Self-Organizing Interference Management For LTE
Self-Organizing Interference Management For LTE
Self-Organizing Interference Management For LTE
for LTE
Francis Dominique, Christian G. Gerlach, Nandu Gopalakrishnan,
Anil Rao, James P. Seymour, Robert Soni, Aleksandr Stolyar,
Harish Viswanathan, Carl Weaver, and Andreas Weber
Introduction
Fourth generation (4G) cellular systems are cur- by modulating each of these sub-carriers. Further,
rently being developed and deployed. Long Term time is divided into slots consisting of a number of
Evolution (LTE) is one such 4G system and is an evo- OFDM symbols, and users are scheduled to transmit
lution of the 3rd Generation Partnership Project on an assigned set of sub-carriers in specific time slots.
(3GPP) Universal Mobile Telecommunications System Scheduling can be either persistent, semi-persistent,
(UMTS) standard [2, 7]. These systems target signifi- or non-persistent.
cantly higher sector capacities and higher per user LTE is generally specified as a universal fre-
data rates compared to third generation systems. LTE quency reuse (or reuse 1) system to attain high
uses orthogonal frequency division multiple access performance through efficient use of frequency
(OFDMA) in the downlink (DL) and single carrier fre- resources. While there is no intra-cell interference
quency division multiple access (SC-FDMA) in the in LTE since users within a given cell are orthogonal
uplink (UL) as the physical layer radio access tech- due to the orthogonality of the sub-carriers, and the
nology. OFDMA has been selected as it provides high enhanced node B (eNB) scheduler ensures that a
spectral efficiency and robust performance in mobili- physical resource is allocated to only one user at any
ty and fading scenarios. Additionally, SC-FDMA pro- given time, there is interference because of trans-
vides a low peak-to-average power ratio. In LTE, the missions on the same physical resources in a neigh-
total system bandwidth is divided into a number of boring cell. This inter-cell interference has a major
orthogonal sub-carriers. Information is transmitted impact, especially on users at the edge of the cell, as
Bell Labs Technical Journal 15(3), 19–42 (2010) © 2010 Alcatel-Lucent. • DOI: 10.1002/bltj.20455
Panel 1. Abbreviations, Acronyms, and Terms
3GPP—3rd Generation Partnership Project OI—Overload indicator
4G—Fourth generation PDRPC—Partial direct reuse with power control
AWGN—Additive white Gaussian noise PRB—Physical resource block
BTS—Base transceiver station PSD—Power spectral density
CQI—Channel quality indicator QoS—Quality of service
DL—Downlink RACH—Random access channel
DPL—Delta path loss RB—Resource block
eNB—Enhanced node B RF—Radio frequency
FPC—Fractional path loss compensation RNC—Radio network controller
FSS—Frequency selective scheduling RNTP—Relative narrowband transmit power
HARQ—Hybrid automatic repeat request SA—Sector autonomous
HII—High interference indicator SC-FDMA—Single carrier frequency division
IBZ—Interference bearing zone multiple access
ICIC—Inter-cell interference coordination SINR—Signal-to-interference-plus-noise ratio
IoT—Interference over thermal SON—Self-optimizing network
IR—Inverted reuse SRS—Sounding reference signal
IRC—Interference rejection combining UE—User equipment
ISD—Inter site distance UL—Uplink
LTE—Long Term Evolution UMTS—Universal Mobile Telecommunications
MGR—Multi-cell gradient System
MGR-UL—MGR for uplink Univr—Universal reuse
OFDMA—Orthogonal frequency division VoIP—Voice over Internet Protocol
multiple access
IoT control
Power control
ICIC component
Scheduler
Figure 1.
ICIC components and interactions.
