GARCIA LAYOUT 8/21/09 12:28 PM Page 110
`
`FEMTOCELL WIRELESS COMMUNICATIONS
`
`Autonomous Component Carrier
`Selection: Interference Management in
`Local Area Environments for
`LTE-Advanced
`
`Luis G. U. Garcia, Aalborg University
`Klaus I. Pedersen and Preben E. Mogensen, Nokia Siemens Networks
`
`ABSTRACT
`
`Low-power base stations such as femtocells
`are one of the candidates for high-data-rate pro-
`visioning in local areas, such as residences,
`apartment complexes, business offices, and out-
`door hotspot scenarios. Unfortunately, the bene-
`fits are not without new challenges in terms of
`interference management and efficient system
`operation. Due to the expected large number of
`user-deployed cells, centralized network plan-
`ning becomes impractical, and new scalable
`alternatives must be sought. In this article we
`propose a fully distributed and scalable solution
`to the interference management problem in
`local areas, basing our study case on LTE-
`Advanced. We present extensive network simu-
`lation results to demonstrate that a simple and
`robust interference management scheme, called
`autonomous component carrier selection, allows
`each cell to select the most attractive frequency
`configuration; improving the experience of all
`users and not just the few best ones, while over-
`all cell capacity is not compromised.
`INTRODUCTION
`Low-power base stations, which are also com-
`monly referred to as femtocells or home base sta-
`tions, are low-cost user-deployed cellular base
`stations using an IP-based wired backhaul such as
`cable or digital subscriber line (DSL) designed to
`provide service in local environments similar to
`existing WiFi access points. In a recent contribu-
`tion [1], the authors indicated the key benefits of
`low-power base stations and outlined the many
`research opportunities as well as technological
`and business challenges associated with femto-
`cells. In [2] an interesting analysis of the financial
`impact of home base stations indicates that cur-
`rent macrocellular network deployment becomes
`less economically viable for increasing data rates.
`In this light, low-power base stations have
`recently reemerged as a promising technology
`component, and many believe it will definitely be
`
`one of the next steps in the evolutionary path of
`cellular wireless systems. Dense deployment of
`low-power base stations offers significantly higher
`capacity per area than macrocells, arising from
`using smaller cell sizes and more efficient spatial
`reuse. On the other hand, installation of many
`low-power base stations also poses new chal-
`lenges in terms of interference management and
`efficient system operation. The latter is especially
`the case for local areas where end users start
`installing home base stations without any prior
`network planning or carefully considering where
`other people in the immediate surroundings have
`installed other home base stations.
`The vast majority of previous contributions in
`the literature focused on solutions for cases where
`the user-deployed cells use the same frequency
`band employed by macrocells, in which case
`capacity and coverage gains can dwindle away if
`macro/femtocell co-channel interference is left
`unchecked. Nonetheless, in [3] the authors point
`out that femto-to-femto interference also becomes
`an important issue for indoor performance, espe-
`cially when femtocells are densely deployed.
`Therefore, we pay special attention to the nuances
`of interference footprint in local area deploy-
`ments, and do not address the complementary
`and equally interesting case of co-channel inter-
`ference to/from macrocells in overlaid networks.
`As demonstrated in [4], the interference foot-
`print is significantly different in such local area
`environments from nicely planned macrocell sce-
`narios, which consequently calls for new self-
`adjusting interference management techniques.
`Early work found in [5, 6] also highlights the need
`for the ability to self-scale and self-adjust, leading
`to a new autonomic paradigm with fully “robotic”
`base stations. The optimal sharing of radio
`resources between low-power base stations
`depend on many factors such as the mutual inter-
`ference coupling among them and the offered
`traffic for individual access nodes. Finding the
`optimal division of frequency resources between
`low-power base stations in a highly dynamic and
`partly chaotic environment is, in general, a non-
`
`110
`
`0163-6804/09/$25.00 © 2009 IEEE
`
`IEEE Communications Magazine • September 2009
`
`IPR2022-00648
`Apple EX1022 Page 1
`
`

`

`The proposed
`scheme uses a
`distributed and fully
`scalable approach.
