November 2007 doc.: IEEE 802.11-07/2752r1 July 2008 Managed Contention Access – A technique to improve Video Streaming Performance Date: 2008-07-013 Authors: Jing Zhu, Intel Corporation Peter Ecclesine, Cisco Systems

November 2007 doc.: IEEE 802.11-07/2752r1 July 2008 Abstract This submission describes Managed Contention Access (MCA), a technique that enables semi-deterministic access to 802.11 wireless medium. Jing Zhu, Intel Corporation Peter Ecclesine, Cisco Systems

What is the problem? Intermittent Channel Access caused by unsaturated / real-time traffic (video, voice, multimedia, etc.) power saving & multi-radio Contending STAs with saturate traffic increased channel access time / jitter reduced sensitivity and increased collisions EDCA/HCCA does not work well in OBSS

Problem Statement: Increased Delay and Jitter Maximum MSDU size: 2304 bytes (at 1Mbps)  18ms 1500 bytes (at 1Mbps)  12ms Data ACK STA1 (FTP) Delay (up to ~18ms) Data ACK STA2 (Video) Arrival General Mouse Hi precision Mouse Keyboard Voice Quality Headset CD Quality headphones Aggregate Throughput < 8kbps 256 kbps 384kbps Latency < 10ms < 3ms < 50ms < 30ms < 100ms Delay Sensitive Throughput Sensitive

Problem Statement (cont’d): Reduced Sensitivity in Post-Absence Channel Access (power save, etc.) Transmission DIFS + BO STA #1: ACK may be out of range of preamble detection (hidden) Data ACK ACK Neighboring STA PLCP Data ACK ACK #2: CCA without preamble detection may not be sensitive enough to detect Data ACK #1 and #2 will cause the performance degradation of the whole BSS Energy Detection Threshold = -62dBm Preamble Detection = -82dBm 20dB loss of CCA sensitivity 90% CCA sensing rage reduction with path loss exponent = 2

How to manage contention access ?

State of the Art: EDCA Admission Control Admissible Parameters Description G.711 (UDP) Data (TCP) Nominal MSDU Size octets 200 (inter-arrival time:20ms) 1500 Minimum Service Interval us (minimum interval between two sucessive SPs) 20ms Maximum Service Interval us (maximum interval between two successive SPs) Inactivity Interval us (minimum inactive time between successive Tx or Rx activities before the TS being deleted from HC) Suspension Interval (< Inactivity Interval) us (minimum inactive time between successive Tx or Rx activities before the polling for this TS being stopped Minimum Data Rate bps (lowest data rate for transport of MSDU) Mean Data Rate bps (average data rate for transport of MSDU) 64Kbps (<10% duty cycle) 100Kbps (10% duty cycle) Burst Size Octets (maximum bytes of the MSDUs arriving at the peak data rate) Minimum PHY Rate desired minimum PHY rate 11Mbps 1 Mbps Peak Data Rate Maximum allowable data rate Delay Bound us (maximum time for a MSDU spent at the MAC layer) 50ms Surplus Bandwidth Allowance 1.X (how much more bandwidth is allowed to be used for the TS) 1.3 (30% retransmissions) 1.3 (30% retransmission) Medium Time 32us/s (the amount of time admitted to access the medium) HC output

How “Medium Time” is used in EDCA admission control ? Two state variables: admitted_time and used_time Two parameters: medium_time and dot11EDCAAveragePeriod Algorithm: a) At dot11EDCAAveragingPeriod (default = 5) second intervals used_time = max((used_time – admitted_time), 0) b) After each successful or unsuccessful MPDU (re)transmission attempt, used_time = used_time + MPDUExchangeTime c) On receipt of a TSPEC element contained in a ADDTS Response frame indicating that the request has been accepted admitted_time = admitted_time + dot11EDCAAveragingPeriod * (medium time of TSPEC).

