergebnisse der einzelnen Module sowie das eigentliche Detektionsergebnis. Zwei der Radardatensätze sind öffentlich zugänglich, sodass alle Ergebnisse reproduzierbar und nachvollziehbar sind. Der dritte Datensatz ist selbst erzeugt und enthält hochauflösende Radardaten von echten Radarsensoren. Die Evaluation zeigt, dass der Objektdetektor auf allen Datensätzen überzeugende Ergebnisse liefert, wobei die Detektionsresultate auf dem selbst erzeugten Datensatz am besten abschneiden. Außerdem beweist eine erfolgreiche Implementierung sowie ein paralleler Betrieb von drei Radar-Objektdetektionssystemen auf drei Sensoren, die in einem realen System für autonomes Fahren installiert sind, dass diese fortschrittliche Technologie echtzeitfähig ist und sich für den Einsatz im realen Straßenverkehr eignet. Beim Einsatz von mehreren Sensoren mit einem sich überlappenden Sichtbereich, liefern die zugehörigen Radar-Objektdetektoren für das gleiche Objekt verschiedene Hypothesen, die sich durchaus widersprechen können. Um ein konsistentes Umgebungsmodell zu erhalten, müssen alle erzeugen Objektdetekionen fusioniert und über die Zeit verfolgt werden. Ein Algorithmus für Multi-Objekt-Tracking ermöglicht es, Messungen aus unabhängigen Quellen zu einem ganzheitlichen Ergebnis zu kombinieren. Solch ein Trackingalgorithmus besteht aus zwei wesentlichen Schritten, der Prädiktion von bestehenden Objekten gemäß einem Bewegungsmodell und der darauffolgenden Korrektur von aktuellen Objektzuständen mit Messungen. Diese Arbeit stellt ein geeignetes Messmodell vor, um alle Hypothesen, die das entwickelte Radar-Objektdetektionssystem liefert, mit dem eingesetzten Multi-Objekt-Trackingalgorithmus zu verarbeiten. Der Multi-Objekt-Tracker erzeugt unter Nutzung der gefundenen Radar-Objektdetektionen ein robustes Umgebungsmodell. Dabei optimiert der Algorithmus die Wahrnehmung von Objekten, sodass die resultierenden getrackten Objekte genau denen in der Realität entsprechen. Der in dieser Arbeit vorgeschlagene Radar-Objektdetektor und das anschließende Multi-Objekt-Tracking funktioniert im realen Betrieb sowohl schnell und präzise, als auch zuverlässig, sodass diese Technik ein unerlässlicher Bestandteil für die Umgebungswahrnehmung in einer realen Anwendung für das automatisierte Fahren darstellt.

Increasing demands in radar sensor performance for automotive applications require reliable test solutions for the development and verification. In this paper, a modulation-based radar target simulator suitable for automotive settings is presented. A cost-effective circuit with low complexity is proposed that modulates the radar signal directly in the E-Band without an intermediate frequency stage. Models are derived for predicting the limitations of such radar target simulators due to hardware nonidealities. The introduced radar target simulator is analyzed with respect to the derived nonidealities in order to predict its performance based on the models. As a verification, radar measurements with the target simulator are conducted, and compared with the models' predictions., acceptedVersion

With radar networks, the resolution of critical radar parameters like Doppler and angle can be improved compared to a single radar sensor. As the network’s aperture is considerably larger than the one of a single radar, a much higher angular resolution is achieved. However, with a large aperture, range- dependent phase deviations, i.e. near-field effects, occur and affect the angle estimation. In this work, these near-field effects are evaluated exemplarily for a coherent network. Furthermore, a new strategy to compensate those network near-field effects is proposed and demonstrated based on measurements. The benefits of the near-field compensation are emphasized by comparing the networks angle-estimation capabilities with and without compensated near-field effects., acceptedVersion

Next-generation radar sensors require imaging capabilities with high angular resolution. As for a single sensor, the aperture, and thus the achievable resolution, is limited due to the constraints of the front end, radar networks consisting of multiple sensors are a possible solution. However, their incoherency usually makes joint angle estimation impossible. This article presents a network concept consisting of an orthogonal frequency-division multiplexing (OFDM) radar and repeater elements, which receive the reflections from targets and retransmit them back to the radar. Thereby, any frequency conversion from radio frequency to baseband and vice versa is omitted such that the signal remains coherent to the initial transmit signal. To distinguish the bistatic signal transmitted by the repeater from the monostatic one of the OFDM radar, the orthogonal subcarrier structure of OFDM waveforms is exploited by combining a sparse radar transmit signal with a low-frequency modulation in the repeater. This allows to evaluate the bistatic signals at the radar with standard multiple-input-multiple-output (MIMO)-OFDM signal processing, leading to separate range-Doppler images for each virtual channel. Finally, it is shown that this method offers a coherent angular estimation based on the extended aperture of the network. For this purpose, a method to establish phase coherency by a reconstruction of the modulation phase is presented. The network concept is proved with measurements at 77 GHz., publishedVersion

