Advanced Research Programs (ARP) has funded a team of UHCL researchers to develop a WSN experiments
over a period of 2007-2009. As the head of the project, Dr. Yang has led the team in developing labs for supporting research
and teaching of network security, including wireless networks. The work proposed in this proposal is focusing on
developing secure and effective algorithms for WSN for Human Detection and tracking, and integrating a WSN Test Bed
into the existing computer security labs. The DCSL, currently providing ample space to host the network devices and
security appliances of the DCSL network, will be used to host the WSN Test Bed. Secure and Optimized Communication &
Organization for Target Tracking
in Wireless Sensor Networks (SOHO) proposal was submitted to the Texas Higher Education Coordination Board (THECB)
in April 2006 for the Advanced Research Programs grant. The proposal was accepted and completed by 2008.
This project decribes an authentication mechanism
to protect the Optimized
Communication and Organization (OCO) method for routing and self-organization of a
wireless sensor network. This includes authentication of sent messages to assure that
they have not been altered (aka. message integrity), and authentication of the sender to
assure that the messages are not forged (aka. origin integrity). The process begins with a
survey of security threats and risk mitigation strategies common to all wireless sensor
networks. This survey includes mainly attacks against origin and message integrity, as well
as those against confidentiality and availability. The risk analysis converges into a risk
assessment of OCO messaging. Selection of an elegant authentication solution requires a survey
of current unicast and broadcast message
authentication protocols for wireless sensor networks. The protocols will be contrasted to
select the most appropriate one for OCO.The project also demonstrates via simulation with the
network simulation software OMNeT++ the
energy costs of integrating message authentication into OCO. Understanding this cost
enables an application owner to evaluate whether to accept the risk of insecure messaging
or bear the cost of authentication.
GENERAL DESCRIPTION
Without authentication mechanisms tailored to
the application, sensor networks will be unreliable for use in critical arenas. The receiver
must be guaranteed that critical messages indeed originated from the claimed source. Conventional
security mechanisms in use on the Internet are usually not applicable to wireless sensor
networks because of the limited resources available in the sensor nodes, such as limited
processor speed, smaller memory size, and limited communication channels and speed.
Security comes at a cost; and that cost must be balanced with the goals of the application.
The goal of this research project is to
produce an efficient, authenticated version of the OCO method, which effectively provides
both message integrity and origin integrity to the wireless sensor network applications.
SECURITY GOALS
Security assessments of any application focus on the five fundamental tenets of
information security: confidentiality, origin integrity, data integrity, non-repudiation, and
availability.
Confidentiality means the concealment of information from unauthorized entities.
Mechanisms used to achieve confidentiality include access control mechanisms and
cryptography. Cryptography scrambles, or encrypts, data to generate ciphertext
unintelligible to any unauthorized viewer. The data can be made comprehensible to an
authorized viewer who knows the secret key.
Origin integrity, also known as authentication, refers to the trustworthiness of the
source of data. It means that the receiver of a message can trust that the sender of the
message is truthfully who it claims. An intruder should not be able to send a fabricated
message and have it treated as a legitimate message from a trusted peer. Data integrity
means that the user of the data can trust that the content of the information has not been
changed in any way by an unauthorized intruder or improperly modified by an authorized
user. Non-repudiation means that the sender of a message should not be able to deny
later that he ever sent that message. In the pre-digital world, one achieved non-repudiation
with a simple hand-written signature. In cryptography, it implies that authentication and
data integrity can be certified with a high level of assurance and it cannot later be refuted.
PROPOSED SOLUTION: s(OCO)
Focusing strictly on authentication, instead of confidentiality or
availability, balances the risk outlined in the risk assessment with the goal of conserving
energy. This project recommends integrating origin authentication and message integrity into
any OCO message with a total risk rating above 250. This captures all messages in the
Processing, Tracking, and Maintenance phases. This authenticated version of OCO, known
as s(OCO), provides individual message authentication with limited overhead. s(OCO) will
protect the network from message fabrication and message spoofing as long as no nodes
are compromised. If an attacker can compromise an active node, it can steal the shared key
and defeat the security protocol
In order to model the system, this proposal imposes a standard TinyOS packet
format onto OCO communications and establishes standard sizes for OCO data fields. This
facilitates calculation of the cost of integrating TinySec-Auth into OCO. Common fields
among packets include destination address, Active Message (AM) type, and packet length.
