Scheduling and Capacity Estimation in LTE
Olav Østerbø
Abstract— Due to the variation of radio condition in LTE the
obtainable bitrate for active users will vary. The two most
important factors for the radio conditions are fading and
pathloss. By including both fast fading and shadowing and
attenuation due to distance we have developed analytical models
to investigate obtainable bitrates for the basic resource unit in
LTE. In addition we estimate the total cell throughput/capacity
by taking scheduling into account. Particularly, the cell
throughput is investigated for three types of scheduling
algorithms; Round Robin, Proportional Fair and Max SINR
(signal-to-interference-plus-noise ratio) were also fairness among
users is part of the analysis. In addition models for cell
throughput/capacity for a mixture of Guaranteed Bit Rate (GBR)
and Non-GBR greedy users are derived.
Numerical examples show that multi-user gain is large for the
Max-SINR algorithm, but also the Proportional Fair algorithm
gives relative large gain relative to plain Round Robin. The Max-
SINR has the weakness that it is highly unfair when it comes to
capacity distribution among users. Further, the model visualize
that use of GBR with high rates may cause problems in LTE due
to the high demand for radio resources for users with low SINR,
i.e. at cell edge. Hence, for GBR sources the allowed guaranteed
rate should be limited.
Index Terms— LTE, scheduling, capacity estimation, GBR.
I. INTRODUCTION
T HE LTE (Long Term Evolution) standardized by 3GPP is
becoming the most important radio access technique for
providing mobile broadband to the mass market. The
introduction of LTE will bring significant enhancements
compared to HSPA (High Speed Packet Access) in terms of
spectrum efficiency, peak data rate and latency. Important
features of LTE are MIMO (Multiple Input Multiple Output),
higher order modulation for uplink and downlink,
improvements of layer 2 protocols, and continuous packet
connectivity [1].
While HSPA mainly is optimized data transport, leaving the
voice services for the legacy CS (Circuit Switched) domain,
LTE is intended to carry both real time services like VoIP in
addition to traditional data services. The mixture of both real
time and non real time traffic in a single access network
requires specific attention, where the main goal is to maximize
cell throughput while maintaining QoS and fairness for both
users and services. Therefore radio resource management will
be a key part of future modern wireless networks.
O. Østerbø is with Telenor, Corporate Development, Fornebu, Oslo,
Norway, (phone: +4748212596; e-mail: olav-norvald.osterbo@telenor.com).
978-0-9836283-0-9 
c 2011 ITC
This paper was peer reviewed by subject matter experts for publication in the Proceedings of ITC 2011
For bandwidth efficiency, where only a single user is
considered, the scheduling is without any significance. This is
not the case when several users are competing for the available
radio resources. The scheduling algorithms studied in this
paper are those only depending on the radio conditions, i.e.
opportunistic scheduling where the scheduled user is
determined by a metrics depending on the SINR (signal-to-
interference-plus-noise ratio). The most commonly known
opportunistic scheduling algorithms are of this type like PF
(Proportional Fair), RR (Round Robin) and Max-SINR. The
methodology developed will, however, apply for general
scheduling algorithms where the scheduling metrics for a user
is a (known) function of SINR [2]. Also the multi user gain
(i.e. the relative increase in cell throughput) due to the
scheduling is of main interest. The proposed models
demonstrate the magnitude of this gain. As for Max SINR
algorithm this gain is expected to be huge, however, the gain
comes always at a cost of fairness among users. Therefore
fairness has to be taken into account when evaluating the
performance of scheduling algorithms.
It is likely that LTE will carry both real time traffic and
elastic traffic. We also analyze scenarios where a cell is loaded
by two traffic types; high priority GBR (Guaranteed Bit Rate)
traffic that requires a fixed data-rate and low priority (greedy)
data sources that always consume the leftover capacity not
used by the GBR sources. This is actual a very realistic traffic
scenario for future LTE networks where we will have a mix of
both real time traffic like VoIP and data traffic.
The remainder of this paper is organized as follows. In
section II the basic radio model is given and models for
bandwidth efficiency are discussed. Section III gives an
outline of the multiuser case where resource allocation and
scheduling is taking into account. Some numerical examples
are discussed in section IV and in section V some conclusions
are given.
