Range-Free Localization with Multidimensional Scaling for Dense NB-IoT Networks in 5G
Emanuel Staudinger, Michael Walter, Armin Dammann, German Aerospace Center (DLR), Germany
Alternate Number 1
Localization is next to communication the most important field of applications for wireless radio transmissions. Global navigation satellite systems (GNSSs) deliver very good location estimates under optimum conditions. It can be achieved by ranging, i.e., the determination of the propagation distance of the radio wave from a transmitter at a known location to the receiver. GNSS localization performance is severely degraded indoors and in urban canyons as satellite visibility is limited and strong multipath is immanent. Terrestrial radio such as 4G and the upcoming 5G cellular system can complement GNSS based localization in various ways: based on cell ID, ranging, and angular information. Localization based on cell ID is not very accurate as only the base station is identified by the mobile user, the cell size is typically in the order of hundreds of meters, and neighboring mobile users are not discovered. Triangulation requires sophisticated antenna array at the base station or at the mobile radio device at high cost. Hence, trilateration based on ranging is most commonly used. The distance between a mobile radio device and the base station can be determined via propagation delay or received signal strength. However, strong multipath limits the achievable ranging accuracy. Moreover, low-power devices envisaged for the Internet of things (IoT) in 5G will not be able to process large signal bandwidths and apply sophisticated multipath estimators. The localization support of 4G cellular narrowband technologies is under study in Release 14 with a signal bandwidth of 1.08 MHz for machine-type communications and 180 kHz of bandwidth for narrowband (NB) IoT transmission. Taking these low bandwidths into account it is obvious that accurate indoor localization based on ranging is very challenging.
Cellular IoT solutions have been standardized to cope with the growing demand of connectivity from low-cost and low-energy devices. 5G will enable networks of massively connected devices and additionally device to device (D2D) connectivity. Those two aspects enable so called cooperative positioning techniques: the distance between each mobile user and the base station is determined via propagation delay based techniques or received signal strength. The location of each mobile user can then be estimated centrally or in distributed fashion. Typically, estimators in a Bayesian framework are used. Another class of estimators is based on multidimensional scaling (MDS) and its representation of similarity or dissimilarity among pairs of mobile users as distances between them in a multidimensional space. For ranging based localization the similarity is expressed as Euclidean distance matrix (EDM). One advantage of MDS is that the similarity can be defined arbitrarily, i.e., by representing connectivity only. Hence, we are interested how well IoTs with little bandwidth can be localized by means of connectivity only. Hence, range-free localization becomes possible with MDS but its performance with depend on various aspects, such as network density and communication range.
In this paper we investigate the achievable range-free localization performance for static IoT devices in an indoor environment based on MDS. The paper organization and key contributions are therefore as follows:
1) An introduction of MDS and EDMs for localization and its adaptation to range-free localization.
2) The first evaluation will focus on a realistic indoor environment with distributed narrowband IoT radio devices and the achievable localization accuracy. Furthermore we will show, how range-free localization performance depends on network density and communication range. As a result we can show which density is theoretically required to achieve the localization accuracy required for narrowband IoTs.
3) In a second evaluation we assume propagation delay based ranging under realistic indoor channel conditions based on the WINNER-II channel model and estimate the location with MDS. This enables a comparison between propagation delay based localization with range-free localization, which to the knowledge of the authors is not available in literature so far in this extent. As a result we can show exactly at which network density and also signal bandwidth we benefit either from range-free or range-based localization.