Zivile Drohnen finden privat wie auch kommerziell zunehmend Verwendung und werden beispielsweise von Sicherheitsbehörden, als Lieferdrohne oder zur Inspektionen von schwer zugänglichen Objekten genutzt. Für einen benutzerfreundlichen und sicheren Betrieb greifen moderne Drohnen auf eine Vielzahl von Sensoren zurück. Aktuelle Systeme nutzen satellitengestützte Positionsbestimmung, barometrische Höhenmessung und zur Detektion der Umgebung Infrarot-, Ultraschall- und Kamerasensoren. Diese Arbeit stellt Untersuchungen zur Anwendung von hochauflösenden Radarsensoren an Drohnen als Altimeter, für die Kollisionsvermeidung und für die Kartierung der Umgebung vor. Radar-Altimeter können gegenüber der barometrischen Höhenmessung die Flughöhe über Grund messen und weitere nützliche Informationen, wie etwa das Vorhandensein von Vegetation, bereitstellen. Basierend auf durchgeführte Untersuchungen zur Detektion von Vegetation, zur Notwendigkeit eines Gimbals, zur Nutzung als Landeradar sowie Frequenz- und Bandbreitenbetrachtungen werden mehrere Konzepte und Umsetzung vorgestellt und die Vorteile gegenüber den aktuellen Sensoren gezeigt. Bei der zwei- und dreidimensionalen Umfelderkennung werden neben der Entfernung und der Geschwindigkeit zusätzlich die Raumwinkel Azimut und Elevation gemessen. Hierfür wurde ein mehrkanaliger Radarsensor mit austauschbarem Antennen-Front-End entwickelt, an dem je nach Anforderung unterschiedliche Antennensysteme montiert werden können. Mit hoher Entfernungs- und Geschwindigkeitsauflösung sowie einem großen Sichtbereich konnte bei der zweidimensionalen Umfelderkennung gezeigt werden, dass die Detektion von stationären, bewegten und schwachen Zielen, z.B. kleine Drohnen oder feine Teile der Vegetation möglich ist. Die direkte Geschwindigkeitsmessung als Alleinstellungsmerkmal der Radarsensorik konnte zudem genutzt werden, um aus der Zielantwort die Eigenbewegung der Drohne zu bestimmen. Um unter Berücksichtigung der limitierten Nutzlast von Drohnen bei der dreidimensionalen Umfelderkennung beide Raumwinkel mit hoher Trennfähigkeit und großem Sichtbereich abzudecken, wurde ein ausgedünntes Antennensystem mit niedriger Kanalanzahl entworfen. Das für die einzelnen Kanäle verwendete Antennenelement sowie die Optimierung des Antennensystems werden detailliert beschrieben und messtechnisch validiert. Für die Erstellung von Karten der Umgebung zur Selbstlokalisierung oder Wegplanung werden an Drohnen häufig zusätzlich Lidarsensoren montiert. Es konnte gezeigt werden, dass mit den Daten des bildgebenden Radarsensors für die Umfelderkennung hochauflösende zwei- und dreidimensionale Radarkarten erstellt werden können. Untersuchungen zur Genauigkeit der Radarkarten zeigen, dass die Kenntnis der Position der Drohne und nicht der Radarsensor ausschlaggebend ist. In Hinblick auf die limitierte Nutzlast von Drohnen konnte selbst mit den satellitengestützten Positionsdaten der genutzten Drohne ohne zusätzlichen Lidarsensor Karten der Umgebung mit einer Genauigkeit von 0,5m erstellt werden. Die in dieser Arbeit gewonnen Erkenntnisse zeigen, dass Radarsensoren einen großen Beitrag für den sicheren Betrieb von Drohnen leisten und neue Anwendungsmöglichkeiten für Drohnen bieten können.

Multiple sensors in a radar network can be jointly evaluated to process the bistatic Doppler information of a target obtained under different aspect angles. This enables the estimation of the target’s vectorial velocity in a single snapshot. As a precursor to the joint evaluation, the monostatic and bistatic Doppler components need to be matched for each target in the range-Doppler map. In this work, a new approach for the target matching in the case of a radar-repeater network is presented. The matching is performed based on the correlation of monostatic and bistatic steering vectors extracted from the range-Doppler map for each receive channel of the radar. After validating the matched responses, the vectorial target velocity is determined. The approach is successfully tested with a 77 GHz radar and two repeaters by measurements of rotating cylinders., acceptedVersion

Recent advances in digital hardware and processing power have enabled the realization of novel digital radars that are superior to radars commercially available today in flexibility and performance. This opens up new opportunities in radar waveform design, modulation, and high-performance signal processing that were previously unfeasible. Probably the most known digital radar modulation concept is OFDM, which has attracted much interest in radar research in the last two decades. The goal of this thesis is to explore new possibilities for OFDM radar waveform design and high-performance digital processing to make future digital radars even more powerful and robust. To this end, the developed novel waveform and signal processing concepts take advantage of the high degree of flexibility and diversity provided by OFDM to overcome previous limitations in radar performance, to enable new MIMO concepts, to enhance radar imaging performance, and to enable robust radar sensing in coexistence with other radars.

A wideband and robust design of a dielectric rod antenna at G-band (140 GHz–220 GHz) for short range radar applications in harsh environmental conditions is presented. The proposed antenna is based on a simple plug-in solution made of PEEK material (εr,PEEK =3.2), which enables the mechanically detachable integration into a glass-encapsulated millimeter wave radar sensor system. To further increase the antenna gain, an optically transparent biconvex lens is presented. The antenna performance has been investigated by full wave simulations and validated through measurements of the fabricated antenna prototype. The experimental results of the realized antenna element show excellent agreement to the simulated values. The rod antenna covers a 3 dB-beam width of 46° and achieves a maximum gain of 12.4 dBi. The additional use of the dielectric lens increases the gain to 30.4 dBi with a stable beam pattern over the entire modulation bandwidth of 20 GHz of the radar sensor. The measured 3D-radiation patterns at various frequencies across the operational band are provided., acceptedVersion

Noncoherent detection schemes are an appealing and low-complexity alternative in multi-user massive MIMO uplink systems compared to classical coherent detection algorithms, since no actual channel knowledge is required at the receiver. For noncoherent multi-user detection to function, the induced power at the base station is utilized to separate the different users. However, spatial separation is impossible when the users are located in the far-field of the receiving antenna array. Consequently, noncoherent detection fails in such scenarios. To this end, beamspace techniques can be applied, focusing the energy of the incident wave to a smaller subset of the receive antennas and enabling again the noncoherent detection scheme. This paper analyzes the beamspace capabilities of a dielectric lens and an analog beamforming network applied at the receiver. Furthermore, a sub-array architecture is proposed, relaxing the design requirements for practical implementation. It is shown that noncoherent detection in combination with beamspace techniques performs comparably to channel-estimation-based detection. In addition, the sub-array architecture revealed a significant performance enhancement accompanied by a reduced user separability., acceptedVersion