`That is, selection of
`primary and
`secondary carriers is
`done locally by each
`cell. Hence, in the
`proposed concept
`there is no need for
`centralized network
`control.
`
`GARCIA LAYOUT 8/21/09 12:28 PM Page 111
`
`linear non-convex NP-hard optimization problem.
`Several interesting contributions are available in
`the literature, where decomposition of this chal-
`lenging problem into subproblems and the use of
`heuristic algorithms are proposed [7–8].
`As a case study, we base our investigations on
`LTE-Advanced, an evolved version of Long
`Term Evolution (LTE) Release 8, offering down-
`link peak data rates in excess of 1 Gb/s in a
`bandwidth of 100 MHz [9]. LTE-Advanced is
`currently in the study item phase in the Third
`Generation Partnership Project (3GPP), and
`design targets and new technology features for
`this system are also aimed at for local area sce-
`narios. We propose a fully distributed and scal-
`able solution based on minimal information
`exchange and negotiation between base stations
`akin to [10] where each individual low-power
`base station autonomously makes decisions with-
`out involving any centralized network control.
`The latter is considered to be the most attractive
`solution, especially for femto-type cells due to
`the expected large number of deployed cells.
`Our scheme mainly relies on measurements col-
`lected as a by-product of normal system opera-
`tion, producing useful statistics for interference
`conditions in the network. In this way each base
`station gathers knowledge about the surrounding
`environment and uses this information in the
`decision making process. We present network
`simulation results to further demonstrate that a
`simple and robust interference management
`scheme, called autonomous component carrier
`selection, is possible for LTE-Advanced, provid-
`ing attractive performance results in local area
`environments. Although the developed scheme
`is equally applicable for uplink and downlink,
`and for frequency-division duplex (FDD) and
`time-division duplex (TDD), we mainly present
`it for downlink TDD in this study.
`The rest of the article is organized as follows.
`We present the system model and outline the
`basic assumptions for autonomous component
`carrier selection. We include more detailed algo-
`rithm descriptions and brief comments on the key
`distinguishing aspects of TDD and FDD deploy-
`ments. System-level simulation results are pre-
`sented for an extended local area residential
`scenario. Finally, the article is closed with con-
`cluding remarks and an outlook on future studies.
`
`SYSTEM MODEL
`The 100 MHz LTE-Advanced bandwidth con-
`sists of five component carriers, each with a
`bandwidth of 20 MHz. The numerology of each
`component carrier is in coherence with LTE
`Release 8. The LTE-Advanced spectrum could
`also be less than 100 MHz, and therefore consist
`of less than five component carriers. The fre-
`quency band and spectrum allocation expressed
`via the number of component carriers and their
`bandwidth are configurable and known a priori
`by all base stations, hereafter denoted eNBs to
`follow 3GPP terminology. An LTE-Advanced
`terminal (user equipment [UE]) can be jointly
`scheduled on multiple component carriers at the
`same time (i.e., using carrier aggregation) or on
`a single component carrier as in LTE Release 8.
`We assume that each eNB always has one
`
`active component carrier, denoted the primary
`component carrier (PCC). The PCC is automati-
`cally selected by the eNB when it is first switched
`on, and is assumed to provide full cell coverage
`as it will be used by the terminals to camp, set
`up new calls, and so on. Depending on the
`offered traffic in the cell and mutual interfer-
`ence coupling with surrounding cells, transmis-
`sion and/or reception on all component carriers
`may not always be the best solution, especially
`for cell edge users. It is therefore proposed that
`each cell dynamically selects additional compo-
`nent carriers for transmission/reception as well
`(i.e., a second step after having selected the
`PCC). The latter is referred to as selection of
`secondary component carriers (SCCs). All com-
`ponent carriers not selected are assumed to be
`completely muted (uplink/downlink) and not
`used by the cell.
`The proposed scheme uses a distributed and
`fully scalable approach. That is, selection of pri-
`mary and secondary carriers is done locally by
`each cell. Hence, in the proposed concept there
`is no need for centralized network control. The
`suggested interference coordination mechanism
`is part of a hierarchical resource management
`process. The (re-)selection of component carri-
`ers is fairly slow and occurs over a longer time
`span than fast packet scheduling, which is free to
`operate within the restrictions imposed by the
`carrier selection process. Our three fundamental
`premises are:
`• Absolute priority of primary over secondary
`component carriers; avoidance of PCC re-
`selection, while SCCs can be reselected on
`a faster basis.
