Optimum Frame-length Configuration in Passive RFID

Systems Installations

M. V. Bueno-Delgado, J. Vales-Alonso, E. Egea-López and J. García-Haro

Information Technologies and Communications Department, Polytechnic University of

Cartagena, Plaza del Hopsital nº1, Cuartel de Antiguones, Cartagena, 30202, Spain

Abstract. Anti-collision mechanisms in RFID, including current standards, are

variations of Aloha and Frame Slotted Aloha (FSA). The identification process

starts when the reader announces the length of a frame (in number of slots).

Tags receive the information and randomly choose a slot in that cycle to

transmit their identifier number. The best performance of FSA always requires

working with the optimum frame-length in each cycle. However, it is not a

parameter easy to adjust in real RFID readers. In this work a markovian

analysis is proposed to find the optimum value of frame-length for the current

readers of the market. Besides, to validate and contrast the analytical results, a

real passive RFID system has been used to get experimental results: the

development kit Alien 8800. The experimental results match the analysis

predictions.

1 Introduction

In passive RFID systems, the communication between readers and tags takes place in

a shared communication channel. When the number of tags in coverage area is high, a

medium access mechanism (MAC) is needed to minimize the collisions that happen

by the simultaneous transmissions. Since passive tags hardware is too simple, the

complexity of the anti-collision protocol must rely on the reader.

Anti-collision mechanisms in passive RFID, including current standards, are

variations of Aloha and Frame Slotted Aloha (FSA) [1]. In FSA the time is divided

into frames (identification cycles) and these are in its turn subdivided into slots.

The identification process starts when the reader announces the length of a frame

or cycle (in number of slots). Tags in coverage receive the information and randomly

choose a slot in that cycle to transmit their identifier number. The FSA throughput

depends on the relationship between the number of tags to identify and the frame

length. If the number of tags in coverage is much more higher than the number of

slots, the identification time is considerable increased because a lot of collisions may

occur and a potentially large number of cycles are necessary to all tags in coverage

successfully dump their data. On the other hand, if the number of tags in coverage is

low and the number of slots to compete is high, a lot of empty slots will succeed, and

this also increases the identification time. The best performance of anti-collision pro-

tocols based on FSA always requires working with the optimum number of slots

V. Bueno-Delgado M., Vales-Alonso J., Egea-López E. and García-Haro J. (2009).

Optimum Frame-length Conﬁguration in Passive RFID Systems Installations.

In Proceedings of the 3rd International Workshop on RFID Technology - Concepts, Applications, Challenges , pages 72-78

DOI: 10.5220/0002203300720078

 SciTePress

(frame-length) in each cycle. However, it is not a parameter easy to adjust in real

RFID readers. Depending on the level of frame-length configuration, the readers

available in the market can be classified as:

• Readers with static and fixed frame, without user configuration [2-6].

Identification cycles are fixed and set up by the manufacturer. It is not

possible to modify by the user (it is usually fixed to 16 slots). Therefore,

these readers are not able to optimize the frame-length.

• Readers with static and fixed cycle with user configuration [6-8]. Before

starting the identification procedure the user can configure the frame length,

choosing between several values, which depend on the manufacturer. Then,

the identification cycle cannot be changed. If the user wants to establish a

different value of frame-length, it is necessary to stop the identification

procedure and restart with the new value of frame-length.

• Readers with dynamic cycle [6-8]. The user only configures the frame-

length for the first cycle. Then the frame-length is self-adjusted trying to

adapt to the best value in each moment, following the standard proposal [9].

We are interested in the testing scenario in which a fixed and known number of tags

N enter in coverage simultaneously. The main goal of this paper is to comparatively

study two different situations:

1. Systems based on a reader with static cycle length. A markovian analysis is

proposed to find the optimum value of frame-length which minimizes the

total identification time.

2. Systems based on a reader with a dynamic frame-length. We assume that the

reader knows in every cycle the number of contending tags (not identified in

previous cycles). Ideally, it is well-known that the frame- length must be

made equal to the number of contending tags to maximize the system

throughput. Nevertheless, the standard constraints the feasible frame-lengths

to a reduced set of possibilities: the natural numbers power of two from 2

. Our study explores this problem, to obtain the feasible frame-length

which maximizes the throughput in each cycle, minimizing the total

identification time.

The results permit to calculate the optimum value of the frame-length for the two

types of readers shown above. Besides, to validate and contrast the analytical results,

a real passive RFID system has been used to get experimental results: the

development kit Alien 8800 [9]. Different frame-length values have been configured

to get the total identification time with several populations of tags. The experimental

results match the analysis predictions.

