Available via license: CC BY 4.0
Content may be subject to copyright.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
QuadScatter: Computational Efficiency in Simultaneous
Transmissions for Large-Scale IoT Backscatter Networks
Ousmane Zeba, Student Member, IEEE, Kentaro Hayashi, Kazuhiro Kizaki, Shunsuke Saruwatari, Member, IEEE,
and Takashi Watanabe, Member, IEEE
Backscatter communication has attracted attention owing to its ultra-low-power consumption ability, which is expected to enhance
internet of things (IoT) technology that aims to enable many novel applications for object-to-object communication. Such a network
with a large and continuously increasing number of connected objects will benefit significantly from resource-saving. This work
introduces a system named QuadScatter, which is a set of algorithms that select and associate transmitters, tags and readers to
enable simultaneous backscatter transmissions and increase network capacity. Consequently, the energy consumption in the network
is considerably lessened. Intensive simulations have been conducted to demonstrate the effectiveness of backscatter simultaneous
transmissions. QuadScatter shows promising results compared to the exhaustive search algorithm. The simulation results highlight
computational time and simultaneous transmission improvements of at least 250x and 2x, respectively. Furthermore, while the
exhaustive search is limited to a few nodes (<20), our proposal uses numerous nodes. Additionally, an implementation of limited
simultaneous backscatter transmissions is conducted to show its feasibility in the real world.
Index Terms—Computational cost, simultaneous backscatter transmissions, quadtree, quadtree quadrant, nearest neighbor search,
exhaustive search, IoT.
I. INT ROD UC TI ON
THE Backscatter device is attracting attention owing to
its ultra-low-power wireless communication capabilities.
The ability to consume the power of a few milliwatts is sig-
nificant for the upcoming internet of things (IoT) technology.
The IoT aims to improve numerous novel industrial or non-
industrial applications, where objects will be able to respond
to the presence of people and other objects, anytime and
anywhere [1], [2].
The impending IoT technology will surely benefit from si-
multaneous backscatter transmissions, where several backscat-
ter devices transmit their data to different readers concurrently
in the same environment with a single carrier signal. This is
because when many tags are deployed in the presence of an
equivalent number of transmitters and receivers, simultaneous
transmissions inevitably occur. Besides, selecting an ideal
threesome source-tag-reader can improve the capacity of the
network. More importantly, energy and time saving is to
be seriously taken into account owing to the exponentially
growing number of connected devices. For all these reasons
many research works are carried out on the backscatter device
for IoT.
In [3], the detection problem of multiple passive and semi-
passive RFID tags with impulsive backscattered signals was
addressed in ultrawideband (UWB) technology. The goal was
to improve the reader’s capability and enable robust tag
detection, even in the presence of multi-tag interference as
well as clock drift effects.
To overcome challenges related to density and transmission
ranges in urban Low-Power Wide Area Networks (LP-WANs),
Choir has been presented in [4]. It allows a maximum of 10
Manuscript submitted on March 4, 2021. This work was supported by JSPS
Science Research Grants JP16H01718.
The authors are with the Department of Information Networking, Grad-
uate School of Information Science and Technology, Osaka University,
Osaka 565-0871, Japan (Corresponding author: Ousmane Zeba, e-mail:
ousmane.zeba@ist.osaka-u.ac.jp).
concurrent transmissions and is capable of retrieving data from
devices located as far as 2.65km. This is achieved by applying
a chirp spread spectrum modulation [5].
Authors in [6]–[8] investigate the ambient backscatter sys-
tems as a promising green communication technology. [6]
explores the channel capacity and outage performance to pro-
pose a numerical method that facilitates capacity assessment.
In [7], protocols are presented to enable energy harvesting
and data transmission for a hybrid ambient backscatter. The
work presented in [8] exhibits a backscatter system including
a full-duplex access point that simultaneously transmits an
orthogonal frequency division multiplexing signal and receives
a time division multiple access signal. The goal is to maximize
the minimum throughput of backscatter devices by optimizing
the backscatter time and reflection coefficient, and power
allocation to the access point.
Works such as [9] and [10] address resource allocation
issues. [9] studies the coexistence of a backscatter network
and a legacy network to derive the outage probabilities related
to each network. [10] considers a wireless powered backscatter
network and maximizes its energy efficiency by optimizing the
transmission power of the power beacon and backscattering
coefficients.
A relay cooperation scheme for performance enhancement
in backscatter communication systems is proposed in [11]. The
backscatter device reflects the incident signal from a power
beacon to the reader and a relay simultaneously. The relay
decodes the received signal and subsequently forwards the
decoded signal to the reader as a redundant signal. The system
throughput maximization problems are formulated considering
a relay with or without a power supply. The aim is to maximize
system throughput by finding optimal time allocation schemes
for the relay.
The study in [12] addresses the capacity maximization
problem by optimizing the deployment of the backscatter
network. A power beacon and backscatter devices are deployed
following a random Poisson cluster process, wherein the power
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
beacons constitute the cluster centers as it is a wirelessly
powered backscatter network. Thus, the maximization of the
transmission capacity relies on the distance between the power
beacon and the backscatter device, the distance between the
receiver and the backscatter device, and the energy harvesting
capability of the backscatter device.
Some works focus on transmission scheduling such as [13]
and [14]. [13] investigates hybrid wirelessly powered IoT
devices that operate in passive and active mode. In passive
mode, the backscatter device acts as an ambient backscatter.
In active mode, it uploads its data to a hybrid access point
that also wirelessly charges the hybrid backscatter tag. It aims
to determine the transmission schedule that maximizes the
minimum throughput. In [14], simultaneously power supplied
backscatter devices that are able to transmit data and backscat-
ter incident signals are studied. The goal is to find the optimal
scheduling between backscattering, transmitting and/or energy
harvesting.
