802.11(ac) "Gigabit" Wi-Fi Throughput

Products are appearing on the market that support the new generation of Wi-Fi standard, 802.11ac  The standard offers more throughput than 802.11(n).  To understand how much this will help, we have to dig into the nuts and bolts behind earlier Wi-Fi standards. 

802.11(b) has a maximum raw throughput of 11Mbps.  The (g)/(a) standards have a maximum throughput of 54Mbps and operate in the 2.4GHz and 5GHz bands respectively.  The (g) and (a) systems send the data in multiple “sub-carriers”, each transmitting 250k symbols per second.  The lower symbol rate makes (g) and (a) more resistance to intersymbol interference (ISI) than (b).  250k sps works out to a symbol period of 4us.  There is a 800ns guard interval between symbols.  It takes radio waves 5us to travel one mile.  The result is if a 802.11g symbol bounces off an object 120m away, that echo will be delayed by less than the guard interval and will not interfere with the next symbol.  Usually reflections from more distant objects are weak enough compared to signals from the more direct path that they do not cause problems.  When Wi-Fi is deployed in industrial environments, (ISI) sometimes presents a problem and requires directional antennas to increase the ratio of signal strength from one signal path to another. 

802.11(n) increased speed in several ways:

1. It offered the option of channels with just over twice the number of subcarriers.  This makes the Wi-Fi signal occupy 40MHz instead of 20MHz when on the 5GHz band.  (The standard provides for ways to use 40MHz channels on 2.4GHz if the channel is not crowded, but I have never seen this implemented, probably because 2.4GHz is generally very crowded.)
2. It added MIMO which allows multiple streams to be transmitted at the same time on the same frequency by creating a matrix of channel functions between each TX and RX antenna.  Inexpensive MIMO is the most amazing engineering wonder I know of, but for it to work you need multiple antennas on both sides.  It’s best if they are far apart and differently polarized.  The distance between the transmitter and receiver must not be too large compared to the antenna separation.  The antenna must have some multipath reflections but not have strong reflections delayed by more than the guard interval.
3. 802.11(n) offers the option of decreasing the guard interval from 800ns to 400ns.

When you dig into commercial equipment, it’s amazing how poorly some fallback algorithms select the data rate and number of streams.  If the environment supports MIMO, it’s usually good to use it, but sometimes it’s better to use a single stream at a higher data rate.  Which guard interval is best depends on the delay spread of the channel.  For all the amazing MIMO technology in the Wi-Fi chipsets, the driver software often does a horrible job of selecting these parameters.  The main benefit from 802.11(n) in a typical use case comes from the 40MHz channel.  Many Wi-Fi products marketed as (n) only support the 2.4GHz (n), which means they have almost no benefit over (g) products. 

802.11(ac) contains two key speed improvements over (n)

1. They again double the number of subcarriers creating 80MHz channels.
2. Higher order modulation (256 QAM) data rates area added.  256 QAM probably requires an SNR approaching the dynamic range of a cheap receiver’s ADC.  802.11n supports 64 QAM.  This provides only some improvement in data rate because much of that speed goes to forward error correction. 

Going from 40MHz channels to 80MHz channels doubles the single-stream max raw data rate from 150Mbps to 300Mbps.  The higher order modulation provides a little more throughput, bringing the raw rate of single-stream (ac) up to 433Mbps.  As with (n), the majority of the improvement comes from the “brute force” increase in subcarriers and consequently signal bandwidth.

If you’ve ever tested Wi-Fi speeds, you know that the actual throughput is less than half of the raw throughput.  If you’re sending TCP over the Wi-Fi link, the latency of the TCP ACK working its way through the radios can significantly reduce performance.  This effect is negligible on an 11Mbps 802.11(b), which provides about 4.5Mbps TCP.  As datarate increases, though, this latency can significantly reduce performance.  Latency is not normally specified on Wi-Fi equipment.

Here are the typical speeds you can expect from Wi-Fi links sending UDP on a clear channel:

The 802.11(n) standard supports up to four (4) streams, but I have not experimented with more than two (2).  The fastest link I’ve worked with, 2-stream 40MHz, was not even close to bottlenecking on the 100Mbps wired ethernet cables connected to the radio when sending TCP.

Typical users of 80211(ac) will use a single stream at 80MHz, so they will experience around 65Mbps of throughput.  You don’t get something for nothing.  When the signal bandwidth doubles, signal strength must double to get the same bit error rate.  If a hotspot is configured to a 80MHz channel, from a range standpoint that’s equivalent to cutting its output power in fourth compared to a 20MHz channel.  Going from 2.4GHz to 5GHz cuts the power in fourth again due to increased path cost at higher frequencies.  The end result is a 80MHz (ac) link will have less range than a (g) link.

MIMO and high-order RF modulation are amazing technical feats that won’t provide that much benefit to the user.  I am very confident that they will find use in industrial equipment.  It remains to be seen how long it takes before crowding pushes commercial users to exploit the technology in these standards.  As of today, much consumer equipment supports the 2.4GHz band only.  Using the less-crowded 5GHz bands is the easiest avenue for the typical user to increase performance.  The (ac) standard requires 5GHz support.  This fact may affect users more than all the amazing technology in (n) and (ac).