mitters (tags) worn by participants and an ultrasound receiver (locator) ... versity of California, Berkeley, CA 94720-7360; email: krksmith@ berkeley.edu.
Original Articles
An Ultrasound Personal Locator for Time-Activity Assessment GIAN ALLEN-PICCOLO, MPH, JAMESINE V. ROGERS, MPH, RUFUS EDWARDS, PHD, MICHAEL C. CLARK, MS, T. TRACY ALLEN, PHD, ILSE RUIZ-MERCADO, MS, KYRA N. SHIELDS, PHD, EDUARDO CANUZ, BE, KIRK R. SMITH, PHD The UC Berkeley Time-Activity Monitoring System (UCB-TAMS) was developed to measure time-activity in exposure studies. The system consists of small, light, inexpensive battery-operated 40-kHz ultrasound transmitters (tags) worn by participants and an ultrasound receiver (locator) attached to a datalogger fixed in an indoor location. Presence or absence of participants is monitored by distinguishing the unique ultrasound ID of each tag. Efficacy tests in rural households of highland Guatemala showed the system to be comparable to the gold-standard time-activity measure of direct observation by researchers, with an accuracy of predicting time-weighted averages of 90–95%, minute-by-minute accuracy of 80–85%, and sensitivity/specificity values of 86–89%/71–74% for one-minute readings on children 3–8 years-old. Additional controlled tests in modern buildings and in rural Guatemalan homes confirmed the performance of the system with the presence of other ultrasound sources, with multiple tags, covered by clothing, and in other non-ideal circumstances. Key words: Exposure assessment; datalogging; indoor air pollution; efficacy; RESPIRE; Guatemala I N T J O C C U P E N V I R O N H E A LT H 2 0 0 9 ; 1 5 : 1 2 2 – 1 3 2
M
onitoring personal exposure to airborne pollutants is not practical or even possible in many populations, due to cost, intrusiveness, and/or lack of appropriate technology. Consequently, investigators often rely on micro-environmental monitoring combined with time-activity assessment.1 Unfortunately, however, the social science methods for timeactivity assessment (questionnaire, diary, observation)
Received from: EME Systems, Berkeley, CA (GAP, TTA), University of California, Berkeley, School of Public Health (JR, MCC, IRM, KNS, KRS), University of California, Irvine, Department of Epidemiology (RE), and Universidad del Valle, Guatemala (EC). Part of this work was supported by the Brian and Jennifer Maxwell Endowed Chair at the UC Berkeley School of Public Health and by the National Institute of Environmental Health Sciences (R01ES010178). Send correspondence to: Kirk R. Smith, School of Public Health, University of California, Berkeley, CA 94720-7360; email: krksmith@ berkeley.edu. Disclosures: The authors declare no conflict of interest.
122
do not work well in some circumstances and themselves can be expensive and intrusive.2 Time stamped voicerecorded activity-location data3,4 and video analysis5 have proven to be useful techniques, but their transcription and interpretation are time consuming. Consequently, there have been efforts to develop more precise, objective, and less intrusive means of time-activity assessment,6 often using devices derived from the information technology revolution, such as GPS,7,8 and more recently in combination with other devices like accelerometers9 and radio frequency identification.10 In our studies of indoor air pollution in developingcountry households in remote settings, we required a method that was inexpensive, reliable, deployable on a routine basis over multiple years by lightly trained personnel, sufficiently non-intrusive to be applied with every population group including infants, and with output easily linked to that from datalogging pollution monitors in the micro-environments of interest (usually rooms within the households). On the other hand, we did not need to know the location of participants at all times, but only when they were in the rooms where pollution was being monitored. Indeed, there would likely be ethical issues raised by deploying devices that reveal people’s locations over 24 hours. Research needs reduced time-activity to a binary function: inside the room or not. We thus worked to develop a system consisting of two components. First, a stationary “locator” mounted in the microenvironment in question. Second, one or more “tags,” each with a signature indicating the identity of the participant wearing it, that could be worn on the clothing of study participants and sense and log when the tag is in the microenvironment. Three types of wireless technology were tested for the system: infrared, radio, and ultrasound. Infrared has the advantage of line-of-sight propagation (it is not able to pass through walls), making it attractive with regards to our binary (in-out) model. Unfortunately, since the tags could become covered by one or more layers of clothing during daily activities, an infrared signal would have a good chance of being cut off. Radio, on the other hand, is too penetrating, easily passing through walls. There-
fore, someone outside could be mistaken for someone inside the room. Ultrasound signals, in contrast, are able to go around corners and through clothes, and are not powerful enough to easily penetrate walls. Available ultrasound technologies are somewhat more expensive than infrared and radio, but not enough to pose a serious handicap.
