Dynamic TDD Transmissions in Homogeneous Small Cell Networks
Dynamic TDD Transmissions in Homogeneous Small Cell Networks
Dynamic TDD Transmissions in Homogeneous Small Cell Networks
Abstract-In this paper, we investigate the performance of recent works [4]. Gains in terms of wide-band (WB) signal-to
dynamic time division duplexing (TDD) transmissions with interference-plus-noise ratio (SINR) and user equipment (UE)
various kinds of interference management strategies, i.e., cell
packet throughput (UPT) have been observed, mostly in low
clustering (ee), power control and interference cancellation
(Ie), as well as their combinations in homogeneous small cell
to-medium traffic load conditions.
networks. We present extensive results on network performance In this paper, we present new results on dynamic TDD
in terms of downlink (DL)/uplink (UL) wide-band (WB) signal transmissions in homogeneous small cell networks. In partic
to-interference-plus-noise ratio (SINR), 95-, 50- and 5-percentile ular, we investigate dynamic TDD schemes with various kinds
user equipment (UE) packet throughputs (UPTs). We also study
of interference management strategies and present results on
the impact of high-order modulation schemes on dynamic TDD.
Our work shows that, when the traffic load is low to medium,
network performance in terms of WB SINR, 95-, 50- and 5-
an advanced dynamic TDD scheme with UL power boosting percentile UPTs. The contribution of this paper is two-fold:
(ULPB), Ie and full flexibility of TDD configuration is shown to 1) The effectiveness of various inter-link interference man
immensely outperform the static TDD scheme by approximately
agement schemes, such as cell clustering (CC) [2],
30-60 % and 210-300 % in terms of the DL and the UL UPTs,
respectively.
power control [5], [6], interference cancellation (IC), as
well as their combinations, is systematically investigated
I. INTRODUCTION and compared.
2) The impact of high-order modulation schemes on dy-
Currently, the Third Generation Partnership Project (3GPP) namic TDD is studied.
sees exciting activities in the design of Long Term Evolution
The rest of the paper is organized as follows: Sections II, III,
(LTE) Release 12 networks [1], in which the small cell
and IV present the network scenario, a formal description on
enhancement (SCE) study item has a lot of momentum. Future
the dynamic TDD DLiUL subframe splitting, and the interfer
small cell networks are envisaged to prioritize time division
ence mitigation schemes evaluated in this work, respectively.
duplexing (TDD) schemes over frequency division duplexing
Section V provides our new results on dynamic TDD, followed
(FDD) schemes because TDD transmissions are more suitable
by some concluding remarks in Section VI.
for hot spot scenarios with traffic fluctuations in both link
directions [2]. In the 3GPP LTE Release 8-11 networks, II. NETWORK SCENARIO
seven TDD configurations [3], each associated with a downlink
During the study item phase of dynamic TDD in the 3GPP,
(DL)/ uplink (UL) subframe ratio in a 10-millisecond trans
a total of eight deployment scenarios were considered for
mission frame, are available for semi-static selection at the
investigation [2]. After preliminary assessment of technical
network side. However, this semi-static selection is not able
feasibility and performance evaluation, the following two
to adapt DLIUL subframe resources to the fast fluctuation in
scenarios were prioritized for further study [1]:
traffic load generated at the small cells due to the low number
of connected UEs and the burstiness of their DLIUL traffic. • Scenario 3: Multiple outdoor picocells deployed on the
In order to allow small cells to smartly adapt their communi same carrier frequency.
• Scenario 4: Multiple outdoor picocells deployed on the
cation service to the quick variation of DLIUL traffic demands,
a new technology, referred to as dynamic TDD, has drawn same carrier frequency, and multiple macrocells deployed
much attention. In dynamic TDD, the TDD configuration on an adjacent carrier frequency, where dynamic TDD
can be dynamically changed in each or a cluster of cells. schemes can only be used in small cells.
