Performance of WLAN RSS-based SON
for LTE/WLAN access network selection
Yu Wang+, Relja Djapicφ, Andreas Bergström+,
István Z. Kovács*, Daniela Laselva*, Kathleen Spaeyx, Bart Sasx
+Ericsson
Research, Stockholm, Sweden; φTNO, Delft, Netherlands;
*Nokia, Aalborg, Denmark; x iMinds/University of Antwerp, Antwerp, Belgium
Introduction
§  Why SON for LTE/WLAN access selection - Motivation
§  What is it – Access selection rule and SON function
§  If it works – Simulation results in a heterogeneous network scenario
§  How does it work – Load balancing leading to user throughput improvement
§  If it really works – Various traffic model parameters
§  How dose it work in practice – Implementation issues
§  How to make it better – Summary and future work
2
SEMAFOUR
Why SON for LTE/WLAN Access Selection?
§  Operators are using WLAN for offloading
–  “Carrier grade Wi-Fi”
§  Great interests in integrating cellular and WLAN networks
§  One of the key technology component: access selection / traffic steering
§  SON for access selection
–  SON: proved working in related areas, e.g. load balancing
–  A good candidate to address the LTE/WLAN access selection
§  Task: design a SON function and prove if it works (or not) by simulation
3
SEMAFOUR
What Is It – Control Parameters and Monitoring KPIs
LTE macro
LTE micro &Wi-Fi
§  Control parameter
–  Wi-Fi RSS admission threshold
§  RSS-based access selection rule
–  Connect to WLAN if RSS > RSS_Thr
–  Access selection at session start
§  Monitoring KPIs
–  LTE: PRB utilization percentage
–  Wi-Fi: Channel busyness percentage
–  Busy: at least one node is
transmitting or attempts to transmit
–  Limitation: may not reflect load especially
with full buffer users
–  Not comparable between them
4
SEMAFOUR
UEs are served by Wi-Fi in locations with
RSS > RSS_thr
What Is It – SON Algorithm
RSS threshold initialization
§  SON function
–  Offloading
–  Onloading
KPI monitor
–  Overload protection
repeat every pSON seconds
§  Important parameters
–  Balanced LTE/WLAN user
throughput
Onloading
§  Operator policy fulfillment
–  Utilize LTE network until highly loaded
Filtering (αW)
Filtering (αL)
Yes
L(n) = (1 − α ) ∗ L(n − 1) + α ∗ Lmeasured (n)
No
Offloading
–  Congestion/High/low load thresholds
LTE load
measurement
(period: pL)
LTE & Wi-Fi
congested
–  Filtering factor
–  SON update period and step size
Wi-Fi load
measurement
(period: pW)
LTE high
load
No
Wi-Fi high
load
No
LTE
LTE low
low
load
load?
No
Wi-Fi low
load
No
Yes
Yes
Yes
Yes
Decrease RSS
threshold
Increase RSS
threshold
Increase RSS
threshold
Decrease RSS
threshold
RSS threshold
regulation
5
SEMAFOUR
If It Works – Simulation Scenario
§  Environment: 1.1 km2 urban area located in city center of Hannover
§  Deployment: 5 LTE Macros (46 dBm), 28 co-located outdoor LTE Micro (33
dBm) / AP (20 dBm)
§  6 dB cell extension offset for Micros
§  Spatial traffic distribution: Outdoor busy hour traffic + a hotspot with 70%
traffic outdoor
§  Offered traffic: Fixed size UDP downloading / uploading (DL: 5 MB; UL: 0.8
MB; 12 arrival/sec)
ground floor only
§  No external interference
§  Mobility: Stationary users only
1200
LTE Macro
Macro: 20MHz@1.8GHz
LTE Micro
& WLAN AP
Micro: 20MHz@1.8GHz
AP:
20MHz@5GHz
Meter
§  Indoor users are located on the
User Location
Macro Index
Micro/AP Index
0
0
6
SEMAFOUR
Meter
1200
If It Works – Fixed RSS Threshold
§  Highest average and 5th percentile user
throughput
–  Fixed RSS threshold = -65 dBm.
