Content uploaded by Fakhrul Alam
Author content
All content in this area was uploaded by Fakhrul Alam on Jan 09, 2018
Content may be subject to copyright.
978-1-5090-5541-8/17/$31.00 ©2017 IEEE
Do RSSI values reliably map to RSS in a Localization
system?
Daniel Konings
1
, Nathaniel Faulkner
1
, Fakhrul Alam
1
, Frazer Noble
1
and Edmund Lai
2
1
School of Engineering and Advanced Technology
Massey University
Auckland, New Zealand
d.konings@massey.ac.nz
2
School of Engineering, Computer and Mathematical Sciences
Auckland University of Technology
Auckland, New Zealand
Abstract—In recent years, research into localization systems
has become more popular as the proliferation of Wireless Sensor
Networks (WSNs) grows. Wireless Localization can refer to either
an “Active” system which tracks a mobile transceiver, or
“Passive” localization which tracks a transceiver free entity by
measuring the changes it makes to the surrounding wireless
environment. Recent work has seen both of these systems
implemented with Received Signal Strength Indication (RSSI)
values from transceivers. Many algorithms and channel models
have been presented to increase the accuracy of a Received Signal
Strength (RSS) based system. In this paper we experimentally
check whether RSSI values map to the expected RSS values within
an IEEE 802.15.4 network. Indoor experiments are repeated
within an ideal outdoor environment, with multiple device
platforms, to eliminate indoor multipath propagation as the cause
for inconsistent behavior of RSSI. We identify 3 key issues with
raw RSSI values and present either a possible solution or a
mitigation strategy to reduce their effect. We conclude that using
raw RSSI values is flawed, as the premise that they uniquely map
to the distance between transceivers is incorrect. However they
may be calibrated to increase their accuracy, and therefore
viability.
Keywords—Indoor Positioning System; Device Free
Localization; Active Localization; Zigbee; Log Distance Path Loss
Model
I.
I
NTRODUCTION
The objective of this paper is to investigate whether raw
Received Signal Strength Indication (RSSI) values reliably map
to Received Signal Strength (RSS), for use in an Indoor
Positioning System (IPS). In 802.15.4 networks, the RSSI is an
8 bit quantized value used to estimate the RSS within the
bandwidth of the channel. We define raw RSS values as the
values received directly from the 802.15.4 receiver, and will use
this interchangeably with raw RSSI within this paper. It is often
implemented in network code as two functions. The first
function estimates the RSS within the channel itself (known as
an ED scan)[1].The second function is used to estimate the RSS
of a received packet, (given by the RSSI field) [1]. This paper
will focus on the second function, power estimates of received
packets. The 802.15.4 standard requires the raw RSSI value to
linearly map (in decibels) to the true RSS and be accurate to ± 6
dB [1]. An IPS using RSSI can be implemented through either
Device-free Localization (DfL) (Passive localization) [2] or
Active localization [3, 4]. DfL is an emerging technology that
can locate moving objects within an area surrounded by wireless
nodes or radios. DfL works by creating a dense network of
“linked pairs” as each radio surrounding the area of interest can
transmit and receive wireless signals. When an object passes
through the links, some of the signal is either absorbed or
reflected by the object, thus resulting in less signal power
(Received Signal Strength) reaching the destination node
(radio). An image of where the power is being absorbed can be
formed by analysing the loss along the “linked pairs”, and thus
a moving object’s location can be detected [5]. Active tracking
utilizes the same information (RSSI), but instead uses it as a
form of wireless ranging where the tracked entity is in contact
with several other nodes at any given time to contribute to the
localization. RSSI Implementations of both methods rely on
either known signal propagation behaviour or wireless ranging.
Wireless ranging is based on the assumption that RSSI values
will monotonically decrease as the separation between the
transmitter and receiver increases [4, 6]. Signal propagation
methods using RSSI are usually implemented by creating an
offline fingerprint of the environment, also known as an RSSI
map. After measuring the environment in an offline setting,
these solutions assume that the variance caused by an entity to
the wireless environment during live operation will be uniquely
identifiable when compared to the stored map.
