Title: EAGER: Predictive Micro Mobility Management in mmWave Cellular Networks, funded by NSF under grant CNS-1837034, 10/01/2018 – 09/30/2021.

Project Summary:

As a means to address the inadequacy of cellular spectrum under 6 GHz, extending cellular communication into the mmWave bands (between 30 and 300 GHz) has received enormous interest recently. Unlike its microwave counterparts below 6 GHz, the mmWave communication is directive and heavily relies on the communication line of sight (LOS), which may be frequently blocked (with a significant probability of 20% to 60%) by obstacles when the user moves locally or in small scale (a.k.a. micro mobility), and hence result in frequent outages (sudden loss of the received signal). The resultant intermittent connection has a devastating effect on the performance of higher layers, e.g., may cause over 10-fold degradation of the TCP throughput. Therefore, unless being addressed properly, user’s micro mobility is the bottleneck to mmWave cellular networks. Although one may switch to a non-LOS (NLOS) path component to continue the communication when the LOS is blocked, the NLOS transmission is either orders of magnitude slower than the rate of the LOS if the same transmission power is used, or consumes much more energy if a comparable transmission rate is retained, because, as revealed by the recent field measurements of NYU, in outdoor communications the strength of a NLOS is in general 20 to 30 dB weaker than that of the LOS for all typical mmWave cellular frequencies (28, 38, and 73 GHz). The overarching goal of this project is to systematically study a suite of cross-layer, seamless, and efficient micro mobility management models, algorithms, and protocols that can minimize the impact of intermittent LOS-blockage on the performance of upper layers (transport layer and above) at the minimum cost of network resources. 

Different from the existing proactive and reactive methods, the main novelty of this project lies in the development of a new class of predictive link/network-layer designs that enables reaction to an outage forehand (before the outage happens), overcoming the weaknesses of existing proactive/reactive methods. The intellectual merits of this project are two folds: (1) The PI proposes an outage prediction mechanism that can accurately predict when the LOS component in a mmWave channel will be blocked and how long the blockage will last in realistic multi-obstacle outdoor environment. This mechanism is achieved by realtime sensing of the geometry of the mmWave multipath channel and by detecting the blockage of peripheral NLOS components. (2) Taking advantage of the predicted blockage information, the PI proposes a multi-scale just-in-time (JIT) predictive outage handling framework. For outage of short durations, the PI proposes cross-layer predictive link-layer designs to hide outage from upper layers. For outage of long durations, the PI proposes a JIT optimal-stopping sequential handover strategy that can maximize the handover efficiency and QoS while accounting for the handover overhead and deadline. The proposed solutions will be thoroughly evaluated using simulations and real mmWave testbeds. 

Project Goals:

This project has the following two major goals:

· A sensing-based LOS blockage prediction mechanism – The PI will develop an outage prediction mechanism that can accurately predict when the LOS component in a mmWave channel will be blocked and how long the blockage will last in realistic multi-obstacle outdoor environment. These predictions are based on the realtime sensing of the geometry of the mmWave multipath channel and by detecting the blockage of peripheral NLOS components. These NLOS components may not be strong enough to be used for high-speed communication, but they are typically strong enough to be detectable.

· A multi-scale just-in-time predictive outage handling framework – The above prediction enables not only forehand, just-in-time (JIT), outage handling but also differentiated handling of outages based on their time scales (durations). For outage of short duration, handover is unnecessary. In this case, the PI will investigate novel cross-layer predictive link-layer designs, including scheduling and traffic shaping that exploit the predicted blockage information to make the outage transparent to upper layers. For outage of long duration, the PI will propose a JIT optimal-stopping sequential handover strategy that can maximize the handover efficiency and QoS while accounting for the handover overhead and deadline.

Project Personnel

PI: Tao Shu, Ph.D.

