Yesterday I had to quickly choose which of three antennas to put on a router being used as a client at a distant site (400 meters) that has marginal reception.
I had asked the people to pick up some 7 dbi omnis before they left the country (we are in the boonies in Mexico). But they didn't. "Maybe next time, but could you please get our laptop online now ?"
So, while setting up the router (WRT54G with DD-WRT) at my house, which is much closer, I brought up the setup pages for DD-WRT for the client router and for the base station router (Buffalo HP on dd-wrt)
As I watched the signal level on both routers (seperated by about 150 meters), I swapped antennas to see which yeilded the high readings in DD-WRT. Speed was fixed at 12 Mbps.
Results given in Signal strength, but sig/noise ratio corresponded. Results wavered by 2 %.
1) Linksys WRT54G stock antenna: 50% on client end, 58% at AP
2) Buffalo WHR-HP-54G stock antenna (supposedly high gain) with sma to tnc adapter: 48% on client end 58% at AP
3) Belkin "9dbi" panel antenna with 4' of LMR 100 and sma to tnc adapter: 48% on client end and 58% at AP
4) Linksys stock antenna with "EZ-10" reflector: 58% on client end and
65% at the AP.
I was surprised. I thought that the 9dbi panel would outperform the others. Of course I did pick it up for $5 as a discontinued item. Maybe that's why. Next I thought that the Buffalo antenna would show better than the linksys.
The reflector was no surprise. They really do help. At the site, the reflector added about 4 dbi to the S/N ratio but that got it up to 20, as shown in DD-WRT. I used it with the original linksys antenna since it seemed as good as the other two I had.
Question: I understand that doing a walk test at fixed connection speed is best, but still, is this approach totally worthless or is it reasonable for me to make comparisons in this manner?
Not totally worthless, but close. The problem is reflections and multipath. It's very difficult to generate real antenna patterns and gain measurements without either an RF anechoic test chamber, or a proper test range. The last time I did it for 2.4Ghz, I had to put both ends of the link on top of a 15ft high fiberglass pole and 12ft ladder to keep the ground bounce out of the picture. This particular test was done in a parking lot. We soon found that moving cars around would create huge variations in apparent gain. You could see people moving around.
For example, here's a good example of how to do it all wrong:
The problem is with the 200 meter test range. The antennas at both ends were well within the fresnel zone. Ground bounce was also evident. There were also some obvious obstructions (trees, utility box) in the path. Judging by the photos, it was done with at least one end at ground level.
Here's a somewhat better way:
See the sections "Measuring Radiation Pattern". Note that the other end of the link uses a high gain yagi antenna to reduce the amount of RF that hits the ground and bounces. They would have done better if they had used something higher than a camera tripod, but good enough.
Try this simple experiment. Take a fiberglass pole and put any antenna on top. Doesn't matter what type as long as the gain isn't too high (
If fact, my test was from an elevated base antenna (12dbi yagi) but the client end was on my tabletop and certainly included all the variables you mentioned.
I guess the implication is that even though various antennas are compared in the same spot, unless they are up high, then just by having different patterns, one is inviting an often severe variation in reflections and refractions.
I know, sorry. My bad. I didn't see a way to use netstumbler on a client router (or would it work?) but DD-WRT does give dB readings. My eye got seduced by the bar graph and I didn't realize I'd be "publishing" numbers.
Next time I'll at least put it up on a pole and use dB readings.
Well, that's better than usual. Here's a fun test. Setup a link between any two machines and continuously record the signal strength over an extended period of time. Use SNMP in DD-WRT and MRTG to make it easy. Instructions on request. Don't change or move anything on either end. What you'll find are radical variations over time. It can be weather related, reflections from moving objects, or just water condensing on the antenna. Lots of sources of errors. Typical is at least 4dB variations in signal strength. If you have bad luck and end up with multiple reflections, probably 10dB overall variations. Kinda hard to use such numbers as some kind of reference to measure antenna gain.
Exactly. That's what the anechoic chamber offers (besides a great place to have lunch without the cell phone ringing). By eliminating reflections, you eliminate most of the major sources of errors and variations. Incidentally, when measuring antenna gain, it's considered good form to use a reference antenna with known gain. You make the initial measurement with the reference antenna, switch antennas, and do it again. The difference in measurments is the relative gain.
Netstumbler gives BOTH the client and the access point's indicated signal strength and signal to noise ratio (SNR). Scroll horizontally with the numerical tables to see them. The column headings are not very clearly labelled so you'll need to do some digging to figure out what is indicated, peak reading, and in what direction. I tried to find a Netstumbler reference page that defined the column headings, but couldn't. You really need all 4 numbers to do a decent job of determining what's happening. For example, an overpowered transmitter will yield a much higher signal strength in one direction than in the other. Any interference might show up only at one end.
