Aalborg Universitet
Experimental validation of a distributed algorithm for dynamic spectrum access in
local area networks
Tonelli, Oscar; Berardinelli, Gilberto; Tavares, Fernando Menezes Leitão; Cattoni, Andrea
Fabio; Kovacs, Istvan; Sørensen, Troels Bundgaard; Popovski, Petar; Mogensen, Preben
Published in:
IEEE VTS Vehicular Technology Conference. Proceedings
DOI (link to publication from Publisher):
10.1109/VTCSpring.2013.6692557
Publication date:
2013
Document Version
Early version, also known as pre-print
Link to publication from Aalborg University
Citation for published version (APA):
Tonelli, O., Berardinelli, G., Tavares, F. M. L., Cattoni, A. F., Kovacs, I., Sørensen, T. B., Popovski, P., &
Mogensen, P. (2013). Experimental validation of a distributed algorithm for dynamic spectrum access in local
area networks. In IEEE VTS Vehicular Technology Conference. Proceedings (pp. 1-5). IEEE.
https://doi.org/10.1109/VTCSpring.2013.6692557
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
- Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
- You may not further distribute the material or use it for any profit-making activity or commercial gain
- You may freely distribute the URL identifying the publication in the public portal -
Take down policy
If you believe that this document breaches copyright please contact us at vbn@aub.aau.dk providing details, and we will remove access to
the work immediately and investigate your claim.
Experimental validation of a distributed algorithm for
dynamic spectrum access in local area networks
Oscar Tonelli*, Gilberto Berardinelli*, Fernando M. L. Tavares*, Andrea F. Cattoni*,
István Z. Kovács§, Troels B. Sørensen*, Petar Popovski* and Preben E. Mogensen*§
* Department of Electronic Systems, Aalborg University, § Nokia Siemens Networks, Aalborg, Denmark
ot@es.aau.dk
Abstract—Next generation wireless networks aim at a significant
improvement of the spectral efficiency in order to meet the
dramatic increase in data service demand. In local area scenarios
user-deployed base stations are expected to take place, thus
making the centralized planning of frequency resources among
the cells, a non-viable solution. Cognitive Radio (CR) and
Dynamic Spectrum Access (DSA) are the research paradigms
which are expected to provide the network nodes the capabilities
for an autonomous and efficient selection of the spectrum
resources. In this paper we present the first experimental
activities with the Autonomous Component Carrier Selection
(ACCS) algorithm, a distributed solution for interference
management among small neighboring cells. A preliminary
evaluation of the algorithm performance is provided considering
its live execution on a software defined radio network testbed.
The obtained experimental results confirm the performance
trends obtained from prior simulation studies. The analysis in
dynamic environment conditions also allowed identifying the
utilization of static thresholds in the decision making process, as
a critical aspect for the optimization of network capacity.
I. INTRODUCTION
Future generation mobile wireless networks are expected to
cope with the dramatic increase of data services demand. In
particular, very high data rate in local-area (LA) is a
challenging requirement when considering the scarcity of
available spectrum resources in licensed bands. The network
setup in indoor home/office environments is expected to be
characterized by dense and uncoordinated deployment of
access points (APs), thus making the inter-cell interference the
major limiting factor for boosting the network performance.
Dynamic Spectrum Access (DSA) and the broader concept of
Cognitive Radio (CR) [1] are research paradigms which are
expected to provide the required spectrum flexibility to cope
with the unplanned usage of frequency resources.
The development of CR/DSA concepts has been characterized
by extensive theoretical work; network distributed algorithms
in particular, have been typically validated by using Monte
Carlo system level simulations. Despite the extensive
simulation results, both the academia and the industry have
shown a growing interest for experimental studies providing
more tangible evidence of the algorithms effectiveness. In this
sense, the implementation of large network testbeds enables
the investigation of environmental properties which are
difficult to comprehensively model in simulators. Examples of
such aspects are the impact of dynamic environment and
terminals mobility. An experimental testbed enables also the
investigation of real-time execution issues as well as practical
system limitations introduced by the hardware inaccuracies. A
considerable amount of work has been presented in literature
in relation to the development of platforms and testbeds for
CR [2]. The majority of the conducted experimental activities
with novel CR concepts, relates to advanced spectrum sensing
techniques, physical (PHY) layer design and DSA solutions
for opportunistic spectrum access. Nevertheless, algorithms
for interference coordination among nodes have, so far,
received little attention. A more extensive coverage of recent
experimental activities with CR/DSA platforms can be found
in [3].