By starting packets with the destination address, nodes may employ early rejection of
messages. When a node determines that it is not the intended recipient, it may conserve
energy by dropping the packet. The active message type, analogous to a TCP or UDP port
in the Internet protocols, specifies the appropriate handler function to extract and interpret
the message on the receiver
Figure illustrates the respective packet formats for TinyOS, TinySec-Auth, and
s(OCO). Shaded fields in the packet diagrams represent fields protected by the MAC.
s(OCO) follows the packet format in TinySec-Auth and increases the TinyOS
headers by one byte. Both proposals drop the 1-byte group ID and the 2-byte CRC fields in
the original TinyOS packet and replace them with a 4-byte MAC. The MAC provides the
packet integrity service of the CRC. The cipher key implicitly replaces the group
membership function provided by the group ID. s(OCO) appropriates bytes from the
payload for additional fields including a counter used as a message id and the packet source
address. Node addresses occupy two bytes. s(OCO) allocates 2-bytes each for time-tosynchronize
and time-to-stay-awake. Node position, node energy level, and notification
timestamps each receive 4 bytes. The standard fields in a s(OCO) packet consume 12 bytes.
PHASES PROPOSED FOR s(OCO)
Table summarizes the 14 OCO messages, their
respective roles, and their packet length.
The Position Collection phase, which only occurs during network initialization,
includes two messages with risks ratings below 250. The base broadcasts message M1, the
Position Request message, immediately following node deployment. The nodes respond by
sending message M2 to their parent, which in turn forwards the message toward the base.
Because of the narrow attack window, the Position Collection phase messages receive a
low total risk rating. Thus, implementation in a standard TinyOS packet format satisfies
security requirements. Message M1 occupies 7 bytes and maps to TinyOS Active Message
(AM) type 1. The Position Reply message, message M2, includes fields for reporting node
ID and that nodes position. These increase packet length of M2 to 13 bytes.
In the Processing phase, the base station sends two topology type packets and three
packets used to assign roles to nodes. s(OCO) must broadcast the topology messages
because, at this point in the network setup, there is no route from the base station to the
destination nodes. The topology information captured during the Position Collection phase
only provided the path for nodes to report their id and position to the base station. Message
M3 advises a child node of the id of its parent. M4 informs a parent node of the id of one of
its children. A parent receives M4 for every one of its immediate children. The packets put
the child node id and parent node id into the message payload, adding 4 bytes and
increasing the length of M3 and M4 to 16 bytes. Because of the lack of a route from the base to the border, the base must also broadcast
these messages. When a node receives on of these messages, it will check the id of the
intended target in the payload and rebroadcast the message if necessary. Nodes use the
counter to track whether or not they have already broadcast the message. M5, which
requires 14 bytes, instructs a border node to activate its tracking sensor and its radio. M6
announces the time to sleep (TTS) and time to stay awake (TTSA) to redundant nodes.
These time fields consume two bytes each and increase packet length to 18 bytes. Message
type 7 consumes 14 bytes to instruct forwarding nodes of their occupation.
The two messages in the Target Tracking phase originate from a border node
alerting its peers of an intruder. M8, sent toward to base station, includes fields for
reporting node id and a 4-byte timestamp. It occupies 18 bytes in a TinySec-Auth format. A
node broadcasts M9, which requires 12 bytes, to its neighbors to inform them of the
intrusion.
The Maintenance phase supports network longevity with keep-alive messages and
notifications when nodes lose their parent or child. Messages M10 through M14 constitute
the Maintenance phase. By way of message M10, a child node can report its health to its
parent. M10 includes a 4-byte field where the child node records its energy level,
increasing total packet length to 16 bytes. A parent informs its children that it is still alive
by broadcasting M11, which requires 12 bytes. Nodes that receive M11 do not rebroadcast
it, as they would when they receive one of the Processing phase messages. However, nodes
that receive the message must still authenticate it to determine if the source address belongs
to their parent. s(OCO) does not define recommended timing interval for sending M10 and
M11, leaving a tradeoff between recovery time and energy use to the implementation
Message M12 and M13 make up the S.O.S. messages in s(OCO). A child node
broadcasts the 12-byte message M12 when it does not receive message M11 from its
parent. Neighboring nodes must authenticate, but not rebroadcast M11. A parent sends
M13 to the base station when its child node fails to report its status. M13 includes 2-bytes
for the lost child node id and 4 bytes for the parent nodes energy level, lengthening it to 18
bytes. Each node that receives M13 must authenticate it and send it to their parent until it
reaches the base. The base station periodically sends message M14 to resynchronize
redundant nodes. This message includes updates to the time to synchronize and time to
stay awake parameters. M14 consumes 16 byes.
