I will attending Cambium training next week. I am taking notes from this discussion.
Jaime Solorza On Jun 9, 2015 8:07 PM, "Dan Sullivan" <[email protected]> wrote: > Hi Craig, > > > > What you describe sounds like an UL interference problem to me, although > you did indicate that you ran eDetect and did not see any interferers. > Here is what I would try. > > > > If you are still setting up your sectors and the channel plan is still > flexible, I would run ACS on each sector to find out the best channel(s) > for each sector on a tower. ACS provides a measurement of every kind of > noise / interferer seen (e.g. 802.11 based, Canopy, etc.). > > > > Once this is done, I would then select pick an ABAB configuration which > optimizes the findings from ACS where the A and B channels are spaced at > least 5 MHz apart. For example, using 2412 and 2437 provides 5 MHz guard > band (i.e. 2422-2427). 5 MHz of guard band is all that is required when > you run in TDD Sync. > > > > eDetect can now be run on each sector in order to detect 802.11 UL > interferers. It will not detect other types of interferers. I would run > eDetect in local mode as all spare time is spent looking for interferers. > > > > These steps should allow you to pick the best two channels for an ABAB > configuration to solve your UL performance issues if it is due to > interference. > > > > With regard to UL RSSI, when using the Subscriber Module Target Receive > Level (TRL) on the AP in TDD Sync mode, this should be set around -60 dBm > on the AP. This field defines the RSSI that the AP will hear from each > SM. Each SM will change its transmit power so that the exact UL RSSI value > as defined by the SM TRL is realized for the SM at the AP. If you set the > value higher than this, then the back side sector AP will start hearing the > SMs from this AP. This is because the front to back ratio of the sector > antenna is 30-35 dB. If the SM TRL is -60 dBm, then the noise floor due to > the backside sector is somewhere between -95 to -90 dBm. If the SM TRL is > raised to -50 dBm, then the noise floor due to the backside sector is > raised to between -85 to -80 dBm. The overall CINR is no different in > either case and additional energy is added to the environment. > > > > ePMP has optimized its sector antennas with front to back ratio. I > recommend using these sector antennas. If you use different sector > antennas, choosing high gain antennas that have poorer front to back will > actually hurt you. Say you choose a sector antenna that gives say 2 dB > better gain, but the front to back suffers by 5-8 dB on average. Then the > CINR will be decreased to 22-30 dB best case and the highest MCS may not be > achieved on both the UL and the DL due to interference from your back > sector. > > > > In the US and everywhere except for ETSI, the ePMP does not support CCA in > TDD mode. It does not wait to transmit based on environmental noise. > Therefore, if throughput and MCS is decreased at high RSSI for a site, this > is most likely due to interferers occurring at the same time and raising > the noise floor. > > > > With regard to 10, 20, or 40 MHz channels, what I would do is look at the > noise level using ACS for each channel size. If you have really clean > spectrum, you could use 40 MHz, but if not you might find cleaner 20 or 10 > MHz channels that favor their use. In general the channel bandwidths > perform comparably in similar noise environments (of course doubling the > channel bandwidth doubles the noise floor). > > > > I hope this helps. > > > > Dan Sullivan > > ePMP Software Manager > > Cambium Networks > > Cambium Networks Community Forum <http://community.cambiumnetworks.com/> > > > > > > > > *From:* Af [mailto:[email protected]] *On Behalf Of *Josh Reynolds > *Sent:* Monday, June 08, 2015 3:41 PM > *To:* [email protected] > *Subject:* Re: [AFMUG] EPMP 10 mhz vs 20mhz > > > > None of our radios have high PPS load. Matter of fact, average packet size > on our network is 1.3k. PPS per radio is very low. > > FWIW, there are no file sharers on our network. If they exist, their > connection is encapsulated over