The goal of this paper is to check whether raw RSSI values
correctly map to their RSS counterparts. This will be done by
taking a common 802.15.4 Transmitter-Receiver (Tx-Rx) pair
(TI CC2530) and checking whether RSSI values uniquely map
to distance measurements in a monotonically decreasing
function in both indoor and outdoor environments. The
experiment is then replicated with another device platform
(Microchip MRF24J40). This ensures that any propagation
irregularities presented are both independent of indoor multipath
effects and are platform independent. This will show whether the
raw RSSI values of an 802.15.4 transceiver provide a good
metric for implementing an IPS. Finally we will analyse the
results and provide suggestions for improvements in an RSS
based IPS implementation.
Here is how the rest of the paper is organized. Section II
discusses previous implementations of a RSSI based localization
system; Section III presents our system implementation; Section
IV covers the implementation results and discussion; Section V
concludes this paper.
II. L
OCALIZATION
M
ETHODS
A. Active Tracking
Active tracking with RSSI works by using wireless ranging
to estimate the distance between beacon nodes and the tracked
node. Once data has been collected for at least 3 separate beacon
links to the tracked node, wireless triangulation is used to
estimate the coordinates of the tracked node. This is based on
the assumption that the power of a received packet decreases as
the distance between the transmitter and receiver increases. The
model commonly chosen to represent this relationship is the
Log-distance path loss model (LdPLM) [7] which can formally
be expressed as:
=
−
−10log
+
(1)
Where
is the power at the receiver in dBm,
is
the transmitted power in dBm,
is the path loss in dB at a
reference distance
, is the path loss exponent and
is a
Gaussian random variable with zero mean.
Common applications will use the model to find distance
estimates between the beacon transmitters and the receiver and
either triangulate an approximate location, or more
simplistically infer a region by checking which beacon(s) the
receiver is closest to. Active tracking can also utilize a radio
mapping approach where offline measurements are taken of the
target in predetermined locations. During online operation,
localization is inferred through similarity matching with the live
RSSI vectors and the offline stored measurements.
B. Device-free Localization
Device-free Localization (DfL) is more difficult than Active
tracking. Since the tracked entity doesn’t carry a transceiver, the
LdPLM cannot directly be used to calculate the entity’s distance
from a known fixed point. These types of systems exploit
wireless signal propagation theory by understanding how RSS
changes as either; Tx-Rx separation increases, or a Tx-Rx link
has an entity pass through it. When an object blocks the line-of-
sight (LOS) path between a transmitter and a receiver, assuming
this is the dominant propagation path (ignoring multipath
propagation), the relative path loss exponent will increase and
therefore the received power at the receiver will be less, as some
has been absorbed by the blocking entity. This behaviour can be
utilized by fingerprinting methods. Fingerprinting creates a
Radio Map similar to that mentioned in Active Tracking. The
difference between the two is that Active tracking stores the
RSSI vector between the static nodes and moving target. DfL
stores the RSSI between all static nodes, whilst the target is
located in a predetermined grid location. Fingerprinting based
on a feature vector set works as follows. First, the localization
area is divided into grids for offline measurements. The target
moves into the first grid and remains stationary. RSSI values are
collected between all static node links and are stored in the
fingerprint database as sample vectors. This is repeated for all
grid locations. When the system goes online, localization is
inferred through similarity matching of the online RSSI vectors
with the stored offline vectors.
A graphical approach to imaging the attenuation has been
developed, known as Radio Tomographic Imaging (RTI)[5].
RTI measures the live variability of the RSSI measurements. By
classifying the variability of the RSSI across multiple Tx-Rx
links, an entity can be located at intersection points when it
attenuates multiple links at the same time.
Both Active Tracking and Device-free Localization assume
that there is a unique relationship between RSSI and distance.
Active Tracking assumes that the received power at a receiver
monotonically decreases (in ideal outdoor free space
environments), as the separation between the transmitter and
receiver increases. DfL assumes that an entity introduced to a
measured area will introduce noticeable and explainable
variability to received RSSI values. This is equivalent to the
(path loss exponent) changing between non-LOS links as the
entity moves through the system for the LdPLM model.
III. I
MPLEMENTATION
We set out to test whether raw RSSI values received from a
common 802.15.4 transceiver do in fact map to the expected
RSS values for a set distance based on the LdPLM. Initial
measurements were taken with a pair of TI CC2530 transceivers.
