In 1995 the US Department of Defence (DOD) developed a satellite-based radio navigation system capable of determining within centimetres a position on the earth's surface. The system consists of 24 active and several "back-up" satellites, that orbit the earth providing all-weather navigation and surveying worldwide 36. A summary of GPS capabilities is provided here, but for detailed information readers are referred to Global Positioning System: Theory and Practice by Hoffman-Wellenhof et al 37.

Global positioning systems (GPS) consist of three components: satellites in space, a ground control system, and the user's instrument. The space component consists of orbiting GPS satellites equipped with atomic clocks, and transmitting two radio frequencies modulated with two types of code: precise and standard. The precise code is reserved for U.S. military operations while the standard code can be used freely by any civilian in possession of a GPS receiver. The civilian code comprises a 50 bs^-1 radio signal transmitted at 1,575.42 MHz carrying three signals: a pseudo-random code, ephemeris data, and almanac data. Together these provide information on the satellites available for fixing a position, the current time and date, and the approximate constellation of the satellites at any time throughout the day. The ground control segment consists of five monitoring stations, three ground antennas, and a master control station. The monitoring stations passively track all satellites in view and accumulate ranging information. This information is processed at the master control station to determine satellite orbit geometry and to update the navigation message broadcast by each satellite. The user segment of the system consists of GPS receivers that calculate their own distance from each satellite based on the travel time of the pseudo-random sequences encoded into the radio signal. Given the geometric positions of the satellites (their ephemeris), four pseudo-ranges are sufficient to correct clock error and to compute the three dimensional position of the receiver with an average accuracy of approximately 10 m 3839. Most modern GPS receivers can track 12 or more satellites simultaneously, improving positional accuracy to within 5 m or less.

Natural Resources Canada, a federal government agency, manages a network of ground stations that transmit differential GPS (DGPS) corrections. These ground stations receive satellite information at a known ground location and estimate the difference between the information received by the satellite receiver and the actual location. Differential-enabled GPS receivers can receive the broadcast DGPS signals from the ground stations and make correction calculations, improving accuracy of position to within 3 m or less. Correction data are also available as public domain information from the Canada-wide differential GPS service, the International GPS Service and the Canadian Coast Guard. GPS signal correction can be performed at the time of measurement, using a DGPS receiver, or data may be post-processed if information on both the position of the DGPS receiver and ground station are collected. Improvements in location accuracy can also be achieved through the reception of signals from wide area augmentation systems (WAAS) and researchers should check to see if this service is available in their geographic region.

GPS signals are not immune to interference. The most severe form can occur from intentional signal degradation, also called "selective availability" by the United States National Space-Based Positioning, Navigation and Timing (PNT) Executive Committee 37. Intentional, slowly changing random errors could be introduced into the pseudorandom code transmitted by each satellite resulting in substantial reductions in positional accuracy of 50 m or more in both the horizontal and vertical directions 40. However, selective availability was removed from the system under executive order on May 1, 2000. The United States Department of Defence has since declared selective availability will no longer be used based on security concerns.

Several additional sources of interference may introduce errors that limit the usefulness of GPS in some spheres of human health and activity research. Variability in atmospheric conditions may affect the velocity of GPS signals. In the troposphere water vapour can slow radiofrequency signals resulting in overestimation of signal range. In the ionosphere different components of a signal can be advanced or delayed when interacting with charged gases. The sum of these atmospheric effects can result in errors of 30 to 60 m and vary depending on the angle of inclination of the satellites in view 41. These effects are greater for satellite signals nearer the horizon where signals travel further through the atmosphere before reaching the GPS receiver. Of particular interest to human tracking studies are multi-path errors arising from the reflection of satellite signals from other surfaces, including buildings, vegetation, the ground or water. GPS signals are also blocked by materials such as concrete and steel thus eliminating reception within many institutional and commercial buildings. GPS reception is nonetheless relatively good in automobiles and public transportation vehicles such as buses and trains. GPS accuracy may also be compromised by poor satellite geometry, viewed from the user's location; precision is greatest when signals are received from satellites that are widely dispersed in azimuth and elevation. Thus, two satellites in the same location relative to the antenna provide similar information. The influence of satellite geometry can be quantified using dilution of precision (DOP) indices. Two DOP measures that are important for human tracking research include the positional and horizontal dilution of position (PDOP and HDOP). The first measure expresses uncertainty in overall position whereas the latter assesses uncertainty on the x and y axes. DOP measures generally range from 1 to 10 so that a location estimated with an HDOP of 2.6 has an uncertainty in the horizontal position that is approximately 2.6 times that of the receiver capability.

