Using Imprecise Location for Emergency Context Resolution
draft-ietf-ecrit-rough-loc-02.txt

Abstract

Emergency calling works best when precise location is available for emergency call routing. However, there are situations in which a location provider is unable or unwilling to provide precise location, yet still wishes to enable subscribers to make emergency calls. This document describes the level of location accuracy that providers must provide to enable emergency call routing. In addition, we descibe how emergency services and non-emergency services can be invoked by an endpoint that does not have access to its precise location.

Status of this Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as “work in progress.”

This Internet-Draft will expire on January 28, 2011.

Copyright Notice

Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

1.  Introduction
2.  Terminology
3.  Determining sufficient location precision
    3.1.  Location filtering
    3.2.  Constructing location filters
        3.2.1.  Civic address considerations
    3.3.  Maintaining location filters
    3.4.  Applying location filters
4.  Requesting emergency and non-emergency services
    4.1.  Emergency calling
    4.2.  Non-emergency services
5.  Acknowledgements
6.  Security Considerations
7.  IANA Considerations
8.  References
    8.1.  Normative References
    8.2.  Informative References
§  Authors' Addresses

1.  Introduction

Information about the location of an emergency caller is a critical input to the process of emergency call establshment. Endpoint location is used to determine which Public Safety Answering Point (PSAP) should be the destination of the call. (The entire emergency calling process is described in detail in [6] (Rosen, B., Schulzrinne, H., Polk, J., and A. Newton, “Framework for Emergency Calling using Internet Multimedia,” July 2010.) and [1] (Rosen, B. and J. Polk, “Best Current Practice for Communications Services in support of Emergency Calling,” July 2010.).) This process is most likely to work properly when the endpoint is provided with the most accurate and precise information available about its location. Using location information with maximal precision and accuracy minimizes the chance that a call will be mis-routed. In addition, when that location is provided to the endpoint, the endpoint is able to verify that the location is correct (to the extent of the endpoint's knowledge of its own location) prior to an emergency call, and is able to perform emergency call routing functions on its own, providing redundancy for network-provided functions.

However, there may be situations in which it is not feasible for endpoints to be provided with maximally precise and accurate location. These cases may arise when computing precise location is an expensive or time-consuming operation (e.g., in the case of wireless triangulation), and location is needed quickly, as is often the case in emergency situations. Or they may arise because the policy of a location provider does not allow precise location to be provided to the endpoint. While it is undesirable to use imprecise location for emergency call routing, the possibility that precise location may not be available to the calling device must be accomodated in order to make emergency calling possible in the largest possible set of circumstances.

This document is concerned with imprecise location only in the context of routing emergency calls, i.e., for determining the correct PSAP to receive a given call (e.g., via a LoST query [2] (Hardie, T., Newton, A., Schulzrinne, H., and H. Tschofenig, “LoST: A Location-to-Service Translation Protocol,” August 2008.)). Depending on the the structure of the local emergency service network, the location information provided to the endpoint may also be used to route the call to an entity that is authorized to request precise location, e.g., an Emergency Services Routing Proxy. The requirements and processes described in this document are the same for both cases.

Location information may also be used in the emergency calling framework to direct the dispatch of emergency responders. This usage is treated separately from call routing for purposes of this document, and this document does not place requirements on the location provided for dispatch, although it should obviously be as precise as possible. The only provision for dispatch in this document is a recommendation that the location provider supply endpoints with a URI that can be used by a PSAP or other emergency authority to obtain a different location for use in dispatch, hopefully more precise than the one used for routing.

This document describes the use of imprecise location information in the emergency call routing system. Section 3 (Determining sufficient location precision) describes how location providers can determine the precision necessary to support emergency call routing, and how they can use this information to optimize location delivery. Section 4 (Requesting emergency and non-emergency services) describes how emergency calls are placed in such an environment, and how non-emergency services can be invoked when precise location is not available to the endpoint by value.

2.  Terminology

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [3] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

We consider in this document patterns of interaction as described in [6] (Rosen, B., Schulzrinne, H., Polk, J., and A. Newton, “Framework for Emergency Calling using Internet Multimedia,” July 2010.). The two main parties of interest are endpoints and location providers. Endpoints are hosts connected to the Internet that originate emergency calls in the emergency calling architecture, while location providers are entities that supply location information that is used for emergency calling. In addition, we will discuss how these parties interact with the LoST mapping infrastructure [7] (Schulzrinne, H., “Location-to-URL Mapping Architecture and Framework,” September 2009.), and with emergency and non-emergency location-based service providers.