Some other works consider coding techniques to perform
concurrent transmissions utilizing one transmitter and one
receiver for multiple tags. In [15], a large scale backscatter
network named NetScatter is presented. NetScatter is a wire-
less protocol that enables a network of 256 backscatters to
be transmitted concurrently on 500kHz channel bandwidth.
The number of concurrent transmissions increases according to
the bandwidth, reaching a thousand concurrent transmissions
for only 2MHz of bandwidth. This performance is achieved
using a distributed chirp spread spectrum coding based on
chirp spread spectrum modulation and ON-OFF keying. Fur-
thermore, NetScatter can operate with weak backscatter sig-
nals around a noisy floor. However, time synchronization is
compulsory for NetScatter. The backscatter devices must be
synchronized when sending data to the access point (AP).
Besides, hardware delays such as clock frequency variations,
inherent to backscatter devices, and propagation delays must
be corrected. The aforementioned timing mismatch is difficult
to avoid with an increasing number of backscatter devices.
[16] studies concurrent backscatter transmission involving one
transmitter and one receiver for several backscatter tags. The
coding technique presented is called Coded-Backscatter Mul-
tiple Access (CBMA), which relies mainly on Code Division
Multiple Access (CDMA).
In most of the aforementioned works, the simultaneous
transmission scheme does not necessarily make use of the
same carrier signal for backscatter transmissions. When the
simultaneous backscatter transmissions utilize the same carrier
signal like in [15] and [16], the backscattered signals are
oriented towards the same reader. Furthermore, signal coding
techniques are performed to achieve their goal. In our case,
the simultaneous backscatter transmissions are oriented to
different readers and achieved through algorithms. Indeed,
in a densely deployed and controlled backscatter network,
algorithms are more appropriate to select and associate trans-
mitter, receivers and tags to enable a backscatter simultaneous
transmission from one transmitter to several receivers. As such,
signal coding and device synchronization are not needed as
each reader receives only one backscattered signal. Moreover,
our proposal is suitable for cases where concurrent backscatter
transmissions similar to [15] and [16] are not applicable.
The contributions of this work are as follows:
•Algorithms that select and associates the best transmitter-
receiver pair to the backscatter device according to the
distance, to maximize link capacity.
•Simultaneous transmissions are enabled by allowing mul-
tiple tags to backscatter the same carrier signal to their
respective receivers whenever it is permissible.
•Furthermore, the proposed algorithms drastically reduce
the computational time and cost in a large-scale backscat-
ter network.
•Finally, an experimental implementation of our proposal
is presented to verify its feasibility in the real world.
The remainder of the paper is structured as follows: Section
II states the problem of simultaneous transmissions in a
large scale backscatter network. Sections III and IV describe
the algorithms designed for this study. Section V draws the
numerical analysis of simultaneous transmissions. Sections VI
presents the simulation and experimental results, whereas the
conclusion follows in Section VII.
II. PRO BL EM STATE ME NT
We consider a scenario where nbackscatter tags and m
APs are deployed in a particular area. Backscatter tags are
passive devices that only reflect [17] incident signals from their
surrounding environment. Similar to the cases in [15] and [18],
the transmitter APs produce an unmodulated carrier signal to
be modulated by the backscatter only. No other devices are
involved in the network or are supposed to communicate with
the APs. Besides, there is a network controller as illustrated
in Fig. 1, which is responsible for tag-to-AP assignment over
the same radio channels used for backscatter communication.
Transmission capacity is known to be an essential parameter
in a network. Similarly, a major factor of the capacity accord-
ing to Shannon’s law is the signal-to-interference-plus-noise
ratio (SINR) [19] [20]. SINR is equivalently replaced by the
signal-to-interference ratio (SIR) in this study, neglecting the
ambient noise for simplicity. Unfortunately, the SIR is subject
to the distance between the transmitter and receiver, and the
interference occurring at the receiver. Obtaining a high SIR,
which in turn provides an interesting and reliable transmission
capacity, requires avoiding or mitigating the interfering signals
and reducing the distance between transceivers. Channel state
information is not considered in the current study due to the
difficulty of generating training symbols with the backscatter
device.
In our bistatic backscatter system, we employ APs for only
transmitting carrier waves or for receiving the backscatter
modulated signal. Therefore, we can consider the signals from
other backscatters as the only interference in addition to the
ambient noise. Fig. 2 presents the devices and the different
signals.
According to [21], considering only the signal from the tag
to the reader, we can write:
SI Rz|n=τ Prd−β
(y,z)
n
iτ P ˙rd−β
( ˙yi,z)
,(1)
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
where Prand τare the received power at the backscatter tag
and a factor introduced by the backscatter tag, respectively.
(0< τ < 1) and β= 2 to 4is a propagation constant. x,y
and zare respectively the positions of the source transmitter,
the tag, and the receiver AP. P˙rand ˙yiare related to the other
tags that are source of interference. PEis the transmit power
from the source transmitter, and d(x,y)is the distance between
source and tag; we have
Pr=PE
dβ
(x,y)
.(2)
Hence Eq. 1 becomes
SI Rz|n=τ PEd−β
(x,y)d−β
(y,z)
n
iτP ˙
Eid−β
( ˙xi,˙yi)d−β
( ˙yi,z)
.(3)
Assuming that all the sources are equal in power then Eq. 3
can be written as:
SI Rz|n=d−β
(x,y)d−β
(y,z)
n
id−β
( ˙xi,˙yi)d−β
( ˙yi,z)
,(4)
which shows that the SIR depends only on the distances
between the nodes involved in the transmission and the in-
terfering signal.
