You are on page 1of 5

A Semi-Analytical Macroscopic MAC Layer Model

for LTE Uplink


S. Louvros
Dept. of Telecommunications,
Systems & Networks, WNA Lab,
Mesolonghi, Greece
splouvros@gmail.com
A.C. Iossifides
Dept. of Electronics
Alexander TEI of Thessaloniki
Sindos, Thessaloniki, Greece
aiosifidis@el.teithe.gr
K. Aggelis
Dept. of Physics,
University of Patras
Patras, Greece
kosangel@gmail.com

A. Baltagiannis
Dept. of Eng. Sciences, Applied Mathematics & Mechanics,
University of Patras,
Patras, Greece
abalt@upatras.gr

G. Economou
Dept. of Physics,
University of Patras,
Patras, Greece
economou@physics.upatras.gr


AbstractThis paper proposes a semi-analytical medium access
control (MAC) layer model for the macroscopic analysis of long
term evolution (LTE) uplink. The model takes into consideration
the expected number of retransmissions arising from channel
disturbances and the number of allocated resources so that to
satisfy a required packet delay budget. The former is computed
in terms of the mean bit error probability while the latter is
calculated through a high order Markov chain incorporating cell
traffic load conditions. The proposed model can be used as a
macroscopic planning decision tool that poses the working limits
of MAC schedulers for different packet delay budget
requirements, in terms of the mean number of allocated
resources, MAC packet length and maximum retransmission
attempts according to the channel conditions. In this context,
based on measured data, we investigate the impact of the MAC
packet size on the total uplink delay budget of a single user
aiming to set-up the basis for future multi-user analysis.
Keywords-LTE uplink; MAC scheduling; delay optimisation;
LTE measurements
I. INTRODUCTION
Mobile broadband has become a reality with the
introduction of High Speed Packet Access Techniques (HSPA)
and lately the Enhanced UTRAN or Long Term Evolution
workgroup of 3GPP, providing advantages to services and
users [1]. This is achieved due to OFDM over the air interface
and the simplified radio network architecture, where the Radio
Network Controller has been removed and an intelligent Radio
Base Station (eNB) has been directly connected to the core
network, reducing thus the deployment cost. Considering a 20
MHz FDD uplink and downlink spectrum allocation, it is
expected to achieve 5 bps/Hz downlink spectrum efficiency
and 2.5 bps/Hz uplink spectrum efficiency [2]. The minimum
used resource for information transmission over the air is
known as resource block (RB), a two-dimensional resource
which has a total size of 180 KHz (12 sub-carriers of 15 KHz
each) in the frequency domain and seven resource elements per
sub-carrier of duration 0.5ms in the time domain, thus one
resource block has 127 = 84 resource elements. Transmission
over radio interface consists of two consecutive resource
blocks in time, known as scheduling block (SB) with time
duration of 1ms (Transmission Time Interval TTI). The most
important element of the eNB is the MAC scheduler which is
responsible for the dynamic allocation of resources into many
users, based on the channel conditions (available through
Channel Quality Indicators CQI), the required Quality of
Service (incorporated in the Quality Class Identifier QCI) [3]
and cell load conditions. Allocation of resources takes place
both in the time domain (TD) and the frequency domain (FD)
providing higher throughput and coverage gains.
Cell coverage and range affect the scheduler decisions and
thus the user throughput due to degraded CQI reports in bad
channel condition areas. Cell planners are very much interested
in predicting the MAC scheduler decisions in order to tune the
cell range properly [4]. System simulation is the most common
technique for cell planners to provide a measure of throughput
for different channel conditions and scheduling techniques. A
lot of scheduler performance analyses are available in
international literature based on simulations. A detailed study
for time-frequency domain resource allocation, together with
different scheduling approaches, have been given in [5]. In [6]
an interesting algorithmic analysis exploiting simul-taneously
throughput optimization and QoS rate restrictions is provided.
Although ideally the scheduler should have exact knowledge of
the channel quality for each sub-carrier and each user [7], [8],
this is not realistic due to increased signalling load. Therefore,
a compromise between good channel quality knowledge and
CQI reports load has to be decided [9].
All the aforementioned efforts mainly focus on the scheduling
algorithm performance optimization based on simulations and
simplified theoretical approaches for CQI, quality and MCS
selection. What is still missing is a simplified but consistent
semi-analytical model to combine all different aspects of
Figure 1. Standardised QCI characteristics
scheduling in order to provide a semi-analytical performance
evaluation for user MAC throughput. IP user data packets have
variable length and RLC/MAC protocol layer accommodate
these IP packets providing segmentation to fit the appropriate
IP segment sizes into variable MAC packet sizes [10]. An
interesting question in such a model might be to find the
optimum MAC packet length size in order to optimize MAC
performance and improve throughput per connection. From an
optimization point of view, transmission delay due to overhead
is inversely proportional to the MAC packet size. Also
selecting larger MAC packet sizes results into less
microprocessor load and less expected delay [11]. On the other
hand, transmission pipelining is inversely proportional to the
MAC size. Hence there should be a compromise between
transmission reliability, which favors large number of bits per
MAC packet, and transmission delay microprocessor load
which favor for larger MAC packet sizes. Microprocessor load
is beyond the scope of this paper, however what is interesting
to measure is the total transmission time (as a measure of
expected delay) and the error performance as functions of
MAC packet size. In this paper we propose a semi-analytical
model incorporating several parameters, including the MAC
packet size, number of retransmissions, bit error rate, and
allocated resources with respect to cell load which provides a
basis for cell planning decisions or scheduling criteria, in order
to satisfy the required delay budget requirements posed by
QCI. As a case study, we use the proposed model to evaluate
the impact of MAC packet size in a realistic scenario, available
through SNIR and BER measurements with an eNB test bed.
The rest of the paper is organised as follows: In Section II a
detailed delay analysis is presented. Section III proposes a
higher layer Markov chain analysis to insert the cell load
conditions into the analysis. Finally, in Section IV results are
presented and comparison with a real drive test analysis is
supported, as a justification to the accuracy of the proposed
semi-analytical model.
II. MAC DELAY SEMI-ANALYTICAL MODEL
LTE traffic is based solely on IP technology. Between one
user equipment (UE) and one eNodeb each MAC packet is
supposed to be transmitted completely over the air interface
before starting transmission of next MAC packet in a duration
of TTI = 1ms. In such a case the advantage is that LTE could
provide error detection and retransmission, not only on
TCP/UDP IP level on the core network but also on the air
interface, resulting into better correction techniques improving
thus the throughput and minimizing delays. Scheduling is
mostly decided based on QoS service profile, CQI reports and
UE uplink buffer sizes (signaled uplink to the eNodeB MAC
layer using the uplink packet physical channel PUCCH) [10],
[12], [13]. Suppose that a TCP/UDP IP packet of M
I
variable
bits per packet be framed in such a way that the resulting MAC
packets of variable length contain M
mac
payload bits and a
fixed number of M
over
header bits per packet [10]. In such a
model then one M
I
packet will be segmented into M
I
/M
mac

