Semantic-Based Bluetooth-RFID Interaction for Advanced Resource Discovery in Pervasive Contexts.

Full text

Page 1

1
SEMANTIC-BASED BLUETOOTH-RFID INTERACTION FOR ADVANCED RESOURCE
DISCOVERY IN PERVASIVE CONTEXTS1

Tommaso Di Noia, Eugenio Di Sciascio, Francesco M. Donini*, Michele Ruta, Floriano
Scioscia, Eufemia Tinelli
SisInfLab -- Politecnico di Bari, Bari, Italy
{t.dinoia, disciascio, m.ruta, f.scioscia, e.tinelli}@poliba.it
*SisInfLab and Università della Tuscia, Viterbo, Italy
donini@unitus.it

Keywords
Knowledge Discovery, Semantic matchmaking, Ontologies, Mobile Technologies, Wireless
Technologies, Knowledge Storage, Knowledge-Based Systems

ABSTRACT
We propose a novel object discovery framework integrating the application layer of Bluetooth
and RFID standards. The approach is motivated and illustrated in an innovative u-commerce
setting. Given a request, it allows an advanced discovery process, exploiting semantically
annotated descriptions of goods available in the u-marketplace. The RFID data exchange
protocol and the Bluetooth Service Discovery Protocol have been modified and enhanced, to
enable support for such semantic annotation of products. Modifications to the standards have

1 This is a draft version of the paper appearing in: International Journal on Semantic Web and Information
Systems (IJSWIS), Volume 4, Number 1, page 50--74 - 2008

Page 2

2
been conceived to be backward compatible, thus allowing the smooth coexistence of the legacy
discovery and/or identification features. Also noteworthy is the introduction of a dedicated
compression tool to reduce storage/transmission problems due to the verbosity of XML-based
semantic languages.

INTRODUCTION AND MOTIVATION
Radio-Frequency IDentification (RFID) is an increasingly widespread and promising wireless
technology interconnecting via radio a transponder carrying data (tag) located on an object, and
an interrogator (reader) able to receive the transmitted data. Tags usually contain a unique
identification code, which can be used by readers to identify the associated object. Since low-
cost tags can be fastened to objects unobtrusively, preserving their common functions, RFID de
facto increases the “pervasiveness” of a computing environment. Current RFID applications
focus on retrieving relevant attributes of the object the tag is clung to, via a networked
infrastructure from a fixed information server. This identification process involves the code
associated to the transponder exploited as index key. Nowadays tags with larger memory
capacity and on-board sensors enable new scenarios and further applications, not yet explored.
We believe that, in the era of semantic technologies and mobile computing, there is room for
more advanced and significant applications of RFIDs extended with structured descriptions, so
that a good equipped with an RFID can semantically describe itself along its whole life-cycle.
We therefore conceived a unified framework where a semantic-enhanced RFID-based
infrastructure and an advanced Bluetooth service discovery –also endowed of semantic-based
discovery features-- are virtually “interconnected” at the application layer permitting innovative
services in u-environments. In our mobile framework, tagged objects expose to a reader not

Page 3

3
simply a string code but a semantically annotated description. Such objects may hence describe
themselves in a variety of scenarios (e.g., during supply chain management, shipment, storing,
sale and post-sale), without depending on a centralized database. Exploiting these annotations
calls for discovery/interaction protocols able to effectively deal with rich and articulated
descriptions. Therefore a novel multi-protocol and interactive discovery mechanism has been
designed. In this effort we borrowed from ideas and technologies devised for the Semantic Web
initiative. To simply illustrate our proposal, we set our stage in a u-marketplace context i, where
objects endowed with RFID tags are dipped into an enhanced Bluetooth framework.
In particular, building on previous works that enhanced the basic discovery features of Bluetooth
with semantic-based discovery capabilities (Ruta et al., 2006a), we propose an extension of
EPCglobal specifications for RFID tag data standards, providing semantic-based value-added
services. Coping with limited storage and computational capabilities of mobile and embedded
devices, and with reduced bandwidth provided by wireless links, issues related to the verbosity
of semantic annotation languages cannot be neglected. Compression techniques become essential
to enable storage and transmission of semantically annotated information on mobile devices. We
hence devised and exploited a novel efficient XML compression algorithm, specifically targeted
for DIG 1.1 (Bechhofer et al., 2003) document instances. Benefits of compression apply to the
whole ubiquitous computing environment, as decreasing data size means shorter communication
delays, efficient usage of bandwidth and reduced battery drain for mobile devices in a Mobile
Ad-hoc NETwork (MANET).
The remaining of the paper is structured as follows. In the next section relevant technological
bricks of the proposed framework are surveyed. Section 3 outlines the framework, explaining the
discovery process as well as proposed semantic-based enhancements to RFID standards. The