Dynamic TDD can thus provide a tailored configuration of Fig. 1 illustrates Scenarios 3 and 4. In Fig. 1, the n-th
DLiUL subframe resources at the expense of allowing inter small cell base station (BS) and the k-th UE are denoted
link interference, i.e., DL-to-UL and UL-to-DL interference. as c(n),n E {l, . . . ,N} and u(k),k E {l, . . . ,K}, re
The application of basic dynamic TDD transmissions in spectively. The UE's reference signal received power (RSRP)
homogeneous small cell networks has been investigated in associated with small cell BS c(n) is denoted as RSRPn,k .
617
ICC'14 - W3: Workshop on Small Cell and 5G Networks
Table I
configurations within cell clusters are required through inter KEY SIMULATION PARAMETERS
cell communications over backhaul links or the air interface.
Parameters Assumptions
A simple method to perform dynamic TDD DL/uL subframe
splitting for cell clusters is to sum d�L (t) and d;;L (t), as well
Cellular model and layout 7 cell sites, 3 cells per cell site
as d�L (t) and d�L (t) over the small cells in the same cell Inter-site distance 500 m
# picocells per cell 4, radius of coverage: 40 m
cluster and proceed accordingly with the algorithm described
in Section III. # UEs per picocell 10, uniformly dropped
System bandwidth 10 MHz
B. Power Control # picocell antenna 4 for both Tx and Rx [3]
The power control strategy includes the DLPR [5] and # UE antenna 1 for Tx, 2 for Rx [2]
ULPB schemes [6]. The DLPR scheme sacrifices DL perfor UE scheduling in each cell Proportional fairness (PF)
mance in exchange of decreasing the DL-to-UL interference Packet scheduling for UE Round Robin (RR)
and thus improving the UL performance. In contrast, the MC schemes QPSK, 16/64/256QAM [I]
ULPB scheme allows more transmit power consumption at Non-data overhead 3 OFDM symbols per subframe
the UE side to combat the DL-to-UL interference coming Small-scale fading channel EPA channel model defined in [8]
from small cells. The implementation of the power control Receiver type MMSE receiver [2]
techniques is relatively simple, i.e., a fixed power offset tlpDL
and tlPUL can be configured on top of the DL and UL power
level, respectively. second. Besides, we assume ,\UL (qn,d 0.5 x ,\DL (qn,k), as
=
capable of cancelling interference coming from neighboring simulations we adopt the common assumption in the 3GPP
BSs. In contrast, the assumption of UEs performing UL-to that UE is equipped with I antenna for transmission and 2
DL IC with regard to other peer UEs would seem to be too antennas for reception (see Table I).
farfetched and impractical. The IC scheme provides the best
ILIM for the UL, but requires good backhaul connections for
A. Performance of DUUL WB SINR
inter-cell information exchange on DL transmission assump
tions, including resource allocation, modulation and coding First, we check the performance of DL/uL WB SINR to get
scheme (MCS), configuration of demodulation reference sig some overall ideas about how dynamic TDD with or without
nals, etc. Besides, strong signal processing modules in small ILIM atlects the link-level performance. For the CC scheme,
cell BSs are needed to detect, reconstruct and cancel the DL the coupling loss threshold P LCC for small cells within a
interference from UL signals. cell cluster is set to 90 dB [2]. For the DLPR and the ULPB
schemes, tlpDL and tlPUL are set to -20 dB and 10 dB,
V. SIMULATIONS AND DISCU SSIONS
respectively. For the IC scheme, since we only consider the
In this section, we conduct system-level simulations and DL-to-UL IC, its impact on the DL WB SINR is none, thus
present numerical results to compare the performance of the we evaluate the IC scheme only for the UL. Besides, since
existing static TDD scheme in LTE Release 11 with that of dy ,\DL (qn,k) : ,\UL (qn,k) 2 : 1, we assume that in dynamic
=
namic TDD transmissions in LTE Release 12 and an enhanced TDD the probability of observed DL subframes and that of
version with full flexibility of dynamic TDD configuration, UL ones are 2/3 and 1/3, respectively. Furthermore, we drop
which probably falls into the scope of LTE Release 13. We the assumption of full-butler traffic load as in the existing
also consider the ILIM techniques presented in Section IV and works [2], [4], and instead we consider a more practical
their combinations, and investigate the performance gains of network scenario with the probability of the occurrence of in
dynamic TDD with basic and combined ILIM schemes. terference being 0.4. The cumulative density functions (CDFs)
We concentrate our analysis on the 3GPP dynamic TDD of the DL and the UL WB SINR performances are shown in
Scenario 3, i.e., a homogeneous layer of outdoor picocells Figs. 2(a) and 2(b), respectively.
