You are on page 1of 9

Adaptive Binary Splitting: A RFID Tag Collision

Arbitration Protocol for Tag Identification


Jihoon Myung and Wonjun Lee

Dept of Computer Science and Engineering, Korea University, Seoul, Korea


wlee@korea.ac.kr



I.

This work was funded by SK Telecom under Contract Number KU-R0405721 to Korea University.

Correspondent Author
Abstract In the RFID system, a reader recognizes tags through
communication over a shared wireless channel. When more than
one tag transmits their IDs at the same time, the tag-to-reader
signals lead to collide and collision disturbs the readers identifi-
cation process. Therefore, tag collision arbitration for passive
RFID tags is significant for fast identification. This paper pre-
sents an Adaptive Binary Splitting (ABS) protocol which is an
improvement on the binary tree protocol. To reduce collisions
and identify tags efficiently, ABS use information which is ob-
tained from the last processes of tag identification. Our perform-
ance evaluation shows that ABS outperforms existing tree based
tag anti-collision protocols.
Keywords RFID, Tag Anti-collision, Tag Identification
INTRODUCTION
Radio frequency identification (RFID) system is an auto-
matic identification system. A RFID reader recognizes an ob-
ject through wireless communications with the tag which has a
unique ID and information and is attached to the object. The
reader must be able to identify tags as quickly as possible.
However, reader-to-tag signals or tag-to-reader signals collide
because readers and tags communicate over the shared wireless
channel. Collisions make both the communication overhead
and the transmission delay of readers and tags have lost their
usefulness. As a result, either the reader may not recognize all
objects or a tag identification process may suffer from long
delay. Therefore, anti-collision protocols which enable the fast
and correct identification regardless of the occurrence of colli-
sions are required.
Collisions are divided into reader collisions and tag colli-
sions [1]. Reader collisions occur where neighboring readers
interrogate a tag simultaneously and confuse it. Tag collisions
mean that more than one tag tries to respond to a reader at the
same time and make the reader unable to recognize any tags.
Reader collisions can be resolved [2][3] because RFID readers
can detect collisions and communicate with one another. Espe-
cially, since low-functional passive tags can neither detect col-
lisions nor figure out neighboring tags, tag anti-collision proto-
cols are important for identification ability of RFID systems [4].
Tag anti-collision protocols can be grouped into two broad
categories: aloha based protocols and tree based protocols.
Aloha based tag anti-collision protocols such as aloha, slotted
aloha, and frame slotted aloha [5][7] reduce the possibility of
the occurrence of tag collisions how tags transmit at the distinct
time. Alohas tags randomly select their transmission time and
tags of slotted aloha can try to transmit only at the beginning of
a timeslot which is a certain time period. Frame slotted aloha
which shows the best performance of aloha based protocols
configures a frame with continual timeslots. As a tag transmits
its ID only at a timeslot in every frame, the frame slotted aloha
reduces collisions. Aloha based protocols, however, cannot
perfectly prevent collisions. In addition, they have the serious
problem that a tag may not be identified for a long time so
called tag starvation problem. On the other hand, the tree based
tag anti-collision protocols such as the binary tree protocol
[5][6][9][10] and the query tree protocol [11] do not cause tag
starvation though they have relatively long identification delay
[8]. They split a group of colliding tags into two subgroups
until the reader receives signals of tags without collisions. In
the binary tree protocol, tags are required to have functional-
ities of managing a counter and a random number generator.
The colliding tags are split according to a randomly selected
number, 0 or 1. The tags which select 0 transmit their IDs im-
