7 mile link with 11a, not up to par

[ginovilla]

Hey list, I wanted to share my experience with my atheros setup.

Im using wrap boards with senao abg cards and Pac Wireless 29 db dishes, I got all from Eje, except the wrap boards... Radio in enclosure besides the antenna, 18" LMR400 jumper

well, this is a 6.89 mile link , and acording to my calcs, I should be getting -54 db of rss. Well guess what!, Someone stole me 18 db somewhere! I just can obtain -72! (using 20 db tx setting). changing the power setting from 0 to 20 and 30 only increased the rss by 5 db each!

Looking for other atheros mini pci cards to test out... are the proxim compatible?

Gino

btw: pulled 16 mbps in 24 mbps mode

[jbland]

I'm currently having a similar problem with a 13 mile link. -90 signal when this supposed to be -68. I'm using the same antennas. I did test at 6 miles and got about the same results you did which does not make any sense.

I don't think it's the card as I have bought 2 different cards and 3 other pigtails. I even clipped the U.Fl connector and soldered straight to the board. Changed out all LMR cables etc... The only thing I haven't done was change the antenna.

I talked to Eje and he mentioned several people using this setup sucessfully for 25-30 miles. Heck with the gain we have this should work well at 40 miles and still have db to spare.

Eje also mentioned that those success stories were all using Mikrotik. The more I play with it the more I'm starting to think it's something with StarOS. If I had 2 copies of Mikrotik I would try it out just to see if the link works, but I'm not about to spend close to $400 just for the purpose of testing.

This is only speculation but I think the tx gain setting is not working right. You had better results than I did but I can go from 0, 10, 18, 20, 30 and see no difference.

[georgew]

Do you have the full fresnel zone? Have you verified your aiming?

When I was being trained to install 38ghz dishes, they had us find the secondary sidelobes, and verify our final aim was indeed the primary lobe. The reason for this was because it is easy to fool yourself into aiming with a sidelobe. 18db is about what you can expect the sidelobe penalty to be.

I've also noticed that the main lobe from most dish's is not exactly where it is supposed to be. For some reason I often find that the best aim is about half of the angular lobe measurement above the horizon. So for an antenna with a 7* lobe, aiming 3 or 4* above the horizon gets the best signal. I assume this is because of fresnel zone clipping on the bottom of the fresnel zone.

Also, each connector is costing as much as 3 to 6db. It might help to directly connect the pigtail to the antenna.

I haven't done any 5ghz tests yet, but when I do, I'm going to mercury wet all of the connectors, and do before and after tests to gauge the effect of a fully soldered signal path.

[jbland]

Make sure you are running the latest firmware, I think it's 1.13.3b10. I was running version b4 and it seems as if there were a lot of fixes including a tx rate fix at b5 that I overlooked. Also I changed a atheros card and pigtail at my far end link and saw a 10db better signal. I suspect the pigtail but won't know until I test it out separately.

[tim]

I haven't done any 5ghz tests yet, but when I do, I'm going to mercury wet all of the connectors, and do before and after tests to gauge the effect of a fully soldered signal path.

Mercury wet ? I have not heard of this before, presumably a small dab of mercury on the connectors will cause them to create an amalgam, which improves the connection ?

Tim

[ginovilla]

I got Full fresnel, speant 2 hours aligning the dishes to make sure we're not getting a lobe. Im running version 1.13.3b8, were i can find vb10?

any ideas?

Im thinking of using another radio, maybe dlink or proxim.

Gino

[jbland]

If you use different radios please post the results. I'm using the same radio's you are. I replaced both radio's and pigtails and got a 5db difference. It's not what I was hoping for but the connection is solid and will easily support the 10 customers I have on it.

Just as you have, I have spent literally hours on the alignment and even borrowed a pair of high power lasers to get them perfect.

[bobbyc]

I just don't trust those tiny pigtails... but there have been others using these same radios/pigtails and I think all have gotten signal levels that are below what they expected... so I don't know if it is the radio/pigtail to blame or the beta drivers.
I have dlink AG520 cards and the signal levels were stronger than what I expected (not by much though) so I am happy. One reason I bought them is because they were easy to get (compusa), but also because these cards are what Tony/Lonnie are playing with and therefore *might* be better supported at this time.
The best thing to remember is that these u.fl connectors are tiny and have a great potential to be the problem of the low signal... but keep in mind that all the atheros releases have been beta thus far, so keep on reporting all the glitches.

