LTE Uplink Pe In-Band Device-Erformance With Interference - To-Device (D2D) Communic From Cations
LTE Uplink Pe In-Band Device-Erformance With Interference - To-Device (D2D) Communic From Cations
LTE Uplink Pe In-Band Device-Erformance With Interference - To-Device (D2D) Communic From Cations
Abstract—Direct device-to-device (D2D) communications network’s effect on the cellular network, and we use Monte
between mobile terminals in cellular networkss allows operators Carlo simulation to obtain values forr these metrics.
to offload proximity traffic from the Radioo Access Network
(RAN) and permits out-of-coverage terminals to maintain peer- II. D2D SCENARIOS AND
D ASSUMPTIONS
to-peer communications. In this study, w we consider D2D A variety of scenarios are possib
ble for the implementation
communications in the context of the Longg Term Evolution
(LTE) RAN and, in particular, the scenariio in which D2D
of D2D communications in an LTE E band. Among the design
communications share LTE uplink resources.. Specifically, we considerations are whether to reusee the uplink, downlink, or
evaluate the performance degradation of the ceellular LTE uplink both, and whether D2D usage will beb permitted for users who
in the presence of interference from in-band d (underlay) D2D are out of coverage of the LTE raddio access network (RAN),
communications. Through physical layer simulations, we in network coverage, or both. Th his section discusses each
quantify the increase in signal-to-noise ratio (SNR) needed to option and the assumptions made in this paper.
maintain a certain data rate or coverage criterioon as a function of
the induced noise rise and under various m multipath channel A. Uplink vs. Downlink Reuse
conditions. The results can be used to devellop physical layer
Assuming the LTE RAN uses frequency
f division duplex
models for network-layer analyses of D2D and d cellular network
performance. (FDD), a key design choice for a D2D system is whether it
shares the uplink band, downlink band,
b or both bands of the
Index Terms—Device-to-device; Long Term Evolution; public LTE RAN. In the case of uplink reuse (Fig. 1(a)), the
safety. potential victims on the infrastructu
ure network are the eNBs.
Sources of interference to the D2D D links are user equipment
I. INTRODUCTION (UE) connected to the LTE RAN an nd communicating with the
eNB. Furthermore, if the cellular system is uplink-limited, as
Cellular networks are continuing to evolvve away from the
is often the case, the interferencee power from an out-of-
traditional “hub-and-spoke” Radio Access Network (RAN)
coverage D2D link, , shown in Fig. 1(a), will be no
architecture consisting of high-power base stations
stronger, on average, than the upliink received signal power
communicating with a large population off user equipment from a cellular UE, . Hence,, the signal-to-interference
(UE). As cellular traffic goes from being doominated by data ratio (SIR) at the eNB due to a sin ngle out-of-coverage D2D
rather than telephony, following the trend of wired networks a emitter is lower-bounded by 0 dB (i.e.,
decade ago, the RAN architecture is changingg in order to meet / 1).
the resulting demand for increases in capaccity and coverage
expansion. The introduction of lower power bbase stations, i.e.,
small cells, is an example of this type of transsformation [1].
Another, more recent, development is thee recent surge of
interest in technologies to enable Device-tto-Device (D2D)
communications [2]. While D2D communnications will be
possible with or without assistance from an eenhanced Node-B
(eNB) (i.e., base station), unassisted comm munication is of
particular interest to the public safety comm munity because it
enables small groups of public safetyy personnel to
communicate in emergency situations even iif they are out of
coverage of any eNBs or relay nodes [3]. Hoowever, networks
of D2D UEs that operate as an underlay withhout coordinating
with any nearby RANs can interfere with tthose networks if
some D2D UEs are located outside but cclose to a cell’s Fig. 1: Interference in uplink and dow
wnlink reuse scenarios
coverage area. How close a D2D network ccan be to a cell’s
coverage area is thus an important architecturral question. In the case of downlink reuse (Fig. 1(b)), the potential
In this paper, we characterize the perform mance of the LTE victims on the infrastructure network are UEs, and eNBs are
uplink in the presence of co-channel interfeerence from D2D the sources of interference to D2DD links. Unlike the uplink
transmitters. We define several performancce metrics that a reuse case, no immediate conclusionns can be made on the SIR
network operator can use to determine the sevverity of the D2D
at the cellular UEs or D2D receivers. However, Doppler et al. located at the cell edge. Then, the noise rise due to a single
[4] report that uplink interference to D2D communications out-of-coverage D2D transmitter can be upper-bounded as
from cellular UEs is much lower than downlink interference follows:
from eNBs and conclude that reusing downlink resources is
therefore more challenging. In addition, as summarized by 1 1 1.