mediately and the tags which select 1 transmit later. Through
continuing to split with a random number, a reader can recog-
nize all tags. The query tree protocol is the deterministic tag
anti-collision protocol. A reader sends a query including a pre-
fix and tags of which ID matches the prefix respond. The prefix
of the reader has the decision power about splitting the collid-
ing tags. The query tree protocol is called the memoryless pro-
tocol because tags need not to have additional memory except
the ID. Since it, however, uses prefixes, the performance is
sensitive to the distribution of tags IDs which the reader is
going to recognize. There are variants of the query tree proto-
col in order to reduce the identification delay [12], but they
simply focus on shortening time taken for transmitting the tags
ID and cannot reduce the number of collisions.
We propose the Adaptive Binary Splitting (ABS) protocol
which restrains tag-to-reader signals from colliding. We focus
our attention on scheduling tags with less collision. By reduc-
ing the number of collisions, a reader can recognize tags faster
and a tag can be recognized with less transmissions. We have
studied the similar approach in the query tree protocol [13].
The essential elements of ABS are the timeslot allocation pro-
cedure and the empty timeslot elimination procedure. To iden-
tify tags efficiently, the timeslot allocation procedure assigns
collision-free timeslots to tags and the empty timeslot elimina-
tion procedure removes unnecessary timeslots where any tags
do not transmit. Tags can quickly reoccupy an exclusive time-
slot with schedular information which is obtained from the last
375 0-7803-9277-9/05/$20.00/2005 IEEE
process of tag identification. The simulation result shows that
our proposed protocol suppresses the occurrence of collisions
and reduces the total delay for identifying all the tags while
preserving low communication overhead. The rest of this paper
is organized as follows. Section II describes ABS. In Section
III and Section IV, the performance analysis including the
derivations of several major equations referred to in the analy-
sis will be explicated. Finally the conclusions of the paper will
be drawn in Section V.
II.
A.
B.
ADAPTIVE BINARY SPLITTING PROTOCOL
ABS schedules tags transmissions via consecutive commu-
nications between a reader and tags. If a tag dwells within the
readers identification range, it is able to communicate with the
reader directly. Tags transmit their own ID and then the reader
detects collision. The reader always informs all tags whether or
not the tag-to-reader signals collide. When tag-to-reader signals
lead to collision, the colliding tags randomly select a binary
number, 0 or 1. Based on this selected number, a group of the
colliding tags is split into two subgroups. By continuing this
split until tags enable to transmit without collision, the reader
can recognize all the tags. Since each tag gets an exclusive time
for transmission, ABS can reduce the number of collisions of
the tag-to-reader signals and identify tags fast. In addition,
ABS accomplishes fast re-assigning the collision-free timeslots
to each of the tags even though a set of tags which inhabit in
the vicinity of the reader becomes different.
Description of ABS
We define a frame as the time elapsed for identifying all the
tags within the readers transmission range. The reader can
adaptively decide the length of a frame with the commands
informing tags of beginning and terminating the frame. The
reader can change the ending point of a frame at any time. The
frame consists of timeslots which are certain time periods. In
each timeslot, tags transmit their IDs and the reader receives
the tag-to-reader signals. According to the number of signals
transmitted in a timeslot, we can categorize the timeslots as
follows.
Empty timeslot: No signals are transmitted by tags in a
timeslot.
Readable timeslot: Only one tag transmits its ID and is
successfully recognized by the reader. Since ABS does
not terminate a frame till all tags are recognized, the
number of readable timeslot equals to the number of
tags recognized by the reader.