The estimation of the direction-of-arrival in a coherent multi-channel radar system requires the distribution of the high-frequency local oscillator signal to all hardware channels. Various system concepts with distribution network or feedthrough topologies are established. While distribution networks are bulky and costly, the time delays in systems with a feedthrough topology cause systematic errors in the range evaluation. In this work, a system concept based on an injection-locked feedthrough and the related radar system analysis is presented. The disadvantage of time delays in common feedthrough architectures can be eliminated by a hardware correction using the phase shifting capabilities of the injection-locked oscillator. In addition, two software correction algorithms are presented and analyzed. The performance of the realized system is shown by radar measurements and direction-of-arrival estimations. It is proven that the phase noise of different injection-locked oscillators is fully correlated within the radar system which results in the same detection performance as in state-of-the-art radar systems using a signal distribution network. Thus, the injection-locked feedthrough imaging radar topology is a suitable concept of realizing flexible radar frontends without bulky distribution networks., publishedVersion

Radar networks offer the opportunity of enhancing the radar sensing performance beyond the inherent limitations of single multiple-input multiple-output (MIMO) radars. Different locations of the individual sensor nodes allow for diverse observation angles on the targets. Moreover, in case of a coherent network, such as radar-repeater networks, the direction of arrival (DoA) capabilities are increased. However, a crucial point in exploiting the full potential of a sensor network is the separability of the signals originating from the different sensor nodes during operation, while preserving reasonable radar sensing performance of the individual sensors and the network. In this paper multiplexing methods for an orthogonal frequency-division multiplexing (OFDM)-based radar network are proposed and investigated. The first method combines traditional time and frequency multiplexing, while the second method applies a random assignment in combination with a compressed sensing-based signal evaluation. Both methods are compared with the traditional FDM multiplexing for MIMO-OFDM. For this purpose, performance criteria are proposed and the concepts are evaluated based on measurements and simulations., acceptedVersion

There is a growing demand for collaborative robots in order to improve the ef���ciency of industrial processes and to push Industry 4.0 forward. The challenge for these robots is the compliance of the safety regulations to protect human workers from injuries. Therefore, a large number of sensors for the close range (several centimeters up to some meters) around the robot are needed. For a protective cover around the robotic arms all sensors must cooperate and process their information together. This paper presents a close range proximity sensor based on a 160-GHz radar MMIC with a transition to two ���exible dielectric waveguides. On the low-loss waveguides, leaky-wave antennas with a large ���eld of view are distributed. This reduces the number of sensors and achieves a coverage area for one sensor of more than 2200 cm�� for a distance of 30 cm. For distances below 30 cm the sensor has blind areas. However, the expected extended targets like body parts can still be detected and the proximity sensor will stop the robot in the danger area. Additionally, the ���exible waveguide offers the possibility to place the antennas almost arbitrarily around the robotic arm. For a single target scenario it is even possible to determine the target position by multilateration with only one receive channel., publishedVersion

Modern consumer and industrial unmanned aerial vehicles (UAVs) are easy-to-use flying sensor platforms. They offer stable flight, good maneuverability, hovering, and even waypoint flights in autopilot mode. For stabilization and localization, sensors such as inertial measurement units (IMUs)-including gyroscopes and accelerometers-barometric sensors, and the Global Navigation Satellite System (GNSS) are used. To sense the UAV's direct environment, e.g., for collision avoidance or fully automated flight, additional sensors are needed. State-of-the-art combinations of infrared, ultrasonic, and vision-based sensors (monocular and/or stereo vision) capture the close vicinity. Using radar sensors is advantageous, as they are able to directly sense range and velocity and are not affected by lighting conditions and contrast. With the help of a multichannel radar, the angular information may also be extracted., acceptedVersion

In order to improve the resolution of imaging radars, electrically large arrays and high absolute bandwidths are required. At a large off-boresight angle of the incident wave, the difference in path length at the antennas becomes a multiple of the range resolution of the radar. Consequently, the radar responses of the targets at the different receiving channels are distributed over a large number of range bins, depending on the direction-of-arrival (DoA) of the target echo. Applying conventional signal processing, the resolution of the radar is considerably reduced. In this paper, the key aspects of DOA estimation are discussed when radars with both a large aperture size and a high absolute bandwidth are employed. A signal processing method is proposed avoiding errors that occur in conventional DOA estimation techniques. As shown by measurements, an angular resolution of 0.4�� can be achieved with an array size of 200 lambda and a bandwidth of 10GHz., acceptedVersion

Flawless operation even with a high density of radar sensors in the vicinity is an essential prerequisite for radar sensors in automotive applications, safety systems, future autonomous driving, and in industry applications. Recent research shows the great potential of digitally modulated radars such as orthogonal frequency-division multiplexing (OFDM). Furthermore, novel frequency agile OFDM approaches such as stepped and sparse OFDM are superior to standard OFDM as they are able to achieve high resolution beyond the hardware-related band limits. In order to prevent possible mutual interference issues of common and novel modulation schemes, the influence of a state-of-the-art FMCW on stepped and sparse OFDM is examined in this paper. The effects of interference are investigated theoretically and through measurements using an experimental OFDM setup at 76 GHz and scenarios with up to two FMCW interferers. To prevent signal deterioration and limited detection performance due to interference, an efficient but low-cost detection and mitigation approach is presented for each of the two approaches., acceptedVersion

