An rfid anti collision algorithm with dynamic condensation and ordering binary tree

RFID simulation An rfid anti collision algorithm with dynamic condensation and ordering binary tree An rfid anti collision algorithm with dynamic condensation and ordering binary tree An rfid anti collision algorithm with dynamic condensation and ordering binary tree

An RFID anti-collision algorithm with dynamic condensation

and ordering binary tree

Department of Information Management, National Taiwan University of Science and Technology, No 43, Sec 4, Keelung Road, Taipei, Taiwan

a r t i c l e i n f o

Article history:

Received 26 February 2013

Received in revised form 20 July 2013

Accepted 2 September 2013

Available online 11 September 2013

Keywords:

RFID

Anti-collision algorithm

Blocking algorithm

Condensation

a b s t r a c t

In many RFID applications, the reader repeatedly identifies the same staying tags Existing anti-collision protocols can rapidly identify the staying tags by remembering the order in which the tags were recog-nized in the previous identification process This paper proposes a novel protocol, dynamic blocking adaptive binary splitting (DBA), based on the blocking mechanism, which prevents the newly-arriving tags from colliding with the staying tags Moreover, DBA utilizes a dynamic condensation technique to reduce the number of idle slots produced when recognized tags leave Following the condensation pro-cess, multiple staying tags may be required to share the same slot, and thus may cause collisions among them Accordingly, an efficient ordering binary tree mechanism is proposed to split the collided tags deterministically according to the order in which they were recognized The analytical and simulation results show that DBA consistently outperforms previous algorithms in all of the considered environments

Ó 2013 Elsevier B.V All rights reserved

1 Introduction

Radio frequency identification (RFID) technology enables

ob-jects to be identified more rapidly and conveniently than traditional

bar-code mechanisms In general, an RFID system consists of a

reader1 and multiple tags, which communicate with one another

over wireless channels Each tag has a unique identification number

(UID), and thus the reader is able to identify all of the tags located

within its neighborhood To identify the tags, the reader emits a

trig-ger signal and then waits for the tags to respond However, if multiple

tags transmit their UIDs simultaneously, their signals collide at the

reader and thus the reader is unable to identify any of the tags and

the collided tags must retransmit their UIDs The collisions not only

delay the identification process but also waste the available

band-width Consequently, in improving the performance of RFID systems,

it is essential to design efficient anti-collision schemes

Broadly speaking, existing anti-collision algorithms can be

clas-sified as either aloha-based [1–9] or tree-based[10–17]

Aloha-based algorithms estimate the number of unidentified tags within

the interrogation zone of the reader and then allocate an

appropri-ate number of slots such that the risk of collisions is reduced By

contrast, tree-based algorithms, e.g., binary tree (BT)[10–13]and

query tree (QT)[14–17], continually divide the collided tags into

two subsets until each set contains at most one tag In general, alo-ha-based algorithms provide an effective means of reducing colli-sions at the beginning of the tag identification process, while tree-based algorithms are effective in avoiding the starvation prob-lem[18], in which a specific tag is not identified for a long time

In many RFID applications, the reader repeatedly identifies the same tags since they do not move out of the interrogation zone between one identification process and the next, i.e., conducting roll calls during a session or monitoring audiences during a show

In general, if the anti-collision algorithm can keep track of all the tags recognized in the previous identification process (i.e., in the previous frame), many collisions in the current identification pro-cess (i.e., the current frame) can be avoided and the identification process completed more rapidly as a result The adaptive binary splitting (ABS) algorithm [18,19] and adaptive query splitting (AQS)[18,20]algorithm, modified from the BT and QT schemes, respectively, both utilize this approach to improve the efficiency

of the identification process In both algorithms, the reader asks the tags already recognized in the previous frame to transmit their UIDs using individually-assigned slots such that collisions among them are avoided However, while both algorithms avoid collisions among the tags which stay within the range between successive identification processes (referred to henceforth as stay-ing tags), they do not prevent collisions between the staystay-ing tags and newly-arriving tags since the latter are allowed to transmit their UIDs using the same set of slots allocated to the recognized tags To resolve this problem, Lai and Lin proposed a single reso-lution blocking (SRB) algorithm[21], in which the transmission of

0140-3664/$ - see front matter Ó 2013 Elsevier B.V All rights reserved.

⇑Corresponding author.

E-mail address: laiyc@cs.ntust.edu.tw (Y.-C Lai).

1

The reader is formally called as the read/write device (RWD) because the data in

the RFID system are transmitted in both directions However, this paper uses the

reader because most previous studies use this term.

