An Adaptive GroupedSubcarrier Allocation Algorithm Using Comparative Superiority

1
An Adaptive Grouped-Subcarrier Allocation
Algorithm Using Comparative Superiority

• Youngok Kim, Haewoon Nam, and Baxter F. Womack
• Dept. of Electrical and Computer Engineering
• The University of Texas at Austin
• MILCOM 2005
• Atlantic City, NJ
• October 17-20, 2005

2
Introduction

• Multiuser OFDM
• ? multiple access technique sharing the
  subcarriers with multiple
• users
• Static v.s. Adaptive subcarrier allocation
  algorithm
• Different users experience mutually independent
  fading
• Given equal power per subcarrier, the channel
  capacity is proportional to the channel gain
• Given channel state information (CSI), capacity
  is enhanced by an adaptive SA algorithm

3
System Model

• K users and N subcarriers in the system

Base station transmitter
Subcarrier Allocation And Adaptive modulation
IFFT
User 1 data
Subcarrier 1
Add cyclicprefix
User 2 data
Subcarrier 2


User K data
Subcarrier N
User 1 CSI feedback
User 2 CSI feedback

User k CSI feedback
Decoder
Subcarrier Selection And Demodu- lation
FFT
Remove cyclicprefix
Subcarrier 1
Subcarrier 2
User k data

Subcarrier N
User k receiver
4
System Model

• Frequency selective fading channel
• Transmitter knows the CSI
• Receiver knows subcarrier allocation (SA)
  information
• Adaptive modulation is considered for capacity
  enhancement
• Moderate correlation value is used for defining
  the
• coherence bandwidth
• ? a variation exists within the coherence
  bandwidth
• ? average channel gain of each group is
  considered
• Coherence bandwidth of channel gt bandwidth of a
  subcarrier

5
Problem Statement

• Optimal subcarrier, bit, and power allocation
  algorithm
• ? Assign adaptively all the subcarriers to all
  users (NP-hard)
• ? High Complexity!!
• Find a sub-optimal algorithm that
• performs close to that of the optimal algorithm
• is computationally not expensive (compared to
  optimal algorithm)

6
Previous Work (1)

• Blockwise Subcarrier Allocation (SA) Algorithm
  Xiaowen et. al VTC 03
• A block consists of a number of adjacent
  subcarriers
• Given CSI, an adaptive block allocation
  (serial process)
• Block allocation Step
• ? Based on the CSI of each user, assign the
  blocks with the highest channel gain to the
• user until the data rate requirement is
  satisfied.
• Iterative improvement Step
• ? Reallocate blocks to minimize the total
  required transmit power while the rate and BER
• requirements

User k
User m
User n
1
8
9
16
24
6
7
2
5
4
3
11
10
13
12
14
15
23
22
21
20
19
18
17
Total Frequency Band
7
Previous Work (1) - continued

• Drawback of Blockwise SA Algorithm
• Assign the blocks to the users via a serial
  process

Without CS
A sample blockwise allocation

• Simple but have a room to be improved in terms
  of system capacity

? Comparative superiority (CS)
8
Previous Work (2)

• Decentralized SA Algorithm Alen et. al Trans.
  on. Broadcast. Dec. 03
• A partition consists of a number of adjacent
  subcarriers
• Given CSI, an adaptive partition allocation
  (parallel process)
• ? Based on the CSI of each user, the usage
  values of all the partitions are initialized.
• ? All users select the partitions with the
  highest usage value independently.
• ? Conflict problem occurs
• Initialization Step
• ? Initialize the usage values of all the
  partitions through the quantitative factors
• (Channel gain, ranking factor) and
  normalization.
• Iteration Step
• ? Update the usage values (cost value,
  weightage factor, noise factor, normalization).
• ? Reallocate partitions to resolve the
  conflict problem among users.

9
Previous Work (2) - continued

• Drawback of Decentralized SA Algorithm
• CS is considered but the usage value is based
  on
• ? The random noise factor and the
  normalization process
• ? Uncertainty exists !!
• Compare all the usage values of partitions
  for all users
• ? Computational expensive processes (e.g.,
  normalization and update)
• ? Still High Complexity !!

10
Previous Work (2) - continued
Channel gain
Ranking factor
Example
update
update
normalize
(noise factor)
normalize
(noise factor)
11
AGSA Algorithm with CS

• Adaptive Grouped SA Algorithm
• A group is determined by coherence bandwidth
• Given CSI, an adaptive group allocation
  (parallel process)
• ? Based on only the average channel gain,
  CS is performed by swapping groups.
• Iteration Step

Build swapping cases within the union set
End
Yes
Are all cases over?
No
Swap groups and update C
Move to next case
Compute after swapping
No
Yes

12
AGSA Algorithm with CS contd
Example
Channel gain

• AGSA is performed via two steps
• Step 1 Select the best group independently
  (sorting)
• ? K groups for K users
  
• Step 2 Reallocate groups via CS (swapping)
• ? L (? K) groups are conflicted/unselected
  
• Total computational complexity

13
AGSA Algorithm with CS contd

• Optional Process
• Some subcarriers in deep fading ? lower the
  ave. channel gain of a group
• The low ave. channel gain ? lower order
  modulation
• Optional Iteration Step swapping subcarriers
• Increase the system capacity v.s. increase
  the computational complexity
• Set the upper limit to avoid the excessive
  number of iterations

14
Simulation Results

• The Comparison of the overall system capacity
• Static v.s. Blockwise v.s. Decentralized v.s.
  Proposed
• Simulation Environment
• Equal amount of power on each subcarrier
• Equal number of subcarriers on each user
• Fair comparison of blocks of partitions
  of groups
• The system capacity is defined as
• where B is a bandwidth, L is of groups, is
  the average channel gain of i-th group
• K 8 , N 1024, and B 10 MHz
• Update the allocation plan after every 1000 OFDM
  symbols

15
Simulation Results - continued

• Frequency selective Rayleigh fading channel with
  an exponential power delay profile

16
Simulation Results - continued

• Performance comparison
• The adaptive schemes outperform the static
  scheme at the same SNR
• The proposed scheme is the best among
  considered schemes even w/o an optional process

17
Conclusion

• All subcarriers are grouped over the coherence
  bandwidth
• All users select the group with the high avg.
  channel gain
• Comparative superiority is adopted to enhance the
  system capacity (swapping groups)
• Simple solution for the conflict problem
• (Reallocate only the conflicted/unselected
  groups)
• Optional process (swapping subcarriers)
• Increase system capacity v.s. increase
  computational complexity

18

Thank you !!