Researchers and GPS users interested in minimizing positional errors can undertake a mission planning exercise. GPS satellite orbits are known and predictable so that the number of available satellites and their geometric position can be computed for any location and any time. Mission planning software is freely available and almanac information can be downloaded electronically in multiple formats. Figure 1 is a typical report from mission planning software showing the number of satellites (visibility) and two DOP values (PDOP and HDOP)for 0700 to 2100 hours on September 25, 2007, at an urban location in Halifax, Nova Scotia. The opportunity for best reception is between 0900 and 1630 hours when seven or more satellites are available at most times within this window and PDOP values are less than 2.3.

Advances in the miniaturization of GPS and related technologies have led to an assortment of applications: timing, surveying, logistics, traffic management and control, security, marketing, and navigation systems 42. Of particular interest here is the development of GPS technology for health-related applications, specifically those concerned with navigation and tracking. Administrators of emergency 911 systems in many larger urban areas use geocoded address and GPS navigation systems to direct emergency response activities. The incorporation of assisted-GPS technology into cellular telephones allows emergency response teams to accurately assess the location of the distressed caller. The U.S. Federal Communications Commission (FCC) currently requires mobile phone service providers to locate emergency (E911) callers with an accuracy requirement of 100 m (67%) or 300 m (95%) for network-based solutions and 50 m (67%) or 150 m (95%) for handset-based solutions (usually GPS-enabled) 43. Geocoded emergency 911 databases can be used to identify the location of an individual at a specific time; this information can be linked to contextual data to explore the role of place as a determinant of health emergencies. GPS in conjunction with other technologies has been used to support tuberculosis control programs in South Africa 4445, and to identify high-risk areas for transmission of vector-borne and environmental diseases 46. GPS is also used to investigate the positional accuracy of geocoding processing in epidemiological research 4748. Finally, microscale positioning systems that use three-dimensional imagery instead of satellite data are showing promise in surgical applications 49.

Although in its infancy, the use of GPS technology for human tracking presents an enormous opportunity for improving understanding of how the characteristics of places and environmental context influence human activity as well as health and well-being. Many technologies and techniques for human tracking have evolved from wildlife tracking research. GPS receivers have been used to track turtles 50, bears and other large mammals 51, farm and pastoral animals 5253, and primates 5455 with some success under a variety of landscape conditions. Very light GPS-enabled air pollution sensors have been fastened to homing pigeons in order to send real-time location-based pollutant information to an online database and mapping server (see http://Pigeonblog.mapyourcity.net).

Innovations in GPS technology have cultivated interest in the development of portable and wearable GPS tracking devices for research on human activity. The majority of development in this area has been devoted to the commercialization of technologies for tracking criminals or persons under care of the courts. Titanium ankle bracelets with embedded GPS are routinely used for real-time tracking of prisoner transfers and for monitoring convicted offenders who are subject to restrictions on movement 56. Several hardware vendors are retailing similar devices to monitor individuals with memory impediments (including Alzheimer's disease), children or individuals at risk of kidnapping, and family pets. Users can establish a "geofence" that generates an alert when a device moves beyond the limits of a predefined geographic boundary.

GPS technology has also been used in studies of physical activity and human exposures. Studies of exercise physiology and nutrition have used lightweight GPS receivers to assess physical activity as measured by the velocity of walking and running 38, to determine the mechanical power of walking 3957, and to geographically contextualize accelerometry data, which indicates the locations where physical activity occurs 58. Time diaries play an important role in epidemiological assessment of exposures to hazardous agents present in the environment. Several studies have employed commercially-available or custom designed wearable GPS data loggers to validate time-activity diaries 3435 or to track individuals in studies of pesticide exposure 59. More recently, GPS-enabled cell phones have been used to track adolescent travel patterns and activity information 60. GPS technologies are now being linked with a variety of sensors to investigate relationships between environmental conditions and human physiology in time and space; these include sensors of environmental factors such as carbon monoxide concentrations or air temperature, and health-related factors such as heart rate 6162.

