IEEE Consumer Electronics Magazine - January/February 2024 - 17

can be observed by a particular device during its
communication with another device. With such
characteristics, the device can calculate how
much it trusts in the other device.
We define the trust of a device j in device i as
FIGURE 1. Trust approach scenario of two-level
approach considering a tree-tier architecture.
changes due to devices' battery draining, devices
entering and leaving the network, connection
disruptions or loss, mobility, scheduled
shutdowns, among others.
An illustration of how the proposal works can
be seen in Figure 1. In the illustrated scenario,
initially, two devices at the thing tier want to
communicate and do not know each other previously.
They need to infer an initial trust value
since they have little or no information about
the other device. Therefore, the operating steps
are as follows.
Step 1: Each IoT device queries the other's
identity in a BC infrastructure that stores the
identities of members.
Step 2: Once a minimum initial trust is established,
the communication can normally start.
Step 3: As the communication between the
devices happens, the low level takes place.
Each device calculates the relative entropy of
the other device's traffic and uses this information
to adjust the respective values of trust. If a
device starts behaving abnormally,a this will
negatively affect trust and may cause communication
to terminate if it decreases beyond a
previously defined threshold. If the communication
is over, a temporal component reduces
the trust value until it reaches the point where
the devices will have to query the BC again and
restart the whole process.
Therefore, to performdevice trust assignment,
we propose a composition of characteristics that
a.
In our case, an abnormal behavior means any traffic pattern whose distribution
diverges from the expected traffic pattern distribution.
TRji. The TRji is initially computed with the first
component that is based on the trust of the i0s
identity, which is obtained from the associated
reputation of the stored identity in a public-permissioned
BC, expressed by C1 in (1). In the very
beginning of any new devices' communication,
the identities of such devices must be stored in
the BC. Then, the trust value that C1 provides is
based on the number of confirmations a transaction
has on the BC. To cope with the high
computational costs commonly found in proofof-work-based
consensus strategies, we envision
the implementation of full nodes at the edge tier
and not at things tier given that the IoT devices
are known as being resource constrained
C1 ¼ # of confirmations the is identity has:
(1Þ
TRji is also influenced by a second component
C2 (2) that is based on the inverse of the relative
entropy of the traffic, which changes when
the current traffic behavior of the device, represented
by qYji
ðxÞ, deviates from the estimated
traffic behavior, represented by pXji ðxÞ, due to
any type of anomalous condition, where qYji ðxÞ
and pXji ðxÞ are nonnegative, integer random variables
that assume values in the interval Sji ¼
½0; D; 2D; ...;Rmax
ji
ji, with D a positive integer and
Rmax the maximum traffic rate. Such component
is also based on Kullback-Leibler31 divergence
DðpjjqÞ, a type of " distance " between two distributions.
The estimated traffic behavior is
obtained using a Kalman filter with mean ¼ 0
and covariance ¼ 1, since it closely tracks traffic
and does not require too many resources
C2 ¼
1
DðpjjqÞ
;
DðpjjqÞ¼
X
x2Sji
pXji ðxÞlog
pXji ðxÞ
qYji ðxÞ
:
(2Þ
The third component (C3) considered to
compute the trust value in our proposal is a temporal
component that works like a timeout by
decreasing the trust value from the moment
devices stop communicating until they reach a
January/February 2024
17
IEEE Consumer Electronics Magazine - January/February 2024

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2024