In this paper we select the Autonomous Component Carrier
Selection (ACCS) algorithm [4] for the implementation and
experimental analysis on a network testbed. ACCS is a DSA
algorithm for inter-cell interference management, which is
characterized by the distributed execution of the decision
making process over the APs aided by an inter-node control
channel for cells coordination. The performed experimental
activities intend to evaluate the algorithm performance with
specific attention to the impact of variable cardinality of the
available set of resources, user terminals position and dynamic
channel conditions.
The paper is organized as follows. An overview of the ACCS
algorithm is given in Section II. A description of the testbed
architecture and the implemented system solutions are
presented in Section III. Section IV describes in detail the
experimental activities and the obtained results from the
algorithm execution. Future developments of the experimental
work and conclusions are reported in Section V.
II.
AUTONOMOUS COMPONENT CARRIER SELECTION
ALGORITHM
ACCS is a DSA algorithm which targets the problem of
spectrum resource management among femtocells in indoor
scenarios. The interest in ACCS relates to its lightweight
distributed decision making process which enables a flexible
reuse of the frequency resources in the network, thus
minimizing the mutual interference and improving channel
capacity [4]. ACCS assumes the available spectrum to be
divided in a number of Component Carriers (CCs) which are
shared among the APs, named evolved NodeBs (eNBs),
according to the 3
rd
Generation Partnership Project (3GPP)
terminology. The DSA decision making process occurs locally
on the eNBs, but relies on shared information about usage of
CCs and inter-cell interference coupling. These features allow
the algorithm to be executed on devices with limited channel
sensing capabilities, at the expenses of a common control
channel in the network. The required sensing information used
by ACCS is limited to the Reference Signal Received Power
(RSRP) measurements in respect to neighboring eNBs. The
ACCS decision process and the control data exchange are
executed periodically on a time-frame basis. The ACCS
framing is supposed to be rather slow compared to baseline
Radio Resource Management (RRM) techniques (e.g.,
time/frequency domain packet scheduling). Further
information about ACCS can be found in [5] and [6]. In the
following subsections, brief descriptions of the main ACCS
procedures implemented in the testbed are provided.
A. Component Carriers selection procedures
In ACCS the available CCs are divided between a Base CC
(BCC), which is intended to be the main communication
carrier between the eNB and the UEs, and a set of
Supplementary CCs (SCCs) providing additional channel
capacity. ACCS aims at ensuring the highest reliability of the
BCC (a single BCC is always allocated in the cell) especially
in conditions of high interference. ACCS also dynamically
enables the allocation of the SCCs in order to meet the cells
data traffic demands. In relation to the BCC and SCCs
selection procedures, the underlying principle of ACCS
consists in enabling the reuse of frequencies in the network
under the condition that no intolerable interference is
generated by an eNB towards a neighboring cell. The
conservative approach in spectrum allocation, enforced by
ACCS, aims at the minimization of the network outage
probability.
B. The Background Interference Matrix
In order to acquire general knowledge about interference
coupling with the neighbors, eNBs can exchange spectrum
allocation information and a background interference matrix
(BIM) over the control channel. The BIM is a data structure
which contains information about inter-cell interference
coupling, in the form of estimations of Carrier to Interference
(C/I) power ratio, i.e. the ratio between the useful and the
interference power which would occur in case two nodes share
the same CC. C/I estimations are computed from RSRP
measurements at the UE side. In ACCS a certain SCC can be
allocated by a node only in case the estimated C/I value, with
respect to all the cells occupying the same CC, is above a
predefined threshold. Different thresholds are defined for BCC
and SCCs. The BCC threshold is typically more restrictive.
