Performance Comparison of Dual

Connectivity and Hard Handover for

LTE-5G Tight Integration!

Michele Polese*, Marco Mezzavilla+, Michele Zorzi*!

*Department of Information Engineering, University of Padova, Italy!
e-mail: {polesemi, zorzi}@dei.unipd.it!
+Tandon School of Engineering, New York University, Brooklyn NY, USA !
e-mail: mezzavilla@nyu.edu!

SIMUTools 2016, Prague, Czech Republic!

August 23rd, 2016!
 Outline!
2!

LTE-5G Tight Integration!

Dual Connectivity Architecture!

Extension of NYU ns-3 mmWave Simulator!

Metrics and Preliminary Results!

Conclusions and Future Work!

 5G networks!
3!

Different targets:!

Very high bandwidth!

mmWave communications!
Ultra-low latency!

Massive number of devices! Sub 1 GHz comms!

Different targets different technologies!

AM
MmWave Communications!
Page 102

4!

3 GHz 57 64 164 200 300 GHz

54 GHz 99 GHz 99 GHz

All cellular mobile Potential 252 GHz

communications available bandwidth

60 GHz oxygen Water vapor (H2O)

absorption band absorption band
(a)

28 GHz!
27.50 28.35 73 GHz!29.10
28.60 29.25 29.50 31.075 31.225 GHz

150 75 150 75
Z. Pi and F. Khan, 850introduction
AntoMHz millimeter-wave
MHzmobile broadband M MHz M
systems, !
Hz Hz
IEEE Communications Magazine, vol. 49, no. 6, pp. 101107, June 2011 !
 A Case for LTE-5G Tight Integration!
5!

mmWave: very high throughput!

BUT!
Variable signal quality, possible link failures!

LTE network as fallback coverage layer!

Hard Handover (HH)! Dual Connectivity (DC)

with Fast Switching!
 Dual Connectivity Architecture!
6!

Single PDCP layer in the coordinator !

(new node or LTE eNB)!

RLC entity in the LTE and mmWave eNBs!

Single connection to Core Network!

Switch: !

a RRC message to UE !

X2 notification to mmWave eNB!

 NYU ns-3 mmWave Simulator!
7!

Channel model based on real measurements!

Fully configurable TDD physical layer!

MAC layer with HARQ, scheduler!

Upper layers (RLC, PDCP, RRC) from LTE ns-3

module!
 Extension to LTE-5G Integration!
8!
J+l2L2i.2pB+2

1T+l2Lb

Gi2l2_`+ J+l2S/+T KKqp2l2_`+

Gi2_H+ KKqp2_H+
Dual Connected UE!
Gi2l2J+ KKqp2l2J+
Gi2l2S?v KKqp2l2S?v

Gi2aT2+i`mKS?v KKqp2aT2+i`mKS?v

JmHiBJQ/2HaT2+i`mK*?MM2H *?MM2H +Hbb2b

New PDCP layer!
JmHiBJQ/2HaT2+i`mK*?MM2H M/ KKqp2"2K7Q`KBM;

Gi2aT2+i`mKS?v KKqp2aT2+i`mKS?v
PDCP-RLC forwarding on X2!
Gi21M#S?v KKqp21M#S?v
Gi21M#J+ KKqp21M#J+

Gi2_H+
Mi2`
7+2
KKqp2_H+ Integration of LTE and
sk B
Gi21M#L2i.2pB+2 J+1M#S/+T JKqp21M#L2i.2pB+2
mmWave channels!
Gi21M#_`+ KKqp21M#_`+

1T+1M#TTHB+iBQM
aR BMi2`7+2
*Q`2 L2irQ`F
 Main Features!
9!

Procedures for Fast Switching:!

Initial Access, Secondary Cell Handover, Switch!

mmWave SINR estimation with reference signals!

RLC with finite-size buffers!

X2-based handover between LTE and 5G!

 Metrics!
10!

Throughput at different layers!

Packet losses!

Latency at different layers!

Control traffic (RRC)!

X2 and S1 traffic!
 20

0
Simulation Scenario! -20
UE UE path at speed s

11!
0 50 100 150 200 250 300 350 400
X [m]

Figure 3: Simulation scenario. The grey rectangles are

buildings

100
Parameter Value
80
Outage threshold 5 dB
60 mmWave carrier frequency 28 GHz
mmWave bandwidth 1 GHz
LTE carrier frequency (DL) 2.1 GHz
Y [m]

40
LTE bandwidth 20 MHz
20 X2 link latency DX2 1 ms
RLC AM buer size BRLC 10 MB
0 S1-MME link latency 10 ms
s UDP packet size 1024 byte
-20 UDP packet interarrival 80 s
UE speed s 2 m/s along the x axis (Fig. 3)
0 50 100 150 200 250 300 350 400 Iterations N = 10
X [m]
Table 1: Simulation parameters

random number generation the channel varies in the same

way. Table 1 contains some parameters of the simulations,
and the scenario is shown in Fig. 3. The trac is gener-
K2i2`b 7Q` i?2 P6.J 7`K2 bi`m+im`2 `2 i?2 bK2 Q7 h#H2 9XkX //B@
ated at a rate of 102.4 Mbit/s at the UDP transport layer,
 Example: throughput over time!