Contents lists available atScienceDirect Computer Communications

j o u r n a l h o m e p a g e : w w w e l s e v i e r c o m / l o c a t e / c o m c o m

the staying tags and the newly-arriving tags were constrained to

different slots; thereby preventing collisions between them As a

result, a lot of collisions can be circumvented, which speeds up

the identification process In a later study, Lai and Lin

proposed a pair resolution blocking (PRB) algorithm [22], in

which the time required to re-identify the recognized tags was

halved by using a pair resolution technique to couple the tags

identified in the previous frame However, the PRB algorithm

be-comes unreliable when channel errors prevent the reader from

decoding the signals of one or more of the tags In such a

situa-tion, PRB misinterprets the decoding failure as a collision between

two staying tags, and yields an incorrect recognition of the tag

UIDs as a result

Although the SRB algorithm reduces the number of collisions

by assigning different ranges of slots to the previously recognized

tags and the newly-arriving tags, respectively, blocking

algo-rithms such as SRB tend to be wasteful of the slots reserved for

the recognized tags Specifically, slots may be reserved for

recog-nized tags which subsequently move out of the interrogation

zone before the following identification frame As a result, the

slots are left idle Accordingly, the present study proposes an

en-hanced form of the SRB algorithm, designated as dynamic

block-ing ABS (DBA) based on a dynamic condensation technique for

reducing idle slots and an ordering binary tree mechanism for

collision resolution The dynamic condensation scheme adjusts

the number of slots reserved for recognized tags dynamically in

accordance with the estimated number of staying tags That is,

when a greater number of tags are expected to move out of

inter-rogation zone, fewer slots are reserved for the recognized tags in

the following frame, and vice versa However, reducing the

num-ber of slots increases the risk of collisions among the staying tags

Therefore, an ordering binary tree scheme based on the order in

which the tags were identified in the previous frame is used to

split the collided tags continuously into two subsets such that

the collisions can be rapidly resolved

The remainder of this paper is organized as follows: Section2

reviews the BT, ABS and SRB algorithms Section3describes the

overall concept and detailed operations of DBA Section4presents

a formal analysis of the identification delays in the BT, ABS, SRB

and DBA algorithms Section5compares the performance of DBA

with that of ABS and SRB using both analytical and numerical

methods, and describes the additional cost of DBA Finally,

Sec-tion6presents some brief concluding remarks

2 Related work

This section reviews the basic operations of the BT, ABS and SRB

algorithms To facilitate the discussions, the following terms are

first defined

 Frame: An identification process comprising multiple slots of

interaction between the reader and the tags for the purpose of

identifying all of the tag UIDs Let fidenote the i-th frame

 Slot: A cycle in which the tags transmit their UIDs to the reader

and the reader responds with a feedback message of collision,

readable, or idle, if multiple tags, one tag or no tag transmit their

UID(s), respectively Let si,jbe the j-th slot in the i-th frame

 Arriving tags in i-th frame: The tags do not appear in fi-1, but

appear in fi

 Staying tags in i-th frame: The tags appear in fi-1and also appear

in fi

 Leaving tags in i-th frame: The tags appear in fi-1, but disappear in

fi

 Possible tags in i-th frame: The tags appear and are recognized in

fi-1, and are likely to appear in fialso Possible tags are the com-bination of staying tags and leaving tags

On identification process, the reader and tags must write some information into memory to maintain their states For simplicity, previous studies all made several assumptions as follows (1) All operations in devices are perfectly correct (2) The time and power consumption of these operations are ignored (3) The information maintained by tags will be preserved until it is re-initialized or the power is exhausted

2.1 Binary tree (BT) algorithm The BT anti-collision algorithm uses a binary random number to divide the collided tags Every tag maintains a counter, which indi-cates the order in which it will be identified among all the unrec-ognized tags and whose value is initialized to zero Only those tags with a counter value of zero can transmit their UIDs to the reader When a collision slot appears, each of the collided tags sets its counter to a random binary number Hence, the collided tags can

be divided into two subsets The reader also has a counter to keep track of the maximal value among the counters of all the tags to determine when to terminate the frame That is, the counter plus one is the number of tag sets to be recognized Once the counter falls below zero, there are no more tags to be identified and thus the current frame terminates

Fig 1(a) illustrates the BT operation for the case of four tags, A,

B, C and D, which are to be identified in frame fi Initially, the four tags all have a counter value of zero and therefore transmit their UIDs in the first slot, si,1 Following the resulting collision, each of the collided tags randomly sets its counter to either zero or one such that the four tags are split into two subsets Assume that tags

A and D select 0 and form the first subset while tags B and C select

1 and form the second subset Since A and D have a counter value

of zero, they are permitted to transmit their UIDs in the next slot, i.e., si,2 In the slot si,2, for example, both A and D transmit their UIDs and a collision occurs, which fails to be resolved if both tags set their counters randomly as one, and further causes the idle slot

si,3and the collision slot si,4 The collision in the slot si,4is then re-solved successfully by selecting different counter values for the tags A and D, and thus they are respectively identified in the read-able slots si,5and si,6.Fig 1(b) shows the slot-by-slot details of the

BT operation, including the counter values of the reader and tags, the tags which transmit their UIDs, and the feedback message pro-vided by the reader

