1. Effective (20us) Preambles for MIMO-OFDM
Seigo Nakao SANYO Electric Co., Ltd. Japan
Yoshiharu Doi SANYO Electric Co., Ltd. Japan
Yasutaka Ogawa Hokkaido University Japan
Presented by David Bagby

2. Presentation Contents
Training sequences - performance motivations (STS and LTS)
Detecting the number of Tx Antennas via STS
AGC performance of proposed STS
Proposed LTS and channel estimation techniques
Conclusions

3. Presentation Contents
Training sequences - performance motivations (STS and LTS)
Detecting the number of Tx Antennas via STS
AGC performance of proposed STS
Proposed LTS and channel estimation techniques
Conclusions

4. Training sequences: performance motivations
PHY overhead restricts high throughput.
Overlapped LTS can reduce the PHY overhead.
More efficient that staggered LTS usage
The proposed STS enable overlapped LTS.
Note: The proposed STS are 20us long – which is the length of the .11a preamble – so this approach allows MIMO without using additional preamble time.

5. Training sequence usage
STS is used for
AGC
Timing detection
Rough frequency offset estimation
Detecting the number of antennas
Useful for overlapping LTS
LTS is used for
Channel estimation
Fine frequency offset estimation
Bullets in black (STS AGC, timing detection and rough frequency offset estimation) were presented in a prior paper.
Please see document r2
This paper extends the prior submission and concentrates on the blue bullets
use of STS for determination of number of Tx antennas and
use of proposed LTS for channel estimation and fine frequency offset determination

6. Overlapped and Staggered LTS
Sig
STS1
LTS1
STS2
LTS2
DATA2
DATA1
Example of overlapped STS and LTS
TX1
TX2
Sig
STS1
LTS
STS2
DATA1
DATA2
Example of overlapped STS, staggered LTS
TX1
TX2
Sig2
Clearly overlapped LTS is a more efficient use of time, hence lower overhead and greater throughput.
Time to detect number of TX antennas

7. STS Implication of overlapped LTS
“Overlapped STS and LTS”, can be easily deployed when the number of antennas can be known during the STS time.
The proposed STS provide this ability.
Overlapped approach implies some LTS considerations for efficient MIMO channel estimation.
Sig
STS1
LTS1
STS2
LTS2
DATA2
DATA1
TX1
TX2

8. Desired STS characteristics:
Each TX antenna should have a unique STS.
The cross-correlation of 1 STS cycle for any pair of STSs should be 0 for
Easy synchronization,
Good frequency offset estimation, and
Optimum AGC implementation.
(As concluded in IEEE r2)
STSs may be used to help distinguish the number of TX antennas.

9. Presentation Contents
Training sequences - performance motivations (STS and LTS)
Detecting the number of Tx Antennas via STS
AGC performance of proposed STS
Proposed LTS and channel estimation techniques
Conclusions

10. Proposed STSs (ex: 4 Tx max)
Used for sync, to find # Tx and fine AGC
(using 6 of 12 carriers of the legacy STS)
1TX
Legacy STS
2TX
STS1
STSa
3TX
STS1
STS2
STSb
4TX
STS1
STS2
STS3
STSc
We have illustrated a maximum of 4 Tx antennas – we believe that this covers the all the pragmatic cases likely to be proposed for TGn.
The desirable STS characteristics are
Low cross correlation between the legacy STS and proposed STSs
Used for Fine AGC
(using the other 6 carriers of legacy STS)
Low cross-correlation
Zero cross-correlation

11. Examples of Proposed STSs (3 antenna max case)
Carrier no.
Legacy
STSa
STSb
STS1
STS2
-24 j j j
-20 j j j
-16 j j j
-12 j j j -8 j j j -4 j j j 4 j j j 8 j j j 12 j j j 16 j j j 20 j j j 24 j j j
sqrt(13/6)=1.472, sqrt(13/3)=2.082

12. How to detect the number of antennas (3 antenna max case)
Correlation
peak
Received signal
Correlator for
legacy STS
(6carrier)
Max
Select
Number of
antennas
Correlator for
STSa
Correlator for
STSb

13. Correlator for Legacy STS detection
Carrier no.
Legacy STS
sequence
Correlator for
legacy STS
(6carrier)
-24 j
This correlator can detect the legacy STS without any effects from STS1 and STS2.
-20 j
-16 j
-12 j j -8 j j -4 j 4 j j 8 j 12 j j 16 j j 20 j j 24 j

14. Detecting Number of Antennas
3 antenna case
Channel model B 5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
-10 -5 5 10 15 20
SNR [dB]
Detection Error

15. Legacy STS compatibility
The Proposed STSs take an approach analogous to that successfully used by .11G
The 11g document defines the same preambles as 11b, but venders don’t actually provide the same preambles – instead only use pure OFDM (like 11a) is used for throughput.
Similarly, The proposed STS can't be directly received by legacy STA.
Therefore the proposed STS have the same degree of backwards compatibility to 11a (as 11g does to 11b).