as illustrated in Fig. l. The full list of system parame As can be seen from Figs. 2(a) and 2(b), the straightforward
ters and the traffic modeling methodology can be found dynamic TDD scheme with no ILIM is actually beneficial
in [2] and [7], respectively. More information on the system for DL transmissions compared to the benchmarking scheme
level simulator used for this analysis can be found in with static TDD. This is because the interference from UEs
http://wnt.sjtu.edu.cn/flintlhtml/index.html. Some key param with relatively low transmission power and high path loss
eters adopted in our simulations are presented in Table I. in the UL is usually weaker than that from picocells in
The traffic model is assumed to be Poisson distributed with the DL. However, the UL WB SINR of the straightforward
>PL (qn,k) taking a uniform value for all UEs. The values of dynamic TDD scheme takes a serious hit compared with
,\DL (qn,d are {0.05, 0.15, 0.25, 0.45} packets per UE per that of the static TDD scheme due to the devastating DL-
618
ICC'14 - W3: Workshop on Small Cell and 5G Networks
-- Baseline: Static TOO (all DL) -- Baseline: Static TOO (all UL)
0.8 ---B- Dynamic TOO (Straightforward) 0.8 ---B- Dynamic TOO (Straightforward)
---B- Dynamic TOO (wiee) ---B- Dynamic TOO (wiee)
0.6 ---"l- Dynamic TOO (wi DLPR) 0.6 ---"l- Dynamic TOO (wi DLPR)
� -- Dynamic TOO (wi ULPB) � ---+--- Dynamic TOO (wI ULPB)
o 0.4 o 0.4 -+- Dynamic TOO (wile)
L-_�___�_�_�
0.2 0.2
-10 o 10 20 30 40 50 -50 20 30
DL Wideband SINR (dB)
to-UL interference. In order to save the seriously interfered configurations defined in LTE Release 12, we add another
UEs from link failure in the UL, a large reduction on the 3 TDD configurations favoring the UL transmissions with
DL power at BSs, e.g., 6..pDL -20 dB, is shown to be
= DLIUL subframe ratios being 1/9, 2/8, and 3/7, respectively.
helpful at the expense of SINR degradation of the non-cell It should be noted that the DLIUL subframe ratio in LTE
edge UEs in the DL. An interesting note is that the cell-edge Release 12 cannot go below 2/3 [3], while in the hypothetical
UEs seem to be immune to the negative effect of the DLPR LTE Release 13 network, the ratio now freely ranges from 1/9
scheme because the benefits of decreased DL interference level to 9/l. Thus, the system can achieve full flexibility of dynamic
generally outweigh the power loss of the useful signals for TDD configuration.
the interference-limited cell-edge UEs. The CC and ULPB Figs. 3(a)-3(c) show the performance results of dynamic
schemes are also proved to be useful to improve the perfor TDD with basic ILIM in terms of 95-, 50-, and 5-percentile
mance in the UL, with little performance degradation in the DLIUL UPTs, respectively.
DL compared with the straightforward dynamic TDD scheme. As can be seen from Fig. 3(a), the performance of the
Finally, the IC scheme brings considerable improvement on the 95-percentile DL UPT is much higher than that of the 95-
UL WB SINR, especially for the cell-edge UEs. An interesting percentile UL UPT because MIMO transmissions can be
observation is that the ULPB scheme and the IC scheme seem activated for the cell-interior UEs in the DL when interference
to complement each other. The ULPB is more etlective for is low.
cell-interior UEs, while IC is more effective for cell-edge UEs.
