January 2005
John S. Sadowsky, Intel
Slide 1
doc.: IEEE 802.11-05/1635r1
Submission
WWiSE Preamblesand MIMO Beamforming?
Notice: This document has been prepared to assist IEEE 802.11. 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 grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE 802.11.
Patent Policy and Procedures: The contributor is familiar with the IEEE 802 Patent Policy and Procedures <http:// ieee802.org/guides/bylaws/sb-bylaws.pdf>, including the statement "IEEE standards may include the known use of patent(s), including patent applications, provided the IEEE receives assurance from the patent holder or applicant with respect to patents essential for compliance with both mandatory and optional portions of the standard." Early disclosure to the Working Group of patent information that might be relevant to the standard is essential to reduce the possibility for delays in the development process and increase the likelihood that the draft publication will be approved for publication. Please notify the Chair <[email protected]> as early as possible, in written or electronic form, if patented technology (or technology under patent application) might be incorporated into a draft standard being developed within the IEEE 802.11 Working Group. If you have questions, contact the IEEE Patent Committee Administrator at <[email protected]>.
Date: 2005-01-15
Name Company Address Phone email John S. Sadowsky Intel Corp. 5000 W. Chandler
Chandler, AZ USA +1 480 554 0842
Tomoya Yamaura Sony 2-17-1 Higashigotanda, Shinagawa, Tokyo, Japan
+81 3 6409 3201
John Ketchum Qualcomm 9 Damonmill Square Suite 2a Concord, MA USA
781 276-0915 [email protected]
Authors:
January 2005
John S. Sadowsky, Intel
Slide 2
doc.: IEEE 802.11-05/1635r1
Submission
Abstract
Interest in beamforming for WLAN products is growing. This is evidenced by the recent launch of several products that provide enhanced rate at range performance via beamforming - using 802.11a/b/g waveforms! It is expected that this trend will continue as MIMO is introduced.
The WWiSE proposal for 802.11n High Throughput WLAN has a preamble structure that can not support of advanced beamforming techniques. This presentation itemizes the problems associated MIMO BF (e.g., SVD BF) with the WWiSE preambles.
January 2005
John S. Sadowsky, Intel
Slide 3
doc.: IEEE 802.11-05/1635r1
Submission
Summary• Problem 1
– WWiSE structure does not allow omni-directional transmission of SIG-N– Result: Hidden node problems
• Problem 2– WWiSE preambles are designed for “per antenna training”– Low overhead BF (beamforming) requires “per spatial stream training”
• Eliminates explicit feedback of BF steering matrices to receiver
– Conjecture: WWiSE could apply ½ symbol cyclic shift training to spatial streams
• Problem 2b– WWiSE channel estimation requires smoothing algorithms– Channel smoothing cannot be applied with MIMO BF (e.g., SVD)
• Problem 3– Explicit feedback CSI (provided by WWiSE MAC management frame)– Inserts MAC latency into time critical processing
January 2005
John S. Sadowsky, Intel
Slide 4
doc.: IEEE 802.11-05/1635r1
Submission
Problem 1: SIG-N is not omni-directional
Time
TransmitAntennas
1
10 x 0.8 = 8s
2
1.6 + 2 x 3.2 = 8s 4s
SS20
SS20(400 ns cs)
LS20
LS20 (1600 ns cs)
GI2
GI2
SIG-N
SIG-N(1600 ns cs)GI GI
GIGI Data
Short sequence Long sequence Signal Data payload
Data
WWiSE ½ symbol cyclic shift training applied to 2 spatial streams to support MIMO BF
SIG-N must then be transmitted in BFNOT omni-directionalThis introduces hidden node problems
Also, how are short symbols transmitted – do 400 ns cs, 1600 ns cs and spatially multiplexed BF data all yield the same Rx power?
