The focus of this thesis is to improve the current mmWave technologies, particularly this thesis focuses on improving mmWave over TCP, initial access, beam training, and multiband location systems. In Chapter 2 we give an in depth introduction to the new IEEE 802.11ad mechanisms and their challenges. The intrinsic characteristics of mmWave set it apart from sub-6 GHz technologies, it has very high bandwidth but the link can degrade very quickly due to blockages or even misalignments. Therefore, we need to understand how current technologies perform under these new characteristics.

In Chapter 3 we look at how mmWave networks perform by evaluating the behavior of the most common transport protocol, TCP. TCP is a legacy technology that gradually speeds up when there are no errors on the transmission and reacts drastically to congestions. From the point of view of TCP, a mmWave suffers from congestion problems when there is a blockage or link degradation, but usually mmWave can recover very quickly, although TCP protocols take some time to ramp up the speed. We do a thorough evaluation of different TCP protocols and scenarios and show the required parameters to improve the performance. We show that there is a need for better parameters or even new congestion protocols that are designed with the characteristics of mmWave in mind. This is only just one of the challenges of mmWave, its directionality makes device discovery and initial access a very complex process. Usually, new devices make use of multiband systems, such as 2.4 and 5 GHz in sub-6 GHz systems. This makes it possible to use out-of-band information to reduce the initial access overhead.

In Chapter 4 we look into this possibility. First we do an exhaustive study of Low Frequency (LF) and High Frequency (HF) channel correlation. We consider 2 sub-6 GHz bands and 2 mmWave bands, we then find the sub-6 GHZ and mmWave combination that offers the highest correlation and then we use LF information to implement better initial access mechanism for mmWave bands. Once we have looked at the initial access, another challenge created by the directionality of mmWave communications is beam training. This mechanism consist on searching for the best combination of beams for a pair of devices. To do so, the Access Point (AP) sends beacons over each one of its sectors (between 32 and 64 in Commercial-Off-The-Shelve (COTS) devices) while the Station (STA) is listening omnidirectionally. Then they switch roles and the STA sends beacons over each one of its sectors while the AP listens omnidirectionally. Due to this complexity the mechanisms to align beams with new devices take most of the time of the connection, which otherwise could be use for communication.

In Chapter 5 we introduce a system that allows to reduce that overhead. We present mm-FLEX, an open, flexible mmWave platform. This platform allows us to implement Ultra Fast Alignment (UFA), an algorithm that completely removes the STA beam training, reducing the overhead on a factor of 8 for dense scenarios. Lastly, we again exploit sub-6 GHZ and mmWave multiband systems. Since we know there is a strong correlation we focus on improving existing mmWave location systems by using out-of-band information. Location systems are important for higher frequencies because it allows better tracking of its users. This, in turn, helps the device discovery and beam training procedures. Knowing the location of the user in real time can even help with known blockages and prevent channel degradation.

In Chapter 6 we present a joint LF and HF location system. As part of this work we reverse engineer a MikroTik COTS device, and we enable Channel State Information (CSI) and Fine Timing Measurements (FTM) extraction. Using CSI and FTM data from the sub-6 GHz band in addition to the mmWave band we can overcome typical problems from a single band location system. Namely, we can improve the accuracy of the LF location system by using HF information when in Line Of Sight (LOS), and we can use LF information to improve the HF accuracy in Non Line Of Sight (NLOS) scenarios. Additionally, we do a comprehensive study on the number of antennas needed in HF to overcome the directionality limitations. Our system has an error of less than 17 cm of location error on median.

About Pablo Jiménez Mateo

Pablo received his BSc in Computational Mathematics and Computer Engineering from the Universitat Jaume I. Castellón de la Plana. He completed an MSc in Intelligent Systems at University Carlos III of Madrid, Spain. Pablo started his PhD at IMDEA Networks and UC3M in September 2016 and his research interests are machine learning, 5G, and mmWave.

Supervisor de tesis: Dr. Joerg Widmer (IMDEA Networks Institute, España)

Universidad: Universidad Carlos III de Madrid, España

Programa de doctorado: Ingeniería Telemática

Miembros del tribunal:

• Presidente: Matthias Hollick; PhD en Ingeniería Eléctrica y Tecnología de la Información por la Universidad Técnica de Darmstadt; profesor titular (W3), FB Informatik, TU Darmstadt
• Secretario: Vincenzo Mancuso; PhD en Ingeniería Eléctrica y Electrónica, Universidad de Palermo, Italia; Profesor asistente de investigación (titular) en IMDEA Networks
• Miembros del panel: Paolo Casari; PhD en Ingeniería de la Información por la Universidad de Padua, Italia; Profesor adjunto, Universidad de Trento, Italia