IEEE Wireless Communications - April 2017 - page 128

IEEE Wireless Communications • April 2017
126
A
symmetrical
O
peration
As illustrated in Fig. 3, the proposed TDD
radio frame structure allows asymmetric link
operation, where users can be scheduled in
the downlink with short TTI sizes (if desirable),
while the minimum TTI size for the uplink is
adjusted in coherence with the users’ coverage
conditions. Hence, the eNB packet scheduler is
responsible for scheduling cell-edge coverage
challenged users with longer TTIs in the uplink.
On a similar note, cell-edge users will also be
configured to send uplink control information
(e.g. ACK/NACK) using longer transmission
times. The transmission time length for uplink
control is assumed to be configured with high-
er-layer signaling per UE. Asynchronous HARQ
is assumed for both the downlink and uplink
in order to have full support for asymmetric
link operation. The use of asynchronous HARQ
offers more flexibility for the timing of ACK/
NACK and retransmissions as compared to syn-
chronous HARQ. Finally, in line with the find-
ings in [11], it is worth mentioning that the use
of the in-resource CCH solution also unleash-
es more flexible use of time-frequency domain
inter-cell interference coordination, as well as it
allows the same beamforming for physical-layer
control and data downlink transmissions.
P
erformance
A
nalysis
In the following, we present example perfor-
mance results to further illustrate the merits of the
proposed solution. We start by addressing latency
performance, followed by presenting examples of
how the overhead varies depending on the use of
different TTI sizes.
L
atency
R
esults
Table 2 summarizes the one-way downlink user
plane latency for different TDD radio frame con-
figurations. UEs at different coverage ranges, enu-
merated as short, medium, large, and extreme
coverage ranges, are considered. The definition
of latency follows that used in 3GPP for the latest
eIMTA studies [9]. The user plane latency is the
sum of five different components, namely the eNB
and UE processing times, the frame alignment
time, the TTI duration, and the average HARQ
RTT. The frame alignment time is the average
waiting time from when data arrives at the eNB
until it can start to transmit the corresponding
payload in the downlink. Hence, the value of this
depends on the TDD radio frame configuration.
The downlink HARQ RRT is defined as time from
the eNB starting to transmit a payload, until it can
start to send the corresponding retransmission
on the same stop-and-wait (SAW) channel. It is
assumed that the average block error rate (BLER)
for a first transmission equals 10 percent, so in cal-
culating the average HARQ RTT, weighting by a
factor 0.1 is applied. The HARQ latency depends
both on the TDD radio frame configuration, as
well as on the UEs’ coverage conditions, since
users at larger coverage ranges can only transmit
their ACK/NACK in a bi-directional TDD block
with a sufficiently long time for uplink resourc-
es. Results are presented for two different TDD
radio frame configurations. The TDD radio frame
configuration with 4x1 ms blocks is composed
of a downlink heavy (D), balanced (B), downlink
heavy (D), and uplink heavy blocks (U). As the
name indicates, the D block has a majority of the
resources for downlink, the B block has similar
downlink/uplink resources, while the U block has
the most resources for uplink transmission. Users
at the short range can transmit their ACK/NACK
in a single subframe (i.e. requires only few uplink
resources), while medium range users can only
transmit their ACK/NACK in B and U blocks, and
large range users can only transmit their ACK/
NACK in U blocks. Given these assumptions, the
values for the frame alignment and the average
HARQ retransmission RTT are obtained from sim-
ple Monte-Carlo simulations. The results in Table
2 clearly show the benefits of having a scheme
that allows asymmetric link operation. The users
at short range can benefit from operation with
low latency, while the users at medium and large
coverage ranges tend to experience longer laten-
cies as they are subject to more constraints on
when they have opportunities for ACK/NACK
transmissions in the uplink. But still, the one-way
user plane latencies for the considered TDD radio
frame of 4x1ms bi-directional TDD blocks is only
on the order of 1.20–1.26 ms, which is significant-
ly lower than the best case 5–6 ms radio latency
for LTE eIMTA.
Table 2 also includes results for a TDD radio
frame configuration consisting of 4x1ms D blocks,
followed by a 1x4 ms D block that contains
approximately 1.2 ms (i.e. corresponding to six
0.2 ms subframes) of time for uplink transmis-
sion. The longer time for uplink transmission is
required for extreme coverage UEs to be able
to transmit their ACK/NACK. The results for this
TDD radio frame configuration therefore further
show the difference in experienced latency for
the short range and extreme range user. It should
furthermore be noticed that the results in Table
2 assume a conservative setting (0.3ms) for the
eNB and UE processing times. If the eNB and UE
processing times are reduced to ~ 0.1 ms, the total
one-way downlink user plane latency is reduced
to less than 1 ms (e.g., as discussed for tactile
Internet use cases).
S
cheduling
O
verhead
vs
L
atency
T
radeoffs
The low latency results are primarily achieved by
using the 1 ms bi-directional TDD blocks and the
even shorter scheduling allocations of 0.2 ms TTIs
(i.e. corresponding to one subframe), using con-
servative link adaptation setting to ensure that
first transmissions have a relatively high success
rate (i.e. low BLER). Scheduling with short TTIs,
does, however, come at a cost of increased CCH
overhead. Fig. 4 shows a bar chart that summa-
F
igure
3.
The basic principle of asymmetric link operation.
Low transmit power UEs
High transmit
power eNB
Uplink transmissions
from cell-edge UE
should be long for
coverage reasons
Can operate
with short uplink
transmissions
Downlink transmissions can
be short for all UEs
1...,118,119,120,121,122,123,124,125,126,127 129,130,131,132
Powered by FlippingBook