TR-IIS-06-014

Design and Implementation of

RFID-Based Object Locator

T. S. Chou and J. W. S. Liu

November 29, 2006 || Technical Report No. TR-IIS-06-014

http://www.iis.sinica.edu.tw/LIB/TechReport/tr2006/tr06.html
 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

Design and Implementation of RFID-Based Object Locator

T. S. Chou and J. W. S. Liu, Fellow, IEEE

Abstract—An object locator is a device designed to assist its interrogator. This situation is clearly not desirable.
user in finding misplaced household and personal objects in a This paper describes three designs and a proof-of-concept
home. This paper describes alternative designs and a prototype of object locators based on the RFID (Radio
proof-of-concept prototype of object locators based on the RFID Frequency Identification) technology. Our object locators do
technology. Advantages of such locators include extensibility
not have the drawbacks of the existing object locator. In
and low maintenance. The numeric model provided here can be
used to determine figures of merits, including costs, search time particular, RFID-based object locators are extensible,
and energy consumption. The results of analysis based on the reusable, and low maintenance. They are extensible in the
model can serve as design guides. sense that the maximum number of tracked objects is
practically unlimited and that a RFID-based object locator
I. INTRODUCTION can support multiple interrogators. The interrogator software
In the coming decades, an increasingly larger number of can run on a variety of platforms (e.g. desktop PC, PDA,
baby boomers will grow into old age. This trend has led to an smart phone and so on). Reusability results from the fact that
increasing demand for products (e.g., [1-6]) and services all RFID tags used for object locators can have globally
designed to help the elderly live independently. One of such unique ids. Hence, tags never conflict, and a tag can be used
products is object locator. An object locator can assist its in more than one object locators. Low maintenance is one of
users in finding misplaced household and personal objects in the advantages of RFID technology. One of the designs uses
a house. Specialty stores and websites now offer object only passive RFID tags; the user is never burdened by the
locators such as the one shown in Fig. 1 for ＄50 US each. A concern that a tag may be out of battery.
locator contains an interrogator with several buttons of The object locator designs described here call for different
different colors and a tag of the color matching the color of hardware components and hence have different overall cost.
each button. By attaching a tag to an object to be tracked, the The differences in their hardware capabilities and object
user can look for the object by pressing the button of search schemes lead to differences in search time and energy
matching color on the interrogator. The tag attached to the consumption. We provide a numeric model that can be used
object beeps and flashes in response and thus enables the user to determine the tradeoff between these figures of merit. The
to find the object. results of analysis based on the model can serve as design
guides to developers of RFID-based object locators. Through
this analysis, we identify the design that is most practical for
the current state of RFID technology. Today, object locators
based on all designs are too costly. We project what it takes to
make RFID-based object locators as affordable as the locators
one can now find in stores.
Our object locator resembles location detection systems in
its goal: assists users to locate objects. Many different
location detection systems are available today. Global
Positioning System (GPS) [7] is the most well known. Priced
Fig. 1 An object locator
at about ＄100 US each, GPS navigators are widely used in
The existing object locator is not ideal in many aspects: cars, buses and so on. However GPS has its limitations.
The number of buttons on the interrogator is fixed; extending Reflection, occlusion and multipath effects seriously interfere
the locator to track more than that number of objects is with distance measurement and make GPS ineffective
impossible. Tags are battery-powered. A tag might become a indoors. For this reason, indoor location detection systems
lost object itself after it runs out of battery. When a tag breaks, use a variety of other technologies, including infra-red (e.g.
a user must purchase a replacement tag of the same color as Active Badge [8]), ultrasound (e.g. Bat [9, 10] and Cricket
the broken one. If a user were to use two tags of the same [11]), and radio frequency (e.g. RADAR [12]). Mote is a
color, both tags would respond to the search signal from the well-known sensor. It is used in MoteTrack [13] for the
indoor location detection purpose. WLAN can be used to
T. S. Chou is with Department of Computer Science, National Tsing Hua
University, Hsinchu, Taiwan (e-email: [email protected]). build location detection system, also. SpotON [14] and
J. W. S. Liu is with Institute of Information Science, Academia Sinica, Nibble [15] are examples. Compared with these location
Academia Road, Section 2, Box 128, Nankang, Taipei, Taiwan (Phone: detection systems, an object locator must be a far more low
+886-2-2788-3799, e-mail: [email protected])

