Chan et al. (Cisco Systems) Slide 1. doc.: IEE

1
New Evidence that 11n Greenfield Devices Causes
False RADAR Detections on DFS Channels

• Date 2008-03-17

Authors
2
Latest tests provide conclusive evidence that
Greenfield causes false detects in 11a devices on
DFS channels

• Since D2.0, weve presented tests that showed
  Greenfield transmissions can cause false detects
  in 11a devices on DFS channels
• 07/0329r2 (Mar 2007) and 07/2849r0 (Nov 2007)
  showed software MATLAB generated GF signals and
  those of VoIP traffic patterns cause false
  detects on 11a devices from different vendors
• There were questions if this issue exists with GF
  transmissions by actual 11n hardware
• So we performed tests with real VoIP traffic on
  WiFi certified Draft 11n Testbed devices and
  observed the same intensity of false detects
• This is documented in 08/0301r1 and presented at
  the TGn pre-meeting last week
• Because these tests involved one VoIP codec and
  were performed in a screen room, there were
  questions whether the codec was cherry-picked and
  whether the screen room represented real-world
  environment
• Over the weekend, we performed the same tests
  with multiple different VoIP codecs and also on
  an open-space of an operating 11a deployed
  network
• We observed the SAME INTENSITY of false detects
  on 11a devices
• This new evidence dispels any uncertainty raised
  by doubters of this effect.
• We can safely conclude that GF on DFS is indeed a
  real and common problem

3
Latest tests with WiFi draft 11n testbed devices
show GF-DFS problem is beyond theoretical and
commonly occurs
Test Setup
Neighboring APs in range but on different
channels
Vendor Y (HT Greenfield Client on laptop)
Bi-directional VoIP streams
WAN
Radar Detects!!!
Vendor Z (HT Greenfield AP)
Channel 52
Vendor Y (HT Greenfield Client on laptop)
Vendor X (802.11a device)
11a clients generating real over the air network
traffic
Channel 52
4
Same intensity of false detects with different
VoIP codecs and open real WLAN environment

• More details on the test and setup
• VoIP streams were generated by IxChariot,
  industry designated network traffic generation
  and testing tool for WiFi certifications
• Codec used include G.711U, G.723.1 (MPMLQ),
  G.723.1 (ACELP) and G.726 with default settings.
• These codecs are very different and have large
  range of variations in their parameters, eg.
  packet size in bytes and time, interpacket
  spacing. (see backup slide for details)
• Tests performed with various MCS (eg. 3, 4, 5, 6,
  7 and 15)
• False radar triggers began on every trial shortly
  after VoIP traffic began, eg. within 5 minutes
• Results did not change when laptop clients were
  loaded with ping traffic
• Vendor X Legacy APs did not have any record of
  falsing before tests commenced or between tests

5
Sample screen shot of Chariot VoIP test
6
Detrimental consequences expected from GF on DFS
Bands

• Operations of legacy 802.11a networks which have
  no concept of Greenfield mode would be disrupted
  by their false detects from GF transmissions by
  moving to another channel each time
• Many mesh network architectures use the 5 GHz
  band for backhaul
• A single voice call using GF transmissions could
  bring down a mesh tree while it changes channel.
• A small number of GF APs using efficient channel
  selection can totally occupy the 5 GHz band and
  cause a mesh network outage.
• This type of behavior also facilitates
  possibilities of simple denial of service attacks
• There is nothing in the DFS regulations that
  indicate radar may be ignored if preceded by MAC
  protection. Therefore protection is ineffective
  for GF preambles in DFS bands.

7
GF on DFS problem is not a late-breaking issue
and is an industry-wide 802.11 concern

• In LB 97 (Draft 1.0), there were CIDs which
  pointed out that GF transmissions can be falsely
  detected by legacy devices in the DFS band as
  radar
• We performed experiments and presented a
  submission, 07/0329r2, in March 2007 (Orlando) to
  discuss the results
• Strawpolls showed a significant fraction of the
  TGn Coex Ad Hoc members are concerned with this
  problem, but more investigations should be done
  to be certain
• We performed a set of measurements with another
  legacy 11a receiver and presented them in
  submission 07/2849r0 in Nov 2007 (Atlanta)
• Strawpoll showed an even more significant
  fraction of the TGn Coex Ad Hoc members a
  majority agreeing for a text change to address
  this
• We hypothesized and presented in Feb 2008 telecon
  with 08/0111r2 that the problem may be related to
  DFS requirement for detecting the bin-5 radar
  profile
• Now we show compelling and conclusive evidence of
  this problem
• Overall our tests have shown at least two
  different 11a chipsets and at least two different
  vendors that would have falsing issues due to
  software generated GF VoIP transmissions
• The problem is probably not limited to VoIP GF
  transmissions too
• This is an industry-wide 802.11 concern

