Resumen de Initial access and beam-steering mechanisms for mmwave wireless systems
Bandwidth in the 4G technology has run short for the vast number of devices and traffic expected for future communication. Therefore, engineers have turned their focus on the wide unlicensed spectrum between 3-300GHz more particularly towards the band around 60GHz. This one should be theoretically be able to support multiple GBps directional wireless connections and pack large antennas in small devices. Future millimeter-wave (mmWave) networks will support very high densities of devices and access points. Considering that communication links will be directional and will require a training to properly steer the radiation patter this will vastly increases the overhead required for access point selection and beam training. Furthermore, due to unfavorable radio propagation properties such as very high attenuation and short wave skin depth, mmWave systems will exploit largescale MIMO and adaptive antenna arrays at both the transmitter and receiver to realize sufficient link margin. On the other hand, the quasi geometrical properties of the channel will make directional communication feasible, making beamforming vital to overcome the high attenuation in wireless millimeter-wave networks by enabling nodes to steer their antennas in the direction of communication. Fortunately, the quasi-optical properties of millimeter-wave channels will also make location-based network optimization a highly promising technique to reduce control overhead in such millimeter-wave WLANs.
In this thesis we explore and present several tools to improve mmWave systems. The thesis begins with a small introduction in the first chapter.
The synthesis of sector beam patterns with antenna arrays is a widely investigated topic in the literature because of its myriad of applications, ranging from massive multiple-input multiple-output (MIMO) to cell sectorization in cellular networks. Very recently, the design of sector beam patterns has also received significant attention in the millimeter-wave (mmWave) context where the use of high-gain, adaptive antenna arrays with configurable beamwidth is essential to cope with the higher propagation loss and unfavorable atmospheric absorption at mmWave frequencies. We start in the second chapter by designing a formula to compute the antenna radiation properties we want to aim for as the complex antenna weight coefficients that generate a sector shaped beam patterns able to radiate in a range of directions. This is computed in a closed form solution to achieve an extremely lightweight implementation. The formula is based in complex numbers analysis to reduce the problem to an equivalent one with real numbers that can then be solved with harmonic analysis. The proposed strategy is more effective than classical approaches, not only in terms of ability to shape sector beam patterns which better comply with a desired mask, but also in terms of reduced computational complexity. This works is then used in our posterior research whenever a sector beam-pattern is required.
The higher propagation loss and unfavorable atmospheric absorption make data transmission over relatively long distances a serious challenge at mmWaves. The short wavelength, however, allows more antenna elements to be integrated into mmWave devices and enables large-scale multiple-input multiple-output (MIMO) and adaptive antenna arrays to overcome range limitations. Traditional MIMO processing as performed in sub-6 GHz systems is, at present, impractical at mmWaves because the high cost, power consumption, and complexity of mixed-signal hardware at such frequencies prevent the use of a dedicated RF chain per antenna element, this is why instead the proposed structure of mmWave antennas is the hybrid analog-digital structure. Due this, in the third chapter we start by dealing with the complications of mmWave hybrid analog-digital transceivers and propose an algorithm to approximate any desired baseline beam-pattern making use of this structure. The hybrid analog-digital structure generally consists on RF-chains connected to multiple (or all) antennas through phase shifters, thus the only options to modify the complex antenna weight coefficients is to select the proper analog configuration for the different phase shifters connected to each RF-chain and the digital configuration corresponding to the coefficient that reaches each RF-chain. This results in a beam-pattern product of quantized power constrained analog beam-patterns defined by the RF-chains with a less constrained transformation that can be applied to the signal in the digital domain corresponding to each RF-chain. To tackle this, we develop a greedy geometric approach based on geometric intuition and a matching pursuit algorithm structure to be able to synthesize parallel beam-patterns make use of this hybrid analog-digital structure. This algorithm does not constrain the beam-pattern’s shape, thus multiple different beam-patterns can be leveraged for simultaneous multi-direction scanning, allowing for a system to be more efficient in terms of synthesizing/gathering information to improve communication. Then we make use of this multi-direction scanning to create a beam training protocol based in bisection search to measure sectors with possible directions for communication paths and narrows them to refine the search and obtain them. By doing this we effectively accelerate the link establishment by exploiting the ability of users to scan multiple ranges of directions at the same time. Then, by learning the appropriate directions for communication we can tune the antenna to exploit them, increasing the communication rate. This resulted in a speed up of the beam training phase, in turn enabled by the multi-beam characteristics of our hybrid codebooks, provided a 25% to 70% increase in spectral efficiency compared to state-of-the-art strategies that adopt sequential, single-direction scanning during beam training.