[georgew]

I haven't done any 5ghz tests yet, but when I do, I'm going to mercury wet all of the connectors, and do before and after tests to gauge the effect of a fully soldered signal path.

Mercury wet ? I have not heard of this before, presumably a small dab of mercury on the connectors will cause them to create an amalgam, which improves the connection ?

Tim

Actually I was going to use enough mercury to create a "soldered" connection. For sensitive signal switching, mercury wetted relays used to be common. The idea is the temporary connection inside the relay had the conductivity of a soldered connection. The seal of the N connectors will contain the mercury, so it won't evaporate.

The mercury will bond to the surface and form an amalgam, but with enough mercury to be wet, the two conductors will have more than just the normal sliding contact, the mercury will wet both parts and fill in the gap to produce a soldered joint. It will be a wet solder that will not solidify, but that's ok.

Mercury is dangerous, and you should not apply it indoors, as any dropped mercury will tend to break up into tiny beads that are difficult to clean up, and will slowly evaporate into the air, contaminating the air in your home/office. Use it in a area that is never closed up, so that concentrations of mercury vapor cannot form. It is my understanding that inhalation of vapors is the main vector of mercury poisoning from the pure metal.

[jbland]

I don't know if the pigtails are to blame. I got 3 more pigtails to play with and all seem to give the same results. On my 2 of them I clipped off the U.FL connector and soldered the end straight to the board. Also removed the N connector and soldered to the LMR cable. The improvement in gain was 1db which tells me that the connectors seem to be working fine.

I think it's safe to say that we are indeed using beta software and each version gets closer and closer to finxing things. On another post Lonnie mentioned that he has a 1 mile connection up that is getting 10db less signal than he expected, so I'm sure they'll be more beta releases in the future that will help our problems.

I know someone who is using the exact same equipment I am and doing a 25 mile shot with -72 signal. The only difference is he is running Mikrotik, which has been supporting atheros longer than Star-OS. The moral is that the equipment can go the distance, we'll just need to wait until the StarOS guys get things tweeked out to perfection.

I have a 13 mile link up with signal around -90, but yet my connection is a solid 10mibt transfer. Based on the card specs this shouldn't even be possible as the 6mbit threshold is -85. So possibly there is an issue with the signal reporting, software driver, etc... I'm just hoping to get 10db more somewhere and I'll be happy.

[ginovilla]

Tomorrow I'll be replacing the senao cards with proxim cards, also im taking with me a pair of proxim mp11.a radios to compare speeds..

let you know

Gino

[georgew]

On my 2 of them I clipped off the U.FL connector and soldered the end straight to the board. Also removed the N connector and soldered to the LMR cable. The improvement in gain was 1db which tells me that the connectors seem to be working fine.


Ever notice that the fatter the coax, the fatter the center conductor? Ever wonder what the difference between 75ohm coax and 50ohm coax?

It's pretty difficult to solder a pigtail coax to a larger LMR 400 cable... without loosing some of the signal at the joint. That you got a 1 db gain is impressive. But I wonder what that really means, since there has bound to be some mismatch... At 5ghz it's going to be difficult to get the connection right.

I suspect that you were loosing 3db before, and now you are only loosing 2db... and this is impressive because I would have predicted the soldered connection to be worse rather than better than the N connectors for the cable.

This is why I talk about mercury wetting the connectors, to preserve the impedance match, and improve the connection at the same time... The mercury soldered connection should impart no additional loss over that of the cable, and since the connector is short, you should be talking about a few hundredths of a db.

Not to mention that the LMR400 center conductor is aluminum, so you can only solder the sides, but the solder won't stick to the end, so symmetry is difficult.

I'm impressed you were able to pull off a coax splice that was better than a connector...

[jbland]

The moral of the test is that there is something more to the problems some of us are having than the N connector. Granted that a loss does occur I doubt that it is in the neighborhood of 20db which I'm seeing on the link. I would speculate that the drivers are not setting the tx gain correctly or the signal reading is wrong.

I have a 13 mile connection that seems to be solid, but yet the signal is at -90. Based on calculations with liberal losses for cable and connecters I should still get -67db. Throughput tests give me 10mbit which is actually not supposed to work after -86 so go figure.