`• When the offered traffic for an eNB
`requires more bandwidth, a cell may aug-
`ment its cell capacity by allocating SCCs.
`• An eNB is only allowed to allocate SCCs
`provided it does not result in excessive
`interference to the surrounding cells, as
`explained later.
`The last item is a policy preventing a so-
`called greedy eNB from using all the available
`component carriers for its own sake, even when
`this results in intolerable interference to the
`neighboring eNBs. Hence, the proposed scheme
`for autonomous component carrier selection
`effectively provides an automatic frequency
`reuse scheme at component carrier resolution.
`This approach ensures protection of both traffic
`and control channels.
`We assume that the allocation of PCC and
`SCCs is signaled among eNBs (either over the
`backhaul or over the air) periodically and/or
`whenever the allocation is changed, so eNBs
`know which component carriers neighboring
`eNBs are currently using. This information is of
`critical importance and is summarized in what
`we refer henceforth as the Radio Resource Allo-
`cation Table (RRAT). Essentially, such tables
`make femtocells aware of the existence of other
`femtocells. Finally, it is assumed that local eNB
`measurements are available, as well as terminal
`measurements for selection of the component
`carriers. The next section on selection of the
`PCC deals with the first premise, whereas the
`SCC selection scheme described later embodies
`the other two assumptions.
`
`IEEE Communications Magazine • September 2009
`
`111
`
`IPR2022-00648
`Apple EX1022 Page 2
`
`

`

`there is marginal interference coupling with
`those. Based on this matrix, we propose the fol-
`lowing procedure for initial primary component
`carrier selection:
`1 If there are row entries in the matrix with
`no selections, the corresponding component
`carrier is selected. (If there are multiple of
`such rows, either select randomly, or select
`the component experiencing the lowest
`uplink received interference power.) Other-
`wise, go to 2.
`2 If there are row entries without P, select
`one of those for primary. Select the row
`entry with the lowest number of S if there
`are multiple rows without P.
`3 If all row entries include P, select the com-
`ponent carrier for primary with maximum
`path loss to the neighboring eNB having
`the same component carrier as its primary.
`4 When there are multiple candidate compo-
`nent carriers for primary according to the
`above rules, select the component carrier
`with the lowest experienced uplink interfer-
`ence, based on eNB measurements of wide-
`band uplink received interference power.
`The above rules essentially assume priority of
`primary over secondary component carriers, as
`each eNB should always have one PCC with full
`cell coverage. The inter-eNB path loss measure-
`ments are used to ensure that only eNBs with
`the largest possible path loss separation select
`the same component carrier for primary.
`It is worth mentioning that the proposed
`method, solely relying on what the eNBs sense,
`was found to be sensitive to the order in which
`eNBs are turned on in case of a very limited
`number of component carriers from which to
`choose. However, with five component carriers,
`the sensitivity was rather small.
`After the new eNB has selected its PCC, the
`cell is configured, and it is ready to transmit and
`carry traffic. In parallel, the eNB shall constantly
`monitor the quality of the PCC to make sure that
`it continues to have the desired quality and cov-
`erage. If poor quality is detected, recovery actions
`will be triggered to improve the situation. Such
`actions can be understood as additional defensive
`measures, not allowing potentially erroneous
`SCC allocations to catastrophically interfere with
`neighboring base stations. Recovery actions are
`the subject of ongoing investigations and out of
`the scope of this contribution; nonetheless, they
`may range from interference reduction requests
`toward neighboring cells where the same compo-
`nent carrier is used as an SCC, to the selection of
`a new PCC with better quality.
`
`SECONDARY COMPONENT CARRIER
`SELECTION
`As stated earlier, our scheme imposes certain
`constraints for selection of SCCs, which basically
`implies that eNBs have to take the interference
`created toward other cells into account. The goal
`is a flexible yet simple and efficient sharing of the
`spectral resources that will not prevent one cell
`from using the entire spectrum when this is a sen-
`sible choice. Granting eNBs the ability to “learn”
`what sensible means is the key aspect here.