Limitations of EDCA-AC “It is recommended that admission control not be required for the access categories AC_BE and AC_BK.” (IEEE 802.11-2007) “Medium Time” (32us/s) usually too long to limit per-packet transmission time TCP duty cycle =10% and dot11EDCAAveragingPeriod = 5 seconds, retransmission = 30% (Surplus Bandwidth Allowance = 1.3) medium time = ceiling(10% x 1000 / 32) = 4 adimitted_time = 4 x 5 x 32 (us) = 640 ms Difficult to choose “TXOP limit” to trade between MAC efficiency and latency: 10ms? or 1ms? No restriction on OBSS traffic No restriction on actual timing  need something that can be controlled by AP but with minimal restrictions on AC_BE and AC_BK

Basic Idea: MCA-Aware Channel Access MCA-aware channel access shall both start after the end of a MCA slot and end before the start of a MCA slot STA DIFS Back-Off Slots Data SIFS ACK MCA slot MCA interval

Managed Contention Access (Cont’d) Beacon Interval Start Time MCA Period Legacy Zone MCA Zone MCA Zone MCA Zone MCA Duration CTS- to-Self MCA- Allowed SIFS MCA interval MCA slot MCA-Unaware STA NAV MCA-Aware STA NAV Slide 11 <first name> <last name>, Intel Corporation

Managed Contention Access MCA Zone is divided into multiple MCA intervals started with a MCA slot with fixed duration MCA slot is a short Guard Interval providing re-sychronization point for STAs contending the channel At the start of a MCA Zone, the AP transmits a CTS-to-Self (to protect the following MCA Zone) Waits for SIFS period and a MCA-Allowed frame The CTS-to-Self shuts off legacy STAs The MCA-Allowed permits the MCA-capable STAs to contend the channel despite of CTS-to-Self Slide 12 <first name> <last name>, Intel Corporation

Example: Mixed VoIP and Data 60 MCA slots, one TX per slot, 30% rTX  ~20 VoIP (bidirectional) connections 102.4ms (Beacon Interval) 22.4ms (Legacy Zone) 30ms (MCA Zone) 20ms (Legacy Zone) 30ms (MCA Zone) 1ms 1ms … … 1ms 1ms 1ms MCA Configurations: start time = 22.4ms, MCA period = 50ms MCA duration = 30ms MCA interval = 1ms

MCA-Aware OBSS MCA-slot schedule shall not be created (owned) by BSS that does not have AC_VI or AC_VO STAs to reduce the number of different MCA-slot schedules in OBSS OBSS MCA-Aware APs shall copy each other’s MCA-slot schedule AP may reuse the existing MCA-slot schedule by other OBSS as much as possible Owner: AP that generates a MCA-slot schedule Follower: AP that copies other AP’s MCA-slot schedule, and needs to update when such information is changed over time due to clock drift. STA may report MCA schedule of other OBSS if such information is not included in the Beacon. OBSS APs may not be in each other’s range

MCA-Aware OBSS (cont’d) T (BSS1 MCA interval) BSS1 T (BSS2 MCA interval) BSS2 d (OBSS MCA interval) T – d (OBSS MCA interval) OBSS It is recommended that MCA-Aware OBSS use the same duration of MCA slot, and synchronize / align with each other as much as possible, i.e., d T-d

MCA-Unaware OBSS CTS-to-Self will prevent MCA-Unaware OBSS to contend with MCA-aware STAs

MCA Violation Problem: Recommendation: OBSS STA may still violate MCA rules unconsciously due to not receiving beacon or CTS-to-Self (collision, PS, etc.) Recommendation: detection: AP may detect it by counting the number of MCA slots being overlapped by transmissions. STA may also help detect it and report such event to AP action: AP may modify MCA slot schedule or reduce / remove MCA zones completely.

Summary Main Changes to IEEE 802.11 Standard Benefit: Cost: MCA-slot Aware Channel Access MCA-Allowed Control Frame New MCA IE in Beacon Frame Benefit: reliability, power efficiency, latency/ duty-cycle OBSS support Cost: efficiency (total network throughput) loss complexity

Straw Poll #1 Would you like to see a normative text proposal on the idea of Managed Contention Access? Yes No Abstain