Submitted to IEEE International Conference on RFID 2007, March 2007 1

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

cost solution and must be ultra easy to set up and use. Many Add operation works in a similar way as the address book
indoor location detection systems (e.g. Bat and Active Badge) of a smart phone. Using this operation, the user can add the
rely on a big infrastructure or a pre-computed database (e.g. registration of an object to be tracked into the interrogator. By
RADAR) to support location estimation. These systems are registration, we mean a mapping between the ID of the tag
too costly to deploy and maintain and hence, unsuitable for attached to an object and the name of the object. The user
home use. Cricket system provides low cost location-aware queries the locations of objects by their names. In response to
service. An object with a receiver can resolve its location. a query, the interrogator uses the object-name-tag-id
This is not what an object locator does. A misplaced object mappings to resolve which one of the registered objects to
does not need to know its own location; the user looking for it search. Fig. 3 shows a scenario: The user picks an unused tag
needs to know. and attaches it to an object to be tracked as shown in Fig. 3(a)
The rest of this paper is organized as follows. Section II and (b). Then, the user puts the tag close to the interrogator
describes scenarios that illustrate how a RFID-based object and selects Add object. This step is shown in Fig. 3(c). In
locator may be used. Section III presents three designs of response to Add object command, the interrogator reads the id
RFID-based object locators. Section IV describes the of the tag, displays a new text field and prompts the user to
implementation of a proof-of-concept prototype based on one enter a name (e.g., Key). When the user confirms the name,
of the designs. It also describes the reader collision problem the interrogator creates a mapping associating the name with
[16] encountered in the prototype and the solution we use to the id of the tag attached to the object, and stores the mapping
deal with to the problem. Section V describes a numeric in its local non-volatile memory. This is illustrated in Fig.
model for computing energy consumption and search time 3(d). The user repeats the above steps to register each object
and compares the merits of the designs. Section VI concludes until all objects to be tracked are registered.
this paper and discusses future works.

II. USER SCENARIOS

The routine usage of an object locator requires only three
operations: Add, Delete and Query. We describe these
operations here to illustrate how a locator may be used.
Without loss of generality, we assume that a new object (b) (a)
locator kit contains a portable interrogator, a dozen of RFID (c) (d)
tags and agents. The interrogator resembles a smart phone.
Like smart phones, it has a small non-volatile storage and a
RF transceiver together with a network address. We will
return in the next section to describe how the RF transceiver
is used, as well as agents and their functions. Unlike common
smart phones, however, the interrogator has a RFID reader.
The reader is used for the Add operation described below.
Fig. 2 shows parts of the user interface on an interrogator
with a LCD touch screen and two buttons. The LCD touch Fig. 3 Add operation
screen is used as both input and output user interface. A user
can use a pen to select an item among the items displayed on Delete operation removes the registration of an object, i.e.,
the screen, the button at the bottom left corner to confirm a the object-name-tag-id mapping stored in the interrogator:
selection, and the button at the bottom right corner to cancel The user can invoke the operation by pressing Delete object
the selection. Some operations need text input. The virtual on the touch screen. In response, the interrogator displays the
keyboard shown on right is for this purpose. list of registered objects, allowing the user to select the object
(e.g. Key) to be deleted. The interrogator deletes the mapping
after the user confirms the selection. Delete operation frees
the tag attached to the now unregistered object and makes the
tag free for use to track some other object.
Query operation, illustrated by Fig. 4, is the work horse.
Virtual The user presses Query object on the touch screen, as
Select area keyboard illustrated by Fig. 4(a), to invoke this operation for assistance
in finding misplaced objects. When the names of registered
objects are displayed, the user selects the object to be
Confirm Cancel
searched; in this example, it is Key. After the user confirms
the selection (as shown in Fig. 4(b)), the interrogator retrieves
Fig. 2 The user interface from its local storage the id of the tag attached to the object

Submitted to IEEE International Conference on RFID 2007, March 2007 2

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

with the selected name and starts a search for the tag with that This allows the interrogator and all agents to form a wireless
id. Hereafter, we call the tag being searched the queried tag local area network (LAN). The network address of the
and the object attached to the tag the queried object. We will interrogator (or each interrogator in a multiple-interrogator
return to describe the search process in the next section. system) is unique and so is the network address of each agent.
Object locators of different designs present the result of We assume that the network provides reliable communication.
Query operation in different ways. As examples, Fig. 4(c) and Other aspects of the wireless LAN are irrelevant to our
(d) shows two different responses. In Fig. 4(c), the queried discussion and hence are omitted here.
tag beeps, allowing the user to look for it by following the The interrogator requests assistance from an agent by
sound. This version works like the existing locator described sending the TID of the queried tag to the agent via the
in Section 1. In Fig. 4(d), the interrogator directs the user to wireless LAN. The RFID reader in the agent enables the
the place (e.g. bedroom 1) where the queried object is. agent to search for the tag within its coverage area.
A. RAIT Locator
A disadvantage of the existing locator is that a user needs
to walk around the house when searching an object and the
interrogator needs to repeatedly send the query signal until
the user hears the queried tag or gives up the search. RAIT
locator is designed to eliminate this disadvantage. As shown
in Fig. 5, RAIT locator uses one or more agents to cover each
room, and the house is fully covered by agents. When the user
(a) (b) invokes Query operation, the interrogator sends a query
(c) (d) message containing the TID of the queried tag to agents and
thus requests the agents to search the queried tag on its behalf.
Each agent broadcasts an addressed mode read request with
the TID retrieved from the query message to read the tags
within range. The tag with id matching the TID beeps upon
receiving a read request as well as responding to the agent.
bi~ bi~ bi~ The agent finding the queried tag reports its network address
to the interrogator. This information enables the interrogator
to display the results as shown in Fig. 4(d), telling the user to
Fig. 4 Query operation go to a specified room in the home.