Thanks. I was looking all over the web and didn't bother to check the help. (argh).
Oops. Looking at the above help page, Netstumbler does NOT show the signal and noise for both ends of the link. It apparently shows only what the local card shows for signal and noise. Well, one of the numerous sniffers that I've used does show everything, but I seem to have muddled them somewhat. I'll dig. It's either Kismet, Wi-Fi Hopper, or one of the network sniffers that can decode 802.11 packets (WireShark). I'll check when I have time.
I sorta considered the Linksys stock WRT-54G antenna to be a reference of sort. I assume (I know, I know) it is a classic 2dBi omni rubber duck. I guess you are saying that the terms and statements should be made as "gain over reference antenna."
I now understand about the reflection, refraction issues, but apart from that is the time issue you mentioned. It seems like if one is not seeing much variation (1-2 db) over the short-term with the control /reference antenna, then swapping the antennas quickly and measuring right away should give a decent comparison? That is if there aren't any moving objects on the scene....??
My biggest question now is about using Netstumbler or Kismet etc. vs using DD-WRT to measure. Are you saying that DD-WRT is NOT an acceptable instrument for signal measurement ?
More specifically; "When running DD-WRT-managed devices on both ends of a wireless link in a controlled environment and then comparing numbers (dB and SN ratio) reported on the "wireless status" page on each end's interface as I switch out antennas (on one end only) would this be an acceptable way to compare antenna gain?"
If not, why? Is DD-WRT's signal level or SN reporting unreliable or poorly derived? Is Netstumbler's or Kismet's much better?
Also, I still don't see how Netstumbler or anything else running on pc would work when using a router in client mode. I'm not able to test it right now, but it seems that Netstumbler would not have any way of seeing the signal strength if the client router is connecting the computer via ethernet cable...
This is particularly useful thread, I think. So many people are buying and building antennas. We all want to be able to tell which antennas deliver measurable gain. Most folks probably depend on marketing hype and anecdotal evidence (I put on one of those machV doodads and it's really humming now, lemme tell 'ya).
And of course we also forget about that four feet of LMR100 and the TNC adapter that we're using.
And we have no way of judging the accuracy of what tests are reported. It would be great to have some understanding and agreement so that various people could be reporting reliable tests with the antennas they have.
I wish I could do that little test again, but the Linksys is in use and cannot be pulled. Has anybody else compared the Buffalo HP's included "hi-gain" rubber duck to the Linksys's included rubber duck? Is it likely that "hi-gain" to Buffalo means 2dBi? Or that the RPA to TNC adapter could eat up any difference?
Nope. It's not a repeatable or well controlled antenna pattern or gain. Inside the rubber radome is just a 1/4 wave piece of coax cable with the braid peeled back over the jacket. There's no way to build two of them that are reasonably identical or any way to prevent other parts of the system (box, coax, connector, table top, etc) from becoming part of the antenna pattern.
Yes. Something like that. Ideally, the reference antenna should be tested in exactly the same position as the antenna under test.
Nope. Cancellation and reinforcement as it goes through nulls is difficult to avoid. A person walking at 1 meter/second will go through about 8 wavelengths of 16 nulls and 16 peaks. It doesn't take much to change the signal levels. I find myself using blue masking tape to old my test setup onto the workbench to keep it from moving and changing the levels. It's much easier to take large number of readings over a fairly long period of time, throw out the very high and very low readings, and average what's left.
Incidentally, I get a kick out of data sheets that claim 3 decimal place accuracy for antenna gain, when they're lucky if they can measure the gain to one decimal place accuracy.
DD-WRT is fine (including Kismet Drone). All that I was concerned with was that signal level and SNR measuring should not be measured just at one end of the link as tx power levels are not guaranteed to be identical.
That's as good as it gets. There's no guarantee that the actual numbers are accurate or even repeatable, but that will have to do under the circumstances.
They're all derived from the RSSI level, which is only 8 bits (0-255) range. It's also not very linear. It's good enough for saying one antenna is xx dB better than another, but not very good for making absolute signal level measurements. SNR should be fairly accurate.
Netstumbler will NOT see a client radio. Netstumbler works by sending a "probe request" to any access point and listening for the responses which contain the SSID. That's called an active probe as the sniffing computer must transmit something in order to get a response.
Kismet is a passive sniffer, which will extract packets out of thin air and does NOT need to transmit anything in order to get a response.