Multiple-Input Multiple-Output (MIMO) radars constitute a proven approach to realize cost-efficient high-precision environment sensing solutions. Its vari- ant with widely distributed antennas benefits from spatial diversity, offering a large field of view and avoiding ghost targets. Until recently, operation of such a MIMO radar required the distribution of a common high-frequency car- rier signal or other elaborate synchronization solutions. This work introduces a system concept and prototypical realization of an incoherent MIMO radar network based on integrated radar sensors without wired high-frequency links. The network simultaneously performs monostatic and bistatic measurements between all pairs of sensor nodes, using independently generated Frequency- Modulated Continuous-Wave (FMCW)-modulated transmit signals with fixed frequency offsets. In addition to optimum parametrization of the network, the complete signal processing chain for the operation of this network is presented. First, co- herency is established by estimating the unknown time, frequency, and phase offsets, yielding high-precision range and speed measurements. Subsequently, these estimates are used to distinguish the contour of targets or estimate their position using multilateration algorithms. The latter include bistatic measure- ments of both range and speed, which is shown to improve the precision of target position estimates by a factor of 10 with the used system parameters. The performance limits of this approach considering random and systematic nonidealities are determined by deducing them from fundamental stochastic processes. They are verified with a newly introduced computationally efficient simulation which regards the correlation in the stochastic processes. This re- veals strict requirements for the purity of the independently synthesized signals to allow bistatic measurements. The prototypical realization of the network with three sensor nodes using integrated radar transceivers operating at 122GHz proves the feasibility of the approach. With chirp sequences of 30 ramps of 1ms length and 1GHz bandwidth, unambiguous range precisions of < 1mm are achieved using a small metal panel at 1m distance. The reconstructed phases exhibit a precision of < 0.1rad, allowing ambiguous range precisions of < 10µm and speed precisions of < 1mm/s. The attainable high precision and comparatively low realization effort make the proposed network approach a favorable design for a wide range of applications.

Although the beginning of research on automotive radar sensors goes back to the 1960s, automotive radar has remained one of the main drivers of innovation in millimeter wave technology over the past two decades. Today, millions of sensors are produced each year, which was made possible by inexpensive and mature millimeter wave technology. The technology maturity, in turn, enables research to be carried out on systems that are considerably more complex and powerful than was possible just a few years ago. The focus of research has thus shifted from purely hardware-oriented and device-level topics to sophisticated millimeter wave systems and RF signal processing topics. This opens up new research topics such as digital modulation schemes, radar networks, radar imaging, and machine learning. In this review paper, we sketch the path from the very beginning through the state of the art with sophisticated multiple-input multiple-output (MIMO) antenna arrays and mature assembly and interconnect concepts to today’s key research topics of automotive radar., publishedVersion

For imaging radars, the calibration effort increases with higher frequency and finer angular resolution demands and is therefore one of the largest cost factors in radar production. In case of high angular resolution, the calibration is also sensitive to array misalignment, resulting in costly calibration procedures. This paper proposes an angle-dependent radar calibration method with significantly reduced calibration effort. The high efficiency and robustness against misalignment is achieved by exploiting phase symmetry of a target in the measured radar response. Based on an initial theoretical formulation followed by an experimental verification, this novel approach does not only yield a cost-effective calibration, but also increases the robustness against array misalignment., acceptedVersion

Die hohe Auflösung, der sehr kompakte Aufbau und die kostengünstige Herstellung von Radarsystemen im oberen Millimeterwellenbereich zwischen 150 GHz und 300 GHz erweiteren das bekannte Anwendungsspektrum von Radarsensoren enorm. Zusätzliche Komponenten wie sperrige Linsen und Gehäuse zum Schutz bei rauen Umgebungsbedingungen beschränken allerdings die Einsatzmöglichkeiten. In dieser Arbeit wird daher ein Radarsystem mit Messsonde basierend auf einem flexiblen dielektrischen Wellenleiter vorgestellt, das beinahe unabhängig von räumlichen Anforderungen und ohne aufwendiges Gehäuse eingesetzt werden kann. Hierbei werden die einzelnen passiven Komponenten ausführlich betrachtet und ihre Auswirkungen auf die gewonnen Radarinformationen analysiert sowie Kompensationsmöglichkeiten für Nichtidealitäten präsentiert. Im Nahbereich unter 50 cm ergibt sich für die Messsonde selbst bei Wellenleitern von bis zu 4 m Länge eine höhere Empfangsleistung als für ein herkömmliches Radar. Durch die geringen Verluste und die hohe mechanische Flexibilität ergeben sich somit völlig neue Anwendungsgebiete für Radarsysteme, wie anhand von Beispielen verdeutlicht wird.

Emerging digital radar concepts such as orthogonal frequency-division multiplexing (OFDM) allow flexible signal generation. This opens up new opportunities in waveform design in a multiple-input multiple-output (MIMO) system such as introducing coding for signal multiplexing. In this article, coded MIMO OFDM waveforms are proposed and investigated that allow continuous and simultaneous wideband transmission for all transmitters of a multiple transmit and receive antenna array for spatial radar environment perception. Challenges for coded MIMO OFDM radar operation are derived, and three coded MIMO strategies are introduced and analyzed. Their potential is validated and compared to the standard subcarrier interleaving OFDM approach using simulations and measurements of an experimental 4×4 MIMO OFDM radar at 77 GHz., acceptedVersion

This paper presents the channel influence when simulating interference between automotive radars in realistic scenarios. As a starting point, a simple approach considering only the line-of-sight (LoS) component of the signal transmitted by the interferer radar is presented. Following, the influence of the ground reflections in interference situations is analyzed by introducing the two-path model. Properties of the road surface such as roughness and the Fresnel reflection coefficient are considered in order to obtain a more realistic simulation. A sweep of all the possible ranges between victim and interferer radars is performed, and the influence of the interferer radar is analyzed. Finally, a statistical analysis of the key radar parameters is carried out in order to evaluate the interference behavior in the presented model., acceptedVersion