III. ALTERNATIVE DESIGNS

Kitchen 1
The three designs of object locator are called Room-level
Agents, Interrogator and Tags (RAIT) locator, Desk-level Bedroom C
Agents, Interrogator and Tags (DAIT) locator and Desk-level Kitchen 2
Living room 1
and Room-level Agents, Interrogator and Tags (DRAIT) Living room 2
locator. As their names imply, each of the locator consists of
tags, agents and at least one interrogator. The adjectives Bedroom B
room-level and desk-level describe the ranges of RFID Bathroom A
readers used by the designs. The ranges of room-level readers Bedroom A-1 Bathroom B
and desk-level readers are sufficient large to cover a Bedroom A-2
typical-size room or desk, respectively.
The term tag refers specifically to RFID tags. Each tag has
a unique id, hereafter called TID. One of the designs uses
only passive tags. The other designs call for tags that can beep Queried tag Agents Beeper Interrogator
upon receiving query messages containing their TIDs. It is
possible to implement such tags using semi-passive RFID Fig. 5 The configuration of RAIT locator

tags since the battery in such a tag can be used not only to Obviously, the agents must be set up for a RAIT locator to
improve read range but also to drive a beeper. be usable. Fig. 6 lists the steps carried out by the user and
An agent is a device that aids the interrogator in locating work done by the system during the set up process. The goal
the queried object (i.e., the queried tag). Each agent has a RF of Steps 3-5 is to make sure that there is no blind region. A
transceiver, together with a programmable network address, a blind region is an area where tags cannot be read by any agent.
RFID reader, and a RFID tag. As stated in Section 2, the The corners of a room are the most probable blind regions.
interrogator also has a RF transceiver with a network address. This is the rationale behind Step 3. When the TEST READ

Submitted to IEEE International Conference on RFID 2007, March 2007 3

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

RANGE switch of an agent is on, the agent repeatedly agents to be queried in n_hops and the TID of the queried tag
broadcasts non-address mode read messages. In this way, the in tag_id to the first agent. The simplest choice is the agent
agent enables the user to determine whether any of the with the smallest address. In response to a query, each agent
corners is a blind region. searches for the tag with the TID in its own cover area. The
1. Choose a location near middle of a room and temporarily attach an agent to the
agent reports its own address to the interrogator if it finds the
ceiling or furniture at the location. tag; otherwise it decreases n_hops by one, increments its own
2. Turn on TEST READ RANGE switch on the agent.
3. Pick up a tag and check whether the tag beeps at each corner of the room.
network address by one to get the address of the next agent
4. If no, adjust the location of the agent or add one more agent at another location in and then forwards the query message to the next agent.
the room and turn on TEST READ RANGE switch on the additional agent. Then
go back to Step 3. If yes, turn off TEST READ RANGE switch.
According to the polling scheme, the interrogator also
5. Securely attach the agents tested in Step 2-5 at their respective locations. sends a query message to the first agent in its polling list,
6. Put the interrogator near the agent and execute Register Agent operation.
7. Repeat step 1 to 6 until all agents covering the house are registered.
provides the agent with the TID of the queried tag and waits
for response from the agent. The agent replies to the
Fig. 6 Agent set-up process interrogator no matter whether it finds the tag or not. If the
The Register Agent operation in Step 6 is similar to Add response from an agent is negative, the interrogator sends the
operation described in Section II. Its goal is to assign a query message to the next agent in its polling list. Advantage
human-readable location name to an agent, so that the of the polling scheme over the relay scheme is that the
interrogator can later generate query results illustrated by the interrogator can dynamically alter the search sequence.
example in Fig. 4(d). During the operation, the interrogator B. DAIT and DRAIT locators
prompts the user to provide a unique name for the location of
DAIT locator, shown in Fig. 8, is an extension of RAIT
each agent. For example, if the living room needs two agents,
locator. The designs are similar in how the Query operation is
Living Room R(ight) and Living Room L(eft) are good names
handled by the interrogator and agents. A DAIT locator can
for them.
also use any of the search schemes mentioned above. DAIT
The interrogator also assigns a unique network address to
differs from RAIT primarily in the required read ranges of
the agent being registered. The id of the tag in an agent is the
agents. The read range of agents used in a DAIT locator is
product serial number of the agent. The interrogator uses the
less than one meter. Agents with such a small range offer
id to distinguish the agent from previously registered agents.
higher accuracy in locations of queried tags. Information on
By assigning successive network addresses to agents as they
the agent that finds the queried tag tells the user the location
are registered and initialized one by one, successive Register
of the searched object within a small vicinity of the agent.
Agent operations enable each initialized agent to join the
Tags in DAIT locators are passive; they do not beep because a
LAN and later compute the addresses of other agents by
user can easily find the misplaced object even though the tag
adding or substituting some number from its own address.
does not beep. Because tags do not need to beep, they can be
Fig. 7 depicts the format of messages in a RAIT locator.
battery free. This is a major advantage of DAIT locator.
This format supports multiple interrogators: the src_addr
allows agents to identify the interrogator issuing the query
message, the dest_addr allows them to address their responses
to a specified interrogator, and the offset and data field allow
interrogators to synchronize their databases created by Add
and Register Agent operations.
1 1 m n
n_hops intr_addr
CMD cflag tag_id
m m 0 3 4 7 offset data
dest_addr src_addr Object Locator Data Unit

