Opportunistic load and spectrum management for mobile communications energy efficiency

Conference Paper (PDF Available) · September 2011with43 Reads
DOI: 10.1109/PIMRC.2011.6140046 · Source: DBLP
Conference: IEEE 22nd International Symposium on Personal, Indoor and Mobile Radio Communications, PIMRC 2011, Toronto, ON, Canada, September 11-14, 2011
Abstract
Dynamic load and spectrum usage management techniques can significantly improve the energy efficiency of mobile communications systems. This paper considers: (i) the opportunistic reallocation of traffic loads between bands to allow radio network equipment in the bands that the traffic is originated from to be powered down, and (ii) the opportunistic selection of more appropriate spectrum based on propagation characteristics to minimize necessary transmission power through improving propagation and/or reducing power leaking into co-channel cells in frequency reuse cases. This paper addresses the simulation of video, FTP and HTTP (web browsing) traffic sources for configurations representing LTE and HSDPA telecommunications networks, and shows that the opportunistic reallocation of users between bands to power down radio equipment achieves a significant saving of 50% or more in from-the-socket power. Furthermore, it shows that the opportunistic reallocation of users/links to minimize transmission power through using more appropriate propagation spectrum leads to a further modest reduction in from-the-socket power consumption.
AbstractDynamic load and spectrum usage management
techniques can significantly improve the energy efficiency of
mobile communications systems. This paper considers: (i) the
opportunistic reallocation of traffic loads between bands to allow
radio network equipment in the bands that the traffic is
originated from to be powered down, and (ii) the opportunistic
selection of more appropriate spectrum based on propagation
characteristics to minimize necessary transmission power through
improving propagation and/or reducing power leaking into co-
channel cells in frequency reuse cases. This paper addresses the
simulation of video, FTP and HTTP (web browsing) traffic
sources for configurations representing LTE and HSDPA
telecommunications networks, and shows that the opportunistic
reallocation of users between bands to power down radio
equipment achieves a significant saving of 50% or more in from-
the-socket power. Furthermore, it shows that the opportunistic
reallocation of users/links to minimize transmission power
through using more appropriate propagation spectrum leads to a
further modest reduction in from-the-socket power consumption.
Index Termsgreen communications, load balancing,
spectrum management
I. INTRODUCTION
ISTORICALLY, the radio spectrum has been managed in
a rather rigid fashion where the primary objective has
been to minimize interference among users and systems to
maintain the spectrum’s viability. This regime is often
extremely inefficient, because at any one time many systems
are not being used thereby leaving the associated spectrum
also unused. Alternative spectrum management, whereby
systems not designated for a particular band may nevertheless
use it if it is available, would greatly increase spectral
efficiency and capacity. Fortuitously, at least from regulatory
and technical perspectives, various recent developments are
leading to the situation where such solutions will be realizable,
albeit initially to a limited extent (see, e.g., [1], [2], [3]).
It is widely expected that communications operators will
own or have access to an increasing range of spectrum bands
in the future, of very different frequencies and physical
characteristics. Already in many major cities around the world,
there are often operators concurrently providing services in
multiple bands, such as GSM 900MHz and 1800MHz, UMTS
2GHz, and 2.4GHz Wi-Fi services among others. Moreover,
the introduction of IMT-Advanced bands in the near future
will further increase the range of spectrum bands available to
the operator: there are already many such bands widely
identified, some examples being 450-470MHz, 790-862MHz,
and 2.3-2.4GHz [4]. The greater use of unlicensed spectrum,
such as UNII bands (5.15-5.825GHz), will also facilitate better
spectrum availability.
It is clear that significant reductions in the energy
consumption of humanity must be made to sustain the planet.
Moreover, the power consumption of mobile and wireless
communications systems can be quite significant. Even an
environmentally-aware operator, such as Vodafone, consumes
approximately 40MW in running its business in the UK, the
majority of which can be attributed to base stations [5].
