When Is 54 Not Equal to 54? A Look at 802.11a, b, and g Throughput
Pages: 1, 2

802.11g

Performing the same calculations for 802.11g is a bit tricky. 802.11g operates in the same frequency band as 802.11b, and is required to remain backwards-compatible. The encoding used by 802.11g will not be recognized by 802.11b stations, so "protection" mechanisms are defined to limit the cross-talk in mixed b/g environments. Essentially, the protection mechanisms require that 802.11g stations operating at high rates pre-reserve the radio medium by using slower, 802.11b-compatible reservation mechanisms.

802.11g SIFS = 10 µs
802.11g short slot time = 9 µs (802.11g-only mode with no legacy stations)
802.11g long slot time = 20 µs (mixed mode requires slow slot time)

802.11g uses many of the same timing parameters as 802.11a. However, it is saddled with backwards compatibility requirements. It inherits the short 10 µs SIFS time from 802.11b, but the high-rate coding in 802.11g needs additional time. Therefore, 802.11g adds a 6 µs "signal extension" time at the end of every frame.

802.11g-only BSS

When no 802.11b stations are present, no protection is required. This situation is extremely unlikely, given the huge installed base of 802.11b cards.

As always, start with the basic timing parameters:

• 802.11g SIFS = 10 µs
• 802.11g fast slot time = 9 µs (can only be used when no 802.11b stations are present)
• 802.11g DIFS = 2 x Slot time + SIFS = 28 µs

The 802.11g ERP-OFDM PHY is nearly identical to the 802.11a PHY, except that it operates in a different frequency band and uses a shorter SIFS time. Physical layer headers are identical, as is the coding. Therefore, the calculation for the time required to transmit a frame is nearly identical, with only minor changes to the interframe space times.

The transaction length of 802.11g is identical to 802.11a. The interframe space is slightly shorter, but this is offset exactly by the signal extension field.

Protection 1: CTS-to-self

Once the first 802.11b station associates with an 802.11g access point, however, protection is required. The minimal protection contemplated by the standard is that 802.11g stations will protect the fast 802.11g frame exchange with a slow Clear To Send (CTS) frame that locks out other stations access to the medium. Protection dramatically reduces the maximum theoretical throughput because the additional CTS transmission is required with its long 802.11b headers.

Longer interframe spacing is required when legacy clients are connected and protection is engaged. The short slot time is only available when no 802.11b stations are present. Once they are present, the frame spacing reverts to the 802.11b standard:

• SIFS = 10 µs
• Slot time = 20 µs
• DIFS = 2 x Slot time + SIFS = 50 µs

A CTS frame is 14 bytes. It will be transmitted at the highest rate understood by all stations attached to the access point, which will be at most 11 Mbps. The CTS frame will be encoded quickly, but only after its long header. Following the CTS will be the Data-ACK sequence at the high 802.11g speeds.

The total transactional time is over twice as long because of the protection mechanism. At 898µs per transaction, only 1,113 transactions can complete per second, and the throughput drops dramatically to 13.0 Mbps.

Protection 2: RTS-CTS

Using only a CTS frame to reserve the medium is the minimum requirement, but it may fail in some cases where there are so-called "hidden nodes" that do not see the CTS. To fully reserve the medium, the initial edition of the 802.11 standard included a two-frame exchange that would fully announce the impending transmission composed of a Request To Send (RTS) frame followed by the CTS frame. Although the standard requires only a CTS-to-self, using the full RTS/CTS will better protect the inner exchange from interference. The final calculation is quite similar to the previous one, with the addition of the RTS frame at the start:

The total transactional time is even longer because a more robust (and hence, time-consuming) protection mechanism is used. A full RTS/CTS exchange makes the TCP ACK require more time than a straightforward 11 Mbps transmission using older 802.11b encoding. Therfore, a transaction would consist of the TCP data protected by RTS/CTS at 774 µs followed by a TCP ACK at the 511 µs previously calculated for 802.11b. The total transaction time is 1,285 µs per transaction, so only 778 transactions can complete per second, and the throughput drops back into single digits--9.1 Mbps

Final Thoughts

No matter how you look at it, 802.11g is significantly faster than 802.11b. However, once an 802.11b station associates to an 802.11g network, the throughput drops dramatically, because protection must be activated. The 802.11b station does not need to actively send data to cut the throughput; it just needs to be associated, so that protection is enabled. Mixed 802.11b/g deployments are likely to be common for the foreseeable future, especially in situations where there is no control over client adapters. 802.11a networks can sustain much higher data rates than 802.11g networks with protection enabled, and 802.11a offers the added advantage of more radio channels for easier layout of high-density deployments. 802.11g offers a worthwhile speed advantage over 802.11b, but it does not challenge 802.11a for the performance crown.

