| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Marianna,
When Eldad did some studies of Bluetooth interference into 802.11n it showed that the effect of the narrowband interference is significant so it would expect that when 802.16h transmitted the effect would be similar.
Steve
From: Mariana
Goldhamer [mailto:marianna.goldhammer@alvarion.com]
Paul, All,
1. The coding effects are ignored in our simulations; however they can make a big difference. For example, you say: There is no concept of 802.11y using the subchannels not used by 802.16h, so even if the 802.16 system is using the minimum, the UL is still occupied from the point of view of an 802.11 system near enough to the 802.16 subscriber to hear it. The introduction of the modifications will report an erroneously low and incomplete measure of occupancy.
However there is no evidence than an 802.11 system cannot overcome the presence of an 802.16 subscriber, because the 802.16 PUSC or AMC sub-carrier partitions concentrate the energy such that only a small number of the 802.11 sub-carriers will be affected. The worst case is PUSC, with 6 tiles. If 802.16 will use 10MHz channels, the number of the relevant tiles will be reduced to 3. There is a big chance that the coding will repair the affected 802.11 sub-carriers. Do not forget that we never place different SS or STA in each-other proximity, being allowed some separation. Also the statistical distribution leads to separation, which also contributes to the interference mitigation. Your example may not be relevant at all for 802.19 scenarios.
2. The BS antenna arrangement, as a measure of coordination between operators (I speak about regular 3.65GHz deployment) can introduce additional separation which will mitigate the interference. 802.19 simulations are based on a worst case which may not exist in real deployments. You can definitely have the situation that the channel is used 100% of time by each network and there is no significant interference between geographically co-located networks. Alvarion has the experience and the solution for real deployments, with co-located scheduled and 802.11-based networks, where the interference was resolved by proper antennae arrangements.
With the antenna arrangement approach, the best case is to have 100% time occupancy, which leads to max. throughput, while you try to convince us that the best case is to use the cannel only a fraction of time.
3. The metrics of channel occupancy has no meaning, as you cannot say which situation is the “good” one.
Mariana From: Mariana
Goldhamer
All,
It just says that if 10 OFDMA stations are occupying a channel in the same millisecond, the channel is occupied by that network for 1 ms.
2. The existing text says that the occupancy “It is calculated for each network by summing up the total transmission time of any station in that network and dividing by the total simulation time.”
This means that if 10 OFDMA stations are occupying the channel in the same millisecond, the equivalent network occupancy time is 10ms, which is obviously not true.
A funny observation: if you apply the existing definition, you can get that an 802.16 system occupies more than the total simulation time (assume it 1s). Based on the 40% up-link occupancy, using the media every second frame, you get time occupancy of 2.3s: 200*(0.4*10 +0.6)*5/2 = 2.3s, where the frame length is 5ms, you have 200 frames in the simulation time and the BS transmission takes 60%. So the simulation time is 1s, but an OFDMA system occupies the media for 2.3s, even if using only ½ of the simulation time!
I do not understand what the problem is with the proposed text, which corrects an evident technical aberration.
4. Steve, I have a question: was the criterion of time occupancy used before 3.65GHz? If yes, where exactly?
Regards,
Mariana
From: Richard Roy
[mailto:dickroy@alum.mit.edu]
From: Kenneth Stanwood
[mailto:KStanwood@nextwave.com]
Please look at the channelization rules for 802.11y and the suggested channelization for 802.16h for the 3.65GHz band – the band of interest here. There is an attempt by numerous people in the community to make the channelization similar. You’re correct that systems that try to use the entire band will cause fairness issues.
[RR] Agreed, and the reason that they are doing this as I see it is because they know that if the channelization schemes are not similar, someone will come out on the short end of the stick. From a game theoretic standpoint, at this juncture of the game, the criterion is min-max, that is, minimize your maximum expected loss (in efficiency) and that is achieved in this case when all systems are “the same”, hence the desire to converge to a common channelization scheme. There is probably a little “it’s just too hard to analyze and predict performance otherwise” thinking going on as well, but that’s only natural. The real shame is that the reason the FCC set downs these CBP rules was something of an attempt to let systems work it out in the marketplace, i.e., let the best system survive. Now we have people trying to harmonize systems, sacrificing optimality for the sake of I’m not sure what. In any case, with reference to the discussion below, instead of having the most efficient communication system win, the result will be some suboptimal compromise. Not sure this was what the FCC had in mind, but if that’s the way it has to be, c’est la vie.
Then think about fairness from the point of view of two or more operators who are geographically co-located and have paid the same for their “nationwide non-exclusive” licenses regardless of which technology they have chosen. Each has legal recourse is they feel they are not given sufficient access to the spectrum, either in frequency or time.
[RR] I can just imagine in a court of law an operator complaining to a jury that his opponent was cheating. What would he show to prove his case that two years ago he was “denied sufficient access”? I can just see it … system log files with thousands of channel busy entries, accompanied by reams of spectrum analyzer outputs from some HP spectrum analyzer … the jury … sound asleep. Interesting to think about, but I know the lawyers will get paid regardless.:^)))
Yes, they can decide they made a bad technology choice and change technologies, but if this were taken to the limit, everyone would eventually end up with the same technology with the same channelization and we’re back to channel occupancy being a very valid metric.