January 2005
John S. Sadowsky, Intel
Slide 5
doc.: IEEE 802.11-05/1635r1
Submission
Problem 1: Omni-directional protection
• MAC Protection?– MAC protection mechanisms assume omni-directional
transmission of control packets
• PHY Protection– Omni-directional part provides PHY protection
– However, the shift to BF part large increase in Rx power
– Can drive receiver into saturation
– WWiSE MM PPDU does not solve this problem
January 2005
John S. Sadowsky, Intel
Slide 6
doc.: IEEE 802.11-05/1635r1
Submission
PHY Protection via Omni-Preamble
• WWiSE MM PPDU does not work due to Rx power transient
omni-preamble BF Part
Rx power~ 6 dB
Rx clipping potential
• TGn Sync HT-STF provides a robust solution to the hidden node problem for MIMO BF transmissions
TGn Sync HT-STF 2nd AGC
January 2005
John S. Sadowsky, Intel
Slide 7
doc.: IEEE 802.11-05/1635r1
Submission
Problem 2: Per Antenna Training
• Per Antenna Training– Requires high overhead of explicit communication of BF matrices
– Complex Rx signal processing
• Per Spatial Stream Training– Receiver directly estimates combined channel
– Low-overhead: no explicit transmission of BF matrices
– Transparency: no prior knowledge of packet BF structure at the receiver• Rx acquisition and equalizer processing is identical for both BF and non-BF packets
BFmatrix
V
ant.-to-ant. ch.
H
H H Vcombined channel
per spatial streamtraining inserted prior
to BF matrix
Conjecture: WWiSE intends to do per SS training with ½ symbol cyclic shift
January 2005
John S. Sadowsky, Intel
Slide 8
doc.: IEEE 802.11-05/1635r1
Submission
Problem 2b: WWiSE Requires Ch. Est. Smoothing
Time
TransmitAntennas
1
10 x 0.8 = 8s
2
1.6 + 2 x 3.2 = 8s 4s
SS20
SS20(400 ns cs)
LS20
LS20 (1600 ns cs)
GI2
GI2
SIG-N
SIG-N(1600 ns cs)GI GI
GIGI Data
Short sequence Long sequence Signal Data payload
Data
1 4 1 3 1 2 1 1 1 0 1 1 1 2 1 3 1 4
2 4 2 3 2 2 2 1 2 0 2 1 2 2 2 3 2 4
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )
H f H f H f H f H f H f H f H f H f
H f H f H f H f H f H f H f H f H f
0.25 0.5 0.25 recovers 1 2( )H f
2 2( )H f
Smoothing Window
-0.25 0.5 -0.25 recovers
1600 ns cs produces a 1 factor on H2(fk), -1 for odd k
January 2005
John S. Sadowsky, Intel
Slide 9
doc.: IEEE 802.11-05/1635r1
Submission
Why is channel est. smoothingbad for MIMO BF?
• Smoothing requires high adj. tone coherence• However, we must estimate the combined channel
– BF matrix has poor adjacent tone coherence
• Why?– Eigen-channel rank reversals
• For each tone, eigen-channels are ranked by singular values• Eigen-channels can reverse ranks on adjacent tones – resulting in an
adjacent tone swap of corresponding columns of BF matrix• Result – very low adjacent tone coherence
– Singular value multiplicity (nearly equal singular values)• Common eigen space - blurs distinction between eigen-channels• Numerical precision issues
H = H V
January 2005
John S. Sadowsky, Intel
Slide 10
doc.: IEEE 802.11-05/1635r1
Submission
SVD BackgroundHH U ΣV
diagonal matrixof singular values
matrix of rightsingular vectors
= BF matrix
U and V are orthonormal matrices; columns are orthonormal
Columns of U and V are left and right singular vectors
V is the optimal BF matrix
Uniqueness:• is unique• U and V are unique up to per column phase factor
January 2005
John S. Sadowsky, Intel
Slide 11
doc.: IEEE 802.11-05/1635r1
Submission
BF Adjacent Tone Coherence
( ) ( 1)Hi SC i SC i SCk k k v v
SCkVSCk= BF matrix for subcarrier
i SCkv SCkV= ith column of
Definition: Adjacent Tone Coherence
Properties:
| | 1 1ji SC i SC i SCk k e k v v
| | 1i SCk i SCk
1 1i SC i SC i SCk k k v v
is a complex number
January 2005
John S. Sadowsky, Intel
Slide 12
doc.: IEEE 802.11-05/1635r1
Submission
What about the arbitrary singular vectors phase?