PRBs. In the restricted PRBs, transmissions are at a Figure 4 shows simulation results for a 7/6
lower power, which results in a soft partial frequency reuse scheme, also known as a 1/7 inverted reuse
reuse scheme, or there may not be any transmissions scheme [5]. The test configuration is 3GPP case 1
at all in the restricted PRBs, which results in a hard [1] with an inter-site distance (ISD) of 500 meters
partial frequency reuse scheme. The neighboring cell for a bandwidth of 10 MHz. Three different CQI
can then schedule its users (typically cell edge users) feedback types are studied. A difference of 6 dB
in the restricted PRBs of its neighbors and realize an between the best server and highest interferer path
improved SINR since the neighboring cells are either loss is used to classify users into cell inner and cell
transmitting with a lower power or not transmitting at outer groups. Additional detail on the simulation
all on those PRBs. Users are generally classified into parameters and results can be found in [5]. Good
two groups: cell inner and cell edge groups. This clas- gains can be obtained with ICIC under conditions
sification can be made based upon a number or com- where it is not possible to exploit frequency selec-
binations of such metrics as: tivity in the propagation channel, such as at high
1. User’s path loss to serving cell, and vehicular speeds, or when use of semi-persistent
2. Differential path loss between the serving cell and scheduling is necessary, such as for Voice over
the strongest neighbors. Internet Protocol (VoIP) applications. One thing to
Fairness:
propfair/
max-min rate
Scheduling ICIC
Network
policy performance
QoS-constraints deployment
(GBR/MBR) Femto/pico/micro/macro
Figure 2.
Determinants of ICIC performance.
be noted (and which is true for all the techniques a.If K is the reuse factor, (K 1) proximate
shown) is that while the simulations assume a neighbor cells each have similar but distinct
hexagonal layout, such a geometric layout is not pieces of cleared spectrum that partition the
possible in the real world. However, since the algo- whole bandwidth.
rithms use metrics (such as the server and neighbor b. Cleared spectrum is assumed frequency diverse.
path loss reports) which reflect the true interference 2. Each cell distributes its total transmit power in
conditions as seen by the users, no issues are antici- the following fashion:
pated with irregular topology. a. A high transmit power spectral density (PSD)
The partial direct reuse with power control static on the cleared spectrum of bandwidth W/K
ICIC scheme. In contrast to the inverted reuse scheme dedicated to edge users.
where each cell has power restrictions on certain
b. A low transmit PSD on the remaining com-
PRBs, the partial direct reuse with power control (K 1)W
(PDRPC) takes a more direct approach. mon spectrum of bandwidth .
K
The PDRPC algorithm implements the following c. Boosts the PSD in the cleared spectrum with
steps, as shown in Figure 5: respect to reuse 1 by a factor of aK. a (0, 1),
1. In each cell, a part of the spectrum is cleared for is a fairness control parameter.
edge users to operate at low interference (cell d. From power conservation (assuming total
coloring). power used is dP), the de-boost factor b for
10
dB
1 2 3 4 5 6 7
10
dB
1 2 3 4 5 6 7
3GPP support:
The transmit PSD mask utilized by each cell is made known
to the neighboring cells through RNTP reporting via the X2
interface (see TS 36.423)
Figure 3.
Downlink static ICIC operation: inverted reuse concept.
100
0.5
0.25
0
0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6
Spectral efficiency (bit/s/Hz/sector)
Figure 4.
Spectral efficiency versus cell border throughput for DL static ICIC for 3GPP case 1.
Figure 6 and Figure 7 show performance results are lumped into one center cell user location with
obtained for two different populations of center cell values for received symbol energy and interference
users using lumped user analysis. In lumped user (relative to thermal noise) representing good radio
analysis, as opposed to using some user distribution frequency (RF) conditions, and the remaining users
across cell geometries, a parametric fraction of users into a cell edge user location representing poor RF
conditions. Figure 6 shows that when the fraction of
center cell users is small, performance gains are seen
simultaneously for the edge user, the center user, and
average sector throughputs with respect to the uni-
Universal reuse Tx PSD versal reuse (Univr) scheme, while variants of the
Tx PSD
Figure 6.
PDRPC lumped user performance for center user fraction of 0.1.
between various cells on a dynamic basis based upon simple and may be built on top of static schemes by
a number of factors such as cell load, traffic distribu- dynamically changing the restrictions on the resources
tions, user distributions, quality of service (QoS) con- through a simple request-grant mechanism employed
straints, and other parameters. For example, between the cells of interest. They can also be more
depending on the load (e.g., a geometric concentra- complex where an optimization problem involving a
tion of terminals at the border between two cells), the utility function is solved.
restrictions are distributed between the two cells Semi static ICIC with request-grant mechanism. The
involved and possibly other neighbors. This allows an semi static ICIC approach [5] builds on top of the
increase in spectrum efficiency. In this way, with dif- static ICIC approach with resource assignments on a
ferent loads, lightly loaded cells can help higher load demand basis. This enables the network to adapt to
neighbors. changing network conditions in a self-optimizing way.