II. SPECTRUM EFFICIENCY
A. Obtainable bitrate per symbol rate as function of SINR
For LTE the obtainable bitrate per symbol rate will depend
on the SINR (for both up-and downlink). The actual radio
signal quality is signaled over the radio interface by the so-
called CQI (Channel Quality Indicator) index ranging from 1
to 15. Based on the CQI value the coding rate is determined on
basis of the modulation QPSK, 16QAM, 64QAM and the
amount of redundancy included. The corresponding bitrate per
bandwidth is standardized by 3GPP [3] and is shown in Table
1 below. For analytical modeling the actual CQI measurement
63
 procedures are difficult to incorporate into the analysis due to also the sending frequency). We first consider variations that
time lags, i.e. the signaled CQI is based measurements taken in are slowly varying over time intervals that are relative long
earlier TTIs (Transmission Time Interval). To simplify the compared with the TTIs (Transmission Time Intervals). Then
analyses, we assume that this time lag is set to zero and that the path loss is usually given in [dB] on the form:
the CQI is given as a function of the momentary SINR, i.e. we
simply take CQI=CQI(SINR). This approximation is justified G = 10 L 10 with L = C − A log10 ( r ) + X t (2)
if the time variation in SINR is significantly slower than the
length of a TTI interval. By applying the CQI table found in
where C and A are constants, A typical in the range 20-40,
[3] we get the obtainable bitrate per bandwidth as function of
and X t a slowly variation normal stochastic process with zero
the SINR as the step function:
[ )
B = f c j for SINR ∈ g j , g j +1 ; j = 0,1,...,15 (1)
mean representing the shadowing effects (i.e. slow fading).
The other important component determining SINR is the noise
where f is the bandwidth of the channel, c j is the efficiency and interference. It is common to split the noise and
[
for CQI index j (as given by Table 1) and g j , g j +1 are the) interference power into two terms: N f = N int + N ext where
corresponding intervals of SINR values. (We also define N int is the internal (or own-cell) noise and interference power
c0 = 0 , g 0 = 0 and g16 = ∞ .) and N ext is the external (or other-cell) interference. In a
TABLE 1 CQI TABLE.
CQI index modulation code rate x 1024 efficiency
CDMA (Code Division Multiple Access) network, the lack of
0 out of range orthogonality induces own-cell interference. In OFDMA
1 QPSK 78 0.1523 (Orthogonal Frequency Division Multiple Access) networks,
2 QPSK 120 0.2344 however, there is a perfect orthogonality between frequencies
3 QPSK 193 0.3770 and therefore the only contribution to N int is the thermal noise
4 QPSK 308 0.6016
5 QPSK 449 0.8770 at the receiver. The interference from other cells depends on
6 QPSK 602 1.1758 the location of surrounding base stations and will typically be
7 16QAM 378 1.4766 largest at cell edges. To make the analysis mathematical
8 16QAM 490 1.9141 traceable we shall assume that the external noise and
9 16QAM 616 2.4063
10 64QAM 466 2.7305
interference is constant throughout the cell or negligible, i.e.
11 64QAM 567 3.3223 we assume the noise N f to be constant throughout the cell.
12 64QAM 666 3.9023 Hence, with the assumptions stated above, we may write
13 64QAM 772 4.5234
14 64QAM 873 5.1152 SINR on the form S t h(r , λ ) where S t represent the
15 64QAM 948 5.5547 stochastic variations which we assume to be distance
To fully describe the bitrate function above we also have to independent capturing the slowly varying fading, and h(r , λ )
[ )
also specify the intervals g j , g j +1 . Several simulation studies represent the distance dependant attenuation (which we also
e.g. [4] suggest that there is a linear relation between the CQI may to depend on the sending frequency). Most commonly
index and the actual SINR limits in [dB]. In the numerical used channel models as described above have attenuation that
examples we therefore take SINR j [dB ] = 10 log10 g j = aj + b follows a power law, i.e. we chose to take h(r , λ ) on the form
aj + b
h ( r , λ ) = h( λ ) r α (3)
or g j = 10 10
for some constants a and b . It is also argued
where α = A 10 is typical in the range 2-4 and
that the actual range of the SINR limits in [dB] is determined
h (λ ) =
N − C 10
10 = 10 Z 10 with Z = 10 log10 ( N ) − 10 log10 ( Pw ) − C given
by the following (end point) observations: SINR[dB]=-6 Pw
corresponds to CQI=1, while SINR[dB]=20 corresponds to [dB], where we also indicate that h(λ ) may depend of the
CQI=15. Hence, with this assumption we then have −6 = a + b
(sending) frequency. With the description above the stochastic
and 20 = 15a + b or a = 13 7 and b = − 55 7 . • ln 10
variable S t = 10 X t 10
= e S with S t• =
t
X t , and hence S t is a
B. Radio channel models 10

