Transcript Media routing - Uppsala University

Adaptive modulation and multiuser
scheduling gains in adaptive
TDMA/OFDMA systems in the WINNER
framework
Sorour Falahati , Mikael Sternad,
Tommy Svensson, Daniel Aronsson
Uppsala University
Chalmers University of Technology
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 1
Outline
Introduction
• FDD downlink and uplink structure
• Timing events in DL/UL transmission
• Key techniques
•
–
–
–
–
Channel prediction
Scheduling
Link adaptation
Compression of feedback information
Simulation results
• Summary
•
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 2
Introduction
•
Predictive adaptive resource scheduling
using TDMA/OFDMA:
– Providing fast link adaptation in an OFDM
system based on the predicted channel state
information of time-frequency chunks
–
Providing multi-user scheduling gain by
allocating the resources to the flows with the
potential of improving the throughput based
on their channel status.
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 3
FDD downlink and uplink
structure
FDD downlink
Chunk
D D
P
U
U
P
P: pilots symbols
8 sub-carriers
BW
Chunk
freq
D: DL control symbols
D D
P
6 TOFDM
U
U
P
U: UL control symbols
time
freq
O
FDD uplink
O
C
C O
C
C O
O: overlapping pilots
C
C O
C
C O
C: DL control feedback
O
O
T chunk
time
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 4
Timing events in DL/UL
transmission
DL
DL
UL
UL
O
O
D D
P
U
U
P
O
D D
P
U
U
O
P
1. DL control symbols:
C
C O
C
C O
Report which present chunks
C
C O
C
C O
belong to which flows
C
C O
C
C O
P C
C O
P C
C O
D D
P
U
U
D D
P
U
U
2. DL pilot symbols:
O
O
Used for channel prediction
O
O
Used for channel estimation
Dl prediction horizon: 2.5X0.3372ms=0.843ms
UL prediction horizon: 2.5X0.3372ms=0.843ms
3. UL control symbols:
Report which next UL chunks
5. DL control feedback symbols:
appointed to which uplink flows
Carry DL channel prediction report
4. Pilot symbols:
Used for coherent detection
And updating predictor states
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
6. UL overlapping pilot symbols:
Used for prediction
Slide 5
Channel prediction
•
•
•
Prediction in frequency domain
A set of linear predictors, one for each sub-band
Kalman predictor:
–
–
Predict the complex channel and its power
Using pilots in parallel sub-carriers:
• Utilizing correlation in frequency and time domain
•
Generalized Constant Gain (GCG) algorithm:
• No need to update a sate-space Riccati difference eq.
• Moderate complexity and negligible performance loss as
compared to Kalman algorithm
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 6
Channel prediction …
• SINR and prediction horizon limit at 5 GHz downlink:
30 km/h
<0 dB, 0.117 
50 km/h
6 dB, 0.195 
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
70 km/h
12.5 dB, 0.273 
Slide 7
Channel prediction …
• SINR and prediction horizon limits at 5 GHz uplink:
2 users
30 km/h
<0 dB, 0.117 
50 km/h
7 dB, 0.195 
8 users
70 km/h
15 dB, 0.273 
30 km/h
3.5 dB, 0.117 
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
50 km/h
11dB, 0.195
70 km/h
20 dB, 0.273 
Slide 8
Scheduling
•
Resource scheduling
– Proportional fair strategy:
• Allocating resources (chunks) to the user with the
highest SINR relative to its average
• For users with the same average SINR, this strategy
reduces to Max. Throughput strategy.
– Allocating chunks to users with the highest MC rate.
• Due to curvature within the chunk, MC scheme is
determined based on:
SINR[dB]  b  SINRav [dB]  (1  b)SINRw[dB]
Chunk Average SINR
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Chunk minimum SINR
Slide 9
Link adaptation
•
Each user selects a modulation and coding (MC)
scheme for each chunk in competition based on the
prediction SINR:
–
The rate limit for a set of MC schemes are adjusted based on
the TBER, average SNR and prediction error variance
–
Based on the predicted chunk SINR, a MC scheme which
fulfills the BER requirement and maximized the throughput is
selected.