Page 4

4
compression algorithm for semantic annotations is outlined in Section 4. Section 5 exemplifies
the approach in a u-commerce scenario. Results on key performance measures to assess the
feasibility of the proposed approach, are provided in Section 6. Conclusions close the paper.

BASICS
In this section we survey relevant aspects of languages, technologies and protocols we use and
adapt, concentrating on key features our proposal is based on. We assume the reader be familiar
with at least basic elements of Semantic Web and ontologies (Berners-Lee et al., 2001; Shadbolt
et al., 2006; Horrocks et al., 2001; McGuinness et al., 2002; Martin et al. 2002), of OWL
(http://www.w3.org/TR/owl-features/) and related languages, such as Description Logics (DLs)
(Borgida, 1995; Donini et al., 1996). We therefore move straightforwardly to analyze issues
closely related to our proposal.

Exploiting semantically annotated descriptions
Given a domain ontology T, DL-based systems usually provide at least two basic reasoning
services: Concept Satisfiability and Concept Subsumption. Using subsumption it is possible to
establish if a description C is more specific than a description D, T Ñ C b D. If the previous
relation holds, then we may say that information C associated to a given resource completely
satisfies what has been requested in D, i.e. a full match occurs. With Concept Satisfiability the
discovery of incompatible resources with respect to a request can be performed. If D 6 C is not
satisfiable w.r.t. the ontology T, then C is not compatible with the request. Obviously full
matches cannot be deemed the only useful, as they will be probably rare in a variety of contexts.

Page 5

5
Given a request and a set of resources, usually C ⋢ D and D 6 C is satisfiable w.r.t. T. That is,
the resource does not completely satisfy the request but it is compatible with it. Hence, a metric
is needed to establish “how much” the resource C is compatible with the request D or,
equivalently, “how much” it is not specified in C to completely satisfy D, in order to make the
subsumption relation C b D true. In (Di Noia et al., 2004) rankPotential algorithm was proposed
to evaluate this measure. Given an ALN (Attributive Language with Number restrictions)
ontology T and two ALN concepts C and D both satisfiable in T, rankPotential(C, D, T)
computes a semantic distance of C from D with respect to the ontology T.
If some requirements in the request D are in conflict with the resource C, rankPotential cannot
be applied. Nevertheless, in looking for “not so much” unsatisfactory matches when recovering
from an initial “no match”, a partial match could still be useful. In (Di Noia et al., 2004) the
rankPartial algorithm was proposed for ranking incoherent pairs of descriptions. Given an
ontology T and two concept expressions D and C, both satisfiable with respect to T, if D is not
compatible with C i.e. their conjunction is not satisfiable with respect to T, then rankPartial
returns a score measuring the semantic incompatibility of D and C.

Semantic based Bluetooth Service Discovery
Usually, resource discovery protocols involve a requester, a lookup or directory server and
finally a resource provider. As a MANET is a volatile environment, a flexible resource discovery
paradigm is needed to overcome difficulties due to the host mobility. Nevertheless, existing
protocols for mobile applications use a simple string-matching, which is largely inefficient in