THE UCB ULTRASOUND TIME-ACTIVITY MONITORING SYSTEM (UCB-TAMS) Each UCB-TAMS system consists of ultrasound transmitting “tags,” and an ultrasound receiving “locator.” Each tag has its own identifying signature (ID), represented by a sequence of beeps, and is worn by a particular study participant. The locator, usually mounted on the wall of the room of interest, “listens” for the ultrasound emitting tags. The locator datalogger tallies the number of times the locator receives a particular tag signal during the course of each minute. At the end of a minute, the datalogger records the date/time and the number of times each tag ID has been received by the locator. The process then repeats for the next minute. In this manner, the time-activity of study participants within the microenvironment of interest can be assessed on a minute-to-minute basis over extended periods, as long as the tags are worn by participants as intended. The UCB-TAMS is inexpensive, reliable and requires a relatively small amount of attention from researchers. It operates above the hearing range of humans and terrestrial animals, and thus is not noticed by domestic animals such as dogs, cats, chickens, etc. As there are no moving parts, sharp edges, or radiation, the system is safe for widespread use. It appears to be a good alternative to more traditional time-activity measurements, especially in cases where a binary inside/outside measurement will suffice. For our purposes, the UCB-TAMS gives more precise results than would a conventional GPS device, which does not have sufficient resolution to measure presence in most indoor locations. Here, after a brief discussion of the way the system operates, we describe tests undertaken in controlled conditions and efficacy tests during actual household activities to validate the accuracy and reliability of the UCB-TAMS system. Field testing took place at our research site in highland Guatemala in households using wood for cooking that were part of the RESPIRE study of woodsmoke and health.11 Standard consents were taken from participants for the field test according to the approval obtained from UC Berkeley’s Committee for Protection of Human Subjects.12
cell battery. The CR2032 battery is mounted under a spring clip and can be replaced in the field. The UCBTAMS tag sleeps between each transmission, so that the average current drain from the battery is 200 microamps, and expected battery life is near 1000 hours. The piezoelectric transducer is driven to a level of twice the battery voltage and a sound pressure level of approximately 105dbm by a differential drive circuit. At 25 g and 5.3 ⫻ 3.5 ⫻ 1.5 cm, the tags are unobtrusive as well as rugged (Figure 1). Each tag is programmed with a distinct binary code (ID) that distinguishes it from other tags. The codes are transmitted as approximately 0.6 second, 40Khz ultrasound “beeps” at random intervals between 0 and 7 seconds. Each “beep” consists of six periods of sound and silence that act to create a binary “on-off” pattern. Each “bit” of the binary codes is approximately 100 milliseconds (ms) in duration. The bit duration is chosen to produce the shortest possible signal transmission time while ensuring integrity of each bit. The signals are emitted at random to minimize interference from multiple tags (Figure 2).
Ultrasound Locator (“Locator”) The ultrasound locator unit has a size and construction much like that of the tag. It consists of a 40 kHz ultrasound receiving transducer and a processing circuit that filters, amplifies and detects tag signals using a phase lock loop. Raw data is in the form of voltage levels that represent the “ones” and “zeros” of the detected ultrasound. The locator and datalogger are typically mounted on a wall in a room, preferably not facing a doorway in order to avoid receiving tag signals from outside the room. The datalogger decodes the bit pattern carried in the ultrasound signal and thereby identifies the tag. In order to simplify data analysis, an
Ultrasound ID Transmitters (“Tags”) Each tag consists of a piezoelectric 40 kHz ultrasound transmitter powered by a commonly available 220 milliamp-hour capacity CR2032 LiMgO2 3-V lithium coin-
VOL 15/NO 2, APR/JUN 2009 • www.ijoeh.com
Figure 1—UCB-TAMS Ultrasound Transmitter (“tag”). Note the small size. The enclosure is ABS plastic. The circular metal protrusion is the ultrasound transducer. The receiver is nearly the same size.
An Ultrasound Personal Locator for Time-Activity Assessment
•
123
Figure 2—Signal outputs from 2 tags and detail of a sample binary code pattern. The figure shows the signals produced by two separate tags: tag A and tag B. Each small rectangle represents an ultrasonic beep that contains the identifying pattern specific to each tag. The beeps occur at a random interval of 0–7 seconds, and are 600 ms in length. Interference can occur if beeps emitted from two tags overlap in time. The tags’ identifying patterns (insets, dashed boxes) are themselves a series of beeps and silent pauses in time slots that are each 100 ms in duration. Interference between tags can create a signal that is represented neither by pattern ID A nor Pattern ID B.
algorithm has been developed that transforms each minute’s tally of detected tag IDs into “present” or “absent.” The algorithm employs a user-selectable threshold (typically 2–5 times a minute) to make this decision as a safeguard against false positives. False positives are reduced by rejecting codes that do not correspond to any of the deployed tags. The cost of parts and labor for a package consisting of 3 tags and one locator is about $80 in lots of 100.
VALIDATION STUDIES Three sets of validation tests were carried out: first in buildings in Berkeley, California to test physical operating conditions; the second set extended these controlled tests to typical households in our Guatemala site; and the third set of “efficacy” tests were conducted in households undertaking normal daily activities, but still under partly artificial conditions due to the presence of a trained observer from the research team. As such, they also did not test compliance by the householders, since the observer served to make sure the tags were used properly. Three or more receptions per minute was the threshold used to categorize a tag as present in the results reported for all the studies here, except Test 11.
124
•
Allen-Piccolo et al.
Performance Tests in Controlled Settings The first set of seven tests was conducted in Berkeley by the authors. 1) Performance of the system under optimal conditions. A tag was mounted 1.5 m high on a wall and 3.5 m directly across from 3 wall-mounted locator/data-logger units. The test was carried out for 24 hours each using 2 different tags. Results. All locators successfully recorded the two tags’ presence for each minute of the 24-hour test interval (100% accuracy). In all, there were 8 false positives and 120 “junk” one-minute readings, making a gross error rate of about 1.5% (n=8640 locator-tagmin). Junk readings are defined as those recorded as tag ID1. These readings are separated from other false positive readings as no real ID1 tag is ever used, and therefore could be easily distinguished and discarded if recorded in the field. Thus, the irresolvable error rate was