Lowering the operational power consumptions of base stations
is therefore particularly important. Enabling sleep modes for
such equipment, and more generally reducing necessary
transmission power, can have a very significant effect on the
overall power consumption of the operator in running its
network. These are the objectives of the concepts discussed in
this paper, whereby given the prior observations on increased
spectrum usage freedom and availability, this is done by
opportunistically reallocating traffic loads between spectrum
bands. This reallocation might either be between bands owned
by a single operator, or ultimately could be through
spectrum/load sharing between operators. In either case, the
concepts and results discussed in this paper still apply.
This paper is structured as follows. In the next section, the
proposed power saving concepts are explained. Section III
investigates aspects of the performances of the concepts,
showing significant potential for power consumption
reduction. Finally, Section IV concludes this paper.
II. POWER-SAVING SPECTRUM MANAGEMENT CONCEPTS
The two concepts studied in this paper to reduce mobile
communications power consumption are introduced as follows.
A. Power Saving by Dynamically Powering Down Radio
Equipment
The first concept, illustrated in Figure 1, is the switching off
(or entering into stand-by) of radio equipment through
Opportunistic Load and Spectrum Management
for Mobile Communications Energy Efficiency
Oliver Holland
1
, Orlando Cabral
2
, Fernando Velez
2
, Adnan Aijaz
1
, Paul Pangalos
1
and A. Hamid Aghvami
1
1
Centre for Telecommunications Research
King’s College London
London WC2R 2LS, UK
{oliver.holland, adnan.aijaz, paul.pangalos,
hamid.aghvami}@kcl.ac.uk
2
Instituto de Telecomunicações, DEM
Universidade da Beira Interior
6201-001 Covilhã, Portugal
{fjv, orlandoc}@ubi.pt
H
reallocating load to other bands at times of low load. This is
extremely promising as it implies a guaranteed power saving
through radio equipment being virtually “switched off at the
socket”. It is noted that for macro-cell base stations in
particular, by far the biggest contribution to power
consumption of the base station is merely it being switched on
and in an operational state; variation of power consumption
with transmission power is relatively less significant although
depends greatly on the exact manufacture of the base station
hence can only be very broadly generalized. The opportunistic
powering down of radio network equipment based on solutions
presented in this paper is therefore a readily achievable way of
reducing actual from-the-socket power consumption.
There are two possibilities considered in this paper
concerning the dynamic powering down of radio equipment:
(i) turning off cells entirely in one network or spectrum band at
that time/location, through traffic being sufficiently carried by
a single network or spectrum band, and (ii) using spare
capacity of one network/band to cover the required drop in
load of another network/band in order to enable that other
network/band operate in omnidirectional mode instead of tri-
sectored mode. The power saving assessments of this concept
in this paper reallocate users between bands whenever possible
to achieve one or both of these objectives.
B. Power Saving by Propagation Improvement
The second concept, illustrated in Figure 2, is the
opportunistic reallocation of links or users to more appropriate
propagation bands at times when that spectrum becomes
available. This decreases necessary transmission power due to
improved propagation, or alternatively in a frequency reuse
scenario, reallocation based on the necessary deployed cell
density/radius and the given local propagation environment
can be used to reduce inter-cell interference through
minimizing power “leaking” into co-channel cells.
Note that these concepts might be implicitly or explicitly
employed together with the concept of reallocation to power
down radio equipment, yielding further improvement in power
efficiency. For instance, the opportunistic reallocation to
power down radio equipment might always leave the band
switched on that has the most appropriate propagation
characteristics given the deployed necessary cell
density/radius. Alternatively, links/users might always be
reallocated between bands for more appropriate propagation
whenever sufficient capacity is available in the target band,
even if moving those users/links does not make it possible to
power down any radio equipment.
III. ASSESSMENT OF POWER SAVING POTENTIAL
Power saving potential of the aforementioned concepts is
assessed through simulations of cellular systems carrying
either video traffic, FTP traffic, or HTTP (web browsing)
traffic. Concerning the traffic models used in simulations, it is
assumed that the average cell load (or in later simulations the
number of users in the cell) at time of day t, L(t), varies
according to a set of statistics on traffic load as a percentage of
busy hour load (BusyLoad) over a 24 hour period, pertaining
to traffic in a 3G network in London, UK, obtained via
interaction with Vodafone representatives within the Mobile
VCE Green Radio research program. Figure 3 depicts these
loads as a percentage of BusyLoad. All simulations assume
that the traffic load is scaled according to the chosen
BusyLoad depending on the time of day, such that if BusyLoad
is 10 users, for example, the number of users in the network at
17:00 hours is ~80% (see Figure 3) of BusyLoad, i.e., 8 users.
radio spectrum
frequency
high transmission
power
low transmission power,
or reduction in density of
active basestations
user/link moved to lower
frequency whenever
possible
Fig. 2: Reallocating users/links to improve propagation.
