The Practicality of Multi-Tag RFID Systems

⋆

Leonid Bolotnyy, Scott Krize and Gabriel Robins

Department of Computer Science, University of Virginia

Abstract. Radio Frequency Identiﬁcation (RFID) is an increasingly popular tech-

nology that uses radio signals for object identiﬁcation. Successful object identiﬁ-

cation is the primary objective of RFID technology (after all, the last two letters

of the acronym “RFID” stand for “identiﬁcation”). Yet, a recent major study by

Wal-Mart has shown that object detection probability can be as low as 66%. In

this paper we address the fundamental issue of improving object detection by

tagging objects with multiple tags. This conﬁrms for the ﬁrst time the practicality

and efﬁcacy of previous works on multi-tag RFID systems. Using different con-

ﬁgurations of commercial RFID equipment, we show signiﬁcant improvements in

object detection probability as the number of tags per object increases. We com-

pare various combinations of multi-tags, readers, and antennas, and demonstrate

that adding multi-tags to a system can improve object detection probabilities more

dramatically than adding more readers. We also address issues such as tag orien-

tation and variability, effects of multi-tags on anti-collision algorithms and on

object detection in presence of metals and liquids, as well as the economics of

multi-tags.

1 Introduction

Radio Frequency Identiﬁcation (RFID) uses radio communication to uniquely identify

objects [1] [2] [3]. A typical RFID system consists of readers (sometimes called bea-

cons), tags (sometimes called transponders), and back-end servers which receive and

process the information that the readers collect from the tags [4] [5] [6]. There are

two coupling mechanisms used by passive and semi-passive tags: inductive coupling

and electromagnetic backscattering (i.e., far-ﬁeld propagation). In inductive coupling

the reader creates a magnetic ﬁeld between itself and the tags which in turn derive

power from this magnetic ﬁeld. In far-ﬁeld propagation the reader sends a signal to a

tag and the tag backscatters (i.e., reﬂects) a response back to the reader. Some of the

major applications of RFID include supply chain management, inventory tracking, ac-

cess control, library book checkout, cattle tracking, passport tagging, and even games

[7] [8] [9] [10] [11].

When bar codes are used for object identiﬁcation, bar code scanners require line-of-

sight visibility of the bar codes, and they usually must be close to the objects. Bar codes

are scanned one at a time, and scanners need to physically move from one bar code to

⋆

This research was supported by a Packard Foundation Fellowship, by NSF Young Investigator

Award MIP-9457412, and by NSF grants CCR-9988331 and CCF-0429737. For additional

related papers see http://www.cs.virginia.edu/robins

Bolotnyy L., Krize S. and Robins G. (2007).

The Practicality of Multi-Tag RFID Systems.

In Proceedings of the 1st International Workshop on RFID Technology - Concepts, Applications, Challenges, pages 100-113

DOI: 10.5220/0002436201000113

 SciTePress

the next in order to read them. Since this is a mechanical process, the read rate is at best

only a few bar codes per second. RFID readers on the other hand, can read hundreds

of tags per second without a line-of-sight or mobility requirement, which allows easy

automation of the reading process and makes RFID-based identiﬁcation very appealing.

However, as the identiﬁcation process is automated, special care is required to ensure

the detection of all objects within the reader’s ﬁeld.

Ubiquitous background radio noise impedes RFID object detection. Moreover, met-

als and liquids tend to reﬂect and/or absorb radio signals, further degrading the reader’s

ability to achieve accurate and complete tag identiﬁcation. Missed items, even at a rel-

atively low rate of 1%, can result in large ﬁnancial losses for stores that rely on RFID-

enabled automatic checkout stations. This situation is real and serious, since milk, wa-

ter, juices, and canned / metal-foil -wrapped (i.e., Faraday caged) goods are commonly

stocked in markets. Practical experiments by Wal-Mart in 2005 showed 90% tag detec-

tion at case level, 95% tag detection on conveyor belts, and only 66% detection rate of

individual items inside fully loaded pallets [12].