The rest of the paper is organized as follows: Section 2 introduces a brief

description about related works. Section 3 shows the markovian analytical study of

the standard, oriented to readers with static frame. Section 4 describes the

experimental results of the real passive RFID System used. Finally section 5 resumes

the main conclusions extracted of this work.

2 Related Work

A relevant set of performance studies has been conducted in the last years for anti-

collision protocols based on FSA for passive RFID systems [10-18]. Most of them

propose variations to the EPCglobal Class-1 Gen-2 standard or new algorithms to

improve the frame adaptation. In [12] the authors propose that if the number of slots

per frame and the number of tags in coverage is known, other standard parameters

can be modified to reduce the identification time. However, nowadays it is not possi-

ble to find commercial readers which permit the tuning of those configuration pa-

rameters. In [13-14] the authors propose a set of anti-collision protocols that require

extra hardware in the tags, which increases the final cost of the RFID systems.

Some works have proposed improvements of the adaptive frame algorithm of EP-

Cglobal Class-1 Gen-2. Some of these add a heuristic to estimate the number of tags

that compete in each cycle [15-18]. In this way, the reader is able to get a more accu-

rate when has to establish the optimal frame-length in each cycle. However, these

works do not take into account that, when these estimation algorithms are imple-

mented with the standard EPCglobal Class-1 Gen-2, the optimum number of slots per

cycle must be adjusted to the power-of-two values permitted by the standard. There-

fore, the results of these works do not correspond to a real behavior of a commercial

reader.

3 Performance Analysis

In this section we describe the markovian analysis for systems with a fixed length

cycle. Our analysis is based on a variation of the analysis in [19]. The identification

process of a RFID system is modeled as (homogeneous) Markov process {X

}, where

} denotes the number of tags unidentified at the s identification cycle. Assuming N

tags enter the coverage area of the reader, the state space of the Markov process is

denoted as {N, N-1,..., 0}.The probability distribution of

indicates that the number

of busy slots with exactly r tags is:

rmrNmKG

irN

),,(

)(

−−

∏

−

⎟

⎠

⎞

⎜

⎝

⎛

⎟

⎠

⎞

⎜

⎝

⎛

(1)

where m=0,...,k and:

() }

∑

−

∏

−

−+=

⎥

⎦

⎥

⎢

⎣

⎢

⎭

⎬

⎫

⎩

⎨

⎧

⎩

⎨

⎧

⎟

⎠

⎞

⎜

⎝

⎛

ivl

jvl

MvlMG

)(

)1(),,(

(2)

In [19] all tags in coverage compete in every identification cycle. We modify and

adapt the analysis proposed in [19] to the case of the EPCglobal Class-1 Gen-2 stan-

dard. Therefore, tags identified in a cycle will not compete in the following ones. The

transition matrix H and transition probabilities are given by:

⎪

⎩

⎪

⎨

⎧

∑

−

+≤<−=

−

otherwise

Kijiij

iNK

),(

(3)

where i =0,…,N. We assume static scenario, where there are not new tag arrivals.

Therefore, the Markov chain is absorbent. If the identification process starts in a non

absorbent state v

(there are not tags identified), the number of steps up to the absor-

bent state (average number of identification cycles)

D is equal to the sum of the

entries of the i-th row of D matrix, which is denoted by:

)(

−

−= FID

(4)

D is the fundamental matrix of the absorbent chain. I is the identity matrix. F is the

submatrix of H with non-absorbent states of transition matrix H (see [20]). Table 1

shows the analytical results with different values of number of slots per cycle (K=2

)

and number of tags. This analysis has been computed with Matlab tool [21].

Table 1. Average number of identification cycles.

Number of slots (K) = 2

Tags(N) 4 8 16 32 64

8.2 3.67 2.44 1.89

1.54

60 8.56 4.11 2.76

2.15

630 19.6 6.15 3.60

2.61

8159 49.4 8.97 4.47

3.06

1.1 10

138 13.03 5.424 3.465

1.6 10

413.9 19.3 6.50 3.90

2.5 10

1304.2 29.41 7.76 4.32

3.8 10

4244.6 46.0 9.26 4.77

6 10

14127 73.81 11 5.23

As Table 1 shows, if we only compare the average number of identification cycles, as

the number of slots per cycle increases, the results are better. However, this evalua-

tion criterion is not significant because each identification cycle has not the same time

duration. The duration depends on the number of slots per cycle, that is, the frame-

length. Table 2 shows the same analysis results but measured as the average number

of slots per frame. In this case, the average number of slots to identify all tags de-

pends on, not only the number of tags that compete to be identified, but also the num-

ber of slots per cycle established.