User Session Percentage (%)
100
80
LTE Macro
LTE Micro
Wi-Fi
60
40
20
0
-92 -85 -75 -65 -55
RSS Threshold (dBm)
-45
5%-ile User
Average User
Througphut (Mbps) Througphut (Mbps)
–  User session split:
–  WLAN: 42%
–  LTE Micro: 37%
–  LTE Macro: 21%
40
20
0
-92 -85
-75
-65
-55
-45
10
5
0
-92 -85 -75 -65 -55 -45
RSS Threshold (dBm)
If It Works – SON Function to Improve User Throughput
§  14% gain in 5th percentile user throughput comparing to the best fixed threshold
reference case
§  Sensitivity to SON parameter configuration
–  Not very sensitive to the load filtering factor
–  Need to be fast enough to capture traffic load variation
§  Balanced DL user throughput among users
§  UL user throughput is not optimized with the DL KPI driven algorithm
SON: fast RSS_Thr update
PSON/PL/PW=0.5s, RSS_Thr_Step=2dB, α=0.9
8
SEMAFOUR
WLAN
users
(%)
0.1 / 11
87
70
7.2 / 31.6
42
60
8.2 / 31.7
41
7.5 / 33.1
45
All (DL)
Macro (DL)
Micro (DL)
WLAN (DL)
All (UL)
Macro (UL)
Micro (UL)
WLAN (UL)
90
80
CDF [%]
Reference: WLAN if coverage
Fixed RSS_Thr = -92 dBm
Reference: Best Fixed Threshold
Fixed RSS_Thr = -65 dBm
SON: reference settings
PSON/PL/PW=1s, RSS_Thr_Step=1dB, α=0.9
SON: smooth load filtering
PSON/PL/PW=1s, RSS_Thr_Step=1dB, α=0.5
SON: slow RSS_Thr update
PSON/PL/PW=2s, RSS_Thr_Step=0.5dB, α=0.9
100
5%-ile / Avg.
user thrp.
(Mbps)
50
40
30
3.7 / 28.2
54
7.1 / 31.2
38
20
10
0
0
5
10
15
20
25
30
User Throughput (Mbps)
35
40
45
How Does It Work – Load Balancing
§  A more balanced load between the two system
–  Load was kept in the configured range by the SON function
60
40
20
0
1 2 3 4 5 1 2 3 4 5 6 7 8 9 10 1112 13 1415 1617 18
-40
-45
-50
-55
-60
-65
-70
-75
-80
100
Cell load
80
60
40
20
0
1 2 3 4 5 1 2 3 4 5 6 7 8 9 10 111213 141516 1718
SON function
9
SEMAFOUR
-40
-45
-50
-55
-60
-65
-70
-75
-80
RSS threshold (dBm)
Fixed Threshold = -65 dBm
100
RSS threshold
WLAN AP load
LTE cell load
80
60
40
20
0
20
40
60
80
Time (sec)
100
-40
-45
-50
-55
-60
-65
-70
-75
-80
120
RSS threshold (dBm)
Cell load
80
Load (%)
100
RSS threshold (dBm)
§  Convergence of the RSS threshold
–  ‘Warming-up’ and stabilization
If It Really Works – Various Traffic Parameters
§  Smaller file size and higher user arrival rate: More simultaneously active users à
Lower WLAN efficiency and higher WLAN load à SON moves more users to LTE
§  Larger file size and lower user arrival rate: Approaching to full buffer traffic à
Resource utilization is not a good load measure
§  Lower offered traffic: SON outperforms the fixed RSS threshold in the 5th percentile
user throughput
[File size: DL/UL,
User arrival rate]
Reference
[5 MB/0.8 MB, 12]
[2.5 MB/0.4 MB, 24]
[1.25 MB/0.2 MB, 48]
[10 MB/1.6 MB, 6]
[5 MB/0.8 MB, 6]
10
SEMAFOUR
SON with the
reference settings
5%-ile /
WLAN
Avg. user
users
thrp.
(%)
(Mbps)
RSS_Thr = -65 dBm
5%-ile /
Avg. user
thrp.
(Mbps)
WLAN
users
(%)
8.2/31.7
41
7.2/31.6
42
7.2/27.8
5.8/23.1
6.6/33.0
10.7/38.1
39
34
40
48
6.6/26.8
3.9/21.8
7.2/35.6
8.3/40.1
42
42
42
42
How it Works in Practice
§  The RSS-based SON algorithm is primarily a distributed solution because it may
be difficult for the central controller, e.g. ANDSF, to execute an algorithm which
updates parameters every few seconds.
§  Implementation of the RSS-based access selection rule
–  Executed in terminals:
–  RSS_Thr is send to terminals via broadcasted or dedicated signalling channels
being standardized in 3GPP
–  Executed in network:
–  Uplink RSS is monitored at the network node and the node controls the access
selection
§  Information exchange between LTE and WLAN
–  Standardization of such an interface is being discussed in 3GPP
–  Proprietary interfaces
–  Terminals as relays
§  WLAN RSS measures
–  RSSI: Received Signal Strength Indicator
–  RCPI: Received Channel Power Indicator
11
SEMAFOUR
How to Make It Better
§  Simulation results showed a RSS-based SON algorithm for access network
selection between the LTE and WLAN systems successfully
–  Balanced the load between the two systems
–  Optimized user throughput statistics with various traffic parameter settings.
§  Future work
–  The optimization of the SON parameters with respect to the traffic parameters is
subject to further study.
–  Better load measure
–  Evaluate the SON function in more dynamic scenarios: mobility and and/or
mixed traffic types.
–  More understanding of the coexistence of distributed and centralized SON
functions is needed to achieve a holistic solution and further optimize the
performance.
12
SEMAFOUR