To reduce the percentage of undetected items, we recently proposed tagging objects

with multiple tags [13], in contrast to previous works on RFID technology that assume

only a single tag per object. Although multi-tags will cost slightly more than single

tags, we experimentally demonstrate in this paper that multi-tags can be very beneﬁcial

for many applications where higher object detection rates are required. The beneﬁts of

attaching multiple tags per object include: (1) greater induced voltages/power aboard

some of the tags, (2) increased tag-reader communication range, (3) larger tag memory

per object, (4) enhanced security, and (5) improved overall detectability, availability,

reliability, and durability of the system [13]. This paper presents the ﬁrst-ever extensive

experimental study that validates the efﬁcacy of a multi-tag RFID system in practical

scenarios.

2 Optimal Placement of Multi-tags

Based on our previous theoretical results for multi-tags, object detection improvement

hinges on the expected grazing angle of the radio signal from the reader to the tag [13].

We performed angle analyses for two power transmission modes, inductive coupling

and far-ﬁeld propagation. Let β be the angle between the arriving signal’s direction (i.e.,

the B-ﬁeld) and the tag’s plane. In the case of inductive coupling, the induced voltage

aboard a tag is proportional to sin(β), and for far-ﬁeld propagation, the induced voltage

is proportional to sin

(β)[14][13][1].

The ﬁrst question is how to orient the tags relative to each other in order to max-

imize the expected grazing angle of the radio wave to one of the tags’ antennas. We

assume a uniform distribution for the signal arrival direction, since in many RFID ap-

plications the orientation of a tag’s antenna to the arriving signal can be arbitrary (e.g.,

products in a shopping cart, RFID-tagged cell phones, etc.). In the case of a single tag,

the tag can be positioned arbitrarily, since its orientation would not affect the expected

(uniformly distributed) signal arrival angle. For two tags, it is optimal to position them

perpendicular to one another in the x-y and x-z planes. Similarly, for three tags, we can

position them pair-wise perpendicularly in the x-y, x-z, and y-z planes. For four tags, it

101

Fig.1. Optimal multi-tag positioning for ensembles of 1, 2, 3 and 4 tags.

turns out that in order to maximize the expected signal incidence angle to at least one

of the tags, it is best to position them parallel to the faces of a tetrahedron, a platonic

solid

(see Figure 1).

The second question asks what is the actual expected maximum grazing angle of the

arriving signal with respect to the antennas of any of the tags, for a given tag ensemble.

To answer this question, we computed the expected incidence angle analytically for one

and two tags, and developed a software simulator that computes the expected angle for

an arbitrary number of tags. The results indicate a two-digit increase in the expected

grazing angle as we move from one tag to two tags, and also as we go from two tags

to three tags, but only a 3 degree average improvement as we move from three tags to

four tags. This suggests that adding an extra tag or two may be beneﬁcial to dramati-

cally improve object detection, but attaching the fourth tag to an object may not garner

substantial detection probability improvement. These improvement trends are indeed

corroborated by our experimental results discussed below.

Many objects are box-shaped, allowing multi-tags to be positioned perpendicular

to each other, but even irregular-shaped objects can beneﬁt from multi-tags, as demon-

strated by our experiments. RFID tags are available in many shapes and sizes, making

it feasible to attach multiple tags to any object.

3 Experimental Equipment and Setup

We performed our experiments using commercial FCC-compliant equipment, namely

Ultra High Frequency (UHF) readers from Alien Technology (model ALR-9800, four

antennas, multi-protocol, 915 MHz) and ThingMagic (model Mercury 4). We deployed

sets of linear and circular antennas from Alien Technology, and circular antennas from

ThingMagic. A single Alien Technology reader antenna can either broadcast or receive

signals, whereas the more versatile ThingMagic antenna can both send and receive sig-

nals. We used several types of tags from UPM Raﬂatac, the world’s leading RFID tag

manufacturer. In particular, we chose unipolar UPM Rafsec UHF tags “Impinj 34x54

ETSI/FCC” and bipolar UPM Rafsec UHF tags “Impinj 70x70 ETSI/FCC” for our ex-

periments.

For ﬁve or more tags, it becomes more complicated to analytically determine the optimal

relative positioning of the tags, except for speciﬁc special cases, such as for N=6 where the

tags should ideally be placed parallel to the faces of a dodecahedron, and N=10 where the tags

should be parallel to the faces of an icosahedron.

102

The experiments were conducted in an otherwise empty room, in order to minimize

radio reﬂection / interference anomalies. We multi-tagged a diverse set of 20 objects

using four tags per object. We positioned tags perpendicular to each other whenever

possible, and spread the tags far apart in space in order to minimize tag occlusion by

other tags and/or objects.