All measurements have been repeated with a second pair of
CC2530 transceivers to ensure that all observed behaviour is
platform specific rather than device specific. Measurements
were also repeated with a pair of Microchip MRF24J40
transceivers to ensure that the results weren’t platform specific.
All measurements were performed with the radios operating
with their respective highest transmission power. Since RSS
based localization is generally considered for indoor
applications, we took measurements from 1m – 10m transceiver
separation at 1m intervals. This provides a good analog of the
propagation behaviour within a home or small office
environment. The experiments were undertaken in an open plan
room, at the university, at night to minimize external
interference. We also used ZigBee channel 26 as it has the best
separation from Wi-Fi since Wi-Fi channels 12 and 13 are not
utilized in New Zealand. All measurements were taken with the
transceivers on stands at 1m above ground level. When a stand
was moved, we ensured the front edge remained perpendicular
to the line of Tx-Rx separation distances. This ensured that the
transceiver antenna orientation remained constant for all
distances. We were aware that the RSS value may deviate from
a monotonically non-decreasing behaviour due to the presence
of multipath propagation. This is usually accounted for in the
LDPLM by the Gaussian random variable. The wavelength of
2.4 GHz ZigBee transmissions is approximately 0.125m. This
means that a propagation path that is approximately 0.0625m
longer than the main propagation path will arrive with an almost
180
o
phase change. Therefore 2.4GHz ZigBee signals are
potentially very prone to multipath fading in indoor
environments. Since this fading could be either constructive or
destructive, there is the potential for large errors to be introduced
when reading RSSI values indoors. There has been research
undertaken that tries to measure RSSI vs Distance by
minimizing potential interference [8, 9]. The problem is that it is
unclear whether their results are caused by the spurious nature
of RSSI, or by potential indoor multipath effects. Therefore to
provide a best case, interference free, free space propagation test
we repeated the experiment in 2 outdoor locations with multiple
device platforms. The first outdoor location was in an empty
stadium carpark and the second was in the middle of a large
field. Both locations should have been completely free of
2.4GHz interference, and we confirmed with an android
application Wi-Fi scanner that there were no Wi-Fi access points
visible. We also ensured that for the carpark, there was no object
located within 30m in any direction that could contribute
reflective multipath components (excluding the ground itself).
Within the field test scenario, there was no object located within
50m in any direction. When an experiment was started, a delay
was used to allow the testing personnel to completely leave the
testing environment before measurements began to ensure their
physical presence did not affect the results. The experimental
setup consisted of 3 CC2530 nodes and a laptop. The first node
which was the transmitter would send a packet to the second
node at a fixed distance away. The second node would measure
the RSSI of the received packet and then forward the recorded
RSSI information to a third node which was connected to a
laptop collecting all data. This was done to remove the data-
logging laptop from the Tx-Rx pair to ensure that no artefact was
introduced from potential reflective signal components off the
laptop itself. At every 1m distance interval, 5000 measurements
were taken to ensure that the measurements were resilient to fast
fading. The antenna orientation was also kept constant with
every set of measurements to prevent any errors due to
anisotropic propagation (even though the antenna was
omnidirectional).
IV. R
ESULTS AND
D
ISCUSSION
For ease of use, when fitting a curve to the raw RSSI results,
we simplified (1) to:
= − 10log
(2)
where represents the transmitted power in dBm minus the path
loss in dB as calculated at a reference
of 1m, and the guassian
zero mean random variable. Figures 1, 2 and 3 all show the fitted
LdPLM as compared to median RSSI values per location.
Median RSSI was chosen over mean RSSI as it is more resilient
to influence from outliers since our sample size was large. To
check that the behaviour of the raw RSSI fluctuations wasn’t
platform specific, we also tested with the Microchip MRF24J40
transceiver. We noticed that the variation of raw RSSI values
was larger on the MRF24J40 than it was on the CC2350
platform. However the collected data were consistent with the
Figure 1 - RSSI values per distance within an indoor environment Figure 3 - RSSI values per distance within an outdoor environment (Carpark)
Figure 2 - RSSI values per distance within an outdoor environment (Field) Figure 4- RSSI values per distance within an outdoor environment (Carpark)
CC2530 results. The CC2530 maps the register RSSI value to a
RSSI dBm estimation by subtracting a constant dB offset. This
means that the RSSI (dBm) is always an integer like the
quantized RSSI register (which has a step size of 1dB). The
MRF24J40 includes both a scaling factor and an offset when
converting from the RSSI register to dBm. This means the
quantized register has an equivalent stepsize of 0.2dB
*
.