VPN. > > Josh Reynolds > > CIO, SPITwSPOTS > > www.spitwspots.com > > On 06/08/2015 11:16 AM, Rory Conaway wrote: > > The pps and cpu load absolutely is another variable you need to take > into acount, especially with 400 and 526mhz atheros �processors that are > also running polling. � Ignore it as part of your overall strategy and > you could be wasting spectrum. �If your ap never exceeds 80mbps, why do > you want 30mhz channels. �Sarcasm aside, does that help you understand my > point. > > > > > > > > Sent from fromm phone where I type with a single digit so please excuse > shortcuts or typos. > > > > Rory Conaway > > Triad Wireless > > > > -------- Original message -------- > From: Josh Reynolds <[email protected]> <[email protected]> > Date: 06/08/2015 2:17 PM (GMT-05:00) > To: [email protected] > Subject: Re: [AFMUG] EPMP 10 mhz vs 20mhz > > I think we are having two different conversations, and I have no idea what > you are talking about right now. > > What we were discussing has to do with channel sizes, epmp, and ubiquiti. > In particular, why UBNT 40mhz isn't any better than 30mhz in terms of > efficiency. > > This part of the discussion has nothing at all to do with any theories on > PPS you may have, other than those you have tried to inject into this > discussion. > > On Jun 8, 2015 10:07 AM, Rory Conaway <[email protected]> > <[email protected]> wrote: > > Excopt that as was mentioned before, the s/n ratio goes down and if you > aren't hitting the limits of the physical layer in 20MHz, why do it? > > > > > > > > Sent from fromm phone where I type with a single digit so please excuse > shortcuts or typos. > > > > Rory Conaway > > Triad Wireless > > > > -------- Original message -------- > From: Josh Reynolds <[email protected]> <[email protected]> > Date: 06/08/2015 12:43 PM (GMT-05:00) > To: [email protected] > Subject: Re: [AFMUG] EPMP 10 mhz vs 20mhz > > I can assure you that on radios connected in a ptp config or small ptmp, > that you will see more throughput on the 30mhz channel given a noise floor > of -97 and signals in the mid -50s, even with nothing connected on the > other side of the radios. > > Its an efficiency issue. > > On Jun 8, 2015 8:13 AM, Mathew Howard <[email protected]> > <[email protected]> wrote: > > I kind of does, the way I understood it, that bottleneck limited you from > really being able to do anything beyond what a 30mhz channel could support. > > Now that I think about it, I have seen 40mhz perform better than 30mhz... > but yes, that was because of RF problems, and neither one was doing > anything close to what it would with a good link. > > > > On Mon, Jun 8, 2015 at 11:09 AM, Josh Reynolds <[email protected]> > wrote: > > That is a bottleneck in the system, but not relevant as far as this > discussion goes. That has nothing to do with the 30/40MHz channel > efficiency per say. > > On Jun 8, 2015 8:03 AM, Rory Conaway <[email protected]> wrote: > > The limitation on the older xm radios was pps.� When you added a lot of > small packets and airmax, you could drop down to as low as 40Mbps.� In > the real world in ptmp mode. We planned for 50mhz per AP with eveything g > taken into account. > > > > > > > > Sent from fromm phone where I type with a single digit so please excuse > shortcuts or typos. > > > > Rory Conaway > > Triad Wireless > > > > -------- Original message -------- > From: Josh Reynolds <[email protected]> > Date: 06/08/2015 11:59 AM (GMT-05:00) > To: [email protected] > Subject: Re: [AFMUG] EPMP 10 mhz vs 20mhz > > This. Thought it was pretty obvious but I guess I assumed too much out of > some on this list ;) > > On Jun 8, 2015 7:33 AM, Jeremy <[email protected]> wrote: > > I think he is talking about using 40MHz channels on the older M series, > that didn't have gig ports.