The mercury wetting of the connection is ovekill, but for the 2-3db gain is it worth it. I would just buy a higher gain antenna.

[georgew]

You are right, it is overkill, and it can evaporate if the connector is not sealed after use. It can also wick away from where it is needed under the right set of conditions.

But to prove the amount of loss you are seeing in your connectors, even short term results can provide a needed sanity check for the purpose of eliminating the connectors as the cause of unexpected loss.

Really, the radio should have a built-in SWR meter, this should be easy to implement, and tell us the health of our entire antenna system.

All the same, I suspect you are right, it's looking like it is not the cable system causing the low signal levels.

[ginovilla]

Well guys, just changed one radio on one of my link ends, swapped the Senao radio with a proxim... The results:

I lost and additional 6 db! my received signal on the end where the radio was not changed went from -78 to -83... so go figure... the cards appear to be ok then...

any ideas? lonnie? tony?


Gino

[lonnie]

No, we are still fighting with this one. Tests with the SA show that noise comes up dramtically as you set power above 20, so 20 seems to be the proper point and that is a bit higher than the cards are supposed to put out. I think they are just weak transmit but very good receivers make up for it.

[ginovilla]

Lonnie,

I just think we're loosing too much signal on this, do you have other cards to test on you're sa? a Dlink maybe? do you want me to send you the proxim one?

Also I noticed, changing the tx power from 10 o 20 only yielded a 1 or 2 db gain....

maybe check your atheros code, i think it needs a 10x multiplier :-)

I tried putting 200 for tx power value but it gave me an error... The cards just appear to be putting out 1, 2 db rather than 10 or 20

Gino

[jbland]

Also I noticed, changing the tx power from 10 o 20 only yielded a 1 or 2 db gain....

maybe check your atheros code, i think it needs a 10x multiplier

I see about the same thing. I've even tried 0 for the tx and only loss about 2db. I can't see how I can go from 20 tx to 0 tx and only have a 2 db loss, it's almost as if I'm completely on antenna power.

For giggles I recalc'd my link using 0 as the TX for the card and I should still get a better signal than what I have.

Please don't take what I have said as a complaint. Just want to provide some feedback and hopefully provide some insight for those struggling with this before you change out everything and still end up back at the same spot.

[lonnie]

0 is "use factory default". The settings are from 1 to 30. The cards only have about 18 dB so anything above that should not do much. We have not seen much change above 20.

[mp3turbo]

hello Lonnie,

are you sure those cards are using EXTERNAL antenna connector as output? I've seen this on 2.4GHz setup : DLink 900AP+, max. 17dB output, 16dB sector, 1m cable (very short), client : Prism PCI card, -88dB sensitivity for 11Mbps, 24dB antenna, very short cable, 1 mile distance (! one mile !) .

Received signal strength : -80 to -83dB. WTF ?!? I should expect around -60dB on client. Problem? Dlink 900AP+ WAS NOT USING EXTERNAL ANTENNA (no matter which antenna I choose, left, right, diversity), I WAS RECEIVING SIGNAL FROM INTERNAL "STICK" !!! AFTER PUTTING HACKED FIRMWARE INTO IT, SIGNAL WENT FROM -82dB (QUALITY AROUND 24/92 IN LINUX) TO -59dB (QUALITY 76/92) !!!

Isn't something like this happening here?

bye, mp3turbo.

[lonnie]

We will double check this but it had been working. Selecting the second antenna would drop the signal to zero.

[lonnie]

What is this hacked firmware? We have used a few 900's and could use that if it makes such a good result.

[bluezx9r]

Here's the link to the hacked firmware.

http://home.earthlink.net/~mlampie/PowerHackDWL-900+.html

ST

[ginovilla]

I just replaced the Wrap boards with senao cards of my links with a Proxim mp11a kit, used same antennas,

I used the 36 mbps data rate on the mp11a, which puts out 15.5 mw . I was able to pull 20 mbps throughput on this.

The only drawback is that the proxim firmware does not gives a rss reading, it gives a snr only. So I cant compare directly with the wrap-staros-senao setup. BUt the snr was around 28-32 and 5.8 is damn noisy around here so we can figure a noise of at least -85 and if we substract the snr reading, it will give us a estimated rss of around -57 to -61

I just wanted to let you know

Gino

[jbland]

I was going to try the open source linux drivers for atheros cards to see if the signal changes, but I couldn't find an easy client mode implementation.