SCCs can be de-allocated in case of decreased traffic demand
or unsatisfactory channel quality experienced in the cell.
III. ACCS
TESTBED SETUP
A network testbed has been designed in order to enable the
runtime execution and support the experimental activities with
ACCS. The developed testbed relies on software defined radio
(SDR) equipment featuring the Ettus USRP N200 hardware
[7] and host computers equipped with the ASGARD software
platform [8]. The radio-frequency front-end consists in the
Ettus XCVR2450 daughterboards, able to operate in the 2.4
and 5 GHz bands. The equipment setup is shown in Figure 1.
A general overview of the ACCS testbed architecture is given
in Figure 2.
The ACCS testbed features two types of nodes:
• eNB. The eNB is responsible for the ACCS algorithm
execution: the ACCS decision engine dynamically selects
the BCC and periodically triggers the reconfiguration of
the used SCCs. The generated ACCS control data is sent
broadcast over the control channel.
• UE. The UE periodically provides the affiliated eNB with
sensing information about the useful signal power as well
as from the interfering eNBs.
The developed PHY layer for both the eNB and UE enables
the RSRP measurements from multiple nodes. Orthogonal
frequency pilot sequences across the allocated CCs have been
univocally associated to the eNBs. Pilots are generated in the
frequency domain and then converted to time domain through
Inverse Fast Fourier Transform (IFFT).
Frequency spacing between the pilots and power sensing at the
receiver have been designed for being robust to the non-
idealities of the USRP boards such as frequency offset and
phase noise. All the USRP boards in the testbed have been
calibrated with the aim of aligning transmit power and
effective receiver gain within 1 dB, thus ensuring consistent
RSRP measurements among multiple nodes. In our current
testbed, the PHY layer is only used for signal power
measurement purposes, while the whole data exchange (e.g.,
spectrum allocation information and BIM) occurs over a
parallel backhaul network (based on Ethernet or WiFi).
The backhaul infrastructure, depicted in Figure 2, is used also
for experiment control and testbed data collection. The ACCS
control channel is emulated by a centralized unit that routes
the control data among the nodes. The feedback channel
connecting the UEs to the affiliated eNBs is also emulated and
enables the reporting of the downlink (DL) RSRP
measurements.
IV. ACCS
EXPERIMENTS
Prior simulation studies [6] allowed characterizing the
performance of a network running the ACCS algorithm. In
particular, improved outage channel capacity in respect to
Reuse 1 (i.e. each node is transmitting over the whole
bandwidth corresponding to the number of configured CCs) is
achieved in case of heavy data traffic conditions.
In this work, a set of experimental trials have been conducted
in order verify these findings in a realistic deployment
scenario. Moreover, experiments have been also performed in
Figure 1 - Testbed node hardware setup: USRP board and host PC
order to evaluate the impact of UE terminal positioning and
channel dynamics. According to the indoor deployment
assumption of ACCS, the testbed nodes have been placed
inside the office premises of Aalborg University. The
environment is characterized by several office rooms on the
same building floor, arranged in a double stripe fashion with a
corridor in between (see Figure 2).
An identical spatial
deployment of the nodes is considered for all the experiment
trials: the setup features 6 testbed nodes deployed across 3
cells obtained by considering 1 eNB and 1 UE per cell. The
cells are confined in 3 separate rooms. Two specific
configurations for cell 3 have been foreseen: in position a) the
eNB 2 is placed very close to the UE 3, thus generating strong
interference. Position b) instead, reduces this effect. Despite
the limited number of nodes, such deployments provide
challenging and diversified interference coupling
combinations for the ACCS performance evaluation.
All experiment trials share common system configuration
parameters which are here briefly described. The I/FFT size is
set to 1024, offering sufficient granularity for the spectrum
division up to 4 CCs and the frequency allocation of the
reference pilots across the same CC. A minimum spacing of
180 kHz between adjacent pilots is set in order to avoid power
leakage due to the USRPs frequency offset.