2.2 Adaptive binary splitting (ABS) algorithm The ABS algorithm, modified from BT, allows each tag recog-nized in the previous frame to remember the order in which it transmit its UID among all tags As a result, the transmission of the staying tag will not collide with each other in the current frame, leading that many unnecessary collisions can be avoided Importantly, once the frame is terminated, each tag remembers its own ASC while the reader remembers the TSC The ASCs of the recognized tags represent the order in which the tags were suc-cessfully identified in the frame and should lie within the range

of 0 to TSC without duplication Since the ASCs have different val-ues, they can be used in the following frame to prevent collisions among the staying tags Meanwhile, the arriving tags in the follow-ing frame can initialize their ASCs to a random number in the range

of 0 to TSC according to the TSC carried to the tags in a start

command sent by the reader Through sharing the slots between

the staying tags and the arriving tags, ABS can reduce the idle slots

due to the leaving tags at the cost of possible collisions between

the staying tags and the arriving tags

Fig 2shows the detailed operation of the ABS algorithm in two

successive frames, i.e., fiand fi+1 Note that an assumption is made

that no tags exist in frame fi1 As shown inFig 2(a), the ABS

pro-cedure in frame fi,is similar to that inFig 1(b) for the BT algorithm

When frame fiis terminated, tags A, B, C and D remember their

ASCs as 0, 2, 3 and 1, respectively, since they were identified in

the order of A(0)–D(1)–B(2)–C(3) In addition, the reader

remem-bers TSC as 3 which is the highest value of the ASCs among the four

tags In frame fi+1, assume that tags C and D move out of the

read-er’s interrogation zone, while two new tags, E and F, arrive (see

Fig 2(b)) As described above, newly-arriving tags select a random

number from 0 to TSC as their ASCs Suppose that in the present

example, tags E and F select ASC values of 0 and 2, respectively

Thus, in the first slot of frame fi+1, i.e., si+1,1, the transmissions of

tags A and E collide, prompting the two tags to adjust their ASCs

by adding a random binary number Suppose that tag A chooses

0 and tag E selects 1 Consequently, the reader successfully

identi-fies tags A and E in slots si+1,2and si+1,3, respectively (seeFig 2(c))

A similar procedure is performed for tags B and F; resulting in their

successful identification in slots si+1,9and si+1,8, respectively

2.3 Single resolution blocking (SRB) algorithm

Although the ABS algorithm prevents collisions among the

stay-ing tags, many collisions may still occur between the

newly-arriv-ing tags and the staynewly-arriv-ing tags Accordnewly-arriv-ingly, Lai and Lin proposed a

single resolution blocking (SRB) algorithm[21], in which separate

slots were allocated to the arriving tags and the staying tags,

respectively Specifically, in SRB each tag identifies itself as a

stay-ing tag or an arrivstay-ing tag by comparstay-ing its reader’s ID in the

previ-ous frame with the reader’s ID broadcasted by the reader at the

beginning of the current frame All of the staying tags keep the ori-ginal ASCs recorded in the previous frame while all of the arriving tags initialize their ASCs to a random number in the range of TSC + 1 to an extended TSC, i.e., TSCEXT SRB will determine an appropriate value of TSCEXT for the following frame based on the number of arriving tags in the present frame In this way, SRB not only avoids collisions among the staying tags (with unique ASCs ranging from 0 to TSC), but also prevents collisions between the staying tags and the newly-arriving tags (with randomly se-lected ASCs in the range of TSC + 1 to TSCEXT)

As in ABS, SRB also maintains two counters for each tag (i.e., PSC and ASC) and two counters for the reader (i.e., PSC and TSC) The operations of these counters are the same in both algorithms with the exception that SRB, being a blocking algorithm, uses additional parameters to maintain separate slots for the staying tags and the arriving tags, respectively, i.e.,

 rRID and tRID: The reader stores its unique ID, denoted as rRID, while each tag also stores the reader’s ID, denoted as tRID

 ArrNum: The reader uses this counter to record the number of newly-arriving tags in each frame

 ANEst: The reader uses the exponential average of ArrNum to estimate the number of newly-arriving tags in the following frame

 TSCEXT: The reader maintains this counter to represent the maximum index of slots initially reserved for possible tags (i.e., both staying and newly-arriving) at the start of each frame That is, slot numbers from 0 to TSCEXT are reserved, where slots

in the range of 0 to TSC are used by staying tags while slots in the range of TSC + 1 to TSCEXT are used by arriving tags For illustration purposes, assume that SRB is used to perform the identification process for the scenario shown inFig 2for frame

fi+1 The tree representations for the staying and newly-arriving tags are shown inFig 3(a), while the detailed operations of the

s i,4

s i,7

s i,1

s i,6

s i,3

1 0

s i,2

s i,11

Readable Idle Collision

s i,8 s i,9

1 0

1 0

(a) Application of binary tree algorithm in frame f i

tag

Feedback message

(b) Slot-by-slot binary tree procedure in frame f i

Fig 1 Illustrative example of binary tree anti-collision algorithm.