Table 2. Average number of slots.

Number of slots (K)=2Q

Tags(N)

4 8 16 32 64

32.8

29.36 39.04 60.48 98.56

240 68.48

65.76 88.32 137.6

2520 156.8

98.4 115.2 167.04

32636 395.2 143.52

143.04 195.84

4.4 10

1104 208.48 173.56 221.76

6.4 10

3311.2 308.8 208 249.6

470.56 248.32 276.48

1.5 10

3.3 10

736 296.32 305.28

2.4 10

1.1 10

352 334.72

It is important to emphasize that EPCglobal Class-1 Gen-2 can work in dynamic

mode, where the number of slots (frame-length) can change cycle by cycle. An empty

slot or a collision slot has not the same duration that a successful slot or a data slot.

To check if the criteria assumed to evaluate if the total number of slots is adequate,

we must show the results in terms of average identification time. This is calculated as:

[

]

ididccvv

total

TkTkTkT ⋅+⋅+⋅⋅≈

(5)

where

k ,

k and

k are the average number of empty, collision and successful

slots. T

is the duration of an empty slot, T

the duration of a collision slot and T

the duration of a slot with a valid data transmission. Since the slot length depends on

the data length sent by the tag, we always suppose the maximum duration, it means,

all tags always transmit a complete EPC code of 96 bits. The time slot depends on the

parameters of the devices employed. In this work we use the default parameters of the

standard specifications [9]. We observed the following results: T

= 2.505·ms and T

=0.575 ms. Since an empty and a collision slot have the same temporal duration,

the equation (5) is simplified:

[

]

ididccv

total

TkTkkT ⋅+⋅+⋅≈ )(D

(6)

where,

∑

−

)(

(7)

is the number of contention slots in the cycle s and n

, 0≤ n

≤N, is the number of

tags identified in each cycle s before the absorption (

D ). The results of average

identification time are shown in Table 3. If we compare Table 2 and 3, it can check

that the criterion to evaluate the total number of slots is valid and the results are ana-

logs. Since the number of operations involved in the last formulas (3) and (4) increase

exponentially with the number of tags and slots, the use of them is limited to low

values of competing tags (N). In order to extend the results of the above analysis, we

have developed a simulator of RFID systems. The objective is to get the average

number of slots when a potentially large number of tags compete to be identified. The

simulator has been developed with the OMNeT++(Objetive Modular Network Test-

bed in C++) tool. The following sections present the results obtained by means of

simulation [22].

Table 3. Average identification time.

Número de slots (K) = 2

Tags(N)

4 8 16 32 64

0.0379

0.0354 0.0395 0.0494 0.0647

0.1742 0.0772

0.0738 0.0869 0.1110

1.5902 0.1476

0.1126 0.1198 0.1443

18.409 0.3011 0.1564

0.1555 0.1830

270.37 0.7223 0.2151

0.1910 0.2176

2.06·10

2.0219 0.2914 0.2316 0.2498

4.23·10

6.3209 0.3988 0.2732 0.2851

94.5·10

19.758 0.5662 0.3209 0.3219

71.2·10

65.765 0.8340 0.3753 0.3598

3.1 Scenario 1: Reader with Static Cycle

In the first scenario we simulate a system with a configurable reader with static cycle.

It permits to configure the number of slots per cycle with the values of Q

∈

[0,..., 15].

We have simulated different population of tags in coverage. Fig. 1 shows the average

number of slots per cycle, as a function of the number of tags in coverage. One curve

is plotted for each value of Q parameter. Each curve defines an interval where the

number of slots wasted to identify the population of tags is minimal with respect to

the other curves. To get the optimum configuration, we have to check the limits of

these intervals, that is, the intersection points with the other curves. From Fig 1 we

get the Table 4 where we show the values of these intersections, and the Q value

associated.

20 40 60 80 100 120 140 160

100

200

300

400

500

600

700

Tags in the coverage area (N)

Average number of slots

Q=3, 8 slots

Q=4, 16 slots

Q=5, 32 slots

Q=6, 64 slots

Q=7, 128 slots

Q=8, 256 slots

Q=9, 512 slots

Fig. 1. Scenario 1. Identification rate vs number of tags for different values of Q.

Table 4. Optimum Q that minimizes the average number of slots.

