Note: The author list will grow to reflect those providing explicit contributions and review comments
Submission Slide 2
Introduction
• The TGn task group has developed a comprehensive MIMO broadband channel models, with support for 40 MHz channelization and 4 antennas.
• The TGac task group is targeting > 1 Gbps MAC SAP throughput using one or more of the following technologies:
– Higher order MIMO (> 4x4)
– Multi-User MIMO with > 4 AP antennas
– Higher Bandwidth (> 40 MHz)
– OFDMA
• We propose some simple modifications to TGn channel models to enable their use for TGac.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 3
Modifications to Handle Large System BW
• TGn systems handled 40 MHz systems BW, assuming tap-spacing of 10 nsec.
• For TGac systems with larger overall system bandwidth (W), we propose to decrease channel tap spacing by a factor of
• The calculation of W and tap spacing is illustrated in the below examples:
– Example : A TGac modem can have 2 channels of 40 MHz each that are spaced by 60 MHz for sufficient isolation.
• W = 40*2+60 = 140 MHz. Channel tap spacing = 2.5 nsec.
• The reduced channel tap-spacing is modeled by linearly interpolating the Cluster channel tap power values, on a cluster by cluster basis.
2log40
1
2W
ceiling
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 4
Higher Order MIMO
• Propose to extend Kronecker models of TGn for higher order MIMO.
• It has shown by measurements [1] that
– TGn channel models tightly bound and sweep the range of MIMO performance observed in real environments.
– Randomly rotating the TGn defined cluster AoA and AoDs is sufficient to emulate the case-by-case variation expected in real-world environments.• Random AoA offsets were distributed uniformly between ±180°
• Random AoD offsets were distributed uniformly between ±30°.
• For each case, the same offset was applied to all clusters.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 5
Higher Order MIMO
Measured Capacity CDFsin Office Environment [1]
Channel Model B Capacity CDFs(Random AoA/AoD)
Channel Model D Capacity CDFs(Random AoA/AoD)
Randomly rotating the TGn defined cluster AoA and AoDs is sufficient to emulate the case-by-case variation expected in real-world environments.
In this figure, capacity calculated assuming SNR = 24 dB and
)det(log *2 HHIC
tr NSNR
N
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 6
Multi-User MIMO Extensions
• Literature Search:
– J-G. Wang, A.S. Mohan, and T.A. Aubrey,” Angles-of-arrival of multipath signals in indoor environments,” in proc. IEEE Veh. Technol. Conf., 1996, pp. 155-159.
• For the same RX location, cluster AoA from 2 different TX locations vary up to 20 degrees in classroom and up to 60 degrees in large halls.
• In Hall, clusters that are relevant for one TX location were absent for another TX location.
• Results directly applicable to MU-MIMO
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 7
Multi-User MIMO ExtensionAoD/AoA vs. Physical Geometry
AP
STA-1
STA-2
AoD1
AoD2
From Physics:
•AP has a different AoD to STA-1 and STA-2. Also, each STA has a different AoA from AP.
The LOS steering vectors to STA-1 and STA-2 are different.
Scenario 1: Pure LOS channel
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 8
Multi-User MIMO Extension AoD/AoA vs. Physical Geometry
AP
STA-1
STA-2
AoD1
AoD1AoD2
AoD2
•Different scatterers may be relevant to different STAs.
•AP may have a completely independent AoD for clusters corresponding to STA-1 and STA-2
•STAs may have completely independent AoA depending on location and device orientation
Scenario 2: NLOS channel with scatterers far away from AP
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 9
Multi-User MIMO Extension AoD/AoA vs. Physical Geometry
AP
STA-1
STA-2
AoD2 AoD1
•AP may have a similar AoDs for clusters regardless of transmission to STA-1 or STA-2.
•STAs may have independent AoAs depending on location and device orientation
Scenario 3: NLOS channel with scatterers close to AP
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission
• Assume TGn-defined cluster AoDs and AoAs for link level simulations.
• For Multi-User MIMO system simulations:– Assume TGn-defined cluster AoDs and AoAs as baseline.– For each client, a single pseudo-random offset is added to all cluster AoDs and
AoAs.• Pseudo-random selection allows comparison across proposals. • Single offset retains TGn angular spacing between clusters.
– NLOS Cluster AoD offsets uniformly distributed between ±30°• Based on experimental results from Wang et al. • Compromises scenarios outlined in slides 3,4,5.
– NLOS Cluster AoA offsets uniformly distributed between ±180° • Clients can see independent AoA depending on orientation and location.
– LOS tap AoA and AoD offsets uniformly distributed between ±180°.• Direct LOS path to each client can have independent AoA/AoD depending on location.
• Pros:– Physically realistic – Introduces statistical AoA/AoD variation across clients– Minimal change to TGn channel model– Simulation complexity increase is reasonable: TX/RX correlation matrix need to be
computed only once per client, for the entire simulation run.