EXPERIMENTAL DESIGN AND TOOLS
The project put forth the hypothesis that securing OCO will increase the total cost of
operating the network to between three percent and thirteen percent. The three percent
lower bound reflects the cost of a packet in TinySec-Auth with a full 24-byte payload. The
thirteen percent upper bound represents the cost increase of s(OCO)s shortest 12-byte
packets. The mean operating cost of s(OCO) should exist within these upper and lower
bounds because of packet length and the influence of the sensor module and the radio
module. The experiments will simulate an OCO network and an s(OCO) network and
evaluate the mean operating costs of both networks under similar circumstances.
The experimental analysis employs the OMNeT++ network simulator for the
implementation and evaluation of the s(OCO) countermeasures on network life span.
OMNeT++, provides a framework that simplifies evaluation of communication protocols.
OMNeT++ supplies a hierarchal set of modules, each interconnected through interfaces called gates.
Since OMNeT++ manages
transmission of messages through the gates, the developer can focus on implementation of
application classes within each module. In this evaluation of OCO, an instantiation of an
OMNeT++ application class randomly distributes nodes across the simulation grid during
the Position Collection phase. It simulates the transmission and reception of Position
Collection messages and tracks the cost of each message throughout the simulation.
A separate C# application reads the output from OMNeT++, constructs the coverage map,
and performs the image processing tasks. This application determines node occupation, and
organizes the network topology. The output from this application is fed back into the
OMNeT++ simulator to evaluate the cost of message passing in the Processing and
Tracking phases. The simulation omits modeling of the Maintenance phase. The
OMNeT++ simulator and the C# image processing application lack automated interfaces
that could allow simple integration of the two components. Without such interfaces, the
network cannot seamlessly notify the base of the need for maintenance, reprocess the
coverage map, and send new topology and occupation messages.
The simulation assesses energy consumed by the nodes radio, its sensor, and its
microcontroller. In the Position Collection and Processing phases, all nodes maintain an
active radio and processor. Thus, a nodes energy consumption in these first two phases
depends mainly on the number of messages it has to send and receive. In the Tracking
phase, a nodes occupation influences its energy usage characteristics. Border nodes
generally consume the most energy because both their sensor modules and radio modules
remain active. Their processor sleeps until it is required to create a message. Forwarding
nodes should consume less since they keep their sensor disabled until one of their
neighbors detects an intruder. Their radio remains enabled to receive and forward
messages. As with border nodes, their microcontroller sleeps except to create messages. All
three components of redundant nodes remain deactivated, although they periodically wake
up to receive commands sent by the base. The simulation assumes that the base station has
unlimited energy and computation resources.
EXPERIMENTAL RESULTS
This analysis of results aims to identify the impact s(OCO) has on individual nodes
and on the network as a whole. According to the TinySec paper, the addition of
authentication increases the cost of sending a single 24-byte packet by three percent.
However, this value does not apply globally to individual OCO or s(OCO) packets, which
range in length from between 12 and 18 bytes total. Shorter packets cost more because
MAC computation must occur early in packet transmission, before the first byte leaves the
mote radio. In longer packets, the cost of computing the MAC averages out over more
bytes. This supports the case for defining a lower bound of three percent. Other services
besides messaging consume energy in a wireless sensor network, such as the sensor and
processor. The experimental results support the hypothesis that s(OCO) costs between three
percent and thirteen percent more of total network energy than the standard,
unauthenticated OCO.
The addition of authentication to OCO increases total energy consumption to
between 8 and 10 percent of all energy consumed during the experiments. While the results
only slightly exceed the costs of maintaining an active processor or an active sensor, they
still negatively influence the longevity of an OCO network. Since the network stabilizes
during the Tracking phase, it should be able to sustain operations for a long duration with
s(OCO). Nodes that transmit or receive a higher number of messages than its peers may
benefit from an alternative risk assessment methodology.
The survey conducted as part of this project tracked the evolution of two disparate
fields in sensor network research: target tracking applications and authentication protocols.
While both areas of research strive for efficiency, the addition of security to a target
tracking mechanisms increases energy consumption. Sensor network authentication protocols similarly strive for
efficiency by reducing computation and communication costs.
