Dear all,
Thank you for the feedback on doc 11-23/2027r2. I am starting this thread to re-summarize the answers to the questions raised (to the best of my recollection), and also to solicit further comments from people
in the queue (Sindhu, George etc.). At a high level, I would like to re-iterate that there are
two separate problems identified in the proposal and correspondingly two separate solutions for them are presented.
Problem 1/Solution 1: Allowing an AP to selectively solicit a response to the Sub-band switch (SBS) initial control frame (ICF) from only a subset of DSO STAs can reduce the required padding delay to be
provided by the AP. Note that any legacy STAs addressed in the ICF still respond as per baseline.
Problem 2/Solution 2: To enable FCS check of SBS ICF before switching, the proposed solution is that DSO STAs are not expected to respond to the SBS ICF and shall perform the switch after the end of the
SBS ICF reception. The AP can use a follow up frame to protect the TXOP for the duration of the switch. In this solution legacy STAs are not addressed in the SBS ICF (unless they jointly span the full TXOP bandwidth).
Comments and their responses during the call:
Maolin: For problem 1, do the non-DSO STAs served after the SBS IC frame need to be (jointly) capable of receiving on the full TXOP bandwidth? Can padding not be used on some of the RUs where non-DSO STAs
are not present?
Ans: Yes, the non-DSO STAs should be capable of transmitting on the full TXOP bandwidth in case an ACK is solicited from them. This is because we would want the ACK to occupy the full TXOP channel width.
If a delayed ACK is used, then we should be able to use the mechanism suggested by Maolin, i.e., use padding on some RUs where non-DSO STAs are not capable of reception (this is depicted in slide 13). Please also note that these images are just some examples
on how the mechanism of “pre-allocation” of sub-band can be leveraged by an AP. There could indeed be other ways.
Minyoung: The padding delay in commercial 11be EMLSR products being shipped is not as high as
s. For DSO switch the padding delay required may be even smaller since the switch is over smaller frequency. So, for
problem 1, trying to minimize the required padding may not provide the level of gains indicated in the example.
Ans: The architecture for DSO can be implementation dependent. In EMLSR there is a weak radio that is already capable of listening on the indicated channel, correspondingly there are some RF components
already tuned to reception on that channel. So, the settling time for these hardware components do not have to be considered. In contrast, for DSO this may not be the case since they switch to a new subband (where there was no weak radio listening). So commercial
EMLSR products needing a smaller EMLSR padding delay doesn’t necessarily imply DSO would require a shorter delay than that. At least for 802.11ax chips and 2.4GHz Bluetooth chips that we have tested, we found the time required for channel switch to be completed
can be longer than
s. In any case, I shall check up on the commercial 11be product EMLSR padding delays.
Yunbo: We have observed that the padding delay for EMLSR products can be longer than
s so the padding overhead can indeed be significant and need to be optimized. In slide 10 (problem 1) lower figure,
why is separate initial control frame sent to STAs 1-2, 3-4 and 5-6? Why not send a single ICF to all?
Ans: Indeed, sending a common ICF to all STAs (independent of whether they are served immediately or later in the TXOP) is the better approach, as shown in Alternative 1. Just to consider all possible cases
here we have also included a comparison to Alternative 2. Please also note that for using Alternative 1, the AP should still provide the indication in the ICF whether a STA is served immediately or later within the TXOP since this will impact the STA behavior
of switching back to primary channel.
Laurent: For solution 1 (slide 7), if you have non-DSO STAs that can already span the full TXOP bandwidth there is no need for performing DSO switch at all. Also, if you have no non-DSO STAs that span the
full TXOP bandwidth then the proposed solution 1 is not useful since the padding is still required. We have also proposed an alternative solution for the FCS check issue (11-23/2003) which should obviate the need for both solutions proposed here.