The reconfiguration of the restrictions can be The network is initially set up for nominal static ICIC
done on a timescale of the order of milliseconds or operation. In addition to the existing exchange of fre-
longer, depending upon the operating scenario. Inter- quency coloring information between the cells
eNB communication corresponds to information required for static ICIC operation, additional messag-
needed to decide on reconfiguration of the scheduler ing is required for this dynamic reconfiguration of
restrictions as well as the actual reconfiguration deci- resources between cells. Two additional messages are
sions. The signaling rate could be on the order of sec- required: a request message from a requesting cell
onds to minutes, depending upon the operating and a grant message from a grantor cell. A requesting
scenario. Dynamic ICIC schemes can be relatively cell is typically a heavily loaded cell or a cell with a
0.5
3 4 5 6 7
Power boost αK (linear)
Figure 7.
PDRPC lumped user performance for center user fraction of 0.5.
localized high concentration of cell edge users. If the the expanded PRB region of operation for its own
grantor cell is lightly loaded at the time of the request, cell edge users. See [5] for detailed simulation
it may change the size of the restricted PRBs in its results.
own cell, thereby enabling users in a neighboring cell The adaptive partial direct reuse with power control
to experience a better SINR over a larger portion of algorithm. An adaptive version of the static PDRPC
the total bandwidth than before. A grantor cell can algorithm can be formulated by adapting the parame-
receive multiple requests simultaneously from neigh- ters a and K. The adaptation would be based upon
boring cells, and hence there has to be a mechanism some measure of the ratio of center cell users to cell
at the grantor to evaluate the priority of the requests. edge users.
As part of this process, every requesting cell sends a The multi-cell gradient algorithm. The multi-cell
soft metric on a per subband basis in its request mes- gradient (MGR) algorithm [9] adjusts the transmit
sage, which is a function of the importance or ben- powers of the different subbands by systematically and
efit of its request. The grantor cell will prioritize its continuously pursuing local maximization of the over-
incoming requests based upon this soft metric and all network utility U, which is the sum of all cell utili-
may also use additional metrics such as application ties U (k). In turn, each cell k utility U(k) is the sum of
type and QoS constraints. Once the grantor cell has utilities Ui(Xi) over all users i in cell k, where Xi are
determined which of the incoming requests it will the users’ average throughputs and Ui are the users’
grant, it sends a grant message back to the appropri- utility functions. This maximization can be done semi-
ate originating cell, which can now take advantage of autonomously by each cell with periodic exchanges
1
∂U(3)
∂Pj(2)
∂U(2)
∂U(3) ∂Pj(3)
∂U(1)
∂Pj(1)
∂Pj(3)
3
MGR—Multi-cell gradient
Figure 8.
Inter-cell coordination of partial derivates for MGR.
between the neighboring interfering cells of a few key j, to each cell m k. It also periodically receives val-
ratios that arise from the optimization approach. An ues of Dj(k,m) for all j from each cell m k, as shown in
important fact to be noted about the MGR and the Figure 8. Cell k maintains the current values of
sector autonomous (SA) algorithms (described in Dkj a D(k,m), for each subband j. Dkj is the estimate
m j
the next section) is that these algorithms are only con- 0U
of the partial derivative (m)
cerned with power allocation (and re-allocation) among 0Pj .
the subbands by each cell, and this is done on a rela- Periodically, cell k does the following:
tively slow timescale. Given the power levels set by Let ∆ 0 be a fixed parameter and denote by
either algorithm, each cell can perform an opportunis- P(k) aj Pj(k) the current total power in the cell. Then
tic propagation channel aware scheduling. No a priori the powers are updated sequentially as follows:
frequency planning is required for this approach. 1. Pick j* (if it exists) such that Djk* is the smallest
Consider a network of K cells and J subbands. among those j with Dkj 0 and Pj(k) 0, and do
W is the bandwidth of one subband and N0 is the
Pj*(k) max5Pj*(k) ¢, 06 (2)
noise power spectral density. Pj(k) is the power allo-
cated in subband j of cell k. 2. If P(k) P*, pick j* (if it exists) such that Dkj* is the
Each cell k constantly adjusts its power allocation largest among those j with Dkj 0 and do
to different subbands in a way that improves the total
utility U of the system. Each cell k maintains the esti- Pj*(k) Pj*(k) min5¢, P * P(k) 6 (3)
mate of the utility U (k) which the cell could poten-
3. If P(k) P* and maxj Dkj 0, pick a pair (j*, j*) (if
tially attain given its current power allocation Pj(k)
it exists) such that Dkj* is the largest, Dkj* is the
among the subbands and current interference level
smallest among those with Pj(k) 0 and Dkj* Dkj*.