Fig. 3: Hourly variation of traffic load as a percentage of busy hour load
over a typical day.
Fig. 1: Reallocating traffic load between bands to enable radio network
equipment to be switched off.
Two simulation approaches are taken. In the first approach
which simulates video traffic, it is assumed that the statistical
number of active users in the cell receiving a video flow is
Poisson distributed, the mean of which at any one time of day
can be taken from the above-mentioned average load at that
time of day, L(t). Using this approach, the probability of there
being k active users in the cell at time of day t is expressed as
!
)(
),(
)(
k
etL
tkP
tLk
. (1).
Under this first approach, our numerical assessment cycles
in outer loops through a 24 hour period in steps of t of one
hour, and uses the value of L(t) (given BusyLoad) at each
hourly time unit to parameterize equation (1). In inner loops,
for each time unit, it then cycles through each possible value of
k representing each possible number of users in the cell, for all
participating frequency bands in the process, and for each set
of ks among the frequency bands ascertains the power
consumption that would be required given the selected
dynamic spectrum access power saving solution being applied.
The actual power consumption for each such case is then given
as this power consumption, multiplied by the probability of it
happening, which is of course the product of the probabilities
of the chosen values of k occurring for the participating
networks/frequencies, i.e., P(k, t)
network1
P(k, t)
network2
. This
result is then summed with equivalent results for all possible
chosen values of k to obtain the overall power consumption at
time unit t. The same operation is performed over all hourly
time units in the 24 hour period, and the average power
consumption is then taken among all time units. Exactly the
same process is also performed to find the average power
consumption of a conventional system without the dynamic
spectrum access power saving operation being applied, then
results are compared.
The second simulation approach assumes a separate
ON/OFF traffic flow to each user, either parameterized as FTP
traffic or HTTP (web browsing) traffic, whereby the number of
users receiving flows varies throughout the 24 hour period
according to L(t) (Figure 3). The chosen FTP and HTTP
ON/OFF model parameterizations are widely used in
literature, taken from [7]. In this second approach, a separate
simulation is performed for each hour in the 24 hour period for
a given number of users being present (obtained from Figure 3
scaled by BusyLoad), where, at each second in the simulation
duration, the simulation tallies the number of users present in
each band according to the ON/OFF model being applied to
each user. It then performs the power saving solution
according to this number of users being present, and ascertains
the power required in the before and after power saving
solution cases. Results are then averaged over results achieved
at each second in the simulation duration, where all such
simulations are typically performed over tens of millions of
seconds. Finally, results are averaged over all simulations
performed at each hour in the 24 hour period.
Configuration parameters applicable to the two simulation
approaches are given in Table 1.
