Vous êtes sur la page 1sur 6

Improving MIMO Spectral Efficiency in

4G Macro-Cellular Networks
Pedro Vieira , Paula Queluz and Antonio Rodrigues

Lisbon Polytechnic Institute (ISEL), Portugal

de Telecomunicaco es / IST, Technical University of Lisbon, Portugal
email: pvieira@deetc.isel.ipl.pt, paula.queluz@lx.it.pt, antonio.rodrigues@lx.it.pt

AbstractThis paper considers the Multiple Input Multiple

Output (MIMO) capacity enhancement considering the Universal
Mobile Telecommunication System (UMTS) Long Term Evolution (LTE) technology, in the downlink, including the effects of
system bandwidth and Signal-to-Noise Ratio (SNR) efficiency.
Firstly, the system bandwidth efficiency is calculated for LTE
framework. It becomes 63% for Single Input Single Output
(SISO) and around 58% for multi-antenna MIMO configurations,
which underlines the importance of considering system bandwidth efficiency when using Shannon spectral efficiency to estimate the system performance of LTE. Then, the SNR efficiency
using Adaptive Modulation and Coding (AMC) is approximated
using curve fitting. The used fitting function is an attenuated and
truncated form of the Shannon bound in order to approximate
the LTE composite spectral efficiency for the Modulation and
Coding Set (MCS). Finally, the total capacity expected results are
calculated for different multi-antenna configurations, revealing
large capacity gains when compared with SISO.
Index TermsWireless Communications, MIMO, LTE, Spectral Efficiency.

Recently, UMTS networks worldwide were upgraded to
High Speed Downlink Packet Access (HSDPA) in order to
increase data rate and capacity for downlink packet data. In
the next step, High Speed Uplink Packet Access (HSUPA) has
boosted uplink performance in UMTS networks. While HSDPA was introduced as a 3rd Generation Partnership Project
(3GPP) release 5 feature, HSUPA is an important feature of
3GPP release 6. The combination of HSDPA and HSUPA is
often referred to as High Speed Packet Access (HSPA).
However, even with the introduction of HSPA, evolution of
UMTS has not reached its end. HSPA+ will bring significant
enhancements in 3GPP release 7. The objective is to enhance
performance of HSPA based radio networks in terms of
spectrum efficiency, peak data rate and latency, and exploit
the full potential of WideBand Code Division Multiple Access
(WCDMA) based 5 MHz operation. Important features of
HSPA+ are downlink MIMO, higher order modulation for
uplink and downlink, improvements of layer 2 protocols, and
continuous packet connectivity.
In order to ensure the competitiveness of UMTS for the
next 10 years and beyond, concepts for UMTS LTE have been
investigated. The objective is a high-data-rate, low-latency and
packet optimized radio access technology. Therefore, a study
item was launched in 3GPP release 7 on Evolved UMTS

Terrestrial Radio Access (E-UTRA). LTE/E-UTRA are a part

of 3GPP release 8 core specifications [1].
One of the most important features of LTE is MIMO. The
MIMO technology, exploiting the spatial multiplexing concept,
is considered the enabling technology for LTE high data rates,
along with the AMC. Hence, its study is of great interest and
constitutes a serious research target.
In the following, this paper studies the MIMO capacity
enhancement considering the UMTS LTE technology, in the
downlink. The spatial multiplexing concept will be developed
and the associated LTE downlink capacity will be calculated
including the effects of system bandwidth and SNR efficiency.
The paper is organized as follows. Section II introduces a
new approach for LTE realistic capacity calculation, using the
spatial multiplexing MIMO setup, AMC and power allocation
technique. Next, the capacity expected values are calculated in
section III for different MIMO setups using an outdoor macrocell environment. Finally, in section IV, conclusions are drawn.
Consider the LTE system, in the downlink [1]. The channel
bandwidth is split into Q flat-fading sub-channels H (fq ), with
a 180 kHz resource block frequency separation. Considering
the channel eigen-analysis [2], the channel matrix H (fq ) may
offer Kq parallel sub-channels with different power gains
(eigenvalues), k,q , where,
Kq = Rank(R (fq )) M in(U, V )