Fig. 7 The message format

A RAIT locator can search a queried object in three ways: Queried object Agent Beeper Interrogator
broadcast, relay and polling. The broadcast scheme is the
most straightforward. The interrogator broadcasts a query
message with the tag_id field filled with TID of the queried Fig. 8 The configuration of DAIT locator
tag. The agents finding the queried tag report their agent ids to However, it is significantly more complicated to set up
the interrogator and the others do not reply. desk-level agents. Blind regions of RAIT locator are easy to
The knowledge on the agent network addresses and the detect and eliminate because a blind region is typically
number of agents enables an interrogator to request assistance created by walls and is near the read boundary of an agent. In
from agents one at a time using the relay scheme: To search the case of DAIT locator, a room cannot be fully covered by
for a queried tag, the interrogator sends a query message one or two agents. Any three adjacent agents may create a
containing its own address in intr_addr field, the number of blind region. Our solution is to give a user a circular thread

Submitted to IEEE International Conference on RFID 2007, March 2007 4

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

whose circumference is less than 3 3 (i.e., the impossible to ensure optimal or near optimal alignment of the
circumference of a regular triangle whose center is one unit tag antennae towards the agents covering their locations.
away from its corners) times their read range and instruct a Hence tags in a DAIT object locator should have
user to set any three adjacent agents within the circular thread. omni-directional antennae. Our DAIT prototype uses only
By doing so, blind regions never occur. tags with directional antennae. (Again, the reason is that such
DRAIT locator is a hybrid design of DAIT and RAIT tags are readily available.) Agents with omni-directional
locators. A DRAIT locator contains both room-level and antennae can be simply set on the furniture as shown in Fig.
desk-level agents. When such a locator is used, the 10(a). Agents with directional antennae should be attached to
interrogator asks desk-level agents to search first. The the ceiling as shown in Fig. 10(b). This arrangement requires
interrogator asks room-level agents only when no desk-level a read range of 2-3 meters. With readers of a sufficiently large
agent finds the queried object. We set up several desk-level read range, RAIT locators can use tags with directional
agents on furniture in addition to setting up room-level agents antennae without performance concern.
as described above. Because misplaced objects are often on
furniture or the vicinities of them, in the majority of the cases,
the queried object can be found by a desk-level agent, and the
tag on it does not need to beep. Thus, this design extends the
life time of a semi-passive tag.