I just want to know if the problem is StarOS or my hardware. From what I've read I would lean towards driver.

-James

[tony]

Try the madwifi project, which supports Atheros cards. It may be a good client platform to compare signals with.

Thanks!

[lonnie]

Anybody have any results of madwifi and identical hardware? How about Mikrotik? I am reluctant to get a license just to test. Hopefully somebody has a Mikrotik license they can use on the same hardware that they were using with our driver.

We now have a solid 1 km 11g through trees. For 11a we have 1.5 km clear LOS, 3.5 km clear LOS and 4.9 km clear LOS. The wierd thing is that they all have about 14 to 20 signal quality.

Each connection has 21 dB SuperPass patches and we pull over 20 mbps, solid. Even though the links do not match the expected levels I am totally pleased and I figure that 20 mbps is 20 mbps, whether you have -50 dB or -80 dB. Heck that dB reading could be all out to lunch as we have no way to verify what it is telling us. It is just a formula that crunches some numbers from the card, but there is no real reference to check against.

The quest goes on. We are pleased so far, since this beats our 11b for backhaul and opens channels for more AP units..

[h3ll]

i have 10 links running mikrotik - all with best signal (54mbps) turbo mode
farthest is around 7 km with near los but with bigger antennas (1.2 m)
getting stable 30mbps+
i'm still waiting to see if any one has some good results with staros to change my backbone. heard that staros can do 50mbps+
will be purchasing 2 licenses next week just for testing

[ginovilla]

Lonnie, list,

I just purchased 2 mikrotik licenses for this purpose... i'll do a head to head comparison... with the same hardware thats running staros, i'll just swap the compact flash... So' hang on to your hats....

Gino

[mp3turbo]

> So' hang on to your hats....

well, you can bet we are hangin'! Let us know, please.

Hellheck, Lonnie, we were able to achieve 75Mbps traffic numbers from your internal throughput test in living room between two 566Mhz celerons running at 85%. How much can we expect, given the ideal conditions, outside? No interference, no problems, short cables, fast machines (no cpu bottlenecks), strong antennas etc. ? Is it just 20Mbps? Or is there any difference between numbers in your throughput test and real tcp/ip performance? I was not able (uhm) to perform tcp/ip throughput test (lack of time), we've just set up machines, installed cards, software, fired it on (all took about 15 minutes!!! I love your code, man!). How much of TCP/IP traffic could I see, given 75Mbps in your throughtput test, between two end-points behind these two computers with atheros radios? Do you have any estimates?

Another thing: these numbers DIDN'T change when we switched off turbo mode, forced 54mbps, 48mbps or 36mbps connect. It was on beta9. Any explanation? (what about the pattern used for test? Is that stream COMPRESSIBLE by on-chip data compression technology?)

bye, mp3turbo.

[lonnie]

The 20 mbps is limited by the WRAP boards doing the throughput test. The test maxes out the CPU with the overhead of keeping tack and dealing with thedata at each end. As a Router we see about 30 mbps through them, meaning they don't do a heck of a lot, just pass the traffic between ports, and maybe do some routing, filtering ,etc but all very efficient code. The speed test is quick and dirty and uses CPU time that would be available to routing.

So we are seeing the solid 20 mbps at about 4.9 km with 21 dB patch antennas for about $150 each from SuperPass. If we put 1.2 m dishes that would add about 15 dB at each end but would add about $600+ to each end in cost.

We have a guy that has linked to 27 km with 1 m dishes but can only see -83 dB signals, which is not quite enough. He still gets about 802.11b speeds but the idea was to exceed that.

[lonnie]

>Another thing: these numbers DIDN'T change when we switched off turbo mode, forced 54mbps, 48mbps or 36mbps connect. It was on beta9. Any explanation? (what about the pattern used for test? Is that stream COMPRESSIBLE by on-chip data compression technology?)

bye, mp3turbo.

The a and g modes use compression and other tricks to be more efficient. They also go faster if they can. The rate you set is the lower limit. Our 20 mbps was obtained at the 6 mbps rate setting. If we set to 54 mbps it is the lowest it goes gets a bit twitchy since 54 is not reliable at the signal levels we have.