Radar sensors are an integral part of driver assistance systems. The performance of modern automotive radar sensors can be improved by incorporating multilayer wideband antennas in these systems. In this thesis, such wideband multilayer antenna array concepts for automotive radar sensors operating in the frequency range 76--81 GHz are investigated. The first two concepts investigated are based on a conventional multilayer PCB. A variable width grid array antenna is used for both designs, with the feed network being the difference between both concepts. For the first design, a two substrate layer PCB is used. The second design employs a three substrate layer PCB. A novel hybrid approach to design multilayer antennas is then introduced. Multilayer antennas are designed using single layer PCB and multilayer thin films attached to the PCB. The single layer PCB houses the complete feed network, whereas the radiating elements are realized on multilayer thin films. Two antenna concepts are presented as examples of this approach. The first is an aperture coupled stacked patch (ACSP) antenna array fed by a grounded coplanar waveguide (GCPW) feed network. A novel compact power divider is included in the feed network to reduce inter-element distance. The second concept is based on a novel antenna element, namely the stacked grid antenna (SGA). Feed networks with three different transmission line types are realized, namely the SIW, GCPW and MS, to create antenna arrays with the SGA to show the flexibility of antenna design. A comparison between the multilayer PCB approach and the hybrid thin film approach is then performed, using the investigated designs as the basis. The comparison shows that both approaches have their benefits. In terms of aperture efficiency, multilayer PCB antennas can achieve better results due to a completely buried feed network. However, it is shown that the hybrid thin film approach achieves similar wideband antenna performance as the conventional multilayer PCB approach, with added benefits of simple design, low cost manufacturing and increased robustness.

Since the number of radar sensors per vehicle has been increasing in order to improve the performance of Advanced Driver Assistance Systems, mutual interference has become one of the most important challenges for near future automotive radar systems. Depending on the sensor parameters and the diverse properties of the traffic scenario, the interference effects can significantly vary. The effort to measure and appraise the interference in real traffic scenarios leads to the importance of designing a tool suited to simulate and analyze the interference. This paper presents a fast simulator capable of modeling, estimating, and evaluating the effects of interference in complex multi-user traffic scenarios. A statistical analysis of the interference in a proposed scenario is performed in order to validate the presented simulator., acceptedVersion

Monostatic,multiple-input-multiple-output (MIMO) radar configurations using the same antenna array for transmission and reception inherently lead to redundant virtual antenna positions within the virtual MIMO aperture. This article presents a new method for designing monostatic MIMO radars where this disadvantage may be overcome and nonredundant virtual arrays are yielded by exploiting biomimetic antenna arrays (BMAAs). The new concept of biomimetic MIMO is introduced and thoroughly investigated in this article. Its working principle is compared with conventional monostatic MIMO and verified by radar measurements. The new approach is intended to improve the angle estimation capability of, in particular, small radar systems. Therefore, the concept is applied to an ultracompact two-channel radar monolithic microwave integrated circuit (MMIC) with fully integrated antennas for range and angle measurements in the 150-GHz range. The chip features two monostatic transmit-receive (TRX) channels with an antenna spacing of ��/4 and occupies less than 3 mm�� of space. By applying biomimetic MIMO, a virtual aperture consisting of four unique antenna positions and an equivalent total size of around 1.5�� is achieved., publishedVersion

The use of radar systems is limited in some applications by spatial constraints or special thermal and environmental conditions. The spatial separation of the sensitive electronics and the more robust antenna by a ���exible connection therefore opens up new applications. A 160-GHz radar system with a mechanically ���exible front end ful���lling these requirements is proposed in this paper. The ���exible front end is an extremely low loss dielectric waveguide feeding a dielectric elliptical lens antenna (28 dBi gain). The dielectric waveguide has dielectric losses of 4.5 dB/m at 160 GHz and is very ���exible, allowing bending radii of down to 1.5 cm with negligible losses. The dielectric waveguide is fed by a 160-GHz radar monolithic microwave integrated circuit (MMIC), which allows bandwidths of up to 20 GHz for a high range resolution. The transition between MMIC and dielectric waveguide is realized with a rectangular-waveguide interface. The radar back end consists of a phased-locked loop (PLL) with standard components, an intermediate frequency (IF) signal conditioning part, and a Xilinx Zynq 7030 System-On-Module (SOM) with an FPGA and an ARM-based processor. The sampled signal is processed directly on the FPGA with a 2D Fourier transform and is available as a UDP stream with an update rate of up to 15 Hz. In addition, a camera image is taken for each radar measurement. The presented system is used to detect and measure the ���ight behavior of honey bees. The electronics are housed in a building whereas the ���exible dielectric waveguide allows the antenna to be placed anywhere around the beehive, where it is exposed to environmental conditions., publishedVersion

Imaging radars are usually realized fully coherently. However, the distribution of one common radio frequency signal to all transmit and receive paths requires a high degree of hardware complexity. In order to reduce the hardware effort significantly, a novel phase synchronization method for incoherent and quasi-coherent frequency-modulated continuous-wave (FMCW) imaging radars with individual signal synthesis per channel is presented. The quasi-coherent setup uses one common oscillator for all frequency synthesizers. It is shown that in the case of the quasi-coherent system, only a phase difference between the calibration and the measurement has to be corrected to achieve coherence. In comparison, an incoherent system causes additional time, frequency, and FMCW ramp slope errors due to the different behavior of the oscillators. In order to achieve phase coherence and to correct the error sources, a calibration-based method using a defined signal path as part of the radar system is proposed. The imaging radar used for verification of the theory consists of individual single-channel radar monolithic microwave integrated circuits (MMICs) at 160 GHz; each MMIC fed by an individual frequency synthesizer. As shown by measurements, it is possible to achieve phase coherence for both system approaches and to perform angle estimation., publishedVersion