Therefore Figure 4 contains fractional dBm values whilst
Figures 1, 2 and 3 are all whole numbers. There are three causes
for concern that our measurements raise when considering the
use of raw RSSI values for localization purposes outlined as
below.
A. Uniqueness Error
We define an RSSI value as unique, if it only maps to 1
corresponding Tx-Rx separation distance. The data shows that
for both indoor and outdoor (ideal) environments, there exist
regions where a raw RSSI value does not map uniquely to a
distance value. This can be seen most clearly in Figure 3 where
distances d
4
, d
7
and d
8
all have the same median RSSI value.
This appears again with d
6
and d
9
, and in Figure 2 with d
8
and
d
10
. Since all experiments were instigated with interference
minimization in mind, we conclude that this behaviour is most
likely due to intrinsic hardware peculiarities rather than an
multipath fading skewing the readings. It should also be noted
that the RSSI values received are well within the acceptable mid-
range, and therefore it is not an issue relating to possible
erroneous readings near the transceiver sensitivity threshold.
This is also backed by our measurements from within a field
which show very low RSSI variation with some distances, even
maintaining the same RSSI reading across all 5000 samples (d
1
,
d
4
, d
5
and d
9
). This suggests that the effect of multipath
propagation and environmental noise is either minimal, or
remained relatively constant over the duration of the testing.
This behaviour can also be clearly seen on the Grenoble and
Strasbourg SensLAB platforms, at set distances, which utilize
256 TI CC1101 radios respectively [8]. It has also been seen to
occur on the popular TI CC2420 [9]. Lui et al experience this
behaviour with Wi-Fi propagation in both indoor and outdoor
environments across multiple device platforms [10].
This is a significant issue that would affect both Active
Localization and DfL. For an Active Localization system
utilizing wireless ranging, this means that for any singular link,
there will be raw RSSI values that map to multiple possible
“actual” distances. This is problematic as the system could
appear to be working correctly for the majority of the time, but
could experience errors whenever a roaming node returns an
RSSI link value which could correlate to multiple physical
locations. A different issue arises due to the non-unique
mapping within a DfL environment. DfL systems work by
measuring the environmental change in RSS when an entity
enters the system. The uniqueness error means an entity
attenuating a Tx-Rx pair by varying amounts could potentially
return the same RSSI values. The is problematic for both
variance and mean based RSSI DfL systems as the system will
return erroneous readings when the RSSI values transition
between unique and non-unique regions. This adds complexity
as Tx-Rx links have to be weighted not only by their own
variance, but also other local link variances, and by whether the
returned RSSI values themselves are unique.
There has been some research done that suggests that radios
may not have necessarily have a linear relationship between raw
RSSI values and RSS for all regions [11]. Chen and Terzis
experimentally verify that non-linarites are platform specific.
Therefore if the non-linear regions are known for the
implemented radio, they can be accounted for accordingly. This
is done by understanding the following relationship for RSSI
given by Chen and Terzis[11]:
=
−
(3)
where RSSI
RX
is the received power, RSSI
TX
is the transmitted
power and PL is the path loss. When an RSSI value is read and
maps to a known non-linear region, the receiver asks the
transmitter to retransmit at a different known power level. This
is repeated until the receiver receives a RSSI value from a known
linear region. This is then used with the known transmitted
power level to calculate the PL factor. Since the PL is constant
irrespective of the transmission power, this can be used
alongside the known original transmission power to calculate an
appropriate received RSSI value (calibrated). This calibrated
value will result in decreased RSSI variance when measuring
between two stationary locations and will change the mean /
median of known erroneous non-linear points. This could
potentially reduce the uniqueness error caused by multiple
consecutive RSSI values falling within a flat non-linear region,
as they will now have unique values.