Generally SINR for a user will be the ratio of the received lognormal process with E[St• ] = 0 and σ = ln 10 σ ( X t ) where
10
signal strength divided by the corresponding noise. The
σ ( X t ) is the standard deviation (given in [dB]) for the normal
received signal strength is the product of the power Pw times
process X t . With these assumptions we have the Probability
path loss G and divided by the noise and interference
Density Function (CDF) and Complementary Distribution
component N f , i.e. SINR = Pw G . Now the path loss G will Function (PDF) of S t as:
Nf
(ln x ) 2
typical be a stochastic variable depending on physical 1 − ~ 1 ln x
s ln ( x) = e 2σ 2 and S ln ( x ) = erfc( ) (4)
characteristics such as rapid and slow fading, but will also 2π σx 2 σ 2
have a component that are dependent on distance (and possible

64 Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011)

 ∞ ∞ ∞ (ln( t x )) 2
where erfc( y ) = 2 ∫e
−x2
is the complementary error ~ ′ 1 −t −

∫ e ts ln (t )dt = ∫e
− xt
dx s su ( x) = − S su ( x) = 2σ 2
dt (9)
π x= y t =0 2π σx t =0
function. and for the PDF we now we take the corresponding truncated
C. Including fast fading integral to be:
T (ln(t x )) 2
1 −t −
There are several models for fast fading in the literature like
Rician fading and Rayleigh fading [5]. In this paper we restrict
s su ( x, T ) =
2π σx t =0
∫e 2σ 2
dt (10)

ourselves to the latter mainly because it describes the fast σ2

−T +
fading variations by a simple negative exponential distribution. In this case we find 0 ≤ s su ( x) − s su ( x, T ) ≤ e as a bound 2

It is possible to include fast fading into the description above. of the truncation error.
To do so we assume that the fast fading effects are on a much By expanding the integral (10) in terms of the exponential
more rapid time scale than the slow fading. We therefore function as above, we now obtain a similar (convergent) series:
assume that the slow fading is constant during the rapid fading ( k +1) 2 σ 2
1 ∞ (−1) k k (k + 1)σ ln( x T )
variations. Hence, condition on the slow fading to be y then s su ( x, T ) = ∑
2 k = 0 k!
x e 2
erfc(
2
+
σ 2
) (11)
for a Rayleigh faded channel the SINR will be exponentially
distributed with mean y h(r , λ ) . We may therefore take SINR D. Distribution of the obtainable bitrate a user located at a
as S t h(r , λ ) where S t = X ln X e is the product of a Log- given distance from the sender antenna
normal and a negative exponential distributed variables. The Below we express the distribution of the obtainable bitrate
corresponding distribution often called Suzuki distribution according to the distribution of the stochastic part of the SINR;
have PDF and CDF given by the integrals: namely S t . From (1) we get the bit-rate Bt (r ) for a channel
∞
1 −
x
~ ∞
−
x occupying a bandwidth f located at distance r as:
s su ( x) = ∫ te t
sln (t )dt and S su ( x) = ∫e t
s ln (t )dt (5)
Bt ( r ) = f c j when S t ∈ [h( r , λ ) g j , h( r , λ ) g j +1 ) ; j = 0,1,...,15
(12)
t =0 t =0

where s ln (t ) is the lognormal PDF above by (4). Since Hence, the DF (Distribution Function) of the bandwidth for a
1 user located at distance r ; B( y, r ) = P( Bt (r ) ≤ y ) may be
s ln ( ) = t 2 s ln (t ) it is possible to express the integrals above in written:
t
terms of the Laplace transform of the Log-normal distribution B ( y , r ) = S ( h ( r , λ ) g j +1 ) for y ∈ ( f c j , f c j +1 ] ; j = 0,1,...,15 (13)
and therefore the CDF (and PDF) of the Suzuki distribution where S (x) is the DF of the stochastic variable fading
may be written as: component. Also we obtain the k ’moment of the obtainable
~
S su ( x) = Sˆ ln ( x) and s su ( x) = − Sˆ ln′ ( x) where bitrate for a user located at a distance r from the antenna as
−t −
(ln(t x )) 2 the (finite) sum:
∞ ∞ 2σ 2
1 e
( )
15
k ~
∫e ∫
− xt
Sˆ ln ( x) = s ln (t )dt =
dt (6)
m k ( r ) = f k ∑ c j − c j −1 S ( h ( r , λ ) g j )
k
(14)
t =0 2π σ t = 0 t j =1
is the Laplace transform of the Log-normal distribution. If we ~
where S ( x) = 1 − S ( x) is the CDF of the stochastic variable
define the truncated transform:
(ln(t x )) 2
fading component.
−t −
T T 2σ 2
~ 1 1 e E. Distribution of the obtainable bitrate for a user that is
S su ( x, T ) = ∫ e −t sln (t x)dt = ∫ dt (7)
x t =0 2π σ t =0 t randomly placed in a circular cell with power-law attenuation
~ ~ Since the bitrate/capacity for a user strongly will depend of
then S su ( x) = lim S su ( x, T ) and further the corresponding
T →∞ the distance from the sender antenna, a better measure of the
error is exponentially small in T . An attempt to expand the capacity will be to find the distribution of bitrate for a user that
integral (6) in terms of the series of the exponential function is randomly located in the cell. This is done by averaging over
∞
( −1) k t k
e −t = ∑ yields a divergent series; however, this is the cell area and the distribution of the corresponding
k =0 k! 1
A ∫A
averaging bitrate Bt is given as B ( y ) = B( y, r )dA(r ) where
not the case for the truncated transform (7). We find the
following series expansion: A is the cell area. For circular cell shape with radius R , and
k σ 2 2 power law attenuation on the form h(r , λ ) = h(λ )r α the
~ 1 ∞ (−1) k k kσ ln( x T )
S su ( x, T ) = ∑ x e ) 2
(8) erfc( + corresponding integrals may be partly evaluated. By defining
2 k =0 k! 2 σ 2 an α -factor averaging random variable S α with DF
Similar the PDF of the Suzuki random variable may be
S α ( x) = P ( S α ≤ x) defined by
found from (6) by differentiation:
2 x 2 1 2
2 − −1 2 −1
S α ( x) =
α
x α
∫ t α S (t )dt =
t =0
α ∫ t α S (tx)dt
t =0
(15)

Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011) 65

and with PDF and we take the truncated approximation of the PDF as the
2
2
− −1
x 2
2
1 2 integral:
sα ( x ) =
α
x α

t =0
∫ t α s(t )dt = α t =∫0t α s(tx)dt (16)
T

∫t
−1 −t
s su α ( x, T ) = e s ln α ( x t )dt (23)
the bitrate distribution will have exact the same form as (13),
t =0
and with moments given by (14) by changing r → R and
and we find the following error bound:
S ( x) → Sα ( x) (and s ( x) → sα ( x) ) in the formulas.
σ2
−T +
1) Distribution of the stochastic variable S α for Log-normal 0 ≤ s su α ( x) − s su α ( x, T ) ≤ e 2 . By the similar approach as for
and Suzuki distribution the CDF we also find the following series expansion of the
Based on the definition we may derive the CDF and PDF of truncated PDF:
stochastic variable S α for the Log-normal and Suzuki
∞
(−1) k x k ( k +1 ) 2 σ 2
(k + 1)σ ln( x T )
distributed fading models. For the Log-normal distribution we s su α ( x, T ) = ∑ e 2
erfc( + )+
x k = 0 ( 2 + ( k + 1)α ) k! 2 σ 2
~ 1 2 −1 ln t (24)
have S ln α ( x) = 2 ∫t α
erfc(
σ 2
)dt . By changing
e
2σ 2
α2 2 − (1+ 2 ) 2σ 2 − α ln( x T )
α x α t =0 γ( + 1, T ) x α
erfc( )
α α ασ 2
variable according to y = ln t in the integral we find:
1⎛ 2σ 2σ 2 − α ln x ⎞
2
~ ln x −2 III. ESTIMATION OF CELL CAPACITY
S ln α ( x) = ⎜⎜ erfc( )+ x αe α2
erfc( ) ⎟⎟ (17)
2⎝ σ 2 ασ 2 ⎠ In the following we assume that the cell is loaded by two
and further the PDF is found by differentiation: traffic types:
1 −( 2 +1) 2σ 2σ 2 − α ln x
2

s ln α ( x ) = x α e α erfc( ) (18)
2
• High priority GBR traffic sources that each requires
α ασ 2 to have a fixed data-rate and
For the Suzuki distribution we have the CDF given by the
∞
• low priority (greedy) data sources that always
integral S~su ( x) = x ∫ t − 2e − t sln ( x t )dt and therefore we have: consumes the leftover capacity not used by the GBR
t =0 traffic.
1 2 ∞
~ 2 −1 ~
This is actual a very realistic traffic scenario for future LTE
∫ t α S su ( xt )dt = x ∫ t e sln α ( x t )dt
− 2 −t
S su α ( x ) = (19)
α t =0 t =0
networks where we actual will have a mixture of both real time
where sln α ( x) is given by (18) above for the Lognormal traffic like VoIP and elastic data traffic. Below, we first
estimate the Resource Block (RB) usage of the high priority
distribution. As for the Suzuki distribution approximation to GBR traffic, and then we may subtract the corresponding RBs
any accuracy is possible to obtain by truncating the integral to find the actual numbers of RBs available for the (greedy)
above: data traffic sources. Then finally we estimate the cell
T
~ throughput/capacity as the sum of the bitrates offered to the
S su α ( x, T ) = x ∫ t −2 e −t s ln α ( x t )dt (20)
t =0
GBR and (greedy) data sources.
and also for this case we find that the truncation error is A. Estimation of the capacity usage for GBR sources in LTE
∞
(−1) k t k
exponentially small in T . By expanding e − t = ∑ and The reservation strategy considered simply reserve recourses
k =0 k! on a per TTI bases and allocate RBs so that the aggregate rate
integrating term by term we find: equals the required GBR rate, i.e. using Non-Persistent
scheduling.
~ ∞
( −1) k x k k 2σ 2 kσ ln( x T )
S su α ( x, T ) = ∑ e 2
erfc( + )+ 1) Capacity usage for a single GBR source
k = 0 ( 2 + kα ) k! 2 σ 2 We first consider the case where we know the location of the
2σ 2 (21)
e α2
2 −2 2σ 2 − α ln( x T ) GBR user in the cell, i.e. is located at a distance r from the
γ ( ,T )x α
erfc( ) antenna. We take B as the bitrate obtainable for a single RB
α α ασ 2
x and consider a GBR source that requires a fixed bit-rate of
where γ (a, x) = ∫ t a −1 e −t dt is the incomplete gamma function. b CBR . We assumes that this is achieved by offering n RBs for
t =0
every k -TTI interval. A simple way of reserving resources to
(Observe the similarity with the corresponding expansion for
~ n
S su ( x) by (8).) GBR connections is to allocate RBs so that B will be close
k
The corresponding integral for the PDF is given by: to the required rate b CBR over a given period. We take
n
∞ N CBR = to be the number of RBs granted for a GBR
∫t
−1 − t
s su α ( x) = e s ln α ( x t )dt (22) k
t =0 connection in a TTI as (the stochastic variable):

66 Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011)

