At the end of last week, I attended the symposium put on by the Wireless Networking and Computing Group in Austin, Texas. On Friday, I spoke as part of the Wireless LAN technical track, comparing security protocols used on 802.11 networks. Immediately before I spoke, I attended a talk by Sean Coffey of Texas Instruments about 802.11n.
802.11 task group N has a goal of 100 Mbps of net throughput, after subtracting all the overhead for protocol management features like preambles, interframe spacing, and acknowledgements. TGn’s goal is interesting in that most other IEEE groups tend to focus on the peak throughput, but TGn was founded with a user throughput goal in mind.
Coffey’s talk was titled “Robust High Throughput High Spectral Efficiency Wireless LANs for 802.11n.” Spectral efficiency has been a sore point for 802.11 users. At the beginning, Coffey noted that although 802.11a and 802.11g have a peak “headline” rate of 54 Mbps, you can only expect about 25 Mbps net throughput, after accounting for the protocol. The net throughput is only about 45% of peak throughput. To get to 100 Mbps net throughput, you have two options: increase the peak speed way past 100 Mbps, or increase the efficiency of the protocol. Different proposals to TGn have placed different emphasis on these tasks.
(His point about efficiency is well taken. As a broad rule of thumb, the 802.11 MAC can move user payload data at about half the peak rate. I recently ran an AP forwarding test that came in at 31 Mbps, but that was in the ideal conditions of an RF isolation unit. 31 Mbps significantly higher than 25 Mbps, but it is still only 57% of the peak rate.)
Efficiency can be improved by reducing the amount of air time devoted to protocol operations. Some of these are already available in proprietary “turbo” modes in existing 802.11g hardware. First, larger frame sizes boost efficiency by improving the ratio of payload data bits to overhead bits. (This is the same argument for jumbo frames on Ethernet.) Second, block acknowledgements can further improve that ratio by requiring fewer acknowledgements. (TCP already does this.) Using larger frames does not help unless they can be reliably transmitted. Modulation must be designed to keep the packet error rate down. Larger frames do improve efficiency, but only if they are received intact. Retransmissions will quickly eliminate the benefit of larger frames, especially if a block acknowledgement is lost and the entire block must be retransmitted.
TGn received several proposals. Four complete proposals were received from WWiSE, TGnSync, Mitsubishi and Motorola, and Qualcomm. (Texas Instruments, with whom Coffey is affiliated, is part of WWiSE.) Several features are common to the leading proposals. All use 2×2 MIMO, which uses two input and two output antennas. (As an interesting aside, a student from UT-Austin who attends IEEE meetings told me that there was a vote on the pronunciation of MIMO, and it’s been standardized as “MyMoe.”) All of them also retain the 20 MHz channels in use by 802.11a. At this point, not all regulators will allow wider channels. Many optional features have also been proposed, most notably 40 MHz channels and the use of more than two antennas.
Mathematically, WWiSE gets to 100 Mbps net throughput by defining a new peak rate modulation that runs at 135 Mbps, and improving the efficiency of the protocol through the use of a burst transmission and block acknowledgements. In the WWiSE proposal, three 4,000 byte frames can be transmitted as a single block. To achieve 100 Mbps throughput, the three frames must be transmitted in 960 microseconds. WWiSE’s 135 Mbps modulation will move the frames in 711 microseconds, with the extra 249 microseconds used for backwards-compatible preambles, the interframe spaces, and the block acknowledgement.
To run at 135 Mbps, WWiSE uses the channel slightly differently from 802.11a. 802.11a divides the 20 MHz channel into 54 subcarriers. It uses 48 to carry data, and four for “pilot” carriers used to calibrate the data carriers. WWiSE divides the channel into 56 carriers. 54 are used for data, and 2 are pilots. Coffey said that the use of MIMO means that the two carriers go through two receivers each, and are as effective as four pilot carriers through one antenna. Each of the 54 carriers can be modulated using the same techniques as 802.11a, but a new 5/6 convolution code is used for the top data rate of 135 MHz. (The top code rate in 802.11a is 3/4.) 40 MHz channels can also be used with WWiSE, and it doubles the channel capacity.
Coffey believes that receivers need to operate at a signal-to-noise 30 dB to be viable. Any less than that, and the range will be short as receivers struggle to decode the radio signal. Part of the reason WWiSE kept the peak data rate low and focused on improving efficiency is so that the receivers could be kept relatively simple and inexpensive. (In fact, channel simulations using various modulation techniques showed that the packet error rate at 30 dB was probably too high using the modulation already used by 802.11a, which is why WWiSE includes a new error correcting code.)
All in all, it was an interesting session for me. My 802.11 background is from the MAC layer up. Although I enjoyed the peek into a new PHY in progress, I haven’t read the technical proposal from TGnSync yet, so I can’t compare the two.