It can't. However, it is possible to sniff the traffic betwenn this client and an access point and extract the signal strength and SNR. Management packets, going in either direction, all contain the signal strength and SNR.
The easiest way to do that is to compare the antenna with a known antenna (as previously described).
That's why I like RF. It's all magic, hype, and anecdotal rubbish. Building and designing antennas is very easy compared to testing them. I've had more than a few antennas that look great on paper (or printout) and don't work anywhere near what I intended.
Some manufacturers also forget about those. They specify the gain of the antenna, without the slightest consideraion of the coax and connector losses.
True. The most common way to compare antennas is to NOT use the test results, but rather to compare the antenna models. There are serious problems with doing it this way, but it does tend to be fairly useful. There's also a big difference between being able to connect at long ranges, and being able to stay connected reliably while doing useful work. Sorry, but I don't have an easy answer. I tend to compare computer models instead of test results. That opens a whole new can of worms, but seems to be adequate for purposes of comparisons.
Dunno about the antenna comparisons. The RP-SMA to RP-TNC adapters have very little loss. Might was well ignore them.
At 2.4GHz, the loss through all those connectors was 2dB or about
0.15dB loss per connector pair. That's essentially zilch.
What would you recommend using as a reference? An isotropic (as if I knew what that is!) radiator? Is there such a thing?
Yep, that's certainly how I shopped for our base antenna. But many antennas don't have any published model or even offer and idea of it's gain (i.e. the Buffalo and Linksys rubber ducks).
And then, sometimes there are numbers (gain) but no model. I wonder if the published or stated numbers are even close to reality. I guess that if one has a fixed location for the antenna already then just swapping out antennas at that location will tell you which to use there.
This sounds interesting, but I don't have the time right now. I can see that I really need to sort out SNMP though...might ask for those instructions later.
I was hopeing you wouldn't ask that question. There are plenty of reference antennas for the usual FCC type certification tests. They start at HF and work their way into microwave land. However, the purpose here is to have a known gain across a wide frequency range. That's not the case if you're only interested in 2400 thru 2483.5MHz which is a much narrower frequency range. So, the ideal antenna for this would have:
A constant or known gain from 2400 to 2483.5MHz.
Very good VSWR characteristics to keep the coax cable from radiating.
Simple and reproduceable construction.
The first two requirements immediately eliminate any high gain antennas. In general, above perhaps 19dBi, the bandwidth and VSWR of the antenna at the band edges is fairly horrible. If you want a wide bandwidth, the antenna has to be a fairly low gain antenna.
For antenna range tests, it's also desireable to have the reference antenna be a close approximation of the antenna being tested in order to reduce or eliminate having each antenna affected differently by the local environment. For example, if the test antenna is highly directional, while the reference antenna is not so directional, differences in ground bounce and reflections from nearby objects will have an effect.
The only antenna that I know for sure will do the trick is a full wave (12.5mm) loop antenna with a reflector. It gets about 8dBi of gain, low VSWR, and is fairly broadband (not sure exactly how wide). It's fairly close to this Maxrad antenna (which is what I've been using).
The PCB antenna can be replaced by a full wave look and some matching stubs. A biquad is a similar antenna.
The problem is that no matter what you decide to use, if you want accuracy, it has to be properly tested on the bench and at an antenna range. However, if the manufactory has done a tolerable job of characterizing the antenna, then those numbers can be used instead.
Yep. That's how I seperate the hype from the technology. Those products without test results (and published testing proceedures) probably have something to hide.
for examples of what proper specs and test patterns should look like. Note that these are test patterns from an antenna range, not from a computah simulation.
They're not. Everybody lies, but that's ok because nobody listens.
That might work in one dimension but not two. For example, if you compare two omnidirectional antennas with the same gain, but different downtilt patterns, you might get the same gain, but at different vertical elevations. Lots of other ways to make comparisons come out weird.
Actually, I've seen simulations that come close. Two pairs of crossed dipoles feed in quadrature works fairly well if the coax cable is decoupled and fairly small. That's easy at HF and VHF, but the coax gets in the way at UHF and up.
Turnstile antenna. It's circularly polarized and has a very wide hemispherical radiation pattern suitable for covering the entire sky. It's usually used with a ground plane reflector, which ruins the isotropic approximation, but adds some gain. I would not want to build the phasing section at 2.4GHz.
I found an isotropic approximation among the samples supplied with
4NEC2 antenna modeling software.
It needs a complex 4 port feed system, is nowhere near 50 ohms, but looks very close to a perfect unity gain sphere. The 4 feed lines are not shown in the model and will probably make a mess of the pattern.