1. Additional Measurements & Analysis Confirming the RLAN Duty Cycle
Sep 12, 2018
Additional Measurements & Analysis Confirming the RLAN Duty Cycle

2. Summary
SE45 agreed to conditionally adopt a [1%] duty cycle, to be validated by measurements using current generation ac equipment available today for 5 GHz.
It was further agreed that a conversion factor would be determined to map actual 2018 test results into results expected in 2025 with ax in 6 GHz.
HPE submitted a test procedure and results using one AP and client type (SE45(18)054A1R1) showing 0.93% duty cycle for each of three ac clients simultaneously streaming video
HPE also provided a conversion factor and associated methodology, yielding 0.72% with 3 clients
Subsequent testing by JRC (SE45(18)070) and HPE with more AP vendors, client vendors and operating systems has shown three broad categories of control frame behavior for ac:
“Limited RTS/CTS” – typical of certain OS/silicon combinations in laptops (best case)
“Heavy RTS/CTS” – typical of certain OS/silicon combinations in laptop/tablet/phones
“Solo UL Ack + Heavy RTS/CTS” – typical of certain OS/silicon combinations in tablets/phones (worst case)
This contribution provides additional measurements of all three types, and a new conversion procedure.
802.11ax has been rearchitected specifically to improve MAC efficiency by AP scheduling of uplink transmissions, as a result of which control frame overhead such as RTS/CTS is minimized
After applying the conversion factor, worst case ac stations still consume 1% or less in ax. The [1%] duty cycle should be confirmed.

3. Differing Client Responses to Video Traffic
11ac clients respond differently to video streaming
Dell Windows laptop has much fewer UL TCP responses and no UL RTS/ DL CTS exchanges, as opposed to Vendor B phone
This results in the Dell laptop having an 11ac duty cycle of 1.16%, where as the Vendor B phone has 1.78% for the same test

4. Vendor B Response to DL TCP
On closer examination, after a large transfer of TCP data to the Vendor B phone, it responds with a series of “solo” and inefficient TCP Ack responses towards the AP
This forces the AP to respond with a Block Ack for each small UL packet
The Vendor B phone also protects each UL Ack transmission with an RTS/CTS exchange
Hence the term “Solo UL Ack + Heavy RTS/CTS”
Other devices with different implementations exhibit more efficient behavior

5. 802.11ax “High Efficiency”
802.11ax is officially known as “High Efficiency” or “HE”
As compared with ac which is known as “Very High Throughput” or “VHT” and focused primarily on raw layer 1 data rate performance
Among other innovations, ax enables the AP to manage the uplink transmissions of the STAs to increase efficiency
UL-OFDMA is a key feature of 11ax which introduces scheduling (by the AP) of uplink transmissions from one or more STAs
UL-OFDMA is a mandatory feature of ax standard, and will be utilized in 11ax networks including enterprise and residential because of the improved MAC efficiency

6. Basic Frame Exchange for 802.11ax Uplink Multi-user
802.11ax supports two uplink multi-user modes - either UL OFDMA or UL MU-MIMO
Trigger frame eliminates the need for the STA to use UL RTS / DL CTS to protect the UL transmission

7. Redesign of TCP “Solo Ack” Packet Exchange from 11ac to 11axAP
TCP1
TCP2
TCP3
TCP4
TCP5
TCPn
CTS BA
CTS BA
11ac: UL
Ack1
UL Ackn
STA BA
RTS
RTS AP
TCP
Trigger BA
11ax:
UL OFDMA
STA BA
Trigger frames are transmitted by AP on regular cadence, e.g. 20 ms

8. Procedure to Convert 11ac Solo Acks/RTS/CTS to 11ax
Search for all UL TCP ACK packets
UL TCP ACK frames are indicated by “Transmitter” address equal to the client MAC address, and “Protocol” equal to “TCP”
Determine 20 msec periods
Compute modulo-20msec of “Time”
Indication of the start of a new 20 ms period will be when result of a) resets (e.g. current value is less than previous value)
Note: Jumps in time larger than 20 ms (e.g. due to a period where only beacons are transmitted) can be eliminated by resetting the 20 ms period
Group the UL TCK ACK frames into 20 ms periods and sum the bytes of the TCP payload
“Length” provides the number of TCP bytes in the packet, but also includes the Wireshark radio tap header
“Header length” provides the Wireshark radio tap header length
Subtract b) from a)
The TCP payload information in a 20 msec period will be transmitted in a single HE UL OFDMA packet
HE preamble + Packet Extension: 58.4 usec
Payload (in usec): (TCP bytes)*8/125Mbps
Remaining UL packet exchange for each 20 msec period
Trigger frame: usec
BA: 32 usec
Sum 4) and 5) for each 20 msec period, and then sum over entire packet capture
In duty cycle computation eliminate contribution from UL RTS, DL CTS, UL TCP ACK, and DL BA or DL ACK and replace with 6)

9. New HPE Device Tests Adjusted for 802.11ax & 3 Client Video
Primary components to measured 11ac duty cycle
AP Beacon
DL TCP
UL BA
11ac UL RTS+DL CTS+UL TCP+DL BA/ACK
“Dummy” beacon
Conversion to 11ax
AP Beacon was previously measured as 0.29%. When amortized over three clients will be 0.1% per client
For DL TCP packets, convert from 11ac data rate to 1411Mbps
As defined in SE45(18)054A1R1
Replace “11ac UL RTS+DL CTS+UL TCP+DL BA” with “11ax trigger + UL OFDMA + DL BA”
As defined on Slide 8
Remove dummy beacon
7 different devices tested

10. Downlink TCP Conversion from 11ac to 11ax

11. Uplink Packet Exchange Conversion from 11ac to 11ax

12. Breakdown of 11ac Measured Data & 11ax Conversions

13. Conclusion
We concur with JRC’s observations that there are three broad categories of Wi-Fi device behavior while streaming HD video. These behaviors broadly result in three “tiers” of duty cycle for the same video file.
Root cause analysis of the behaviors shows that increased duty cycle for certain silicon/device/OS combinations is due to implementation differences within the ac standard.
The ax standard mitigates/eliminates these behaviors by imposing UL-OFDMA with its scheduled PHY/MAC protocol as a mandatory feature.
We have provided the conversion procedure to take a ac measured packet capture and calculate an ax duty cycle for HD video streaming.
We have shown that even worst case ac devices streaming HD video are expected to consume 1% or less duty cycle with ax in 6 GHz.
The [1%] duty cycle should be confirmed.