|
Very good points, to which I cannot help but add a
few more :-) We develop different protocols, different PHYs, different
MACs to address different application spaces (here a "space" is can be
thought of as a common set of requirements). It is important to keep in
mind that each type of system is solving a different problem.
When we talk about the impact one system has on
another, it is clear that the amount of time each spends transmitting affects
potential impact on the other, of course. What do we mean by
"impact"? This seems to be a fundamental term and I'm not sure we
are all working from the same definition. ...
Building on what was said at the teleconference, the important thing to
consider is the percentage decrease in the throughput capacity of the systems
being simulated. An optimal point to consider as a design goal is for
all systems to experience an equal percentage drop in throughput
capacity.
I might be mis-reading this, but I'm going to
disagree that "throughput capacity" is the most important performance metric.
Not every application space has throughput (bits per second) as the primary
performance criteria. For dot-11 and dot-16, yes. For other things which
may exist in our spectrum (like 802.15.4a) bps is really low on the list of
optimization priorities. These standards are intended for low bit rates and high
reliability, low cost, baterry life, etc. In some applications latency may
be more important than bps, and so on.
This goal can be incorporated in the protocol design of the systems in a
community if information related to this throughput metric is shared between
systems, and used to implement appropriate evasive action. What we have
here is essentially a distributed optimization problem, so it is important
that each system has the right information to make local decisions that
improve the common good.
This kid of "cooperative coexistence" is an
interesting distributed optimization problem. Even more challenging (and
interesting) is if when there is no ability to communicate between the
coexisting systems. As Harry points out, there are open loop
methods we can use (whcih can be either defensive or propactive) to make
decisions on how to best use the spectrum based on what we can sense. Our
current focus should be when/where/if there is a problem, not how to mitigate
it. But we should consider mitigation factors that are part of the respective
systems, whether it be protocol solutions (such as DAA) or convenient PHY
characteristics, likely proximity, etc. Not designing/specifying mitigation, but
certainly we should consider what may be there!
The channel occupancy metric is a blunt instrument for measuring
coexistence problems because different systems use different PHY technology.
This means that for each combination of system type, we could
potentially determine a different coexistence sensitivity to channel
occupancy. Even still, the metric is useful in helping us identify which
combinations are the most problematic if we do an exhaustive
search.
Blunt, yes. I wasn't thinking of occupancy as
a metric at all, but as a variable in the simulation that will affect
impact on other systems. I can build you a case where we have simultaneous
occupancy and still coexist pretty well (code separation in a 15.4a system, for
example, provides pretty good like-system coexistence without explicit
coordination). And other cases where one system will preclude all others when it
is on the air. So how much it is on the air may be significant. As I said,
throughput may not be the most relevant performance criteria, depending on the
appicaiton (and standard). "No
offered load" is not the same as zero throughput or zero band occupancy. I
read that as "no user payload bits" but inclusive of all the overhead
traffic. I think these are important inputs to the simulation.
With that said, Harry's
interpretation for using thoughput capacity as a performance
metric also makes sense for some "victem" systems, and would be very
useful.
Even some very simple simulation/analysis can be
very useful. It might be good to start simple, and build more
sophisiticated models as the need becomes clear. Just one guys opinion, of
course :-).
-Ben
What gets tricky is when we make adjustments to the protocol design and
implementation of systems. Any modification would require a complete
redo of our simulations, since the channel occupancy metric will have a new
impact on coexistence. For example, if a system disables certain tones
when it experiences high levels of interference, two systems can share the
same band by subdividing it in an FDD fashion. A similar modification
could take place in the time domain by synchronizing transmission
opportunities. Without this disabling, the systems might interfere.
In both cases, the channel occupancy has not changed.
Mind you, we could modify the metric to account for percentage of the
band occupied, but the real issue is not just how much of the band is occupied
by a system, but where in the band a system occupies the channel, and when it
is occupied, in relation to all neighboring systems.
To address your issue of zero throughput systems that impact neighbors,
the throughput capacity metric should be studied. A zero throughput
system will reduce the throughput capacity of neighbors. Its throughput
capacity will also be affected by its neighbors. We can therefore study
the phenomenon you mentioned with this metric, and use it as a tool to better
understand how well the community is getting along.
Harry
On Aug 6, 2008, at 7:36 AM, Kenneth Stanwood wrote:
I am still very concerned about the
insistence to remove the channel occupancy metric without an appropriate
replacement. As was discussed in Denver, it is very valuable to assess how a
system implementing one technology may block the use of all or a part of the
channel by systems implementing other
technologies.
In particular, it is important to see
how a technology blocks access by others when the first technology has
little or no actual data to send. As we are already aware, any
802.16/WiMAX based technology should be suspect due to the designed
operation of unmodified WiMAX systems. Modified WiMAX systems should
be required to prove they do not limit access by other technologies under
low demand situations within the WiMAX system. NextWave proposed the
channel occupancy metric in question because inclusion of that metric in our
simulations was key to understanding and modifying the WirelessMAN-UCP
protocol in section 6.4 of 802.16h so that it would not excessively block
access by similar channel bandwidth 802.11 systems when the 802.16 system
had low demand. It is not clear to me that the WirelessMAN-CBP
protocol described in section 15 of 802.16h has addressed this problem.
Lack of an appropriate metric could hinder the identification and
resolution of issues in currently proposed systems and systems likely to be
proposed in the future.
I understand the reluctance to accept
the metric. At NextWave we had heated internal debate over whether or
not to include the metric in our simulations, but when the results came in
and the value was overwhelmingly apparent argument
ceased.
At a minimum, if a throughput metric is
used instead, there must be a solid requirement that system A’s impact on
the throughput of system B must be simulated for a variety of cases where
system A has low or no demand.
Thanks,
Ken
Hi All,
Regarding the yesterday teleconf and
the way forward:
1. In my view metrics which do not have
a clear interpretation as coexistence criteria shall be omitted, and this is
the case of the Medium occupancy where two opposite target criteria were
proposed by Paul and Eldad (Paul – 50% of time in case of two collocated
systems – was no agreement on this, because is ignoring the antenna
separation, powers, modulation, coding, etc.; Eldad 100% or similar
occupancy). In addition it is not defined yet what the occupancy is, and we
spent 45min. just with a discussion on the different possibilities. This in
addition to the time spent in the meeting.
2. I (and probably many others)
appreciate the straightforward metrics and criteria proposed by Richard,
looking at the relative throughput degradation of the two systems as result
of interference. Similar degradation means acceptable
coexistence.
3. The hidden nodes are also important,
being the exact image of the “harmful interference”. The reception of the
signal is directly affected by the hidden nodes. The hidden node statistics
is “hidden” in the averaged throughput results, and this is why we need this
metrics in addition to the throughput.
4. I think that for the next teconf. we
need to invest the time in advancing the CA document itself. We need to
discuss the simulation results and agree on the CA
text.
5. We spent already more than one year
on this issue (Apr. 07 is the date of the first parameter document), in
meetings and bi-weekly teleconf. I hope that we will be able to finalize
this process asap. We have committed for a much shorter process and the
planned resources have gone. Probably the 802.19 guys have also other issues
to address.
Regards,
Mariana
Minutes
posted on the server,
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).
************************************************************************************
|