and the functions Rank () and M in () return the rank of the

matrix and the minimum value of the arguments, respectively;
U and V are the number of Base Station (BS) and Mobile
Station (MS) antennas. This corresponds to the spatial multiplexing concept.
In the situation where the channel is known at both Transmitter (TX) and Receiver (RX) and is used to compute the
antennas optimum weight, the power gain in the (k, q)th
eigenmode channel is given by the k,q eigenvalue, i.e., the
SNR for the (k, q)th sub-channel, k,q , is equal to,
k,q = k,q k,q



where k,q is the fraction of power assigned to the (k, q)th

sub-channel, and S/N is the radio link SNR. The k,q values
depend on the sub-channel power allocation scheme; if spatial

water-filling is used, the k,q values are calculated as a function of the eigenvalues, k,q . On the contrary, if uniform power
allocation is considered, the k,q are constant, corresponding
to the uniform power distribution between all the sub-channels
In order to evaluate the performance bound of different
power allocation methods and MIMO configurations, the normalized Shannon channel capacity is usually used (in bit/s/Hz).
The Shannon Capacity bound can not be reached in practice
due to several implementation issues. To represent these losses
mechanism, a modified Shannon capacity expression for the
k th spatial sub-channel and also for the total capacity will be
used as,

Ck =
log2 1 + k,q k,q SN R
Q q=1

Kq Q
log2 1 + k,q k,q SN R



where BW , SN R adjust for the system bandwidth and SNR

implementation efficiency of LTE, respectively.

Most of the spectral efficiency and digital modulation performance calculations are based on symbol error probability,
P (e). To allow comparisons between modulation schemes
with different values of M and hence, whose signals carry
different numbers of bits, a better performance measure is the
bit error probability Pb (e), often also referred to as Bit Error
Rate (BER). This is the probability that a bit emitted by the
source will be received erroneously by the user.
In summary, the evaluation of a modulation scheme may
be based on the following three parameters: the bit error
probability, Pb (e), the b /N0 necessary to achieve Pb (e),
and the bandwidth efficiency Rb /Bw . The first tells us about
the reliability of the transmission, the second measures the
efficiency in power expenditure, and the third measures how
efficiently the modulation scheme makes use of the bandwidth.
LTE uses link adaptation considering several MCSs. The
implemented modulations are Quadrature Phase-Shift Keying
(QPSK) and M -Quadrature Amplitude Modulation (QAM),
with M = 16, 64. Hence, the two modulation types will be
The error probability of QPSK can be determined explicitly.
From [3],

A. SNR and Error Probability

Transmission over the Additive White Gaussian Noise
(AWGN) channel with a two-sided noise power spectral density N0 /2 with perfect Channel State Information (CSI) is
assumed, when modeling the LTE channel.
The bit rate of the source, Rb , is related to the number of
waveforms used by the modulator, M , and to the duration of
these waveforms, T , by the equality,
log2 M
Rb =
This is the rate in bit/s that can be accepted by the modulator.
The average power expended by the modulator, S, is

where is the average energy of the modulator signals. Each
signal carries log2 M bits. Thus, defining b as the average
energy expended by the modulator to transmit one bit, so that
= b log2 M , we have
S = b

log2 M
= b R b .


The SNR (symbolically S/N ), is defined as the ratio between the average signal power and the average noise power.
The last equals N0 Bw , where Bw is the equivalent noise
bandwidth of the receiving filter. Hence, we have
b R b

N0 Bw
N0 Bw


Expression (8) shows that the SNR is the product of two

quantities, b /N0 , the energy per bit divided by twice the
power spectral density, and Rb /Bw , the spectral efficiency of
a modulation scheme.