We ﬁrst describe our experiments involving the Alien Technology hardware, since

this equipment allows us to collect data for both circular and linear antennas. A similar

experimental setup was used with the ThingMagic equipment. In the discussions below,

we will implicitly assume that the Alien Technology equipment was used in each ex-

periment, unless explicitly stated that the ThingMagic hardware was used instead. Sim-

ilarly, all the experiments discussed below have used the unipolar UPM Rafsec UHF

tags “Impinj 34x54 ETSI/FCC”, unless explicitly stated that bipolar tags were used.

We positioned Alien Technology reader antennas side-by-side in pairs, with each

pair consisting of a sending and a receiving antenna. In one experiment we used linear

antennas, and in the other experiment we used circular antennas. We positioned each

pair of antennas 55 inches from the center of a plastic bag containing all 20 objects,

20.5 inches above the ﬂoor, and perpendicular to the bag. The reader was operating in

“inventory mode” using Gen-2 tag reading protocol. We allowed sufﬁcient time for the

reader to read all the tags within its range by performing many tag reads and maintaining

adequate timeouts between reads to make sure that the effects of the environmental

noise are minimized

We randomly (re)shufﬂed the tagged objects multiple times to change the tags’ ori-

entations with respect to the reader’s antennas in order to improve the statistical signif-

icance of the results (the values reported in the tables and graphs below are averages

over all object shufﬂings). We also varied the power emitted by the antennas, keeping

in mind that the distance at which tags can be detected is proportional to

√

power. We

performed our experiments for linear as well as for circular antennas using 7 different

power levels ranging from 25.6dBm to 31.6dBm, in increments of 1dBm.

4 Experimental Results

4.1 Linear Antennas

Our experiments show that multi-tags considerably improve object detection probabil-

ities for linear antennas. The detection probabilities for different numbers of tags per

object, different numbers of reader antennas, and various reader power levels are sum-

marized in Figure 2. This table shows that switching from one to two tags per object

produces a high double-digit increase in tag detection probability, and a low double-

digit increase when moving from 2 to 3 tags, but only single-digit increase from 3 to 4

tags. These results corroborate our theoretical expectations [13].

The multi-tagged objects included soap bars, cereal boxes, paper plates, plastic boxes, pack-

aged foods, clothing items, etc.

To enable others to reproduce our results, we specify here the Alien Gen-2 algorithm

parameter settings used in our experiments: TAG

TYPE = 16, ACQ G2 CYCLES = 10,

ACQ

G2 COUNT = 100, ACQ G2 Q = 2. Our source codes and scripts are available upon

request.

103

Fig.2. Detailed statistics showing the average detection probability for linear antenna(s) as a

function of the power level for different antenna conﬁgurations and for different numbers of tags

per object.

Figure 3(a) graphically shows the increase in object detection probability for each

object (the objects are sorted along the X-axis according to their detection probabili-

ties). Again, we observe signiﬁcant separations between the ﬁrst three curves. In Figure

3(b), we compare object detection improvements between two tags per object versus

two reader antennas. From this data we can see a dramatic double-digit improvement

from adding a second tag to each object, and only a low single-digit improvement from

adding a second reader, yielding almost a factor of 4 improvement in object detection

probability using multi-tags as compared to multi-readers.

4.2 Circular Antennas

The detection probability statistics for circular antennas are given in Figure 4. As with

linear antennas, experiments with circular antennas show a dramatic double-digit aver-

age improvement in object detection as the number of tags per object increases. How-

ever, the detection probabilities for circular antennas are higher than for linear ones,

since the orientation of objects with respect to the reader antennas varies widely. From

the comparisons of different numbers of multi-tags and multi-readers (Figure 5), we can

see that for circular antennas the advantage of adding a tag is on par with that of adding

a reader. Figure 6 gives two graphs that depict changes in object detection probability

as a function of power for various multi-tag combinations. The graphs show that the

average object detection probabilities decrease more rapidly for circular than for linear

antennas, as a function of decreasing antenna power.

