1. Constrained Distributed MU-MIMO
Month Year
doc.: IEEE yy/xxxxr0
September 2018
Constrained Distributed MU-MIMO
Date:
Authors:
Name
Affiliations
Address
Phone
Ron Porat
Broadcom
Srinath Puducheri
Ron Porat (Broadcom)
John Doe, Some Company

2. September 2018
Abstract
In the July meeting contribution number 1116 we introduced the topic of constrained distributed MU-MIMO
In this presentation we provide a more detailed description of how this scheme works using building blocks we have in 11ax
Ron Porat (Broadcom)

3. Constrained Distributed MU-MIMO - Reminder
September 2018
Constrained Distributed MU-MIMO - Reminder
Distributed MU-MIMO means MU-MIMO running jointly on multiple distributed APs.
Constrained – we propose that we study this topic with a constraint that all AP participating can hear one of the APs (which we call the master AP).
Note this doesn’t necessarily mean all AP can hear all AP
Proposal aims to enhance current mesh AP deployments or any other deployments that can share data among several APs (wirelessly or through a wired backhaul).
Typical configuration may consist of a 4-antenna master AP and two 2-antenna slave APs (total 8 antennas). A more advanced configuration may consist of antenna APs (total of antennas)
Main problem – APs are not synchronized and drift independently from time of CSI feedback and that drift needs to be corrected at time of joint AP transmission
Ron Porat (Broadcom)

4. Proposed Design Framework - Concept
September 2018
Proposed Design Framework - Concept
Leveraging the concept of 11ax trigger frame, the Master AP sends a frame which we here call a ‘slave trigger’ to enable slave APs to initially sync and subsequently re-sync their timing, CFO and phase. The slave trigger enables a fresh accurate estimate of the current value of the phase relative to the initial value. This is easy to do based on e.g. the legacy preamble.
Slave APs pre-compensate for the total phase drift and for the CFO/SFO offsets (as is done in 11ax TB PPDU) and join the master in joint transmissions.
The slave trigger is transmitted in each part of the process – sounding and every joint MU-MIMO transmission thereafter.
The reason/benefit of transmitting the slave trigger before every joint transmission is threefold
Tells slave AP that this is going to be a joint transmission
Allows slave AP to calculate exact total phase rotation from time of sounding due to oscillator drift
Allows slave AP to calculate impact of new timing synchronization and apply a phase correction
Ron Porat (Broadcom)

5. Proposed Design Framework – Sounding
September 2018
Proposed Design Framework – Sounding
The diagram shows the sequence of transmissions building on 11ax sounding process with joint transmissions highlighted in green.
A slave trigger starts the process and enables slave APs to jointly transmit the NDPA for maximum reach
Not necessary if the master AP can reach all STA.
A second slave trigger is transmitted to enable fresh sync and storage of a reference H after which the master and slaves jointly transmit an EHT NDP.
Transmission of NDP resembles how LTFs are modulated in 11ax UL MU-MIMO
Note that storage of H (e.g. from L-LTF) is for illustration purposes, we may store only some phase related parameters
Note as described in slide 4 that slave APs pre-compensate CFO/SFO before all joint transmissions
Master
Slave Trigger
SIFS
NDPA
SIFS EHT NDP
Slaves
EHT NDP
Estimate CFO
Estimate H_ref, CFO
Ron Porat (Broadcom)

6. Proposed Design Framework – CSI Feedback
September 2018
Proposed Design Framework – CSI Feedback
This stage is similar to 11ax as a joint master-slave trigger will follow to collect CSI feedback from all STA using UL OFDMA/MU-MIMO
Master and slave APs now have access to the entire CSI from all participating APs to all participating STAs. They can now independently compute precoding vectors for joint MU-MIMO transmission.
Some STA feedback may not be heard by some AP (although with UL OFDMA that is less likely) but this is OK as we may assume that those AP anyway don’t benefit/impact those STA in the subsequent DL transmission
Ron Porat (Broadcom)

7. Proposed Design Framework – Joint MU-MIMO Transmission
September 2018
Proposed Design Framework – Joint MU-MIMO Transmission
The diagram shows the sequence of transmissions
The slave trigger precedes the joint MU-MIMO transmission and enables an estimation of the total phase drift by comparing H_new to H_ref
Slave APs then join the master in transmission after initially correcting for the phase and timing drift and subsequently correcting for CFO/SFO during the joint transmission
Note that slave APs also transmit the preamble for maximum reach.
What if the packet is so long that slave AP clocks drift reduces performance?
We can leverage the 11ax concept of midambles except in here these will enable the slave AP to re-sync (so not intended for STA and only transmitted by the master AP).
Slave AP may hear the master AP enough to drive the residual CFO to very low levels.
Master
Slave Trigger
SIFS
EHT preamble
Joint MU-MIMO Data
Slaves
Estimate H_new, CFO
Ron Porat (Broadcom)

8. Further Notes September 2018
For simplicity and similarity with 11ax concepts we assumed a separate slave trigger is always transmitted but it may also be combined into one larger preamble
e.g. the second slave trigger on slide 5 may be combined with the EHT NDP preamble
Payload data needs to be shared with the slave APs. In here we primarily build on mesh-AP deployments currently being sold by many vendors. The advantage lies in improving the second stage of mesh deployments whereby instead of each AP separately sends data to its own STA a much more efficient and robust stage can take place with joint MU-MIMO transmissions to all STAs (including the master AP STAs)
However we are not restricting ourselves necessarily to mesh deployments. Several deployments e.g. enterprise and hot spots can leverage a wired backhaul to transfer payload to multiple AP or alternatively use mmW. All these cases can leverage the design here.
Ron Porat (Broadcom)

9. September 2018
Summary
We provided a framework solution for joint distributed MU-MIMO transmissions that solves the synchronization problem between APs by leveraging and enhancing 11ax design concepts such as the trigger frame, TB PPDU and UL MU-MIMO channel estimation.
We recommend inclusion of this topic in EHT as it provides a direct extension of 11ac based DL MU_MIMO into future generations of connected APs to provide best performance/coverage.
Ron Porat (Broadcom)