IV. PROTOTYPE IMPLEMENTATION

(a) (b)
We implemented a proof-of-concept prototype of DAIT
locator. We chose to implement this design because it does Fig. 10 The arrangement of agents
not require customized semi-passive tags. Indeed, all
components used in our prototype are readily available today. Close proximity of readers (i.e., agents) is necessary in
Parts (a) and (b) of Fig. 9 shows an agent and the portable order to avoid blind regions. Our DAIT prototype is no
interrogator of our prototype, respectively. The agent is exception. When RFID readers have overlap coverage areas,
composed of a microcontroller, a RF transmitter, a RF signals sent at the same time from them to tags in the overlap
receiver and a RFID reader module. The microcontroller is region interfere with each other. This is called the reader
ATMEL ATmega128. It runs at 8MHz and has 128k bytes collision problem [16]. Fortunately, only the broadcast
flash / 4k bytes EPPROM. The RF transmitters and receivers scheme suffers this problem. Our prototype uses the polling
interconnecting interrogator(s) and agents are LINX scheme to avoid the problem: According to the polling
TXM(RXM)-433-LR, which use 433MHz ASK. RFID reader schemes (or the relay scheme), agents search the queried tag
modules are MELEXIS EVB90121, which is in sequence; signals from readers never interfere.
ISO15693-compliant and uses a directional antenna. We use We will experiment with the broadcast scheme in the future.
TI OMAP5912 and NEC Q-VGA to implement the portable There are many ways to circumvent the reader collision
interrogator. The current version of our prototype supports problem. For example, the DAIT prototype can let each agent
the three operations described in Section II and uses the delay transmitting its query signal by an amount of time that
polling search scheme. is a function of its network address. In this way, agents try to
avoid transmitting query signals at the same time. This
solution is practical and easy to implement and is the solution
we plan to implement. Another solution requires each agent
to know the network addresses of its neighbors. Each agent
can be viewed as a node in a connected graph. There is an
edge between two nodes when the agents represented by them
have overlapping coverage regions. A graph coloring
algorithm can be used to assign different colors to adjacent
nodes. The reader collision problem never occurs as long as
agents labeled by different colors do not transmit query
signals concurrently. This solution is likely to have a better
(a) (b)
response time than the solution mentioned above or the relay
Fig. 9 The agent and interrogator
and polling schemes. However it requires additional
The lack of customized antenna design for tags and readers hardware for each agent to automatically detect its neighbors
and the reader collision problem seriously affects the or connectivity information entered by the user manually. The
performance of our prototype. When the antennae of tags and additional hardware makes agents more costly, and
readers are directional, the read performance of agents complicated operations by the users make an object locator
depends on the orientation of the antennae. Clearly, it is hard to use.

Submitted to IEEE International Conference on RFID 2007, March 2007 5

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

V. RELATIVE MERITS agent finding the queried tag consumes EAI to send a response
We use search time and energy consumption of a single back to the interrogator.
query to measure the relative merits of object locator designs. In the expressions, pAi denotes the probability that the i-th
Search time and energy consumption per query depend on agent in the search sequence finds the queried tag. In general,
many factors including the number of agents, search scheme, this probability is a function of the number and location
search sequence and locations of misplaced objects. distribution of objects (i.e., tags) in the house. (To keep the
expressions simple, our notations do not show this
A. Search Time and Energy Consumption dependency.) The expression of the expected time taken by
The expressions of energy consumption and search time the locator using the broadcast scheme to respond to a Query
per query according to broadcast, relay and polling schemes operation assumes that agents search the queried tag in
are listed in Table 1. The expressions assume that agents and sequence in order to avoid the reader collision problem. The
interrogator(s) are battery powered and communicate in the first term in the expression is the time taken by the query
manners described in Section III. The notations used in the message from the interrogator to reach all the agents. If the
expression are defined in Table 2. first agent finds the queried tag, which occurs with
TABLE 1 probability pA1, the addition delay is DArfid + DAI. This is the
Expressions for search time and energy consumption reason for the second term in the expression of Tavg. In general,
the probability that the queried tag is found by the i-th agent is
i −1

Etotal E IA + NA(x, y, r)E Arfid + E AI ∏ (1 − pA ) pA . When this occurs, each of the i agents spends
k i
k =1

broadcast DArfid amount of time to search for the queried tag before the
DIA + pA 1(D Arfid + DAI) +
n i −1
i-th agent can respond to the interrogator. Hence, the delay is
Tavg
∑ ( ∏1 − pA
i =2 k =1
k )pA i (iD Arfid + DAI) iDArfid + DAI .
The average search time of an object locator that uses the
E IA + pA 1(E Arfid + E AI) +
n i −1
relay and polling scheme are estimated by the expressions in
E avg
∑ ( ∏1 − pA
i =2 k =1
k )pA i (iE Arfid + (i − 1)E AA + EAI) the fourth and sixth rows, respectively. Relay and polling
relay scheme also lets all agents search the queried tag in sequence.
DIA + pA 1(D Arfid + DAI) +
n i −1
This is why the coefficients in these expressions are the same
Tavg
∑ ( ∏ 1 − pA
i =2 k =1
k )pA i (iD Arfid + (i − 1)D AA + DAI) as the coefficients in the expression of Tavg for the broadcast
scheme. The expressions of the average energy consumption
E IA + pA 1(E Arfid + E AI) +
n i −1
can be derived from the expressions of the average search
E avg
∑ ( ∏1 − pA
i =2 k =1
k )pA i (iE Arfid + (i − 1)E IA + iE AI) time by substituting energy consumption for message
polling transmission delay because sending a message cause both
DIA + pA 1(D Arfid + DAI) +
n i −1
transmission delay and energy consumption.
Tavg
∑ ( ∏1 − pA
i =2 k =1
k )pA i (iD Arfid + (i − 1)D IA + iD AI) As stated earlier, Table 1 is based on the assumption that
agents, like the interrogator, are battery powered. Hence, the
TABLE 2 Notations
expressions for energy consumption listed in the table include
energy consumption of both agents and an interrogator.
•DIA: Delay of a message transmission from an interrogator to an agent
•EIA: Energy consumption of a message transmission from an
However, agents can be connected to wall plugs, especially
interrogator to an agent when the number of agents is smaller, as in the case of RAIT
•DAA: Delay of a message transmission from one agent to another locators. The interrogator using relay and broadcast scheme
•EAA: Energy consumption of a message transmission from one agent to
another consumes exactly EIA to search a queried tag. The interrogator
•DAI: Delay of a message transmission from an agent to an interrogator using polling scheme consumes at least EIA to search a
•EAI: Energy consumption of a message transmission from an agent to
an interrogator
queried tag. Thus, the polling scheme is suitable for stationary
•DArfid: Time for an agent to use its RFID reader to search a queried tag interrogator(s) and the relay and broadcast scheme are
•EArfid: Energy consumption for an agent to use its RFID reader to search suitable for portable interrogator(s) if we do not need to
a queried tag
account for the energy consumption of agents.
The total energy consumed by the object locator for
B. Model of Object Locality
processing a Query operation according to the broadcast
scheme is the sum of the three terms in the first row of Table 1. The probability pAi of that an agent Ai finds the queried tag,
In this case, the interrogator transmits only one query and hence the misplaced object, depends on where the object
message when processing a Query operation. The energy it is at the time. To calculate this probability, we use a locality
consumes is EIA. The energy consumed by each agent in the model of tracked objects. The model gives the spatial
search is EArfid. The total energy consumed by all agents is probability density of the locations of each object. For the
NA(x, y, r)EArfid, where NA(x, y, r) is the number of agents with sake of simplicity and without noticeable lose of accuracy, we
range r in a rectangular space of dimensions x and y. The partitions the space in the search area into unit squares, rather
than treating the coordinates of a location as continuous