For convenience, we say that location information, either in LoST queries or in service boundaries, is provided "in geodetic form" if it is provided in the "geodetic-2d" LoST location profile, and "in civic form" if it is provided in the "civic" profile.

3.  Determining sufficient location precision

A location provider wishing to provide location information usable for emergency call routing requires a mechanism for determining when a description of location (e.g., a polygon) is precise enough to be used for emergency call routing. This mechanism might be used to decide when to terminate a positioning process that converges over time, or to choose a polygon larger than the known location of the endpoint (in order to obscure the known location of the endpoint), while preserving the utility of the location for emergency call routing.

There are three basic requirements for a location to be usable for emergency call routing:

1. The location SHOULD be sufficiently precise that a LoST request with the location and any service URN will return a unique URI mapping value. This may not be possible in all cases, e.g., because of overlapping service boundaries creating areas with non-unique mappings, or because of positioning limitations that prevent sufficiently precise positioning.
2. When the location of the endpoint is known by the provider to greater precision than is being provided, the provided location MUST return the same mappings from LoST, for all service URNs, as the known location.
3. When the location of the endpoint is known by the provider to greater precision than is being provided, the provided location MUST contain the precise location (as a geographic subset).

These requirements lead naturally to the idea of a "location filter". A location filter is a collection of geographical regions satisfying the following criteria:

1. For any location value that is a subset of a filter region, a LoST request for any service will return a unique mapping result.
2. Any two locations within the same filter region receive the same LoST results for all services

Given a location filter, it is easy to determine when a given location value is sufficiently precise, or to create a less precise version of location that is still precise enough. Namely, a location value is precise enough when if fits within a given filter region, and any superset of a location value (e.g., a polygon containing a point) can be used as a less precise version of the location value as long as it still fits within the same filter region.

For example, a simple fuzzing algorithm that maintains sufficient precision for emergency services is to replace a given location value with the filter region that contains it. This way, the server can compute the filter off-line (as described below), then provision the location of each possible target by storing a pointer to the filter region that contains the target's location.

The remainder of this section discusses the concept of location filtering in more detail, and describes how a location server can construct and maintain a location filter based on information from the LoST mapping infrastructure.

3.1.  Location filtering

With each service-to-URI mapping, a LoST query provides a service boundary that represents the set of locations in which that mapping is valid. A consequence of this is that given a set of service boundaries for different services, the intersection of those service boundaries is the region in which two mappings are valid. If one service boundary corresponds to the area where "urn:service:sos.fire" is served by "sip:fire@example.com" and another maps "urn:service:sos.police" to "sip:police@example.com", then the intersection is the are where both of these mappings are valid ("urn:service:sos.fire" maps to "sip:fire@example.com" and "urn:service:sos.police" maps to "sip:police@example.com"). Outside that area, one or more of the mappings is invalid. So as was suggested above, the intersection of two service boundaries defines a set of mappings, and any two locations within that intersection are equivalent for the purpose of LoST mapping (i.e., emergency call routing).

               Service boundaries for individual services

             urn:service:sos.police    urn:service:sos.fire

                  +-------+                +-------+
                  | A     |                | C     |
                  |       +---+            |   +---+---+
                  |       |   |            | X |       |
                  +---+---+   |            +---+       |
                      |     B |                |     D |
                      +-------+                +-------+

                        |                        |
                        |                        |
                        +-----------+------------+
                                    |
                                    V

                            +-------+
                            | A,C   |
                            |   +---+
                            |   | +---+
                            +---+ |A,D| +---+
                                  +---+ |   |
                                    +---+   |
                                    |   B,D |
                                    +-------+

                     Resulting Location Filter Regions

The regions in a location filter can thus be constructed by taking intersections of service boundaries. Figure 1 (Generating a filter from service boundaries) shows a simple location filter: Starting with a set of four service boundaries for two different services. The filter that results from taking intersections of these boundaries has three regions:

1. A region where police calls are directed to A and fire calls are directed to C.
2. A region where police calls are directed to A and fire calls are directed to D.
3. A region where police calls are directed to B and fire calls are directed to D.

These regions satisfy the criteria for a location filter because each one has a unique set of mappings and those mappings are valid across the entire region. The service regions for B and C do not overlap -- there is no place where police calls go to B and fire calls to C -- so there is no (B,C) region.

More generally, a filter region is the intersection of the service boundaries for all services available within the region. A filter can used to determine whether a location is usable for emergency call routing in the following way:

1. The location SHOULD be contained in exactly one of the regions in the filter. This guarantees that LoST mappings are unique.
2. When the precise location of the endpoint is known, the provided location MUST be contained in the same region(s) of the filter as the known location. This guarantees that LoST queries with the provided location return the same results as those done with the known location.
3. When the precise location of the endpoint is known, the provided location MUST contain the precise location (as a geographic subset).