Optimal Q Number of slots (K)= 2

Tags in coverage (N)

1 2

≤

2 4

≤

N < 8

3 8

≤

N < 19

4 16

≤

N < 38

5 32

≤

N < 85

6 64

≤

N < 165

7 128

165

≤

N < 340

8 256

340

≤

N < 720

9 512

720

≤

N < 1260

10 1024

1260

≤

N < 2855

11 2048

2855

≤

N < 5955

12 4096

5955

≤

N < 12124

13 8192

12124

≤

N < 25225

14 16384

25225

≤

N < 57432

15 32768

57432

≤

3.2 Scenario 2: Reader with Dynamic Cycle

Currently, most of readers incorporate the adaptive algorithm proposed by the EP-

Cglobal Class-1 Gen-2 standard [9]. This algorithm works to get the Q value that

maximizes the identification rate, that is, the number of tags identified per slot. This

result is useful in scenarios where the reader has a priori knowledge about the num-

ber of tags that compete in each cycle. In this case, we can get the Q value that maxi-

mizes the identification rate. We use the simulator to evaluate this metric, establishing

the same value of tags competing in each cycle. Hence, we get the average identifica-

tion rate for different Q values. From these simulations we get the Fig. 2 where we

check that, for different values of Q, the maximum identification rate is 0.36, that is,

the best theoretical value of FSA. To get a better view in Fig 2, we only show the

values between Q

∈

[2,...,7]. From Fig 2 we extract the Table 5 where we indicate

the optimum Q value that maximizes the number of identifications in a cycle. The

limits established in each Q value are the intersections between two consecutives

curves of Q.

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Tags in the coverage area (N)

Identification rate

Q=2, 4 slots

Q=3, 8 slots

Q=4, 16 slots

Q=5, 32 slots

Q=6, 64 slots

Q = 7, 128 slots

Fig. 2. Scenario 2. Identification rate vs number of tags for different values of Q.

Table 5. Optimum Q that maximizes the identification rate per cycle.

Optimal Q Number of slots (K)= 2

Tags in coverage (N)

1 2

≤

2 4

≤

N < 4

3 8

≤

N < 9

4 16

≤

N < 20

5 32

≤

N < 42

6 64

≤

N < 87

7 128

≤

N < 179

8 256

179

≤

N < 364

9 512

710

≤

N < 1430

10 1024

1430

≤

N < 2920

11 2048

1430

≤

N < 2920

12 4096

2920

≤

N < 5531

13 8192

5531

≤

N < 11527

14 16384

11527

≤

N < 23962

15 32768

23962

≤

4 Experimental results: RFID System Alien 8800

An experimental validation of the results has been conducted. We have used the pas-

sive RFID kit Alien 8800 [9]. It works with EPCglobal Class-1 Gen-2 and its FSA

anti-collision protocol in UHF band (868-929MHz). The kit is composed by two

circular polarization antennas, installed face to face, at two meters of distance. One

acts as transmitter and the other one as receiver. The reader only permits to choose

among Q

∈ [4,..., 7]. For each Q value and different populations of tags we have

measured the total identification time because this is the only parameter that the Alien

reader gives us. Each experiment has been repeated up to 100 times. The results are

shown in table 6. The experimental results are close to the analysis and simulation

results. We have to take into account that the ambient conditions can affect to the

final experimental results.

Table 6. Average identification time in experimental results.

Number of slots (K) = 2

Tags(N)

8 16 32 64

0.041 0.042 0.067 0.089

0.091

0.089 0.911 0.142

0.157

0.139 0.143 0.163

0.340

0.172 0.192 0.189

0.823 0.243

0.201 0.241

2.021 0.311

0.253 0.279

5.213 0.414 0.298

0.296

12.25 0.697

0.348 0.349

53.66 0.921 0.394

0.379

5 Conclusions

In this work we analyzed the optimum configuration of frame-length (Q parameter) in

different RFID readers, attending to the population of tags to identify. We have stud-

ied two operation modes in RFID readers that correspond to the readers of the real

market: readers with fixed cycle and readers with frame-length adjusted dynamically.

Our results show that there are significant differences between the two scenarios

studied and permit to calculate the optimum value of the frame-length in each sce-

nario. For instance, if we work with a reader with static cycle and we set Q=5, we get

that the better response will be with a population of tags between 20 and 42. On the

contrary, in a reader with dynamic cycle, the same Q value determines that the popu-

lation should be between 38 and 85 tags.