With the aforementioned issues and the given parameters,
we focus on maximizing the SIR of the receiver to improve
the transmission capacity of the network. Some works have
already addressed the issue of interference mitigation such as
[22]–[24]. Thus, we opted to treat that issue in a simple but
different manner as follows:
•First, we build an assignment or selection algorithm to
optimize the received power on the reader’s side. The
received power depends on the distance between the
sender and receiver, assuming that all the carrier signals
are emitted with the same power. Thus, if the nearest
transmitter and reader in the vicinity of the tag can
be selected, it would be easier to obtain a maximized
received power. Note that the backscatter tag is neither
able to select a carrier signal nor decide which reader
should receive the reflected signal. Therefore, there is a
controller that controls all Network components.
•Second, we determine the maximum number of simul-
taneous backscatter transmissions that can be tolerated
by the reader. By simultaneous transmissions, we refer
to multiple tags reflecting the same incident signal to
different readers at the same time. The signal from other
tags creates interference to the inbound signal to each tag.
According to the SIR threshold for successful decoding of
the signal, there is a limit to the acceptable interference,
which corresponds to a certain number of interfering
signals and backscatter devices.
III. PROP OS ED AL GO RI TH M
This section describes our proposed algorithm that max-
imizes the transmission capacity and enables simultaneous
transmissions in a backscatter network.
ĐĐĞƐƐƉŽŝŶƚ;WͿ ĂĐŬƐĐĂƚƚĞƌƚĂŐ EĞƚǁŽƌŬĐŽŶƚƌŽůůĞƌ
Fig. 1: Backscatter network model.
Transmitter
signal Backscatter
signal Interference
signal
Fig. 2: Simultaneous backscatter transmissions: Signals and
interferences.
In an already deployed dense network, redeploying to a
more suitable topology would cost tremendous time and effort.
To avoid redeployment and improve the transmission capacity
of the network, we apply our AP selection technique that
maximizes transmission capacity and allows simultaneous
transmissions.
The AP selection algorithm searches for APs (transmitter
and receiver) that are most suitable to each tag in terms of
distance. It selects the two closest APs to the tag. For a large
number of tags and APs, finding the nearest neighbor can
become NP-hard owing to the computational complexity. For
nAPs and mtags, finding two nearest neighbors of each
tag comes with the complexity of O(2n)by making a naive
comparison with the exhaustive search algorithm. The high
computational cost is attributable to the process of finding
the most appropriate AP. Therefore, a nearest neighbor search
algorithm would considerably reduce the computing cost.
A. Nearest Neighbor Search
There are many algorithms for the nearest neighbor search.
Among them are Locality Sensitive Hashing (LSH) and
quadtree.
The LSH algorithmic technique creates a bucket of input
items with a high probability of similarity using hash functions
[25]. Since it separates similar items from dissimilar ones,
LSH can be used for the nearest neighbor search. However,
LSH is more convenient in high-dimensional space, which is
different from our two-dimensional space. For example, the
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
Python implementation of LSH in [26] is only suitable for
data samples with high dimensionality.
Quadtree is an algorithmic technique used to partition
a two-dimensional space by recursively subdividing it into
four quadrants or regions [27]. The subdivided regions may
have arbitrary shapes. Quadtree decomposes the space into
buckets. Each bucket has a maximum capacity, upon reaching
which the bucket divides into four quadrants. The quadtree
algorithm is perfectly suited to our situation because it is
conceived for two-dimensional space. For its suitability and
simplified mechanism, we retain quadtree to be a part of our
assignment algorithm for the process of finding the nearest
neighbor. Furthermore, the quadtree algorithm has a worse
case complexity of O(n), which is significantly lower than
the complexity of the exhaustive search algorithm.
B. QuadScatter
QuadScatter is the algorithm we define for assigning APs to
backscatter tags. Mainly four phases compose the QuadScatter
algorithm: Applying quadtree to create quadrants, nearest
neighbor selection phase, common transmitter search phase,
and transmitter-receiver assignment phase.
1) Define and create an adequate number of quadrants
This phase is a crucial step in the entire process because
it subdivides the target area into quadrants allowing an easier
localization of nodes. An implementation of a quadtree allows
one to decide the maximum number of nodes in each quadrant
after receiving the coordinates of the targeted area. Addition-
ally, it is possible to limit the number of divisions by stating
the maximum range of divisions. This phase of QuadScatter
receives the area size and the coordinates of nodes deployed
on it as an input, Algorithm 1 line 2. As an output, it provides
the resulting quadrants and the points associated with each of
them, Algorithm 1 line 3.
Algorithm 1: CRE ATEQUADRANT creates quadrants
in area A
Input: A two-dimensional area Aand the coordinates
of its network nodes
Output: A finite set Q={q1, q2, . . . , qn}of quadrants
1Q← ∅
2Quadtree(A)
3Q← {q1, q2, . . . , qn}
4return Q
2) Nearest neighbor selection
Instead of randomly selecting the transmitters and readers, a
thorough search of the nearest APs would provide a better re-
sult since SIR and capacity depend on the distances separating
the two nodes.
As already mentioned in section III-A, we apply quadtree to
an area to lower the computational complexity of our proposed
algorithm. The quadrants or regions resulting from quadtree
contain a few devices that we address with less complexity. For
a particular quadrant, containing mtags and nAPs, Algorithm
2 compares the distances separating tags and APs, in lines 3 to
6. Subsequently, it returns an array of the nearest APs to each
AP’s location Tag’s location
Fig. 3: Quadrants for nearest neighbor search.
tag, Algorithm 2 line 7. In that array, each line corresponds
to a tag in the quadrant. The same line is a sorted list of APs
where the first AP is the nearest one.