An essential prerequisite for radars in automotive applications, safety systems, future autonomous driving, and industry applications is the reliable operation even under difficult conditions such as interference from other radars. Recently, digital radar principles such as orthogonal frequency-division multiplexing (OFDM) gained increasing attention. In this letter, an interference-robust processing for OFDM radar signals using compressed sensing (CS) is presented, which is particularly suitable for random OFDM but also applicable to stepped and standard OFDM. It allows robust radar operation in presence of interfering signals that have a comparable low instantaneous bandwidth and high narrowband power. The method is validated experimentally using measurements of an OFDM radar setup at 76 GHz and a state-of-the-art FMCW radar as an interferer., acceptedVersion

Radar sensors are crucial components of advanced driver assistance systems. In order to cope with the future requirements of highly automated and autonomous driving, today’s automotive radar sensors performance needs to be improved. This dissertation addresses the design and development of robust mm-wave antenna systems suitable for automotive radar applications. This study presents a microstrip antenna array with wideband and stable radiation performance. It proposes also an optimization strategy for finding the position of transmitting and receiving antennas in a MIMO radar while taking the angular resolution requirements into consideration. Finally, it analyzes the problem of the integration of the radar sensor in the vehicle and introduces the use of low-RCS antenna arrays to mitigate the multiple reflections between the bumper of the car and the radar sensor, when the latter is hidden integrated in the vehicle. Some low-RCS antenna designs are proposed and discussed.

The system concept of a compact three-dimensional (3D) imaging radar in the millimeter-wave range around 150 GHz is presented. The multiple-input multiple-output (MIMO) radar system is composed of two identical two-channel radar monolithic microwave integrated circuits (MMICs) with on-chip antennas arranged perpendicular to each other. Typically, such arrangements suffer from angular ambiguities due to large antenna spacings that come along with the MMIC footprint and printed circuit board (PCB) routing constraints. The concept of biomimetic antenna arrays (BMAAs) is used to mitigate the problem of ambiguities. The influence of applying a spherical lens to the radar system is discussed in the paper, and radar measurements show the applicability of the proposed concept., acceptedVersion

In order to improve the resolution of imaging radars, electrically large arrays and a high absolute modulation bandwidth are needed. For radar systems with simultaneously high range resolution and very large aperture, the difference in path length at the receiving antennas is a multiple of the range resolution of the radar, in particular for off-boresight angles of the incident wave. Therefore, the radar response of a target measured at the different receiving antennas is distributed over a large number of range cells. This behavior depends on the unknown incident angle of the wave and is thus denoted as range-angle coupling. Furthermore, the far-field condition is no longer fulfilled in short-range applications. Applying conventional signal processing and radar calibration techniques leads to a significant reduction of the resolution capabilities of the array. In this article, the key aspects of radar imaging are discussed when radars with both a large aperture size and a high absolute bandwidth are employed in short-range applications. Based on an initial mathematical formulation of the physical effects, a correction method and an efficient signal processing chain are proposed, which compensate for errors that occur with conventional beamforming techniques. It is shown by measurements that with an appropriate error correction an improvement of the angular resolution up to a factor of 2.5 is achieved, resulting in an angular resolution below 0.4�� with an overall aperture size of nearly 200 \lambda., publishedVersion

Especially for short-range radar sensors, a protection of the electronics against environmental damage like heat or humidity as well as a robust antenna front end are essential. In order to separate the sensitive electronics and the robust antenna spatially, a system concept for a 300-GHz radar with mechanically ���exible front end is presented in this paper. The ���exible front end is realized with a low-loss dielectric waveguide (DWG) with losses of around 15 dB/m. It is bendable up to a radius of 1 cm. An elliptical lens (30.9 dBi) at the end is used to focus the beam. The front end is fed by a radar MMIC in a QFN-like package with an integrated silicon lens. The wide beam is focused by simply placing a multimodal DWG above package and lens. The intentionally caused interference of the individual modes then results in a ���eld maximum in the center, and the fundamental mode can be fed into the DWG., acceptedVersion

Immer komplexere und leistungsfähigere Fahrerassistenzsysteme in Fahrzeugen sind auf eine möglichst genaue Erfassung der Umgebung angewiesen. Eine wichtige Sensorklasse stellen dabei die Radarsensoren dar, da diese durch den Dopplereffekt direkt die radiale Geschwindigkeit von Zielobjekten erfassen können. Zukünftige Radare müssen über eine immer bessere Auflösung für die Bestimmung der Entfernungen, der Geschwindigkeiten und der Winkel, unter der Ziele erfasst werden, verfügen. Dies ermöglicht eine Klassifikation von Zielobjekten, beispielsweise die Unterscheidung zwischen Fußgängern und Fahrzeugen. Diese Arbeit stellt optimierte Ansätze der Signalverarbeitung vor, um unter anderem die Geschwindigkeitsauflösung durch eine geschickte Messauswertung zu verbessern, ohne die Messdauer zu verlängern. Weiterhin werden zwei Verfahren betrachtet und verglichen, die den Eindeutigkeitsbereich der Geschwindigkeit bei der Verwendung mehrerer Sendeantennen vergrößern. Eine verbesserte Winkelauflösung wird durch eine gezielte Unterabtastung und anschließende Rekonstruktion erreicht, sodass in der gleichen Entfernung und Geschwindigkeit unterschiedlich starke Ziele getrennt werden können. Zum Schluss werden zwei verschiedene Ansätze zur Unterabtastung in allen Messdimensionen und der anschließenden Rekonstruktion verglichen.