Slide 10
MU-MIMO Channel Model Proposal
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 11
MU-MIMO Simulation Overview
• Assumptions:– 16 TX antennas, 8 STAs, 2 RX antennas per STA– TGn channel models B, D (LOS and NLOS scenarios) used as baseline
• AoD and AoA as specified in the channel model document– Composite multi-user channel matrix constructed from vertical concatenation of 8
2x16 channel matrices• Clients are effectively uncorrelated from each other
• Capacity Analysis:– For each channel model, 5 cases of random per-user AoA and AoD generated– 200 channel realizations generated per case– MMSE precoder applied to each 16x16 channel instance
• Post-processing SINRs calculated for each stream and subcarrier• PHY capacity for each stream/subcarrier calculated as log2(1+SINR)
– For each instance, sum-average channel capacity calculated by averaging across subcarriers and summing across spatial streams
– CDFs generated across all 200 channel instances
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 12
MU-MIMO Model B Results – Capacity CDFs
• Model B: 2 clusters, 0dB K factor in LOS case• Capacity CDF varies by +20% depending on user selection and their AoA/AoD
– Note #1: AoD variation in LOS channel component leads to variation of steering vectors across clients and hence improves MU-MIMO capacity.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 13
MU-MIMO Model D Results – Capacity CDFs
• Model D: 3 clusters, 3dB K factor in LOS case
• Capacity CDF varies by +/-10% depending on user selection and their AoD/AoA. – Note #1: Artifact of TGn model: TGn AoA specification result in optimal per-user MIMO capacity.
Any AoA offset tends to degrade per-user MIMO capacity.
– Note #2: AoD variation in LOS channel component leads to variation of steering vectors across clients and hence improves MU-MIMO capacity.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 14
MU-MIMO Model F Results – Capacity CDFs
• Model F: 6 clusters, 6 dB K factor in LOS case
• Capacity CDF varies by negligible amount depending on user selection and their AoD/AoA. – Note #1: 6 clusters, each with a large AS of 30-60 degrees, make the capacity CDF less sensitive to
AoA/AoD variations.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 15
MU-MIMO Summary• Equal AoD for all STAs is not physically realistic.
– In pure LOS scenarios, such a model will “break” MU-MIMO by mandating equal steering matrices across clients.
• Diversity of AoD/AoA across STAs impacts MU-MIMO performance:– Capacity improves in LOS scenarios and models with small # of clusters.
• 20% improvement in LOS channel model B.
• Recommend using a pseudo-randomly selected AoDs/AoA offset across users in MU-MIMO model
– NLOS Cluster AoD offsets uniformly distributed between ±30°– NLOS Cluster AoA offsets uniformly distributed between ±180°– LOS tap AoA and AoD offsets uniformly distributed between ±180°.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 16
Incorporating Dual Polarized Antennas
• Dual-Pol antennas are likely to be employed in TGac devices.
– Dual-polarized antennas improve MIMO channel capacity, especially in LOS channel conditions. E.g: Hallways.
– Co-located dual-pol antennas minimize real estate in devices.
• Measurements indicate that Dual-Pol TGn channel models suggested in Erceg et al., is applicable to 8x8 MIMO. – We assumed the following, while comparing measured data with
simulation results:• XPD value of 10 dB for the steering matrix HF,• XPD value of 3 dB for the variable matrix Hv. • 0.2 correlation for co-located cross-polarized antenna elements.
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 17
Incorporating Dual Polarized AntennasMeasurements vs. Simulations
• CDFs in “light lines” indicate measured capacity CDFs in office environment [1]
• TGn channel models B, used as baseline with AoD and AoA as specified in the channel model document
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 18
References
1. Breit, G. et al. “802.11ac Channel Modeling.” Doc. IEEE802.11-09/0088r1.
2. Erceg, V. et al. “TGn Channel Models.” Doc. IEEE802.11-03/940r4.
3. Kenny, T., Perahia, E. “Reuse of TGn Channel Model for SDMA in TGac.” Doc.IEEE802.11-09/0179r0.
4. Schumacher, L.; Pedersen, K.I.; Mogensen, P.E., "From antenna spacings to theoretical capacities - guidelines for simulating MIMO systems," Personal, Indoor and Mobile Radio Communications, 2002. The 13th IEEE International Symposium on , vol.2, no., pp. 587-592 vol.2, 15-18 Sept. 2002.
5. Jian-Guo Wang; Mohan, A.S.; Aubrey, T.A., "Angles-of-arrival of multipath signals in indoor environments," Vehicular Technology Conference, 1996. 'Mobile Technology for the Human Race'., IEEE 46th , vol.1, no., pp.155-159 vol.1, 28 Apr-1 May 1996.
6. Offline discussions with Vinko Erceg (Broadcom) and Eldad Perahia (Intel).
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 19
Appendix
MU-MIMO Code Changes
March 9, 2009 doc.:IEEE 802.11-09/0309r1
Submission Slide 20
MU-MIMO Code Changes• Channel generation by offsetting TGn-defined cluster AoAs/AoDs
– IEEE_802_11_Cases.m function altered
• Four new function arguments defining angular offsets: Delta_AoD_LOS_deg, Delta_AoA_LOS_deg, Delta_AoD_NLOS_deg, Delta_AoA_NLOS_deg
– “LOS” arguments specify offset in degrees added to steering matrix AoD and AoA
– “NLOS” arguments specify common offset in degrees added to all cluster AoDs and AoAs
• Composite MU-MIMO channel generation– Example scenario
• 8 TX antenna, 2 users, 4 RX antennas per user
• Generate independent 4x8 channel matrices for each user, denoted as H1 and H2
– Each user channel formed assuming different cluster AoA and AoDs