5 Importance of Tag Orientation

One of the major claims made in our original theoretical paper on multi-tags [13] is

that tags need to be oriented perpendicular to each other to obtain the most beneﬁts in

object detection. Here we experimentally conﬁrm this claim by varying the tag orienta-

tion, collecting tag identiﬁcation data, and calculating object detection probabilities for

different multi-tag orientations. We performed experiments with unipolar tags (UPM

Rafsec UHF tag Impinj 34x54 ETSI/FCC) whose plane orientation matters, and with

bipolar tags (UPM Rafsec UHF tag Impinj 70x70 ETSI/FCC) whose plane orientation

has no effect on tag detection.

104

(a) Comparison of multi-tags

(b) Multi-tags versus multi-readers

Fig.3. (a) Average object detection probability improvements for linear antennas as the num-

ber of tags per object increases. (b) Comparisons of multi-tags with multiple readers for linear

antennas. Note that attaching multiple tags to an object yields higher average object detection

probabilities than adding more readers.

Fig.4. The statistics table shows the detection probability for circular antenna as a function of

the power level for different antenna conﬁgurations, and for different numbers of tags per object.

With unipolar tags we ran experiments comparing differently oriented pairs of tags.

One orientation which we call 180-same refers to two tags positioned on the same

105

(a) Comparison of multi-tags

(b) Multi-tags versus multi-readers

Fig.5. Figure (a) shows the improvement in object detection probability for each object for cir-

cular antennas as the number of tags per object is increased. Figure (b) compares multi-tags

with multiple readers for circular antennas. Attaching multiple tags to an object produces higher

object detection probability than adding more readers.

plane and having identical orientation. The second orientation 180-diff refers to two

tags positioned on the same plane, but one of the tags is rotated 90 degrees relative

to the orientation of the other tag. The third orientation 90-same refers to two tags

having identical orientation, but positioned on perpendicular planes. Finally, the forth

tag orientation 90-diff refers to two tags positioned on perpendicular planes with one

tag rotated 90 degrees relative to the other tag. In our experiments we compared these

four different tag orientations, and the results are presented in Figure 7(a). The results

show that tags perpendicular to each other yield a higher probability of detecting at least

one of them than tags that have identical orientation. In addition, to increase detection

probability, it is better to position tags on perpendicular planes, rather than to locate all

the tags in the same plane.

With bipolar tags we compared two possible tag orientations - 180, where tags are

positioned on parallel planes, and 90, where tags are positioned on perpendicular planes.

106

(a) Two linear antenna pairs

(b) Two circular antenna pairs

Fig.6. The two graphs show the detection probability for linear and circular antennas as a func-

tion of reader power for different numbers of tags per object. Observe that the detection proba-

bility decreases at a faster rate for circular than for linear antennas, as expected.

These are the only possibilities since tag orientations within the plane have no effect

on bipolar tag detection. The results of the experiments shown in Figure 7(b) demon-

strate no difference between tag orientations for omni-directional/circular antennas, but

a drastic advantage for perpendicular 90 tags over parallel 180 tags for directional/linear

antennas. These results show that multi-tags improve object detection not only because

they increase the total antenna size per object and decrease the probability of antenna

occlusions, but also because the expected grazing angle between the signal from the

reader and one of the tags increases, which in turn raises the expected power on-board

one of the tags. These ﬁndings conﬁrm our theoretical expectations.

6 Controlling the Variables

It is important in RF experiments to carefully isolate and control the variables in order to

ensure the accuracy of the results. Speciﬁcally, we controlled the effects of radio noise,

reader variability, tag variability, the number and type of reader antennas, reader power

107

(a) Unipolar tag orientation comparison.

(b) Bipolar tag orientation comparison.

Fig.7. The two tables comparing object detection probabilities for unipolar and bipolar tags

for different multi-tag orientations. The results show the signiﬁcance of perpendicular multi-tag

orientation, especially for directional/linear antennas. In Figure 7(a), 180-same refers to identi-

cally oriented tags positioned on parallel planes; 180-diff refers to perpendicularly oriented tags

positioned on parallel planes; 90-same refers to identically oriented tags positioned on perpen-

dicular planes; 90-diff refers to perpendicularly oriented tags positioned on perpendicular planes.