The algorithm execution iteration period (ACCS frame) is of
400 ms while the UE measurements reporting period is of 200
ms. The considered traffic model is full-buffer: once the UE
connects to the cell, the eNB attempts to allocate the
maximum number of SCCs allowed according to the channel
configuration.
The ACCS algorithm is executed on the testbed in real-time,
thus generating time data traces of the eNBs control data and
UEs RSRP measurements. These experimental results allow
generating network-wide statistics about downlink SINR
experienced in the cells, and computing the corresponding
estimated capacity which is obtained through Shannon
mapping [9]. It is to be noted that the Signal to Interference
and Noise Ratio (SINR) is first measured on the narrowband
pilots, and then scaled to the effective emulated bandwidth of
the used CC configuration. The total system radio bandwidth
is of 12.5 MHz, while the effective bandwidth occupied by the
transmitted pilots and also the emulated bandwidth for
capacity estimation, is of 10 MHz. Transmission power per
CC is 0dBm.
A. Static environment algorithm analysis
The goal of the first experiment (Experiment 1) is to verify the
behavior of the network in a static environment scenario. The
impact of a different number of CCs on the network
performance is investigated in particular. In order to meet the
static environment assumption, the experiments runs have
been executed during night hours. The experiment deployment
considers the UE 3 placed in position a according to Figure 3.
Iterations with 2, 3 and 4 CCs have been performed. ACCS
performance is then compared with standard Reuse 1. A single
run of the experiment consists in the steps reported below:
•
eNBs are activated sequentially and a single BCC is
selected per cell.
•
UEs are activated sequentially within an interval of few
ACCS frames: the maximum number of SCCs allowed
by the algorithm is allocated in the cell as soon as the
UE connects, and before the following UE is activated.
Nodes activation sequences have a considerable impact on the
final CC allocation for the different cells especially among
highly-coupled cells. The full buffer traffic model favors
indeed a wider allocation of resources for the first cell
activated in time. Allocated SCCs are not released unless the
channel quality becomes unsatisfactory. In order to cover all
the possible combinations of eNB and UE activation
sequences over 3 cells, 36 experiment runs were designed.
Multipath fading variability is introduced by repeating the
experiment over 10 different carrier frequencies ranging from
4.91 to 5.81 GHz. The amount of experiment runs for each CC
configuration is therefore 360, for total of 1080 runs
considering the aforementioned 3 configuration cases.
The obtained results in terms of downlink Shannon channel
capacity are presented in Figure 4. Every point in the
Figure 2 - ACCS Testbed architecture. Connections between network nodes
and the testbed backhaul infrastructure are visible. All units in the testbed
can be time synchronized for the automatic execution of experimental runs.
Testbed Manager and Data Collection
Comm.
Client
ACCS
Decision
PHY
eNB 2
Control Data
Control
Channel
Emulator
Control Data
Comm.
Client
ACCS
Decision
PHY
eNB 1
DL RSRP Measurements
Feedback
Channel
Emulator
Network
Interface
PHY
UE 1
Nodes Control
Figure 3 - Experiments Deployment Scenario. UE in cell 3 is moved from
position a (Experiment 1) to position b (Experiment 2)
1
1
2
2
3
3
a) b)
eNB
UE
1
2
3
3
cumulative distribution functions (CDFs) corresponds to the
capacity estimated for a specific cell (eNB-UE link) at the end
of the experimental run, after all nodes have been activated
and SCCs selection has occurred. The curves show a general
gain of ACCS in comparison with Reuse 1. As expected,
ACCS performs better in the lower percentile of the CDF,
meaning in the case of highly interference-coupled cells.
Lower gain is instead achieved in the best cases. The obtained
results also show that moving from Reuse 1 to the 2 CCs
configuration is sufficient to orthogonalize the allocation of
frequency resources in most of the high interference cases,
thus obtaining the major gain contribution. Such behavior is
also coherent with the network topology where 2 cells (2 and
3) are strongly interference-coupled while one (cell 1) is more
isolated. Increasing the CCs cardinality provides a decreasing
gain margin, enabling however, further chances of fractional
frequency reuse and thus improving capacity.