P (e) = erf c



r 2
erf c



where erf c(.) is the complementary error function. A simpler

approach for general M -ary Phase-Shift Keying (PSK) is to
use the upper bound to P (e) [3],
P (e) erf c


log2 M sin


The QAM is a linear modulation scheme such that the

source symbols determine the amplitude as well the phase of
a carrier signal. Contrary to PSK, the signal envelope is not
constant. Several QAM families are available. We will focus
the two considered LTE square constellations with M = 16
and M = 64. A simple bound to P (e), which is only an
approximation useful for large M , will be used for the two
modulations [3],
3 log2 M b
P (e) 2erf c
2 (M 1) N0
Considering a certain maximum BER target, Pb,target with
Gray Mapping and expressions (8), (10) and (11), the spectral
efficiency, Rb /B, is calculated in function of the SNR as,

log2 M
f orM P SK
2 sin

(erfc (Pb,t arg et ))
3 log2 M
2 N

f orM QAM

Pb,t arg et log2 M

2(M 1) erfc

where B is the transmission bandwidth and it was considered
equal to the equivalent noise bandwidth, Bw .


B. LTE Bandwidth Efficiency

The LTE bandwidth efficiency, BW , is decreased by several
issues that will be described below. Due to the constraints
of Adjacent Channel Leakage Ratio (ACLR), the bandwidth
occupancy is reduced. The ACLR is a measure of TX performance for wideband networks, and it is defined as the
ratio and the transmitted power to the power measured after a
RX filter in the adjacent radio-frequency channel. Considering
the parameters for LTE downlink transmission scheme, the
bandwidth efficiency due to ACLR, ACLR is calculated as,

1 f
= 0.9

where Nsc
is the number of downlink data subcarriers, f
is the subcarrier bandwidth and B is, again, the transmission
Furthermore, the bandwidth efficiency due to the Cyclic Prefix (CP) overhead, CP , will be considered. CP is calculated
Tslot Nsymb
CP = 1

Tslot is the LTE 0.5 ms time-slot and a subcarrier spacing f = 15 kHz corresponds to a useful symbol time
Tu = 1/f 66.7s. The overall Orthogonal Frequency
Division Multiplexing (OFDM) symbol time is then the sum
of the useful symbol time and the cyclic-prefix length TCP .
LTE defines two CP lengths, the normal CP and an extended
CP, corresponding to Nsymb
equal to seven and six OFDM
symbols per slot, respectively. Furthermore,
Normal cyclic prefix: CP = 0.93, for TCP 5.2s
(OFDM symbol #0) , TCP = 4.7s (OFDM symbol #1
to #6);
Extended cyclic prefix: CP e = 0.80, for TCP e
16.7s (OFDM symbol #0 to OFDM symbol #5).
The obtained channel parameters in [4] point to delay
spreads from 0.5 s up to 0.7 s. Hence, the normal CP will
be considered, since it represents a 10 times value greater than
the delay spread.
To carry out downlink coherent demodulation, the MS needs
estimates of the downlink channel. A straightforward way
to enable channel estimation in case of OFDM transmission
is to insert known reference symbols into the OFDM timefrequency grid. The reference symbol density depends on the
number of TX antennas used in the multi-antenna configuration. In the sequence, the bandwidth efficiency caused by
reference symbols insertion using r TX antennas , RS,r is,
r = 1, 2
1 N RB4r
sc Nsymb
RS,r =
r = 3, 4




are 12 subcarriers per resource block
and 7 OFDM symbols per subcarrier, respectively.
At the system level, additional overhead related to common control channels is added such as the Synchronization
Channel (SCH) and Broadcast Channel (BCH). Besides other

Bandwidth Efficiency
ACLR efficiency (ACLR )
CP efficiency (CP )
Reference signal efficiency (RS,r )
L1/L2 efficiency(L1/L2 )
Total link-level efficiency (BW,r )
Total system-level efficiency