from other cells. Cell k maintains estimates of partial
0U(k) Then
derivates Dj(m,k) (m) of its utility on the power levels
0Pj
Pj*(k) max5Pj*(k) ¢, 06 (4)
Pj(m) in all cells m (including self, m k) and all sub-
bands j. Cell k periodically sends values of Dj(m,k) for all Pj*(k) Pj*(k) min5¢, Pj*(k) 6 (5)
SA
being assigned to cell edge users.
400 MGR The sector autonomous algorithm. The SA algo-
rithm [9] adjusts powers in each subband indepen-
350 dently in each cell using a non-trivial heuristic. This
algorithm is completely autonomous, desirable in situa-
300 tions where information exchange between the
relevant cells is not possible. No a priori frequency
250 planning is required for this approach.
A virtual scheduling algorithm (different from
200 that employed by MGR) is run in each cell, which
2 3 4 5 6 7 8 9 10
5% throughput (bits/slot)
tries to selfishly solve an “artificial” optimization prob-
(b) Fast fading
lem. This problem is not the “true” original cell utility
384 bits/slot 1 bit/s/Hz maximization but has the following features:
AWGN—Additive white Gaussian noise
1. It is “highly correlated” with the original prob-
DL—Downlink lem, and
MGR—Multi-cell gradient
2. It inherently “encourages” an uneven power allo-
SA—Sector autonomous
cation to subbands (when such is beneficial).
Assume a cell operates the following way.
Figure 9. Suppose a parameter P0, P*J P0 P*, is fixed. In
DL throughput versus cell edge performance for MGR
and SA algorithms for full buffer best effort traffic. each (virtual) time slot, in each subband j, the cell
either transmits to exactly one of the users i at power
level P0 (and then the transmission rate is Rij, as it is
To estimate and update the values of the partial dependent upon the measured SINR of user i) or does
derivatives Dj(m,k), each cell k continuously runs a vir- not transmit to any user at all (in which case the
tual scheduling algorithm (see [9] for details). power used is 0). Given this setting, a scheduling strat-
Figure 9 shows simulation results for the MGR egy is employed which, over time, solves the follow-
and SA algorithms for an ISD distance of 2.5 km, ing problem:
penetration loss of 10 dB, lognormal (8.9 dB standard Maximize ΣiUi(Xi), where Xi are the users’ aver-
deviation) shadow model, and bandwidth of 1.25 MHz age throughputs and Ui are the users’ utility func-
with six subbands for full buffer traffic with 20 users tions, subject to the constraint on the total average
γ α
β β OI (overload indicator) are per-
PRB quantities exchanged on X2
interface to allow frequency
γ α γ α IoT shaping of interference
β
Sector α 0 1 2 3
γ α
β β 4 5 6 7 8 9 10 11
γ α γ α
β
IoT
γ α Sector β 4 5 6 7
Try to concentrate
0 1 2 3 8 9 10 11
interference in
these interference
bearing zones
IoT
Sector γ 8 9 10 11
0 1 2 3 4 5 6 7
Figure 10.
Uplink static ICIC operation: the interference bearing zone concept.
power, Σj P0j P*, where P0j is the average power (per interference bearing zone to bear the brunt of
virtual slot) allocated in subband j. the interference from neighboring cells. Through
Then, the actual per-subband power levels Pj are event-triggered reporting, the serving cell knows the
set and adjusted to be equal to the average powers P0j , identity of the strongest neighboring cell for each
continuously produced and adjusted by the virtual mobile. For mobile devices at the cell border, the
scheduling. See [9] for a detailed description of the uplink scheduler preference is to assign resources in
virtual scheduler. the IBZ of the mobile’s strongest neighboring cell. If
the scheduler needs to assign the device outside the
Uplink ICIC IBZ, it does so with a reduced transmit PSD level,
In the uplink, it is the UEs near the cell border implemented through an absolute power control com-
which cause most of the interference to adjacent cells. mand in the UL scheduling grant. The concept here is
Figure 10 shows the concept of the interference bear- to concentrate the bulk of the inter-cell interference in
ing zone (IBZ) [5]. The concept here is to designate a a small portion of the total bandwidth, thereby pre-
portion of the bandwidth in each cell known as the venting any impact to the majority of users since the
FPC α 0.8
100
90
80
No ICIC (PF scheduler) FPC α 0.7
70 ICIC inverted reuse-3 (PF scheduler)
ICIC inverted reuse-9 (PF scheduler)
60
3500 4000 4500 5000 5500 6000
Sector throughput (kbps)
Figure 11.