Figure 1. Three states of the tag
Collisional timeslot: More than one tag transmits and
then the tag-to-reader signals collide. The reader is un-
able to recognize any tags.
The reader, in the end of a timeslot, sends a feedback informing
all the tags of the type of the current timeslot. After receiving
the feedback, tags operate the timeslot allocation procedure and
the empty timeslot elimination procedure so that a timeslot will
carry only one tags signal. To realize the fast identification,
ABS holds down empty and collisional timeslots.
The tag maintains values of a progressed-slot number and an
allocated-slot number. The progressed-slot number represents
the number of timeslots passed in a frame and is initialized
with 0 at the beginning of a frame. The progressed-slot num-
bers of all the tags are always equal. To put it concretely, the
value of the progressed-slot number is not increased in every
timeslot and is only increased by 1 in the readable timeslot, i.e.,
when a tag is successfully identified. The allocated-slot number
signifies the sequence that the tag can access a channel to
transmit. In other words, the tags of which the allocated-slot
number is the same value as the progressed-slot number can try
to transmit at the beginning of the timeslot. As shown in Fig. 1,
the tag has one of three states as follows:
Wait state: The tag has the allocated-slot number
greater than the progressed-slot number. It does not
transmit any signal and waits for its turn.
Active state: The tag has the allocated-slot number
equal to the progressed-slot number and tries to trans-
mit its own ID.
Sleep state: The tag has the allocated-slot number less
than the progressed-slot number. Since the tag has al-
ready recognized in the ongoing frame, it does not
transmit any signal until the completion of the frame.
As the timeslot allocation procedure and the empty timeslot
elimination procedure of ABS change the progressed-slot num-
ber and the allocated-slot number, the tag takes possession of
the favorable timeslot to transmit. In the collisional timeslot,
the colliding tags, i.e., the tags of the active state, add a ran-
domly selected binary number (0 or 1) to the allocated-slot
number. Therefore, the active tags which select 1 convert their
state into the wait state. The tags in the wait state, when colli-
sion occurs, increase the allocated-slot number. The tags in the
wait state, when the received feedback points that any signals
are not carried in the current timeslot, decrease the allocated-
slot number. The detailed description and the example of two
procedures are given in Section II-C and Section II-D. After all,
each of allocated-slot numbers is assigned to only one tag.
There exist no ownerless allocated-slot numbers less than any
allocated-slot number which a certain tag owns. The tags pre-
serve the allocated-slot number at the beginning of the next
frame and the timeslot allocation procedure and the empty
timeslot elimination procedure re-arrange allocated-slot num-
bers fast. Consequently, tags are fast recognized in next frames.
Frame Termination
The reader determines the end point of a frame with a pro-
gressed-slot number and a terminated-slot number. The pro-
gressed-slot number of the reader represents the number of
timeslots passed in the ongoing frame like one of the tag. If the
376
reader successfully recognizes the tag which transmits the ID
alone, the progressed-slot number is increased by 1. The termi-
nated-slot number signifies the last timeslot number of the
frame. As soon as the progressed-slot number is greater than
the terminated-slot number, the reader concludes that all tags
are recognized and transmits the command terminating the
frame to all the tags.
The frame should have the proper length regarding the num-
ber of tags in the identification range of the reader. If the ter-
minated-slot number is too big compared with the number of
tags, the empty timeslots is increased and the reader suffers
from long identification delay. On the other hand, if the termi-
nated-slot number is too small, the reader is unable to recog-
nize all tags. The number of tags, however, is not correctly
predicted earlier. For this reason, the timeslot allocation proce-
dure and the empty timeslot elimination procedure adaptively
change the terminated-slot number during a frame in order to
identify all the tags and to eliminate the unnecessary empty
timeslots. The reader, in the end of a timeslot, changes the pro-
gressed-slot number and the terminated-slot number according
to the type of the current timeslot. For terminating a frame after
identifying all the tags, the reader acts as the tag which has the
largest allocated-slot number. At this time, the reader updates
the terminated-slot number as if the tag updates the allocated-
slot number. The reader closes the current frame in its own
timeslot. By changing the values of the progressed-slot number
and the terminated-slot number, the reader can adaptively de-
termine the size of the frame.
C.
D.
Timeslot Allocation
Fig. 2(a) describes a tag identification process that the exist-
ing tree based anti-collision protocols are recognizing four tags
(tag A, B, C, and D) through splitting a group of colliding tags
into two subgroups. A circle in the figure means a timeslot
when some tags transmit and a number in a circle corresponds
to the number of tags which transmit simultaneously. A num-
ber in a rectangle corresponds to the sequence that tags are
recognized. The reader recognizes four tags after detecting
three collisions. Since the binary tree protocol and the query
tree protocol do not exploit information which is obtained from
the last identification process, they make three collisional time-
slots again whenever the reader recognizes tag A, B, C, and D.
We are motivated by the observation that the reader may en-
able to recognize four tags without collision by using the iden-
tification sequence of tags as shown in the dotted line of Figure
2(a). The identification sequence is also a good tool for reduc-
ing collisions even though new arriving tags, which have not
been recognized by the reader in the last frame, come into the
identification range of the reader. Consider the scenario that tag
E arrives at the area where tag A, B, C, and D inhabit. The
reader attempts to recognize five tags. Figure 2(b) shows that
the increment of the number of tags causes more collisions and
the reader recognizes five tags with four collisions in the exist-
ing protocols. The use of the identification sequence, however,
enables to make the reader recognize five tags with only one
collision.
If more than one tag has the identical allocated-slot number,
the signals of these tags cause collision. The reader cannot rec-
ognize any tags due to collision, but can detect the occurrence
of collision. The colliding tags should be allocated different
timeslots to transmit successfully. Therefore, ABS splits a
group of the colliding tags into two subgroups and creates a
new timeslot for the allocation to one of two subgroups. If the
current timeslot is the collisional timeslot, where the tag-to-
reader signals lead to collide, the tags of the active state ran-
domly select a binary number, 0 or 1 and then add the selected
number to their own allocated-slot numbers. Since the pro-
gressed-slot number is not changed, the tags which select 0 re-
try to transmit at the following timeslot. On the contrary, the
tags which select 1 increase their own allocated-slot number by
1. These tags get to have the same allocated-slot number as
other tags which already have had the allocated-slot number
equal to the progressed-slot number + 1 at the beginning of the
timeslot. For preventing two subgroups from combining, the
tags of the wait state add 1 to the allocated-slot number when-
ever they receive the feedback pointing the collisional timeslot.
When collision occurs, the reader adds 1 to the terminated-slot
number in order to increase the length of the frame. This opera-
tion of splitting tags and creating a timeslot is continued until
each of the timeslots is allocated to only one tag.