[lonnie]

Results of our overnight ping to the new system. No other traffic but the notice not one pcket dropped and also note the very consistent ping times. Not bad for a FULL 1500 byte packet. :D The signal quality is 12-15 and is not as high as I would like it makes a rock solid system.


10.10.31.1 : [48452], 1500 bytes, 1.30 ms (1.40 avg, 0% loss)
10.10.31.1 : [48453], 1500 bytes, 1.28 ms (1.40 avg, 0% loss)
10.10.31.1 : [48454], 1500 bytes, 1.05 ms (1.40 avg, 0% loss)
10.10.31.1 : [48455], 1500 bytes, 1.28 ms (1.40 avg, 0% loss)
10.10.31.1 : [48456], 1500 bytes, 1.05 ms (1.40 avg, 0% loss)
10.10.31.1 : [48457], 1500 bytes, 1.30 ms (1.40 avg, 0% loss)
10.10.31.1 : [48458], 1500 bytes, 1.04 ms (1.40 avg, 0% loss)
10.10.31.1 : [48459], 1500 bytes, 1.06 ms (1.40 avg, 0% loss)
10.10.31.1 : [48460], 1500 bytes, 1.04 ms (1.40 avg, 0% loss)
10.10.31.1 : [48461], 1500 bytes, 1.31 ms (1.40 avg, 0% loss)
10.10.31.1 : [48462], 1500 bytes, 1.05 ms (1.40 avg, 0% loss)
10.10.31.1 : [48463], 1500 bytes, 1.31 ms (1.40 avg, 0% loss)
10.10.31.1 : [48464], 1500 bytes, 1.30 ms (1.40 avg, 0% loss)
10.10.31.1 : [48465], 1500 bytes, 1.29 ms (1.40 avg, 0% loss)
10.10.31.1 : [48466], 1500 bytes, 1.29 ms (1.40 avg, 0% loss)
10.10.31.1 : [48467], 1500 bytes, 1.29 ms (1.40 avg, 0% loss)
10.10.31.1 : [48468], 1500 bytes, 2.26 ms (1.40 avg, 0% loss)
10.10.31.1 : [48469], 1500 bytes, 1.87 ms (1.40 avg, 0% loss)
10.10.31.1 : [48470], 1500 bytes, 1.29 ms (1.40 avg, 0% loss)
10.10.31.1 : [48471], 1500 bytes, 1.28 ms (1.40 avg, 0% loss)
10.10.31.1 : [48472], 1500 bytes, 1.05 ms (1.40 avg, 0% loss)
10.10.31.1 : [48473], 1500 bytes, 1.42 ms (1.40 avg, 0% loss)
10.10.31.1 : [48474], 1500 bytes, 1.04 ms (1.40 avg, 0% loss)
10.10.31.1 : [48475], 1500 bytes, 1.28 ms (1.40 avg, 0% loss)
10.10.31.1 : [48476], 1500 bytes, 1.05 ms (1.40 avg, 0% loss)
10.10.31.1 : [48477], 1500 bytes, 1.30 ms (1.40 avg, 0% loss)

10.10.31.1 : xmt/rcv/%loss = 48478/48478/0%, min/avg/max = 1.03/1.40/57.6

1 targets
1 alive
0 unreachable
0 unknown addresses

0 timeouts (waiting for response)
48478 ICMP Echos sent
48478 ICMP Echo Replies received
0 other ICMP received

1.03 ms (min round trip time)
1.40 ms (avg round trip time)
57.6 ms (max round trip time)
1718.236 sec (elapsed real time)

[bobbyc]

My 1.5 mile 5.3ghz backhaul link is real similar to lonnies. Every once in a while you'll get a 40-50ms hiccup.... but who cares.
I see that 5.x is a little more prone to fog. My -79 signal is now -81 and it's FOGGY!
Bob C

[ginovilla]

Well guy's

Finally returned from a long day at the mountain sites where I have my 802.11a test link. Here's he scoup:

Got to site a, another tech on site b, replaced both staros Compact flashes with mikrotik ones, no movement on antennas , no change in cable, no change in radios....

Results:

Rssi on both ends avg of -79 with swings to -83, same numbers as Staros!

