| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
Hi Dibakar, Thanks for your insight. In general, latency improvement translates to throughput improvement. In case of non-STR, as you yourself point out and we note in our email, chances of getting access to both the links simultaneously would be small. It is clear that for a no-congestion scenario like a single AP connected to a single non-AP STA and with unidirectional traffic, non-STR ML throughput would be higher than that of TDM ML. Also, the fact that the throughput using this principle is close to that of conventional ML, is also demonstrated in the contribution 562r2 authored by you, Minyoung, et al. In your second point, you have suggested allowing a CTS transmission to impair in a limited manner an ongoing reception on another link (given this is a non-STR non-AP STA). So, you would like a non-STR device to behave as an STR device at the expense of fidelity of certain transmissions. There are many knobs and decision points that are necessary to make such a scheme work in an intended manner. For example,
So, in our view, the most straightforward mode of operation is to have a clear demarcation of non-AP STAs: STR and non-STR; and to ensure that the latter does not transmit and receive at the same time. Regards,
From: Das, Dibakar [mailto:dibakar.das@xxxxxxxxx]
In general, it may be good to clarify that the performance you mention below is mostly for latency. For throughput there could be low congestion scenarios where non-STR is clearly beneficial compared to having data exchange on just one link at a time. It could be argued though whether such low congestion scenarios are rare or not. “If the AP uses RTS/CTS (which is the default configuration in most cases): non-STR ML will follow the same procedure as TDM ML. This is because, if there is any difference between channel access times on the two links, RTS/CTS cannot be exchanged with a non-STR non-AP STA on the link which gains access later.” è This is not true for throughput. It is possible that the STA in second link can send CTS while Rxing a PPDU on link 1. Since the CTS packet is short, the impact is typically not high (i.e., only few MPDUs lost). Please refer to doc: 106-r4.
From: Sindhu Verma <000011381223f2e2-dmarc-request@xxxxxxxxxxxxxxxxx>
Dear All, This email is to start a discussion on the “TDM Multilink operation” scheme as proposed in 0659r1. The basic principle of this scheme is very similar to the “Enhanced Multilink single radio operation” scheme proposed in 0562r2. The common principle in these contributions is listen on one or more links but Tx/Rx of data on only 1 link at a time. Different radio + baseband architectures can be used to achieve this principle and the specific architecture is not integral to the principle. Following are the advantages of this principle which we call TDM ML: TDM ML at a non-AP STA provides the same or similar performance as conventional non-STR ML, but at a lower cost and lower complexity at the non-AP STA. The complexity at an AP to support a TDM ML STA is the same as what is required to support a conventional non-STR non-AP STA. STR ML performance can be better than TDM ML. However, STR ML is not expected to be implemented in R1 for any non-AP STA in 5G/6G due to the cost and complexity involved. Given the above, TDM ML provides a more feasible and cost effective means at a non-AP STA to achieve non-STR ML performance in 5G/6G. The equivalence of performance between TDM ML and non-STR ML at a non-AP STA can be explained intuitively as follows: Untriggered UL: UL for non-STR ML will follow the same procedure as UL for TDM ML, except when a non-AP gains channel access on both links at exactly the same time. The chances of the latter happening are statistically miniscule. Triggered UL: UL for non-STR ML will follow the same procedure as UL for TDM ML, except when the AP gains channel access on both links at exactly the same time. The chances of the latter happening are statistically miniscule. This is because the triggered UL transmissions must start at the same time, as once a non-STR non-AP starts transmitting on one link, it cannot receive DL data or a DL trigger or perform CCA for untriggered UL on the other link. DL: The difference between non-STR ML and TDM ML will arise only when the AP wins a second link to transmit to a non-STR STA while it has an ongoing transmission to the same STA on another link. In this case: If the AP uses RTS/CTS (which is the default configuration in most cases): non-STR ML will follow the same procedure as TDM ML. This is because, if there is any difference between channel access times on the two links, RTS/CTS cannot be exchanged with a non-STR non-AP STA on the link which gains access later. If the AP does not use RTS/CTS (as is common mostly for short transmit durations): The AP in this case can start transmitting in parallel at different times on both links to a non-STR ML STA, but has to ensure that the AMPDUs on both links have the same end time. The short duration of AMPDUs for high bandwidths + the restriction of making the AMPDU on the second link shorter due to end alignment with the AMPDU on the first link, leads to insignificant gain for non-STR ML. Simulations presented in 0659r1 (slides 11 and 12) also show this equivalence. Please let us know your comments. Regards,
To unsubscribe from the STDS-802-11-TGBE list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-11-TGBE&A=1 To unsubscribe from the STDS-802-11-TGBE list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-11-TGBE&A=1 To unsubscribe from the STDS-802-11-TGBE list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-11-TGBE&A=1 |