`
`GARCIA LAYOUT 8/21/09 12:28 PM Page 112
`
`SP
`
`S S
`
`eNB #4
`
`New
`eNB #5
`is switched
`on
`
`eNB #1
`
`P S
`
`SP
`
`S
`
`eNB #3
`
`eNB #2
`
`S P
`
` Figure 1. Simple illustration of the autonomous component carrier concept.
`All eNBs announce their existence and current resource allocation. Addition-
`ally, eNBs that are being switched off could signal their leaving.
`
`PRIMARY COMPONENT CARRIER
`SELECTION
`The proposed autonomous component carrier
`selection scheme is illustrated in Fig. 1 with a
`simple example. Here there are four existing
`eNBs, while a new eNB, #5, is being switched
`on, and hence is ready for first selecting its PCC.
`The current selection of PCC and SCCs is illus-
`trated for each eNB with P and S, respectively.
`Component carriers not allocated as PCC or
`SCC are completely muted, and not used to
`carry any traffic.
`As the eNB is being initialized, it clearly can-
`not rely on UE assisted mechanisms; therefore, in
`addition to the information available in the
`RRAT, we propose new inter-eNB measurements
`based on reference signal received power levels
`for the purpose of estimating the path loss
`between neighboring eNBs. In FDD systems this
`implies that eNBs are able to listen to the down-
`link band as well. Conversely, in TDD systems,
`this is not an additional requirement, since uplink
`and downlink use the same band. It is proposed
`that the new eNB carry out the measurements on
`the PCCs of the surrounding cells and that knowl-
`edge of their corresponding reference symbol
`transmit power is available (signaled between
`eNBs) so that the inter-eNB path loss can be esti-
`mated. Notice that these inter-eNB path loss mea-
`surements need not be frequent as they are only
`required by new eNBs when they are switched on.
`Given the aforementioned information, a
`matrix for initial PCC selection is formed as
`illustrated in Fig. 2, where the eNBs are sorted
`according to the path loss experienced from the
`new eNB. As depicted in Fig. 2, only the neigh-
`boring eNBs within a certain path loss threshold
`are considered relevant. Neighboring eNBs with
`higher path loss are not taken into account as
`
`112
`
`IEEE Communications Magazine • September 2009
`
`IPR2022-00648
`Apple EX1022 Page 3
`
`

`

`#7
`
`#6
`
`S
`
`P
`
`S
`
`S
`
`P
`
`Increasing
`path loss
`
`P#
`
`5
`
`S
`
`#1
`
`#2
`
`#3
`
`P
`
`S
`
`S
`
`P
`
`S
`
`S
`
`P
`
`P#
`
`4
`
`S
`
`S
`
`S
`
`CC #1
`
`CC #2
`
`CC #3
`
`CC #4
`
`CC #5
`
`Path loss threshold
`(only those eNBs are
`considered when
`selecting primary)
`
` Figure 2. Matrix for initial primary component carrier selection.
`
`matically affect only few UEs (e.g., those near
`windows in a tall building).
`In addition to the incoming BIM, eNBs also
`maintain another BIM table that lists all the
`potentially interfered cells. This BIM is known as
`the outgoing BIM. Basically, it allows a cell to
`estimate how much interference it generates
`toward each of its neighbors if it decides to use
`the same CC the neighboring cell already uses.
`It is linked to the incoming BIM as follows: At
`the same time an interfering cell entry (cell 2) is
`added or modified into the incoming BIM of the
`interfered cell (cell 1), the corresponding inter-
`fered cell (cell 1) is added as an entry into the
`outgoing BIM of the interfering cell (cell 2). The
`relation between the incoming and outgoing
`BIMs is illustrated in Fig. 3.
`It is assumed that the reporting of measure-
`ments from the UE to the eNBs for the purpose
`of BIM is fairly slow in order to minimize the
`control signaling overhead and measurement
`burden from this. Similarly, the update rate of
`the local BIM information in each eNB is also
`anticipated to be rather slow compared to, say,
`packet scheduling. However, the ideal update
`rate is the subject of future investigations.