number of MAC packets. Considering that this division will
rarely allocate an integer number then one extra MAC packet
will be needed in most of the cases to accommodate the
remaining bits out of the division. The total number of
transmitted bits then will be M
I
+ M
I
/ M
macM
over
, where the
factor M
I
/ M
macM
over
indicates the overhead created by the
MAC layer for the M
I
size TCP/UDP IP packet transmission.
Considering ideal radio channel conditions (no
retransmissions) the expected whole TCP/UDP IP packet
transmission time is:



+
= +

I mac over
delay
TTI
I
s s
M M M M
T T nT
M N n
(1)
In (1) n
TTI
is the number of transmitted bits per SB
depending on Link Adaptation Modulation Scheme. N is the
average allocated number of 180 kHz radio block units of
bandwidth per TTI, considering also the constraint that
0 0.18N BW < where BW is the allocated bandwidth in MHz
(1.4 to 20) and M is the number of antenna ports (in case of
MIMO implementation). Finally, n is an integer indicating the
number of TTIs a MAC packet is waiting to be scheduled by
the scheduler in a total scheduling period and T
s
is TTI duration
of 1ms. In uplink, n depends mainly on the service QCI, on
CQI reports, on UE transmitter mean packet waiting time on
the buffer and on cell load.
The average transmission delay is expected to be increased
due to existing retransmissions (HARQ functionality) [10].
Indeed real transmission environments with dispersive channel
characteristics introduce ISI and thus Bit Error Rate (BER) on
the receiver especially in low SNR cellular areas. Inevitably
retransmissions are going to be present resulting in increase of
transmission delay and degradation of service throughput.
Retransmissions are created when errors are present in the
received packets, the total resulting BER is higher than a
threshold based on the QoS profile of the supported service [3]
and the error control techniques fail to recover errors. In this
scenario we also consider that corrupted packets are
uncorrelated between each other, meaning that if one MAC
packet is corrupted and retransmission is requested, next MAC
packet of the TCP/IP original packet could be corrupted or not
without any previous memory of the previous packet condition.
Assuming that each MAC packet can be retransmitted v times
at most and the average number of retransmissions is n
mac
(1),
TCP/IP packet transmission delay time can be calculated as:
( ) )
1 (1 +