� It was my understanding that the processor > would get taxed as well on a 40MHz channel, making 30MHz actually work > better. > > > > On Mon, Jun 8, 2015 at 9:24 AM, Josh Luthman <[email protected]> > wrote: > > Ubnt and epmp have gig ports. > > Josh Luthman > Office: 937-552-2340 > Direct: 937-552-2343 > 1100 Wayne St > Suite 1337 > Troy, OH 45373 > > On Jun 8, 2015 11:20 AM, "Josh Reynolds" <[email protected]> wrote: > > I don't know how epmp does it. > > For UBNT, a 30mhz channel is just a "fat" 20mhz channel in the atheros > chip. Single operation. For� a 40mhz channel, it's really two 20s, > meaning radio operations are ran twice. Loss in efficiency, also marred by > the lack of gigabit port. > > On Jun 8, 2015 7:13 AM, Mathew Howard <[email protected]> wrote: > > I've never seeing much difference in performance on the ubnt M series > between 30mhz and 40mhz channels, so yes, I would say that is true... but > I'm not sure how much applies to ePMP - they do have a much a faster > processor and on a software level they are very different. > > So far, I have been running all of our ePMP APs on 20mhz channels and PTP > links on 40mhz or 20mhz, depending on how much capacity they need. I > haven't really seen much need to go down to 10mhz channels with ePMP. > > > > On Mon, Jun 8, 2015 at 8:43 AM, Shayne Lebrun <[email protected]> wrote: > > I seem to recall that with the M series, at least, a 30 mhz channel works > 'better' than a 40 because the 40 is really two 20 mhz channels bonded > together, where a 30 mhz channel is a 30 mhz channel. > > -----Original Message----- > From: Af [mailto:[email protected]] On Behalf Of Rory Conaway > Sent: Saturday, June 6, 2015 8:32 PM > To: [email protected] > Subject: Re: [AFMUG] EPMP 10 mhz vs 20mhz > > I'm not that familiar with the ePMP's yet but I can tell you some things > that we saw with Ubiquiti.� One is that channel width does not scale with > bandwidth that that Atheros chipset.� For example, 40MHz channels rarely > hit their theoretical maximum due to a variety of factors, noise, lower > s/n, processor limitations, etc...� Second, 20MHz channels seem to be the > sweet spot but even with GPS sync, you have to deal with reflections.� > Third, 10MHz channels have more overhead as a percentage of total capacity > and don't handle a lot of users well (above 40 for example with the older > 400MHz chipsets. I'm starting to deploy XW radios with the 520MHz > processors but everything is 20MHz now so I don't have a comparison).� We > did see peaks of 32Mbps with some customers on 10MHz channels but that's > non-peak times.� In peak times, we were seeing 8Mbps when more users were > online. > > Rory > > > -----Original Message----- > From: Af [mailto:[email protected]] On Behalf Of Craig House > Sent: Saturday, June 06, 2015 5:20 PM > To: [email protected] > > Subject: [AFMUG] EPMP 10 mhz vs 20mhz > > We have deployed 6 towers to begin our new EPMP network and 4 of those > towers have a full cluster of 2.4 90 degree EPMP sectors.� They are > configured with ACS turned off now because in several cases they all ended > up on the same or very close to the same channel.� I have Front back > designations and non overlapping channels set up on all towers.� I have > tried 40 mhz 20 mhz and now 10mhz channels and while the customer stability > has gotten better the more I play with settings I have kind of hit a point > I dont know what else to try.� I have some that the uplink quality will > vary wildly from 100% to 0%.� Most have gotten better since I went to a > 10mhz channel.� Most of the customers get 12MB -30mb down in the wireless > link test but the uplinks are as bad as .17.� �What is the cause of > this poor uplink quality?� Is it interfernece?� My one 5ghz AP does not > have this problem but even with noise many of these customers have -50 > signals and oddly enough the ones with the great signals seem to be the > ones that have the poorest link tests on the up link side.� I also have > customes with -65 or -72 signals that get 5MB up on the same sectors?� Im > scratching my head a bit on what the fix is for this?� Should I leave ACS > on and change everything to 10mhz channels?� Will a full cluster with ACS > on work all on the same channel? > I'm used to FSK where you pick your channel and any channels that are > adjacent will cause problems with connected SM's.� So am I just applying > old knowledge to a technology that it doesn't apply to? > > Craig > > > > > > > > >