a) Recognizing tags (A, B, C and D)

b) Recognizing tags after Tag E comes
Figure 2. Causing collisions
Empty Timeslot Elimination
When a group of colliding tags is split into two subgroups;
one subgroup, which contains all the colliding tags, and the
other, which contains no tags, the tree based tag anti-collision
protocols cause the empty timeslot. In Fig. 3(a), tag A and tag
B transmit at the same time and then they are selected into the
377
left subgroup. The timeslot for the right subgroup does not
carry any tag-to-reader signals. In addition to that, ABS which
exploits the identification sequence creates additional empty
timeslots as tags go out of the readers range. Fig. 3(b) shows
the occurrence of the empty timeslot after tag B at Fig. 3(a)
moves out of the readers range. The empty timeslots do not
make a reader fail to notice a tag, but extend delay which is
required to recognize all tags. Therefore, the empty timeslot
elimination procedure of ABS re-adjusts the identification se-
quence of tags in order to minimize the impact of empty time-
slots. If the current timeslot is the empty timeslot, the tags of
the wait state decrease the allocated-slot number by one. Since
the progressed-slot number is not changed in the empty time-
slot, the decrement of the allocated-slot number gets to pull the
schedules of transmissions of the tags which have the allo-
cated-slot number greater than the progressed-slot number.
When the reader does not receive any tag-to-reader signals in
the current timeslot, it also reduces the terminated-slot number
by 1. ABS can eliminate empty timeslots regardless to the fail-
ure of splitting collisions or the disappearance of tags.
E.
III.
Impact of tag movements
As objects which have tags move, the set of tags which the
reader has to recognize in the following frame may be different
from the set of tags which are recognized in the previous frame.
If the set of tags at the next frame is equal to the set of tags at
the last frame, ABS does not make any empty timeslots and
any collisional timeslots. Since all the tags maintain the allo-
cated-slot number, they can transmit without collision. The
readable timeslot is the only component to consist a frame and
the tags are recognized with ideal delay. When the tags which
have not recognized in the last frame come into the vicinity of
the reader, they can use their possessive allocated-slot number
(may be set up by other reader) without errors. Only if the allo-
cated-slot number of the new arriving tag is greater than the
terminated-slot number of the reader, the reader is unable to
recognize the new tag. To cope with this problem, the reader
offers its terminated-slot number with the command beginning
a frame. The arriving tag which has the allocated-slot number
greater than the terminated-slot number resets the allocated-slot
number with 0. When the tags of the old set which has identi-
fied in the previous frame go out of the readers range, the
frame has the empty timeslots. However, the empty timeslot
elimination procedure removes the empty timeslot and adapts
the number of reduced timeslots to the length of the frame.
Therefore, ABS is not affected by the movement of tags.
a) Empty timeslot by wrong splitting

b) Empty timeslot by tags movement
Figure 3. The occurrence of empty timeslots
IDENTIFICATION DELAY ANALYSIS
We analyze the average identification delay of ABS. The av-
erage delay is proportional to the duration of the frame because
ABS terminates the frame as soon as the reader recognizes all
tags.
Definition 1: Let A
i
be the set of tags which are recognized
in the ith frame. The total delay d
total
(A
i
) for identifying is
( ) ) ( ) (
i tag reader i total
A T d d A d + =

(1)
where d
reader
is the delay of delivering a readers command, d
tag

is the delay of delivering a tags message, i.e., ID, and T(A
i
) is
the number of timeslots required for identifying all tags of A
i
.
The total delay is determined by the number of timeslots, T(A
i
)
because d
reader
and d
tag
are constant.
Lemma 1: The number of timeslots, T(A
i
) is the sum of the
number of empty timeslots, E(A
i
), the number of readable time-
slots, R(A
i
), and the number of collisional timeslots, C(A
i
).
) ( ) ( ) ( ) (
i i i i
A C A R A E A T + + = (2)
Proof: The number of tag-to-reader signals in a timeslot can-
not be negative. If the number of signals is zero or one, it is the
empty timeslot or the readable timeslot respectively. When a
timeslot carries two signals or more, it is the collisional time-
slot. Therefore, timeslots of a frame is one of the three.
A. No tag movement
At first, consider the case that any tags do not cross the
readers identification range during two consecutive frames. In
other words, A
i
which is the set of recognized tags in the ith
frame equals to A
i+1
which is the set of recognized tags in the
i+1th frame.
Lemma 2: Let C
binary
(A
i
) be the number of collisional time-
slots when the terminal-slot number of the reader and the allo-
cated-slot numbers of A
i
are initialized with 0 at the beginning
of ith frame. C
binary
(A
i
) is