Algorithm 2: FIN DNE AR ES T finds the nearest APs in
the quadrant
Input: A quadrant qof a particular area and its points
Output: An array Aof nearest APs to each tag
1A← ∅
2list ← ∅
3for i←1to mdo
4for j←1to ndo
5compute distance dij
6list ←dij
7sort list
8A←list
9return A
3) Common transmitter search and assignment
The main objective of this stage is to avoid the usage
of multiple transmitters while only one suffices for several
backscatter tags. The corresponding algorithm (Algorithm3)
is in charge of finding that common transmitter and works as
described in the following lines. Algorithm 3 selects the first
element of the array returned by Algorithm 2 and compares
it to the elements of the following line of the array, lines 7
to 8 in algorithm 3. In case it matches any element in that
line it jumps to the next line, line 9 of Algorithm 3. If the
selected element matches no element in the following lines,
Algorithm 3 returns to the subsequent element in the first line
of the array, as described in lines 11 and 12. After any line is
finished, Algorithm 3 ensures to always select the first element
of the next line as shown in line 14. The algorithm stops
searching for the common element when no common element
is found before reaching the last line or after comparing the
last selected element to the last line of the array, lines 15 and
16.
4) Transmitter and receiver assignment
At this stage, the appropriate tags and APs are assigned to
each other by the network controller. This process completes
the backscatter network and enables communications between
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
Algorithm 3: FIN DCO MM ON finds the common ele-
ment to the lines of an array
Input: An array of ncolumns and mlines
Output: The common element to all the lines
1i←0
2j←0
3line ←0
4common ← ∅
5for k←1to n−1do
6while i≤nand j≤mdo
7common ←A0,k
8if common ̸=Ai,j then
9j←j+ 1
10 else
11 i←i+ 1
12 j←0
13 line ←i
14 j←0
15 if line =mthen
16 break
17 return common
network nodes as described in algorithm 4. Algorithm 4
assumes a set of APs and tags associated with one quadrant
as an input. First, it ensures that a common AP is either found
or not. If a common AP exists, it is set as the transmitter for
all the tags in the quadrant. Once the common transmitter is
designated, the nearest AP to each tag other than the common
AP is set as the receiver for the appropriate tag, as illustrated in
lines 1 to 8 of Algorithm 4. In the case where no common AP
is found, the nearest AP to each tag is assumed as a receiver
and the second one as a transmitter. The latter case is described
in lines 9 to 14 of Algorithm 4.
Algorithm 4: ASS IG N assigns APs to tags
Input: A finite set Aof nAPs and mbackscatter tags
Output: Selected and assigned APs and tags
1if common ̸=∅then
2common ←transmitter
3for i←1to mdo
4for j←1to ndo
5if Ai,j ̸=transmitter then
6Ai,j ←receiver
7Ai,j ←tagi
8break
9else
10 for i←1to mdo
11 Ai,0←receiver
12 Ai,1←transmitter
13 Ai,0←tagi
14 Ai,1←tagi
15 return 0
Channel(ch)
161
11 11
11
11
11
11
11
1
1
1
1
1
1
6
66
6
6
6
ch #
Fig. 4: Quadrant channel assignment in ISM 2.4 GHz band.
IV. INT ER -QUADRANT INT ER FE RE NC E
As depicted in the above sections, QuadScatter focuses
on the devices inside a quadrant to select transmitters, as-
sign transmitters to tags, and eventually trigger simultaneous
transmissions. Thus, the intra-quadrant interference problem is
solved. The number of simultaneous transmissions is defined
by the amount of interference. Simultaneous transmissions
are permitted only if the sum of interference is below the
threshold.
Nonetheless, since quadrants are contiguous to each other
as shown in Fig. 3, chances are high that inter-quadrant
interference will occur. The occurrence of inter-quadrant inter-
ference is harmful to both quadrants and can drastically impact
their transmissions, resulting in transmission failures and/or
packet drops. In the worst case, it may reduce the number
of simultaneous transmissions or even annihilate simultaneous
transmissions, which in turn eliminates the effect of QuadScat-
ter and its advantage. To remedy such an extreme drawback,
we propose to assign different channels to adjacent quadrants
as depicted in the following subsection.
A. Quadrant Channel Assignment
This subsection introduces a channel assignment algorithm
that solves the inter-quadrant interference issue that might be
encountered with QuadScatter. Therefore, a node located in a
particular quadrant might suffer from the interferences from
nodes located in its neighboring quadrants. This issue can
be solved by alternately assigning different channels to the
quadrants. Considering the industrial, scientific, and medical
(ISM) 2.4GHz band and its 3[28] orthogonal channels, the
channels are assigned so that a quadrant and its immediate
neighbors - that might interfere with each other - use the three
orthogonal channels- Fig. 4. In the case that the ISM 5GHz
band is applicable, it becomes easier as the 5GHz has more
orthogonal channels. Clusters of quadrants will be created
with a maximum cluster size equal to the available number
of orthogonal channels in the 5GHz. Special attention should
be provided to quadrants on either side of a cluster border to
ensure that the same channels are not assigned to them- Fig.
5.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
60 36 108
56 104
48
108
48
56
40
52
40
36
64
100
52
104
64 44
60
100
44
Cluster 1
Cluster 2
Fig. 5: Quadrant channel assignment in ISM 5 GHz band.
B. Channel Assignment Algorithm
We describe the channel assignment algorithm in the sub-
sequent lines. The algorithm allows the usage of either the
2.4GHz or the 5GHz frequency bands. For the 2.4GHz band,
only the first 3non-overlapping channels are considered. The
first line of Algorithm 5 checks the decided ISM frequency
band. When the chosen frequency band is 2.4GHz (B1),
the quadrants are organized into columns and rows. The first
quadrant is assigned to the first channel, while the quadrant
immediately on its left assumes the second channel, and the
quadrant immediately on its right assumes the third channel,
lines 2to 4of the algorithm. When the target frequency
band is 5GHz (B2), all the quadrants in the area have to
be organized into clusters as depicted in Fig. 4. In line 7, the
algorithm creates clusters according to the number of usable
non-overlapping channels in (B2). The cluster size (number of
elements in the cluster) is equivalent to the number of channels
selected from (B2). Therefore, a channel is used only once in
each cluster 5, Algorithm 5- lines 8to 10. Note that in Figs.