[RR] The point you are making is a good one. Even a posteriori, a stable set point from a game theoretic viewpoint is one where all operators deploy the same technology/system. Then each knows exactly what the other is doing … and, by the way, I believe now transport layer throughput is a very valid metric as well (and I’d suggest perhaps a lot easier to measure) since the system efficiencies are identical and 1/n is the “optimal” ratio. As long as the market is allowed to seek out these optimal set points, I believe you are right and convergence to a common system will occur.
I think I’m over a dime by now …
RR
Ken
From: Richard Roy
[mailto:dickroy@alum.mit.edu]
Hi Ken,
I’m going to assume that you’ll read subsequent messages in this thread, so I won’t repeat what’s therein. My reason for chiming in was to provide an alternative way of thinking about the issue of fairness, since Marianne (and I presume others) had raised and were raising issues related to the definition of a “channel”. My thought was a different perspective might help to resolve the issue. It has clearly sparked some interesting exchanges if nothing else.
More below …
From: Kenneth Stanwood
[mailto:KStanwood@nextwave.com]
The problem with not counting channel occupancy is that we then revert to viewing fair access to the network in terms of throughput. Throughput as a fairness indicator, while important, doesn’t show the entire story because you have differences in efficiencies between two technologies.
[RR] This is where it gets interesting. At some point, either explicitly or implicitly, a choice of definition of “fairness” must be made. A question such as: If system A is 10 times more efficient than system B, should it get 10 times the resources or should the resources be divided evenly (or some other ratio)? From everything I have read to date on this thread, it seems the group is thinking it should be divided evenly. I’d argue from the perspective of maximizing utility while still providing opportunity that a weighted division is “optimal”. The weighted division is expressed in the previous postings. That having been said, my guess is that the efficiency discrepancies in systems today are not nearly as large as a factor of 10, and are more likely in the 0%-100% range. It’s hard to imagine a system with less than half the efficiency of another lasting very long in the marketplace after all. In this case, the weighted allocations could be reasonably approximated by 1/N and now you’re almost back to “fairness” as “equal time on the channel”. There is a difference however … continued below.
Neither 802.16 nor 802.11 should need to know how efficiently the other system uses the medium and adjust algorithms based on that. It’s simply too complex and constantly being upgraded. Channel occupancy works very well as the first fairness metric because it totally eliminates the discrepancies, whether PHY or MAC, between technologies.
[RR] I would argue that channel occupancy as defined in the postings is no simpler. To paraphrase your first sentence: “Neither 802.16 nor 802.11 should need to know how the other system defines a “channel” and how efficiently it’s MAC and PHY use the resources available, and adjust its algorithms based on that. It’s simply too complex and, in many of the newer technologies, constantly changing in real time … on the fly (cf., SDMA, OFDMA, etc.)” I believe this was what Marianne was trying to address in her posting.
As a simple example of how channel occupancy can be problematic, just look at the coexistence issue in 11n with the 40MHz channels and the legacy 20’s. Consider a single 40MHz band and system A (a 40 MHz channelized “11n” system) and system B (a 2 x 20MHz channelized “11b” system). Since in real systems the two 20MHz channels are not orthogonal, just independent (cf., adjacent channel power leakage in real devices), it is generally not possible for a 20MHz system to operate simultaneously on adjacent channels at the same AP. So it operates on one at time, and when it does so, on either channel., system A is denied use of its only channel. The equal channel occupancy argument is tantamount to agreeing that it is “fair” that system B should be able to deny system A access half the time when using only half the resources system A would be able to use. That is, it is fair for system B to reduce the overall spectrum efficiency by 25% (assuming A and B have the same bits/sec/Hz/km^3 throughput rates). Stated another way, it is “fair” for system B to force half the resources to lay fallow while it operates at half the efficiency of system A. Most economists would probably suggest this is not “fair” use of a common resource for the common good. Note that if, based on this or a similar argument, the thinking is changed to “weighted” use of the resources, then the same argument applies to the network and transport efficiencies and we get to weighted throughput at the transport layer as a reasonable definition of “fairness”. It’s really all about changing the way one thinks about “fairness”.
Two systems operating co-channel should each expect a goal of getting half the band width, and with current technologies that means half the time.
[RR] First I assume you meant “Two systems operating co-channel should each expect a goal of getting half the band width all the time, and with current technologies that means all the bandwidth for half the time.” Marianne was pointing out that with current technologies, this is unlikely to happen and that there are alternative more complex measures that could be considered (e.g., time weighted by amount of spectrum actually occupied for one). However, I interpret from your statement that you have in mind that there are similarities in how technologies define a “channel”. I would like to suggest that this is precisely where the maximum differences in systems are likely to occur, and it is these differences that Marianne and others were trying to address in their postings. For two systems to be operating “co-channel”, assuming by co-channel you mean operating on the same channel, they must have the same definition of “channel”. This is almost never the case for different systems, unless by channel you mean ALL the resources available, i.e., the entire band allocated to the service over all time (for the duration of the license or an unlicensed equivalent) and space (in a given geographical area). (And when you include smart antenna capabilities, NO two channels are the “same”, ever, because the location of the antenna elements matters!)
