standard handsets were able to simultaneously observe towers from the two or more GSM network
operators that often compete in an area.
As is the case for 802.11 systems, the key drawback of GSM fingerprinting systems is the
need to collect a large number of measurements to train the system. One way to ameliorate this
problem is to use a radio propagation model to derive a detailed radio map from a much smaller set
of measurements. Chen et al. [23] developed a system that models signal propagation using Gauss-
ian Processes, which are nonparametric models that estimate Gaussian distributions over functions
based on the training data. Because the Gaussian processes algorithm models signal strengths with
continuous functions, this technique works better in more open environments, such as residential
areas. In high-density downtown environments, however, the technique fails to capture the sharp
changes in signal strengths due to obstructions. Similarly, Zhu and Durgin [179] developed a sys-
tem that uses the Hata signal propagation model to generate the radio map.
Standards for the deployment of location technologies by cellular providers have been published by
the Third Generation Partnership Project (3GPP) and the 3GPP2, which concentrate on the GSM
family of protocols (including W-CDMA) and CDMA, respectively [178]. These specifications
call for the implementation of three location services: Cell-ID, TDOA, and A-GPS. Each of these
types of location services strikes a different trade-off between ease of deployment, coverage, and
accuracy. Cell ID provides the lowest accuracy, but because it does not require additional hardware
on the network side, it is the one method that is universally supported by all network providers.
TDOA provides higher accuracy, but because it requires changes to the network, the handset, or
both, it is not yet universally supported. A-GPS provides accuracy in the tens of meters, but is still
only available on a small fraction of mobile handsets [57].
A-GPS and TDOA complement each other to a large extent. A-GPS is expected to perform
best in rural and suburban environments where few obstructions provide for good satellite vis-
ibility. In this environment, however, low base station density leads to poor TDOA performance.
Conversely, in downtowns and other dense urban environments, TDOA has higher coverage than
A-GPS, as it works (albeit at reduced accuracy) indoors and in other obstacle-rich environments.
Whereas location systems based on Cell ID, TDOA, and A-GPS are commercially available,
mobile phone fingerprinting systems are still at the stage of experimental prototypes. Radio fin-
gerprinting systems can achieve high accuracy, but the need to collect training measurements limit
their coverage. While these systems do not require additional hardware, the lack of a standardized
interface for accessing signal strength information represents an important obstacle to their com-
mercial deployment. Figure 5.5 shows a table summarizing the performance characteristics of these
four approaches to cell-based location estimation.

Get Location Systems now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.