So we know for sure the code is ok...., the funny thing is, I replaced the Pac Wireless 29 db dish on one of the sites with a Proxim (mars rebranded) 23db 5.2 - 5.8 flat panel... well guess what... got the same rssi...-79 ...

Go figure, any ideas?

I think we are dealing with not 1 but 2 issues:

1- really low powered cards
2- not so 29db solid dish

comments?

Gino

[ginovilla]

BTW Lonnie,

How can I use the recently approved FCC channels on StarOs?

5.4 to 5.7 ghz, I know that the card supports them

Gino

[bobbyc]

hmmm, all of a sudden the mikrotik a,b minipci card (AR5111, AR5211, AR2111) looks like it could be a better choice than the senao.
There's a couple things I'd like to rule out though; Lonnie, jbland, gino: what board are you guys using with your senao minipci radios? WRAP? Is it possible the board is underpowering the minipci radios?

EDIT: I just did some back reading and see that jbland is using a routerboard and PC, lonnie and gino are using WRAP.
So maybe it isn't the PCs if jbland is seeing low signals as well. Perhaps it is the extra cable/connector loss of 5.x? Or just bad cards.
Are the 2.4ghz portion of the cards underpowered as well?
Bob C

[lonnie]

The OFDM modulation is the cause for the somewhat low readings. There are several ways to make power.

One way is to have a large single frequency signal and it is simply Average Voltage * Average Current.

Now reduce the voltage but add in another signal and you suddenly have a lower peak voltage but higher average current, thus the power is the same. The trouble with this is that the receiver works on voltage (typically spec'ed as uV per meter), so it reports a lower voltage as a lower signal. The power delivered can be the same but the method that uses many signals spread simultaneously over the band will have higher current but lower voltage.

I am sure this is what is happening.

[georgew]

It could also be that the radios simply don't have accurate signal meters. Either by not measuring the peaks fast enough, or by simply not measuring anything past a given point. Perhaps they only measure marginal signals, and anything better than marginal is off scale.

Many A to D conversions have a limited range, and for whatever reason most of the range is on one end or the other of the maximum/minimum limits. I have seen A to D's that are simply not linear, and that could be the problem. The engineers may have only cared about the signals at the low end of the scale, and not cared about linearity past that point.
Also, it might be a 4 bit A to D internally, so that there are only 16 steps, and the -79 everyone is seeing is just the notch that most signals fall in... For example there may be no notches between -79 and -60, so it stays at -79 untill the signal passes -60.

The d2d cards do something similar... most links show 85%, regardless of being next to the AP, or 500 feet away behind a tin roof.

the probably intended -79 or better to light a green LED, and didn't anticipate the end user expecting precise DB measurements.

[tony]

We'll be happy to add the new range.

Can you provide us with the nessesary docs or FCC web site URL referring to this new range?

Thanks!

BTW Lonnie,

How can I use the recently approved FCC channels on StarOs?

5.4 to 5.7 ghz, I know that the card supports them

Gino

[ginovilla]

Not direct form the fcc but form a reliable source:

http://www.infoworld.com/article/03/11/14/HNfccspectrum_1.html

Gino

[ginovilla]

Fcc's order:

http://216.239.39.104/search?q=cache:OZmMS83qXd4J:hraunfoss.fcc.gov/edocs_public/attachmatch/DOC-241220A1.doc+5.470+5.725+order&hl=en&start=5&ie=UT F-8

[jbland]

Been on a slight vacation so I didn't get to respond. With recent updates posted by "ginovilla" I'm now leaning towards an antenna issue. If the antennas are truely 29db then these links should work perfectly with only a 1db tx power but that hasn't been true.

I'm going to purchase the Equinox 39" Dish to test with. Only going to change out 1 side, but if antenna gain is the problem this should work perfectly. I might also be looking into the 20db cards from Mikrotik, technically 18 to 20db is a big difference in gain even tho it's only two number away.

I posted a Question to John at Superpass about antenna sizes being that the PAcWireless 29 is 24" and other 29db are about 36". John being a nice guy said it was possible with good tuning etc... but I'm a skeptic and think there's something fishy when others use a larger dish.

For reference my link is 13 miles (21km). The signal is -88 plus or minus 2. I've even seen this at -91/-92 at times but yet I get a full 10mbits of throughput with 6mb set. For my application 6mb is all I need and I went to A for the frequencies.

-James