1. <month year>
doc.: IEEE <04-106>
March 2004
Project: IEEE P Working Group for Wireless Personal Area Networks (WPANs)
Submission Title: [WiQuest Proposal for Reduced PAR Channel Estimation Sequence for MB-OFDM]
Date Submitted: [ 13 March, 2004]
Source: [Matthew B. Shoemake and Sridhar Rajagopal] Company [WiQuest Communications, Inc.]
Address [8 Prestige Circle, Suite 100, Allen, Texas, 75013, USA]
Voice:[ ], FAX: [ ],
Re: [N/A]
Abstract: [The channel estimation (CE) sequence of the current MultiBand OFDM proposal to Task Group 3a has been found to have a very large peak to average ratio (PAR) of 8.86 dB. This high PAR increases the probability of clipping the CE sequence on transmit and on receive thereby increasing the probability of packet decode failure. An improved CE sequence is proposed herein that provides a 5.43 dB improvement in PAR for the CE sequence. This improved CE sequence has been identified via an extensive computer search. The proposed improvement does not increase computational complexity at the transmitter or the receiver.]
Purpose: [The authors would like the MultiBand OFDM proposal to be updated to include the channel estimation sequence described herein. The authors believe that such action will increase the level of support for the MultiBand OFDM proposal to Task Group 3a.]
Notice: This document has been prepared to assist the IEEE P It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein.
Release: The contributor acknowledges and accepts that this contribution becomes the property of IEEE and may be made publicly available by P
Shoemake and Rajagopal, WiQuest
<author>, <company>

2. Outline Overview of current packet preamble
March 2004
Outline
Overview of current packet preamble
Desired properties in packet preamble
Peak to Average Ratios for MB-OFDM and IEEE a/g
The problem of high PAR
Improved Channel Estimation Sequence with PAR reduction of 5.42 dB
Conclusions and Proposal
Shoemake and Rajagopal, WiQuest

3. Current PLCP Preamble (268r3)
March 2004
Current PLCP Preamble (268r3)
Symbols of Packet Sync Sequence
Packet Detection
AGC
Frequency Estimation
3 Symbols of Frame Sync Sequence
Boundary Detection/FFT Placement
6 Symbols of Channel Estimation Sequence
Channel Estimation
Shoemake and Rajagopal, WiQuest

4. Desired Properties of Preambles
March 2004
Desired Properties of Preambles
Packet Detection
Good Correlation Properties
Low Cross-Correlation (SOP)
Frequency Estimation
Multiple repeats
Frame Sync
Low correlation with packet sync
Channel Estimation/AGC
Low PAR
Should have significantly lower clip probability than payload
Shoemake and Rajagopal, WiQuest

5. March 2004
Why low PAR?
Why is low PAR more important for the channel estimation sequence than it is for the data portion of the packet?
Channel estimation errors increase the probability of error on all data bits
A clip in the payload affects the probability to decode a small number of bits
The interleaver helps prevent payload clipping from causing errors
FFT
Channel
Inversion
Deinter-
leaver
FEC
Decode
Data
CE Sequence
Channel
Estimator
Shoemake and Rajagopal, WiQuest

6. March 2004
Peak to Average Ratio
PAR Calculation
Possible for a given set of tones
Difficult in general
PAR is usually calculated numerically
PAR is dependent on the sampling phase
Must oversample the signal to cover different phases and find the true PAR
Shoemake and Rajagopal, WiQuest

7. Legacy Channel Estimation Sequence (1 of 2)
March 2004
Legacy Channel Estimation Sequence (1 of 2)
Legacy sequence PAR:
dB at NyQuist
dB oversampled
Oversampling approaches:
- Use larger FFT
- Interpolate in the time domain
Results here use larger FFT
Average energy: 122
Can see visually that signal must be
oversampled to obtain accurate PAR
Shoemake and Rajagopal, WiQuest

8. Legacy Channel Estimation Sequence (2 of 2)
March 2004
Legacy Channel Estimation Sequence (2 of 2)
The current sequence has a spike at the end of the sequence
The spike at the end of the sequence causes variation in PAR measurements under various interpolation methods
The spike generates a transient that is a function of design decisions and may affect performance
The spike at the end of the sequence should and can be avoided
MBOA PAR
SRRC
7.31 dB
SRRC (repeat)
8.35 dB
SRRC (zero pad)
7.36 dB
IFFT (4x upsample)
8.85 dB
Sinc (zero pad)
7.29 dB
There is 1.49 dB difference in spike between two reasonable but different implementations.
Shoemake and Rajagopal, WiQuest