16. Proposed STS advantages
The proposed STSs support overlapped LTS usage and can be used to detect any number of Tx antennas.
The required STS correlators can be cost effectively implemented
by defining STSa and STSb such that the correlator implementation cost can be much less than the cost of n separate correlators.
(e.g. STSa_I = STSb_Q, STSa_Q = STSbI)

17. Presentation Contents
Training sequences - performance motivations (STS and LTS)
Detecting the number of Tx Antennas via STS
AGC performance of proposed STS
Proposed LTS and channel estimation techniques
Conclusions

18. AGC performance
Generally, AGC calculates signal power over 16 FFT points (i.e. 1 STS).
The average power of the STS term should be the same as that of the average power of of the DATA term.

19. Other candidates of new STSs
Frequency cyclic use of legacy STS
Which doesn’t have 16 FFT samples cycle
(IEEE r0)
Polarity changed STS of legacy STS
(IEEE r1)

20. Proposed STS AGC performance (channel model B)
0.5 1
1.5 2
2.5 3
3.5 4
Ave.power of 1 STS cycle
Ave.Power of Data

21. Other proposal’s AGC performance (channel model B)
Polarity changed STS of legacy STS
Frequency cyclic use of legacy STS
0.5 1
1.5 2
2.5 3
3.5 4
Ave.power of 1 STS cycle
Ave.Power of Data
0.5 1
1.5 2
2.5 3
3.5 4
Ave.power of 1 STS cycle
Ave.Power of Data

22. Proposed STS AGC Advantages
Proposed STSs have good performance characteristics for AGC implementations.
Proposed STSs have an advantage over frequency cyclic use of legacy STS because the proposed STS is a 16 FFT points cycle.
No disadvantage results from using 6 carriers (vs. 12) for STS (for MIMO)
For 1 Tx, 12 carriers are better than 6
For 2+ Tx, 6 carriers are better than 12 because of cross correlation advantages.

23. Presentation Contents
Training sequences - performance motivations (STS and LTS)
Detecting the number of Tx Antennas via STS
AGC performance of proposed STS
Proposed LTS and channel estimation techniques
Conclusions

24. Reference
This portion of the paper is derived from the work of Yasutaka Ogawa, Hokkaido University Japan.
For reference, please see:
Ogawa, et al. “A MIMO-OFDM System for High-Speed Transmission,” VTC2003-Fall, Oct

25. Overlapped LTS usage:
LTS should be used for fine frequency offset estimation.
LTS should have good performance for MIMO channel estimation.
At the start of the presentation we said that the LTS should be used for these purposes.

26. Proposal for new LTS and Channel Estimation:
TX1
GI21
T1,1
T1,2
(32)
(64)
TX2
GI22
T2,1
T2,2
(32)
(64)
The Impulse response between each TX and RX antenna pair is estimated by the minimum mean square error scheme using 2 LTSs in the time domain (128 samples).
Channel at each subcarrier is obtained by FFT of the zero padded impulse response.
The notation is:
GI = guard interval

27. LTS channel estimation notes
On the previous page diagram, the color differences indicate the difference in sequence.
4 sequences are shown: T1,1 , T1,2 , T2,1 and T2,2.
where T x,y is from TX antenna “x” with LTS type “y”
There are 160 LTS FFT points each antenna (32GI original LTS). The impulse response (propagation channel) is calculated using a Minimum Mean Square Error (MMSE) scheme.
Using the MMSE, one can get the impulse response.
Then the impulse response is put into the FFT

28. Channel and Frequency Offset (Df ) Estimation:
Coarse Df estimation is carried out using 3 cycles of the STS.
Phase rotation is compensated by the coarse Df estimator before the next steps.
Channel estimation is done using the 2 LTS portions (Ti,1 and Ti,2) assuming that Df = 0.
Replica of the time-domain LTS sequences with Df = 0 is calculated by the channel estimator.
Continued…

29. Channel and Frequency Offset (Df ) Estimation:
Fine Df is estimated from the phase rotation in the LTS period using the replica.
Phase rotation is compensated again using the fine Df estimator.
Channel estimation is done again using the 2 LTS portions.

30. Average BER Performance of Proposed method:/ N [ d ] D f s t i m W h p c k n o w l 1 2 3 4 - 5
TX 4, RX 4
Df = 50kHz
QPSK
9 data symbols / subcarrier
16 multipath signals (Average power of the multipath signals decreases successively by 1 dB.)
No channel coding
Spatial filter (MMSE)

31. Presentation Contents
Training sequences - performance motivations (STS and LTS)
Detecting the number of Tx Antennas via STS
AGC performance of proposed STS
Proposed LTS and channel estimation techniques
Conclusions

32. Proposed STS Advantages
Enables easier use of overlapped LTS for PHY overhead reduction.
Detects the # Tx signals during STS time
Easy synchronization.
Good course frequency offset estimation.
Provides optimum AGC implementation.
Has good cross correlation characteristics w.r.t. to legacy STS.
Have cost effective correlator implementations.
Are appropriate for use with MIMO-OFDM

33. Proposed LTS Advantages
Are appropriate for “overlapped LTS” usage.
Offers fine frequency offset estimation.

34. References IEEE 802.11-04-0002r2 IEEE 802.11-03-0999r0
Ogawa, et al. “A MIMO-OFDM System for High-Speed Transmission,” VTC2003-Fall, Oct
IEEE r2
IEEE r0
IEEE r1

35. Questions (for Mr. Nakao)?