4 and 5, the displayed channel numbers are the real numbers
according to the international regulations.
Algorithm 5: ASS IG N_C H assigns channels to quad-
rants
Input: A finite set Qof quadrants
Output: Assigned channels to quadrants in 2.4GHz
band or 5GHz band
1if B̸=B2then
2for qi,j in Qdo
3qi,j ←B1ch1
4qi+1,j ←B1ch2
5qi,j+1 ←B1ch3
6else
7create (Ca set of lclusters of size n)
8for k←1to ldo
9for j←1to ndo
10 Ck,j ←B2chj
11 return 0
Fig. 6: Simultaneous transmissions and formulation of the
distances: The distance is evaluated following the path of the
signal from the transmitter to the receiver
V. MATHE MATI CA L ANALYSI S
Eq. 4 provides the SIR in the theoretical analysis of the
capacity of backscatter communication as presented in [21].
Even though this formulation is the appropriate one for the
current situation, it might become arduous to manipulate
when dealing with a large number of devices, owing to the
product of powered distances dβ
(x,y)and dβ
(y,z)and the sum
idβ
( ˙xi,˙yi)dβ
( ˙yi,z). For simplicity and to reduce the heaver of
mathematical manipulations, we consider a slightly different
formulation: as shown in Eq. 4, the factor τinduced by the
backscatter disappeared as if the backscatter itself was no more
involved in the transmission. Thus, we assume the distance
from the source xto the reader zto be a straight line instead
of considering distances from source to tag and from tag
to reader. Therefore, we obtain a simple and mathematically
convenient method to manipulate this version of the SIR.
SI Rz|n=d(x,y)+d(y ,z)−β
id( ˙xi,˙yi)+d( ˙yi,z )−β.(5)
A. Simultaneous Transmissions
According to Eq. 5, the SIR of 4 simultaneous backscatter
communications, as illustrated in Fig. 6, is as follows: This
considers one source as a transmitter and follows the distances
derived in the same figure.
SI Rz|4=d−β
2(ρd +d1 + ρ2)−β+ (2ρd +d)−β.(6)
Since the propagation constant β= 2, we obtain the following
simplified version of Eq. 6
SI Rz|4=1
1
(1+2ρ)2+2
(√1+ρ2+ρ)2
.(7)
While looking for the upper bound of SI Rz|4we obtain the
following. The details are provided in the appendix section of
this paper.
SI Rz|4<8ρ(3 + 2√2)
16ρ+ (3 + 2√2),(8)
with ρz|4>3+2√2
8√2−3≈0.7945.
For five simultaneous transmissions, under similar condi-
tions as explained above, the SIR appears to be:
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
Fig. 7: Average number of simultaneous transmissions in
random 30mx30m area with 6 APs and 12 APs. Exhaustive
search was unable to go beyond 3 and 9 tags for respectively
12 and 6 APs.
Fig. 8: Computational time of exhaustive search and Quad-
Scatter in random 30mx30m area with 6 APs and 12 APs.
SI Rz|5=1
2
1−ρ√5−1
2+ρ2+ρ2+2
1+ρ√10−2√5
2+ρ2+ρ2
.
(9)
For clarity of SI Rz|5in Eq. 5, we compute its upper bound:
SI Rz|5<(ρ+√2)2(4 + 2√3)
8 + 2√2 + 2(ρ+√2)2.(10)
It yields ρz|5>6+8√3
3−√2≈1.158 for a minimum SIR
of 3dB, which clearly suggests the limitation of simultaneous
backscatter transmissions.
Note that we use an upper bound estimation ρ. Ideally,
the real values are supposed to be inferior to the provided
values. Even though ρz|5appears to be greater than 1, at least
five simultaneous communications can be considered possible,
which is verified by the simulation results of QuadScatter in
section V.
VI. PE RF OR MA NC E EVALUATI ON
A. Simulation Results
Here, we evaluate the results of our proposed method. We
compare the number of simultaneous transmissions and the
computational time of our algorithm named QuadScatter to
those of the exhaustive search or brute force algorithm. We
simulated a network of randomly deployed tags and AP nodes
on a 30mx30m area. The data collected are the average values
from one hundred different random topologies; meaning that
the program is executed one hundred times and at each time a
different random topology of APs and tags is created. While
the number of APs remains unchanged, the number of tags is
incremented by one after one hundred trials until the desired
number of tags is attained. We conducted simulations in three
cases: case 1with only 6APs, case 2with 12 APs, and
case 3with an increasing number of APs. The program is
required to provide the execution time in each case to measure
the computational performance. 6is the minimum number of
APs that provided 5simultaneous transmissions following our
numerical study in section V-A. The simulation parameters
are set as follows: Transmit power = 20dBm equivalent to
100mW, sensitivity =−94dB, noise floor =−97dB, SNR
threshold = 3dB, and path loss exponent β= 2.
Concerning the results, we found that exhaustive search
reaches a maximum of 3simultaneous transmissions whereas
QuadScatter outputs a maximum of 7simultaneous transmis-
sions, in some occurrences, outperforming exhaustive search
and confirming our theoretical analysis. Hence, if the suitable
topology is found with QuadScatter, at least twice the per-
formance of exhaustive search can be attained. Additionally,
note that the values shown by Figs. 7 and 9 are average
values of 100 trials, which is more realistic for a random
deployment scenario. Fig. 7 presents a better performance of
exhaustive search over QuadScatter for a reduced number of
nodes (<20). However, it failed to produce any results for
more than 3tags with 12 APs, and for more than 9tags with
6APs (Fig. 7). As illustrated in Fig. 8, the computational
cost of exhaustive search does not comply with a large-
scale network. When the network size grows, QuadScatter
renders an agreeable performance in terms of simultaneous
transmissions while maintaining an ultra-low computational
cost. In Figs. 9 and 10 QuadScatter shows its ability to handle
a network of a hundred nodes, delivering approximately 5
simultaneous transmissions with a computational time lower
than 0.01 sec, whereas exhaustive search fails to produce
any result in a reasonable amount of time. The amount of
time needed by exhaustive search scales to hours or days,
in this case, making it impossible to be displayed in the
same graph for comparison. QuadScatter is suitable to a
network of numerous nodes while exhaustive search deals
only with a network of a restrained number of nodes without
a time constraint. Naturally, the exhaustive search does not
exceed 10 backscatter tags and its computational time extends
considerably with an increased number of APs. Consequently,
we deliberately restrained the number of tags for exhaustive
search, in all the cases, to avoid running the experiment for
several days.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
Fig. 9: Average number of simultaneous transmissions in
random 30mx30m area with 30 tags. Exhaustive search shows
no output in reasonable time.