`In possession of the information just
`described, an eNB is now able to decide whether
`or not the new allocation(s) will jeopardize any
`existing allocations based on the target SINR
`values. As explained, we assume a priori knowl-
`edge of the minimum SINR targets (C/I)PCC and
`(C/I)SCC for primary and secondary component
`carriers, respectively. The process is fairly
`straightforward, and the interested reader can
`find a somewhat more formal mathematical
`description in [11]. In the following we provide a
`simplified description of the process.
`In essence, for each component carrier not
`yet allocated to the cell, the eNB calculates a set
`of four differences (in dB). These differences
`can be understood as neighbor-specific BIM
`
`GARCIA LAYOUT 8/21/09 12:28 PM Page 113
`
`One of the design targets is to maximize the
`cell throughput for each eNB, but always ensur-
`ing that the experienced signal-to-interference-
`plus-noise ratio (SINR) on PCC and SCC equals
`at least (C/I)PCC and (C/I)SCC, which represent
`minimum SINR targets expressed in decibels for
`the PCC and SCCs, respectively. These are con-
`sidered as configurable parameters that could
`come from operations and maintenance (O&M),
`for example. Without loss of generality, we
`assume that (C/I)PCC is higher than (C/I)SCC as
`the PCC is assumed to always have full cell cov-
`erage while the SCCs may have reduced cover-
`age (i.e., use less transmit power).
`Once it is detected that the capacity offered
`by the PCC alone is not sufficient to carry the
`offered traffic, the eNB will use two information
`sources to autonomously decide whether it can
`allocate additional SCCs. The first source is the
`aforementioned RRAT, which provides real-
`time information on the usage of component
`carriers by neighboring eNBs. The second piece
`is the background interference matrix (BIM),
`which essentially expresses the interference cou-
`pling between cells. Now, unlike the selection of
`the PCC, UE assistance comes into the picture
`during the creation and maintenance of BIMs.
`Each active UE connected to a cell performs
`downlink measurements of reference signal
`received power levels which are reported to its
`serving eNB. These measurements are conduct-
`ed both towards the serving cell and the sur-
`rounding cells (e.g., for handover purposes).
`Given these UE measurements, the serving eNB
`calculates a ratio expressed in decibels of own to
`other cell received signal power. We call it a
`conditional C/I sample. That essentially allows
`eNBs to produce an estimate of potential signal
`quality as perceived by their served UE. Each
`time a certain (quantized) value is calculated, an
`occurrence counter is incremented. Eventually,
`given enough samples, empirical C/I distribu-
`tions are generated locally by each eNB, one for
`each detected neighbor. A matrix is then built;
`we call it the incoming BIM.
`The C/I value stored in the BIM for each
`neighboring cell is the value corresponding to a
`certain outage probability of , say, 95 percent.
`The C/I value is a measure of mutual interfer-
`ence coupling between a pair of cells. Therefore,
`each cell maintains local information on all
`potential interfering cells and a corresponding C/I
`value. In this example only 5 percent of users are
`likely to experience C/I values in the downlink
`lower than the value stored in the BIM. Notice
`that this C/I is only realized if the interfered cell
`and the interfering cell use the same component
`carrier simultaneously. As component carriers
`are likely to experience the same path loss condi-
`tions, the BIM is component-carrier-independent
`as it is only based on path loss types of measure-
`ment (i.e., it is sufficient for the UE to measure a
`single component carrier per cell).
`Alternatively, in a more dynamic setting the
`C/I value stored in the BIM for each neighbor-
`ing cell could correspond to near-real-time con-
`ditional C/I values reported by the served UE
`most severely impacted by that particular neigh-
`bor. This approach would better capture the
`effects of faraway yet strong femtocells that dra-
`
`IEEE Communications Magazine • September 2009
`
`113
`
`IPR2022-00648
`Apple EX1022 Page 4
`
`

`

`GARCIA LAYOUT 8/21/09 12:28 PM Page 114
`
`If for any given
`neighbor using
`that particular
`component carrier
`either as a PCC or
`SCC, any of the four
`margins is found to
`be negative, that
`particular compo-
`nent carrier is not
`taken into use and
`another component
`carrier is evaluated.
