1. Aggregate Block-ACK definition
July 2010
doc.: IEEE /0788r2
July 2010
Aggregate Block-ACK definition
Date:
Authors:
Jochen Miroll
Jochen Miroll, Saarland University

2. July 2010
doc.: IEEE /0788r2
July 2010
Abstract
This presentation explains how a Leader-Based Protocol (LBP) that aggregates feedback within the same time slot should be incorporated into TGaa
Normative text will follow based on discussions
Revisions of this document
Rev. 1 provides additional examples for frame exchange sequences
Rev. 2 has errors on slide 15: Frame Exchange Sequences (5) fixed
Jochen Miroll
Jochen Miroll, Saarland University

3. Feedback aggregation in the same time slot
July 2010
Feedback aggregation in the same time slot
How to get rid of the dependency on n?
All receivers provide feedback, but this feedback from k≤n STAs is aggregated in a single time slot
overhead(k) = overhead(1)
Introduction of NACK
AP transmits a data frame
Then, AP asks for ACK/NACK
If STA i has received the data frame: it responds with an ACK
If STA j did not receive the data frame: it responds with NACK at the same time ?
AP1
ACK
STA 1
NACK
STA 2
STA 3
NACK
STA 4
Jochen Miroll

4. Leader-based feedback cancellation
July 2010
Leader-based feedback cancellation
Groupcast becomes unicast to leader
non-leaders transmit a negative ACK if a frame is lost after being asked to do so – introduce the question “did you receive the frame?”
Target: Larger groups (than 11aa MRG-BA should practically handle)
Overhead for asking about frame reception – same thing for 11aa MRG block-ACK request
In this presentation: Show how this overhead is reduced by data frame aggregation similar to 11aa MRG-BA
Leader selection: choose the „weakest“ receiver (as seen by the AP)
So this STA’s ACKs can be cancelled with very high probability
But: No error correction guarantees in this scheme
“More reliable”, as compared to Multicast
Jochen Miroll

5. Recap: Results in the LBP-worst-case
July 2010
Recap: Results in the LBP-worst-case
Worst case results
where leader-selection would not be able to reliably determine SNR difference between receivers
Ns2 – scenario:
Rayleigh fading channel, equal AP-STAs distance
feedback cancellation rate is about 76% for 2,
more than 90% for more than 2, and
already 99% for 5 receivers
Measured: 3 rcvrs, 1 leader
~89% feedback cancellation success
Measurement parameter
Rate
Software ACK loss avg.
Software NACK loss avg.
Hardware ACK-6 loss avg.
Hardware ACK-12 loss avg.
Jochen Miroll

6. Recap: Hybrid LBP (HLBP)* cf. doc.: IEEE 802.11-09/0290r1
July 2010
Recap: Hybrid LBP (HLBP)* cf. doc.: IEEE /0290r1
Phase I Transmit a block of frames, as in MRG BA. Here: systematic FEC part
Phase II Parity phase. Instead of BAR/BA, do AggregateAckRequest/AggregateAck
* Assume e.g. DVB-IPDC or Raptor code on upper layer, MAC somehow knows which packets are systematic (DATA) or parity
Jochen Miroll

7. Motivation for (H)LBP with aggregation through cancellation
July 2010
Motivation for (H)LBP with aggregation through cancellation
Due to aggregation: Scales with the number of receivers
Only the number of retransmissions may increase with increasing group size, but not the protocol overhead
Due to additional FEC: Degraded channel at one STA does less harm to overall performance
both in terms of errors and delay
Hybrid LBP FEC
Enable cross-layer error correction through systematic packet level FEC controlled by upper layers (e.g. the application or a transport protocol other than TCP/UDP)
The remaining MAC mechanism is simple and easily implementable
Predictable delay in error correction
due to aggregation of feedback in the same time slot
application layer is able to control the error correction delay
Jochen Miroll

8. Motivation for (H)LBP with aggregation through cancellation cont‘d
July 2010
Motivation for (H)LBP with aggregation through cancellation cont‘d
About the error floor of feedback cancellation:
If leader is the weakest receiver, it probably looses packets at higher probability than the rest
If the weakest receiver is significantly weaker: cancellation success increases and the error floor becomes negligible (now also due to the capture effect!)
With increasing block length, independent errors at all stations within block become more likely
How to choose the leader?
Incorporate aggregate Block-ACK in a way that AP can switch from ABA to BA without overhead. MRG BA feedback used to measure signal strength (or any other traffic from resp. STAs)
Leader selection has further benefits
Rate adaptation: for the leader (since it’s the weakest STA in the group)
If AP detects that the stream rate is higher than what is served to the current leader, it can just select another STA in the group as the leader for the next block
Jochen Miroll