In Figure 7(b), 180 refers to tags positioned on parallel planes; 90 refers to tags positioned on

perpendicular planes.

level, and the distance from the reader antennas to the objects. To control the effect

of ambient radio noise, we ran our experiments multiple times, sometimes even across

multiple days to ensure that statistical properties of the data are stable. To accurately

calculate improvements in object detection with multi-tags, we allowed sufﬁcient time

for the reader to read the tags. The reader parameters were carefully selected to ensure

that all tags within a reader’s detectability range are read. To ensure that our results

are independent of the particular reader and antenna manufacturer/brand, we ran our

experiments with readers and antennas from two different manufacturers. In all of our

experiments we used consistent tag types and ensured that tag variability does not affect

our experiments. We will discuss tag variability further below. The reader and identical

reader antennas were carefully selected and objects were placed on a rotating platform

at a ﬁxed distance from the reader. The reader power levels were carefully controlled

via a parameter in the software driver.

6.1 Tag Variability

To determine tag properties and control tag variability we performed multiple tag vari-

ability tests. RFID tags with different chip manufacturers and antenna geometries have

108

different detectability/receptivity properties [15]. The importance of tag receptivity and

its use as a tag performance metric is addressed in [16]. Similarly, no two chips are

truly identical due to inherent VLSI manufacturing variations [17]. Indeed, we found

differences in tag detectability among tags of the same type, even among ones com-

ing from the very same tag roll. In fact, these inherent tag receptivity differences were

surprisingly high, with up to an order-of-magnitude difference in detectability between

the “best” and “worst” tags. These ﬁndings provide yet another incentive for deploying

multi-tags in order to ensure consistent object detection.

In our tag variability experiments, we used a ThingMagic reader, one circular Thing-

Magic antenna, and “UPM Rafsec UHF tag Impinj 34x54 ETSI/FCC” tags. Tags were

elevated 26 inches from the ﬂoor, and positioned perpendicular to the antenna at a

distance of 59.5 inches from the antenna center. The reader power level was set to

31.6dBm. Each tag was read 200 times and the number of successful reads was recorded.

We paused for 50ms between reads to allow tags sufﬁcient time to lose power and ini-

tialize their state. The reader was allotted 10ms to read a tag. In this way, we computed

the detectability/receptivity of 75 seemingly identical tags. To ensure data consistency,

each experiment was performed twice and repeated the next day with the tags rotated

180 degrees.

The smallest number of successful reads out of 200 was 8 and the largest was 91.

The average was 43.44 and the standard deviation was 23.92. The Pearson product-

moment correlation coefﬁcient between two reads of each tag on the same day was

0.99 and the correlation between reads across two days was 0.98. Figure 8(b) shows

the distribution of the number of successful tag reads. Figure 8(a) compares the num-

ber of successful reads for each tag across the two sets of experiments conducted on

consecutive days, and Figure 8(c) depicts how the number of successful reads varies

across different days. To magnify the visual spread between tags, we show the number

of successful tag reads out of 400 by summing the detectability across the two runs of

each day. Similarly high tag detectability variations were found in other UPM Rafsec

tag types.

6.2 Reader Variability

To ensure that our results are not dependent on the reader/antenna manufacturers, we

repeated our experiments using ThingMagic readers and ThingMagic circular antennas.

Since the tag detection algorithms used by ThingMagic and their implementations are

different from those of Alien Technology, and since ThingMagic antennas are much

bigger than those by Alien Technology, the detection probabilities we obtained differed

between these two systems. However, the percentage improvements of multi-tags ver-

sus single-tagged objects were similar for both systems, supporting our hypothesis that

the percentage improvements in object detection using multi-tags is mostly indepen-

dent from the speciﬁc equipment used. Figure 9 shows the statistics of object detection

improvements using circular ThingMagic antennas for a different number of tag en-

sembles per object. Note that the ThingMagic equipment enabled the collection of data

for 3 and 4 antennas, whereas the Alien Technology readers work with only 1 and 2

antennas.

109

(a) Comparison of the number of successful reads per tag across two

days. The tags are sorted based on the number of successful reads to

better illustrate the data.

(b) This graph shows the distribution of successful tag reads across two days and

two tag orientations. The number of successful reads shown is out of a total of 400

attempted. We observe a signiﬁcant separation between several “clusters” of tag per-

formance levels.

day deviates from the number of successful reads on the ﬁrst day.

Fig.8. Characterizing tag detectability/receptivity by comparing the number of successful tag

reads, the distribution of these reads, and the deviation in the number of reads across multiple

experiments.