 ⎧α bCBR 1) Scheduling of based on metrics
⎪ if CQI > 0
NCBR =⎨ B (25) First we consider the case with K users that are located in
⎪0 if CQI = 0 the cell with distances (from the sender antenna) given by a
⎩
where we have introduced a scaling factor α so that on the vector r = (r1 ,...., rK ) and we assume that the user scheduled in
long run we obtain the desired GBR-rate b CBR . By choosing a TTI is based on:
~ i schedul = arg max{M i }
α = p CQI −1 where pCQI = P(CQI > 0) = S (h(r , λ ) g1 ) then (32)
i =1,.., K

E [N CBR B ] = b CBR and hence we also have: where M i = M i (r ) is the scheduling metric which also may
depend on the location of all users (through the location vector
[
E N CBR CQI > 0 = ] b CBR
p CQI
[
E B −1 CQI > 0 ] (26) r = ( r1 ,...., rK ) ). Hence, for the scheduler to choose a user i ,
the metric M i must be larger than all the other metrics (for the
The mean numbers of RBs is therefore:
other users), i.e. we must have M i > U i where
β = β (r , b CBR ) = b CBR m −CQI
1 (r ) (27)
U i = max M k
where the conditional moments m kCQI (r ) = E B k CQI > 0 is [ ] k =1,.., K
k ≠i
(33)
found as Since we assume that a user is granted all the RBs when
fk
(
⎛ k~
) ⎞
scheduled, this gives the cell throughput when user i is
15
~
CQI
m (r ) = ~ ⎜ c1 S (h(r , λ ) g1 ) + ∑ c j k − c j −1k S (h(r , λ ) g j ) ⎟
(28)
k ⎜
S (h(r , λ ) g1 ) ⎝ ⎟
scheduled (located at distance ri ) to be NB (ri ) , where B (ri )
j =2 ⎠
Note that by conditioning on having CQI > 0 we exclude the is the corresponding obtainable bit-rate per RB. Hence, cell
users that are unable to communicate due to bad radio bit-rate distribution may then be written as:
conditions, and avoid the problems due to division of zero in K

the calculation of the mean of 1 B . For circular cells and Bg ( y, r ) = ∑ Bi ( y, r ) where (34)
i =1
power law attenuation we obtain the corresponding result as Bi ( y, r ) = P( NB (ri ) ≤ y, M i (r ) > U i (r )) (35)
above by changing r → R and S ( x) → Sα ( x) .

2) Estimation of RBs usage for several GBR sources is bitrate distribution when user i is scheduled. Unfortunately,
In the following we first estimate the RB usage for a fixed in the general case, explicit expression of the probabilities
number of M GBR sources located at distances r j from the Bi ( y, r ) is difficult to obtain, mainly due to the involvement
antenna with bit-rate requirements b CBR
j ; j = 1,..., M . The of the scheduling metrics. However, for some cases of
particular interest closed form analytical expression is possible
total usage of RBs β CBR will be the sum the individual to obtain. For many scheduling algorithms the scheduling
contribution from each source as given by (27): metrics are only function of the SINR for that particular user
M
(and does not depend of the SINR for the other users) and for
β CBR = ∑ β (rj , bCBR
j ) (29)
j =1
this case extensive simplification is possible to obtain. In the
For the case with random location the expression gets even following we therefore assume that the metrics M i only are
simpler: functions of their own SINRi and the location ri of that
M particular user, i.e. we have M i = M ( Si , ri ) , where we (for
β CBR = β ( R, ∑ b CBR
j ) (30) simplicity) also assume that M ( x, ri ) is an increasing function
j =1

i.e. we may add the GBR rates from all the sources in the cell. of x with an unlikely defined inverse function M −1 ( x, ri ) .
The corresponding throughput for the GBR sources is taken as The distribution functions for M i and U i = max M k are then
k =1,.., K
the sum of the individual rates i.e. k ≠i
M
b CBR = ∑ b CBR
j (31) M i ( x, ri ) = P ( M i ≤ x) = S ( M −1 ( x, ri )) and (36)
j =1 K
U i ( x, r ) = P (U i ≤ x) = ∏ S (M −1
( x, rk )) (37)
B. Estimation of the capacity usage for a fixed number of k =1,k ≠i

greedy sources If we now condition on the value of S i = x in (35), we find

In the following we shall estimate the capacity usage for a the distribution of the cell capacity when user i is scheduled
fixed number of greedy sources by assuming: as:
• There are totally K active (greedy) users that are ∞
y
placed randomly in the cell and always have traffic to Bi ( y, r ) =
x
∫=0 P( B(r ) ≤ N S
i i = x)U i ( M ( x, ri ), r ) s ( x)dx (38)
send, i.e. we consider the cell in saturated condition.
By using (12) as the obtainable bit-rate per RB we find:
• There are totally N available RBs and the scheduled
user is granted all of them in a TTI interval.

Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011) 67

 h ( ri , λ ) g j +1
commonly known scheduling algorithms, namely Round
Bi ( y , r ) = ∫ Fi ( x, r ) s ( x)dx if y N ∈ f c j , f c j +1 ( ]; j = 0,1,...,15 (39) Robin (RR), Proportional Fair (PF) and Max SINR by
x =0

where we now have defined the multiuser “scheduling” applying the cell capacity models described above.
function Fi ( x, r ) by:
a) Round Robin
K
For the Round Robin algorithm each user is given the same
Fi ( x, r ) = U i ( M ( x, ri ), r ) = ∏ S (M
k =1,k ≠i
−1
( M ( x, ri ), rk )) (40)
amount of bandwidth and hence this case corresponds to
Finally, by assuming that all users are randomly located taking K = 1 , i.e. the results in section II may be applied by to
throughout the cell the corresponding bit-rate distribution is find the cell capacity with f → Nf and S ( x) → S su ( x) and
found by performing a K -dimensional averaging over all also Sα ( x) → S su α ( x) .
possible distance vectors r over the cell;
1
Bg ( y ) = K ∫ K ∫ r1 K rK Bcell ( y, r ) dA1 L dAK , where A here b) Proportional Fair (in SINR)
A A A Normally, the shadowing is varying over a much longer time
is the cell area. Due to the special form of the function scale than the TTI intervals, and hence we may assume that the
K slow fading is constant during the updating of the scheduling
Fi ( x, r ) = ∏ S (M
k =1, k ≠ i
−1
( M ( x, ri ), rk )) the “cell averaging” over
metric M i and therefore should only account for the rapid
the K −1 dimension variables r1 ,.., ri−1 , ri +1 ,..., rK (not fading component. This means that the shadowing effect may
) be taken as constant that may be included in the non varying
including the variable ri ) yields the product S (M ( x, ri )
K −1
[ ] part of the SINR over several TTI intervals. Hence, we take
where
SINR as St h(r , λ ) where S t = zX e conditioned that the
) 1
S ( y ) = ∫ uS ( M −1 ( y, u ))dA(u ) (41) shadowing X ln = z . By assuming X ln = z is constant over the
AA
short TTI intervals, the scheduling metrics will be
Hence, for the case when user i is located at distance r and
zX e h( ri , λ ) Xe
all the K − 1 other users located at random, then we find Mi = = . In the final result we then
zE[ X e ] h( ri , λ ) E[ X e ]
Bi ( y, r ) = B( y, r ) is independent of i and:
h ( ri , λ ) g j +1 “integrate over the Log-normal slow fading component”. We
B( y, r ) = ∫ [S) (M ( x, r ))] K −1
(
s ( x) dx if y N ∈ f c j , f c j +1 ; ] find that the probability of being scheduled is p(r ) = 1 and
x=0 (42) K
j = 0,1,...,15 , that the conditional bandwidth distribution for a user at located
) at distance r (and the K − 1 users random located) is given
∫ [S (M ( x, r ))]
∞
K −1
and moreover p( r ) = lim Bi ( y , r ) = i s ( x) dx is by the results in section II.D with f → Nf and
y →∞
x=0
S ( x) → S K ( x) with:
the probability that a particular user is scheduled. ∞ ∞
x and sK ( x) = K x x
For circular cell size the cell bit-rate distribution is therefore S K ( x) = ∫ S (t ) ∫ Se ( ) K −1 se ( ) sln (t )dt
K
e sln (t )dt (47)
given as: t =0 t =0
t t t
h ( r , λ ) g j +1
) where S e ( x) = 1 − e − x and s e ( x) = e − x .
[ ] ];
R
Bg ( y) =
2
∫r ∫ K S ( M ( x, r ))
K −1
(
s ( x ) dxdr if y N ∈ f c j , f c j +1
R2 r =0 x =0 (43)
Further the distribution of the cell capacity is given by the
j = 0,1,...,15
results in section II.E with f → Nf and further the
where we now have
corresponding α -averaging is given by the integrals:
) 2
R
~ ∞
~
∫ uS (M
−1
S ( y) = 2 ∫ KS (t )
K −1
( y, u ))du (44) S K α ( x) = se (t ) Slnα ( x ) dt and (48)
R
e t
r =0 t =0

The moments of the cell capacity (when the users are ∞

located according to the vector r = ( r1 ,...., rK ) may be written ∫ KS (t )
K −1
sK α ( x) = se (t )t −1slnα ( x )dt (49)
e t
t =0
as:
h ( ri , λ ) g j +1
K 15
E [ B g (r ) k ] = f k N k ∑ ∑ c j ∫ F ( x , r ) s ( x ) dx
k
i (45) c) Max SINR algorithm.
i =1 j =1 x = h ( ri , λ ) g j
For this algorithm the scheduling metric is M i = S i h( ri , λ ) .
The corresponding moments for cell capacity when the users By assuming circular cell size and radio signal attenuation on
are randomly located in a circular cell is given by:
the form h(r , λ ) = h(λ )r α gives:
⎧⎪ h ( r , λ ) g ) ⎫⎪
[ ]
R j +1
2 f kNk 15
K −1
E [ Bg ] =
k

R2
∑ c j ∫ r ⎨ ∫ K S ( M ( x, r ))
k
s ( x ) dx ⎬ dr (46) ) 2 R
j =1 r =0 ⎪⎩ x = h (r ,λ ) g j ⎪⎭ S ( M ( x, r )) = 2
R ∫ uS ( x (u
r =0
r )α )du = Sα ( x ( R r )α ) (50)
2) Examples
Below we consider and compare three of the most We find that the probability of being scheduled