`
`Outgoing BIM (cell 1)
`
`Incoming BIM (cell 1)
`
`Outgoing BIM (cell 2)
`
`Cell Id C/I
` 2 10 dB
` 3 8 dB
` 4 7 dB
`
` ... ...
`
` N 20 dB
`
`Cell Id C/I
` 2 6 dB
` 3 13 dB
` 4 25 dB
`
` ... ...
`
` N 30 dB
`
`Interference
`relations
`
`Cell Id, C/I
` 1 6 dB
`
` ... ...
`
`Outgoing BIM (cell 3)
`
`Cell Id, C/I
` 1 13 dB
`
` ... ...
`
`Outgoing BIM (cell 4)
`
`Cell Id, C/I
` 1 25 dB
`
` ... ...
`
`Outgoing BIM (cell N)
`
`Cell Id, C/I
` 1 30 dB
`
` ... ...
`
` Figure 3. Relation between incoming and outgoing BIM entries.
`
`entry margins with respect to (C/I)SCC in incom-
`ing interference evaluations, and with respect to
`either (C/I)PCC and (C/I)SCC in outgoing inter-
`ference evaluations, depending on the compo-
`nent carrier usage of the interfered neighbor. If
`for any given neighbor using that particular com-
`ponent carrier as either a PCC or SCC, any of
`the four margins is found to be negative, that
`particular component carrier is not taken into
`use, and another component carrier is evaluated.
`The four differences mentioned earlier corre-
`spond in fact to estimated downlink incoming,
`downlink outgoing, uplink incoming, and uplink
`outgoing SINR margins. It is important to stress
`that all uplink estimations are rough approxima-
`tions of the actual uplink interference situation
`based on measurements UE has made on the
`“interfered” side. The rationale behind this is
`that incoming/outgoing downlink interference
`propagates through the same path as the outgo-
`ing/incoming uplink interference; thus, the
`downlink C/I estimate contains correlated and
`useful information. Now, given the hypothetical
`C/I values in Fig. 3, a simple example illustrates
`the proposed concept. Let us assume cell 1 is
`evaluating a component carrier that is currently
`only in use by cell 3 as its PCC, and (C/I)PCC
`and (C/I)SCC are set to 10 dB and 8 dB, respec-
`tively. Since cell 1 intends to use this component
`carrier as an SCC, the estimated downlink
`incoming C/I margin is positive, since 13 dB is
`above (C/I)SCC. However, allocation will be
`denied because the estimated downlink outgoing
`C/I margin is negative, for 8 dB is lower than
`(C/I)PCC. Uplink incoming and outgoing SINR
`margins are calculated similarly.
`
`PERFORMANCE RESULTS
`We study the potential benefits of our proposed
`autonomous component carrier selection (ACCS)
`for LTE-Advanced femtocells using system-level
`simulations. Our system operates at 3.4 GHz car-
`rier frequency with up to 100 MHz bandwidth,
`the maximum transmission power of eNBs is 200
`mW (23 dBm), and 3dBi antenna gain is
`assumed. Even though our scheme does not pre-
`clude other power allocations, for simplicity,
`there is no downlink power control, and the total
`transmission power is evenly divided among the
`component carriers into which the bandwidth is
`divided; hence, eNBs will only transmit at full
`power if they employ all component carriers. A
`simple full-buffer traffic model (i.e., eNBs and
`UEs always have data to transmit) and a simple
`round-robin packet scheduler are considered.
`Figure 4 depicts the topology of our refer-
`ence residential scenario. It represents the model
`for a single indoor floor layout with one eNB
`(small circle) randomly placed in each 10 m × 10
`m four-room residence. The number of uniform-
`ly distributed users per residence is fixed to 4.
`The indoor path loss and slow fading models
`used are based on A1-type generalized path loss
`models for the frequency range 2–6 GHz devel-
`oped in WINNER [12].