Compared with the baseline scheme with static TDD
(Scheme 1), the straightforward dynamic TDD scheme with Tl
Hence, combined ILIM schemes are envisaged to be more
powerful than individual ones, which will be investigated in
(Scheme 3) shows solid gains in most performance categories.
following sub-sections.
However, it shows a large performance degradation of more
than 60 % in terms of the 5-percentile UL UPT when the traffic
B. DUUL UPTs with basic fLIM
load is medium to high, e.g., ,,\DL (qn,k) > 0.15. This is due to
In this sub-section, we investigate the performance of lack of ILIM to mitigate the DL-to-UL interference. Moreover,
DLiUL UPTs for dynamic TDD with various basic ILIM a faster dynamic TDD configuration time scale (Scheme 3) is
schemes. The periodicities of dynamic TDD reconfiguration shown to outperform a slower one (Scheme 2), as previously
are To 200 ms and Tl
= 10 ms for comparison purposes.
=
reported in [4].
The parameters of all ILIM schemes are the same as those in
Regarding the basic ILIM strategies, the CC scheme
Section V-A. First, we consider the following schemes with
(Scheme 4) brings a considerable improvement of more than
basic ILIM for benchmarking:
50 % on the 5-percentile UL UPT compared with the straight
• Scheme I: LTE Release 12 static TDD with TDD con- forward dynamic TDD (Scheme 3), at the expense of minor
figuration #3 [3], where the DLIUL subframe ratio is 7:3 sacrifice in DL UPTs. However, the effectiveness of the CC
• Scheme 2: LTE Release 12 dynamic TDD (To) wlo ILIM scheme declines when the traffic load is medium to high,
• Scheme 3: LTE Release 12 dynamic TDD (T1) wlo ILIM e.g., ,,\DL (qn,k) 0.25
= 0.45, because the flexibility of
rv
• Scheme 4: Scheme 3 with CC dynamic TDD is compromised considering that all the small
• Scheme 5: Scheme 3 with DLPR cells in a cluster adapt their TDD configuration according
• Scheme 6: Scheme 3 with ULPB to the aggregated traffic in the cluster rather than to their
• Scheme 7: Scheme 3 with IC individual traffic conditions. The DLPR scheme (Scheme 5) is
• Scheme 8: Hypothetical LTE Release 13 dynamic TDD shown to noticeable outperform the CC scheme (Scheme 4),
(Td with IC but with an alarming performance deterioration of around
Note that the assumed TDD DLiUL subframe splitting in 20 % in the 95-percentile DL UPT compared with Schemes 3
Scheme 1 matches the ratio of DLIUL traffic arriving rates in and 4. This is due to the poor performance of the DL
case of T 10. In Scheme 8, apart from the existing 7 TDD
= WB SINR shown in Fig. 2(a), which greatly hinders the
619
ICC'14 - W3: Workshop on Small Cell and 5G Networks
�i;�dJ
results for Schemes 3 and 6 with and without 256QAM
II
.Scheme2
when ,\DL (qn,k)
ii 0.15 .Scheme3 = 0.05. As can be observed from Table II,
.Scheme4 without 256QAM, the performance gains of the ULPB scheme
ii���
.SchemeS
(Scheme 6) over the straightforward dynamic TDD scheme
II I
0.25
I I I
.Scheme6
(Scheme 3) is around 0 % and 3.3 % in terms of the 95- and
IIII�
.Scheme7
0.45 .SchemeS
50-percentile UL UPTs, respectively, while the corresponding
� gain in terms of the 5-percentile UL UPT is around 60 %. In
so 60 40 20 o 0 20 40 contrast, with 256QAM, Scheme 6 results in about 7.56 %,
DL UPT (Mbps) UL UPT (Mbps)
30 % and 64 % higher performance than Scheme 3 in terms of
(a) 95-percentile DL and UL UPTs (basic TUM) the 5-, 50- and 5-percentile UL UPTs, respectively, showing
A,DL ( S the benefits of 256QAM. However, it is important to note that
q" ,)
Scheme 6 provides a moderate gain of 7.56 % with respect
-- 0.05 ......,. 1 Schemel to Scheme 3 in terms of the 95-percentile UL UPT. This is
1 • Scheme2
I ....