8
802.11n should be changed to prevent GFs
potentially disruptive effects to legacy 11a
devices in the DFS bands

• Therere two options to solving this problem
• 1. Prohibit Greenfield operations in DFS bands
• or
• 2. Define a suitable mechanism to prevent
  Greenfield operation in DFS bands in the presence
  of legacy 11a devices
• Simple to implement since it reuses existing 11n
  schemes to signal when GF can be used.
• Involves only a software upgrade/change.
• More importantly, this mechanism doesnt affect
  11n GF evolution path, as 11a devices get phased
  out in the next few years, GF wouldnt be
  prevented from use due to this prohibition.
• True to the definition of having a greenfield.

9
Illustration of Option 2s proposed mechanism
AP detects non-HT OBSS (1/4)
Operation on a DFS Band
Covered by proposed text changes in 08/0302r5.
HT Greenfield Transmissions
HT Greenfield Transmissions
Beacon
HT Greenfield AP
Non-HT AP
HT Greenfield Clients
During operations or when establishing a BSS, an
HT Greenfield AP receives beacon from a non-HT
AP, thus detecting a non-HT OBSS.
10
Illustration of Option 2s proposed mechanism
AP detects non-HT OBSS (2/4)
Operation on a DFS Band
Covered by proposed text changes in 08/0302r5.
Beacon
HT Greenfield AP
Non-HT AP
HT Information Element
OBSS Non-HT STAs Present
HT Greenfield AP sets its Greenfield support bit
from 1 to 0 and OBSS Non-HT STAs Present bit from
0 to 1.
0 1
11
Illustration of Option 2s proposed mechanism
AP detects non-HT OBSS (3/4)
Operation on a DFS Band
Covered by proposed text changes in 08/0302r5.
HT Mixed Mode Transmissions
HT Mixed Mode Transmissions
Beacon
HT Greenfield AP
Non-HT AP
HT Greenfield Clients
Greenfield transmissions are then suppressed
across this BSS. Non-HT OBSS is not disrupted by
11n BSS.
12
Illustration of Option 2s proposed mechanism
AP detects non-HT OBSS (4/4)
Operation on a DFS Band
Covered by proposed text changes in 08/0302r5.
After waiting 30 min
HT Greenfield AP
Non-HT AP
HT Information Element
OBSS Non-HT STAs Present
If non-HT AP does not exist anymore, HT
Greenfield AP can revert to its previous settings
after thirty minutes. (Thirty min is a value
defined as a the MIB variable that has range from
30 min to 48 hrs.)
1 0
13
Proposed prevention mechanism is simple and low
impact to existing 11n implementations

• Merits of the proposed prevention mechanism
• Simple to implement and deploy via a software
  upgrade
• No monitoring or scanning on clients at all
• Monitoring of non-HT OBSS only on the AP
• Nothing required of clients other than changing
  their behavior according to APs beacon (which is
  the default and expected behavior for 11n STAs),
  and disabling GF for other (i.e. DLS) links on
  DFS channels when GF is disabled by the AP
• No new fields but re-uses existing 11n bits and
  signaling schemes
• Minor changes from D2.0 behavior
• Preserves 11n GF evolution path
• Achieves true definition of having a greenfield.

14
References

• Compliance Measurement Procedures for
  Unlicensed-national Information Infrastructure
  Devices Operating In The 5250-5350 Mhz and
  5470-5725 Mhz Bands Incorporating Dynamic
  Frequency Selection, Appendix to Revision of
  Parts 2 and 15 of the Commissions Rules to
  Permit Unlicensed National Information
  Infrastructure (U-NII) devices in the 5 GHz band,
  FCC 06-96, June 30, 2006.
• Submission 07/0329r2
• Submission 07/2849r0
• Submission 08/0111r2
• Submission 08/0301r0
• Submission 08/0302r0

15
Backup slides
16
Default Parameters of VoIP Codecs show large
variations
Excerpted from http//www.cisco.com/en/US/tech/tk6
52/tk698/technologies_tech_note09186a0080094ae2.sh
tml
17
Excerpted from Nov 2007 (Atlanta) Coex Ad Hoc
Minutes
18
Recap of previous investigations on this issue
19
Recap of previous investigations on this issue