Qianxi et al. [5], a recent contribution to 3GPP on the D2D
interference issue [6] also found that the interference seen by For example, if 1.6 dB—which according to our
cellular UEs from D2D UEs transmitting on the downlink can simulations achieves a 2 % block error rate (BLER) with the
be greater than the interference seen by eNBs from D2D UEs 3GPP Extended Vehicular A (EVA) channel model [8] and
modulation-coding scheme (MCS) 0 using five resource
transmitting on the uplink, and that downlink interference is
blocks and two-antenna receive diversity1—then the noise rise
harder to control. The authors in [6] also note that allowing
is at most 2.3 dB.
D2D UEs to transmit on the downlink requires adding an
additional transmitter to the UE if FDD is being used, and that B. Increase in Required SNR
there may be regulatory obstacles to allowing D2D UEs to use An increase in the noise rise naturally translates to a
the downlink. For these reasons, we consider the uplink reuse decrease in the signal-to-interference-plus-noise ratio (SINR).
scenario in this paper. From another point of view, it also translates into an increase
in the required SNR to maintain some coverage contour or
B. In-coverage vs. Out-of-coverage
some data rate, as illustrated in Fig. 2. Typically, thermal
Another aspect of the D2D system is whether the UEs in the noise is modeled as an additive white Gaussian process, and
D2D networks are all in coverage of the RAN, out of the BLER is some function of the SNR: ⁄ . If
coverage, or a mixture, where some are in coverage and some the interference can be modeled as additive white Gaussian
are out of coverage. A key advantage of in-coverage D2D noise (which is not the case, in general), then we have
communications is tighter control of reuse of the band through
network directed resource allocation, potentially leading to ·
more efficient use of the spectrum and reduced interference.
The main motivation for out-of-coverage D2D where ⁄ is the SNR in the absence of interference and
communications is to provide the opportunity for peer-to-peer is the noise rise due to interference from the D2D
communication in unserved areas (e.g., remote areas, indoor network. If is the required SNR to achieve some target
environments, and underground). The out-of-coverage BLER, then to maintain the same BLER in the presence of
scenario is of particular interest to the public safety D2D interference, the new required SNR must be ·
community, which currently relies on the direct-mode of , and the increase in required SNR, ⁄ , is just the noise
communications of its legacy narrowband technology. While rise, . Since, in general, the interference does not fit a
the analysis of this paper is agnostic with respect to in- white Gaussian noise model, the increase in required SNR will
coverage versus out-of-coverage scenarios, the latter scenario be some unknown function of the noise rise, , and
lends itself to convenient bounds on interference, as noted · . The objective of the performance
above and discussed further below. evaluation below is to characterize the increase in required
SNR, ⁄ , as a function of the noise rise.
III. PERFORMANCE METRICS Fig. 2 shows illustrative receiver BLER curves versus SNR
When evaluating the performance of the infrastructure when a given MCS is used, for the cases where a D2D
network in the presence of D2D reuse of the uplink, a number interferer is present and where there is no D2D interference.
of metrics can be considered. This section describes the As the figure shows, the addition of an interference signal
performance metrics used in our analysis. pushes the BLER curve outward, so that the SNR is
required to achieve the target BLER that can be achieved with
A. Noise Rise SNR when no interference is present. Note that the increase
A common metric of the uplink environment in cellular in SNR, , can change with the required BLER and also
systems is the noise rise due to out-of-cell emissions. Noise depends on the MCS in use. Part of the analysis is thus
rise is defined here as the ratio of the interference-plus- determining which subset of the parameter space we need to
thermal-noise power to the thermal noise power: examine.
1
C. Decrease in Cell Coverage/Throughput/Capacity
where is the average receiver noise power due to thermal While noise rise and required SNR are fundamental metrics
energy. at the physical layer, of ultimate concern to a cellular operator
In the out-of-coverage scenario, we have observed that the and cellular users is the impact of D2D interference on higher
average interference power at the eNB from a D2D transmitter layer metrics such as throughput, cell capacity, and cell
is upper-bounded by the average received cellular UE signal coverage. From a design point of view, it may be desirable to
power. Let ⁄ be the minimum required signal-to- map these higher layer metrics to margins on lower layer
noise ratio (SNR) on the cellular uplink for a cellular UE
1
For comprehensive simulation results without D2D interference, see [7].