`The simulation tool relies on series of “snap-
`shots.” During each snapshot, path loss, shadow-
`ing, and the location of devices remain constant.
`In practice, various system-level practical aspects
`such as the effects of achievable bandwidth effi-
`ciency, control channel overhead, and receiver
`algorithms all limit the achievable system-level
`
`114
`
`IEEE Communications Magazine • September 2009
`
`IPR2022-00648
`Apple EX1022 Page 5
`
`

`

`GARCIA LAYOUT 8/21/09 12:28 PM Page 115
`
`5 m
`
`5 m
`
`20 m
`
` Figure 4. Example of a residential deployment scenario with 16 eNBs with
`uncoordinated location planning. Walls between residences are modeled dif-
`ferently from the other internal walls.
`
`20 m
`
`6,67
`
`3,94
`
`2,25
`
`3,95
`
`3,16
`
`5,40
`
`Reuse 1/2
`ACCS
`
`1,55
`
`2,55
`
`1,40
`
`1,20
`
`1,00
`
`0,80
`
`0,60
`
`0,40
`
`ormalized average DL cell throughput
`
`0,20N
`
`0,00
`
`0%
`
`25%
`
`50%
`75%
`Activity factor
`
`100%
`
` Figure 5. Simulation results for the 100 MHz configuration for variable
`deployment density levels.
`
`er choice as its overall cell capacity is quite limit-
`ed. Similar results to those in Figs. 5 and 6 have
`also been generated for the uplink. Based on
`those results, we draw similar conclusions; that
`is, the ACCS approach is equally valid for the
`uplink.
`Finally, we highlight that it is possible to
`trade off overall cell capacity for cell edge capac-
`ity in a controllable manner, by varying the C/I
`targets of primary and secondary component
`carriers.
`
`spectral efficiency, and a modified Shannon
`capacity formula according to [13] maps the
`SINR to corresponding throughput values. Spec-
`trum efficiency is limited to 5.4 b/s/Hz since only
`a single transmit and receive antenna configura-
`tion has been considered.
`Two different spectrum settings are used in
`our simulation. The first one is the general case
`of 100 MHz system bandwidth and 5 component
`carriers of 20 MHz each. In the second one the
`available spectrum is 60 MHz and therefore con-
`sists of 3 component carriers of 20 MHz each. In
`all cases (C/I)PCC and (C/I)SCC are set to 10 and
`8 dB, respectively. Additionally, we consider dif-
`ferent deployment densities to evaluate the flexi-
`bility and scalability of the proposed concept. In
`both cases we assumed private access, also known
`as closed subscriber group (CSG) mode, whereby
`UE can only connect to the eNB in the same res-
`idence. Private access is far more challenging
`than open access from an interference manage-
`ment perspective, since in the latter UEs are
`served by the eNB with the strongest signal ame-
`liorating the interference scenario. In our simula-
`tions all cells first select their PCC and only then
`the SCC selection starts. Because of the full load
`assumption, a cell will always allocate as many
`SCCs as possible given the existing allocation of
`its neighbors and interference coupling.
`The results are summarized in Figs. 5 and 6.
`The activity factor in the x-axis indicates how
`dense the deployment is as it represents the share
`of eNBs that are active. For example, activity fac-
`tors of 25 and 75 percent mean that on average 4
`and 12 of the 16 eNBs are active, respectively.
`Given the private access assumption, eNB inactiv-
`ity in a given residence, implies inexistence of
`UEs in that residence. The y-axis is the normal-
`ized downlink average cell throughput. The bub-
`ble size is proportional to the number inside it,
`which represents the normalized cell edge user
`throughput (5 percent outage). All values are nor-
`malized with respect to the corresponding
`throughput figure achieved when the entire avail-
`able spectrum is used by all cells (reuse 1/1).
`For the sake of comparison, Fig. 5 also pre-
`sents the performance achieved by genie-aided
`hard frequency reuse 1/2, whereby a severe
`interfering pair of cells each uses complementary
`halves of the spectrum. The results clearly show
`that our concept (ACCS) renders overall cell
`throughput nearly insensitive to the activity fac-
`tor, while retaining the benefit of higher cell
`edge user throughput. It achieves near four
`times the throughput provided by reuse 1/1 when
`all 16 eNBs are active. Despite being a very
`attractive solution for 100 percent activity factor,
`the hard limit of 50 MHz imposed by reuse 1/2
`severely limits the overall cell throughput in
`sparser deployments.