B. Deviation Error
We refer to a RSSI value as a potential deviation error when
the median (and / or mean) of the RSSI at any set distance lies
significantly away from RSSI values either side of it. This is
most visible for d
7
in both Figure 1 and Figure 2 which we term
as deviation locations. When propagation peculiarities are seen
in indoor environments, there are often assumed to be due to
multipath behaviour. Since the deviation error seen indoors in
Figure 1 is also apparent in outdoors in Figure 2, multipath
propagation is unlikely to be the cause. This again suggests a
hardware specific peculiarity that should be considered
whenever using RSSI values. This behaviour has also been seen
in RSSI vs Distance measurements taken on the popular TI
CC2420 [12] and in a variety of Wi-Fi platforms [10].
For an Active system utilizing ranging, this phenomenon would
introduce large localization errors at the deviation locations if
raw, uncalibrated RSSI values were used within a range based
estimation system. For a variance based DfL system, larger
variation within links would be seen when an RSSI link value
was received which had the same value as a deviation location
from a standard RSSI vs Distance plot. As long as the weighting
was not placed solely on the magnitude of the RSSI variance and
rather on other known spatial properties, this wouldn’t cause
large localization errors, but it needs to be accounted for. Whilst
the underlying erroneous raw RSSI values are hard to correct in
this scenario, preventative measures can be taken to mitigate
their effect. By implementing sanity checks and using a
probabilistic movement approach (by ensuring any localized
entity must travel through a surrounding region to reach a far
location), pathing can be established [13]. This will improve the
* The MRF24J40 is based on the UZ2400. The UZ2400 datasheet provides a
register to dBm RSSI conversion whilst the MRF24J40 does not. Thus it is
assumed the MRF24J40 follows the same mapping.
accuracy in an Active ranging system since walking through a
deviated location will not result in a predicted location set far
away from the last known location. This error can also be
minimized by utilizing more sensors, thus increasing the system
resolution and making it easier to filter out suspected erroneous
raw RSSI values. Finally, Zanella and Bardella show that the
accuracy of range based localization can be improved by
averaging the RSSI across multiple channels [14]. They show
that this approach reduces the RSSI variance at a location, which
could fix the problem of deviation locations if they prove to be
a channel specific error. Though this reduces the effect of indoor
multipath, it would not fix non-injunctive regions. Their results
show large improvements on the raw RSSI values, but
Uniqueness Error and Deviation Error still appear to be present.
C. Asymmetrical Error
The 802.15.4 standard requires the RSSI dBm value to be
within ±6dB of the true value. The CC2530 claims to meet ±4dB
of the true value. A 12dB swing is very large considering that
ideal free space propagation assumes 6dB loss per octave. We
did not have the equipment to accurately read the received RSS
of a node (as this would require an accurate spectrum analyser).
But we did notice that the raw RSSI within a link was not
symmetrical, i.e. the RSSI value read from a packet at node B,
sent from node A was not equal to the RSSI read at node A, from
a packet sent by node B. This behaviour was also seen in a large
scale indoor environment implementing a system based on TI
CC1101 radios [8]. Through experimentation we found that
though the links were asymmetrical, they were separated by a
linear dB offset (O
dB
). Thus we propose the following formula:
=
+
=
+
(4)
This shows that whilst any CC2530 device offset (DO
dB
) will be
within 4dB of the true value, it doesn’t mean that all devices
from the same product line will have the same offset (otherwise
all links would be symmetrical in an ideal environment for
identical radios). In our testing, the dB offset (O
dB
) between 2
nodes, appears to be constant across the range (1-10m).
The results show that the Asymmetrical Error will have little
effect for variance based DfL systems, as the term cancels out
when taking the difference between two RSSI measurements,
measured from the same node. It will however have a large
effect on range based Active Localization systems. This finding
means that either each link needs to be calibrated with its own
model specific coefficients, or the raw RSSI values need to be
pre-calibrated to the same mapping, before use in a range based
system. This can be done by choosing a reference node, and
calibrating all offsets to be equal to the chosen node. Care needs
to be taken as even after calibration, it may not be advisable to
deploy a single model (such as LdPLM) over the global system.