Page 6

6
advanced scenarios (Ruta et al., 2006b). With specific reference to the Bluetooth service
discovery protocol (SDP), it is based on a 128 bit Universally Unique Identifier (UUID)
associated to single service classes. Resource matching in Bluetooth is hence strictly syntactic,
and SDP manages only exact matches. In (Ruta et al., 2006a) a framework has been proposed
that allows the management of both syntactic and semantic discovery of resources, by integrating
a semantic layer within the OSI Bluetooth stack at application level. The Bluetooth standard has
been enriched by new functionalities which permitted to maintain a backward compatibility
(handheld device connectivity), adding the support to discovery of semantically annotated
resources. Unused classes of 128 bit UUIDs in the original Bluetooth standard were exploited to
mark each specific ontology thus calling this identifier OUUID (Ontology Universally Unique
IDentifier). By means of the OUUID matching the context was identified and a preliminary
selection of resource referring to the same request's ontology was performed. The fundamental
assumption is that each resource is semantically annotated. A service provider stores annotations
within resource records, labelled with unique 32-bit identifiers. Each record contains general
information about a single semantic enabled resource and it entirely consists of a list of resource
attributes. In addition to the OUUID attribute, there are a ResourceName (a human-readable
name for the resource), a ResourceDescription (expressed using DIG syntax) and a variable
number of ResourceUtilityAttr i attributes, i.e., numerical values used according to specific
applications. In (Ruta et al., 2006a), by adding four SDP Protocol Data Units (PDUs)
SDP_OntologySearch (request and response) and SDP_SemanticServiceSearch (request and
response) to the original standard (exploiting not used PDU ID), together with the original SDP
capabilities, further semantic enabled discovery functionalities were introduced. The overall
interaction was based on the original SDP in Bluetooth. No modifications were made to the

Page 7

7
original structure of transactions. In fact, semantic-based micro-layer has been built over the
standard SDP recycling its basic parameters, data structures and functions, just differently using
the basic framework.

RFID features
In our framework we refer to RFID transponders compliant with EPCglobal standard for Class 1-
Generation 2 UHF tags (Traub et al., 2005). Tag memory is divided in four logical banks
(EPCglobal Inc., 2005a): (1) Reserved. It is optional; if present, it stores 32-bit kill and access
passwords. (2) Electronic Product Code (EPC). It stores, starting from address 0: (i) 16 bits for
a Cyclic Redundancy Check (CRC) code; (ii) a 16-bit Protocol Control (PC) field, composed of
5 bits for identification code length, 2 bits reserved for future use and 9 bits of numbering system
identification; (iii) an EPC field for the identification code. (3) Tag identification (TID). It stores
at least tag manufacturer and model identification codes. This bank may be enlarged to store
other manufacturer or model-specific data (e.g. a tag serial number). (4) User. An optional bank
that stores data defined by the user application. Memory organization is user-defined. EPCglobal
air interface protocol is an Interrogator-Talks-First (ITF) protocol: tags only reply to reader
commands. Here we briefly outline basic protocol features.
An RFID reader can preselect a subset of the tag population currently in range, according to
user-defined criteria, by means of a sequence of Select commands.
Select command sends a bit string to all tags in range. Each tag will compare it with the content
of a memory area specified by the reader, then it will assert/deassert one of its status flags
according to the comparison result (match/no-match). Command structure is shown in Table 1;
parameters are as follows: (i) Target determines which tag status flag will be modified by the