+ +
= +

mac mac I mac over
retr
delay
TTI
I
s s
n M M M M
T T nT
M N n
n

(2)

Figure 2. Higher Order Markov Chain; Cell load analysis
The average number of retransmissions n
mac
is a function of
the MAC packet error rate. Let p be the packet success
probability. The MAC packet is received correctly with
probability p at the first transmission interval, with probability
p(1p) at the second transmission interval and so on up to vth
transmission interval with probability p(1p)
v1
. If after the
maximum number of transmissions v, the packet is still
corrupted it will be finally forwarded to the upper RLC layer
with probability (1p)
v
since MAC layer does not discard
corrupted packets even after the maximum number of
retransmissions v [10]. Therefore, the mean number of
retransmissions can be calculated as:

1
1
1
2 (1 ) ... (1 ) (1 )
(1 ) (1 )
v v
mac
v
v k
k
n p p p vp p v p
v p kp p

=
= + + + + =
= +

(3)
Setting a = 1p, (3) can be rewritten as

1
1
1
(1 )
(1 )
v
v k
mac
k
v
v k
k
n va a ka
d
va a a
da

=
=
= + =
= +

(4)
which, taking into account the geometric series formula

1
1
1
1
K K
k
k
a
a
a

(5)
And after some straightforward calculations yields

1
(1 ) (1 ) 1 1
v
mac
p v v p pv
n
p

+ +

= (6)
Considering that
1
1
1
p
p
+

for 1 p , (6) leads to



1 (1 )
v
mac
p
n
p

= (7)
The average number of retransmissions n
mac
depends
explicitly on the maximum number of attempts v and on the
size of the MAC packet M
mac.
. The LTE MAC Scheduler
follows rules of priorities on scheduled packets per TTI. A
common practice is to assign retransmission packets with
higher priority than new packets on the transmitter bucket
when they are uplink scheduled by scheduler. A very strict
restriction on LTE non guaranteed bit rate (non-GBR) data
services regarding delay budget, as described in Fig. 1, is
approximately
max
= 300 ms. Hence due to HARQ function
one MAC packet will be retransmitted a maximum number of v
times as long as delay budget never exceeds 300 ms:

max
max
s
s s
s
nT
vT nT v
T

= + = (8)
Then, equation (7) becomes

( )
( )
( )
( )
( )
( )
max
1 1 1
1
1 1 1
1
mac
mac
s
mac
s
mac
v
M
b
mac
M
b
nT
M
T
b
M
b
p
n
p
p
p


= =

(9)
where p
b
, the average bit error probability of the MAC bits, is
necessary for the computation of average number of
retransmissions n
mac
. Finally, the IP transmission data rate R
data
is given by

I
data
retr
delay
M
R
T
= . (10)
Overall, the net service throughput is less than the
calculated data rate because of higher layer overhead [14],
additional delays introduced in core network for information
management and security functions, e.g. [15], which consist a
crucial factor of modern wireless system designs, However
further study of the core network impact is beyond the scope of
this paper.
III. CELL LOAD CONSIDERATIONS
Consider a cell with a specific bandwidth of N
RB
total
available radio blocks. The model maps each available radio
block into one E
i
state of a higher order birth-death process.
Each user theoretically could demand up to the total N
RB
RB, if
available, more practical however might be the allocation of
i < N
RB
with a probability
i
. Considering the higher order
queue [16] of Fig. 2 and assuming that each user is entering the
cell with arrival rate and the service time equals , the
probability the system is at state E
i
is provided by the
following recursive formula
1
( 1)
0
1
1
1
1
( , ) ( 2)
( 1)
1
, 2
( 1)
subject to ( , ) 1
RB
RB
N i
mi RB m i j
j
i
m i RB

N
mi RB
i
N i
i
i N
i
N
+

=
=

= +









=


(11)

Figure 4. BER measurements, TU3 model

Figure 3. Average number of reserved resources vs cell load.