Throughput [Mbit/s]
400
3
Cell Throughput
9!
Cell

2 200

DC setup! 1
1.2
104 0
34
100 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

Throughput [Mbit/s]
400
3 Time t [s]
Cell Throughput
LTE eNB 1
3 80 8 102

RRC throughput [bit/s]

Cell

2 CellId Latency 200 DC, s

1

Latency [s]
6 102
DC, s
60
Y [m] Cell

2
mmWave eNB 2 mmWave eNB 3 4 102 DC, s
1
0
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 DC, s
40 Time t [s] 0.8 2 102

1 HH, s
0
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
HH, s
20
Current cell in time! PDCP throughput over time!
HH setup!
3
CellId
Time t [s]
Latency
8 102
HH, s

Latency [s]
(a) Dual Connectivity setup 0.6
6 102
HH, s
0
Cell

Throughput [Mbit/s]
2 400
4 102
3
Cell UE UE path at speed s Throughput
300
-20 0.4
2 102
Cell

1 0.1 1 10
234 0
200
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

0 50 100 150 200 250 300 350

Time400
t [s] DX2 [ms]
100

1 X [m] (a) Dual Connectivity setup 0

34 35 36 37 38 39 40 41 42 43 44
Figure
45
4: 46Average
47
RRC
48
trac
49
per
50
user for DC

Throughput [Mbit/s]
400
3 Time t [s]
Cell Throughput
Figure 3: Simulation scenario. The grey rectangles are 300

buildings
3
PDCP Throughput
0.4
RLC
Cell

2 CellId Latency 200

Latency [s]
0.3

Dual Connectivity 106.70 Mbit/s0.2 100 5.1 m

Cell

2
Parameter
1
34 35 36 37
Value
38 39 40 41 42 43
Hard
44
Handover
45 46 47
104.98
48
Mbit/s
49 50
0 18.1
0.1
Time t [s]
 LTE eNB 1

Throug
RRC throughput [bit/s]
DC, s =
1
1

Example: latency over time!

0
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 400

Throughput [Mbit/s]
3
Cell Time t [s] Throughput
DC, s =
mmWave eNB 2
Cell mmWave eNB 3 DC, s =
8 102
9!
32
CellId Latency
200 DC, s =
0.8

Latency [s]
6 102
HH, s =
Cell

21 4 102
HH, s =
Different scales!
0
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

DC setup! 1
Time t [s]
0.6
HH, s = 2 102

34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
0
50 8 102 HH, s =
3
CellId Time t [s] Latency
UE UE path at speed s

Latency [s]
6 102
(a) Dual Connectivity setup
Cell

2 4 102
0.4

Throughput [Mbit/s]
400
3 0.1 1 10
Cell Throughput 2 102
0 50 100 150 200 250 300 350 400
1
DX2 [ms]
0
300
Cell

34
2 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 200

X [m] Time t [s]

Figure 4: Average RRC traffic per user
100
for DC and
(a) Dual Connectivity setup

HH setup! Current cell in are

time! RLC latency over time!
1
0
Simulation scenario.
The grey rectangles
35 34 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 400

Throughput [Mbit/s]
3
Cell Time t [s] Throughput
PDCP Throughput300 RLC Lat
0.4
Cell

32
CellId Dual Connectivity 106.70 Mbit/s
Latency
200
5.1 ms

Latency [s]
0.3
Hard Handover 104.98 Mbit/s 100
18.1 ms
Value
Cell

21 0.2
0
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50

shold 5 dB Time t [s]

Table 2: Throughput and latency for the two setups,
0.1

1
rrier frequency34
3
35 28 GHz36 37 38 39 40 41 42
parameters
43 44
in Table
45 46
1, average
47 48
over
49
N =
0
50 0.4
10 simulatio
Time t [s]
ndwidth 1 GHz
CellId Latency

Latency [s]
0.3

frequency (DL) 2.1 GHz (b) Hard Handover setup

Cell

2
idth Figure 2: Examples 20 MHz RLC Latency 0.2

of simulation output: throughput and latency over time, over the cell to which the UE is connected.
0.1 Cells
ncy DX2 2 and 31 are the 1 ms eNBs, cell 1 is the LTE eNB
mmWave Dual Connectivity 5.1 ms
0
uer size BRLC 34 35
10 MB 36 37 38 39 40 41 42
Time t [s]
43 44
Hard45
Handover
46 47 48
18.1
49
ms50

k latency 10 ms
 RRC Traffic!
12!

.*- s = 2 Kfb >>- s = 2 Kfb

.*- s = 4 Kfb >>- s = 4 Kfb
.*- s = 8 Kfb >>- s = 8 Kfb
104 .*- s = 16 Kfb >>- s = 16 Kfb
1.2

1
__* i?`Qm;?Tmi (#Bifb)

0.8

0.6

0.4 0.1 1 10

DX2 (Kb)
 Conclusions!
14!

LTE-5G Dual Connectivity architecture!

Extension of NYU mmWave ns-3 simulator!

Examples of metrics that can be collected!

DC performs better than HH, for more results see [1]!

Flexible framework, it opens many research directions!

[1] M. Polese, Performance Comparison of Dual Connectivity and Hard Handover for LTE-5G
Tight Integration in mmWave Cellular Networks, Masters thesis, Dept. of Information
Engineering, University of Padova, July 2016. Available at http://arxiv.org/abs/1607.04330 !
 Performance Comparison of Dual
Connectivity and Hard Handover for
LTE-5G Tight Integration!

Michele Polese*, Marco Mezzavilla+, Michele Zorzi*!

*Department of Information Engineering, University of Padova, Italy!
e-mail: {polesemi, zorzi}@dei.unipd.it!
+Tandon School of Engineering, New York University, Brooklyn NY, USA !
e-mail: mezzavilla@nyu.edu!

SIMUTools 2016, Prague, Czech Republic!

August 23rd, 2016!