0.15
� • Scheme3
because 256QAM may still underestimate the achievable UL
UPT of the cell-interior UEs in Scheme 6. The modulation
II� ..
0.25
Q[ I
• Scheme4
• SchemeS
• Scheme6
ceiling created by 256QAM may also occur for Schemes 7
and 8. Thus, an even higher modulation scheme than 256QAM
I I
I • Scheme7 may be required, e.g.,1024QAM, though its implementation
0.45 • SchemeS
I feasibility is currently unclear due to the error vector magni
30 20 10 o o 10 20 30
tude (EVM) issues at transmitters [1].
DL UPT (Mbps) UL UPT (Mbps)
I I I .Scheme6
• Scheme 9: Combined Schemes 4 and 6
I II I III .Scheme7
.SchemeS
• Scheme 10: Combined Schemes 4 and 7
I • Scheme 11: Combined Schemes 4 and 8
20 10 0 o 10 20 • Scheme 12: Combined Schemes 6 and 7
DL UPT (Mbps) UL UPT (Mbps)
• Scheme 13: Combined Schemes 6 and 8
(c) 5-percentile DL and UL UPTs (basic TUM) • Scheme 14: Combined Schemes 4, 6 and 7
• Scheme 15: Combined Schemes 4, 6 and 8
Figure 3. DL and UL UPTs (basic rUM)
Figs. 4(a)-4(c) show the performance results of dynamic
TDD with combined ILIM in terms of 95-, 50-, and 5-
MIMO operation in the DL. The ULPB scheme (Scheme 6) percentile DLiUL UPTs, respectively.
is shown to be quite useful when the traffic load is low, e.g., As can be observed from Figs. 4(a)-4(c), the combination
,\DL (qn,k) 0.05. However, with a heavier traffic load, the
= scheme of CC and IC (Scheme 10) is strictly inferior to the
cell-edge UEs suffer from large performance degradation since combination scheme of CC and ULPB (Scheme 9), because
the power headroom of a cell-edge UE tends to be quickly the CC scheme already eliminates a few strong interfering
drained up and increasing UL power leads to more serious small cells for the UL, and thus IC only treats minor interferers
UL interference. The IC schemes (Schemes 7 & 8) are proved in Scheme 10. On the other hand, in Scheme 9, on top of
to be very effective, especially when the traffic load is not CC, UEs are granted to use a larger power that leads to a
heavy, e.g., ,\DL (qn,k) < 0.45. Scheme 8 with full flexibility better performance. When the traffic load is not heavy, e.g.,
,\DL
of dynamic TDD configuration can even achieve remarkable (qn,k) < 0.45, Scheme 9 greatly outperforms the static
gains around 30 % and 200 % for the DL and the UL UPTs, TDD scheme (Scheme 1) by around 20-40 % and 100-140 %
respectively, when ,\DL (qn,k) 0.05. = in terms of the DL and the UL UPTs, respectively. The
combination scheme of CC and IC with the hypothetical LTE
C. Impact of 256QAM on DUUL UPTs
Release 13 dynamic TDD (Scheme 11) outperforms or shows
The ULPB scheme was reported to be less effective in [6] comparable performance as Scheme 9 because of the full
than in our study. This is because the ULPB scheme in [6] was flexibility of dynamic TDD configuration.