68 Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011)

 ∞
)
where β = β ( r , b CBR ) is given by (27) and further S ( M ( x, r ))
∫ [Sα ( x ( R ]
α K −1
p(r ) = r) ) s( x)dx (51)
x =0 is defined by (44). Observe that the GBR traffic only will
M
and that the conditional bandwidth distribution for a user
located at distance r (and the K − 1 other users random
affect the cell throughput by the sum ∑b
j =1
CBR
j of the rates and
located) is given by the results in section II.D with f → Nf
not the actual number of GBR sources.
and S ( x) → S c ( x, r ) with:

IV. DISCUSSION OF NUMERICAL EXAMPLES

[ ]
x
1 K −1
S c ( x, r ) = ∫
p ( r ) y =0
Sα ( y ( R r )α ) s( y ) dx (52)
In the following we give some numerical example of
It turns out that extensive simplifications occur for the case downlink performance of LTE based on the models described
in sections II and III.
where all the users are randomly located in the cell and we find
TABLE 2 INPUT PARAMETERS FOR THE NUMERICAL CALCULATIONS
that the distribution of the cell capacity is given by the results Parameters Numerical values
in section II.E with f → Nf and further the α -averaging is Bandwidth per Resource Block 180 kHz=12x 15 kHz
given by taking Sα ( x) → Sα ( x) K i.e. is simply the K ’th Total Numbers of Resource Blocks 100 RBs for 2Ghz
power of the α -averaging of S ( x ) . Distance-dependent path loss. (The L=C +37.6log10(r),
actual model is found in [6].) r in kilometers and
C. Combining real-time and non real time traffic over LTE C=128.1 dB for 20MHz
We are now in the position to combine the analysis in Lognormal Shadowing with 8 dB (in most of the
sections III.A and III.B to obtain complete description of the standard deviation cases)
resource usage in a LTE cell. The combined modeling is based Rayleigh fast fading
on the following assumptions: There are M GBR sources Noise power at the receiver -101 dBm
applying the allocation described in section III.A and K Total send power 46.0 dBm=(40W)
active (greedy) data sources which always have traffic to send, Radio signaling overhead 3/14
i.e. we consider the cell in saturated conditions. Further the 0.8

number of available RBs is taken to be N . Located at cell

0.7
Since the GBR sources have priority over the data sources, edge
they will always get the number of RBs they need and hence
Mean throughput per RB @MbitêsD

0.6

the leftover RBs will be available for the Non-GBR data

0.5
sources. By conditioning on the RB usage of the GBR sources Random
we may apply all the results derived in section III.B with location
0.4

available RBs taken to be the leftover RBs not used by the

GBR sources. Then we may find the average usage of RBs for 0.3

the GBR traffic as done in section III.A. 0.2

We consider first the case where the location of the sources

0.1
is given; where the GBR sources are located at distances s j
from the antenna with bit-rate requirements bCBR
j , j = 1,..., M , 1 2 3 4 5
Distance @Km D

and the greedy data sources are located at distance ri ,

Figure 1: Mean throughput right and std. left per RB for a user random
i = 1,..., K . located, and fixed located as function of cell radius with 2 GHz sending
With these assumptions the mean cell throughput is given frequency and Suzuki distributed fading with std. of fading σ=0dB, 2dB,
as: 5dB, 8dB, 12dB from below.
M K 15
h ( ri , λ ) g j +1
M
In Figure 1 we have depicted the mean bitrate per RB as
Bcell = f ( N − ∑ β ( s j , b CBR )) ∑ ∑ c j ∫ Fi ( x, r ) s ( x )dx + ∑ b j
CBR
function of the cell radius. The maximum bitrate is just below
j =1
j
i =1 j =1 x = h ( ri , λ ) g j j =1
(53)
0.8 Mbit/s for excellent radio conditions. The mean bitrate per
RB have decreased to 0.1 Mbit/s for cell sizes of
where β (r , bCBR ) is given by (27) and further Fi ( x, r ) is
approximately 2 km and shadowing std. equal 8 dB and when
defined by (40). For circular cells and power law attenuation users are random located. The corresponding bit-rate for users
on the form h( r , λ ) = h(λ )r α and randomly placed sources the at the cell edge is approximately 0.04 Mbit/s.
corresponding cell throughput is found to: As seen from Figure 2 the Max-SINR algorithm will over
⎧⎪ h ( r ,λ ) g ) ⎫⎪
perform the PF algorithm when it comes to cell throughput.
[ ]
M 15 R j +1
2 K −1
Bcell = f ( N − β ( R , ∑ b CBR
j )) ∑ c ∫ r⎨ ∫ j K S ( M ( x , r )) s ( x ) dx ⎬dr But if we consider fairness among users the picture is complete
j =1 R2 j =1 ⎪ ⎪⎭
λ r =0 ⎩ x=h ( r , ) g j

M (54) different. When considering the performance of users located

+ ∑ b CBR
j at cell edge the Max-SINR algorithm actually performs very
j =1

badly. While PF give equal probability of transmitting in a TTI

Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011) 69

 80 80

for all active users, the Max-SINR strongly discriminate the 70 70

user close to cell edge.