Filter regions can be deduced constructed from LoST mappings for a sample location by intersecting all the service boundaries for services available at that point. Figure 2 (Generating a filter region from a sample point) illustrates how the filter region containing the point X is the intersection of the service boundaries for police and fire services that serve X.

If the server also stores the lists of URN-URI mappings for each region, x then the filter can also be used as a cache for LoST mappings; the LoST mappings for a location are the mappings bound to the region(s) containing it.

      sos.police           sos.fire          sos.ambulance

   +-------+           +---------------+
   | A     |           |             B |
   |       |           |               |       +-------+
   |     X |           |     X         |       | X     |
   +-------+           +---------------+       |       |
                                               |     C |
                                               +-------+

           |                   |                   |
           |                   |                   |
           +-------------------+-------------------+
                               |
                               V

                       +-------+-------+
                       | A     |     B |
                       |   +-------+   |
                       |   | X |   |   |
                       +-------+-------+
                           |     C |
                           +-------+

                               |
                               |
                               V

                           +---+
                           | X |
                           +---+
                    Resulting filter region
              (police=>A, fire=>B, ambulance=>C)

When the location of the endpoint is known to more precision than the location provided to the endpoint, although any location meeting the two criteria above is equivalent to the known location for purposes of LoST, the provided location MUST contain the known location in order to avoid errors if the location is used for other purposes in the course of an emergency (e.g., if the location is provided to first responders for dispatch). This guarantee also allows the endpoint to do some course verification that the provided location is correct (in order to prevent very gross errors in routing). Thus, any location that (1) contains the known location and (2) is contained in the same filter region as the known location is allowable. Locations that also are contained in only one filter region are preferred. Adding randomness to the provided locations may have privacy benefits in some cases, as discussed in the security considerations below.

3.2.  Constructing location filters

For simplicity, we assume that the entity performing filtering will only be using the filter to test locations contained within a particular geographic "coverage area". In principle, this coverage area could be the entire world, but assuming a more limited coverage area allows for a filter to be built more quickly.

Given a coverage area and the ability to act as a LoST client, a location service provider can autonomously compute a location filter by constructing regions around sample points until it has a collection of filter regions that collectively cover its service region. (The process for an individual point is illustrated in Figure 2 (Generating a filter region from a sample point).)

In order to ensure that all services boundaries are taken into account, the server starts by issuing a <listServicesByLocation> query, and caching the list of services that it returns, along with the corresponding service list boundary [4] (Wolf, K., “LoST Service List Boundary Extension,” February 2010.). The server then samples points within that service list boundary, retrieving mappings with service boundaries for each service in the service list and intersecting the service boundaries to obtain a new filter region. In pseudocode, the algorithm is as follows:

Set FILTER = the empty set
While filter does not cover LS coverage area
    Choose a random uncovered point X in the LS coverage area
    Perform a LoST <listServicesByLocation> query for X
        Set SL = <serviceList> from LoST response
        Set SLB = <serviceListBoundary> from LoST response
	    If SLB is not provided, choose new point X and re-query
	    If more than 100 points X have been tried
	    	Set R = uncovered area
	    	Add R to FILTER
	    	END
    While filter does not cover SLB
        Choose a random uncovered point Y in SLB
        Set R = SLB
        For each service S in SL
            Perform a LoST <findService> query for Y and S
                Set SB = <serviceBoundary> from LoST response
            If SB is not provided, return an error
            Else set R = intersection( R, SB(S,Y) )
        Add R to FILTER

If the LoST servers have been provisioned properly then this algorithm will terminate successfully. If LoST mapping do not cover part of the service region, then the <serviceListBoundary> will not be returned, and the algorithm will give up after 100 queries. This limit on queries introduces some risk that a small covered area will be left out of the filter and marked as uncovered; if this is a concern, then the query limit can be increased.

Of course, if the location server operator has information about service boundaries through some channel other than LoST, then the LoST queries above can be replaced by queries to a local store of mapping information. The choice of random points can also be guided to ensure that all mapped areas are covered even if there are some uncovered areas. The location server can also cache service boundaries acquired during the algorithm to avoid unnecessary LoST queries.

3.2.1.  Civic address considerations

This algorithm actually results in two filters -- one for geodetic service boundaries and one for civic service boundaries -- since civic and geodetic boundaries cannot be directly compared or intersected. It is RECOMMENDED that location servers always compute a geodetic filter for use with emergency services, since the notion of civic service boundaries have some inherent ambiguity.