SRB algorithm are shown inFig 3(b) Assuming that the number of

newly-arriving tags is estimated to be 2, and thus TSCEXT is set

equal to 3 + 2 Assume that in randomly setting their ASCs between

TSC + 1 and TSCEXT, both newly-arriving tags (E and F) select a

va-lue of 4 Two idle slots exist among the first four slots since tags C

and D leave Furthermore, since tags E and F both have the same

value of ASC, a collision occurs in slot si+1,5 Thus, the two tags

are finally identified in slots si+1,6and si+1,7, respectively

Compar-ingFigs 3 and 2, it is seen that SRB eliminates the risk of collisions

between the staying tags and the arriving tags, and therefore is

able to produce fewer collision slots

3 Dynamic blocking ABS (DBA) algorithm

In the SRB algorithm, the staying tags and arriving tags are

as-signed to the slots in different ranges in order to prevent collisions

between them However, SRB reserves a slot for each possible tag;

resulting in a few idle slots if some of these tags actually leave

To minimize such waste, the present study proposes a dynamic

blocking ABS (DBA) algorithm based on a dynamic condensation mechanism and an ordering binary tree (OBT) collision resolution scheme In the proposed approach, the dynamic condensation mechanism adaptively adjusts the number of slots (referred to as condensed slots) in accordance with the estimated number of stay-ing tags, rather than reservstay-ing a slot for each possible tag How-ever, since the condensation process may result in some condensed slots being shared by multiple tags, collisions may oc-cur among the staying tags Accordingly, the OBT scheme divides the collided tags deterministically into two subsets in accordance with their ASC values remembered from the previous frame in or-der to resolve the collisions in a quick and efficient manner 3.1 Dynamic condensation mechanism

Fig 4illustrates the basic concept of the dynamic condensation mechanism Assume that there exist a total of n possible tags with ASC values in the range of 0 to n  1 Assume also that the number

of condensed slots is denoted as m In accordance with the

tag

Feedback message

s i+1,5

s i+1,3

s i+1,2

s i+1,6

s i+1,7

s i+1,9

s i+1,8

Readable Idle Collision

Slot

TSC

PSC

tag

Feedback message

Fig 2 Illustrative example of ABS anti-collision algorithm.

dynamic condensation mechanism, the n possible tags are

parti-tioned equally into m groups and each group of tags uses the same

condensed slot to transmit their UIDs Since the possible tags are

partitioned equally, each group contains n/m tags in average Thus,

each tag can determine the condensed slot to which it belongs

sim-ply by referencing its ASC, i.e., ASC/(n/m) Since the result of the

division process may not be an integer, the floor function

bASC=ðn=mÞc is applied Consider a hypothetical example in which

n = 18 and m = 5 In this case, the possible tags with ASC values of

0–3, 4–7, 8–10, 11–14 and 15–17 are assigned to the condensed

slots 0, 1, 2, 3 and 4, respectively Since there may be some tags

leaving, only a few tags among the possible tags stay in the current

frame and can transmit using the assigned slots

In general, implementing the dynamic condensation

mecha-nism requires the resolution of two problems, namely determining

an appropriate mapping of the staying tags to the condensed slots

(seeFig 4) and correctly estimating the number of staying tags In

practice, the number of staying tags is heavily dependent on the

number of possible tags Thus, the actual number of staying tags may vary significantly from one frame to another, and is difficult

to estimate as a result However, a detailed investigation shows that the staying ratio, i.e., the number of staying tags divided by the number of possible tags, generally remains stable from one frame to the next Thus, DBA uses an exponential averaging

meth-od to estimate the staying ratio Specifically, the estimated staying ratio in frame i, denoted as SREsti, is estimated as SREsti¼v SRi1þ ð1 vÞ  SREsti1, wherevis the exponential weight, while SRi1and SREsti1are the actual and estimated stay-ing ratios in frame fi1, respectively Having determined the esti-mated staying ratio, the estiesti-mated number of staying tags is then computed as SREst  n, where n is the number of possible tags Intuitively, it seems reasonable to expect that the number of condensed slots should be set equal to the estimated number of staying tags However, this approach does not lead to the optimal identification efficiency since the OBT scheme can resolve the collisions among the staying tags very effectively Thus, for any

Tag D leaves

Tag C leaves

s i+1,1

Tag A

s i+1,5

s i+1,7

s i+1,6

Readable

Idle

Collision

s i+1,3

Tag B

s i+1,8

No arriving tag selects this slot

At identifying staying tags At identifying arriving tags

Blocking Mechanism

Tag E Tag F

tag

Feedback message

Fig 3 Illustrative example of SRB anti-collision algorithm.

Condensed ASC

Condensed slots

Possible tags

Fig 4 Basic concept of dynamic condensation mechanism.