Cheers,
RR
What they do with that bandwidth is up to the designers of the respective technologies.
Regards, Ken
From: Richard Roy
[mailto:dickroy@alum.mit.edu]
I don’t understand why “the time a network occupies a medium” is as important to a concept of fairness as suggested in the document. Also, I’m not quite sure why “This is the total time in which a given network is transmitting and those transmissions are successfully received at the destination divided by the total simulation time.” isn’t simply “This is the total length (in time unit) of the successfully received frames divided by the total simulation time.” I can’t imagine successfully receiving a frame that was never transmitted.
The more important question is how fairness is defined. If you try to get too detailed, defining it at the PHY level as some measure of “time on the channel”, then disparate PHYs like .11 and .16 present all these “yucky” details. (It gets incredibly more complex if you really want to do it right and include the spatial dimension in the definition of a channel … which you really should since both 16 and 11 have smart antenna technology at their core!)
If, on the other hand, one considers fairness as a property of a collection of systems at the transport layer and higher, one can avoid all these issues and get right to the point. I’d suggest that fairness could be thought of (and ultimately defined precisely in mathematical terms) as follows:
Given a frequency band that is to be shared in a given geographical area, and given a set of systems that are desiring to share that band in that area, and given the maximum attainable throughput (at the transport layer) of useful user data for each system operating alone in that band in that area (in bits/sec/Hz/km^2(or 3)), then the systems are said to be “playing fair” if, when all systems are operating in the same area in the same band at the same time at “full load” (that is, adding more load does not increase throughput), then the amount of successful throughput for each system is in the same proportion as their maximum successful throughputs when operating alone. This avoids all the PHY issues of “time transmitting”, “amount of band used by an OFDMA system”, “how the spatial dimensions are or are not used by each MAC & PHY”, etc..
My two cents …
RR
From: Shellhammer,
Steve [mailto:sshellha@qualcomm.com]
Paul,
Is it correct to say then that you would prefer to reject the outstanding comments regarding the Medium Occupancy Metric and keep the text the way it is?
Regards, Steve
From: Paul Piggin
[mailto:ppiggin@nextwave.com]
These are our comments on the proposed changes to the Medium Occupancy Coexistence Metric as detailed in the attached contribution:
- Medium Occupancy is an important metric for determining fair sharing of the channel and of high importance in the 3.65-3.7GHz band. - The idea behind the introduction of the Medium Occupancy Metric was to abstract the underlying radio access technology. This proposal is a step backwards and moves away from this intention. - Because Medium Occupancy abstracts radio access technology discussion of TDMA and OFDMA is not valid. - Medium Occupancy is not intended to measure efficiency – but coexistence fairness. - For 802.16h the system either has the frame or not. When a system has the medium, be it 802.16 or 8021.11, the frame occupancy is 1 - not some fraction based on the number of devices transmitting within a system. The metric is Medium Occupancy – medium meaning the whole channel. - There is no concept of 802.11y using the subchannels not used by 802.16h, so even if the 802.16 system is using the minimum, the UL is still occupied from the point of view of an 802.11 system near enough to the 802.16 subscriber to hear it. The introduction of the modifications will report an erroneously low and incomplete measure of occupancy.
From these points the suggested changes are not acceptable.
Kind regards, Paul.
From: Shellhammer,
Steve [mailto:sshellha@qualcomm.com]
Here is a document from Marianna.
Steve
From: Mariana
Goldhamer [mailto:marianna.goldhammer@alvarion.com]
Hi Steve,
Please consider the attached contribution as response to my A.I.
Regards,
Mariana
From: Shellhammer,
Steve [mailto:sshellha@qualcomm.com]
The Minutes and the updated comment resolution database have been uploaded.
https://mentor.ieee.org/802.19/file/08/19-08-0002-09-0000-conference-call-minutes.doc
Steve
************************************************************************************ This footnote confirms that this email message has been scanned by PineApp Mail-SeCure for the presence of malicious code, vandals & computer viruses(190). ************************************************************************************ ************************************************************************************ This footnote confirms that this email message has been scanned by PineApp Mail-SeCure for the presence of malicious code, vandals & computer viruses(42). ************************************************************************************ ************************************************************************************ This footnote confirms that this email message has been scanned by PineApp Mail-SeCure for the presence of malicious code, vandals & computer viruses(43). ************************************************************************************ ************************************************************************************ This footnote confirms that this email message has been scanned by PineApp Mail-SeCure for the presence of malicious code, vandals & computer viruses(190). ************************************************************************************ ************************************************************************************ This footnote confirms that this email message has been scanned by PineApp Mail-SeCure for the presence of malicious code, vandals & computer viruses(42). ************************************************************************************ ************************************************************************************ This footnote confirms that this email message has been scanned by PineApp Mail-SeCure for the presence of malicious code, vandals & computer viruses(42). ************************************************************************************ |