`Figure 6 presents the simulation results for a
`system with 3 component carriers of 20 MHz
`each. In this case reuse 1/2, which given the pre-
`vious results seemed to be a nearly optimal
`choice for this particular environment assuming
`100 percent activity factor, cannot be achieved in
`a straightforward way. Now the comparison is
`performed against reuse 1/3, entailing a hard
`limit of 20 MHz per cell. The trend is nearly the
`same with the exception that reuse 1/3 is a poor-
`
`IEEE Communications Magazine • September 2009
`
`115
`
`IPR2022-00648
`Apple EX1022 Page 6
`
`

`

`REFERENCES
`[1] V. Chandrasekhar, J. G. Andrews, and A. Gatherer,
`“Femtocell Networks: A Survey,” IEEE Commun. Mag.,
`Sept. 2008, pp. 59–67.
`[2] H. Claussen, L. T. W. Ho, and L.G. Samuel, “Financial
`Analysis of a Pico-Cellular Home Network Deployment,”
`IEEE ICC ’07, June 2007, pp. 5604–09.
`[3] Y. Shu-ping et al., “WiMAX Femtocells: A Perspective on
`Network Architecture, Capacity, and Coverage,” IEEE
`Commun. Mag., vol. 46, no. 10, Oct. 2008, pp. 58–65.
`[4] Y. Wang et al., “Fixed Frequency Reuse for LTE-
`Advanced Systems in Local Area Scenarios,” IEEE Proc.
`VTC-Spring, Apr. 2009.
`[5] F. J. Mullany et al., “Self-Deployment, Self-Configura-
`tion: Critical Future Paradigms for Wireless Access Net-
`works,” Proc. IFIP TC6 WG6.6 Int’l. Wksp. Autonomic
`Commun. WAC, Oct. 2004, Berlin, Germany, and in
`LNCS, Springer Verlag, vol. 3457, July 2005, pp. 58–68.
`[6] H. Claussen, “Distributed Algorithms for Robust Self-
`Deployment and Self-Configuration in Autonomous
`Wireless Access Networks,” Proc. IEEE ICC ’06, vol. 4,
`June 2006, Istanbul, Turkey, pp. 1927–32.
`[7] G. Li and H. Liu, “Downlink Dynamic Resource Alloca-
`tion for Multi-Cell OFDMA System,” IEEE Trans. Wireless
`Commun., vol. 5, Dec. 2006, pp. 3451–59.
`[8] M. S. Kang and B. C. Jung, “Decentralized Intercell
`Interference Coordination in Uplink Cellular Networks
`Using Sub-band Exclusion,” IEEE WCNC, Apr. 2009.
`[9] S. Parkvall et al., “LTE-Advanced — Evolving LTE
`Towards IMT-Advanced,” IEEE Proc. VTC., Sept. 2008.
`[10] M. Salmenkaita, J. Gimenez, and P. Tapia, “A Practical
`DCA Implementation for GSM Networks: Dynamic Fre-
`quency and Channel Assignment,” IEEE Proc. VTC, Apr.
`2001, pp. 2529–33.
`[11] 3GPP tech. docu. R1-090235, “Use of Background
`Interference Matrix for Autonomous Component Carrier
`Selection for LTE-Advanced,” Nokia Siemens Networks,
`Feb. 2009; http://www.3gpp.org
`[12] WINNER II, D1.1.2, “WINNER II Channel Models part I-
`Channel Models,” Tech. Rep., Sept. 2007.
`[13] P. Mogensen et al., “LTE Capacity Compared to the Shan-
`non Bound,” IEEE Proc. VTC, Apr. 2007, pp. 1234–38.
`[14] T. H. Kim and T. J. Lee, “Throughput Enhancement of
`Macro and Femto Networks by Frequency Reuse and
`Pilot Sensing,” IEEE Proc. Int’