9. Aggregate Block-Ack Request
July 2010
Aggregate Block-Ack Request
Define an aggregate-BAR: “ABAR”
MRG BAR information field: (Figure 7-13aa)
Use reserved bit to indicate the request for aggregate feedback to MRG group members
Bitmap Offset becomes ABAR leader indicator
Bitmap preceded by ABAR minimum
Bits: 8 1 7
Variable
MRG BAR Information Length
reserved
MRG BAR Bitmap Offset
MRG BAR Partial Bitmap
Bits: 8 1 7 8
Variable
MRG BAR Information Length
ABAR
Flag
(set to 1)
MRG BAR Bitmap Offset
(ABAR leader)
ABAR minimum
MRG BAR Partial Bitmap
Jochen Miroll

10. Aggregate Block-Ack Request Minimum
July 2010
Aggregate Block-Ack Request Minimum
Define an aggregate-BAR minimum (m): “m is the minimum number of frames a station has to have received within the last k frames, such that the upper layers can decode the current FEC block”
The AP may, after transmitting k frames, ask the group members to (Negatively-)AggregateBlockACK m < k frames
Jochen Miroll

11. Frame exchange sequences (1)
July 2010
Frame exchange sequences (1)
„Did you get 5 since BAR start seq#?“
BAR start seq#
Aggregate BA is cancelled
frame exc. end … D D D D D
ABAR P P D
AP1 D D D D D
ABA P
Leader D D D
N-ABA P P
STA 2 D D D D D P
STA 3 D D D D
N-ABA P P
STA 4 D D D D D P P D
Packets avaliable in AP buffer for currently serviced MRG group:
Jochen Miroll

12. Frame exchange sequences (2)
July 2010
Frame exchange sequences (2)
„Did you get 5 since BAR start seq#?“
(note that 6 have been sent)
„Did you get 5 since BAR start seq#?“
BAR start seq#
N-ABA is received
frame exc. end D D D D D P
ABAR P
ABAR
AP1 D D D D P
ABA
Leader D D D P
N-ABA P
STA 2 D D D D D P
STA 3 D D D D P P
N-AB
STA 4
AP can decide whether to add redundancy/parity a-priori
Reception of N-ABA and lack of ABA are treated as the same case
Jochen Miroll

13. Frame exchange sequences (3)
July 2010
Frame exchange sequences (3)
„Did you get 6 since BAR start seq#?“
sent: 7
Would ask „got 5 now?“
sent: 6 = 5 data + 1 parity
„Did you get 5 since BAR start seq#?“
interrupted by beacon
BAR start seq#
Aggregate BA
is cancelled
frame exc. end D D D D D
ABAR P
beacon P
ABAR
AP1 D D D D D
ABA P
ABA
Leader D D D
N-ABA P P
STA 2 D D D D D
Assume everyone received the beacon
Increases our seq# counter by one P
STA 3 D D D D
N-ABA P P
STA 4
Jochen Miroll

14. Frame exchange sequences (4)
July 2010
Frame exchange sequences (4)
„Did you get 5 since BAR start seq#?“
BAR start seq#
frame exc. end D D D D D
ABAR P
ABAR P
ABAR
AP1 D D D P
Leader D D D
N-ABA
N-ABA P
N-ABA
STA 2 D D D
N-ABA P
N-ABA
N-ABA
STA 3 D D
N-ABA P
N-ABA P
N-ABA
N-AB
N-AB
N-AB
STA 4
AP is limited in parity delivery as provided by the application layer
Application layer can implicitly control MAC error correction delay
Jochen Miroll

15. Frame exchange sequences (5)
July 2010
Frame exchange sequences (5)
„Did you get 6 since BAR start seq#?“
„data“ packet carries parity
„Did you get 6 since BAR start seq#?“
BAR start seq#
frame exc. end D D D D D P
ABAR P
ABAR
AP1 D D D D P
ABA
Leader D D D P
N-ABA P
STA 2 D D D D D
N-ABA P
STA 3 D D D D P
N-ABA P
N-AB
STA 4
Application layer can enforce a-priori packet FEC
Jochen Miroll

16. July 2010
Conclusion
Reception of N-ABA and lack of ABA are treated as the same case
It is not always necessary that feedback from the leader is cancelled
If packets are (somehow) marked or the AP (somehow) determines that IP packets carry systematic FEC data or respectivly parity and it has some of them buffered
Application layer block coding and MAC Block-ACK can work more effectively together
AP can decide whether to add redundancy/parity a-priori
How this is implemented can be left to the manufacturer
Marking packets needs additional mechanisms, but this is out of scope of 11aa
If application layer is not satisfied by MAC layer performance
it can conceal parity as data (purposely false marking) to enforce packet FEC
Jochen Miroll

17. Questions and discussion
July 2010
Questions and discussion
Jochen Miroll