Considering a really tight LTE planning (1/1 frequency
planning) of maximum cell frequency band capacity of
20MHz, the number of available OFDM resources per cell will
be N
RB
= 110. The average number of used OFDM resources
given a certain amount of traffic load into the cell can be
calculated by [4]

1
( , )
RB
N
mi RB
i
N i N
=
=

(12)
Fig. 3 provides a plot of the average number of reserved
resources for different cell load conditions considering service
requests of 1, 2 and 5 resources.
IV. RESULTS AND DISCUSSION
The analysis presented in the previous sections can be used
in macroscopic planning decisions that pose the limits of
short-term real-time scheduling decisions that relate channel
conditions (in terms of bit error probability), allocated resource
elements and MAC packet size so that to guarantee that, in
average, the predefined maximum service delay is satisfied.
Given the bit error probability p
b
, we can calculate (9) the
average number of retransmissions n
mac
that satisfy a
predefined packet delay budget
max
with respect to the MAC
packet size M
mac
. This will lead us to the average number of
resources to be allocated, namely N, with respect to the MAC
packet size, through (2). In turn, given the available number of
resources for a specific traffic load, e.g. by using (12), we can
evaluate, via (2), (9) and (10), the necessary mean MAC packet
size and the average number of retransmissions that lead to a
certain throughput while satisfying packet delay budget
criteria. The above mentioned cases may be either used in cell
planning procedures or as input to MAC schedulers for setting
the limits for the required QoS.
In an example scenario of the aforementioned second case
type, we evaluated, in the following, the impact of MAC
packet size. In order to compute n
mac
, the bit error probability is
required. This was calculated with the use of Fig. 4 that was
extracted from drive tests using TEMS (Ascom S.A.). The
measurements have been conducted in an urban environment
which is highly dispersive. For the drive test trial a test eNB of
Teledrom AB was used with a rooftop antenna. Additionally,
an LTE UE category 4 [17] (max uplink bit rate = 50 Mbps,
uplink higher supported modulation 16QAM, 22 spatial
multiplexing) by an in-car moving user with an external
rooftop antenna was engaged, in order to avoid extra in-car
penetration losses. The car speed during the trial was less than
5 Km/h, trying to fulfill the Typical Urban channel model
(TU3 model, 3 Km/h) requirements [18]. In such a case
Doppler shift effects can be considered negligible.
TEMS investigation Data Collection software was used for
data gathering and MapInfo was used for geographical post
processing in order to collect real E
b
/N
0
data. Throughout the
test the average E
b
/N
0
was reported to be equal to 10 dB
indicating a relative good quality. From Fig. 4, E
b
/N
0
= 10 dB
corresponds to an approximate BER of 210
1
. Therefore, for a
maximum delay budget of
max
= 300 ms (the test included an
FTP upload to a collocated to the eNB server), we can calculate
the number of retransmissions n
mac
(9). On the other hand, it is
obvious form Fig. 3. that resource unavailability arises only for
high cell load. The test eNB allows only test SIM cards to be
connected. Thus, continuous scheduling of the test UE was
achieved, meaning that n = 0 in (1) and (2). M = 2 was set for a
22 MIMO solution (supported by both UE category 4 and test
eNB). Average IP packet length was considered to be M
I

