Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 1 ‘Cyclic Prioritized Multiple Access (CPMA): An Access Mechanism for Contention-Free Sessions’ Mathilde Benveniste AT&T Labs, Research

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 2 CFS and PCFS: Definitions Contention free session (CFS) Any frame exchange sequence that may occur without contention following a successful channel access attempt. A CFS may involve one or more stations. A CFS may be initiated by any station. –A Contention-Free Burst and an RTS/CTS exchange are both examples of a CFS Periodic contention free period (PCFS) A CFS that must occur at regular time intervals. –A Contention-Free Period is an example of a PCFS Both PCFSs and CFSs are needed; the former for periodic traffic, the latter in order to use channel time efficiently, as channel availability permits –When restricting the time to the next access attempt, the channel cannot be used sooner, even if needed and available; it limits efficiency of channel re-use For simplicity, but without loss of generality, we assume that CFSs/PCFSs are initiated by APs

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 3 Introduction In a multi-BSS system, there will still be interference between BSS assigned the same channel once channels have been assigned to the BSSs Allocation of the channel time is achieved through dynamic bandwidth allocation, which enables sharing of the channel time among co-channel BSSs efficiently; no channel time is left idle Because there is no central controller, distributed prioritized dynamic bandwidth allocation algorithms in order to coordinate multi-BSS channel reuse.

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 4 Overview This is one of 5 independent submissions relating to HCF access and OBSS mitigation. These are: (1)CPMA: An Access Mechanism for Contention-Free Sessions  CPMA a protocol for prioritized contention-based access (2) An access mechanism for Periodic Contention-Free Sessions Regularly spaced sessions (3) ‘Shield’: Protecting High-Priority Channel Access Attempts Prevents CFS/PCFS corruption in case of collision with an (E)STA (4) ‘Neighborhood capture’ in wireless LANs Preventing a capture effect (5) HCF Access Mechanisms: Inter-BSS NAV protection Virtual carrier sense for CFS/PCFSs

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 5 Channel Access for CFSs A CFB is a special case of a CFS that is started by an HC Ideally we want: –CFSs/PCFSs to have priority access over (E)DCF transmissions –(E)DCF transmissions to access channel at assigned priority –CFSs to be able to regain control of the channel periodically conflict-free No conflicts with CFSs from other BSSs or (E)DCF transmissions –Efficient channel re-use (channel not left idle)  “Dynamic bandwidth allocation” As proposed by others, EDCF can be used by the CFSs to access the channel –The CFSs are placed in the highest priority class which is above the highest priority in EDCF –Shorter AIFS are used for CFS access, which helps avoid collisions with (E)DCF transmissions –Backoff helps deal with CFS conflicts among BSSs

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 6 Concerns with other proposals Concerns about random backoff: –With a random backoff, (E)STAs may access the channel before the HC as (PIFS + 1)=DIFS allows a (E)DCF station to transmit –A long backoff leaves many idle slots; (E)STAs would transmit before HCs –A short backoff causes collisions between CFSs Concerns about fixed/minimum re-visit time: –When restricting the time to the next access attempt, the channel cannot be used sooner, even if needed and available –While this is a good approach for PCFSs, it limits efficiency of channel re-use

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 7 CFS Channel Access Mechanism: Key features Cyclic Prioritized Multiple Access - CPMA This prioritized distributed medium access protocol consists of 3 features 1Fixed Deterministic Post-Backoff Reduces conflicts between APs 2Staggered Start-up (Optional) Contiguous sequences of CFSs to deter collisions with (E)STAs 3‘Pegging’ (Optional) Preserves CFS sequence contiguity

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 8 Assumptions Inter-BBS NAV protection We assume that there exists a mechanism for ‘busy’ channel detection (detection of the start and end of a CFS) ‘Ideal’ backoff conditions We assume fully overlapped BSSs or partially overlapping BSSs with IBNAV protection Otherwise, we assume ‘parallel’ backoff [see paper on “Neighborhood Capture”]

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 9 Fixed Deterministic Post-backoff Repeating in cycles, –Contention-free sessions (CFSs) are generated, one from each overlapping BSS Each active AP engages in fixed deterministic post-backoff –Post-backoff is ON –A fixed deterministic backoff delay, Bkoff, is used by all APs Bkoff > number of overlapping BSS Bkoff should large enough to enable the traffic that can be accommodated by the channel –The channel is accessed and the backoff timer is counted down using the shortest AIFS possible –A CFS is initiated when backoff expires and the backoff is reset to Bkoff; this starts a new cycle –A cycle is measured in terms of idle time slots; it does not represent a fixed time interval –(E)DCF transmissions are attempted by their assigned priority while the AP is counting its backoff down A new AP can get started and resolve possible collisions by a small random backoff

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 10 AP2 sets backoff = Bkoff CFS of AP1 PIFS CFS of AP2 AP1 sets backoff = Bkoff PIFS SIFS CFS of AP1 PIFS CFS of AP2 PIFS SIFS Backoff of AP1 expires & is reset Backoff of AP2 expires & is reset CFS sequence Cycle for AP1 [fixed number of idle time slots] Time for DCF transmissions Non-conflicting CFSs Future CFSs will not conflict, given a sequence of non-conflicting CFSs –Because their previous CFSs did not conflict, the follower AP’s backoff delay exceeds that of the leader’s by at least 1

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 11 Staggered Start-up - 1 In general, the procedure can get started by a random backoff, 0, 1, … [small value] To reduce the probability of collisions with (E)DCF transmissions, a contiguous sequence of CFBs (consecutive CFBs separated by idle gaps <= PIFS) is generated by observing the following startup procedure: If there is no other AP present –The first AP will get started after waiting for a cycle, which is the time it takes Bkoff idle time slots to expire (with deferral time) If one AP is operating and a second AP powers on –AP2 listens to the channel until a PIFS Idle following a busy channel, or for another indication of a CFS –Then it looks for the first idle longer than PIFS, and sets its post- backoff delay to transmit always right after AP1 An idle period X= PIFS + x, x>0, has been detected at time t, AP2’s backoff at time t is set at Bkoff - x If several APs power on during the same cycle –Collision between new APs is possible –It can be resolved by a random backoff, 0, 1, … [small value].