1
0
2
1
1
2
1
2
1
1 1 2 ) (
n
k k
k
n
k
k
i binary
n
A C

(3)
where n is the number of tags of A
i
, |A
i
|.
Proof: When the terminal-slot number and the allocated-slot
numbers are initialized with 0, the operation of ABS is as same
as one of the binary tree protocol. Therefore, timeslots of the
378
frame can be represented in a binary tree and the total number
of the collisional timeslots of the frame is

=
=
0
) , ( ) (
k
binary i binary
k n C A C (4)
where C
binary
(n,k) is the number of collisional timeslots in the
depth k of the binary tree. Since the total number of nodes in
the depth k of the binary tree is 2
k
, by lemma 1,
( ) ( ) ( ) k n R k n E k n C
binary binary
k
binary
, , 2 , = (5)
where E
binary
(n,k) and R
binary
(n,k) are the number of empty time-
slots and the number of readable timeslots in the depth k of the
binary tree, respectively. The probability that the mth timeslot
of the depth k, s
k,m
(1 m 2
k
) is the empty timeslot is the
probability that n tags tries to transmit at s
k,1
,s
k,2
, ,s
k,m-1
,s
k,m+1
,
, s
k,2^k-1
, and s
k,2^k
. E
binary
(n,k) is
n
k
k
binary
k n E

=
2
1
1 2 ) , ( (6)
If A
i
has n tags, A
i
consists of tag a
1
, a
2
, , a
n-1
and a
n
. The
probability that s
k,m
is a readable timeslot is the probability that
only a
j
(1 j n) transmits in s
k,m
and tag a
1
, a
2
, , a
j-1
, a
j+1
,
, a
n-1
and a
n
tries to transmit at s
k,1
,s
k,2
, ,s
k,m-1
,s
k,m+1
, ,
s
k,2^k-1
, and s
k,2^k
.
1 1
2
1
1
2
1
1
2
1
2 ) , (

=
n
k
n
k k
k
binary
n n k n R (7)
By (5), (6) and (7),

=
1
2
1
1
2
1
2
1
1 1 2 ) , (
n
k k
n
k
k
binary
n k n C (8)
By substituting (8) into (4), lemma 2 is derived.
Theorem 1: For any A
i
,

+ =

1
0
2
1
1
2
1
2
1
1 1 2 2 1 ) (
n
k k
k
n
k
k
i binary
n
A T

(9)
where n = |A
i
|.
Proof: Every intermediate node of the binary tree is only the
collisional timeslot because only a node of a collisional time-
slot has two child nodes. Therefore, the tree is the full binary
tree and the total number of nodes of the tree is 2C
binary
(A
i
) + 1.
By (4),

=
+ =
+ =
0
) , ( 2 1
) ( 2 1 ) (
k
binary
i binary i binary
k n C
A C A T

(10)
From (8), T
binary
(A
i
) is proved.
Theorem 2: When A
i+1
is equal to A
i
and the reader and tags
maintain their terminated-slot number or allocated-slot num-
bers at the beginning of i+1th frame, the number of timeslots
required to identify A
i+1
, T
ABS
(A
i+1
|A
i
) is
1 1 1
) ( ) | (
+ + +
= =
i i i i ABS
A A R A A T (11)
Proof: The timeslot allocation procedure and the empty
timeslot elimination procedure make n tags transmit by follow-
ing the identification sequence. There are neither unused allo-
cated-slot numbers nor common allocated-slot numbers. The
frame has only readable timeslots.
B. Arriving tags
Consider that the set of tags which the reader needs to rec-
ognize expands. New tags, i.e., tags which are not recognized
in the previous frame, come into the vicinity of the reader from
outside.
Theorem 3: When A
i
is {a
1
, a
2
, , a
n-1
, a
n
} and A
i+1
is {a
1
,
a
2
, , a
n+-1
, a
n+
}, T
ABS
(A
i+1
|A
i
) is
( )