For unmanned aerial vehicles (UAVs), grid maps can be a versatile tool for navigation and self-localization. In general, payload is critical for UAVs and every additional sensor limits the flight duration. Due to its robustness and the ability to directly measure velocities, radar sensors are well suited for sense and avoid applications (SAAs) for UAVs. It would be advantageous if these sensor data could be used to generate grid maps instead of mounting additional sensors such as light detection and ranging (LiDAR). This letter demonstrates that using the data from high-resolution multiple-input-multiple-output (MIMO) imaging radars, high-resolution 2-D and 3-D radar grid maps can be created. The necessary adaption of the sensors free-space model for MIMO radar-based occupancy grid maps is presented in detail. UAV-borne measurements resulting in 2-D and 3-D grid maps with an adequate representation of the environment validate this approach., acceptedVersion

Have you read everything about direction-of-arrival estimation in textbooks but are you still uncertain how to realize it in practice? This tutorial-like paper will help to link the theory with a practical approach for direction-of-arrival estimation using mm-wave radar systems and it deals with arising challenges. Step by step, it is explained how to move from the measured time domain data to the estimated angular position of the target. Since the target angle estimation is usually carried out after range-velocity processing, the required pre-processing steps are described. The assessment criteria to form the relationship between array design and radar performance are expressed with the Rayleigh criterion for angular resolution and the ambiguity function as a measure for the unambiguous angular range. An in-depth instruction on how to carry out the sensor calibration is provided. Besides the choice of a proper calibration object, the required number of measurements and signal-to-noise ratio related issues are considered and supported by measurement results. The discussion of single- and multi-target measurements with the calibrated sensor concludes this work., acceptedVersion

This paper presents an approach to estimate the true direction of motion and the velocity of a target within a single measurement snapshot. This is achieved by using a single radar sensor in combination with an active repeater element that is operated as a displaced virtual sensor. Coherency between sensor and repeater tag is only obtained by signal processing; on a hardware level both operate incoherently. This setup provides the necessary multi-perspective view on the target at low effort and large freedom of placement. The results of simulations and measurements in the 77 GHz band with one radar sensor and two repeater elements are presented to validate the theory., acceptedVersion

Multiple-input-multiple-output (MIMO) radars show improved angular resolution compared to traditional phased-array radar systems. Exploiting the advantages of MIMO radars could thus significantly improve the performances of the next generation of automotive radar sensors. The placement and the characteristics of the antenna arrays affect the performance of the MIMO system. For such applications, different metrics like the ambiguity function of the antenna system must be accurately evaluated. This letter presents a detailed characterization of an MIMO antenna configuration at millimeter-wave frequencies. To prove the effectiveness of this method, a five-element nonuniform antenna configuration based on cavity antenna arrays has been designed and tested., acceptedVersion

Die Radarsensorik stellt aufgrund ihrer hohen Reichweite und Robustheit eine Schlüsseltechnologie für moderne Fahrerassistenzsysteme und automatisierte Fahrfunktionen dar. Dem gegenüber steht das begrenzte laterale Auflösungsvermögen der Radarsensorik als ein limitierender Faktor. Gleichzeitig sorgt eine immer größer werdende Marktdurchdringung von Assistenzsystemen für neue Herausforderungen bei der Interoperabilität und der Vermeidung von Störungen. Ziel dieser Arbeit ist es, mittels Methoden der digitalen Signalverarbeitung die Auflösung und Robustheit von Automobilradaren zu steigern. Hierzu werden zunächst die radartechnischen Grundlagen beschrieben und ein Signalmodell für Radare mit Chirp-Sequence-Verfahren und Gruppenantenne aufgestellt. Darauf folgt eine vergleichende Untersuchung von hochauflösenden Detektionsalgorithmen. Dabei wird u. a. eine Erweiterung des Relax Algorithmus präsentiert, um eine speichereffiziente Implementierung bei gleichzeitig hoher Nebenläufigkeit zu ermöglichen. Außerdem wird auf den Aspekt der Kalibrierung eingegangen, weil eine gute Kalibrierung unabdingbar für eine gute Auflösung ist, da andernfalls Unschärfe bei den Detektionen auftritt. Hinsichtlich der Interoperabilität behandelt diese Arbeit das Thema der Stördetektion und -unterdrückung. Dabei wird untersucht, wie sich weitere Radarsensoren negativ auf das betrachtete System auswirken können. Schließlich folgt die Vorstellung eines neuartigen Verfahrens zur Stördetektion, welches auf Methoden der Mustererkennung beruht. Dadurch lassen sich Störungen wirksam unterbinden und die Maskierung von schwachen Zielen vermeiden. Dazu wird anhand eines Beispiels demonstriert, wie Störungen einzelne Detektionen maskieren können. Darüber hinaus werden Anwendungen für hochauflösende Radardaten vorgestellt. Zum einen wird demonstriert, welches Potential zur Konturbestimmung besteht, und weitere Aspekte diesbezüglich werden diskutiert. Zum anderen wird ein probabilistisches Framework vorgestellt, welches zur Eigenbewegungsbestimmung genutzt werden kann. Abschließend folgt ein Ausblick auf potentielle Zukunftsthemen im Bereich der automobilen Radarsensorik., Radar sensors render a key technology for advanced driver assistance systems and automated driving due to their high range and robustness. However, the limiting factor for radar technology in automotive applications is the lateral resolution. Further the growing market penetration of driver assistance systems with radar sensors provides new challenges regarding interoperability and the avoidance of interference. The goal of this work is to increase the resolution and robustness of automotive radar sensors by means of digital signal processing. Therefore the basic radar principles are explained first and a signal model for chirp sequence radars with array antennas is introduced. Then high resolution detection algorithms are investigated and compared to each other. As an additional benefit improvements are proposed such as a new implementation of the Relax algorithm, which allows high parallelism even with constraints on the amount of memory, that can be used. As an add-on to the detection algorithms the calibration is investigated, because this is an essential to achieve a good resolution. Regarding the interoperability the interference is investigated. It is shown how other radar sensors can negatively effect the considered radar system. To overcome this issue a novel method is proposed to detect interference with algorithms for pattern recognition. Based on this method the influence of the interfering signal can be suppressed and the masking of weak reflecting targets is avoided. By an example measurement is demonstrated, how interference can mask detections. Furthermore applications of high resolution radar data are presented. The high potential for object contour estimation is demonstrated. Further a probabilistic framework for ego motion estimation is derived. Finally an outlook to future research topics in the field of automotive radar is given.