Sector throughput versus cell edge bit rate for UL static ICIC.
interference is now localized to certain sub-carriers For situations where FSS gains can be exploited, there
and the sub-carriers are orthogonal to each other. is no ICIC gain over FSS with universal reuse.
Static ICIC Dynamic ICIC Algorithms
In static schemes, the size and location of the IBZ As seen in the downlink discussion, better per-
are distributed to the different cells and are generally formance can be obtained by adapting the restrictions
constant on a timescale corresponding to days. on a dynamic basis, based on factors such as cell load,
Different kinds of restriction distributions can be used user distributions, traffic distributions, and other
which involve frequency or cell planning in an area, parameters. This improves spectral efficiency globally
resulting in different reuse factors. across cells since lightly loaded cells can help higher
Figure 11 shows simulation results for different load neighbors.
reuse factors. The test configuration is 3GPP case 1 Semi static ICIC. A concept similar to the downlink
[1] with an ISD of 500 meters for a bandwidth of semi static ICIC can be applied to the UL. In the UL, it
10 MHz. Vehicular speed is 120 kph with a typical is the size of the IBZs that is being adjusted dynami-
urban multipath profile. The various points in the cally. A larger IBZ in a particular cell implies that a
curves correspond to different FPC alpha factors. More neighboring cell can now allocate cell edge users in its
details of the simulation parameters and a detailed own cell over a larger portion of the bandwidth,
discussion of the results can be found in [5]. A small thereby providing improved performance.
improvement in edge bit rate is seen with a bigger The multi-sector gradient for uplink algorithm. The
gain for sector throughput. The concentration of multi-sector gradient for uplink (MGR-UL) algorithm
the interference in the IBZ benefits all the users in the [8] continuously tries to improve system-wide utility
cell, leading to an improvement in sector throughput. U, which is the sum of all cell utilities U(k). In turn,
Number of subbands 6
Figure 12.
Sector throughput versus cell edge bit rate for the MGR-UL ICIC algorithm.
Figure 13.
High level concept of UL IoT control.
results in increased inter-cell interference that begins 1. Each eNB must make measurements of the cur-
to impact eNB receiver performance. The point at rent IoT level and signal OIs to its neighbors if the
which this transition occurs is a complex function of IoT level is greater than an IoT threshold.
cell size, propagation channel conditions, and cell 2. Each eNB processes the OIs from neighboring
load. Hence a single value of IoT is not suitable for all eNBs and adapts the open loop power control
deployment scenarios. The optimization of the choice parameters that it is broadcasting to UEs it is serv-
of the IoT level is something that lends itself readily to ing.
SON operation, where various parameters across dif- 3. Inter-eNB communications.
ferent cells can be optimized in a self-organizing and IoT measurements and emitting the IoT overload
self-optimizing manner. indicator. IoT is defined on a per-resource unit basis.
Figure 14 shows the various components and IoT is the total uplink interference measured in a
interactions of a self-organizing IoT control scheme. resource unit divided by the thermal noise estimate
The purple circles represent the basic components of over the bandwidth of a resource unit (12 subcarriers
a SON IoT control scheme while the gray circles rep- 180 kHz). Defining IoT on a per resource unit basis
resent the other major functional areas of the eNB provides flexibility when supporting ICIC techniques
that interact with IoT control. There are three basic where significantly different planned average inter-
components to a SON ICIC implementation: ference levels are used in different parts of the total
1. Core IoT control algorithm, bandwidth.
2. IoT measurement, and A measurement of the total uplink interference
3. Inter eNB communications. plus noise power per resource unit is available from
IoT control performance is dependent upon a num- layer 1 processing. To obtain the thermal noise esti-
ber of factors, as shown in Figure 15. Cell sizes, cell mate, techniques similar to those used in the radio
load, propagation conditions, and other parameters play network controller (RNC) of the UMTS system can
an important role in determining IoT performance. be used.