+ =

=
+
+

k k
k
k
k
i i ABS
n A A T
2
1
1
2
1
1
2
1
1 1 2 2 ) | (
0
1
1
(12)
Proof: ABS operates the timeslot allocation procedure for
a
n+1
, a
n+2
, , a
n+-1
and a
n+
in the i+1th frame. For simplicity,
we assume that a
n+1
, a
n+2
, , a
n+-1
and a
n+
initialize their
allocated-slot number with 0. Tag a
1
, a
2
, , a
n-1
and a
n
, i.e., n
tags which have recognized in the ith frame, exploit their allo-
cated-slot numbers which are determined in the ith frame.
Therefore, the n-1 tags of A
i
except the tag of which allocated-
slot number is 0 transmit in the readable timeslots. The tag of
which allocated-slot number is 0 in the ith frame and a
n+1
, a
n+2
,
, a
n+-1
and a
n+
has the same allocated-slot number, 0.
) 1 ( 1 ) | (
1
+ + =
+

binary i i ABS
T n A A T
(13)
From (9), the number of timeslots for arriving tags is calculated.

C. Leaving tags
Next, consider that some recognized tags go out of the
readers identification range and there are no arriving tags.
Theorem 4: When A
i
is {a
1
, a
2
, , a
n-1
, a
n
} and A
i+1
is A
i
-
{a
f(1)
, a
f(2)
, , a
f(-1)
, a
f()
} (1 f(x) n), T
ABS
(A
i+1
|A
i
) is
i i i ABS i i ABS
A A A T A A T = =
+
) | ( ) | (
1
(14)
Proof: One frame is required in order that ABS operates the
empty timeslot elimination for nonexistent tags. Therefore, the
number of required timeslots of the following frame after some
tags go out is equal to one of the last frame.
D. Tag movements
Finally, we derive the average delay when tags move ran-
domly.
Theorem 5: When A
i
is {a
1
, a
2
, , a
n-1
, a
n
} and A
i+1
is {a
1
,
a
2
, , a
n+-1
, a
n+
} - {a
f(1)
, a
f(2)
, , a
f(-1)
, a
f()
} (1 f(x) n),
T
ABS
(A
i+1
|A
i
) is
( )

+ =

=
+
+

k k
k
k
k
i i ABS
n A A T
2
1
1
2
1
1
2
1
1 1 2 2 ) | (
0
1
1
(15)
379
Proof: If B
1
is {a
1
, a
2
, , a
n-1
, a
n
}, B
2
is {a
n+1
, a
n+2
, ,
a
n+-1
, a
n+
}, and B
3
is {a
f(1)
, a
f(2)
, , a
f(-1)
, a
f()
}, by theorem 3
and theorem 4, the number of timeslots in the i+1th frame is
( 1 1
) | ( ) | (
2 1
1 2 1 1 3 2 1
+ + =
+ = +
B T B
B B B T B B B B T
binary
ABS ABS
)
IV.
A.
B.
C.

(16)
By theorem 4, the number of timeslots is not affected by the
number of leaving tags.

PERFORMANCE EVALUATION
We evaluate the performance of ABS compared to the bi-
nary tree protocol and the query tree protocol. To measure effi-
ciency of identifying tags in the tree based protocols, we con-
sider the following aspects.

