I believe OFDMA (orthogonal frequency-division multiple access) will be one of the most significant improvements to Wi-Fi in a long time. Throughout the evolution of the 802.11 protocol new standards have tended to focus on improving speed through the use of more complex modulation. Complex modulation by its very nature is more susceptible to noise and interference making it more prone to retransmissions or rate shifting to less complex modulation. Retransmissions and rate shifting lead to overhead which degrades the performance of wireless networks. The 802.11ax amendment, also referred to as High Efficiency (HE) wireless, takes a different approach and focuses on improving efficiency (although it does also introduce QAM 1024).
While 802.11ax contains many improvements the new OFDMA physical layer (PHY) specification is a major factor in improving efficiency. The improvement is gained by allocating spectrum over both time and frequency. A key component to this spectrum allocation is the use of Resource Units (RU). Before discussing the benefits of RUs it’s important to look at some fundamental changes to the PHY specification.
The 802.11n/ac (HT/VHT) OFDM PHY specifies the width of each subcarrier at 312.5 kHz. A 20 MHz channel is comprised of 64 subcarriers with 52 carrying data, 4 pilots, and 8 guard bands (null) as illustrated below. This great illustration was taken from a presentation given by Perry Correll of Aerohive Networks at WLPC 2018 in Prague (slides available here).
With OFDMA the subcarrier width decreases to 78.125 kHz resulting in 256 subcarriers for a 20 MHz channel. The new subcarrier composition includes 234 carrying data, 8 pilots, 3 DC (direct current; used to locate the center of the OFDM frequency), and 11 guard band. That’s 4.5 times as many data carrying subcarriers! Now back to RUs.
With OFDM, an entire 20 MHz channel was allocated to a single station transmission. With OFDMA a channel can be divided into multiple RUs, each comprised of various subcarrier widths. Valid OFDMA RU allocations for a 20 MHz channel width include 26, 52, 106, and 242 subcarriers (note: a client station can be assigned to only one RU). What I find interesting is the RU allocation across the channel can mix and match although there are fixed combinations. What’s even more interesting is that an AP can transmit each RU at a different power level and data rate. The chart below from the fantastic Rohde & Schwarz IEEE 802.11ax Technology Introduction white paper illustrates a 20 MHz channel and the various resource unit allocations.
Another helpful illustration is shown below, also taken from Perry Correll’s presentation.
For example, with a single 20 MHz channel one AP could assign four client stations a 26 subcarrier RU and two client stations a 52 subcarrier RU on the same channel. The RU allocation could be based on differing throughput needs per client. Alternatively, one could assign nine client stations a 26 subcarrier RU (the two 13 subcarrier RUs at the center of the channel represent make up a 26 subcarrier RU). The following image from the same Rohde & Schwarz white paper demonstrates one possible RU allocation combination across three clients.
Imagine the efficiency improvements from a medium contention and overhead perspective of simultaneously transmitting to nine client stations. OFDMA is effectively breaking down a 20 MHz channel (or 40/80/160 MHz) into smaller sub-channels. This a complete reverse course from 802.11n and 802.11ac where channels were bonded to improve overall throughput for a single station transmission.
After reading about OFDMA, the first thing that came to my mind was is this applicable in the downlink only, or uplink as well? With CSMA/CA in mind, a single 802.11ax AP station contending for airtime and transmitting an OFDMA transmission containing multiple RUs destined for multiple clients seems feasible. Lets take a look at how this multi-user communication process takes place downlink (DL).
With 802.11ax the AP will send a multi-user request-to-send (MU-RTS) trigger frame (control frame) including RU allocations for specific client stations. Each client will then respond with a CTS frame. The AP then sends a multi-user downlink PPDU followed by a block acknowledgement request (BAR). Finally, the clients will respond with a block acknowledgement to complete the exchange. This exchange is illustrated below and was also taken from Perry Correll’s presentation at WLPC 2018.
What’s fascinating to me is that in the example above, four client stations are transmitting to a single AP station radio simultaneously within the same transmit opportunity (TxOP). This is pretty slick compared to the traditional OFDM behavior where only a single station can transmit at once. While the multi-user downlink seemed intuitive (with the exception of the simultaneous CTS frames) the uplink direction initially had me scratching my head. As any wireless professional knows with CSMA/CA only one station can transmit at a given time.
Before we look at the mechanics of uplink multi-user OFDMA we need to understand how the AP learns about client uplink traffic needs. The AP can learn which clients have traffic to send one of two ways: explicitly and implicitly. One method involves the AP sending a Buffer Status Report Poll (BSRP) trigger frame, to which clients will respond with a Buffer Status Report (BSR). This explicit method (shown below) is inefficient due to the control traffic associated with polling clients but there is a more efficient alternative discussed later.
Now for the mechanics of the exchange. To coordinate uplink MU-OFDMA the AP will again contend for airtime. Once the AP has obtained a TXOP it may send a BSRP trigger to learn about traffic needs with clients responding with a BSR. Next the AP will send another MU-RTS trigger frame indicating which client stations can simultaneously transmit back. The client stations will respond by sending a CTS to the AP. The AP will then send another trigger frame effectively initiating the simultaneous uplink MU-OFDMA transmissions. At this point each client will send their data upstream and add padding as necessary (each uplink PPDU must be the same size). Finally the AP sends a multi-user block acknowledgement. 
Once clients are on the network and passing traffic the AP can glean information contained in the QoS Control Field of client transmissions to learn about traffic buffers. This implicit method reduces, and possibly eliminates, the need for the AP to poll clients with a BSRP thereby further improving efficiency in the uplink direction.
To me this AP coordinated trigger frame mechanism seems a little like the Point Coordination Function (PCF) MAC architecture that never caught on. The AP is almost acting as an arbitrator, directing multi-user transmissions in both the downlink and uplink.
One question that come to mind include how does one ensure client A cannot snoop in on a RU destined to client B in the downlink? I have not found any documentation to confirm this but I would suspect each RU within an AP downlink transmission is encrypted with a key unique to the client. I would imagine this increases the load on the AP CPU, with the AP processing multiple data streams through an encryption algorithm before a transmission with the reverse decryption process as well. However, 802.11ax will require new hardware which will likely have increased CPU and RAM to handle the additional load.
Another question that comes to mind is how does the mult-user aspect of OFDMA interact with WMM? If nine clients are transmitting simultaneously upstream the contention window mechanism is not in use. The same could be said for the downstream. I suppose the AP will schedule clients with high priority traffic in the multi-user transmissions more frequently than clients with lower priority traffic?
The drawback at this time is that in order to take advantage of this efficiency improvement the environment needs 802.11ax compliant, or Wi-Fi 6 certified, client stations which are not widely available at the time of this writing. I’m excited about the possibility of transmitting to nine times as many clients in a single TXOP. Time will tell if this produces a net gain in terms of overall throughput delivered on a channel during a given transmission. It will be interesting to see how we measure any performance improvement OFDMA has over allowing a single client to transmit at the highest MCS rate possible.
Hopefully you found this post informative. If you’re interested in learning more I strongly encourage you to read the white paper referenced in the post, and to watch the presentation from Perry Correll at WLPC.
cradtech 