Page 8

8
Select command; (ii) Action tells how a tag is required to modify the flag (assert, deassert, do
nothing) for either positive or negative match outcome (a three-bit field is thus required to
encode the six cases); (iii) MemBank indicates what memory bank must be compared; (iv)
Pointer is the address of the first bit of MemBank tag memory area that must be compared; (v)
Length is the length of the bit string to be compared; (vi) Mask is the bit string to be compared
with the content of the memory area selected by MemBank, Pointer and Length values; (vii)
Truncate tells the tag to send only part of its EPC code in the following protocol step; (viii) CRC,
used for command data integrity protection.
Opcode
10102
Target
3 bits
Action
3 bits
MemBank
2 bits
Pointer
bit vector
Length
8 bits
Mask
1-255
bits
Truncate
1 bit
CRC
16
bits
Table 1. Select command structure in RFID protocol
After this phase, the inventory loop begins. In each iteration the reader isolates one tag in range,
reads its EPC code and can access its memory contents. Among available commands, only Read
and Write are relevant for our purposes.
Read command allows to read from one of the four tag memory banks. Command structure is
shown in Table 2; parameters are as follows: (i) MemBank indicates the bank data must be read
from; (ii) WordPtr points to the first 16-bit memory word to be read; (iii) WordCount is the
number of consecutive 16-bit memory words that must be read (if it is 0, then the tag will send
data stored up to the end of the memory bank); (iv) RN, random number used as access
transaction identifier between reader and tag; (v) CRC.
Write command allows a reader to write a 16-bit word to one of the four tag memory banks.
Command structure is similar to Read, as shown in Table 3.
Opcode
110000102
MemBank
2 bits
WordPtr
bit vector
WordCount
8 bits
RN
16
bits
CRC
16
bits
Table 2. Read command structure in RFID protocol

Page 9

9
Opcode
110000112
MemBank
2 bits
WordPtr
bit vector
Data
16
bits
RN
16
bits
CRC
16
bits
Table 3. Write command structure in RFID protocol
Together with tag data and air interface protocol, the EPCglobal standard defines a support
infrastructure for RFID applications, where a key role is played by Object Naming Service (ONS)
(EPCglobal Inc., 2005b). It is based on the Domain Name System adopted to solve symbolic
Internet addresses. ONS allows to retrieve services related to a specific object using the EPC
code stored within the tag as a URI. EPCglobal Network Protocol Parameter Registry is
maintained by EPCglobal consortium and contains suffixes identifying all valid service types
(e.g., ws for a Web Service, html for a Web Page of the manufacturer, epcis for a EPCglobal
Information Service providing authoritative information about the object associated with an EPC
code).

Figure 1. Infrastructure elements: semantic-enhanced RFID tags; air-interface EPCglobal RFID
protocol; middleware stratum; Bluetooth SD protocol, hotspot enriched with semantic
matchmaking capabilities.

FRAMEWORK AND APPROACH
We designed a unified semantic-aware framework, comprising modified RFID and Bluetooth
based infrastructures that are virtually “interconnected” at the application layer permitting
innovative services in u-environments. Our framework introduces a proposed extension of

Page 10

10
EPCglobal standard, allowing a semantic-based object discovery. Protocols to read/write tags
have been preserved maintaining original code-based access (so keeping a compatibility with
legacy applications practically without modifications). A good can be easily and thoroughly
described by means of a semantic annotated description stored within the tag it is associated
with. Main elements of the proposed framework, see Figure 1, are: 1) goods equipped with
semantic-enhanced RFID tags, 2) a middle tier component provided with an RFID reader and
Bluetooth connectivity, 3) hotspot enriched with semantic matchmaking capabilities. Two
identification/discovery paradigms are involved: EPCglobal air interface protocol for RFID tags
and semantic-enhanced Bluetooth Service Discovery Protocol. Interaction can be triggered by
the user by means of either an implicit or an explicit request. The simplest --though not trivial, as
obviously requests may change over time and during the product life-cycle-- form of interaction
is querying the tag (of the good) for some information, exploiting user’s mobile handheld device.
In implicit requests the framework can be used to recognize choices she performed so
intercepting and interpreting them as a preliminary interaction aimed at discovery of goods
similar or to be combined with the chosen one. In the first case the user can directly interact with
the hotspot, issuing requests to it via the semantic-enhanced Bluetooth SDP and waiting for
replies. In the latter one the user plays a more passive role as the “Environment” (in the sense of
a pervasive and intelligent context, a marketplace in our example scenario) is able to perceive
modifications w.r.t. an earlier situation. RFID tags are required for hosting product features and
to set a link between the real and the digital world, whereas the middle tier is a double-faced
component. It listens for descriptions directly coming from the objects (by reading the tag
memory content), issues requests to the service provider and finally records and displays results
to the user. The RFID reader, scanning characteristics of a selected product, enables the further