Identification delay: This metric is the total delay re-
quired to recognize all tags. We measure the delay by
the timeslot because three protocols consist of time-
slots of which each has a time period for carrying tag-
to-reader signals and a time period for carrying the
reader-to-tag signal. The fast identification is the most
significant factor in the tree based anti-collision proto-
cols because they do not cause tag starvation problem.
Tags communication overhead: This metric is the av-
erage number of bits transmitted by each of the tags. A
tag of the tree based protocols transmits with its own
ID. The tags communication overhead influences the
amount of power consumption. Due to the lack of the
power source of tags, it must be low.
In our simulations, a single reader tries to recognize tags in
its identification range. Tags have randomly selected IDs. The
signals of the tags include only their IDs. The reader transmits
a startFrame signal at the beginning of a frame and an end-
Frame signal at the end of a frame. Tags which receive start-
Frame signal can try to transmit their IDs during that frame.
The lengths of the readers signals are different according to
the used protocol. In a timeslot of the binary tree protocol and
ABS, tags transmit first and then the reader responses. The
reader of the binary tree protocol transmits only one bit in order
to indicate whether collision occurs or not. Similarly, the reader
of ABS transmits two bits whose values match the empty, read-
able, or collisional timeslot respectively. On the other hand, the
reader, in the query tree protocol, transmits first and then tags
response it with their IDs. The readers signal contains one
prefix (the maximum length of the prefix is equal to the length
of tags ID). Therefore, the timeslot period of the query tree
protocol may be, relatively, longer than the period of other pro-
tocols. The notations of Table I are used for description of the
simulation result.
No tag movement
First, we simulate the performance with changing the num-
ber of tags in the readers range when a set of tags does not
become different between two consecutive frames. In other
words, there are neither any leaving tags nor any arriving tags.
All the tags are in the vicinity of the reader and enable to di-
rectly communicate with the reader. The reader has recognized
all the tags in the first frame and recognizes the tags in the sec-
ond frame again. Fig. 4(a) and Fig. 5(a) show the identification
delay and the tags communication overhead in the second
frame. Note that all the tags are recognized because three tree
based protocols do not cause tag starvation problem. As the
number of tags is increased, the delay gets longer and the tag
transmits more. The binary tree protocol imposes a heavy
communication overhead on a tag through the binary tree pro-
tocol and the query tree protocol show almost similar identifi-
cation delay. This implies that the binary tree protocol has
more colliding tags in each collisional timeslot than the query
tree protocol. ABS degenerates very slowly and takes the
shortest time for identifying. In addition to that, the tags of
ABS transmit very small bits because the timeslot allocation
procedure and the empty timeslot elimination procedure regu-
late communications between the reader and tags without both
empty and collisional timeslots.
TABLE I.
NOTATIONS
Description
B
ID
The number of bits of the tags ID
N
all
The total number of tags
N
last
The number of tags recognized in the last frame
N
in
The number of tags which arrive in the readers range
after the beginning of the last frame and before the
beginning of the new frame
N
out
The number of tags which are recognized in the last
frame and leave before the beginning of the new
frame
T
a
The total number of timeslots required for tag identifi-
cation in the new frame
T
r
The number of readable timeslots in the new frame (T
r

= N
last
+ N
in
N
out
)
T
c
The number of collisional timeslots in the new frame
T
e
The number of empty timeslots in the new frame
B
tx
The average number of bits transmitted by a tag in the
new frame
Arriving tags
Fig. 4(b) and Fig. 5(b) are the simulation results when none
of tags recognized in the last frame go out of the readers range
and the number of new arriving tags varies. After having rec-
ognized 500 tags in the first frame, the reader recognizes 500
tags and new arriving tags in the second frame. The arriving
tags set their allocated-slot number with 0. As the number of
arriving tags increases, the identification delays of three tree
based protocols are increased by the same incremental ratio.
This results from the same number of empty timeslots and col-
lisional timeslots created among new tags. ABS, however, has
less delay and less overhead because of neither the empty time-
slots nor the collisional timeslots among tags which have been
recognized in the first frame.
Leaving tags
Fig. 4(c) and Fig. 5(c) give the identification delay and the
tags communication overhead when there are no arriving tags
and the number of tags which go out of the readers range vari-

380



a) B
ID
=96, N
all
=1000, N
in
=0, N
out
=0



b) B
ID
=96, N
all
=1000, N
last
=500, N
out
=0



c) B
ID
=96, N
all
=1000, N
last
=500, N
in
=0




Figure 4. Identification delay



a) B
ID
=96, N
all
=1000, N
in
=0, N
out
=0



b) B
ID
=96, N
all
=1000, N
last
=500, N
out
=0



c) B
ID
=96, N
all
=1000, N
last
=500, N
in
=0




Figure 5. Tags communication overhead

381
es. There exist 500 tags in the simulation area. After the reader
has recognized all the tags in the first frame, some leaving tags
do not transmit any signals since then. As the number of leav-
ing tags increases, the identification delay of the binary tree
protocol and the query tree protocol decrease. (Note that the
binary tree protocol and the query tree protocol do not exploit
any information from the last frame.) On the other hand, the
reader, in ABS, does not know nonexistent tags in the follow-
ing frame as soon as some tags have leaved. Since the states of
the timeslots for nonexistent tags change into the empty time-
slots, the total number of timeslots is not reduced in the second
frame by the leaving tags. The empty timeslot elimination pro-
cedure is operated in the second frame and then the unneces-
sary empty timeslots for the leaving tags are removed. When
about 300 tags of 500 tags go out, ABS causes larger delay
than other protocols. However, when both the leaving tags and
the arriving tags are considered, ABS has better efficiency even
though the majority of the tags go out.
D.
V.
Tag movements
Next, we vary the number of tags when tags randomly move.
As shown Fig. 6, ABS has the best performance of three tree
based protocols. As the number of tags recognized by the
reader is increased, the gap of the identification delay between
ABS and other protocols is much larger. The identification
delay is similar with the result when the tags do not go out of
the readers range although ABS causes a little longer identifi-
cation delay when a set of tags decreases rapidly as known in
the previous simulation. The tags communication overhead of
ABS is much smaller like other scenarios. The high perform-
ance of ABS results from efficiently reducing the number of
the empty timeslots and the number of the collisional timeslots.
ABS is able to eliminate 72% of collisional timeslots and 65%
of empty timeslots of the binary tree protocol and the query
tree protocol.