estimated staying ratio, it is necessary to establish the optimal

con-densation ratio which minimizes the overall identification delay

Fig 5shows the variation of the optimal condensation ratio with

the staying ratio Note that the results are obtained using a binary

search method based on values of the estimated staying ratio in the

range of 0–1 (in incremental steps of 0.01) and identification delay

is computed using the analytical method presented in (8) of

Sec-tion4 It is observed that the optimal condensation ratio varies

with a ladder-like characteristic as the value of SREst is increased

In implementing the DBA algorithm proposed in this study, the

re-sults presented inFig 5(a) are listed as 101 entries in a table

map-ping the estimated staying ratio to the optimal condensation ratio,

and this table is then stored at the reader.Fig 5(b) shows a

simpli-fied version of the table for SREst units of 0.1 Having computed

SREst, the reader locates the nearest staying ratio within the table

and retrieves the corresponding optimal condensation ratio The

appropriate number of condensed slots, m, is determined by

mul-tiplying the optimal condensation ratio by the total number of

pos-sible tags

3.2 Ordering binary tree mechanism

Due to the condensation of the slots, the risk of collisions among

the staying tags is increased since the condensed slots may be

shared by multiple possible tags The collisions can be solved using

the traditional BT approach However, the OBT mechanism

pro-posed in this study resolves the collisions in a more efficient

man-ner by splitting the collided staying tags deterministically in

accordance with their unique ASC values Specifically, given a

col-lision among the staying tags with ASCs ranging from x to y, OBT

divides the collided tags into two groups, namely one group

con-taining the tags with ASCs ranging from x to bðx þ yÞ=2c and

an-other group containing the tags with ASCs ranging from

bðx þ yÞ=2c þ 1 to y This deterministic partitioning approach

re-sults in a faster separation of the collided staying tags than the

ran-dom partitioning approach used by BT As a result, the overall

identification delay is effectively reduced

To achieve the deterministic partitioning of the collided tags in

a distributed manner, OBT must maintain some additional

param-eters When a collision occurs, each tag involved in the collision

should use its original ASC value in the previous frame to

deter-mine whether it belongs to the first group or the second group

However, the value of the ASC is initially condensed in the dynamic

condensation process and then subsequently updated during the

identification procedure in the ongoing frame Thus, in DBA, each

tag not only maintains the ASC for the current frame like SRB,

but also records the original ASC (denoted as OASC here) which

is the final value of ASC of the tag in the previous frame and

re-mains fixed during the current frame Meanwhile, it maintains

two additional parameters, GL and GH, which are set equal to the lowest and highest OASCs, respectively, among all the tags within the group to which it belongs Let GM be the middle of the two parameters, i.e., GM ¼ bðGL þ GHÞ=2c In the event of a collision, a collided tag with an OASC value less than or equal to GM retains its ASC value and adjusts its GH parameter to GM Conversely, a col-lided tag with an OASC value larger than GM increases its ASC by one and adjusts its GL to GM + 1 Thus, the collided tags can be split into two groups based on OASC, GL and GH in a distributed manner Fig 6illustrates the detailed operations of the OBT mechanism Assume that eight staying tags with OASC counter values of 0–7 (i.e., the order recognized in the previous frame) are assigned to

a single condensed slot and initialize their ASCs to 0 Thus, in the first slot of the frame, i.e., si+1,0, all of the tags transmit their UIDs

to the reader simultaneously To resolve the resulting collision, the tags partition themselves into two groups by comparing their OASC values with the GM value (equal to 3 in the present example) The tags with OASCs in the range of 0–3 retain their ASC values (i.e., ASC = 0), while those with OASCs in the range of 4–7 increase their ASC by one (i.e., ASC = 1) The values of GL or GH for each tag are then adjusted accordingly Thus, the collided tags are split into two groups, i.e., one group containing tags with ASC = 0 and a sec-ond group containing tags with ASC = 1 The splitting procedure is repeated iteratively in this way until all of the collided tags are resolved

3.3 DBA procedure

In both SRB and DBA, each tag must have the ability to deter-mine whether it is a staying tag or an arriving tag In SRB, this is achieved by means of the reader’s ID only However, this approach may cause confusion when a tag leaves the interrogation zone of the reader and then re-enters the interrogation zone several frames later without entering the interrogation zone of another reader in the meantime Since the value of tRID is unchanged in this case, the tag interprets itself as a staying tag However, the reader should in fact regard this tag as a newly-arriving tag because the reader did not recognize the tag in the previous frame Thus this tag will pos-sibly collide with the actual staying tag in the current frame, lead-ing to a wrong interpretation DBA resolves this problem by uslead-ing both the reader ID and a frame number parameter (tFN) to distin-guish the staying tags from the arriving tags In the proposed ap-proach, each tag sets its tFN to the current frame number when

it has been recognized In a later frame, the tag compares its tFN with the current frame number received from the reader, i.e., rFN If tFN + 1 is equal to rFN, the tag infers that it was recognized

in the last frame, and therefore identifies itself to be a staying tag However, if the tag leaves the interrogation zone of the reader and then returns in a later frame without entering the interrogation

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

SREst

SREst Optimal condensation

ratio

Fig 5 Mapping of optimal condensation ratio vs estimated staying ratio.

zone of another reader in the meantime, the value of tFN + 1 is not

equal to rFN Thus, the tag interprets itself as an arriving tag even

though its tRID value is equal to rRID

As in the SRB algorithm, DBA uses a blocking mechanism to

pre-vent collisions between arriving tags and staying tags

Conse-quently, it inherits all of the SRB parameters, i.e., PSC, ASC, TSC,

TSCEXT, tRID, rRID, ArrNum and ANEst (see Section 2.3) Also

DBA maintains the following additional parameters:

 rFN and tFN: The frame numbers stored in the reader and tag,

respectively

 StayNum: A parameter used by the reader to count the number

of staying tags

 SREst: The estimated staying ratio calculated by the reader

based on the exponential average of the actual SRs in earlier

frames

 SR and CR: The staying ratio (SR) and condensation ratio (CR)