Ans: Regarding the first point, the BSS having some STAs that can receive on full TXOP bandwidth doesn’t make the DSO feature unnecessary. For example, the number of DSO devices in the BSS may be much larger
than the number of full bandwidth non-DSO STAs. Also, the AP may have limited buffered traffic to send to full-bandwidth non-DSO STAs. So even if there are a few full bandwidth non-DSO STAs who are served within the TXOP, for proper utilization of the TXOP
resources the AP may still want the DSO devices to perform sub-band switch. Regarding second point, even if there are no non-DSO STAs to be served at all, the ability to pre-allocate the sub-bands to DSO STAs and only solicit a response from the DSO STAs served
first is beneficial. In this case the padding corresponding to only the DSO STAs served first needs to be provided. The gains for this case are exemplified in slides 9 and 10. For the third point, I don’t see how FCS check solution proposed in 11-23/2003 resolves
problem1 which is about reducing the required padding to be used in SBS ICF. 11-23/2003 indeed provides an alternative solution to problem 2, and I agree we should weigh the benefits of solution 2 with the one in 11-23/2003 to decide the approach to pursue.
This however doesn’t impact problem 1/solution 1.
Morteza: For solution 1, allowing an AP to schedule some DSO STAs first and some DSO STAs later in a TXOP based on their padding delay will cause unfairness issues. Also, if ACK is not transmitted by some
of the STAs how will the AP know whether the corresponding STAs have received the SBS ICF and have switched or not? Finally, there may be many complicated error situations to handle if all STAs are not expected to behave the same way to SBS ICF.
Ans: Regarding first point, scheduling decisions by an AP are implementation dependent and are outside the scope of the spec. Also, I don’t think scheduling one STA earlier in a TXOP and one later within
the same TXOP causes fairness issues. Also by that line of argument, many existing mechanisms can be considered unfair. The proposed solution is to provide the AP the flexibility to optimize its scheduling, the exact implementation of the scheduler is up to
the AP vendor. Regarding second point, if the SBS IC is an MU-RTS frame, the CTS is sent by all the addressed STAs in parallel. So, the AP can’t know who sent the CTS even now as per baseline. In fact, allowing only the STA which is served first to respond
within each sub-band (solution 1) is better than baseline solution since AP can be at least aware of whether the STA to be served first has sent a CTS. Regarding point 3, I don’t see how the error handling becomes more complex. Whether all STAs are required
to respond to ICF with a CTS or only some STAs are required to respond have identical behavior from the AP side since AP doesn’t know who sends the CTS. The situation may be different if BSRP is used as the SBS ICF, but that is only one case, and it suffers
from other issues (I plan to present another contribution on this soon).
Gaurang: For solution 1, when would the DSO STAs who are indicated to be addressed later in the TXOP switch back? They may not know how long to wait before switching back. In our previous proposal we proposed
a behavior to be followed by DSO STAs after switching.
Ans: A DSO STA addressed in an SBS ICF can behave differently depending on whether it has been indicated to be served immediately or later within the TXOP. A DSO STA which has been indicated to be served
later within the TXOP can wait till the end of the TXOP to switch back. There can be exceptions of course that cause earlier switching back, but these are not specific to this proposed solution.
Sindhu: I agree with some of the previous questions asked. Also, the proposed solutions in this document are not generic and, for example, cannot be used to serve 11be EMLSR devices using the same initial
control frame. They cannot be, for example, decoded only on the primary 20MHz channel. Also, [written in chat] if BSRP is chosen as the ICF for DSO then the AP can identify who has responded to the ICF.
Ans: Proposed solution 1 in no way prevents simultaneous triggering of DSO STAs, legacy STAs and EMLSR STAs by the SBS ICF. Please also note that the ICF is still sent in duplicate non-HT PPDU format. All
solution 1 suggests is to provide a STA-specific indication (e.g., in User Info field) in the ICF about whether a STA is served immediately and so if it is expected to respond to the ICF. This indication is only applicable to DSO STAs and legacy STAs may behave
as per their regular operation (i.e., they respond). With regard to solution 2, I agree that not soliciting any response to the SBS ICF may not be compatible with pre-11bn STAs. In this case, the first SBS ICF may be addressed only to DSO STAs (and any new
UHR STAs that require channel switch), and the follow-up frame (that reserves the medium for the channel switch) can be the trigger frame meant to trigger the pre-11bn STAs. I do acknowledge that the pre-FCS proposal in 11-23/1873 and 11-23/2003 is a valid
alternative to solution 2, and we should indeed discuss either approach to determine the best one to pursue. However, this does not obviate the need for solution 1. Regarding the use of BSRP as the preferred ICF for DSO, I think the group still needs to debate
on this. I do plan to share more detailed thoughts on this in another contribution.
Regards,
Vishnu
To unsubscribe from the STDS-802-11-TGBN list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-11-TGBN&A=1