9. PAR Properties for MB-OFDM and 802.11a/g
March 2004
PAR Properties for MB-OFDM and a/g
Oversample Factor
CE r2
CE r3
LS 11a 1
5.44
7.09
3.17 2
8.0
8.85 4
1024
8.01
8.86
3.18
MB-OFDM CE r2: Channel Estimation Seq in rev 2 (112 tones)
MB-OFDM CE r3: Channel Estimation Seq in rev 3 (122 tones)
LS a/g: Long (CE) Sequence of a/g (48 tones)
Current CE PAR is too high thereby reducing probability of good
channel estimate and increasing packet error probability
Shoemake and Rajagopal, WiQuest

10. PAR and Conjugation (1 of 2)
March 2004
PAR and Conjugation (1 of 2)
Plot shows that the constraint of conjugate symmetry tends to increase the PAR on average by dB
Result of unnecessarily forcing the time domain signal to be real valued
1.7 – 2.0 dB
Shoemake and Rajagopal, WiQuest

11. PAR and Conjugation (2 of 2)
March 2004
PAR and Conjugation (2 of 2)
In addition to increasing the PAR
Conjugation does not help channel estimation
Conjugation does not reduce complexity or cost
Mandatory rates of 110 and 200 Mbps do not use conjugation
Conjugation should be removed from the
MB-OFDM channel estimation sequence
Shoemake and Rajagopal, WiQuest

12. Improved Channel Estimation Sequence
March 2004
Improved Channel Estimation Sequence
Tones(-61:-1 1:61) = {   1 -1  1  1  1 -1  1  1 -1  1  1
-1 -1    1 -1  1  1    1  1  1  1  1  1
-1  1  1     1  1  1  1    1 -1  1
-1    1  1 -1  1  1  1  1 -1  1  1  1 -1  1 -1  1  1  1  1
-1  1  1 -1    1 -1  1    1  1  1 -1
-1 -1      1  1    1 -1  }
New sequence without conjugate symmetry has PAR of dB measured with oversampling
Shoemake and Rajagopal, WiQuest

13. Non-Conjugate Symmetric
March 2004
Reduced PAR Sequences
5.42 dB
Oversample Factor
CE r3
WiQuest
Non-Conjugate Symmetric
Improvement
1024
8.86 dB
3.44 dB
5.42 dB
Shoemake and Rajagopal, WiQuest

14. March 2004
Receive PAR
The PAR at the receiver depends on the multipath channel
To confirm that a PAR reduction is maintained at the receiver, cumulative distribution functions were generated for the PAR at the receiver
The reduced PAR CE sequence was compared to the CE sequence in the current draft in AWGN, CM1 and CM2
A PAR reduction of 0.7 to 3.0 dB exists at the receiver
~3 dB
~ 1 dB
~ 0.7 dB
Shoemake and Rajagopal, WiQuest

15. Reuse of the Constellation Mapper
March 2004
Reuse of the Constellation Mapper
Rotation of the CE sequence does not change the PAR
The proposed non-conjugate symmetric sequence is specified over {+1, -1}
Rotating the proposed non-conjugate symmetric sequence by 45 degrees would cause the sequence to be over {+1+j, -1-j}/sqrt(2)
If the sequence is rotated by 45 degrees, the current QPSK mapper can be used without modification
The authors are open to either approach
Shoemake and Rajagopal, WiQuest

16. Conclusions and Proposal
March 2004
Conclusions and Proposal
Current MB-OFDM channel estimation (CE) sequence has:
A very high PAR of 7.36 to 8.86 dB
An undesirable spike near the end of the sequence
A higher probability of clipping on TX and RX
Sequences exist that provide dramatic reductions in the PAR
A 5.42 dB decrease
Provides reduced DAC and ADC resolution
The authors propose to replace the channel estimation sequence in the current MB-OFDM proposal with the non-conjugate symmetric sequence on slide 11
Shoemake and Rajagopal, WiQuest