a) Identification delay
b) Tags communication overhead
Figure 6. Tags move randomly (N
last
, N
in
, and N
out
are determined ran-
domly and B
ID
=96)
CONCLUSION
In this paper, an adaptive binary splitting collision arbitration
protocol for passive RFID tags has been proposed and evalu-
ated. We develop a novel and enhanced binary tree protocol to
reduce identification delay by exploiting information obtained
in the last process of identification. The timeslot allocation
procedure assigns the collision-free timeslots to each of tags
and the empty timeslot elimination procedure removes unnec-
essary timeslots. The simulation results show that the timeslot
allocation procedure and the empty timeslot elimination proce-
dure of ABS largely diminish the identification delay.
ACKNOWLEDGMENT
This work was supported by grant No. R01-2005-000-
10267-0 from Korea Science and Engineering Foundation in
Ministry of Science and Technology.


REFERENCES
[1] S. Sarma, D. Brock, and D. Engels, Radio Frequency Identification and
the Electronic Product Code, IEEE Micro, v.21 n..6, p. 50-54, Novem-
ber 2001.
[2] D. W. Engels and S. E. Sarma, The Reader Collision Problem, in
Proceedings of IEEE International Conference on System, Man and Cy-
bernetics, Hammamet, Tunisie, October 2002.
[3] J. Waldrop, D. W. Engels, and S. E. Sarma, Colorwave: An Anticolli-
sion Algorithm for the Reader Collision, In proceedings of IEEE Inter-
national Conference on Communications (ICC), Anchorage, Alaska,
USA, May 2003.
[4] H. Vogt, Efficient Object Identification with Passive RFID Tags, in
Proceeds of the International Conference on Pervasive computing, p. 98-
113, April 2002.
[5] Information technology automatic identification and data capture tech-
niques Radio frequency identification for item management Air inter-
face - Part 6: Parameters for Air interface communications at 860-960
MHZ, Final Draft International Standard ISO 18000-6.
[6] Draft protocol specification for a 900 MHz Class 0 Radio Frequency
Identification Tag, Auto-ID Center, Feb 2003.
[7] EPC
TM
Radio-Frequency Identification Protocols Class-1 Generation-2
UHF RFID Protocol for Communications at 860MHz-960MHz Version
1.0.8, EPCglobal, December 2004.
382
[8] T. A. Scharfeld, An Analysis of the Fundamental Constraints on Low
Cost Passive Radio-Frequency Identification System Design, M.S. the-
sis, Massachusetts Institute of Technology, p. 92-100, August 2001.
[9] D. R. Hush and C. Wood, Analysis of Tree Algorithms for RFID Arbi-
tration, in Proceedings of IEEE International Symposium on Informa-
tion Theory, p. 107-, 1998.
[10] M. Jacomet, A. Ehrsam and U. Gehrig, Contactless Identification De-
vice with Anticollision Algorihm, in Proceedings of IEEE Conference
on Circuits, System, Computers and Communications, p. 269-273, Ath-
ens, Greece, July 1999.
[11] C. Law, K. Lee, and K.-Y. Siu, Efficient memoryless protocol for tag
identification. in Proceedings of the 4th International Workshop on
Discrete Algorithms and Methods for Mobile Computing and Commu-
nications, p. 75-84, Boston, Massachusetts, USA, August 2000.
[12] F. Zhou, C. Chen, D. Jin, C. Huang, and H. Min, Evaluating and Opti-
mizing Power Consumption of Anti-Collision Protocols for Applications
in RFID systems, in Proceedings of the International Symposium on
Low Power Electronics and Design, Newport Beach, California, USA,
August 2004.
[13] J. Myung and W. Lee, An Adaptive Memoryless Tag Anti-Collision
Protocol for RFID Networks, IEEE Conference on Computer Commu-
nications (INFOCOM-2005), Poster Session, Miami, Florida, USA,
March 2005.


383

You might also like