The purpose of this study is to develop and pilot test a customized wearable GPS data logger suitable for tracking human subjects over lengthy periods of time. The ability to track people over extended periods of time facilitates the development of individualized spatial units for place and health analyses. In more urbanized areas the application of GPS technology to accurately measure location over time requires evaluative pilot testing for reliability and validity to ensure feasibility of the technology under actual conditions. Here we propose a general framework of dynamic and static tests for evaluating and testing human tracking devices based on GPS technology. A novel contribution of this work is the testing of a wearable GPS across multiple modalities of dynamic measurement among a variety of urban contexts. Knowledge of time-location patterns plays a critical role in understanding how people interact with, and use, space, and reveals the 'places' relevant to the study of variations in health outcomes.

Technological features relevant to the development of wearable GPS for exposure assessment research have been discussed elsewhere 3559. The following list incorporates features from previous work, and introduces several additional physical and performance-based features judged to be critical for human tracking studies: a) size and weight (relatively light and unobtrusive, <0.5 kg), b) logging capability (configurable logging frequency and adequate data storage), c) run-time (minimum 2 d battery capacity at frequent sampling intervals, quick recharge), d) passive (simple to operate and requiring little or no interaction during logging), e) durable (resistant to vibration, minor impacts and water resistant), f) fast time-to-first-fix (obtain fix quickly after signal loss), g) accurate (2–5 m resolution and precise among a variety of built and natural environments).

After constructing several prototypes we developed a wearable GPS data logger instrument called the HeraLogger. The HeraLogger comprises a PVC case (165 mm × 71 mm × 25 mm) containing the GPS module, data logger and battery pack, and an external magnetic patch antenna (Figure 2). The instrument weighs approximately 170 g and easily fits into a jacket pocket or small bag. The antenna has a 2 m cable and can be positioned appropriately to maximize visibility of the sky and satellite signal reception. The GPS module can be configured to output position information at sampling rates up to a maximum of four times per second; data are logged to a removable SD card with capacity up to 1 GB. The instrument can accommodate multiple battery capacities, ranging from 2.4 Ah to 10.4 Ah; this range corresponds to 16 h to 71 h of runtime, or 57 600 to 248 400 data points using a one per second sampling rate. An on/off switch initiates the logging of geographic position and the instrument can be left on while recharging the batteries. The GPS module is a 16-channel receiver with a rated time-to-first-fix of less than 34 s for a cold start, less than 3.5 s for a hot start, and is accurate to within 2 to 5 m of its actual position 63. Software was developed to read, parse and write satellite data to a text file suitable for import into a geographic information system or statistical software package. The cost of each instrument is approximately $450 not including labour costs associated with assembly. Four GPS instruments were assembled for further testing.

Static testing of the Heraloggers was conducted during the summer of 2007 at Dalhousie University, Point Pleasant Park and at the Art Gallery of Nova Scotia in Halifax, Nova Scotia. Three static tests were performed to assess instrument accuracy and precision under field conditions commonly experienced by human subjects. These conditions included the edge of an urban forested park (some obstruction from trees), an open rooftop with no obstructions, and near a building wall to simulate urban canyon conditions (Figure 3). GPS performance improves as the percentage of open sky increases 36.

Precision was estimated for all three sites; accuracy was assessed only at the park location using a known geodetic location, in this case a municipal survey monument maintained by the Province of Nova Scotia (Northing: 4940643.46, Easting: 454981.21, NAD83, UTM Zone 20 N). The four instruments were assessed simultaneously to control for the effects of weather (atmospheric interference) and variations in position estimation resulting from dilution of position (DOP). DOP refers to the geometric strength of satellite configuration on GPS accuracy 37. None of the logged synchronous data were filtered for GPS signal quality so as to simulate actual field conditions. Sampling periods were not selected for optimal signal reception.