Page 11

11
discovery phase which is aimed at identifying resources similar to the chosen one or to be
combined with it. Via the semantic based Bluetooth SDP and exploiting non standard inference
services outlined above, best matching resources of the marketplace will be discovered and
returned to the user. Hence the middleware integrates RFID and Bluetooth environments at the
application layer: data coming from RFID tags are extracted, processed and reformatted.
Furthermore they are arranged to enable the interaction with the service provider (hotspot) via
the semantic-enhanced Bluetooth SDP. The hotspot keeps track of resources within the
marketplace and replies to a submitted request with the best matching products for similarity and
association. To this aim, it is equipped with a DL reasoner able to provide previously introduced
services. Such an approach may provide several benefits. Information about a product is
structured and complete; it accurately follows the product history within the supply chain, being
progressively built or updated during the good life cycle. This improves traceability of
production and distribution, facilitates sales and post-sale services thanks to an advanced and
selective discovery infrastructure.

Semantic-enhanced EPCglobal RFID standard
In this subsection we outline the proposed backward-compatible extensions to EPCglobal RFID
standards enabling the framework described above. It is noteworthy that our semantic enabled
descriptions are expressed in DIG formalism (Bechhofer et al., 2003), a more compact syntactic
variant of OWL.
Two reserved bits in the EPC area within each tag memory are exploited. The first one – at 15h
(101012) address – is exploited to indicate if the tag has a user memory (bit set) or not (bit
cleared). The next one – at 16h address – is asserted to mark semantic enabled tags. In this

Page 12

12
manner, by means of a Select command (see Table 4), a reader can easily distinguish semantic
based tags. In particular Target and Action parameters have the effect to assert the SL tag status
flag only for semantic-enabled tags and deassert it for remaining ones. The following inventory
step will skip tags having SL flag deasserted, thus allowing a reader to identify only semantic-
enabled tags (protocol commands belonging to the inventory step have not been described,
because they are used in the standard fashion).
Parameter
Value
Description
Target
1002
SL flag
Action
0002
assert in case of
match, deassert
otherwise
MemBank
012
EPC
memory
bank
Pointer
000101012
initial
address
Length
000000102
number of
bits to
compare
Mask
112
bit
mask
Table 4. Select command parameters to detect semantic enabled tags
The EPC standard for UHF-Class 1 tags impose the content of TID memory up to 1Fh bit is
fixed. As said above, optional information could be stored in additional TID memory. We use the
TID memory area starting from 1000002 address. There we store the identifier of the ontology
(OUUID) w.r.t. the description contained within the tag is expressed. In order to make RFID
systems compliant with the ontology support system proposed in (Ruta et al., 2006a), we define a
bidirectional correspondence of OUUIDs stored in RFID transponders with those managed by
Bluetooth devices. To retrieve the OUUID value stored within a tag, a reader will exploit a Read
command with parameters as in Table 5:
Parameter
Value
MemBank
102
WordPtr
0000000
102
initial
address
WordCount
000010002
Description
TID memory
bank
read up to 8
words (128 bits)
Table 5. Read command parameters to extract OUUID from TID memory bank
Within the user memory bank together with the semantically annotated description of the good
the tag is clung to (opportunely compressed) will be stored also contextual parameters (whose
meaning depends on the specific application).

Page 13

13
The extraction or the storing of a description within a tag can be performed by a reader through
one or more Read or Write commands, respectively. Both commands are used in compliance
with the standard air interface protocol. In Table 6, parameters of the Read command for
extracting a compressed description are reported.
Parameter
Value
Description
MemBank
112
User memory
bank
WordPtr
0000000002
initial
address
WordCount
000000002
read up to
the end
Table 6. Read command parameters to extract semantic annotations from the User bank
In our approach the ONS mechanism is considered as a supplementary system able to grant the
ontology support. In case the reader does not manage the ontology the description within the tag
refers to, it may need an Internet connection in order to retrieve the related DIG file, which will
then remain stored for further usage on other goods of the same category. For this purpose we
use the ONS service and we hypothesize to register within the EPCglobal Network Protocol
Parameter Registry a new service suffix, the dig one, that will contain the URL of the DIG file
ontology. Of course the same can be done for OWL.
In case of EPC code families derived from the GS1 standard (formerly EAN.UCC) for barcode
product identification, we assume that the pair of fields used for ONS requests – which refer to
the manufacturer and to the merchandise class of the good – will correspond to a specific
ontology. In fact that pair exactly identifies the product category. Two goods with the same value
for that field parameter will be surely homogeneous or even equal. Note that the vice versa is not
verified, but this is not a concern for our purposes because ONS searches proceed only from the
EPC code toward the ontology. Hence we can surely have an unambiguous correspondence.
Deploying the approach
In our case study framework, we hypothesize a “smart shopping cart” is equipped with a sensor
and a tablet computer, which integrates an RFID reader and Bluetooth connectivity. When a

