IEEE Consumer Electronics Magazine - March 2018 - 74

The signal is generally strong enough
to offer coverage of a few meters,
and it has a strong penetration
power that allows sensing signals
to pass through walls.
gyroscopes, infrared sensors, radio sensors, and camera sensors. However, these methods have limitations, such as high
cost, limited communication distance, and obstructed signals.
A wireless sensor and Wi-Fi-RSSI recognition technology
can provide an alternative method for the interpretation of
hand gestures.
The RSSI represents the power contained in a received
radio signal. In a Wi-Fi signal, the RSSI is measured by the
energy observed at the antenna. Furthermore, RSSI values
change with respect to alterations occurring in the environmental conditions (e.g., there are objects moving) according
to the Doppler effects and Wi-Fi modulation schemes.
Wi-Fi is a popular communication technology that is widely adopted in different kinds of services, and an RSSI can be
easily obtained by using any Wi-Fi-enabled device (e.g.,
smartphones). The signal is generally strong enough to offer
coverage of a few meters, and it has a strong penetration
power that allows sensing signals to pass through walls.
Moreover, based on the Wi-Fi-RSSI recognition technology,
we can build a hand-gesture sensor interface on Wi-Fi-enabled
devices for various purposes and test the interface to determine which of the gestures are more identifiable.
A user application is developed in this study that turns a
Wi-Fi-enabled device into a hand-gesture sensor, and the application continuously detects RSSI values. Whenever a user
performs a hand gesture, the RSSI values change, and by
detecting the change, the application recognizes the corresponding hand gesture. Furthermore, the application performs different system functions based on the different hand
gestures performed.

WAYS TO MEASURE AN RSSI
The changing point detection algorithm [1] uses overlapping
sliding windows to increase the signal intensity. It uses the
Bayesian information criterion to detect the changing point and
compares the signal to the predefined models in the database.
This algorithm also used the maximum likelihood method to
improve its recognition accuracy. Moreover, the algorithm also
provides noise detection, which can further improve the accuracy. However, it needs an additional Wi-Fi transmitter and
cannot distinguish between noise and small movements.
The Doppler-only tracking algorithm [2] detects the Doppler shifts of Wi-Fi signals and mobile phone signals to track
the movements of people, and it uses the time vector and the
location of the access point to locate people. It also applies
different particle filters to achieve a higher accuracy and
74 IEEE Consumer Electronics Magazine

MARCH 2018

detect changes of directions. The advantages of this algorithm
are that it can achieve a higher accuracy when using the
sequential importance resampling (SIR) particle filter and
support the detection of multiple targets. However, this algorithm is mainly designed for location tracking and not handgesture recognition. Moreover, the complexity of this algorithm
increases the processing power and feedback time.
The passive system approach [3] does not require a specific
transmitter, because it receives the RSSI from peripheral access
points. In contrast, the approach needs to change the firmware
of the receiver's wireless adapter to monitor the mode and run
packet-capturing tools on the receiver to obtain RSSI values.
This algorithm has the advantages of being lightweight and
flexible. First, users do not need to buy a specific transmitter. In
addition, the approach only retrieves the RSSI values without
performing any filtering or post-processing functions, thus
making the approach more flexible and extendable.

WAYS TO RECOGNIZE RSSI CHARACTERISTICS
After evaluating the changes of the Wi-Fi-RSSI values
under different settings (e.g., different obstacles and hand
gestures), some characteristics of the RSSI were identified.
To recognize these characteristics, three methods were used.
The first utilized the maximum likelihood estimation, which
is a well-known method. It is measured by taking the mean
and variance of samples to generate the observed results
that are most probable. The second method used was the
hidden Markov model, which is a statistical model
employed to find a certain pattern in unobserved parameters. It includes states that can represent the relationship
between the time and variable. The third method used classification accuracy [4] to measure the classification performance. It was determined by the number of correct
predictions from all of the predictions made, as shown in
the following equations:
Sensitivity = TP/ ^TP + FNh,

(1)

Specificity = TN/ ^FP + TNh,

(2)

Positive predictive value = TP/ ^TP + FPh,

(3)

Negative predictive value = TN/ ^FN + TNh,

(4)

Efficiency = ^TP + TNh /
^TP + TN + FP + FNh .

(5)

In the equations, TP stands for true positive prediction, TN is
the true negative prediction, FP stands for false positive prediction, and FN is the false negative prediction.

RELATED WORKS
There are real-life applications that currently use the features
of a Wi-Fi RSSI. A detection system in [5] is devised using
the Doppler-shift characteristics of Wi-Fi signals to locate
malefactors hidden in the dark. Moreover, the passive radiation from household Wi-Fi routers under IEEE 802.11 b, g, n,

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - March 2018