A. Power Saving by Dynamically Powering Down Radio
Equipment
Simulating the LTE system under the assumption of video
traffic sources to users, Figure 4 gives the proportion of from-
the-socket power that is saved through moving users between
TABLE I
SIMULATION CONFIGURATION PARAMETERS
Parameter
Value
Simulated Systems
System configuration
Broadly reflecting LTE (Section
III.A) and HSDPA Rel. 5 (Section
III.B)
Bandwidth per band (LTE)
20MHz
Bandwidth per band (HSDPA)
5MHz
Channel path loss models for
HSDPA (path loss is not relevant
to the LTE simulations, as they
concentrate only on opportunistic
radio equipment powering down)
2GHz:
128.1+37.6∙log(distance
km
) [6]
5GHz:
141.52+28∙log(distance
km
) [6]
HSDPA pilot power
20% of cell power budget
Traffic Models
Streaming video traffic rate per
user (LTE)
384kbps
Data (FTP) traffic OFF duration
Exponentially distributed, mean
180s [7]
Data (FTP) traffic ON duration
Pareto distributed file size of
mean 2MB [7], α=1.5 (unless
otherwise stated) and k calculated
from the mean and α, with ON
duration calculated from each
sampled file size assuming a
fixed data rate of 1MB per user
Data (HTTP) traffic reading time
(OFF duration)
Exponentially distributed, mean
30s [7]
Data (HTTP) traffic parsing time
(OFF duration)
Exponentially distributed, mean
0.13s [7]
Data (HTTP) traffic main object
size (contributes to ON duration)
Truncated Lognormally
distributed, σ=1.37, μ=8.35,
min=100B, max=2MB [7]
Data (HTTP) traffic embedded
object size (contributes to ON
duration)
Truncated Lognormally
distributed, σ=2.36, μ=6.17,
min=50B, max=2MB [7]
Data (HTTP) traffic number of
embedded objects per page
(contributes to ON duration)
Truncated Pareto distributed,
α=1.1, k=2, max=55 (k subtracted
from each sampled value) [7]
Per-user rate in FTP/HTTP ON
durations
1Mbps (LTE), 64kbps (HSDPA)
Fig. 4: Power saving against busy hour load for network powering down
solutions (streaming video traffic).
bands to dynamically power down radio equipment, for 2, 3,
and 4 spectrum bands participating in the process, whereby the
2-band case considers the network powering down solution
with and without the sectorization switching solution operating
in tandem. For ease of depicting results, all results in Figure 4
are for the case where BusyLoad is the same for all channels.
From these results, it is clear that very significant power
savings of up to 50% or more can be achieved, these savings
being more typically being in the range of 20-50% if there are
lesser bands participating in the process or there is a greater
network loading. For the 2-band case, it is noted that if the
networks are heavily loaded the sectorisation switching
solution considerably improves performance compared with
the network powering down solution operating alone, but
offers no improvement if networks are lightly loaded. Other
simulations that we have performed, not presented in this
paper, show a large additional improvement in power saving
attained by the sectorisation switching solution, if there is a
significant difference in BusyLoads for the participating bands.
Figure 5 plots power saving results for the FTP and HTTP
(web browsing) ON/OFF traffic models over the LTE
configuration, where 2 bands are participating in the process
and the assumption is that the network powering down solution
only is employed. Results again show a significant power
saving potential of up to 50% for low network loads. In the
FTP case, power saving begins to reduce at a BusyLoad of ~20
users, reaching as low as 10% at a BusyLoad of ~50 users. In
the HTTP case, power saving begins to reduce at a BusyLoad
of ~150 users, and hits 10% at a BusyLoad of ~500 users. It is
emphasized here that per-user traffic load for the HTTP (web
browsing) case is very light compared with FTP downloads.
It might be noted that the benefits of these power saving
solutions are greatly accentuated if there is a difference in the
from-the-socket power consumptions of equipment operating
at the two bands. This is because the solution can always
choose to switch off the equipment at the more power hungry
spectrum band whenever it is possible for either band to
support the sum of the offered traffic loads. Figure 6 has
investigated this phenomenon for FTP and HTTP traffic by
varying the power consumption difference factor between the
two bands. This figure shows a very significant increase in
power saving if the power consumption difference factor is
high, particularly if the network is lightly loaded. This increase
can be a high as 30%, for the investigated range of values.
0
10
20
30
40
50
60
20
25
30
35
40
45
50
55
60
65
70
75
80
85
90
95
100
power saving (%)
busy load, users
(a)
(b)
Fig. 5: Power saving against busy hour load for network powering down
solution: (a) FTP ON/OFF traffic, (b) HTTP ON/OFF traffic.
(a)
(b)
Fig. 6: Power saving against busy hour load and power consumption
difference, for network powering down solutions: (a) FTP ON/OFF traffic,
(b) HTTP ON/OFF traffic.
