Professional Documents
Culture Documents
System Level Simulation of LTE Networks: June 2010
System Level Simulation of LTE Networks: June 2010
System Level Simulation of LTE Networks: June 2010
net/publication/224149232
CITATIONS READS
345 4,425
3 authors:
Markus Rupp
TU Wien
584 PUBLICATIONS 7,835 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Mobile channel analysis and modelling in millimeter wave band View project
All content following this page was uploaded by Josep Colom Ikuno on 05 June 2014.
Abstract—In order to evaluate the performance of new mobile The LTE system-level simulator [1] supplements an already
network technologies, system level simulations are crucial. They freely-available LTE link-level simulator [7]. This combination
aim at determining whether, and at which level predicted link allows for detailed simulation of both the physical layer
level gains impact network performance. In this paper we present
a M ATLAB computationally efficient LTE system level simulator. procedures to analyze link-level related issues and system-level
The simulator is offered for free under an academic, non- simulations where the physical layer is abstracted from link
commercial use license, a first to the authors’ knowledge. The level results and network performance is investigated.
simulator is capable of evaluating the performance of the Down- The license under which the simulators are published al-
link Shared Channel of LTE SISO and MIMO networks using lows for academic research and a closer cooperation between
Open Loop Spatial Multiplexing and Transmission Diversity
transmit modes. The physical layer model is based on the post- different universities and research facilities. In addition, devel-
equalization SINR and provides the simulation pre-calculated oped algorithms can be shared under the same license again,
”fading parameters” representing each of the individual inter- facilitating the comparison and cross validation of algorithms
ference terms. This structure allows the fading parameters to be and results and making them more credible.
pregenerated offline, vastly reducing computational complexity The LTE system-level simulator implementation offers a
at run-time.
high degree of flexibility. For the implementation, extensive
I. I NTRODUCTION use of the Object-oriented programming (OOP) capabilities of
The Long Term Evolution (LTE) standard, specified by M ATLAB, introduced with the 2008a Release have been made.
the 3rd Generation Partnership Project (3GPP) in Release 8, Having a modular code with a clear structure based in
defines the next evolutionary step in 3G technology. LTE offers objects results in a much more organized, understandable and
significant improvements over previous technologies such as maintainable simulator structure in which new functionalities
Universal Mobile Telecommunications System (UMTS) and and algorithms can be easily added and tested.
High-Speed Packet Access (HSPA) by introducing a novel This paper is organized as follows: in Section II we describe
physical layer and reforming the core network. The main the overall structure of the LTE system-level simulator. In
reasons for these changes in the Radio Access Network Section III we show how the physical layer has been abstracted
(RAN) system design are the need to provide higher spectral in the link measurement model. Afterwards, we present the
efficiency, lower delay, and more multi-user flexibility than the link performance model in Section IV, and Section V presents
currently deployed networks [2]. the main uses of the simulator as well as some conclusions.
In the development and standardization of LTE, as well
as the implementation process of equipment manufacturers, II. S IMULATOR OVERVIEW
simulations are necessary to test and optimize algorithms and While link-level simulations are suitable for developing
procedures. This has to be performed on both, the physical receiver structures [8], coding schemes or feedback strate-
layer (link-level) and in the network (system-level) context. gies [9], it is not possible to reflect the effects of issues
While link-level simulations allow for the investigation such as cell planning, scheduling, or interference using this
of issues such as Multiple-Input Multiple-Output (MIMO) type of simulations. Simulating the totality of the radio links
gains, Adaptive Modulation and Coding (AMC) feedback, between the User Equipments (UEs) and eNodeBs is an
modeling of channel encoding and decoding [3] or physical impractical way of performing system level simulations due
layer modeling for system-level [4], system-level simulations to the vast amount of computational power that would be
focus more on network-related issues such as scheduling [5], required [10]. Thus, in system-level simulations the physical
mobility handling or interference management [6]. layer is abstracted by simplified models that capture its essen-
Along with the standardization process, commercially avail- tial characteristics with high accuracy and simultaneously low
able LTE simulators have been developed. Equipment vendors, complexity.
to this effect, have also implemented their own, proprietary Figure 1 depicts a schematic block diagram of the LTE
solutions. Some universities and research centers have also system-level simulator. Similarly to other system-level sim-
developed such simulators, but to the authors’ knowledge none ulators, the core part consists of: (i) a link measurement
with publicly available source code. model [11] and (ii) a link performance model [12].