calculated by the reader when applying the dynamic

condensa-tion mechanism

 OASC, GL and GH: Parameters maintained by the tags when

applying OBT mechanism

Fig 7presents the pseudo code of the DBA algorithm Note that

Fig 7(a) presents the pseudo code for the reader operations, while

Fig 7(b) presents that for the tag operations At the beginning of

each frame, the reader calculates SREst, CR, the new condensed

TSC and TSCEXT in accordance with the staying ratio (SR) and

the number of arriving tags (ArrNum) calculated in the previous

frame (see lines 7, 8, 9 and 10 inFig 7(a)) The reader then

trans-mits a start command containing TSC, TSCEXT, CR, rRID, and rFN to

all of the tags within interrogation zone (see line 14) On receiving

this command, each tag compares the received rRID and rFN values

with their own tRID and tFN values If either of the values does not

match, the tag interprets itself as an arriving tag Thus, it set its ASC

to a random number in the range of TSC + 1 to TSCEXT (see line 7 in

Fig 7(b)) and then updates its tRID and tFN counters once it has

been successfully identified (see lines 34–35 inFig 7(b)) If the pair

(tRID, tFN) equals the received pair (rRID, rFN), the tag interprets

itself as a staying tag, and already has an appropriate ASC

determined in the previous frame In the dynamic condensation procedure, the tag condenses this ASC as bASC  CRc in order to determine the condensed slot in which it should attempt to trans-mit its UID Additionally, the tag sets its GL and GH counters to the lowest and highest OASCs in the group, respectively, for the pur-pose of resolving collisions using OBT procedure In the event of

a collision among the staying tags, OBT is applied to resolve the collision using the procedures described in lines 25 to 31 in Fig 7(a) Note that the manipulations of the PSC and TSC counters

in the reader and tags are very similar to those in SRB, and thus a discussion of the related procedures inFig 7(a) and (b) is deliber-ately omitted here

3.4 Illustrative example of DBA procedure Fig 8shows the DBA identification process in frame fi+1for the example considered inFig 3involving two staying tags (A and B) and two arriving tags (E and F) Assume that the optimal conden-sation ratio, CR, corresponding to the estimated staying ratio, SREst, is found from the mapping table to have a value of 0.5 Fur-thermore, assume that the estimated number of arriving tags, AN-Est, is equal to 2 At the beginning of frame fi+1, the reader issues a start command containing TSC ¼ b3  0:5c ¼ 1, TSCEXT ¼ TSC þd0:88  2e ¼ 3, CR = 0.5, rRID and rFN to all four tags After receiving this command, each tag checks whether it is a staying tag or not by comparing its tRID and tFN values with those received from the reader The staying tags, i.e., A and B, then condense their ASCs from 0 and 2 to 0 and 1, respectively, by applying the conden-sation ratio (i.e., CR = 0.5) The successful identification of tags A and B requires only two slots, si+1,1and si+1,2, because in this case the two tags are assigned to different slots in accordance with their new ASCs following condensation, and thus no collision occurs be-tween them Comparing the results obtained using the DBA algo-rithm with those obtained using SRB (seeFig 3(a)), it is found that the dynamic condensation mechanism results in the saving

of the two idle slots generated by leaving tags C and D, respec-tively Regarding the arriving tags, E and F set their ASCs to random numbers in the range of TSC + 1 = 2 to TSCEXT = 3 Assume that both tags set their ASC to a value of 2 Thus, arriving tags E and F

GL=0

ASC=0

OASC=7 OASC=6

OASC=5 OASC=4

ASC=1

ASC=0

GL=0 GH=1 GL=2 GH=3

ASC=1

OASC=7 OASC=6

OASC=5

ASC=2

OASC=3 OASC=2

GL=2 GH=3

ASC=2

OASC=4

OASC=7 OASC=6

OASC=5

ASC=3

OASC=4

Si+1,0

Si+1,3

Si+1,2

Si+1,1

Fig 6 Use of OBT scheme in resolving collisions among staying tags.

collide in slot si+1,3, but are finally identified in slots si+1,4and si+1,5,

respectively, (say) following the random resolution by the

conven-tional BT method

4 Performance analysis

This section presents a formal analysis of the average

identifica-tion delay in the ABS, SRB and DBA algorithms, respectively All

three algorithms are based partially on the BT scheme Thus, for

convenience, the section commences by deriving the average

iden-tification delay of BT, even though the corresponding derivation is

already available in previous studies[18,19]

4.1 BT

Let the total number of slots required in a frame to identify n

tags be denoted as DBT(n) At the start of the identification process,

the n tags collide in the first slot and are randomly divided into two

subsets containing i tags and n  i tags, respectively The

probabil-ity that i tags out of n tags are selected through binary random

tests follows a binomial distribution B(N, P), i.e., Prði; N; PÞ

¼ ðNiÞPið1  PÞNi with parameters N = n and P = 1/2 Each subset

is split repeatedly until the resulting sets contain either no tags

or just one tag, i.e., DBT(0) and DBT(1), and require an idle slot or

a readable slot, respectively Thus DBT(n) can be expressed as the following recursive function,

i¼0

i

 