Fig. 10: Computational time of QuadScatter in random
30mx30m area with 30 tags. Exhaustive search shows no
output in comparable time.
In summary, 3 major improvement points are introduced
by QuadScatter compared to the exhaustive search: An in-
creased number of simultaneous backscatter transmissions and
backscatter nodes in a controlled network, and significantly
faster computational time; QuadScatter is at least 250 times
faster than exhaustive search.
B. Emulation of two simultaneous backscatter transmissions
This section describes an emulation of simultaneous trans-
missions in a reduced size. We conducted only an emulation of
two-simultaneous transmissions due to time constraints. The
design and conception of a backscatter modulator require a
considerable amount of time. The goal of this experiment is
to show the practicability of simultaneous transmissions in
the real world. The backscatter modulator is composed of 4
circuits as illustrated in fig. 13a. The conceived backscatter
modulator is shown in Fig. 13b.
1) Emulated experimental environment
The emulated network consists primarily of one transmit-
ter(Tx), two receivers(Rx), and two backscatter modulators
(Mod1 and Mod2) as shown in Fig. 11. Five attenuators
are used to reproduce the signal attenuation due to path
loss according to the connections between the devices. It is
worth mentioning that in this case the signals are transmitted
over cables rather than wireless transmissions as illustrated in
Fig. 12. We consider and environment with no obstacles that
would create reflections and no interference between wireless
channels. The cables associated to attenuators play the role
of wireless channels. Therefore, the emulated network can be
seen as a wireless network.
According to the configurations, Mod1 is supposed to trans-
mit its data to Rx1, and Mod2 is required to send its data to
Rx2. Attenuator ATT1 introduces some attenuation before the
transmitted signal reaches the backscatter modulators Mod1
and Mod2. ATT2 attenuates the signal between Mod1 and the
receiver Rx1 while ATT3 does the same between Mod2 and
receiver Rx2. ATT4 produces the path loss from Mod1 to Rx2
and ATT5 generates the path loss from Mod2 to Rx1.
The attenuation created by attenuators ATT4 and ATT5
varies from 1 dB to 11 dB during the experiment. ATT2 and
ATT3 are successively given the values 10 dB, 20 dB, and
30 dB while experimenting. ATT1 received a fixed value of
10 dB. The backscatter devices modulate 1000 times at each
attenuation pattern and we analyze the received packets at each
receiver.
2) Results of the emulated implementation
Fig. 14, Fig. 15 and Fig. 16 render the results of two
backscatter simultaneous transmissions. As already mentioned
above, these results take into account the signal attenuation
between the transmitter and the backscatter tags (ATT1), and
between the tags and the receivers (ATT2 and ATT3). They
also include the interferences from one backscatter modulator
to the other (ATT4 and ATT5).
The backscatter devices deliver a significant amount of
packets to their intended receivers when the attenuation to
their signals is kept at a low level and the attenuation to the
interfering signals is kept at a high level- Fig. 14. When the
direct attenuation to the modulator’s signal increases from 10
dB to 20 dB, the receiver receives the signal from the other
backscatter when the attenuation to the interference is very
low, less than 4 dB in Fig. 15a and less than 6 dB in Fig. 15b.
The receiver starts to detect the intended signal only when the
attenuation to the interference is increased to more than 5 dB
in Fig. 15a and more than 6 dB in Fig. 15b.
Likewise, Fig. 16 confirms the trends in Fig. 14 and Fig.
15. When the immediate attenuation to the modulators’ signals
becomes important (30 dB), only the interfering signals arrive
at the receivers. however, the interfering signals are important
only when they are not attenuated or when their attenuation
is very low.
The most important point to retain from this experiment
is that simultaneous transmissions are practically possible in
the real world. Although some important parameters are to be
considered during the deployment of the backscatter network,
it is safe to say that simultaneous backscatter transmissions
can be implemented in the real world. The results collected
and analyzed in this implementation provide the following
information. First, the receiver needs necessarily to be in
the transmission range of the modulator to reduce signal
attenuation due to path loss. Second, the backscatter devices
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
Fig. 11: Configuration of the emulation
Fig. 12: Emulated experimental testbed
need to be conveniently distanced to each other to weaken as
much as possible the reciprocal interferences.
VII. CON CL US IO N
In this study, we analyzed the computational cost and
simultaneous transmissions in a large scale backscatter net-
work. Computational cost and simultaneous transmissions are
important factors to consider in increasing a network capacity.
Simultaneous transmissions reduce the usage of network re-
sources such as energy and time, and improve the throughput
owing to the simultaneousness of several transmissions. Net-
work latency is also reduced as the transmission time of the
whole network is reduced. Indeed, simultaneous transmissions
imply no waiting time before transmissions. The performance
of the whole network is attributable to the results of the
algorithms developed in this study. Additionally, the results
exhibit significant improvements compared to the exhaustive
search algorithm in terms of computational time. A real world
implementation of backscatter simultaneous transmissions was
included to prove its feasibility. Future work will focus on
increasing the number of simultaneous transmissions and
exploring its limits. Another future project is to envision a
3-D scenario for QuadScatter.