B. Power Saving by Propagation Improvement
In assessing the power saving of reallocating users/links to
improve propagation, we have opted to study an HSDPA
system. This is because an HSDPA base station was the most
modern specification base station for which detailed data was
available on from-the-socket power consumption against
transmission power; no such data was available for LTE base
stations. For an anonymous manufacturer responding to a call
for information, internal documentation within the Mobile
VCE Green Radio research program indicates from-the-socket
power consumption for an HSDPA basestation at 100%
transmission power to be 857W, and at 20% transmission
power to be 561W. It is widely observed that from-the-socket
power consumption against transmission power broadly varies
with an mp+c relationship, comprising a fixed term c that is
independent of transmission power p, and a term that varies
with transmission power, mp. Given this, the above figures
regress to give 487W as the fixed part from-the-socket power
consumption c, and the gradient of variation with transmission
power m as 9.25 from-the-socket Watts per transmission Watt.
These values are used throughout this section.
In ascertaining necessary transmission power, we use values
in Table 3 of reference [6], with 80% of the power budget
being scaled by the number of users present in the system and
20% being allocated to pilot transmission. The comparison in
[6] is between full HSDPA networks operating at 2GHz (as
per current HSDPA deployments), and at 5GHz as argued is a
good future deployment option in [6]. A 600m cell radius is
chosen, where again we assume the aforementioned FTP
ON/OFF traffic model as described at the start of Section III.
Results in Figure 7 show that there is significant
transmission power saving potential through the opportunistic
reallocation scheme. Power saving initially increases to some
58% as the busy hour load is increased to 30; this is because it
is always possible to reallocate users to power down radio
equipment so adding more users simply increases the number
that are reallocated to better spectrum thereby saving
additional transmission power. However, as the traffic load is
increased further power saving decreases and a difference
begins to emerge in performance for the solutions with and
without opportunistic reallocation to save transmission power.
It is noted that especially if the networks are experiencing
moderate load, the opportunistic reallocation of links to save
transmission power saves up to an additional 10% compared
with just opportunistically reallocating users/links to be able to
power down radio equipment. This additional saving would be
far higher if the value of m, the gradient of the from-the-socket
power to transmission power relationship, were greater than
the modest value of 9.25 assumed here.
IV. CONCLUSION
Energy consumption in mobile communications can be
reduced significantly by the better use of spectrum.
This paper has investigated various concepts through which
dynamic adaptation of spectrum allocations among the range
of bands available to the operator (or, indeed, spectrum
sharing among different operators’ bands) can reduce
operators power consumption in providing services. Results
have shown considerable power saving potential, of up to 50%
or more in base station power. Among additional observations,
it has been shown that if there is a significant difference in
from-the-socket power consumption of the network equipment
utilized at the different participating spectrum bands, the
potential for power saving is further accentuated.
ACKNOWLEDGMENT
This work has been supported by the Green Radio Core Research Program
of the Virtual Centre of Excellence in Mobile & Personal Communications,
Mobile VCE, www.mobilevce.com, the ICT-ACROPOLIS Network of
Excellence, FP7 project number 257626, www.ict-acropolis.eu, UbiquiMesh,
OPPORTUNISTIC-CR, COST Actions IC0902 and IC0905 “TERRA”, and
by Marie Curie Reintegration Grant PLANOPTI (FP7-PEOPLE -2009-RG).
The authors are thankful to Dr. Terence E. Dodgson of Roke Manor Research
Ltd., part of the Chemring group, for partaking in very useful discussions.