Highly directional communications complicate the link establishment and maintenance between an Access Point (AP) and a User Equipment (UE). In fact, AP and UE must perform a time-consuming beam training procedure to determine the best directions of transmission/reception, which incurs significant overhead (and waste of network resources). The problem is exacerbated in scenarios with mobility, since even a slight beam misalignment or environmental changes, such as link blockage, device rotation, etc., can cause considerable signal drop. To sum up, fast and efficient beam training/tracking strategies are of paramount importance to maintain seamless connectivity in a mm-wave network with node mobility. In the fourth chapter we propose smart beam training and tracking strategies for fast mm-wave link establishment and maintenance under node mobility. We achieve this by leveraging the ability of hybrid analog-digital transceivers to collect channel information from multiple spatial directions simultaneously and formulate a probabilistic optimization problem to model the temporal evolution of the mm-wave channel under mobility, this one makes use of the learnt channel information and evolution models to improve the beam-training process in which the devices need to learn about the channel before they can communicate. We do this by combining smart beam-pattern designs in order to reduce the search space based in constraints that guarantee a full spectrum search while ensuring their perfect design with a hybrid analog-digital structure and the extra information generated by the same hybrid analog-digital structure when beam-forming is a desired direction combining several RF-chains. This information is then combined with previous information of the channel skeleton consisting on directions and powers of different geometric paths and fed to an evolution probabilistic model that allows us to consider a maximum likelihood estimation of the new state of the channel skeleton without increasing too much the required number of measurements. Simulation results, obtained by a custom simulator based on ray-tracing channel modeling, demonstrated that the proposed solution is effective to keep the average communication rate only 10% below the optimal bound. Compared to both the IEEE 802.11ad standard and the state of the art, our solution provides a 40% to 150% performance increase while at the same time using lower complexity hardware.
Beam training, also called SLS in infrastructure-based 802.11ad networks, works as follows. The Access Point (AP) transmits beacon messages using each of its available beam patterns sequentially, while the station (STA) listens with a quasi-omnidirectional beam pattern. After that, the STA repeats the same process but includes in each of its messages the identifier of the beam pattern that it received best from the AP. Finally, the AP replies with the identifier of the best beam pattern of the STA in a dedicated control message. While this mechanism is straightforward, it clearly does not exploit the full potential of the antenna array of IEEE 802.11ad devices. First, none of the available beam patterns in the codebook may steer exactly towards the receiver. Second, strong reflectors in the environment may remain unused even though they could contribute to the received signal strength. Third, existing reflections may result in destructive interference at the receiver, causing significant harm to the communication. If devices were to adapt their beam patterns to the specific environment in which they operate, they could easily mitigate the above issues. However, this often requires full Channel State Information (CSI) at the transmitter, which is particularly challenging to obtain in millimeter-wave systems. The reasons are twofold. First, the feedback overhead of full CSI is significant. Second, devices must probe enough orthogonal beam patterns to cover all the dimensions of the channel. The channel information extracted from all the probes must be coherent in phase. Channel state information is the main source of information when it comes to beamforming as some of the most common techniques like eigen beam-forming or spare path decomposition can be applied. Then, in order to be able to apply all these techniques to commercial devices without full information in the fifth chapter we propose a mechanism to extract full channel state information (CSI) regarding phase and magnitude from coarse signal strength readings on off-the-shelf IEEE 802.11ad devices, the requirements for this are simple signal magnitude measurements such as SNR or RSS together with mathematical tools based in harmonic analysis to the extraction of the desired channel properties. The way it works is by first obtaining a reference beampattern that we can obtain from previous beam training, then we cycle over several phase shift values of a given antenna while measuring the signal strength on those different cases and by combining the measurements we are able to retrieve the module and phase shift of that antennas. The way they are obtained is through complex number and Fourier analysis, we find a relation between the discrete Fourier transform coefficients and by inverting it we can retrieve the original desired information. Then we design a beam-training method (ACO) that directly makes use of this technique to extract CSI without incurring in unnecessary measurements. We implement ACO on TP-Link Talon AD 7200 tri-band routers by obtaining full access to the beamforming control of the integrated 32-element phased antenna array. To this end, we disassembled the phased antenna array and reconstructed the antenna weighting network experimentally. Our evaluations in a real-world office environment show that ACO increases the SNR by factor 2.5 and achieves a 2x higher TCP throughput. To support the community and allow other researchers to benefit from our results, we make our framework and firmware patches as well as the source code of our actual implementation publicly available.