Nr. of TX antennas
0.95 0.90 0.89
0.80 0.76 0.75
0.63 0.60 0.59


operations, these transport channels are used for cell search

Furthermore, the more essential control signaling overhead
in LTE is related with the transmission of downlink transport
channels, more specifically the Downlink Shared Channel
(DL-SCH) transmission. This control signaling is normally
referred to as the Layer 1 (L1)/Layer 2 (L2) control signaling.
The L1/L2 control channels are mapped to the first (up
to three) OFDM symbols within each 1 ms sub-frame. The
overhead depends on the number of MSs to be scheduled,
the MIMO/beamforming schemes selected and the resource
assignment. The resource assignment indicates which resource
units (and layers, in case of multi-layer transmission) the MSs
should demodulate. Additionally, it also depends on the used
modulation scheme, payload size and the Hybrid Automatic
Repeat-reQuest (H-ARQ) process the current transmission is
addressing [5].
Hence, the maximum overhead scenario will be considered,
3 OFDM symbols within each 1 ms Transmission Time
Interval (TTI) will be occupied with L1/L2 control signaling,
leading to a bandwidth efficiency, L1/L2 , equal to
L1/L2 = 1


= 0.7857.


Similar L1/L2 control signaling efficiency results were reached

in [6] using system simulators, for a simultaneously scheduling
in the order of 10 MSs using a 10 MHz bandwidth.
The total results for the LTE link-level bandwidth efficiency
are given by,


and listed in Table I. The associated system -level results are

reached by multiplying BW,r by the L1/L2 control signaling
parameter, L1/L2 .
Including the additional system overhead, the LTE bandwidth efficiency becomes 63% for SISO and around 58% for
multi-antenna MIMO configurations. Additionally, the L1/L2
control signaling overhead was considered constant and independent from r, the number of TX antennas, for simplistic
reasons. Since the worst-case scenario was set, the bandwidth
efficiency value for a lower number of TX antennas will be
slightly pessimistic.
C. LTE SNR Efficiency
The previous section presented the LTE bandwidth efficiency calculation. It is a straightforward process, since most


Carrier Frequency [MHz]
OFDM Parameters
Channel Model
Channel Estimation

short CP, 7 data symbol per 0.5 ms
sub-frame, 1 ms TTI
Extended COST 273 DCM [8]
QPSK: 1/8, 1/5, 1/4, 1/3, 1/2,2/3,4/5
16-QAM: 1/2, 2/3, 4/5
64-QAM: 2/3, 3/4, 4/5
20 km/h average speed, vehicular [9]

of the LTE link and system level parametrization allows a

direct calculation of the several spectral efficiency losses.
On the other hand, the SNR impact on spectral efficiency
depends on a higher number of factors, some of them much
more difficult to model than the previous bandwidth efficiency
Consider the basic LTE downlink physical layer parameters
in Table II. There are constraints to the maximum spectral
efficiency from the supported MCSs and also the used MIMO
mode. Additionally, there are performance aspects related with
the TX and RX algorithms (linear, non-linear, etc) [1] which
makes the SNR efficiency definitely more complex to directly
calculate than the bandwidth efficiency. The presented method
will use curve fitting as in [7] and will be explained in the
following. LTE downlink system uses several MCSs that can
be characterized by a modulation type and a code rate, see
Table II.
In what concerns the LTE coding scheme, Release 6 HSDPA
compliant turbo coding is used. For each encoding scheme, the
coded bit error probability is often determined using system
level simulations [10], which allow to compute the coding
gain for certain BER. The coding gain, GM CS (Pb,target ), is
defined as the relation between the SNR levels associated to
the uncoded and coded system required to reach the same
Pb,target . Hence, using system level results, a coding gain can
be set, for each MCS and Pb,target . Moreover, the maximum
spectral efficiency of a given MCS is the product of the code
rate (k/n) and the number of bits per modulation symbol,
log2 (M ).
The spectral efficiency performance using coded modulation
schemes can be estimated using the previous inputs. From
equation (12), the LTE spectral efficiency with coded QPSK
(M=4), 16-QAM (M = 16) and 64-QAM (M = 64) is,

3 log2 M

Pb,t arg et log2 M
= min 2(M 1) erfc
S k
k GMCS (Pb,t arg et ) N , n log2 M
The spectral efficiency results are presented in Fig. 1 as a
function of the SNR and for the LTE MCSs (see Table II); The
BER target, Pb,t arg et is 103 . Fig. 1 also shows the Shannon
bound (in yellow), which represents the maximum theoretical
throughput that can be achieved over an AWGN channel
with a given SNR. The performance of LTE using AMC is
approximated in the following, using curve fitting as presented

Fig. 1. Spectral efficiency, Rb /B as a function of SNR, for the LTE MCS.