110

Fig.9. The detection probability statistics for circular ThingMagic antennas as a function of the

power level for different antenna conﬁgurations and for a different number of tags per object.

7 Effects of Multi-tags on Anti-collision Algorithms

Anti-Collision algorithms allow a reader to uniquely identify tags while minimizing the

number of tag broadcasting collisions (i.e., simultaneous interfering transmissions by

the tags). Multi-tags have no effect on two variants of Binary Tree-Walking [1] [18],

and may at most double/triple the total read time for dual/triple-tags over single tags

for Slotted Aloha [1] and for Randomized Tree-Walking [19] [20] [21]. Our current

experimental study of multi-tags addressed how multi-tags improve object detection. It

is worth noting, however, that since not all tags are detected, the time required to identify

all reader-visible tags is considerably less than double (or triple) the time needed to

identify single-tagged objects.

In particular, from our experiments we observed that 25% to 65% of all tags are

detected with one reader antenna, depending on its type and power level. Therefore,

attaching two tags to each object may not add any signiﬁcant overall time delay for

object identiﬁcation. In addition, current RFID technology can read hundreds of tags

per second, making the increase in the number of tags insigniﬁcant, even for real-time

systems. Moreover, in many scenarios the beneﬁts of successfully identifying all the

objects certainly justiﬁes a modest increase in identiﬁcation time.

8 Object Detection in Presence of Metals and Liquids

So far, our discussion of multi-tags has been restricted to scenarios where the objects to

be identiﬁed contained no metal or liquid materials. In practical scenarios however, sets

of items to be identiﬁed can contain mixtures of non-metallic objects, as well as partly

metallic and liquid objects, making reliable object identiﬁcation more problematic. It

is more difﬁcult to detect metals and liquids because they tend to interfere with ra-

dio signals, thus preventing readers from receiving accurately decodable tag responses.

Metallic and liquid objects can also occlude other non-metallic objects, and thus inter-

fere with the detection of these as well.

To detect metallic and liquid objects in our experiments, we had to considerably

reduce the distance from the objects to the readers and to operate readers at high power

111

levels. Based on our experimental results, multi-tags are highly effective in improving

object detection in the presence of metallics and liquids. We observed an almost lin-

ear improvement in metallic and liquid object detection when the number of tags per

object is increased, as compared to the rapidly increasing and then leveling curve for

solid non-metallic objects. In addition, when metals and liquids are present, the detec-

tion probability curve for solid and non-metallic objects drops considerably, due to the

radio interference created by the metallics and liquids. A more detailed treatment of the

effects of multi-tags on metallic and liquid object detection will be reported elsewhere.

9 Economics of Multi-tags

Based on the results presented in this paper, we see that object detection probabili-

ties are far from perfect even when multiple antennas and readers are used. Multi-tags,

potentially in conjunction with multiple readers, can provide a viable solution to this

problem. The cost of RFID tags in 2007 is around 10-20 U.S. cents a piece, making

multi-tagging of high cost items viable even now. In addition, the cost of tags is drop-

ping at an exponential rate following Moore’s law, allowing for the cost-effective tag-

ging of even low-cost objects in the near future. Also, RFID tags decrease in cost at a

substantially faster rate than RFID readers, due to the economy-of-scale and improved

yield trends inherent in their manufacturing; moreover, this price gap is expected to

continue to widen. The future omnipresence and ubiquity of RFID tags is expected to

eventually bring down the cost of RFID tags into the sub-penny level.

Many RFID tag types are delivered to the customer on a continuous paper roll,

and the customer later programs the tags with unique IDs. We envision that tags will

soon be cheap enough to embed into adhesive packaging tape that wraps around pack-

ages/containers, thus simplifying the multi-tagging of boxed objects and enabling auto-

matic tag diversity and orientation selection to greatly improve object detection at negli-

gible cost. Also, with higher tag ubiquity and the multi-tagging of objects, the testing of

RFID tags will not be required as tag production yields become almost irrelevant, thus

further reducing the cost of tag manufacturing and ensuring high object detection prob-

abilities as well as improved dependability and reliability of RFID systems. In short,

multi-tags are absolutely economically viable, and these beneﬁts are bound to become

even more dramatic over time.