Due the fact that channel properties and antenna design at 60GHz are ideal for path angular information extraction, following an almost ideal geometric channel model we present in the sixth chapter some localization methods specifically designed for the 60GHz band, these are based on angle difference of arrival information and due their ability to estimate the surrounding environment require no input being ideal for applications that need working without a proper calibration such as navigation in an unknown environment such as an office or virtual reality fast deployment. These location algorithms also aim to be very useful for channel estimation and prediction since the channel properties at 60GHz resemble a sparse geometric channel. The location algorithms presented are two, and they both must estimate two sets of variables: the access points locations and the user locations over time. The first one is base on a variable extension to linearizes an inverted geometry equivalent problem when optimizing by projections, then it refines the solution by making use of these linear cut optimization. In simple words, we extend the number of variables to simplify the formulation and then while considering one set of variables constant we find the optimum solution for the other set of variables and iterate over the two sets of variables. The second one finds a polynomic trigonometric relationship between two sets of variables: the measured angle differences and the angles the access points form. This allows us to write the problem in a quadratic form with coefficients computed from the measured angle differences and variables being a polynomial trigonometric transformation of the angles between access points. Thus we minimize the quadratic form for the variables computed from a set of access points locations in order to obtain the access points location solution to our problem and then apply angle difference of arrival techniques to estimate the user location over time given the already compute access points location. The advantage of this method is that increasing the complexity of the problem by adding more measurements does not increase the complexity of the formulation since the quadratic form simply gets updated by new measurements.
As already mentioned location can be used to speed up the beam training and that’s exactly what we do in the seventh and last chapter, we merge the ideas presented in this thesis and by extracting channel state information from off-the-shelf routers we estimate the user location to manage a location aware beam-training and device handling method. Here we design a location system to work along a beam training protocol. The location system is fed with angular information provided by the beam training system to compute the location of the access points by making use of a formulation that gets reduced to an MMSE problem. Once the access points are located the user location tracked. The tracking algorithm is a triangulation-based filter in which an evolution (movement) model is assumed, and the measured angles are used to compute the likelihood of a given location to later have a maximum “a posteriori” of the location. Then the coverage of the room is learnt through machine learning to have an estimate of the SNR per point in the room. To ease the learning process both cartesian coordinates of the estimated location and estimated distance to the access points is fed to help creating the connections between linear wall blockage and fading due to distance, respectively. By making use of the evolution model we make a prediction of the immediate future set of locations and take a decision on which access point to connect to based of available SNR and also reduce the beam training searching space to the range in which the access point location has been estimated. This optimizes access point association and selects the most suitable antenna beampatterns while significantly reducing the beam training overhead. By doing this we can anticipate changes in the communication link and anticipate them aiming to maximize the average throughput in the immediate future by selecting the proper access point to communicate while narrowing the training overhead.
Overall, this thesis proposes a wide variety of methods to improve mmWave communications, these ones are: baseline beampattern design, hybrid analog-digital structure coding algorithm, several beam-training algorithms with and without mobility, a way to retrieve phase information from smartly designed amplitude (SNR or RSS related) measurements, several location algorithms aiming to different problem conditions and a final location aided environment learning beam training algorithm that condensates the previous ideas.