= 1500 bytes and M
over
= 40 bits. Finally, the number of
transmitted bits per RB over a SB interval (1 ms) for a UE
supporting 22 MIMO with 16QAM is calculated as: (144 user
plane symbols per SB) (4 bits per symbol) (2 antenna
layers) = 1352 bits.
Fig. 5 illustrates the semi-analytically calculated delay as a
function of different allocated OFDM resources N and M
mac
packet lengths for n = 0, 1, 2, 3, 4 (each surface corresponds to
a value of n). It is obvious that delay is very low at 5000 <
M
mac
< 12000 packet sizes and 3 < N < 100. The drive tests
showed an average test uplink service rate of 10 Mbps with an
Figure 5. Analytical model delay analysis
average allocated number of resources equal to N = 55.
Following (10), the drive test average delay is estimated to be
retr
delay
T = 0.0012s. Using the analytical plot data (Fig. 5), the
required average MAC packet size for a total delay of 0.0012s
at N = 55 OFDM allocated resources and n = 0 is predicted to
be M
mac
= 6738 bits. It can be easily seen (Fig. 5) that for a an
average MAC packet size lower than 5000 bits the total delay
is increasing significantly, although still under the service
requirements of 300 ms.
V. CONCLUSION
This paper introduced a semi-analytical model for the
analysis of LTE uplink. The model takes into consideration the
expected number of retransmissions arising from channel
disturbances and the number of allocated resources so that to
satisfy a required packet delay budget. Based on this model and
measured data, we investigated the impact of the MAC packet
size on the total uplink delay budget of a single user aiming to
set-up the basis for future multi-user analysis.
ACKNOWLEDGMENT
The authors of this paper would like to express their
gratitude to Mr. Panos Kostopoulos, C.E.O. of Teledrom AB,
Sweden, for its prompt help on setting up the LTE eNB for the
drive test.
REFERENCES
[1] Dahlman, Parkvall, Skold, and Beming, 3G Evolution: HSPA and LTE
for Mobile Broadband, Academic Press, Oxford, UK, 2007.
[2] 3GPP TS 25.104, Base Station (BS) Radio Transmission and Reception
(FDD), V9.2.0, 2009.
[3] 3GPP TS 23.203, Policing and Charging Control Architecture, Rel-11,
V11.4.0, 2011
[4] S. Louvros, K. Aggelis, and A.Baltagiannis, LTE Cell Coverage
Planning Algorithm Optimising Uplink User Cell Throughput,
Proceedings of the 2011 11
th
International Conference on
Telecommunications (ConTel), pp. 51-58, June 2011.
[5] G. Monghal, K. I. Pedersen, I. Z. Kovacs, and P. E. Mongensen, QoS
Oriented Time and Frequency Domain Packet Schedulers for the
UTRAN Long Term Evolution, IEEE Vehicular Technology
Conference (VTC Spring 2008), pp. 2532-2536, May 2008.
[6] G. Wunder, C. Zhou, H-E. Bakker, and S. Kaminski, Throughput
Maximazation under Rate Requirements for the OFDM Downlink
Channel with Limited Feedback, Eurasip Journal on Wireless
Communication & Networking, vol. 2008, January 2008.
[7] K. Homayounfar and B. Rohani, CQI Measurements and Reporting in
LTE: A New Framework, IEICE Tech. Rep., vol. 108, no. 445, pp.
191-196, March 2009.
[8] T. Mohammad Kawser, Nafiz Imtiaz Bin Hamid, Md. Nayeemul Hasan,
Md. Shah Alam, and Md. Musfiqur Rahman, Downlink SNR to CQI
Mapping for Different Multiple Antenna Techniques in LTE,
International Conference on Future Information Technology (ICFIT),
December 2010.
[9] N. Kolehmainen, J. Puttonen, P. Kela, T. Ristaniemi, T. Henttonen, and
M. Moisio, Channel Quality Indication Reporting Schemes for UTRAN
Long Term Evolution Downlink, IEEE Vehicular Technology
Conference (VTC Spring 2008), pp. 2522-2526, May 2008.
[10] 3GPP TS 36.321, Evolved Universal Terrestrial Radio Access (E-
UTRA); Medium Access Control (MAC) protocol specification (Release
8), V8.1.0, 2008.
[11] D. Szczescny, A. Showk, S. Hessel, A. Bilgic, U. Hildebrand, and V.
Frascolla, Performance Analysis of LTE Protocol Processing on an
ARM Based Mobile Plattform, International Symposium on System-
on-Chip (SOC 2009), pp. 56-63, Tampere, October 2009.
[12] 3GPP TS 36.211, Evolved Universal Terrestrial Radio Access (E-
UTRA); Physical Channels and Modulation , Rel-10, V10.4.0, 2011.
[13] 3GPP TS 36.300, Evolved Universal Terrestrial Radio Access (E-
UTRA) and Evolved Universal Terrestrial Radio Access Network (E-
UTRAN); Overall Description; Stage 2, Rel-11, V11.0.0, 2011.
[14] M. Wang, M. Georgiades, and R. Tafazolli, Signalling Cost Evaluation
of Mobility Management Schemes for Different Core Network
Architectural Arrangements in 3GPP LTE/SAE, Vehicular Technology
Conference (VTC-Spring), Singapore, 11-14 May 2008, pp. 2253-2258.
[15] N. Sklavos, "On the Hardware Implementation Cost of Crypto-
Processors Architectures", Information Systems Security, The official
journal of (ISC)2, A Taylor & Francis Group Publication, Vol. 19, Issue:
2, pp. 53-60, 2010.
[16] O. Allen, Probability, Statistics, and Queueing Theory with Computer
Science Applications, Academic Press, Boston, 1994.
[17] 3GPP TS 36.306, Evolved Universal Terrestrial Radio Access (E-
UTRA) User Equipment (UE) Radio Access Capabilities, Rel-8,
V8.0.0, 2007.
[18] 3GPP TR 45.050, Background for Radio Freequency (RF)
Requirements, Rel-10, V10.0.0, 2011.