IoT Control via Adaptation of Open Loop Power Two things must be decided when signaling an
Control Parameters OI: a desired IoT operating point and the bandwidth
To implement IoT control, three basic components over which to make IoT measurements. The desired
are required: IoT operating point is chosen from link budget
IoT control
IoT
core ICIC
measurement
algorithm
Figure 14.
UL IoT control components and interactions.
considerations as well as taking into account the serving cell and the UE’s strongest neighbor cell as
normal system stability considerations required for follows [8]:
systems with power control. Once an average IoT
measurement is available, it is compared to the TARGET_SINR(dB)
1 (1 a)
(9)
desired IoT operating point and an OI is signaled if (PLstrongest-non-serving
PLserving
(dB) (dB)
d )
the measurement is greater than the operating point.
Adaptation of open loop power control parameters. Or another improvement is:
A number of power control schemes can be used for
LTE. FPC [8] is the standard power control technique TARGET_SINR(dB)
1 (1 a)(
DL(dB)
d ) (10)
defined in 3GPP. However, FPC only accounts for
DL(dB) 10 log a a b
ps
serving cell path loss. A simple modification to regu- (11)
is pi
lar FPC that reduces the variance of inter-cell inter-
ference is to set the target SINR of the UE as a where ps is the linear path loss to serving eNB, pi is the
function of the path loss difference between the UE’s linear path loss to eNB i, ΓDL(dB) is the downlink SIR
IoT control
ICIC strategy performance Cell size
QoS Propagation
constraints environment
Cell edge
load
# users
Figure 15.
Determinants of UL IoT control performance.
3 dB FPC_856 m ISD
2
CQI_1520 m ISD
2 dB
1.5 FPC_1520 m ISD
1
40 60 80 100 120
Edge rate (kb/s)
Figure 16.
Performance of UL IoT control for different UL power control algorithms.
overload indicators arrive from neighboring eNBs, and Simulation results. Simulation results show that
the mechanism is to adjust Γ1 as follows: edge throughput has a maximum at some IoT level. In
_StepUp_dB
_StepUp_dB * addition, sector throughput may also have a maxi-
(13) mum at some IoT. This is shown in Figure 16, where
TARGET_IoT_OvershootRate the performances of IoT control with FPC- and CQI-
(1 TARGET_IoT_OvershootRate) based power control methods are compared in sector
Where Γ_Step Down_dB and TARGET_IoT_OvershootRate throughput and edge rate for various IoT targets and
are parameters specified a priori. The target SINR com- ISDs for an AWGN channel. The various points on a
ponent Γ1 is then adjusted using the equations below. curve in the figure correspond to different IoT targets
IF (overload indicators are set), adjust in units of 0.1 dB. Point values are shown only for
the CQI-based method since they are the same for the
1(dB)
1(dB)
_StepDown_dB (14) FPC case. A value of 20 in the figure maps to a value
ELSE of 2.0 dB, and so forth. One can see that the optimal
IoT varies with cell size. At an 856 meter ISD, the best
1 (dB)
1(dB)
_StepUp_dB (15)
IoT level is 9.5 dB. At larger ISDs, the best IoT is less.
These updates in general ensure that the eNBs The FPC- and CQI-based methods provide similar
do not exceed a specified IoT target more than edge performance, but the CQI-based method—
TARGET_IoT_OvershootRate fraction of the time. Con- through better choice of the target SINR—provides
straints are also usually applied on the dynamic range higher sector and user peak throughputs. Results for
that Γ can be adjusted over to ensure that UL SINR the DPL method are not shown since the DPL and CQI
targets are bounded. Variations of this approach such methods yield very similar results when ICIC is not used
as unequal step sizes or adaptive step sizes can also be and both methods are simulated in an ideal fashion
implemented. assuming continuous amplitude and timing knowledge
13 dB
14
11.5 dB 11 dB
12
CQI_428 m ISD
9 dB
7.5 dB CQI_856 m ISD
8
7 dB FPC_856 m ISD
6 6 dB CQI_1520 m ISD
5.5 dB 5 dB
5 dB
FPC_1520 m ISD
4
4 dB
3.5 dB
3 dB
2
2 dB
0
20 40 60 80 100 120
UL edge rate (kb)
CQI—Channel quality indicator based power control ISD—Inter site distance
FPC—Fractional path loss compensation power control LTE—Long Term Evolution
IoT—Interference over thermal UL—Uplink
Figure 17.
Mean IoT levels for different cell sizes for different UL power control algorithms.