macroscopic pathloss [dB] shadow fading [dB]
network layout 1000 140 40
800
base-station deployment 130 30
antenna gain pattern interference macro-scale fading 600
20
tilt/azimuth structure antenna gain 400 120
shadow fading
y pos (m)
10
200
110
link-measurement 0 0
micro-scale fading
mobility management model −200 100
−10
−400
precoding 90 −20
−600
traffic model 80 −30
−800
−1000 −40
70
resource scheduling −1000 −500 0 500 1000 −1000 −500 0 500 1000
x pos (m) x pos (m)
strategy
in a realistic way a two-dimensional Gaussian process with x̃ = (Ĥ0 F )+ (H0 F )x0 + (Ĥ0 F )+ n + (Ĥ0 F )+ (Hi F )xi ,
appropriate spatial correlation is desirable [18]. For our mod- i=1
eling, a low-complexity method capable of introducing space
where ’+’ denotes the pseudoinverse. Denoting A =
correlation into the Gaussian process while still preserving its
(ĤF )+ (HF ), B = (HF )+ and Ci = (Ĥ0 F )+ (Hi F ), and
statistical properties as well as inter-site correlation has been
denoting the matrix elements as aij , A[i, j] we can express
used [19].
the SINRi for the symbols received in layer i as
Figure 2 (right) depicts the resulting space-correlated
shadow fading map for a given eNodeB. A UE traversing the
ROI will experience a slowly changing pathloss (LS,b0 ,u0 ) due |aii |2 Pi
SINRi = ν N ν
to shadow fading, LS,b0 ,u0 being correlated with LS,bi ,u0 , i = X X Xint X
1, . . . , Nint . Thus avoiding the unrealistic simulation scenario |aij |2 Pj + σ 2 |bik |2 + |cl,im |2 Pl,m
where spatially close UEs would have uncorrelated shadow j6=i k=1 l=1 m=1
(1)
fading losses.
Where Pi is the power received at layer i after macro
and shadow fading losses and σ 2 the receiver noise, assumed
C. Channel modeling uncorrelated. Assuming a homogeneous power distribution
While the losses caused by the macroscopic pathloss and Pl = Ptx /ν, we defined Pthe following fading parameters:
the shadow fading are position-dependent and time-invariant, ζ = |aii |2 and ξ = |a
j6=i ij | 2
, which model channel
small-scale fading is modeled as a time-dependent process. estimation errors
Pν (A = I ν×ν for perfect channel knowl-
2
For each of the modeled MIMO transmission modes edge), ψ = |b
k=1 ik | , which
Pν models the ZF receiver
2
(Transmission Diversity (TxD) and Open Loop Spatial noise enhancement, and θ = m=1 |cl,im | , modeling the
Multiplexing (OLSM)), a model based on a simple Zero interference. We can then express SINRi,u for UE u as
Forcing (ZF) receiver has been developed. As of this version,
systems with two transmit antennas have been modeled, but ζi LM,0,u LS,0,u Pl
SINRi,u = (2)
the derived SINR expressions can be easily extended for the N
Xint
LTE transmit modes using four antenna ports. Based on the ξi Pl + ψi σ 2 + θi,l LM,l,u LS,l,u Pl,m
derived models, a trace of fading parameters modeling the l=1
time-and-frequency variant behavior of the channel has been Where LM,bi ,u and LS,bi ,u represent the macro and shadow
generated. These fading-parameters furthermore allow for a fading pathlosses between the UE u and its attached eNodeB
generation prior to the system level simulation itself, which (for bi = 0) and its interferers (bi = 1, . . . Nint ) respectively.
reduces the run-time computational complexity significantly.
The channel modeling aims at computing a per-layer SINR. 2) MIMO TxD modeling: For TxD, the precoding operation
In LTE, a spatial layer is the term used for the different for the two TX antenna case uses the Alamouti scheme [21],
streams generated by spatial multiplexing. A layer thus can be [22], which can be written as
described as a mapping of symbols onto the transmit antenna
ports. Each layer is then identified by a (precoding) vector of
(0)
h(1)
y0 h x n
size equal to the number of transmit antenna ports [20]. = · 0 + 0
y1∗ h(1)∗ −h(0)∗ x1 n1
| {z } | {z } | {z } | {z }
ỹ H̃ x̃ ñ
1) MIMO OLSM modeling: The LTE OLSM MIMO ¸ trans-
mission mode consists of a precoding for Spatial Multiplexing Where h(0) and h(1) contain the channel coefficients from
(SM) with large-delay Cyclic Delay Diversity (CDD) [21]. In the first and second transmit antennas to all NR receive
this mode, the precoding is defined by: antennas
Similarly to the OLSM case, for the H̃0 channel and H̃i
ˆ + H̃ , B = H̃
ˆ+
y(0) (i) x(0) (i) (i = 1, ..., Nint ) interfering channels, A = H̃0 0 0
.. .. ˆ
= W (i)D(i)U
+
. . and Ci = (H̃0 ) (H̃i ) have been defined. Then, similarly
y(Nt −1) (i) x(ν−1) (i) to Equation (1), we obtain
LTE BLER, CQIs 1-15, SISO AWGN, 5000 subframes SNR−CQI measured mapping SNR−CQI mapping model
0
10 15 15
14 14
13 13
12 12
11 11
10 10
−1 9 9
CQI
CQI
10 8 8
7 7
6 6
BLER
5 5
4 4
3 3
2 2
−2
10 1 1
0 0
−20 −10 0 10 20 30 −20 −10 0 10 20 30
SNR [dB] SNR [dB]
−3
10
Fig. 4. CQI mapping. BLER=10% points from the BLER curves (left) and
−10 −5 0 5 10 15 20 25 SINR-to-CQI mapping function (right)
SNR [dB]
Fig. 3. BLER curves obtained from 1.4 MHz, single-user, 5000 subframes
long, SISO AWGN simulations for all 15 CQI values. From CQI 1 (leftmost) to map to the BLER obtained from AWGN link-level simula-
to CQI 15 (righmost) tions. The effective SINR γeff is obtained by performing the
following non-linear averaging of the several Resource Block
(RB) SINRs:
|aii |2 Pl N
!