AWGN channels assumed.

Nr. of TX antennas, r


SN R,r

Correlation Coef.

in Fig. 2. The first stage is to calculate the composite spectral

efficiency for the AWGN codeset, using max(.) (maximum)
function for each SNR value (black curve), this maximizes
the throughput for each SNR. The second stage is to use an
attenuated and truncated form of the Shannon bound in order
to approximate the composite performance. The proposed
function is,

BW,r log2 1 + SN R,r N

, f or N
N >


where BW,r is the bandwidth efficiency parameter for r TX
antennas, as calculated in section II-B; S/N (max) is the SNR
at which maximum spectral efficiency, RBb (max), is reached
(equal to 4.8 bit/s/Hz using MCS-13).
SN R,r is the SNR efficiency parameter for r TX antennas.
It will be calculated in the following using curve fitting over
the composite spectral efficiency for the AWGN codeset. The
c , and provides a simple and
fitting application is Ezyfit
efficient way to perform quick curve fitting with arbitrary (nonlinear) fitting functions. The fitting function is the truncated
Shannon bound type presented in (19) with SN R,r working
as the only calibrating parameter, since BW,r was previously
The results for the curve fitting correlation and SN R,r
calculation are presented in Table III. Fig. 2 presents the
fitting function for one TX antenna ( RBb (BW,r , SN R,r )) and
also the Shannon bound. The fitting procedure correlation

Fig. 2. Composite Spectral efficiency, Rb /B as a function of SNR, for the

LTE MCS and associated curve fitting.

Fig. 3. Total Capacity Expected Value, E(C), for the SISO and MIMO nn
setup, n = 2, 3, 4, using uniform and water-filling power allocations.

values are quite good, which supports a simple but also

realistic approach on estimating the LTE spectral efficiency
performance for different SNR conditions.

gain is defined as the relation between the MIMO configuration expected capacity and the SISO expected capacity.
The water-fill gain relates, for each MIMO configuration, the
mean capacity using water-filling with the equivalent mean
capacity using an uniform power allocation. Starting with a
SISO reference spectral efficiency of 1.24 bit/s/Hz, the average
capacity using spatial multiplexing and water-filling is 2.38,
3.33 and 4.16 bit/s/Hz using MIMO 2 2, 3 3 and 4 4,
respectively. This represents a serious 92%, 169% and 235%
MIMO gain for these configurations.
The uniform power allocation results are slightly moderate,
but still impressive. 1.97, 2.61 and 3.21 bit/s/Hz average
spectral efficiency was achieved after simulation, for two,
three and four multiple antenna systems, respectively, with
associated 59%, 110% and 159% MIMO gain.


The modified Shannon capacity expression in order to
predict the LTE link level spectral efficiency, see equation
(4) has been presented. It depends on a number of important
factors which are the radio channel, the power allocation
scheme and the radio link SNR, properly represented by the
sub-channel eigenvalues (k,q ), the power fraction (k,q ) and
S/N , respectively. The remaining parameters, BW and SN R
are calibrating values which were previously calculated in
sections II-B and II-C for LTE and different MIMO configurations.
The LTE capacity will be calculated using an existing
MIMO channel model framework [2], which was developed in
c and requires building data information, terrain data,
street data, BS and MS information. The BS and MS data
store the geographical positioning, height, power parameters,
MIMO antenna configuration, radiation pattern, and system
parameters like handover hysteresis and offset.
The tested urban Radio Environment (RE) is located in the
city of Lisbon, Portugal, and several virtual test MSs travel
around the city for a period of 10 minutes. To this end, a
large number of MIMO channel realizations were generated
for each cluster oriented large-scale fading set of parameters,
that simulate the MIMO channel conditions as the MS moves
while connected to the BS. For each channel iteration, an
eigenvalue sample is recorded.
The total capacity expected results, E (C), were calculated
for the SISO and MIMO n n setup, n = 2, 3, 4 and are
displayed in Fig. 3. 5 and 0.5 antenna spacings were used
for the BS and MS, respectively. For comparison purposes,
MIMO and Water-filling Gains were calculated. The MIMO