Matthew Gast is the director of product management at Aerohive Networks responsible for the software that powers Aerohive's networking devices.

Return to the Wireless DevCenter

Comments on this article

Showing messages 1 through 10 of 10.

• What about backoff?
  2007-06-07 10:05:01  foobarian [View]
  
  
  Where are you accounting for the random backoff? Stations are supposed to pick a random number from 0-31 and count down that many slots before sending.

• What is speed negotiation's influance
  2006-02-28 13:28:03  Dick9 [View]
  
  
  I understand that weaker stations initiate raw speed negotiations (down from 54mb) that are then used for all others. What is the influance of this downgrading/upgrading process on throughput of 802.11g?

• Wireless G in Mixed Mode
  2003-12-11 14:38:12  anonymous2 [View]
  
  
  I'm currently using a LinkSys "G" series in mixed mode. Looks like I can speed up my wireless throughput by changing to G only since I don't have any B devices.

• question about cts-to -self
  2003-09-08 20:35:43  anonymous2 [View]
  
  
  In IEEE802.11g specification,cts-to -self mechanism is described as "when a node want to distribute NAV information,the node may first transmit a cts frame with the RA address equal to its own mac address and with a duration time that protect the pending transmition".
  From this article,i don't understand why the author said"Once the first 802.11b station associates with an 802.11g access point, however, protection is required. The minimal protection contemplated by the standard is that 802.11g stations will protect the fast 802.11g frame exchange with a slow Clear To Send (CTS) frame that locks out other stations access to the medium".
  I don't completely understand cts-to-self mechanism.Would you please explain it to me?
  my email:stella.li@netac.com
  
  
  
  
  

• Corrections made/new revision date
  2003-08-15 16:15:40  Matthew Gast | [View]
  
  
  The date on the article now reads "revised 8/14/03" because two corrections were made.
  
  The smaller correction is that the analysis did not mention the six OFDM tail bits used in 802.11a and 802.11g. Six bits are added as a pad at the end of the data for use with the convolution code. As it turns out, adding six bits to the end of any of the frames discussed in this article does not change the number of symbols, so it does not change any of the calculations.
  
  More substantially, I had neglected the 6 microsecond signal extension time in 802.11g. One anonymous talkback comment as well as a private e-mail pointed out my error. The tables and text have been revised to correct the oversight. The effect of the correction is that an 802.11g-only network will have identical throughput to an 802.11a network.

• Apple's Airport Extreme Basestation
  2003-08-12 03:01:27  anonymous2 [View]
  
  
  I would assume this to be the case. My reasoning is that my linksys B/G AP has the same option and in one of the early reviews (tomshardware I believe) it was mentioned that transfer was slower in mixed mode than when it was set to G only.

• Correct?? Pure 802.11g Goodput in this article has some errors...
  2003-08-11 03:56:49  anonymous2 [View]
  
  
  802.11g(D8.2) pure network (ERP-OFDM case) has 'Signal Extension'(=6us)time for convolutional decoding per all OFDM packet(All PHY scheme so called XXX-OFDM in 802.11g spec. has Signal Extention.). 802.11g's goodput almost equals to 802.11a's goodput.

• Apple's Airport Extreme Basestation
  2003-08-10 17:14:01  anonymous2 [View]
  
  
  I understand that Apple's Airport Extreme basestation allows you to specify the mode you want it to run in - 802.11b compatible, mixed 802.11b/g, or 802.11g only. If I were to specify the basestation to run in only 802.11g mode, would this allow it to utilize the full speed of 802.11g because it would disregard any attempt for a 802.11b card to be associated with it?
  
  Hope this makes sense.
• Apple's Airport Extreme Basestation
  2003-08-12 02:57:47  anonymous2 [View]
  
  
  I would assume this to be the case. My reasoning is that my linksys B/G AP has the same option and in one of the early reviews (tomshardware I believe) it was mentioned that transfer was slower in mixed mode than when it was set to G only.
• Apple's Airport Extreme Basestation
  2003-08-10 23:38:27  anonymous2 [View]
  
  
  
  I think you are correct but I don't know for sure.