Submitted to IEEE International Conference on RFID 2007, March 2007 6

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

variables. (Except for where it is stated otherwise, the the average search time and energy consumption of the object
dimension of a unit square is 1 cm by 1 cm.) This allows us to based on the probability pAi for all agents.
model a house as a finite, discrete and planar search space.
C. Evaluation Environment and Results
We denote the space by Z = {Z x, y } ⊆ N × N . Each element
The environment we used to evaluate the relative
Zx,y of the space is a unit square; its location is given by the performance of our designs has a 10m by 10m search space,
coordinate (x, y) where both x and y are integers. All agents containing 1000 × 1000 unit squares of size 1 cm by 1 cm.
are at fixed and known locations. A misplaced object may be Agents are placed according to the arrangement in Fig. 12(a).
placed anywhere within the search space. The number of agents is NA(1000, 1000, r). Again, r is the
We call the probability of finding a queried object at Zx,y read range of an agent. The ranges of desk-level and
the (existence) probability of the object at Zx,y. (For example, room-level agents are 100 and 350, respectively, the typical
if we find an object at Zx,y on the average 10 times in 100 number of room-level agents in a RAIT locator is
searches for the object, the (existence) probability of the NA(1000,1000,350) = 6, and the typical number of agents in a
object at Zx,y is approximately 0.10. We use pZx,y(j) to denote DAIT locator is equal to NA(1000,1000,100) = 42.
the existence probability of an object with a tag of id = j at Zx,y.
We ignore the probability of a registered object being outside
A A A
the search space when the object is being searched. (In other A A
words, we do not consider the situation where someone has
taken some registered object shopping, for example, while A A A
someone else is searching for it in the house.) For every
object, the sum of probabilities of it at all locations in the A A A A A
search space equals to 1. (a) (b) (c)
Fig. 11 gives an illustrative example. The figure is not
drawn in scale, and each unit square in this example is 10 cm Fig. 12 The possible arrangement of agents