Indeed, the notion of intersection of civic service boundaries has some dependence on the jurisdiction within which the service boundaries are defined. Civic service boundaries are comprised of a set of <civicAddress> elements, each defining a set of civic addresses that are within the boundary, namely those that match the civic elements provided.

When computing the intersection of two civic service boundaries, any <civicAddress> elements that are shared between the two service boundaries MUST be included in the resulting intersection. When two <civicAddress> elements in the service boundaries being compared are different from each other, then their intersection must be computed according to local addressing standards.

Note that the resulting filter regions SHOULD still cover the location server's coverage area, i.e., there should be a filter region that contains every civic address within the coverage area. In particular, the server SHOULD NOT use a specific address to represent a filter region: Such an address would not include many points in the service region (i.e., it would not meet the third rules from the lists of rules above). If the server creates a PIDF-LO document describing a civic address that does not contain the precise location of the target, then it MUST set the 'method' element of the PIDF-LO it returns to value 'area-representative' registered in Section 7 (IANA Considerations).

3.3.  Maintaining location filters

As the LoST mappings that underlie the filter change, the filter will need to be updated. The entity maintaining the filter MUST obtain a new mapping for a region when an existing mapping expires. The service boundary from the new mapping is compared to the service boundary from the old mapping: If they are the same, then the filter need not be updated. If they differ, then regions in the filter that intersect either the old service boundary or the new service boundary will need to be recomputed. Note that since this operation only requires the server to determine if two service boundaries are identical, the server need only store a hash of the old boundary to which it can compare a hash of the new boundary.

3.4.  Applying location filters

After constructing a location filter, a location server can use it to optimize how it delivers location. When the location server is using a positioning algorithm that grows more accurate with time, the filter tells it how long to run the algorithm. Namely, the algorithm can be terminated when the estimated location (that is, an uncertainty region containing the target's location) is within one of the regions in the filter.

When the location provider knows the precise location of the caller, a location filter can also be used as a "location cache". That is, the location provider can simply look up which of the filter regions contains the caller's precise location and return that region as the caller's location, or some subset that contains the precise location.

This caching strategy allows an additional optimization in some cases: If the location server knows that the caller's precise location will be within the same region for a period of time, it can instruct the client not to re-query in that time. For instance, if the server is delivering location over HELD, then it can use the HTTP cache-control headers (e.g., Expires). However, the location server MUST NOT instruct the client to wait for longer than the current filter is valid; the expiry time of the location MUST be before the earliest expiry of a LoST mapping used in the filter.

4.  Requesting emergency and non-emergency services

When a location provider wishes to deliver endpoints location information that is below its maximum available precision while still supporting emergency calling, it MUST provide to the endpoint both a location (by value) that is sufficient for emergency call routing (as defined above) and a location reference (i.e., a URI) that can subsequently be used by authorized parties to obtain more precise information about the location of the endpoint. The endpoint then can then use both the location value and the location reference to request emergency services and other location-based services (LBS).

4.1.  Emergency calling

The overall procedure for placing an emergency call is identical to that described in [6] (Rosen, B., Schulzrinne, H., Polk, J., and A. Newton, “Framework for Emergency Calling using Internet Multimedia,” July 2010.). In particular, the endpoint requirements in Sections 8 and 9 of [1] (Rosen, B. and J. Polk, “Best Current Practice for Communications Services in support of Emergency Calling,” July 2010.) still apply to an endpoint that receives imprecise location.

In addition, an endpoint that receives location both by value and by reference from its location provider MUST include both the location value and the location reference in the SIP INVITE message that initiates an emergency call, as specified in [8] (Polk, J., Rosen, B., and J. Peterson, “Location Conveyance for the Session Initiation Protocol,” July 2010.). When the endpoint supports LoST, it MUST use the location value to obtain a PSAP URI for LoST queries before attempting to dereference the location reference. Note that the caller would also have to add the "used-for-routing" parameter to the geolocation header that points to the location value as inserted into the INVITE message. Note that this process crucially relies on the location value having sufficient precision for routing emergency calls (see Section 3 (Determining sufficient location precision) for techniques to ensure the location value is suitable for emergency call routing).

When a PSAP receives a SIP INVITE that contains both a location value and a location reference, and the value is too imprecise for use in dispatch then the PSAP SHOULD dereference the LbyR to obtain more precise information. In turn, the location provided by the location provider MUST allow access by all PSAPs whose service boundaries overlap with the region served by the location provider. This means that either the provider must supply a reference that can be dereferenced by any party, or else the provider must establish explicit authentication and authorization relationships with all PSAPs in its service area. It is RECOMMENDED that location providers establish similar relationships with other PSAPs in adjoining jursidictions -- even if their service regions do not overlap with the location provider's -- in case such a PSAP needs access to precise location information, for example, if it is acting as a backup for one of the location provider's normal PSAPs.