B. UE position impact
The deployment scenario and the achieved network
performance in Experiment 1 are characterized by the high
interference from eNB 2 to UE 3. Different interference
conditions may impact the ACCS allocation of resources thus
providing significant variations in the available capacity in the
cells. A second experiment (Experiment 2) has been then
conducted in order to investigate such aspect. In respect to
Experiment 1, the UE 3 has been moved to the other side of
the room 3 (position b in Figure 2) in order to maximize the
pathloss with respect to eNB 2. The execution of Experiment 2
follows the same procedure as experiment 1. A single channel
configuration with 4 CCs is considered.
Statistics about CCs utilization obtained from the experiments
have been summarized in Table I and Table II. The values
reported are averaged over the entire experimental session
considering 360 runs. Data in the tables show that the
variation in the level of interference experienced in cell 3 from
Experiment 1 to Experiment 2 has a widespread impact on the
amount of resources allocated in the entire network. In the first
case, cell 2 and cell 3, being extremely interference-coupled,
trigger a perfectly orthogonal allocation of their resources by
ACCS. In experiment 2 instead, the more isolated position of
the UE diminishes the cell coupling thus enabling a better
utilization of the spectrum by all the cells. Opportunities for
frequency reuse among the cells are also increased, especially
looking at cell 1 which is the most isolated in the considered
scenario. According to the previous analysis, the results in
Figure 4 show an increase in cells’ capacity mostly affecting
the upper percentile. These specific values are related to the
performance of cell 1 which greatly benefits from the
increased average spectrum allocation. The lower percentile of
the CDF is almost non-affected by the UE movement. This
situation is due to the unmodified eNBs interference
contributions on the heavily interfered UE in cell 2, which
indeed is unable to take advantage from the new scenario
conditions.
C. Dynamic environment algorithm analysis
The effect of rapid variations in the experienced C/I values in
the cells, and SINR estimations over the CCs, can impact the
effectiveness of the ACCS allocation. A third experiment
conducted with the ACCS testbed aimed then at evaluating the
impact of a dynamic environment on the algorithm execution.
Human presence in the rooms has been then allowed. The
relation between the capacity estimation obtained in such
dynamic context, in comparison to the previous static-
environment studies has also been analyzed.
Experiment 3 features the same testbed setup as in experiment
1. A fixed channel configuration with 4 CCs, and fixed
frequency carrier setting at 5.41 GHz have been considered. In
order to acquire results in realistic scenario conditions, the
experiment has been executed during working hours.
The office rooms are characterized by different degrees of
human activity: cells 2 and 3 are on average, more crowded
than cell 1. An experimental run of 1 hour duration (3600 sec)
is considered for the analysis.
An overview of the experiment results is provided in Table III.
The table reports the time average of cell capacity values (X
�
),
together with the reference values obtained by the same
experimental scenario (carrier frequency and activation
sequences) in static environment conditions (from Experiment
1). Standard deviation (σ) is also included.
The obtained results confirm the average capacity, for cell 1,
comparing to the static environment case. The behavior of
Figure 4 – CDFs of the downlink Shannon Capacity in the cells
0 10 20 30 40 50 60 70 80
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Capacity [Mbps]
CDF
Reuse 1
ACCS - 2 CCs - exp 1
ACCS - 3 CCs - exp 1
ACCS - 4 CCs - exp 1
ACCS - 4 CCs - exp 2
TABLE I. EXPERIMENT 1 SPECTRUM USAGE OVERVIEW
Cell
Spectrum usage
(4CCs=100%)
Shared spectrum
resources (4CCs=100%)
Cell 1
Cell 2
Cell 3
1
84%
-
32%
20%
2
48%
32%
-
0%
3
48%
20%
0%
-
TABLE II. EXPERIMENT 2 SPECTRUM USAGE OVERVIEW
Cell
Spectrum usage
(4CCs=100%)
Shared spectrum
resources (4CCs=100%)
Cell 1
Cell 2
Cell 3
1
95%
-
45%
53%
2
49%
45%
-
5%
3
54%
53%
5%
-
