PDCCH LINK ADAPTATION tuning

Report
PDCCH LINK ADAPTATION tuning
Summary
• In the attempt to improve the performance (throughput) some
parameters controlling the use of the PDCCH resources were tuned.
• The results is indicating that significant benefits are possible
– Highest gains can be expected in multi user scenarios
– Exact gain will depend on traffic pattern and radio environment
• The parameter changes and tests was done lab in a network without
commercial traffic.
• The following parameters were tuned:
– pdcchCfiMode
– pdcchLaGinrMargin (system constant)
– nrOfTransmissionsSib1
• Some of the parameters might not be suitable for a live network
deployment
Lab Setup & Execution
Single cell – L12B ICP4 software – 5MHz Bw @2.1GHz
•
•
•
Spirent VR5 Channel Emulator (mainly EPA5 channel used)
Full buffer UDP or FTP traffic
Programmed Attenuation sweep
– from ~100dB path loss  UE release (~150dB) (1dB steps, 30sec/step)
•
UE’s and Logging Tools
– Qualcomm 8960 (FFA) & Quanta (Qualcomm chip)
•
All data shown from UE side (not network side data included)
Definitions & Data Sources
• Path Loss
– As reported by Qualcomm UE (based on RS power)
• SINR
– As reported by Qualcomm UE (probably based on C-RS?)
• UL & DL Scheduling Ratio
– Ratio of subframes when the UE DRB data is scheduled to
total subframes (ms) in averaging window
• BLER
– Ratio of successful Redundancy Version 0 (RV0)
transmissions to all RV 0 transmissions
Characteristics – Default parameters
5MHz Bw – EPA5 – Single UE - UDP DL full buffer traffic
UDP Downlink – Expected peak TP
(35M) not reached
Only RLC status and polling
sent in uplink
Uplink Limited  lowest MCS not reached
No. of PRB reduced in good RF env.
Sched. ratio decreased
with decreasing RF env.
Scheduling Ratio & PRBs
Detailed View
Sched. ratio reduced in poor RF
especially for subframe 0 & 5
In good RF 25, 22, 19 and 5 PRBs are used
In poor RF only 25 PRBs are used
The PRB reduction in good RF is mainly
in subframe 0 and 5
System Information (SI) is scheduled in subframe 0 and 5 with various repetition
PDCCH Control Chanel elements
Control Channel Element
1
2
3
4
5
6
7
GOOD Radio Env.
Low Robustness
1 CCE – 8 spaces