Graduate Students

· Li Sun

· Jing Hou

· Xueyang Hu

Project Activities and Results

1. Fast and high-resolution non line-of-sight (NLOS) beam switching over commercial off-the-shelf (COTS) mmWave devices

The high frequency and high directionality of mmWave communication make its line-of-sight (LOS) path susceptible to blockage. When LOS is blocked, a good solution is to promptly steer the antenna beam towards a strong NLOS signal propagation path to maintain the communication. To find such a strong NLOS path, existing methods in the literature rely on resolving the angle of arrivals (AoAs) and angle of departures (AoDs) of the mmWave multipath channel, which are computed by running sophisticated array signal processing algorithms such as MUSIC over phased array antennas. Because under these algorithms the spatial resolution of the multi-path channel resolving outcome is bounded by the beam width of the phased array antenna or by the measurement resolution of the minimum phase difference at different antenna elements, paths that are not separated far apart would not be distinguishable when these algorithms are used over current COTS mmWave devices, which are commonly equipped with only coarse-grained wide-beam antennas (e.g., a quasi-omnidirectional antenna for reception, and a 60◦-beam width phased array antenna for transmission). These consumer-level antennas also lack the capability of accurately measuring the phase difference at different antenna elements. Therefore, when the existing spatial channel resolution algorithms are used on COTS mmWave devices, the low-resolution multi-path channel resolving outcomes would fail to identify the exact direction of the strong NLOS path. For this reason, today’s COTS mmWave products do not provide the capability of prompt NLOS beam switching in response to LOS blockage. When LOS is blocked, there are usually two options: keep the current LOS path or exhaustively scan for an alternative NLOS path. The former would result in a degraded throughput during outage, while the latter would bring long switching time and a degraded link stability.

Keeping the above limitation of existing methods in mind, in this research we are interested in developing a new method to support high-resolution mmWave multipath channel resolving based on coarse-grained wide-beam phased array antennas that are commonly equipped on today’s COTS mmWave devices. Based on this new method, we further propose a computation-based beam switching algorithm that can directly predict a strong NLOS path whenever LOS blockage happens and a strong NLOS backup path is needed.

The following accomplishments have been achieved:

(1) We have developed a new method to support high-resolution mmWave multipath channel resolving based on coarse-grained wide-beam phased array antennas. In particular, we propose to perform fine-grained spatial scanning of the antenna array and exploit the high-spatial-resolution differential received signal strength (RSS) information measured when the antenna array is pointed at different directions with small steps. The key insight here is that the wide beam-width of the antenna array does not prevent the array from scanning the space in a fine resolution (e.g., with a step of 1◦ increment in the direction of the antenna beam). The differential RSS information associated with the spatial scanning process, which naturally has a high spatial resolution (e.g., in a resolution of 1◦), is then exploited by a novel two-step multi-path channel resolving algorithm. In particular, a low-resolution out-lobe resolving step is first performed to identify the clusters of paths that are separated more than the beam width of the antenna array. Then, for each cluster, a high-resolution in-lobe resolving step is performed, which utilizes reverse engineering to compute the optimal in-cluster fine-grained paths that offer the closest match with the measured differential RSS of that cluster.

(2) Based on the above channel resolving process, we further developed a prediction-based fast NLOS beam switching algorithm. In particular, our method consists of two phases: the offline site survey phase and the online operational phase. In the site survey phase, our model aims to construct a reflector map by estimating the locations and reflection coefficients of the dominant reflectors in the environment, through a sequence of coordinated differential RSS measurements at multiple locations. At each location, the above channel resolving process is called to compute the top-K strong NLOS paths generated by the dominant reflectors. Exploiting the sparse nature of the mmWave channel, the NLOS paths computed at different locations are then used to estimate the location of the dominant reflectors by a specular reflection model. Furthermore, based on Fresnel reflection model assumption, the reflection coefficient of each dominant reflector is calculated by a minimum mean square error (MMSE) regression based on the RSS measurements. This reflector map is subsequently used in the operational phase to calculate the supposedly strongest NLOS path at the current location of the user. The main beam of the transmit antenna is then steered accordingly to maintain the ongoing connection when the LOS is blocked.

To verify the performance of the proposed method, we implemented our algorithm on a commercial off-the-shelf mmWave device MikroTik WAP 60G transceiver set. A HTC VIVE VR system is used as our localization anchor, and a 6-axis robot arm is used to perform mechanical beam steering. A picture of our testbed is shown in Figure 1 below. The system was tested in a lab environment that has multiple dominant reflectors for both static and mobile applications. The measurement shows that our system is able to accurately estimate the locations of the top 2 reflectors in the test environment. In case of LOS blockage, by steering the transmit antenna towards the directions indicated by the proposed beam-switching algorithm, the system is able to achieve a 200% to 300% throughput gain over the case that the proposed algorithm is disabled so that the transmit antenna is always pointing to the LOS direction. Our received signal strength indicator (RSSI) measurement at the physical layer also shows that the link is more stable under the proposed algorithm. A paper describing this research is under revision and will be submitted shortly:

Xueyang Hu, Tao Shu, and Tian Liu, “Fast and high-resolution NLOS beam switching over commercial off-the-shelf mmWave devices,” under revision, to be submitted.