( )( 1) ( 1), (0) (0)i SCj ki SC i SC i ik e k v v v v
( ) | ( ) |i SC i SCk k
Columns of V are unique up to an arbitrary phase. Select these phases to maximize phase coherence between adjacent frequencies.
Problems? YES!This is an additional non-linear processing step in a time-critical operation.
Result:
| ( ) |i SCk is an optimistic measure of adjacent tone coherence.
January 2005
John S. Sadowsky, Intel
Slide 13
doc.: IEEE 802.11-05/1635r1
Submission
Example: 2 x 2, Model B
-20
-15
-10
-5
0
5
10
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
Eigen-channel rank reversal
January 2005
John S. Sadowsky, Intel
Slide 14
doc.: IEEE 802.11-05/1635r1
Submission
Example: 2 x 2, Model D
-20
-15
-10
-5
0
5
10
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
Nearly equal singular values coherence breakdown
January 2005
John S. Sadowsky, Intel
Slide 15
doc.: IEEE 802.11-05/1635r1
Submission
Example: 2 x 2, Model D
-20
-15
-10
-5
0
5
10
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
Eigen-channel rank reversal
January 2005
John S. Sadowsky, Intel
Slide 16
doc.: IEEE 802.11-05/1635r1
Submission
Example: 4 x 2, Model D
-20
-15
-10
-5
0
5
10
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
Loss of coherence in 2nd eigen channel only
January 2005
John S. Sadowsky, Intel
Slide 17
doc.: IEEE 802.11-05/1635r1
Submission
Example: 4 x 2, Model D
-20
-15
-10
-5
0
5
10
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
Eigen-channel rank reversal
Loss of coherence in 2nd eigen channel only
Joint loss of coherence due to rank reversal
January 2005
John S. Sadowsky, Intel
Slide 18
doc.: IEEE 802.11-05/1635r1
Submission
Example: 4 x 4, Model D
-20
-15
-10
-5
0
5
10
15
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
January 2005
John S. Sadowsky, Intel
Slide 19
doc.: IEEE 802.11-05/1635r1
Submission
Example: 4 x 4, Model D
-20
-15
-10
-5
0
5
10
15
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
January 2005
John S. Sadowsky, Intel
Slide 20
doc.: IEEE 802.11-05/1635r1
Submission
Example: 4x2 Model B
-20
-15
-10
-5
0
5
10
-10 -8 -6 -4 -2 0 2 4 6 8 10Frequency (MHz)
Sin
gu
lar
Val
ue
(dB
)
0.75
0.80
0.85
0.90
0.95
1.00
-10 -8 -6 -4 -2 0 2 4 6 8 10
Frequency (MHz)
abs(
rh
o )
January 2005
John S. Sadowsky, Intel
Slide 21
doc.: IEEE 802.11-05/1635r1
Submission
Response to 1645 r1 Simulation
• 1645 r1 provides simulations of WWiSE ch. est.– Calculations of MSE with SVD beamforming
– Shows only a small degradation in MSE
– Simulated only 4x2 Model B example• High spatial diversity & low frequency diversity
– a best case scenario for MSE calculation
• 1645 r1 does not specify how BF phase is selected– Are there addition constraints to SVD calculation? Complexity?
• The MSE results are misleading– We show that for most tones there is very high adjacent tone coherence
• This will dominate MSE averaging to produce an optimistic result
– However, when coherence breaks down, as in eigen-channel rank reversal, we loose all tones within the smoothing window
– How does this impact PER?
January 2005
John S. Sadowsky, Intel
Slide 22
doc.: IEEE 802.11-05/1635r1
Submission
Problem 3: Explicit Feedback CSI
• WWiSE provides explicit CSI (channel state info.)– CSI transmitted in a MAC management frame– Huge Overhead!– Huge MAC latency problems!