Acknowledgements

This research has been supported by project grant DEP2006-56158-C03—03/EQUI,

funded by the Spanish Ministerio de Educación y Ciencia, project TEC2007-67966-

01/TCM (CON-PARTE-1), funded by the Spanish Ministerio de Industria, Turismo y

Comercio, project TSI-020301-2008-16 (ELISA) funded by the Spanish Ministerio

de Industria, Turismo y Comercio, project TSI-020301-2008-2 (PIRAmIDE) funded

by the Spanish Ministerio de de Industria, Turismo y Comercio, and it is also devel-

oped within the framework of "Programa de Ayudas a Grupos de Excelencia de la

Región de Murcia, de la Fundación Séneca, Agencia de Ciencia y Tecnología de la

RM (Plan Regional de Ciencia y Tecnología 2007/2010).

References

1. Wieselthier, J. E., Ephremides, A., Michaels, L. A., “An exact analysis and performance

evaluation of framed ALOHA with capture”, IEEE Transactions on Communications, vol.

37(2), pp. 125-137, 1988.

2. Symbol available online in: http:// www.tecno-symbol.com/.

3. ThingMagic Mercury4 available online in: http://www.thingMagic.com.

4. Caen available online in: http://www.caen.it/rfid/index.php.

5. Awid available online in: http://www.awid.com.

6. Samsys available online in: http://www.samsys.com.

7. Intermec Reader available online in: http://www.intermec.es,

psfiles.intermec.com/eps_files/eps_brochure/RFIDReader_brochure_web.pdf.

8. Alien Reader: http://www.alientechnology.com/readers/index.php.

9. EPCglobal: Class 1 Generation 2 UHF Air Interface Protocol Standard Version 1.0.9: Gen

2. Online: http://www.epcglobalinc.org/standards.

10. Kawakita, Y., Mitsugi, J. “Anti-collision performance of Gen 2 Air Protocol in Random

Error Communication Link”, Proc. Int. Symposium on Applications on Internet Work-

shops, pp. 68-71, 2006.

11. Mitsugi, J., Yumoto, Y., Hada, H., Murai, J. “Auto-ID Labs. Activities and collaborations

in wireless technology for the harmonized deplyment of UHF RFID system”, Auto-ID Labs

Research Workshop, Zurich, 2004.

12. Joe. I, Lee, J. “A novel Anti-Collision Algorithm with Optimal Frame Size for RFID Sys-

tem”, Fifth International Conference on Software Engineering Research, Management and

Applications, 2007.

13. Jacomet, M., Ehrsam, A., Gehring, U. “Contactless Identification device with anticollision

algorithm”, en Proc. of IEEE Conference on Circuits, Systems, Computers and Communi-

cations, pp. 269273. Athens, Greece, July 1999.

14. Jihoon Myung, Wonjun Lee. “Adaptative Binary Splitting: A RFID tag Collision Arbitra-

tion Protocol for tag Identification”. Mobile Networks and Applications Journal, vol. 11,

pp. 711722. May 2006.

15. G. Khandelwal, L. Kyounghwan, A. Yener, S. Serbetli, “ASAP: A MAC Protocol for dense

and Time Constrained RFID Systems”, EURASIP Journal on Wireless Communications

and Networking, vol. 2007, Article ID 18730, 2007.

16. J. Cha, J. Kim, “Novel Anti-collision Algorithms for Fast Object Identification RFID Sys-

tem”, en Proc. of the 11th Conference on Parallel and Distributed Systems, vol. 2, pp. 63-

67, 2005.

17. C. Floerkemeier, “Transmission control scheme for fast RFID object identification”, en

Proc. of the Fourth Annual IEEE International Conference on Pervasive Computing and

Communications Workshops (PERCOMW'06), pp. 457-462, Pisa, Italy, March 2006.

18. C. Floerkemeier, M. Wille, “Comparison of Transmission schemed for framed ALOHA

based RFID protocols”, in International Symposium on Applications and the Internet

Workshops (SAINT'06)}, pp. 92-95, Phoenix, Ariz, USA, January 2006.

19. Vogt, H., “Efficient Object Identification with Passive RFID Tags”, Lecture Notes in Com-

puter Science, vol. 2414, pp. 98-113, 2002.

20. Norris, J. R., Markov Chains, Cambridge University Press, 1997.

21. Matlab available online in: http://www.mathworks.com

22. A. Vargas, “The OMNeT++ Discrete Event Simulation System”, European Simulation

Multiconference ESM 2001, Prague (Czech Republic), June 2001.