If the system is implemented within an indoor environment, the
path loss exponent could vary due to walls, and multipath
propagation will have an effect on the system. Therefore it may
still be advisable to use multiple ranging propagation models for
various links in a system where propagation path loss could vary
greatly between different nodal links.
V. C
ONCLUSION
As we have shown, raw RSSI values do not follow a
monotonically decreasing function. We have also shown that
propagation peculiarities still occur in ideal outdoor
environments and therefore are not due to multipath fading
effects. This is significant, as it shows that raw RSSI values do
not provide a consistently accurate measure of true RSS in
802.15.4 devices. Three major concerns were raised and either a
solution or mitigation strategy was proposed. We conclude that
raw RSSI should not be used for localization purposes, but it
may be possible to calibrate the raw RSSI values to provide a
more usable localization metric. Users need to be wary of the
limitations of device reported raw RSSI values, and address
them accordingly to increase the accuracy of RSSI values.
VI. R
EFERENCES
[1] "IEEE Standard for Low-Rate Wireless Networks," IEEE Std
802.15.4-2015 (Revision of IEEE Std 802.15.4-2011), pp. 1-709,
2016.
[2] N. Pirzada, M. Y. Nayan, F. S. M. F. Hassan, and M. A. Khan,
"Device-free Localization Technique for Indoor Detection and
Tracking of Human Body: A Survey," Procedia - Social and
Behavioral Sciences, vol. 129, pp. 422-429, 5/15/15 May 2014
2014.
[3] T. Alhmiedat, G. Samara, and A. O. A. Salem, "An indoor
fingerprinting localization approach for ZigBee wireless sensor
networks," arXiv preprint arXiv:1308.1809, 2013.
[4] O. Hernandez, V. Jain, S. Chakravarty, and P. Bhargava, "Position
location monitoring using ieee® 802.15. 4/zigbee® technology,"
Beyond Bits, issue IV, pp. 67-69, 2009.
[5] J. Wilson and N. Patwari, "Radio Tomographic Imaging with
Wireless Networks," IEEE Transactions on Mobile Computing, vol.
9, pp. 621-632, 2010.
[6] Y. Huang, J. Zheng, Y. Xiao, and M. Peng, "Robust Localization
Algorithm Based on the RSSI Ranging Scope," International
Journal of Distributed Sensor Networks, vol. 2015, 2015.
[7] T. S. Rappaport, Wireless communications: principles and practice
vol. 2: Prentice Hall PTR New Jersey, 1996.
[8] K. Heurtefeux and F. Valois, "Is RSSI a Good Choice for
Localization in Wireless Sensor Network?," presented at the
Proceedings of the 2012 IEEE 26th International Conference on
Advanced Information Networking and Applications, 2012.
[9] A. T. Parameswaran, M. I. Husain, and S. Upadhyaya, "Is rssi a
reliable parameter in sensor localization algorithms: An
experimental study," in Field Failure Data Analysis Workshop
(F2DA09), 2009, p. 5.
[10] G. Lui, T. Gallagher, B. Li, A. G. Dempster, and C. Rizos,
"Differences in RSSI readings made by different Wi-Fi chipsets: A
limitation of WLAN localization," in 2011 International
Conference on Localization and GNSS (ICL-GNSS), 2011, pp. 53-
57.
[11] Y. Chen and A. Terzis, "On the mechanisms and effects of
calibrating RSSI measurements for 802.15.4 radios," presented at
the Proceedings of the 7th European conference on Wireless Sensor
Networks, Coimbra, Portugal, 2010.
[12] M. Petrova, J. Riihijarvi, P. Mahonen, and S. Labella, "Performance
study of IEEE 802.15.4 using measurements and simulations," in
IEEE Wireless Communications and Networking Conference, 2006.
WCNC 2006., 2006, pp. 487-492.
[13] P. Gao, W. Shi, W. Zhou, H. Li, and X. Wang, "A Location
Predicting Method for Indoor Mobile Target Localization in
Wireless Sensor Networks," International Journal of Distributed
Sensor Networks, vol. 9, March 1, 2013 2013.
[14] A. Zanella and A. Bardella, "RSS-based ranging by multichannel
RSS averaging," IEEE Wireless Communications Letters, vol. 3, pp.
10-13, 2014.