Positional data from each of the four GPS instruments were collected for four transportation modes: walking, cycling, automobile, and transit bus. Walking and cycling data were collected along a route approximately 5 km in length. The average time to complete the route was 66 min for walking and 24 min by bicycle. Test participants were asked to walk in the middle of the sidewalk, unless passing another pedestrian, and while cycling to maintain a consistent distance away from the curb unless changing lanes or turning. The automobile test route was approximately 40 min in length; drivers were instructed to abide by posted speed limits and to select the lane closest to the curb on roads with more than two lanes. Transit bus data were collected along an urban downtown bus route with a total loop time of approximately 80 min. Riders were not provided with any specific instructions regarding seating placement in order to avoid special efforts to improve signal reception by selecting a window seat.

Table 1 shows the number of points logged by each HeraLogger to determine the accuracy of the instruments. The average duration of the test was 68 min. The loggers received information from an average of 10.3 satellites during the course of this evaluation resulting in accurate position estimates and nominal PDOP and HDOP values (i.e. below 4.0 and 2.0 respectively). Results revealed horizontal position accuracies of 2.65 ± 0.25 m for 50% of logged values, 7.83 ± 1.17 m for 98% of logged values, and an average distance of 2.82 ± 0.40 m away from the known reference location.

Table 2 gives details of the precision of the GPS instrument under three built environment scenarios. Precision was best in the park setting. The average duration of the logs was 115 min, and the mean number of satellites obtained, as well as HDOP and PDOP values, were similar to those data from the accuracy tests. Half of the values lie within 2.00 ± 0.35 m and 98% lie within 5.45 ± 1.06 m. Logged data from the rooftop location are to some extent less precise than those from the urban park location. Logging took place over a 24 h period and, on average, one less satellite was used to provide a location solution. The mean CEP value is 2.35 ± 0.11 m and 98% of the location data fall within 6.18 ± 0.29 m. Position precision estimates worsened under urban canyon conditions where multipath effects are expected to be more of an issue. Data were logged on average for 63 min; three fewer satellites were available to derive location solutions, in comparison with the urban park test. PDOP values almost doubled compared to other locations. The radius of the circle required to capture 50% of the position data was roughly nine times larger than for data from the urban park. A circle with an average of radius of 53.14 ± 25.06 m captured up to 98% of the GPS data.

Table 3 shows the performance of the GPS instruments using different forms of transportation in the three built environments. Positional accuracy under open sky conditions was best when cycling and walking (72% to 99.1% of all positions within 5 m). No data are available for transit bus tests in open sky conditions. In mixed density areas, determinations of position were more accurate from automobile logs (89% within 5 m), followed by cycling (81% within 5 m), walking (74.5% within 5 m) and then transit bus (65% within 5 m). In urban canyon areas where GPS receivers are most challenged, the greatest positional accuracy was attained from automobile (82.6% within 5 m) and transit bus modes (60.2% within 5 m), followed by walking and then cycling (57% and 53.7% within 5 m, respectively).

Satellite reception was best for automobile travel (10 satellites) followed by walking, cycling and then transit bus transportation (8.3 satellites); however, this difference in reception did not always translate into reduced horizontal and positional dilution of position values. Figure 4 shows a close-up perspective of a pedestrian path under open sky conditions. The dotted line shows the true path (sidewalk) bounded by 2 m, 3.5 m, and 5 m buffers, as well as the actual GPS locations logged at 5 s intervals. As expected, signal accuracy for walking and cycling deteriorated as the potential for the built environment to interfere with satellite reception increased. Differences in signal reception are less apparent under varying built environments for automobile trips.

Figure 5 shows the relationship between horizontal (HDOP) and positional dilution of position (PDOP) under static and dynamic GPS instrument testing conditions. Static HDOP varies in a log-linear fashion with PDOP, so that reductions in horizontal accuracy diminish at a value of approximately PDOP = 2.3; this implicates vertical dilution of position as the reason for decreased PDOP in static conditions. PDOP varied less with HDOP under dynamic conditions with no clear association between dilution of position values. HDOP values were greater under dynamic as compared to static conditions.