One common problem of frequency modulated continuous wave radar is leakage from the transmitter to the receiver. The leakage power is orders of magnitude larger than the target return power and appears as a very strong signal in the first few range bins. Additionally, the residual phase noise density of the local oscillator occurs around the leakage signal, which often raises the noise floor and limits the dynamic range of a radar system at the close proximity of the sensor. In this paper a novel system concept that cancels the phase noise around the dominating leakage path is proposed, mathematically derived, and proven by radar measurements with a radar demonstrator at 77 GHz., acceptedVersion

Anti-personnel fragmentation mines are relatively large metallic mines, which are only partially buried and often triggered by a metallic tripwire. In humanitarian mine clearance, the search for the wires is usually carried out manually. As a new approach, an airborne system for the detection of tripwires using a synthetic aperture radar is presented. The system consists of an industrial multicopter, a frequency-modulated continuous-wave radar, and a real time kinematic global navigation satellite system. For image formation, a back-projection algorithm is used. Measurements with tripwires attached to a dummy mine successfully demonstrate the functionality of this system approach. In addition, the influence of wire length, vegetation, and incidence angle are investigated. It is shown that several overflights with different directions of flight are required to detect randomly oriented tripwires., acceptedVersion

Viele Anwendungen im industriellen Umfeld können durch kostengünstige Radarsensoren, die in Silizium-Germanium hochintegriert sind, erschlossen werden. Ausgehend von den typischen Anforderungen und Randbedingungen wird ein Systemdesign vorgestellt, mit dem ein möglichst leistungsstarker Radarsensor erreicht werden kann. Dabei werden die Erzeugung eines geringen Phasenrauschens bei der Betriebsfrequenz und der systematische Vergleich eines monostatischen und eines bistatischen Radarchips diskutiert. Durch die Integration der Antennen auf den Radarchip ist das gesamte Millimeterwellen-Frontend auf einem MMIC realisiert. Die Chip-integrierten Antennen werden innerhalb der Aufbau- und Verbindungstechnik umfassend erläutert. Als kostengünstige Häusungstechnologie werden QFN-Gehäuse betrachtet. Sowohl für offene als auch teilvergossene QFN-Gehäuse werden einfach umsetzbare Gehäuseanpassungen vorgestellt, die eine sinnvolle Verwendung über 100 GHz ermöglichen. Abschließend wird der modulare Radardemonstrator vorgestellt und u. a. in Feldtests in einer industriellen Tankanlage charakterisiert.

In automotive sensor networks the spatial distribution of the radar sensors allows to unveil additional information about the target objects. For example, the different angles to a target provide an increased robustness, and the wide sensor distances allow for a single snapshot motion estimation. This is even possible with single-channel radar sensors, but the reliability of the output highly depends on the condition that single scattering points are jointly detected. This paper presents an algorithm which associates single detections of a target over multiple sensors by utilizing the ego-motion of the vehicle. Further, it is shown how all target detections can be assessed and how reliable detections can be identified to improve subsequent processing steps like target localization. The proposed processing is validated by radar measurements at 77 GHz., acceptedVersion

As high resolution automotive radars become more common, so does their usage for next-generation functionalities like advanced driver assistant systems and autonomous driving. This creates the need for robust clustering techniques to distinguish among multiple extended objects like vehicles in the same scenario. One especially challenging scenario is that of separating two extended targets close to each other, each following its own trajectory. This paper proposes a clustering algorithm based on the analysis of the velocity profile to divide target points of multiple vehicles into sub-clusters. The theoretical background is explained and shown on simulation data. The algorithm is verified using radar measurements of two extended vehicular targets., acceptedVersion

In order to enhance the resolution of radar sensors, the sampling of the different measurement domains range, Doppler, and direction-of-arrival can be improved. This, for example, leads to the application of a sparse array for an increased angular resolution and an enhanced multiple-input multiple-output switching scheme for an increased maximum Doppler velocity. In this paper, an investigation is performed in order to determine the dimension where undersampling is most effective. As the measurement dimensions are linked, two different reconstruction approaches of the subsampled data are presented and evaluated. Measurements validate the effective reconstruction of the subsampled data., acceptedVersion

The multiple-input and multiple-output (MIMO) principle is a well-established method to improve the angular resolution of radars. Due to hardware limitations the maximum number of transmit and receive channels is often limited. In order to further increase the angular resolution with a given number of hardware channels, sparse arrays can be employed. However, the optimal placement of the antenna positions is a critical design parameter and it is difficult to consider non-idealities and mutual coupling within the design process. This paper addresses the non-idealities and the influence of coupling on the virtual array positions of a MIMO radar. An efficient calibration method to determine the real antenna positions is proposed. The discrete Fourier transform (DFT) based direction-of-arrival (DoA) estimation is shown with measurements of a 77 GHz and a 160 GHz MIMO radar for a uniform linear array (ULA) and two sparse arrays in comparison., acceptedVersion

In this paper, a coherent multistatic radar network with a novel system architecture is presented, which circumvents the general problems of clock distribution and phase noise related signal-to-noise ratio (SNR) issues. The proposed network consists of a variable number of multiple-input multiple-output (MIMO) radar sensors and a variable number of repeater tags, all of which operate incoherently on the hardware level. In a minimum configuration, the network only consists of one MIMO radar sensor and a repeater tag. The theory behind such a multistatic network is mathematically derived and simulations are presented to show key aspects of the network, i.e. multistatic range and Doppler measurements as well as high resolution angle estimation exploiting a very large virtual aperture spanning the whole network. Measurements with one sensor and one repeater tag at 77 GHz are carried out to verify the simulations. The measurements show that the bistatic path between the sensor and repeater tag retains coherency., publishedVersion