ð1Þ

4.2 ABS Assume that n tags are recognized in the previous frame and that the numbers of arriving tags and staying tags in the current frame are denoted asaand b, respectively In ABS, an individual slot is assigned to each of the n possible tags, and thus a total of

bslots are occupied by staying tags, while the remaining n  b slots are empty For each of the b occupied slots, if i arriving tags select the slot, the slot will contain a total of 1 + i tags, and thus D (1 + i)

/*Dynamic Blocking ABS Reader Operation*/

rRID, rFN

/* Dynamic Blocking ABS Tag Operation*/

CR, rRID, rFN

Fig 7 Pseudo code of DBA anti-collision algorithm.

slots are required to achieve collision resolution using the BT

scheme For each of the n  b empty slots, if the slot is selected

by i arriving tags, the slot will contain i tags, and therefore DBT(i)

slots are required to achieve collision resolution Since each

arriv-ing tag selects any one of the n slots with an equal probability of

1/n, the probability that i out of theaarriving tags select a specific

slot follows a binomial distribution B(N, P), with parameters N =a

and P = 1/n Therefore, the average identification delay for ABS is

given as

i¼0

n

i¼0

n

4.3 SRB

Since SRB is a blocking algorithm, the staying tags and arriving

tags are initially assigned to the slots of two separate regions,

cor-responding to the ranges of 0 to TSC and TSC + 1 to TSCEXT,

respec-tively Assume there are n recognized tags in the previous frame

Then in the current frame, the first region contains n slots allocated

to n possible tags, and thus the identification delay is simply equal

to n (n = TSC + 1 here) In the second region, given an estimated

number of arriving tags equal to ^a, the optimal number of slots

allocated for these arriving tags is equal to d0:88^ae[21,22] Since

each arriving tag selects any one of these d0:88^ae slots with equal

probability, the probability that i out of theaarriving tags select a

specific slot follows a binomial distribution B(N, P), with

parame-ters N =a and P ¼ 1=d0:88^ae Therefore, the identification delay

of SRB is equal to

i¼0

The minimal identification delay is obtained under perfect esti-mation conditions, i.e., ^a¼a Thus, the optimal value of the SRB identification delay, denoted as D

SRBðn;a;bÞ, is given as

i¼0

4.4 DBA

As with SRB, DBA is also a blocking algorithm, and thus the identification process is again performed on the slots in two sepa-rate regions The identification delay in the second region of the frame is the same as that for SRB However, the identification delay

in the first region of the frame requires elaborate calculation As-sume that m condensed slots are reserved for n possible tags, i.e., the number of possible tags for each condensed slot (denoted as k) is equal to n/m For analytical convenience, suppose that k is

an integer When b staying tags are randomly selected out of n pos-sible tags, the probability that a specific condensed slot has exactly

i staying tags (out of k possible tags for this slot) is equal to

ðkiÞðnkbiÞ=ðnbÞ If i > 1, then i staying tags in the condensed slot collide and need to be resolved using the OBT mechanism Let DOBT(k, i) de-note the number of slots required to resolve the collisions of i tags among k possible tags Therefore, the identification delay in the first region of slots, DDBA1(n,a, b), can be calculated as

DDBA1ðn;a;bÞ ¼ mXk

i¼0

n

In accordance with the OBT operation, the i (i P 2) collided tags among the k possible tags in a condensed slot are split into two groups, namely one group containing j collided tags among bk=2c possible tags and a second group containing i  j collided tags

Tag D leaves

Tag C leaves Tag A

s i+1,3

s i+1,5

s i+1,4

Tag B

s i+1,6

No arriving tag selects this slot

s i+1,2

At identifying staying tags At identifying arriving tags

Blocking Mechanism

Tag E Tag F

Dynamic Condensation

Readable

Idle

Collision

Condensation-Staying tags

Condensation- Leaving tags

Slot

TSC

PSC

ASC value

Responding tag

Feedback message

Fig 8 Illustrative example of DBA anti-collision algorithm.

among dk=2e possible tags Since the probability for such a division

is equal to ðbk=2c

j Þðdk=2e

ij Þ=ðkiÞ, the total number of slots required for OBT collision resolution is given by

j¼maxð0;ik h Þ

ð kl

j Þð kh ij Þ

ð k

i Þ

1

0 B

1 C A;

2

 

2

 :

ð6Þ

In practical applications, k is unlikely to have an integer value

In this case, the m condensed slots can be divided into m1and

m2 slots, with each slot containing k1¼ bn=mc possible tags and

k2¼ dn=me possible tags, respectively m1and m2can be obtained

by solving the equations m1þ m2¼ m and m1k1þ m2k2¼ n The

identification delay in the first region of the frame, DDBA1(n,a, b),

can then be obtained as

DDBA1ðn;a;bÞ ¼ m1

i¼0

ð k1

i Þð nk1 bi Þ

i¼0

ð k2

i Þð nk2 bi Þ

ðnbÞ DOBTðk2;iÞ;