10 Conclusion

Our experiments indicate that multi-tags are highly effective in improving object de-

tection probability, yielding double-digit improvements over traditional single-tagged

object RFID systems using both linear and circular antennas. Moreover, multi-tags can

offer signiﬁcantly larger improvements in object detection as compared to adding ex-

tra readers, even in the presence of objects containing metals and liquids, without ex-

acerbating the burden on anti-collision algorithms. Multi-tagging of some objects is

economically viable today, and as the cost of tags decreases rapidly, a wider range of

applications will become possible with each year. We conclude that multi-tags can be

112

an effective, and economically viable solution for RFID applications that require higher

object detection probabilities.

References

1. Finkenzeller, K.: RFID Handbook. Wiley & Sons (2003)

2. Shepard, S.: RFID - Radio Frequency Identiﬁcation. McGraw-Hill, New York (2005)

3. U.S. House of Representative Subcommittee on Commerce, Trade, and Consumer Protec-

tion: RFID Technology - What the Future Holds for Commerce, Security, and the Consumer.

University Press of the Paciﬁc, Honolulu, Hawaii (2004)

4. Furness, A., Smith, I.G.: RFID Compendium - The Technology and Where to Use It. Auto

ID Service Providers, Halifax, England (2004)

5. Garﬁnkel, S., Rosenberg, B.: RFID - Applications, Security, and Privacy. Addison-Wesley

(2006)

6. Polizzi, T.: RFID in the Enterprise for Cross-Enterprise Business Automation. WCCN

Publishing, Irvine, CA (2004)

7. Arneson, E.: Hyperscan - rﬁd game system from mattel.

http://boardgames.about.com/od/news/a/hyperscan.htm (2006)

8. Bhuptani, M., Moradpour, S.: RFID Field Guide - Deploying Radio Frequency Identiﬁcation

Systems. Sun Microsystems Press, New Jersey (2005)

9. Heinrich, C.: RFID and Beyond. Wiley Publishing, Indianapolis, IN (2005)

10. Jilovec, N.: EDI, UCCnet and RFID - Synchronizing the Supply Chain. 29th Street Press,

Loveland, CO (2004)

11. Kleist, R.A., Chapman, T.A., Sakai, D.A., Jarvis, B.S.: RFID Labeling - Smart Labeling

Concepts & Applications for the Consumer Packaged Goods Supply Chain. Printronix, Inc.,

Irvine, CA (2004)

12. IDTechEx: Rﬁd progress at wal-mart. http://www.idtechex.com/products/en/articles/

00000161.asp (2005)

13. Bolotnyy, L., Robins, G.: Multi-tag radio frequency identiﬁcation systems. In: Proc. IEEE

Workshop on Automatic Identiﬁcation Advanced Technologies (Auto-ID). (2005) 83–88

14. Balanis, C.A.: Antenna Theory Analysis and Design. Wiley & Sons (1997)

15. Schuster, E., Scharfeld, T., Kar, P., Brock, D., Allen, S.: Analyzing the rﬁd tag read rate

issue. (http://mitdatacenter.org/CutterITAdvisor.pdf)

16. Impinj: Receptivity - a tag performance metric.

http://www.impinj.com/ﬁles/MR MZ WP 00005 TagReceptivity.pdf (2005)

17. Chen, Y., Kahng, A.B., Robins, G., Zelikovsky, A.: Area ﬁll synthesis for uniform layout

density. IEEE Trans. Computer-Aided Design 21 (2002) 1132–1147

18. Juels, A., Rivest, R., Szedlo, M.: The blocker tag: Selective blocking of rﬁd tags for con-

sumer privacy. In Atluri, V., ed.: Proc. ACM Conference on Computer and Communications

Security, Washington, DC, USA (2003) 103–111

19. Bolotnyy, L., Robins, G.: Randomized pseudo-random function tree walking algorithm for

secure radio-frequency identiﬁcation. In: Proc. IEEE Workshop on Automatic Identiﬁcation

Advanced Technologies (Auto-ID). (2005) 43–48

20. Center, A.I.: (2003) Draft Protocol Speciﬁcation for a 900 MHz Class 0 Radio Frequency

Identiﬁcation Tag.

21. Weis, S.: Security and privacy in radio-frequency identiﬁcation devices. Master’s thesis,

MIT (2003)

113