by 10 cm in dimension. Two agents A1 and A2 are at their In Section III, we said that the agent with the smallest
locations. The id of A1 is 1 and the id of A2 is 2. The rectangle network address is the first agent and the other agents are
models a desk. It contains 15 unit squares. The number in asked to search for the object one by one in order of agent ids.
each square gives the probability of a queried object being at We call this search order sequential. Alternatively, we can
the location. Since the numbers add up to 1, they tell us that ask the agents in non-increasing order of their empirical
the object is surely some where in the rectangle. We want to existence probabilities. This search sequence is called
calculate pAi, the probability that the agent with id = i can find profiling.
the queried tag. Using Fig. 11 for example, we see that pA1 Our evaluation program assumes that object regions are
equals to the sum of all existence probabilities within the read circular for the sake of simplicity. The center and radius of an
range of the agent A1; in other words, pA1 is about 0.87. object region are randomly generated. The variables DIA, DAA
Similarly, we find that pA2 is about 0.68. and DAI in Table 2 have the same values because both
interrogators and agents use the same kind of RF transceiver.
Z1,1 Z2,1 Z100,1
For the same reason, EIA, EAA and EAI have the same value. For
A1 convenience, we use DArfid and EArfid as base units of delay and
Z1,2
energy consumption. The ratio of DIA/DArfid (DAI/DArfid and
0.05 0.05 0.1 0.05 0.03
DAA/DArfid) is called DRatio and the ratio of EIA/EArfid is called
0.05 0.05 0.2 0.05 0.05 ERatio. The evaluation program needs only these two
0.02 0.05 0.1 0.05 0.05 parameters rather than all variables.
Fig. 13(a) and (b) show the average search time for
broadcast scheme, relay scheme, and polling scheme (i.e.,
A2 polling in sequential order), as well as polling scheme with
profiling. The search time of relay and polling schemes is
Z1,100 Z100,100
higher than broadcast scheme for all values of DRatio. The
Fig. 11 The locality model search time of polling scheme with profiling is less than that
We call the area where a misplaced object might be placed of broadcast scheme when DRatio is less than about 1 (100)
an object region. The size of an object region is the total area for NA = 42 and 1.25 (100.1) for NA = 6.
of the region in number of unit squares. We characterize the Fig. 14 shows the average energy consumption consumed
locality of a misplaced object by the size and shape of its by agents when NA is 42 and 6. The energy consumption
object region and its existence probabilities of being at each consumed by agents is the same, when the relay and polling
unit square within the region. Once we know the locality scheme is used. As Fig. 14(a) depicts, the energy
parameters of an object and coverage area of each agent Ai, consumption of relay scheme, polling scheme and polling
the terms pAi can easily be calculated. We can then calculate scheme with profiling is less than that of broadcast scheme

Submitted to IEEE International Conference on RFID 2007, March 2007 7

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

when ERatio is less than 1.99 (100.3) and 7.94 (100.9), Table 3 gives a summary. The table suggests the broadcast
respectively. ERatio at the intersections of the curves in Fig. scheme when DRatio is high and search time is more
14(b) are about 3.16 (100.5) and 15.85(101.2). important than energy consumption. When DRatio is low, the
differences among the search times of all search schemes are
Broadcast Relay Polling Polling with profiling small. Energy consumption becomes the dominant factor for
4 comparison. It is possible for agents in a RAIT locator to
3.5 connect to power source. For energy saving on interrogators,
Log(search time unit)

3
we suggest polling scheme with profiling for stationary
2.5
interrogators and relay or broadcast scheme for portable
2
interrogators. As for DAIT locators, we consider energy
1.5
consumption of an interrogator and agents. We suggest
1
polling scheme with profiling when ERatio is low and the
0.5
same suggestions as that for a RAIT locator if ERatio is high.
0
-2 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2 TABLE 3 Summary of results
Log(DRatio)
Low DRatio Low DRatio High DRatio High DRatio
(a) NA = 42 Low ERatio High ERatio Low ERatio High ERatio
Broadcast Relay Polling Polling with profiling RAIT locator PI: b or r PI: b or r
broadcast broadcast
3 (fewer agents) SI: pp SI: pp
2.5 DAIT locator PI: b or r
Log(search time unit)

pp broadcast broadcast
2 (more agents) SI: pp

1.5 PI: portable interrogator b: broadcast r: relay

SI: stationary interrogator pp: polling with profiling
1

0.5
VI. CONCLUSION
0
-2 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2 We described here three alternative RFID-based object
Log(DRatio) locator designs. RFID-based object locators are extensible,
(b) NA = 6 reusable and low maintenance. They are very easy for users to
Fig. 13 Search time Vs DRatio use and set up. Our analysis shows that search time and
energy consumption for all designs and search schemes
Broadcast Relay Polling Polling with profiling
3.5 depend the capabilities of RFID readers and RF transceivers
used by agents. Roughly speaking, polling and relay scheme
Log(energy consumption unit)

3
is competitive to broadcast scheme only when DRatio or
2.5
ERatio are less than 10.
2 We implemented a proof-of-concept DAIT prototype
1.5 object locator to demonstrate the object locator concept and
1 designs. The prototype uses only readily available hardware
0.5
components, including readers and tags with directional
antennae. The performance of the prototype is far from ideal,
0
-2 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2
primarily for this reason. Because it is impossible to control
Log(ERatio)
the orientation of tag antennae, omni-directional antennae are
better suited for our application.
(a) NA = 42
The total cost of an object locator depends on many factors.
Broadcast Relay Polling Poling with profiling
2.5 The total hardware cost of a minimum object locator is the
Log(energy consumption unit)

sum of the costs of an interrogator and required number of

2
agents and tags. Compared with the costs of interrogator and
1.5 agent, the hardware cost of tags is significantly lower and, for
1 the discussion here, can be neglected.
Currently, the total hardware cost of an object locator is
0.5
dominated by the total cost of agents, and the cost of an agent
0 is dominated by the RFID reader in the agent. The number of
-2 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2
Log(ERatio)
agents required to fully cover a house depends on dimensions
(b) NA = 6
x and y of the house, the read range of the agents and the way
Fig. 14 Energy consumption of agents Vs ERatio agents are placed. To get a rough estimate, we assume that the