4.2.  Non-emergency services

Non-emergency LBSs may require more precise information than is required for emergency call routing. Therefore, when requesting a non-emergency LBS, the endpoint SHOULD include the location reference provided by its location provider, and MAY additionally provide the location value. If the provided location value is not sufficiently precise to deliver the requested service, then the LBS provider should then dereference the location value to request location information of sufficient precision from the location provider. If the dereference fails, then the request for service may fail as well.

Note that when the location reference provided by the location provider is access-controled, this dereference may require a pre-existing authentication and authorization agreement between the LBS provider and the location provider. In such a case, the endpoint may not know whether a given non-emergency service is authorized to obtain the endpoint's precise location using the location reference. The endpoint is always capable of requesting services without knowing whether they are authorized; in this way, the endpoint can discover authorized services by trial and error. In order to simplify this process, a location provider may supply the endpoint with references to authorized service providers, although there is currently no standard protocol for this transaction.

5.  Acknowledgements

This document generalizes the concept of "rough location" that was originally discussed in the context of the location hiding problem. This concept was put forward by Henning Schulzrinne and Andy Newton, among many others, in a long-running ECRIT discussion.

6.  Security Considerations

The use of imprecise location provides a security trade-off for location providers. When location providers are required to provide location in support of emergency services, they have to balance that requirement against the risk that location information will be disclosed to an unauthorized party. The use of location configuration protocols inherently introduces some risk that an entity other than the target will be able to masquerade as the target (e.g., another host behind the same NAT or malicious software on the host) [9] (Tschofenig, H. and H. Schulzrinne, “GEOPRIV Layer 7 Location Configuration Protocol: Problem Statement and Requirements,” March 2010.). In some cases, the location provider may not authorize the target itself to access precise location. At the same time, because endpoints can roam between networks, it is not generally possible to have strong client authentication for LCPs.

Using of rough location to support emergency calling enables a location provider to provide low-precision location with low assurance (e.g., without client authentication)and high-precision location with higher assurance. Because lower-precision location generally has lower value -- to location providers and LBS providers as a commercial asset, and to targets as private information -- this trade-off allows a location provider to avoid the cost of protecting location with high-assurance access controls when this location has low value.

However, in order to support emergency services, location providers cannot provide only low-precision location; they also have to provide PSAPs with access to high-precision location information. Because PSAPs require high-precision location for emergency response, a location provider that normally provides imprecise location to clients MUST also provide them a location URI that a PSAP can use to obtain high-precision location. This constraint means that the provided URI MUST have either no access control at all or a policy that allows access by appropriate PSAPs and other emergency response systems, e.g., ESRPs. That is, if such a location URI is access controlled, then the location provider MUST be able to authenticate requests from PSAPs.

The use of location by reference introduces some risk that the reference will be used by an attacker to gain unauthorized access to the target's location. These risks are not specific to emergency service, however; general risks and mitigations for location by reference are discussed in [10] (Marshall, R., “Requirements for a Location-by-Reference Mechanism,” May 2010.)

As described in Section 3.1 (Location filtering) above, the location provider choosing to provide a less precise location than a known location has a significant amount of choice in deciding which location to provide: Any location that contains the known location and is in the same filter region will do. When the provider is reducing precision for privacy purposes, there is a some privacy benefit to choosing a random location meeting these criteria. If a watcher is interested in whether or not the endpoint is moving, an imprecise location may still reveal that fact if it is constant when the endpoint is at rest. If the provided location is randomized each time it is provided, then the watcher is unable to obtain even this level of information. An algorithm for securely fuzzing a target's location can be found in [11] (Schulzrinne, H., Tschofenig, H., Morris, J., Cuellar, J., and J. Polk, “Geolocation Policy: A Document Format for Expressing Privacy Preferences for Location Information,” January 2010.); for emergency services, the additional constraint must be added that the fuzzed location must remain in the same filter region as the original.

7.  IANA Considerations

This document requests that IANA register a new PIDF-LO 'method' token in the registry defined by RFC 4119 [5] (Peterson, J., “A Presence-based GEOPRIV Location Object Format,” December 2005.)

area-representative:

Location chosen as a representative of a region in which the target is located; may not be the target's location.

8.  References

8.1. Normative References

8.2. Informative References

Authors' Addresses