.

Figure 1. mmWave testbed based on off-the-shelf devices.

2.  Spatial and temporal context-based handovers in ultra-dense mmWave cellular networks

Although handovers are frequent in ultra-dense mmWave cellular systems, it has been shown that about 61% handovers made under existing handover algorithms are unnecessary or could have been avoided if the UE had made a better choice regarding which BS it should handover to. Such an over-frequent handover problem is mainly caused by the fact that conventional handover mechanisms are based on measurement of signal strength, and do not perform well in mmWave networks since it may cause “short-sighted” handover decision. For example, a base station (BS) with the highest signal strength would be chosen by conventional solutions as the handover target even if the line-of-sight (LOS) link associated with it will be lost in the next second after the handover, leading to another handover right after the current one. In this research, we aim at developing optimal handover policies that avoid the above unnecessary handovers by taking into account not only the current instantaneous state of the candidate BSs, but also the future change of the state, so as to reach a “far-sighted” handover decision. In particular, without prior knowledge of user’s mobility and the environment, the proposed handover mechanisms exploit the empirical distribution of user’s post-handover trajectory and LOS blockage, learned online through a multi-armed bandit (MAB) framework. The main novelty of our method lies in the construction of a new set of spatial and temporal features, which are computed based on user’s received signal strength (RSS) in the signal space rather than on the user’s actual location coordinates in the Euclidian space. This new set of features enables the proposed MAB to learn both spatial and spatial-temporal contexts of users’ mobility and the environment without requiring any exact location information of the users. Comprehensive performance evaluation are performed based on computer simulations. The results demonstrate that the proposed contextual handover mechanisms significantly outperform existing counterparts on reducing handovers in all simulated scenarios..

This research has led to the publication of two papers, one in IEEE Globecom 2019 Conference, and the other on the IEEE Transactions on Mobile Computing:

Li Sun, Jing Hou, and Tao Shu, “Optimal handover policy for mmWave cellular networks: A multi-armed bandit approach, ” Proc. of the 2019 IEEE GLOBECOM, Waikoloa, HI, USA, Dec. 2019, doi: 10.1109/GLOBECOM38437.2019.9014079.

Li Sun, Jing Hou, and Tao Shu, “Spatial and temporal contextual multi-armed bandit handovers in ultra-dense mmWave cellular networks,” accepted by IEEE Transactions on Mobile Computing, to appear, doi: 10.1109/TMC.2020.3000189.

Broader Impacts

Frequent outage caused by user’s micro mobility is not only the bottleneck to the performance of mmWave cellular networks, but also one of the main factors that limit the feasibility of mmWave cellular communication in many situations. If successful, this project will provide the much needed solution to this fundamental issue, and hence pave the way towards a feasible and efficient cellular network in the mmWave bands. The research outcome will increase the per-user wireless capacity by at least two orders of magnitude, and will create a profound positive impact on the nation’s economy. The research will also generate new insights for predictive network design, an area under-explored before. To broaden its impacts, this proposal will also carry out a comprehensive education plan, with a special emphasis on underrepresented and minority groups.

The following activities have been taken to broaden the impacts of this project so far:

· The PI has presented part of the research outcomes at the 2018 semi-annual meeting of NSF I/UCRC Center for Fiber Wireless Integration and Networking (FiWIN) for Heterogeneous Mobile Data Communications in Nov., 2018.

· Part of the research outcomes are also integrated with the networking and security courses the PI is teaching at Auburn University, including COMP 4320 (Introduction to Computer Networks), COMP 5320/6320/6326 (Design and Analysis of Computer Networks), and COMP 7370/7376 (Advanced Computer and Network Security).

· This project was introduced to over 1000 high-school students and their parents during the 2019 Open House Engineering Day (E-day) of the Samuel Ginn College of Engineering at Auburn University. This helps to foster the high-school students' interests in taking science and technology as their future career.

· This project was introduced to over 1700 high-school students and their parents during the 2020 Open House Engineering Day (E-day) of the Samuel Ginn College of Engineering at Auburn University. This helps to foster the high-school students' interests in taking science and technology as their future career.

· One of the Ph.D. students supported by this project, Jing Hou, is a female. Therefore this project has helped to increase the diversity in the STEM areas and promote women in engineering.