Page 14

14
customer picks up a product, the system assists her in discovering additional items, either similar
or to be combined with the selected one. To this aim, a two-step discovery is performed,
exploiting two different but related ontologies. In the first step, rankPotential algorithm is
exploited to retrieve correspondences with the request. Resources analogous to the one selected
by the user are identified, but – at the same time – semantically incompatible goods are
recognized. Their descriptions are submitted to the second matchmaking step. It exploits
rankPartial over a differently modeled ontology so allowing to discover products to be
associated with the chosen one. The hotspot will return two different lists of resource records
respectively for objects in a potential correspondence with the request and in a partial one.
In advanced mobile scenarios, usually the match between a request and a provided resource
involves not only the description of the resource itself but also data-oriented contextual
properties. In fact, it would be quite strange to have a mobile commerce application without
taking into account for example price or delivery time, among others. Hence, the overall match
value should depend not only on the semantic distance between the description of the demand
and of the resource, but also on those subsidiary values. An overall utility function has to
combine them with semantic matchmaking results, in order to give a concrete match measure
(Ruta et al., 2006b). In the proposed case study – referred to a u-commerce electronic product
store – the utility function adopts three contextual parameters: price (in US dollars), estimated
delivery time (in days), and product category, as shown in Table 7. They are exploited in a post-
processing phase following the semantic-based matchmaking and aimed to better agree
discovery results with user needs.
The proposed utility function (whose formulation derives from common sense considerations)
has two expressions, for potential and partial matches respectively:

Page 15

15
R
OR
R
O
OR
OR
POT
p
pp
α
p
p
ttu
tt
match
2
pot
f
)1 ( 3
)
+
1 (
) 5 . 0
−
()(
3
) tanh(
_
) (
⋅
α
β
−+
+−
−
+=

) 2 ( 3
1
)(
6
) tanh(
2
_
) (
⋅
OR
OR
OR
OR
PAR
cc
cc
ttu
tt
matchpar
f
−+
−−
+−
−
β
+=
γ

where pot_match and par_match are the potential and partial match values, p is price, t is
delivery time and c is product category. The index R is referred to the request whereas the O one
is referred to the supply and
) (⋅
u
is Heaviside step function. Parameters α, β, γ can be used to
fine-tune the utility function. Values we experimentally experienced with good results are
1 . 0
=α
,
10
=β
,
2 . 0
=γ
. They have been determined by means of empirical tests through the
comparison of system results with human users judgement. The higher the utility value the better
the obtained match. In both formulas the leading term is represented by the semantic match.
Product category
Value
Table 7. Product category contextual parameter
phones
1
computers
2
photo
3
audio/video
4
hobbies
5
The second term depends on the estimated delivery time and it is differently weighted in
proposed formulas. In the first one (discovery of goods similar to the request) a late delivery is
more penalized. On the other hand, partial matches refer to items that can be used together with
the selected one (such as accessories or complements), therefore a delay is less of a concern.
The last term is different in the two formulas. For potential matches, it is related to product price.
The price imposed by the requester is increased with a factor α on the assumption that, usually,
the demander is willing to pay up to some more than what she originally specified, on condition
that she finds the requested item or something very similar. Supplies with a much lower price
than request (less than 50%) are penalized since they likely represent items in a different market
segment. In the formula for partial matches, the last addend considers product category. Products