Low PDCCH
Capacity Consumption
0
4 CCE – 2 spaces

8 CCE – 1 space
POOR Radio Env.
High Robustness

High PDCCH
Capacity Consumption
2 CCE – 4 spaces
Fewer CCE used for each PDCCH allow for more users scheduled in parallel
PDCCH Link Adaptation
•
•
•
•
One PDCCH can be mapped to 1, 2, 4 or 8 CCEs
Selection of number of CCEs is done based on the same GINR estimate used for
PDSCH link adaptation
A margin (back-off) is added to PDSCH GINR to compensate for different
interference scenarios between the two channels
Degraded RF conditions 
– Higher CCE Aggregation level
– Lower MCS (for the same number of PRB’s)
•
If the required number of CCE’s are not available in the control region (1) the UE can
not be scheduled on the PDSCH (2) in that subframe
DL scheduler
& Link Adaptation
1CCE
2CCE
4CCE
8CCE
eNodeB
UE
PDCCH Candidates for 5MHz
Example
•
•
•
•
•
•
Exact candidate positions depend on x-RNTI and subframe number
Number of candidates increasing with number of PDCCH OFDM symbols (CFI)
Number of candidates is dependent on number of PHICH groups configured (def. 3)
Both Uplink and Downlink Scheduling need the PDCCH resources
Uplink scheduling has priority over downlink scheduling
Common is always using 8 CCE to ensure full cell coverage (Ericsson)
CFI = 1
No Common scheduling possible
No Poor/Medium coverage UE Scheduling possible
CCE
0 1 2 3 4 5 6 7
1U U U
2U
4
8
CFI = 2
One 8 CCE UE’s or common can be scheduled in the
same Subframe
CCE
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1
U U U U U U
2U
U
U
U
U
U
4
CU
CU
C
8
CU
CFI = 3
Two 8 CCE UE’s (or common + one 8 CCE UE)
can be scheduled in the same subframe
CCE
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
1
U U U U U U
2U
U
U
U
U
U
4
C
C
CU
CU
8
CU
CU
U
C
UE specific PDCCH candidates
Common PDCCH candidates
CCE Aggregation Level vs. SINR and
MCS
CCE Use increasing at low SINR
PDSCH Link Adaptation (MCS) follow SINR
PDCCH LA follow PDSCH LA
Problem Summary
Reduced Sched. ratio due to
uplink scheduling (higher priority)
Reduced Sched. ratio due to
SI scheduling (higher priority)
•
In good radio environment (1 & 2 CCE required)
–
–
•
DRB (user) data can be scheduled on remaining PRB:s when SI is scheduled since there is enough space
in the PDCCH
Possible to fit both uplink and downlink scheduling in the PDCCH
In poor radio environment (8 CCE required)
–
–
–
No DRB (user) data can be scheduled when the SI is schedule due to lack of PDCCH resources (only one
8 CCE candidate)
No downlink data can be scheduled when there is uplink data to schedule
Reduced downlink scheduling ratio
Solution Summary
Optimized Parameter Setting
•
•
•
Increase max number of PDCCH symbols to 3
– pdcchCfiMode: 0  5
– 0: (Static by bandwidth – 2 for 5Mhz), 5: (Adaptive CFI, max 3)
Reduce SIB Scheduling
– Sib 1 is scheduled every 80ms and repeated 4 times, i.e. every 20ms
(default)
– nrOfTransmissionsSib1: 4 1 (i.e. 1 repetition = every 80ms)
Reduce PDCCH Link Adaptation Margin (relative PDSCH LA)
– Delaying the CCE increase to a lower SINR will
make the PDCCH less robust
– Analysis needed to find suitable tradeoff
Tuning of PDCCH LA Margin
•
Low Margin  UE will fail to decode the PDCCH
– UE will consider itself not scheduled  Reduced Scheduling Ratio as seen from the UE side
•
High Margin  UE will use unnecessarily many CCE’s
– Increasing the probability of lack of PDCCH resources  Reduced Scheduling Ratio
– In case of a single UE with DL UDP traffic this will only happen when SI is scheduled*
•
Strategy:
– Reduce margin until a distinct decrease in Scheduling Ratio is seen*
– Initially test in large steps, reduce later
•
Margin steps: 10dB, 6dB, 3dB, 0dB -3dB & -5dB
DL Scheduling Ratio vs. MCS
10- -5dB Margins - EVA70 Evaluation
Reference: 10dB
(default)
3dB – no degradation
-3dB –obvious
degradation
6dB – no degradation
0dB – no obvious
degradation
-5dB –severe
degradation
analysis
EVA70 Evaluation
• Based on the EPA70 channel tests it was found that the
Scheduling ratio was reduced with a PDCCH LA margin of
between 3dB and -3dB
• A focused test in the range of 4dB to -3dB margin was made
– Step size: 0.5dB
• Channel model was changed to EPA5
• Results are shown as
– Downlink Scheduling Ratio vs. MCS
– Distribution of PDCCH Aggregation level vs. MCS
DL Scheduling Ratio & CCEs vs. MCS
4- 3dB Margins – EPA5
Reference 4dB Margin
3.5dB Margin
No degradation
3dB Margin
No degradation
DL Scheduling Ratio & CCEs vs. MCS
3- 2dB Margins – EPA5
3dB Margin
No degradation
PDCCH decoding failures
2.5dB Margin
PDCCH decoding failures
2dB Margin
DL Scheduling Ratio & CCEs vs. MCS
1.5- 0dB Margins – EPA5
PDCCH decoding failures
1.5dB Margin
PDCCH decoding failures
1dB Margin
PDCCH decoding failures
1dB Margin
LA Margin Selection
• Based on the results the PDCCH LA margin was selected as 3dB
– Slightly lower margin might be possible without major degradation
• Considering the limited test scenarios and channel models evaluated the
chosen setting was considered reasonably safe for the non commercial
network
3dB Margin
No degradation relative to
the reference case
Evaluation of performance improvement
multi ue scenario
• 2 UE’s in the same radio environment (EPA5 channel)
– UE1: Full buffer TCP Downlink Traffic
– UE2: Full buffer TCP Uplink Traffic
– In this scenario the UE’s are competing on the PDCCH space rather than PDSCH
• TCP Traffic is generating additional traffic in the reverse direction
– TCP ACK’s
• Programmed Attenuation Sweep of both UE’s together (splitter)
– From ~100dB path loss  UE release (~150dB) (1dB steps, 30sec/step)
• 3 Cases Evaluated:
– Default Parameter Set
– Flexible CFI with a max of 3 symbols
– Fully Optimized Parameters
• PDCCH LA margin – 3dB
UE2
eNodeB
UE1
UE1: TCP Downlink UE
Downlink & uplink Throughput
Improved Thp. (~ 2MBps) from
Improved scheduling ratio
Increased UL Thp. Is an effect of
increased rate of TCP ACK’s
Increased sched. Ratio at low
SINR due to uplink PRB reduction
UE1: TCP Downlink UE
Downlink Scheduling ratio per subframe number
Default Parameters
Flexible CFI (max 3)
Optimized Parameters
UE2: TCP Uplink UE
Downlink & uplink Throughput
Increased DL Thp. Is an effect of
increased rate of TCP ACK’s
Improved Thp. (~ 1MBps) from
Improved scheduling ratio
Main improvement from Flexible CFI
Field performance
Filed performance
optimized parameter setting
• The optimized parameter setting was tested in the field
– ~1 Hour drive route in suburban-rural area
• Both unloaded and 100% OCNG downlink load was tested
• Single UE FTP download
– Limited file size causing idle periods during drive test
• EPA5 Lab result curve included as reference
Field: Throughput performance
Suburban-rural area
The radio environment in field appears a bit less challenging for the UE than
EPA5 in the Lab
Field: Scheduling ratio
Suburban-rural area
In general lower scheduling ratio during field test relative to lab
Handovers, quick variations in radio environment in connection to the TCP protocol
etc. could account for this lowered scheduling ratio
Samples with very low scheduling ratio is due to idle times between filed
downloads
THNAK YOU

similar documents