Estimation of User Velocity for Mobility Management in Heterogeneous Networks

Abstract

The rapid growth of bandwidth-hungry applications being installed in current generation of mobile digital gadgets have generated an exponential increase in demand for cellular data-traffic. This trend is foreseen to continue in future due to the emergence of applications like Internet-of-things (IoT), Cyber physical systems, etc. However, due to limited radio spectrum the only promising way to keep pace with the future demand is through aggressive spatial reuse of available spectrum. One way to realize spatial reuse is by densification in network deployment. The deployment of small cell Base-stations (BSs) inside macro-cell network is a viable solution to address the increasing traffic demand, and such cellular networks are known as Heterogeneous-networks (HetNets). More specifically, HetNets is a combination of arbitrary and non-arbitrarily deployed BSs like, picocells, femtocells, etc. which are supported by diverse radio access technologies with hierarchical power levels. Due to promising architecture, HetNets aims to provide a major contribution towards achieving the ambitious 1000-fold capacity improvement for next-generation 5G cellular networks. On flip-side, due to BS densification mobile users on several occasions faces unnecessary handovers and service failures, which makes mobility management a critical challenge in HetNets. Since user velocity plays a crucial role in handover process, its knowledge is imperative for effective mobility management. In addition, due to limited capacity of batteries in mobile devices, more emphasis is given to the velocity estimation techniques that takes place at the service provider end. The primary objective of this doctoral work is to investigate and evaluate the methods that can be used to estimate the user velocity at service provider end while serving as a guideline for future research endeavours.

Long-term-evolution (LTE) specification standardizes the mobility state estimation and detection via handover-count measurements. Hence, to address the mobility management issue, we derive the velocity estimator on the basis of handover-count and sojourn-time measurements available at service provider end. The velocity estimation techniques presented in this work can be broadly classified as, (i) classical estimators; and (ii) Bayesian estimators. Considering classical estimation technique Maximum-likelihood (ML) and Minimum-variance-unbiased (MVU) estimator are proposed on the basis of measurement data. The velocity in these estimation techniques is assumed to be deterministic but unknown constant. The proposed ML estimators are observed to be asymptotically unbiased and efficient. In our analysis, we also notice that sojourn-time based velocity estimator is more accurate in comparison to handover-count based estimator as both handover-count and sojourn-time information are accommodated in statistics of sojourn-time measurements. Considering practical limitation in availability of large records of measurement, MVU estimation is proposed in which we observe tight closeness of its variance with Cramer-rao-lower-bound (CRLB). Our numerical analysis illustrates that the proposed classical estimators are more accurate (i) in hyper-dense network, and (ii) with increase in time-span for collecting measurements.

Next, considering availability of prior information about the user velocity, we investigate Bayesian approach in which we propose Minimum-mean-square-error (MMSE) and Maximum-a-posterior (MAP) estimates of the user velocity. The proposed MMSE estimator exploits location based velocity information in the form of prior density function and along with recent handover count measurements minimizes Mean-square-error (MSE) in estimated velocity. Since MMSE estimator is difficult to derive for other than prior Gaussian density functions, we propose MAP estimator in which we consider Next-generation-Simulation (NGSIM) program gathered velocity dataset as prior information. Here, we have approximated the probability density curve obtained from dataset to following best fitted standard distributions: (i) Cauchy, (ii) Gamma, (iii) Gaussian, (iv) Laplacian double exponential. Considering these standard distributions as prior density function, we have derived corresponding MAP estimates and their Cramer-rao-bounds (CRB). Our observation in numerical analysis illustrate that for large measurements records there is very small difference between CRB and estimator variance and thus proposed estimator can be identified as asymptotically efficient. We also observe that the proposed Bayesian estimators outperforms classical estimators based solely on measurement data as appropriate consideration of prior information reduces the uncertainty about user velocity.

Finally, we proposed an efficient mobility management strategy in which estimated velocity is exploited to predict the sojourn time of upcoming BSs and reduces the unnecessary handovers and handoff failures. Here, we have also designed a cost function based approach in which we consider various Quality-of-service (QoS) parameters like, available power, bandwidth, security etc. in final selection of the target BS for handover completion. We validated our approach via simulation where proposed scheme outperforms the fixed Received-signal-strength (RSS) based approach by reducing the unnecessary handoff. Further, we also proposed the user mobility state detection technique using sojourn-time measurement and compare it with that of previously proposed handover-count based detection. We observed that our proposed sojourn-time based approach is more accurate, and its detection performance improves with increase in BS density.