In earlier blog posts on 802.11ac practical considerations, we reviewed 80 MHz channels, 256 QAM and 5 GHz migration. Continuing the 802.11ac insights series, in this post we will look at some practical aspects of MU-MIMO, which is the star attraction of the impending Wave-2 of 802.11ac.
MU-MIMO mechanics and 802.11ac standard
At a high level, MU-MIMO allows AP with multiple antennas to concurrently transmit frames to multiple clients, when each of the multiple clients has lesser antennas than AP. For example, AP with 4 antennas can use 2-stream transmission to a client which has 2 antennas and 1-stream transmission to a client which has 1 antenna, simultaneously. Implicit requirement to attain such concurrent transmission is beamforming, which has to ensure that bits of the first client coherently combine at its location, while bits of the second client do the same at the second client location. It is also important to ensure that bits of the first client form null beam at the location of the second client and vice versa.
What does 802.11ac standard offer for implementing MU-MIMO
- The standard provides Group ID Management procedure to form client groups. Clients in a given group can be considered together for co-scheduling of transmissions using the MU-MIMO beamforming.
- To be able to perform peak/null adjustments in MU-MIMO beamforming as described above, the AP needs to have knowledge of Tx to Rx antennas channel matrix to each client in the group. For this, the standard provides well defined process for channel learning wherein AP transmits sounding packet called as NDP (Null Data Packet) to which clients respond with channel feedback frames (this is called explicit feedback mechanism).
What the standard does not specify
There is more to MU-MIMO implementation that is outside of the scope of the standard. The true promise of MU-MIMO is also dependent on these additional implementation factors:
- AP has to identify clients that can be co-scheduled in a group. How to form these groups is implementation specific. It is dependent on prevalent channel conditions to different clients. AP will have to make smart decisions on group formation.
- AP has to keep track of channel conditions for clients in different groups by sending regular sounding packets and receiving explicit feedback to the sounding packets from the clients. Various implementations may differ based on how frequent channel learning is required in them. Frequent learning increases channel overhead, but may result into cleaner (non-interfering) MU-MIMO beams. Slow learning can result in stale information thereby causing inter-beam interference during concurrent transmissions.
- When channel conditions change, re-grouping of clients is required. Implementations can differ based on re-grouping triggers and method of re-grouping.
- Implementations can also differ based on how total antennas on AP are used for beamforming within any given group.
- The performance of MU-MIMO also depends to some degree on the client side implementation. For demodulating the MU-MIMO signal, clients can implement additional techniques such as interference cancellation to eliminate inter-beam interference.
- The formation of MU-MIMO groups at physical/MAC layer has to also coincide with traffic and QoS requirements of the clients at higher protocol level.
The above considerations are at practical implementation level. Many of them are in the domain of chip design. How well different chip vendors address them could differentiate them from one another in the MU-MIMO era.
They can also impact Wi-Fi chip design paradigm, which traditionally used similar designs for AP and client radios. With MU-MIMO, there will be bulk of tasks that will be performed at AP, resulting in significant design differences between AP side chipset and client side chipset.
Due to all the nuances of implementation and sensitivity to channel conditions, comparing different MU-MIMO implementations in practical network is difficult task. Notwithstanding, I can imagine MU-MIMO becoming table stake in RFPs after Wave-2 arrives, to which everyone will answer “yes” without heed to the exact implementation details. :-)
One radical thought
Given the cost and complexity of chip level tasks required in MU-MIMO, could there be some chip family which may just use all antennas on the AP to form beam to single client at a time. That is, sequential SU-MIMO, instead of parallel MU-MIMO. What will be pros and cons? Will MU-MIMO be only incrementally or significantly better than sequential SU-MIMO? Time will tell.
Devil is in Detail!
- 802.11ac (Wave-1): Network Engineering Insights
- 802.11ac (Wave-1): MORE Network Engineering Insights
- Wi-Fi networks in 5 GHz: a few observations
- The 802.11ac Paradox by Lee H. Badman @wirednot
- 11 Commandments of Wi-Fi Decision Making
- Next Generation Wireless LANs - 802.11n and 802.11ac, excellent book by Eldad Perahia and Robert Stacey that gets into deep details of 802.11ac and MIMO