ð7Þ

In (5) and (7), the optimal value of the identification delay,

D

DBA1ðn;a;bÞ, is obtained when using the optimal value of m⁄,

i.e., the value of m for a perfect estimation of the staying ratio

As described in Section3.1, after obtaining the staying ratio, the

reader searches the mapping table to find the corresponding

opti-mal value of the condensation ratio (CR) corresponding to the

stay-ing ratio (SR) as depicted inFig 5, and m⁄can then be obtained by

n  CR The optimal DBA identification delay, D

DBAðn;a;bÞ, can then

be obtained as

DBA1ðn;a;bÞ þ mXa

i¼0

5 Simulation and performance comparison

In this section, the identification performance of DBA is

com-pared with that of ABS and SRB, respectively Note that BT is not

included for comparison purposes since it identifies all of the tags

from scratch in every frame In other words, it is not intended to

enhance the efficiency of the re-identification procedure, and

therefore inevitably achieves a much poorer performance than

either ABS or SRB[20–22] Showing the results of BT will cause that

the curves of ABS, SRB, and DBA shown in the figure are

indistin-guishable Also, currently no any aloha-based algorithm is

de-signed for the re-identification scenario Therefore, the

aloha-based algorithms, including frame slotted aloha-aloha-based algorithm

[1–9]and the Q-algorithm in EPCGen 2 tags[3,25], are not

com-pared in our simulation for the same reason above The

perfor-mance of the three algorithms is evaluated in terms of three

metrics, namely the number of collision slots, the number of idle

slots, and the total number of slots required to identify all the tags

Note that the number of readable slots is not considered since this

metric is the same for all three algorithms

The simulations commence by investigating the effect of the

number of staying tags and arriving tags on the identification

per-formance of the three schemes In performing the simulations, it is

assumed that the staying ratio and the number of arriving tags in

SRB and DBA are correctly estimated such that the optimal

identi-fication performance is obtained Let Z be the total number of tags

to be identified in the frame The identification performance of

each scheme in the i-th frame, fi, is then evaluated by varying the

staying ratio r and the arriving ratio r, where r is defined as the

ratio of the number of staying tags over n and rais defined as the ratio of the number of arriving tags over Z  n, where n is the num-ber of recognized tags in the (i  1)th frame

In the ABS, SRB and DBA schemes, the staying ratio and the number of arriving tags are both estimated using an exponential averaging method, and are therefore dependent on the mobility

of the tags Therefore, in the second series of simulations referred

to[21–24], the effects of tag mobility on the identification perfor-mance of the three schemes are systematically explored As the environment in [18,21,22], the simulations consider a total of Z mobile tags located randomly within an area measuring

10 m  10 m Each tag is assigned a stationary probability which determines whether or not it moves during the period of one frame Moreover, the tag velocity is defined as the distance moved

by each tag during one frame if it moves The reader is located in the center of the simulation environment and is assumed to have

an interrogation zone of just 3 m Hence, some tags enter and leave the interrogation zone as they move during the frame The initial position and direction of the tags are specified randomly within the simulation area If a tag touches the border of the simulation area, it randomly selects a new direction of travel Fig 9shows

an example of the simulation environment with 15 mobile tags

in a 10 m  10 m simulation area A reader, with 3 m interrogation zone, is located in the center All tags move according to the sta-tionary probability equal to 0.2 and the tag velocity equal to 1.5 m per frame InFig 9, 12 tags move with random directions while three tags do not move between two continuous frames The 12 moving tags includes 3 tags entering, 3 tags leaving, 2 tags staying in, and 4 other tags locating outside, the interrogation zone Three unmoving tags contain 1 tag staying in the interroga-tion zone and 2 tags locating outside the zone Thus, the reader has to recognize 6 tags, including 3 arriving tags and 3 staying tags

in the current frame

Each simulation is performed over a total of 106frames The simulations consider the effects of three specific parameters, namely the tag velocity and the stationary probability It is worthy

to be mentioned that all algorithms compared in this paper are based on BT, so their performance is not affected by the tag UID distribution Thus even when the UID distribution of leaving tags

is not uniformly distributed, the performance of these algorithms are not changed

5.1 Effect of arriving tags Fig 10shows the effect of the arriving ratio, ra, on the identifi-cation performance of the three schemes given a constant staying ratio of rs= 0.5 Note that Z = 1000 and n = 500 In general, the re-sults show that for every scheme, the identification delay increases with an increasing arriving ratio as the result of a greater number

of collision slots, idle slots and total slots, respectively.Fig 10(a) shows that ABS results in the greatest number of collision slots for most values of radue to the large number of collisions which occur between the arriving tags and the staying tags As the value

of raincreases, the probability of the arriving tags colliding with the staying tags also increases, and thus the performance gap be-tween ABS and SRB and DBA, respectively, increases Of the three schemes, SRB results in the lowest number of collisions since its blocking strategy ensures that collisions are limited only to the arriving tags In the DBA algorithm, the dynamic condensation mechanism produces additional collisions among the staying tags Thus, for a given value of ra, the number of collisions in DBA is greater than that in SRB However, since the probability of colli-sions among the arriving tags in DBA is the same as that in SRB and the dynamic condensation procedure only generates collisions among the staying tags, the performance gap between the two schemes is independent of the arriving ratio