620
ICC'14 - W3: Workshop on Small Cell and 5G Networks
Table II
COMPARISON OF THE UL UPT ( w i OR wlo 256QAM)
XDL (q""
S
DL performance. When the traffic load is low to medium,
0.05 e.g., ,\DL (qn,k) = 0.05 0.15, Scheme 13 is shown to
rv
Schemel
CScheme9
immensely outperform the static TDD scheme (Scheme 1) by
iI I,
0.15 o Scheme10
DScheme11
approximately 30-60 % and 210-300 % in terms of the DL
and the UL UPTs, respectively.
I 0.25
.Scheme12
o Scheme13
Finally, the combination of all three ILIM schemes
I I (Scheme 14) only gives similar performance compared with
I • Scheme14
I ; II I
0.45 .Scheme15 the combination of CC and ULPB (Scheme 9). Only with the
hypothetical LTE Release 13 dynamic TDD activated can the
80 60 40 20 0 0 20 40
full combination scheme (Scheme 15) outperform Scheme 9 in
DL UPT (Mbps) UL UPT (Mbps)
the UL. This is because the CC scheme is not very compatible
(a) 95-percentile DL and UL UPTs (combined IUM)
with the IC scheme, i.e., the CC scheme eliminates dominant
mnTI���ADL (q';k���mn interfering small cells for the UL, rendering the IC process
less effective.
0.05
Scheme1
0 Scheme9
To sum up, if it is preferable to find an easy-to-implement
0.15
���mm=n=ml1 0 Scheme10 scheme with reasonable performance gains, Scheme 9 should
-H 0 Scheme11
FR'ffi=fi=fi=f++t+t-t++ be called upon. But if complexity issue is a minor concern,
0.25
• Scheme12 Scheme 12 or 13 should be engaged to realize the full potential
o Scheme13 of dynamic TDD.
,+--+t++-H--+H • Scheme14
0.45
• Scheme15 V I. CONCLUSION
FH''H-1-++t+++H-++t+
-H+ • Scheme14 The authors would like to thank Jian Huang and Ruiqi
• Scheme15 Xue from Shanghai Jiao Tong University for their selfless
assistance in the simulations.
20 10 0 o 10 20
DL UPT (Mbps) UL UPT (Mbps)
REFE RENCES
(c) 5-percentile DL and UL UPTs (combined IUM)
[1] ETSI MCC, Draft Report of 3GPP TSG RAN WGI #75. Nov. 2013.
Figure 4. DL and UL upn (combined ILIM) [2] 3GPP, "TR 36.828 (Vll.0.0): Further enhancements to LTE Time Divi
sion Duplex (TDD) for Downlink-Uplink (DL-UL) interference manage
ment and tratlic adaptation," Jun. 2012.
[3] 3GPP, "TS 36.213 (VI1.2.0): Physical layer procedures," Feb. 2013.
The combined ULPB and IC schemes (Schemes 12 & 13) [4] Z. Shen, A. Khoryaev, E. Eriksson, X. Pan, " Dynamic uplink-downlink
seem to be the most powerful combinations, which substan configuration and interference management in TD-LTE," IEEE Commu
nications Magazine, vol. 50, no., II, pp. 51-59, Nov. 2012.
tially increase the UL performance due to the larger transmit
[5] Sharp, " R1-132350: DL power control based interference mitigation for
power at UEs and the IC capabilities at BSs. Some of the eIMTA," 3GPP RANI Meeting #73, Fukuoka, Japan, May, 2013.
tremendous performance gain in the UL is also shown to [6] Sharp, " R1-132351: UL power control based interference mitigation for
eIMTA," 3GPP RANI Meeting #73, Fukuoka, Japan, May, 2013.
be transferred to the DL by means of the traffic-adaptive
[7] 3GPP, "TR 36.814 (V9.0.0), Further advancements for E-UTRA physical
dynamic TDD scheduling. To be more specific, since the layer aspects," Mar. 2010.
performance in the UL is enhanced, some UL subframes [8] 3GPP, "TS 36.104 (V1l.4.0), Base Station (BS) radio transmission and
can be transformed into DL subframes thus improving the reception," Mar. 2013.
621