• in a time critical processing step• Requires 2x variable queuing and channel access delays
(~1 ms or more)
– Limited to 4 BF antennas
• TGnSync is a channel reciprocity system– Very low overhead– SVD processing contained entirely in the PHY– Broad Utilization
• TGn Sync give Basic BF receive to ALL STAs• Essentially no cost/complexity burden to BF Rx only client STA• WWiSE cannot do this!
January 2005
John S. Sadowsky, Intel
Slide 23
doc.: IEEE 802.11-05/1635r1
Submission
Explicit CSI Feedback Overhead
Num. Tx Ant. 2 4 4 2 4 4
Num Rx Ant. 2 2 4 2 2 4
Ch. BW (MHz) 20 20 20 40 40 40
Bytes per
CSI FDBK896 1792 3584 1792 3584 7232
Overhead Rate@ 2 ms FDBK
3.6 Mbps
7.2Mbps
14.3Mbps
7.2Mbps
14.3Mbps
28.9Mbps
Overhead Rate@ 4 ms FDBK
1.8Mbps
3.6Mbps
7.2Mbps
3.6Mbps
7.2Mbps
14.5Mbps
TGn Sync has NO CSI FEEDBACK OVERHEAD
January 2005
John S. Sadowsky, Intel
Slide 24
doc.: IEEE 802.11-05/1635r1
Submission
SVD Tx & Immediate Training
STA A
STA B PPDU 1
PPDU 2SVD
Reciprocity SystemA→B channel (from reciprocal B→A channel) available to STA A hereSVD processing commences immediately in STA A PHYMAC is not involved in time critical operations
Explicit CSI Feedback SystemA→B CSI only after the frame 1 is decodedThere may be additional MAC processing delaysToo late for immediate use on PPDU 2
SVD BF matrix must be available here
Required SVD processing time
Note: SVD computation is comparable to MMSE equalizer setup calculations (nxn Hermitial inverse). For n = 2, there is a direct closed form solution for both. For n > 2, both problems are efficiently solved via Jacobi iteration – hence there is possibility for hardware reuse between the two algorithms.
January 2005
John S. Sadowsky, Intel
Slide 25
doc.: IEEE 802.11-05/1635r1
Submission
Explicit Feedback SVD Protocol ??
Is this what WWiSE is proposing? If not – then what?
Mgnt frame
ACK
CF poll
Mgntframe
ACK SVD Data
BA
CSI Feedback896 – 7232 bytes
CSI Request
Age of CSI
Protocol Overhead!
Not included: MAC protection protocol
January 2005
John S. Sadowsky, Intel
Slide 26
doc.: IEEE 802.11-05/1635r1
Submission
Broad Utilization CSI Buffer
PPDU w/ CSIfeedback request
Channel estimates available here.These are applied to the equalizer, then typically thrown away.
CSI feedback request known here(after MPDU is decoded)
All channel estimates must be buffered for this period
CSI Buffer for a 2 Rx ant. STA: 1792 bytes (20 MHz), 3584 bytes (40 MHz)
This CSI buffer is required for all STAs to receive BF from a 4 Tx device.Does WWiSE propose to add this requirement?
January 2005
John S. Sadowsky, Intel
Slide 27
doc.: IEEE 802.11-05/1635r1
Submission
Conclusions
• The WWiSE proposal presents several barriers for MIMO BF– Non-omni-directional SIG-N– Required smoothing in channel estimation– High overhead explicit CSI feedback solution– Unspecified packet exchange protocol, large MAC latencies– MIMO BF is an after thought in the WWiSE proposal
• MIMO BF is important for future extensibility of the 802.11n standard– Recently launched BF products have demonstrated enhanced rate-at-
range performance in standards based (a/b/g) solutions– BF allows concentration of cost and power consumption to the AP for
downlink intensive applications (e.g., video)
• The 11n should provide seamless support MIMO Tx beamforming
January 2005
John S. Sadowsky, Intel
Slide 28
doc.: IEEE 802.11-05/1635r1
Submission
References
• IEEE 802.11-04/0886r6,“WWiSE Proposal: High throughput extension to the 802.11 Standard”
• IEEE 802.11-05/1645r1,“Preambles, Beamforming and the WWiSE Proposal”
Top Related