Submitted to IEEE International Conference on RFID 2007, March 2007 8

 Institute of Information Science, Academia Sinica, Technical Report No. TR-IIS-06-014, November 2007

coverage area of each agent is a circle. Fig. 12 depicts three [4] P. H. Tsai, H. C. Yeh, C. Y. Yu, P. C. Hsiu, C. S. Shih and Jane W. S.
Liu, “Compliance Enforcement of Temporal and Dosage Constrains,”
ways to place agents. Putting agents further apart than In Proc. IEEE RTSS, December 2006
locations shown in Fig. 12(a) can create blind regions. Putting [5] http://www.sharperimage.com/us/en/catalog/product/sku__SI676FUN
more agents closer than those indicated in Fig. 12(c) is not “Now You Can Find It!”
[6] http://www.epill.com/ e-pill Medication Reminders
necessary since the space is covered by at least two agents.
[7] Getting I. “The Global Positioning System,” IEEE Spectrum 30, 12
We need six room-level agents to cover a 10m x 10m space (December 1993), 36–47.
even when we place agents as shown in Fig. 12(a) (i.e., as far [8] Want, R., Hopper, A., Falcao, V. and Gibbons J., ”The Active Badge
as possible without creating blind regions). The existing Location System,” ACM Transactions on Information Systems 10, 1
(January 1992), 91–102.
object locator costs ＄ 50 US. A RAIT locator is not [9] Harter A. and Hopper A., “A New Location Technique for the Active
competitive to the existing locator unless the cost per Office,” IEEE Personal Communications 4, 5(October 1997), 43–47.
room-level agent is about ＄10 US. As for DAIT locator, the [10] Andy Harter, Andy Hopper, Pete Steggles, Andy Ward and Paul
Webster, “The Anatomy of a Context-Aware Application,” In Proc.
cost per desk-level agent must be much lower. We are ACM MOBICOM, August 1999
optimistic that the cost of agents will become sufficiently [11] Nissanka B. Priyantha, Anit Chakraborty, and Hari Balakrishnan, “The
Cricket Location-Support System,” In Proc. ACM MOBICOM, August
lower in the coming decade as the need for more and more 2000
products (e.g., Smart pantry [1], dispenser in [4]) containing [12] Bahl P., and Padmanabhan V., “RADAR: An In-Building RF-based
RFID readers are developed to take advantage of this User Location and Tracking System,” In Proc. IEEE INFOCOM, March
technology [17]. 2000
[13] Konrad Lorincz and Matt Welsh, “MoteTrack: A Robust, Decentralized
Approach to RF-Based Location Tracking,” In Proceedings of the
ACKNOWLEDGMENT International Workshop on Location and Context-Awareness at
Pervasive, May 2005
This work is partially supported by the Taiwan Academia [14] J. Hightower, G. Borriello, and R.Want, “SpotON: An indoor 3-D
Sinica thematic project SISARL (http://www.sisarl.org). location sensing technology based on RF signal strength,”
Univ.Washington, Seattle, Tech. Rep. 2000-02-02, February 2000
[15] P. Castro, P. Chiu, T. Rremenek, and R. R. Muntz, “A probabilistic
REFERENCES room location service for wireless networked environments”, In Proc.
[1] C. F. Hsu, H. Y. Liao, P. C. Hsiu, Y. S. Lin, C. S. Shih, T. W. Kuo and ACM UBICOMP, 2001
Jane W. S. Liu, “Smart Pantries for Homes,” In Proc. IEEE SMC, [16] Leong Kin Seong, Ng Mun Leng and Cole Peter H., “The Reader
October 2006 Collision Problem in RFID Systems”, Auto-ID Labs, August 2005
[2] H. C. Yeh, P. C. Hsiu, P. H. Tsai, C. S. Shih and Jane W. S. Liu, [17] http://www.abiresearch.com/products/market_research/RFID_Readers
“APAMAT: A Prescription Algebra for Medication Authoring Tool,” “The Market for RFID Readers”, ABI research
In Proc. IEEE SMC, October 2006
[3] Y. T. Hsu, S. F. Hsiao, C. E. Chiang, Y. H. Chien, H. W. Tseng, A. C.
Pang, T.W. Kuo and K. H. Chiang “Walker’s Buddy: An Ultrasonic
Dangerous Terrain Detection System,” In Proc. IEEE SMC, October
2006

Submitted to IEEE International Conference on RFID 2007, March 2007 9