Cel l capa city @MbitêsD

Cel l cap aci ty @Mbi têsD

60 60

50 50
GBR=1 Mbit/s
80 40 -- Non-Persistent,
40 cell edge

30 -- Non-Persistent,
30 random

70
20 GBR=3 Mbit/s -- mean PF 10 users
20

-- PF 10 10

60 -- Max-SINR 0.5 1
Distance
1.5
@Km D
2 2.5 0.5 1
Distance
1.5
@Km D
2 2.5

-- RR 80 80
Cell capacity @MbitêsD

70 70
50

Cel l ca pa ci ty @Mb itêsD

Ce ll capa cit y @Mbi têsD

60 60
GBR=0.3 Mbit/s GBR=0.1 Mbit/s
50 50
40
40 40

30 30

30 20 20

10 10

20
0.5 1 1.5 2 2.5 0.5 1 1.5 2 2.5
Distance @Km D Distance @Km D

10
Figure 3: Mean cell throughput for PF, 10 users and a GBR user of 3.0, 1.0,
0.3, 0.1 Mbit/s using non-persistent scheduling, for 2 GHz and 100 RB and
1 2
Distance @Km D
3 4 5 Suzuki distributed fading with std. σ=8dB. Red curves corresponds to
random location and blue for user located at cell edge.
Figure 2: Cell capacity as function of cell radius for Max-SINR (red), PF
(blue) and RR (black) scheduling, 2GMHz frequency with 100 RB and with
Suzuki distributed fading with std. σ=8 dB. The number of users is 1, 2, 3, 5,
V. CONCLUSIONS
10, 25, 100 from below. By extensive analytical modeling where both fading and
As seen from Table 3 below, for the case with 10 active attenuation due distance is included we obtain performance
users in a cell, the PF fairly give each user 10% chance of models for:
accessing radio resource while the Max-SINR only give 1.2% • Spectrum efficiency through the bitrate distribution
chance of accessing the radio resources if a user is located at per RB for users that are either randomly or located at
cell edge. a particular distance in a cell.
TABLE 3: PROBABILITY THAT A USER IS SCHEDULED AS FUNCTION OF
NUMBERS OF USERS AND LOCATION FOR PF AND MAX-SINR SCHEDULING • Cell throughput/capacity and fairness by taking the
ALGORITHMS, AND SUZUKI DISTRIBUTED FADING WITH STD. OF 8DB. scheduling into account.
Number PF MAX-SINR
of users • Cell throughput/capacity for a mix of GBR and Non-
r/R=1 r/R=0.5 r/R=0.25 r/R=0.1 GBR (greedy) users.

2 0.50 0.308708 0.594756 0.82579 0.96119 The usage of GBR with high rates may cause problems in
LTE due to the high demand for radio resources if users have
3 0.33 0.147869 0.414839 0.71126 0.92784 low SINR i.e. at cell edge. For GBR allocation the allowed
5 0.20 0.055113 0.245871 0.56102 0.87130
guaranteed rate should be limited. It seems that a limit close to
1 Mbit/s will be a good choice.
10 0.10 0.012690 0.104912 0.36531 0.76418
REFERENCES
25 0.04 0.001356 0.025222 0.16326 0.56989
[1] H. Holma and A. Toskala, Eds.: “LTE for UMTS, OFDMA
100 0.01 0.000019 0.001293 0.02453 0.24325 and SC-FDMA Based Radio Access”, Wiley, 2009.
[2] H. Kushner and P. Whiting: “Asymptotic Properties of
In Figure 3 we consider the cases where 10 greedy users are Proportional-Fair Sharing Algorithms,” Proc. of 2002
Allerton Conference on Communication, Control and
scheduled by the PF algorithm together with a GBR user with
Computing, Oct. 2002.
guaranteed rate of 3, 1, 0.3 or 0.1 Mbit/s. We consider the
cases where either the GBR user is located at cell edge or have [3] 3GPP TS 36.213 V9.2.0 (2010-06), Physical layer
random location throughout the cell. We observe that thin procedures, Table 7.2.3-1: 4-bit CQI Table.
GBR connections do not have big impact on the cell
[4] C, Mehlfuhrer, M. Wrulich, J. C. Ikuno, D. Bosanska, M.
throughput, and it seems that GBR bearers up to 1 Mbit/s Rupp: “Simulating the Long Term Evolution Physical
should be manageable without influencing the cell Layer”,17th European Signal Processing Conference
performance very much. Higher GBR rates, however, (i.e. 3 (EUSIPCO 2009); Glasgow, Scotland, August 24-28, 2009.
Mbit/s) will reduce the cell throughput by a quite large factor
[5] B. Sklar, “Rayleigh Fading Channels in Mobile Digital
especially if the user is located near the cell edge.
Communication Systems Part I: Characterization and Part II:
Mitigation”, IEEE Communications Magazine, July 1997.

[6] 3GPP TSG-RAN1#48, R1-070674: “LTE physical layer

framework for performance verification”, Feb. 12-16 Feb.,
2007, St. Louis, MI, USA.

70 Proceedings of the 2011 23rd International Teletraffic Congress (ITC 2011)