APP EN DI X
Upper bound proof of ρz|4and ρz|5in section IV.
0<ρ<1gives 1+4ρ < (1 + 2ρ)2<8ρwhich gives
1
8ρ<1
(1+2ρ)2.0< ρ < 1also gives 1<(1 + ρ2+ρ)2<
(√2 + ρ)2, which gives 2
(√2+ρ)2<2
(√1+ρ2+ρ)2and 2
3+2√2<
2
(√2+ρ)2. By adding 2
3+2√2to 1
8ρwe obtain the lower bound
of 1
ρz|4=1
(1+2ρ)2+2
(√1+ρ2+ρ)2. Then by taking the inverse
of that sum, we obtain the upper bound 8ρ(3+2√2)
16ρ+(3+2√2) of ρz|4.
The upper bound of ρz|5is computed in the same manner.
REF ER EN CE S
[1] O. Zeba, S. Saruwatari, and T. Watanabe, “QuadScatter for Simultaneous
Transmissions in a Large-Scale Backscatter Network,” in ICC 2020-2020
IEEE International Conference on Communications (ICC). IEEE, 2020,
pp. 1–6.
[2] C. Perera, C. H. Liu, S. Jayawardena, and M. Chen, “A Survey on
Internet of Things from Industrial Market Perspective,” IEEE Access,
vol. 2, pp. 1660–1679, 2014.
[3] F. Guidi, N. Decarli, S. Bartoletti, A. Conti, and D. Dardari, “Detection
of Multiple Tags Based on Impulsive Backscattered Signals,” IEEE
Transactions on Communications, vol. 62, no. 11, pp. 3918–3930, 2014.
[4] R. Eletreby, D. Zhang, S. Kumar, and O. Ya˘
gan, “Empowering Low-
Power Wide Area Networks in Urban Settings,” in Proceedings of the
Conference of the ACM Special Interest Group on Data Communication.
ACM, 2017, pp. 309–321.
[5] B. Reynders and S. Pollin, “Chirp Spread Spectrum as a Modulation
Technique for Long Range Communication,” in 2016 Symposium on
Communications and Vehicular Technologies (SCVT). IEEE, 2016, pp.
1–5.
[6] W. Zhao, G. Wang, R. Fan, L.-S. Fan, and S. Atapattu, “Ambient
Backscatter Communication Systems: Capacity and Outage Performance
Analysis,” IEEE Access, vol. 6, pp. 22695–22 704, 2018.
[7] D. Li, W. Peng, and Y.-C. Liang, “Hybrid Ambient Backscatter Commu-
nication Systems With Harvest-Then-Transmit Protocols,” IEEE Access,
vol. 6, pp. 45 288–45 298, 2018.
[8] G. Yang, D. Yuan, Y.-C. Liang, R. Zhang, and V. C. Leung, “Optimal Re-
source Allocation in Full-Duplex Ambient Backscatter Communication
Networks for Wireless-Powered IoT,” IEEE Internet of Things Journal,
vol. 6, no. 2, pp. 2612–2625, 2018.
[9] Y. Ye, L. Shi, X. Chu, and G. Lu, “On the outage performance of ambient
backscatter communications,” IEEE Internet of Things Journal, 2020.
[10] H. Yang, Y. Ye, and X. Chu, “Max-min energy-efficient resource
allocation for wireless powered backscatter networks,” IEEE Wireless
Communications Letters, vol. 9, no. 5, pp. 688–692, 2020.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
0&38
670I5(
5)6ZLWFK
&*[
60$
&RQQHFWRU
3LQ
&RQQHFWRU
(a) Architecture of backscatter device.
(b) Backscatter device.
Fig. 13: Backscatter device used for the experiment.
(a) Packets received at Rx1 (b) Packets received at Rx2
Fig. 14: Packets received by each receiver when the attenuation from ATT2 and ATT3 is 10 dB
(a) Packets received at Rx1 (b) Packets received at Rx2
Fig. 15: Packets received by each receiver when the attenuation from ATT2 and ATT3 is 20 dB
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
(a) Packets received at Rx1 (b) Packets received at Rx2
Fig. 16: Packets received by each receiver when the attenuation from ATT2 and ATT3 is 30 dB
[11] B. Lyu, Z. Yang, H. Guo, F. Tian, and G. Gui, “Relay Cooperation
Enhanced Backscatter Communication for Internet-of-Things,” IEEE
Internet of Things Journal, vol. 6, no. 2, pp. 2860–2871, 2018.
[12] K. Han and K. Huang, “Wirelessly Powered Backscatter Communication
Networks: Modeling, Coverage, and Capacity,” IEEE Transactions on
Wireless Communications, vol. 16, no. 4, pp. 2548–2561, 2017.
[13] C. Yang, X. Wang, and K.-W. Chin, “On max–min throughput in
backscatter-assisted wirelessly powered iot,” IEEE Internet of Things
Journal, vol. 7, no. 1, pp. 137–147, 2019.
[14] N. Van Huynh, D. T. Hoang, D. Niyato, P. Wang, and D. I. Kim, “Op-
timal time scheduling for wireless-powered backscatter communication
networks,” IEEE Wireless Communications Letters, vol. 7, no. 5, pp.
820–823, 2018.
[15] M. Hessar, A. Najafi, and S. Gollakota, “NetScatter: Enabling Large-
Scale Backscatter Networks.” in NSDI, 2019, pp. 271–284.
[16] N. Mi, X. Zhang, X. He, J. Xiong, M. Xiao, X.-Y. Li, and P. Yang,
“Cbma: Coded-backscatter multiple access,” in 2019 IEEE 39th Interna-
tional Conference on Distributed Computing Systems (ICDCS). IEEE,
2019, pp. 799–809.