670
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals
metrics. For example, one might like to kknow how much performance, and average perform mance can be obtained by
noise rise can be tolerated by the cellular ssystem subject to assigning a probability of resource collision.
c
some maximum degradation in cell coverage, link throughput, For each scenario, we obtain a family of BLER-vs-SNR
and/or cell capacity. Answering such quuestions requires curves for a range of noise rise values by Monte Carlo
additional assumptions, such as a propagationn model to predict simulation. Fig. 3 illustrates an example of these BLER
cell coverage. Section V presents an exampple application of curves for the scenario in which the cellular uplink uses MCS0
the results below to the question of cell coveraage degradation. over the EVA70 channel. From this t family of curves, we
obtain the increase in required SN NR, , to maintain a
given BLER for each noise rise value. To quantify the
uncertainty of the values, wee derived 95 % confidence
intervals by treating the BLER estim mates as Gaussian random
variables with the mean and varian nce of a binomial random
variable having parameters and , where is the number of
LTE blocks simulated and is the raatio of blocks in error to .
Typically, the SNR pertaining to a target
t BLER is obtained by
interpolating between neighborin ng points. The linear
interpolation of the SNR is a functtion of the aforementioned
Gaussian random variables. We obttain the confidence interval
of the interpolated SNR by a Monte Carlo method [11].
0
10
671
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals
MCS25, possibly due to operation in a higher SNR regime. EPA5 (D2D) --> EPA5 (LTE)
1.2
The increase in required SNR is higher for MCS0 compared
with MCS15 and MCS25, which have a higher required SNR 1.15
EPA5 (D2D) --> ETU70 (LTE)
2% BLER
to begin with. The implication is that cell-edge users, who 1.1
EPA5 (D2D) --> EVA70 (LTE)
g(NR)/NR
interference. The increase in required SNR at MCS0 1.05
10% BLER
0.9
1.4
0.85
1.35 0 2 4 6 8 10
2% BLER Noise Rise (dB)
EPA5 (D2D) --> EPA5 (LTE)
1.3
EPA5 (D2D) --> ETU70 (LTE) Fig. 6: g ( NR) / NR with MCS25 due to a single interferer, with
1.25
EPA5 (D2D) --> EVA70 (LTE) 95 % confidence intervals
1.2
g(NR)/NR
1.15
1.5
1.1
0.9
0 2 4 6 8 10 1.2
Noise Rise (dB)
1
1.25
1.05
1.5
1
two interferers
0.8 1.2
0 2 4 6 8 10
Noise Rise (dB)
95 % confidence intervals
1
0.9
0 2 4 6 8 10
Noise Rise (dB)
672
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals
relative powers of the D2D transmissions at the eNB will same total average power as the single interferer has a gamma
likely differ, depending on their locations, transmission distribution with shape parameter and scale parameter 1⁄ .
powers, and fading channels. If the power of one of these The probability density functions of the received power from
interfering transmissions dominates the others, the scenario one to four interferers are shown in Fig. 9(b). As the number
can be approximated by the single interferer case examined of interferers grows, the density becomes narrower and peaks
above. However, if the interfering signals are comparable in around the total normalized mean power, and the variation in
power, their effect on the LTE uplink should also be interference power decreases. Because the probability of
considered. Following are results for the case of two equal- outliers also decreases, we expect the impact of a large
power, uncorrelated interferers. We argue that the likelihood number of interferers on the cellular uplink to be less than that
of a few interferers.
of larger numbers of equal-power interferers using the same
resource blocks as a cellular uplink transmission diminishes
V. APPLICATION OF RESULTS
rapidly, especially for cells that are not heavily loaded on the
The preceding results can be used in higher layer and other
1
exponential (single interferer)
0.9 gamma (two interferers)
gamma (three interferers)
1800 0.8 gamma (four interferers)
Single interferer
1600 Two interferers 0.7
1400 0.6
1200
f(x)
0.5
1000
Count
0.4
800
98th-percentile thresholds 0.3
- Single interferer: 2.7
600
- Two interferers: 2.0
0.2
400
0.1
200
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Normalized Power x
(a) (b)
Fig. 9: (a) Histogram of normalized interference power with EPA5 channel model;
(b) Probability density functions of cumulative interference power resulting from independent flat Rayleigh fading channels
673
2015 IEEE Wireless Communications and Networking Conference (WCNC 2015) - Track 1: PHY and Fundamentals
674