SINRi = 1 X − SINRi
ν Nint ν γeff = EESM(γi , β) = −β · ln · e β .
X X X X N i=1
Pl |aij |2 + σ 2 |bik |2 + |cl,im |2 Pl,m
j6=i k=1 l=1 m=1 Where N is the total number of sub-carriers to be averaged
SINRi can then be expressed in an identical form as and β is calibrated by means of link level simulations to fit
2 the compression function to the AWGN BLER results [13].
for the
P OLSM2 case (Equation
Pν Pν ζ = |aii2| ,
(2)) by having
2 It is possible to consider not all of the TB sub-carriers
ξ = j6=i |aij | , ψ = k=1 |h̃ik | and θl = m=1 |cl,im | .
but only a subset, as long as the frequency spacing between
3) SISO modeling: As for the Single-Input Single-Output two SINR values does not exceed half of the coherence
(SISO) case, the SINR for a given subcarrier can be written bandwidth [13]. Thus, in our channel traces, we reduced the
as amount of memory needed for a simulation by using only two
sub-carrier SINRs per RB to obtain γeff .
Ptx Using AWGN BLER curves, γeff is mapped to BLER. It
SINR = ,
1 2 X
Nint
|hl |2 is then decided via a coin toss whether the given received
σ + Ptx,l TB was received correctly and ACK reporting is subsequently
|h0 |2 |h0 |2
l=1 generated.
thus only a trace of the noise and the interference parameters Related to the link performance model, the CQI feedback
1 |hl |2 reporting provides the eNodeB with a figure of merit of the
and is needed.
|h0 |2 |h0 |2 state of the channel of the UE. For the CQI feedback strategy,
the SINR-to-CQI mapping is realized by taking the 10% points
IV. L INK PERFORMANCE MODEL
of the BLER curves, obtaining the mapping shown in Figure 4.
The link performance model determines the BLER at the The obtained CQIs are afterwarfds floored to obtain the integer
receiver given a certain resource allocation and Modulation CQI values that are reported back to the eNodeB.
and Coding Scheme (MCS). For LTE, 15 different MCSs are
defined, driven by 15 Channel Quality Indicator (CQI) values. V. M AIN USES , SIMULATION RESULTS AND CONCLUSIONS
The defined CQIs use coding rates between 1/13 and 1 com- In this paper we present a LTE system level simulator
bined with 4-QAM, 16-QAM and 64-QAM modulations [23]. capable of simulating LTE SISO and MIMO networks using
To asses the BLER of the received Transport Blocks (TBs), TxD or OLSM transmit modes and offered for free under an
a set of Additive White Gaussian Noise (AWGN) link-level academic, non-commercial use license.
performance curves are employed. The SINR-to-BLER map- The main purpose of this tool is to assess the network
ping then requires of an effective SINR value γeff , obtained performance increase of new scheduling algorithms (Figure 5).
from mapping the set of sub-carrier-SINRs assigned to the UE Testing Fractional Frequency Reuse (FFR) strategies im-
TB to an AWGN-equivalent SINR. Figure 3 shows the SISO plemented at the scheduler level, as well as the network
AWGN¸ BLER curves the link performance model utilizes [7]. impact of different receiver types and channel quality feedback
The Exponential Effective Signal to Interference and Noise strategies, provided accurate modeling of those, can also be
Ratio Mapping (EESM) [24], [25], [26] is the method currently tested. SINR optimization (Figure 6) via electrical [27] and
used to obtain a TB effective SINR γeff which can be used mechanical tilting and with pathloss maps imported from
UE and sector throughput, 20 UEs/sector, 5 MHz bandwidth
1 [3] J. Colom Ikuno, M. Wrulich, and M. Rupp, “Performance and modeling
of LTE H-ARQ,” in Proc. ITG International Workshop on Smart
0.9
Antennas (WSA), Berlin, Germany, Feb. 2009.
0.8 [4] M. Wrulich and M. Rupp, “Computationally efficient MIMO HSDPA
0.7 system-level evaluation,” EURASIP Journal on Wireless Communica-
tions and Networking, vol. Volume 2009, 2009.
0.6
[5] C. Shuping, L. Huibinu, Z. Dong, and K. Asimakis, “Generalized
F(x)