This paper studies the MIMO capacity enhancement considering the UMTS LTE technology, in the downlink. The
spatial multiplexing concept is presented, and the associated
LTE downlink capacity is calculated including the effects of
system bandwidth and SNR efficiency.
A new approach is introduced for LTE realistic capacity
calculation, using the spatial multiplexing MIMO setup, AMC
and power allocation technique. First, the system bandwidth
efficiency is calculated for LTE framework and multiple antenna configurations. It becomes 63% for SISO and around
58% for multi-antenna MIMOconfigurations, which underlines
the importance of considering system bandwidth efficiency
when using Shannon spectral efficiency to estimate the system
performance of LTE.
The SNR efficiency using AMC was approximated using
curve fitting, where the used fitting function is an attenuated
and truncated form of the Shannon bound in order to approximate the LTE composite spectral efficiency for the AWGN
MCS codeset.

Finally, the total capacity expected results were calculated

for different multi-antenna configurations in a macro-cell
scenario, revealing impressive capacity gains when compared
with SISO. The capacity calculation procedure presented in
this paper is LTE oriented. Nevertheless, the presented framework is generic and applicable to other radio systems.
[1] Dahlman E.; Parkvall S.; Skold J.; Beming P. 3G Evolution: HSPA and
LTE for Mobile Broadband. Academic Press, 2007.
[2] Vieira P.; Queluz M.P.; Rodrigues A. MIMO antenna array impact
on channel capacity for a realistic macro-cellular urban environment.
accepted for 68th IEEE Vehicular Technology Conference Fall 2008,
Calgary, Canada, September 2008.
[3] Benedetto S.; Biglieri E. Principles of Digital Transmission with
Wireless Applications. Kluwer Academic, Plenum Publishers, New York,
[4] Vieira P.; Queluz M.P.; Rodrigues A. Terrain and clutter Impact on
Joint Statistical Properties of Azimuth Spread and Delay Spread in
Macro-cell Bad Urban Environment. IEEE International Symposium
on Wireless Communication Systems 2007, Trondheim, Norway, pages
798802, October 2007.
[5] Technical Specification Group Radio Access Network 3GPP. TR 25.814:
Physical Layer Aspects for Evolved Universal Terrestrial Radio Access
(UTRA). 3GPP, September 2006.
[6] Pokhariyal A.; Kolding T.E.; Mogensen P.E. Performance of Downlink
Frequency Domain Packet Scheduling for the UTRAN Long Term
Evolution. 2006 IEEE 17th International Symposium on Personal,
Indoor and Mobile Radio Communications,, pages 15, September 2006.
[7] Mogensen P.; Wei Na; Kovacs I.Z.; Frederiksen F.; Pokhariyal A.;
Pedersen K.I.; Kolding T.; Hugl K.; Kuusela M. LTE Capacity Compared
to the Shannon Bound. IEEE 65th Vehicular Technology Conference
2007, pages 12341238, April 2007.
[8] Vieira P.; Queluz M.P.; Rodrigues A. Improving a Cluster Based Directional Channel Model in Realistic Macro-cell Environment. submitted
for Wireless Personal Communications Journal, Springer, Special Issue,
[9] Vieira P.;Vieira M.A.; Queluz M.P.; Rodrigues A. A Novel Vehicular
Mobility Model for Wireless Networks. Wireless Personal Communications Journal, Springer, Vol. 43, pages 16891703, August 2007.
[10] Technical Specification Group Radio Access Network 3GPP. TR
36.942: Evolved Universal Terrestrial Radio Access (E-UTRA); Radio
Frequency (RF) system scenarios (release 8). 3GPP, June 2007.