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 10
Link adaptation …
•
BER performance of MC schemes for perfect and
imperfect prediction (NMSE=0.1)
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 11
Link adaptation …
•
•
Variation of rate limits of MC schemes with prediction quality:
SNR=10 dB and TBER=0.001
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 12
Compression of feedback
information
•
Tricks or tools to reduce downlink
overhead:
– Use implicit signaling of utilized modulation
rate whenever possible
– Contention-band: The active terminals are in
competition for only a part of the total BW
– Use short-hand addresses to indicate
identities of active users whenever possible.
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 13
Compression of feedback
information …
Tricks or tools to reduce uplink overhead:
– Contention-band
– Compression of feedback information
104
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Subsampling
104
Chunk
SNR
104
Tx feedback
Scheduling
IDCT
Scalar
Quantizers
DCT
104
104
Upsampling
Received
feedback
104
Q-1
Rx
Calculate
chunk SNR
• Discrete cosine transform: utilizing correlation in frequency
• Sub-sampling of transform coefficients in the time domain
FFT
•
Slide 14
Compression of feedback
information …
•
•
THP as a function of feedback rate:
ITU VA channels, v=50 km/h, sub-sampling factor of 2
10 users
5 users
1 user
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 15
Simulation results
•
Simulation set-up
–
–
–
–
Wide-area full-duplex FDD downlink
WINNER Urban Macro channel model
Single cell (sector) and SISO
Users with equal velocities and average SINRs
Center frequency
5.0+/-0.384 GHz
Number of OFDM sub-carriers
1024
FFT BW
20 MHz
Signal BW
16.25 MHz paired
Number of used sub-carriers
832
Sub-carrier spacing
19531 Hz
OFDM symbol length (exc. CP)
51.20 microseconds
Cyclic prefix (CP) length
5.00 microseconds
Physical chunk size
156.24kHz x 337.2 microseconds
Chunk size in symbols
8 x 6=48
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 16
Simulation results …
•
•
Multi-user diversity, channel variations:
THP versus SNR for 2 and 8 users
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 17
Simulation results …
•
•
Multi-user diversity, channel variations:
BER versus SNR for 2 and 8 users
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 18
Simulation results …
•
•
Prediction quality, multi-user diversity, channel variation:
THP versus number of users (19 dB)
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 19
Simulation results …
•
•
Prediction quality, multi-user diversity, channel variation:
BER versus number of users (19 dB)
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 20
Simulation results …
•
•
Prediction quality, multi-user diversity, channel variation:
THP versus number of users (10 dB)
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 21
Simulation results …
•
•
Prediction quality, multi-user diversity, channel variation:
BER versus number of users (10 dB)
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 22
Simulation results …
•
•
TDMA/OFDMA versus use of TDMA:
THP versus number of users (19dB)
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 23
Summary
•
An adaptive transmission based on TDMA/OFDMA
using multiuser scheduling is investigated.
–
Predictive adaptation to the short-term fading and
frequency-domain channel variability leads to significant
multi-user diversity gain.
–
With TDMA instead of TDMA/OFDMA, only half of these
gains are realized for channels with Urban Macro
scenarios.
–
Predictive adaptation can use MC rate boundaries
adjusted so that BER constraints are fulfilled in the
presence of SINR prediction uncertainty.
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 24
Summary…
– Feasibility of adaptive transmission is limited by
prediction accuracy.
• Prediction accuracy is determined by SINR and
terminal velocity.
• For realistic SINR values, transmission at 50 km/h is
feasible at 5 GHZ in FDD DL.
– A solution to reduce the required feedback rate:
• To feed back the required SINR and source code it by a
combination of transform coding in the frequency
direction and sub-sampling in the time direction.
Event: CUBAN/WIP Workshop - Tunisia
Date: 23-25 May 2005
Slide 25