REFERENCES
[1] I. Aklildiz, W. Lee, M. Vuran, and S. Mohanty, “NeXt
generation/dynamic spectrum access/cognitive radio wireless networks:
A survey,” Computer Networks, Vol. 50, No. 13, September 2006
[2] Federal Communications Commission, Second Memorandum Opinion
and Order, “In the Matter of Unlicensed Operation in the TV Broadcast
Bands; Additional Spectrum for Unlicensed Devices Below 900 MHz
and in the 3 GHz Band,” September 2010, accessible at:
http://www.fcc.gov/Daily_Releases/Daily_Business/2010/db0924/FCC-
10-174A1.pdf, accessed April 2010
[3] Federal Communications Commission, Notice of Inquiry, “In the Matter
of Promoting More Efficient Use of Spectrum Through Dynamic
Spectrum Use Technologies,” November 2010, accessible at:
http://www.fcc.gov/Daily_Releases/Daily_Business/2010/db1130/FCC-
10-198A1.pdf, accessed April 2010
[4] Results of WRC-07, online presentation available at:
http://www.itu.int/en/ITU-R/space/Presentations/resultsWRC07.pdf,
accessed April 2011
[5] Vodafone UK Corporate Responsibility Report 2009-10, accessible at:
http://www.vodafone.com/content/dam/vodafone/UK%20CR/download
s/online.pdf, accessed April 2010
[6] Orlando Cabral et al., “Integrated Common Radio Resource
Management with Spectrum Aggregation over Non-Contiguous
Frequency Bands,” Springer Wireless Personal Communications
Journal, February 2011
[7] 3GPP TR 25825, “Technical Specification Group Radio Access
Network; Dual-Cell HSDPA Operation,” V1.0.0, May 2008
0
10
20
30
40
50
60
70
0
10
20
30
40
50
60
70
80
90
100
110
120
130
140
150
power saving (%)
busy load, users
with opportunistic link reallocation to
save transmission power
without opportunistic link reallocation
to save transmission power
Fig. 7: Power saving against busy hour load through opportunistic link
reallocation to use better propagation bands (FTP ON/OFF traffic).
    • "These concepts were introduced in authors' previous work in [6], [14] where we consider (i) the opportunistic reallocation of loads between bands to allow radio network equipment in the bands that the traffic originated from to be powered down, and (ii) the opportunistic selection of more appropriate spectrum band based on propagation characteristics to minimize necessary transmission power through improving propagation. In [7] we developed ON/OFF traffic models for different services like FTP, HTTP, video streaming etc. and investigated the proposed concepts under these models to evaluate the energy saving potential. In [8] we evaluated the concept of dynamically powering down the radio network equipment by opportunistically shifting users from cellular to Wi-Fi networks. "
    [Show abstract] [Hide abstract] ABSTRACT: This paper investigates the green or energy efficient operation of cellular access networks through dynamic spectrum and traffic load management techniques. Different energy saving techniques are discussed along with applications in realistic multi-cellular scenarios. The use of base station cooperation is discussed. The outage probability analysis is also discussed for different energy saving techniques in order to satisfy a particular Quality-of-Service (QoS) threshold for different services. The analysis is supported by different base station power consumption models, traffic models, and realistic statistics of different base stations in London, UK. Results show a considerable power saving potential of up to 65% or more in base station power.
    Full-text · Conference Paper · Sep 2013
    • "Assuming that a web session for a mobile user lasts, on average, five minutes, we fixed l w to a value of 1/300. The model for FTP sessions is similar to that for web sessions, except for the fact that there is no parsing time [27,13]. The download time exclusively depends on the size of the object to be transferred, which follows a Pareto distribution. "
    [Show abstract] [Hide abstract] ABSTRACT: Cellular networks are rapidly evolving towards the fourth generation, thus providing a global infrastructure for wideband mobile network access. Currently, most of the energy consumption of such technology is by cellular base stations, which are not energy efficient—at least in terms of the transmission energy to “from-the-socket” energy consumption ratio. This paper addresses the problem of energy efficiency in cellular networks by taking advantage of the principles of cognitive networking, which promotes the creation of intelligent networks capable of self-configuration with minimal human intervention. In particular, this paper uses the concept of fuzzy cognitive maps to decide upon opportunistic traffic and user reallocations between radio network equipment operating in different spectrum bands to enable power saving modes by some subsets of the radio network equipment, and to utilize spectrum of more appropriate propagation characteristics to save transmission energy. The feasibility and performance of the proposed approach is investigated through simulations. Significant energy savings of some 25–30% are shown over a 72-h period, and blocking rate under the concept is shown to remain reasonable albeit exhibiting a high variance.
    Full-text · Article · May 2013
  • [Show abstract] [Hide abstract] ABSTRACT: This paper investigates the energy savings a cellular operator can achieve in the access network by dynamically offloading users or traffic loads to Wi-Fi networks. The method used to analyze energy savings is presented, followed by detailed simulations using different traffic types. Results show that significant savings of up to 65-70% can be achieved by opportunistically powering down cellular radio network equipment as facilitated by the offloading of users to Wi-Fi.
    Conference Paper · Jun 2012
Show more