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 12 Staggered Start-up - 2 AP2 listens to the channel until a PIFS Idle following a busy channel Then it looks for the first idle longer than PIFS, and sets its post-backoff delay to transmit always right after AP1 –An idle period DIFS= PIFS + 1, has been detected at time t, AP2’s backoff at time t is set at Bkoff - 1 CFS of AP1 AP1 resets backoff = Bkoff CFS of AP1 Backoff of AP1 expires & is reset CFS of AP2 Backoff of AP2 expires & is reset AP2 sets backoff = Bkoff-1 PIFS SIFS PIFS SIFS CFS CFS sequence Cycle for AP1 [fixed number of idle time slots] Time for DCF transmissions PIFS SIFS Channel busy AP1 is operating AP2 is a newly installed HC

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 13 Staggered Start-up - 3 Future CFSs will be contiguous, given a sequence of contiguous CFSs –Because their previous CFSs were contiguous, the follower AP’s backoff delay exceeds that of the leader’s by exactly 1 – NAV protection and longer AIFS prevent DCF transmissions from conflicting with new CFSs AP2 sets backoff = Bkoff CFS of AP1 PIFS CFS of AP2 AP1 sets backoff = Bkoff PIFS SIFS CFS of AP1 PIFS CFS of AP2 PIFS SIFS Backoff of AP1 expires & is reset Backoff of AP2 expires & is reset CFS sequence Cycle for AP1 [fixed number of idle time slots] Time for DCF transmissions Non-conflicting contiguous CFSs

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 14 AP3 backoff AP1 backoff AP2 backoff AP2 AP3 AP1 Contention-free bursts EDCF Transmissions Top priority Lower priority Peg SIFS +1 Time AP2 transmits a peg when it has no data to transmit Contiguity through ‘Pegging’ Pegging maintains contiguity If an AP has no traffic, it will transmit a short packet - “peg”- and set its backoff=Bkoff No gaps of length DIFS+1 are left idle, and thus (E)DCF stations cannot seize the channel, until all APs have completed one CFS per cycle Example: Bkoff = 8

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 15 Collisions from ‘gaps’ in a sequence If an AP retires or does not use pegging, –‘Gaps’, Idle periods >PIFS, will occur –A collision between its follower AP and an EDCF station is possible Newly activated APs can help take up excess void (two retirements back to back) in the sequence CFSs can be protected from (E)DCF transmissions by using a ‘shield’ [See 01/570] AP3 backoff AP1 backoff AP2 backoff X Time AP2 retires, giving rise to a ‘gap’ AP3 beacon collides with an EDCF station

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 16 CFS of AP1 Collisions at head of sequence & rotation Given a contiguous sequence of CFSs (no gaps due to retirements), only the first AP can collide with (E)DCF stations –in lighter traffic (low probability!) –Subsequent APs do not collide with (E)DCF stations because of their shorter AIFS CFSs can be protected from (E)DCF transmissions by using a ‘shield’ [See ‘Shield’] If an AP experiences a collision with (E)DCF transmission –It may reset its backoff = Bkoff -x, thus moving from the head of the sequence behind another AP –Another AP will thus take the lead position, sharing the collision probability AP1 and AP2 are operating An ESTA transmits after a DIFS idle DCF Transmission DIFS PIFS CFS of AP2 DIFS CFB of HC1 PIFS AP1 sets backoff= Bkoff -1 Collision! Expiration of AP1’s backoff CFS retrial X

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 17 Summary of CFS Channel Access Mechanism Fixed deterministic post-backoff –Prevents collisions among the different APs Staggered ‘start up’ –Achieves contiguous CFS sequences –Contiguity decreases probability of collision with (E)DCF transmissions as no idle gaps are left the size of the AIFS of (E)DCF stations ‘Pegging’ –Maintains contiguity of CFS sequences If collisions occur, –A small random backoff is used to resolve conflicts. The probability of conflicts is still less than with simple random backoff Note: CPMA can be combined with the following feature: ‘Shield’ –Protects CFSs from collisions with (E)DCF transmissions

Submission doc.: IEEE /599r1 November 2001 M. Benveniste -- AT&T Labs, ResearchSlide 18 QoS Management PCFSs provide regular access to the channel for periodic traffic The use of PCFSs alone cannot not provide efficient dynamic bandwidth allocation CFSs generated on a contention-basis must complement PCFSs PCFSs and CFSs access the channel with the shortest AIFS QoS requirements are met by an AP scheduling traffic as follows: –Periodic traffic is transmitted in PCFSs –Non-periodic traffic is placed either in a PCFS or in its allotted CFSs according to traffic priority –Delay-sensitive traffic is scheduled first, followed by traffic of lower priorities