[17] C. Xu, L. Yang, and P. Zhang, “Practical Backscatter Communication
Systems for Battery-Free Internet of Things: A Tutorial and Survey of
Recent Research,” IEEE Signal Processing Magazine, vol. 35, no. 5, pp.
16–27, 2018.
[18] Z. Ma, L. Feng, and F. Xu, “Design and Analysis of a Distributed
and Demand-Based Backscatter MAC Protocol for Internet of Things
Networks,” IEEE Internet of Things Journal, vol. 6, no. 1, pp. 1246–
1256, 2018.
[19] C. E. Shannon, “A mathematical theory of communication,” Bell system
technical journal, vol. 27, no. 3, pp. 379–423, 1948.
[20] T. M. Cover and J. A. Thomas, Elements of information theory. John
Wiley & Sons, 2012.
[21] A. V. Padaki and M. J. Zawodniok, “Theoretical Capacity Analysis for
Multi-hop Backscatter Communication Networks,” in 2011 Proceedings
of 20th International Conference on Computer Communications and
Networks (ICCCN). IEEE, 2011, pp. 1–6.
[22] W. Liu, K. Huang, X. Zhou, and S. Durrani, “Full-duplex Backscatter
Interference Networks Based on Time-hopping Spread Spectrum,” IEEE
Transactions on Wireless Communications, vol. 16, no. 7, pp. 4361–
4377, 2017.
[23] M. R. Souryal, D. R. Novotny, D. G. Kuester, J. R. Guerrieri, and K. A.
Remley, “Impact of RF Interference between a Passive RFID System
and a Frequency Hopping Communications System in the 900 MHz ISM
Band,” IEEE Electromagnetic Compatibility Magazine, vol. 1, no. 3, pp.
97–102, 2012.
[24] A. Varshney, O. Harms, C. Pérez-Penichet, C. Rohner, F. Hermans, and
T. Voigt, “LoRea: A Backscatter Architecture that Achieves a Long
Communication Range,” in Proceedings of the 15th ACM Conference
on Embedded Network Sensor Systems. ACM, 2017, p. 18.
[25] Wikipedia contributors, “Locality-sensitive hashing — Wikipedia,
The Free Encyclopedia,” https://en.wikipedia.org/w/index.php?title=
Locality-sensitive_hashing&oldid=905340247, 2019, [Online; accessed
16-August-2019].
[26] Avinash Kak, “LocalitySensitiveHashing — LocalitySensitiveHash-
ing.py version: 1.0.1,” https://engineering.purdue.edu/kak/distLSH/
LocalitySensitiveHashing-1.0.1.html, 2017-May-25, [Online; accessed
8-October-2019].
[27] Wikipedia contributors, “Quadtree — Wikipedia, The Free Ency-
clopedia,” https://en.wikipedia.org/w/index.php?title=Quadtree&oldid=
910044683, 2019, [Online; accessed 16-August-2019].
[28] ——, “List of wlan channels — Wikipedia, the free encyclopedia,”
https://en.wikipedia.org/w/index.php?title=List_of_WLAN_channels&
oldid=965323971, 2020, [Online; accessed 5-July-2020].
Ousmane Zeba received a Diplôme d’Ingénieur in
Electronics from the University of Tlemcen, Algeria.
He worked as an IT technician and has experi-
ence in network design and deployment. He has
also been involved in GPS vehicle tracking system
management. He joined Osaka University in 2016
as a research student and received his M.S. degree
in 2019. He is currently pursuing a Ph.D. degree
with the Department of Information Networking at
the Graduate School of Information Science and
Technology, Osaka University. His research interests
include wireless mesh networks, cognitive radio networks, IoT, and M2M
communications.
Kentaro Hayashi received the B.E. degree from
Osaka University, Osaka, Japan, in 2020, where
he is currently pursuing the M.S. degree with the
Department of Information Networking, Graduate
School of Information Science and Technology. He
joined the Information Processing Society of Japan
in 2018. His research interests include microwave
power transfer.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/OJCS.2021.3104986, IEEE Open
Journal of the Computer Society
Kazuhiro Kizaki is a Project Research Associate
with the Graduate School of Information Science
and Technology, Osaka University, Japan. He joined
Communication Equipment Works, Mitsubishi Elec-
tric Corporation, in 1970. He specializes in RF
circuits and antennas.
Shunsuke Saruwatari received a B.E. degree from
the University of Electro-Communications, Japan,
in 2002, and M.S. and Ph.D. from the University
of Tokyo, Japan, in 2004 and 2007, respectively.
He is an Associate Professor with the Graduate
School of Information Science and Technology, Os-
aka University, Japan. In 2007, he was a visiting
researcher with the Illinois Genetic Algorithms Lab-
oratory, University of Illinois at Urbana-Champaign.
From 2008 to 2011, he was a research associate
with the Research Center for Advanced Science and
Technology, University of Tokyo, Japan. From 2012 to 2015, he worked as an
assistant professor at the Graduate School of Informatics, Shizuoka University,
Japan. His research interests are in the areas of wireless networks, sensor
networks, and system software. He is a member of ACM, IPSJ, and IEICE.
Takashi Watanabe received a B.E., M.E., and Ph.D.
from Osaka University, Japan, in 1982, 1984, and
1987, respectively. He has been a professor with
the Graduate School of Information Science and
Technology, Osaka University, Japan, since 2013.
He joined the Faculty of Engineering at Tokushima
University in 1987 and moved to the Faculty of
Engineering at Shizuoka University in 1990. He
was a visiting researcher with the University of
California, Irvine, CA, USA, from 1995 to 1996.
He has served on many program committees for
networking conferences, IEEE, ACM, IPSJ, and IEICE. His research interests
include mobile networking, ad hoc sensor networks, IoT/M2M networks, and
intelligent transport systems, especially MAC and routing. He is a member
of IPSJ and IEICE.
