From a reliable source I am told that the various companies specs on the power or range of various radios cannot be trusted.
Trouble is in weeks of searching various wifi questions I have found no objective tests of various client mode radios.
Anyone know of any published tests where the vendor is not doing the testing? Or even a company that publishes tests on it's products? I know of one vendor that does, however he does not include most of the products he sells and the tests are pretty vague.
If there is a way to force the wifi card to put our a constant carrier, measuring the power output isn't all that difficult. Measuring sensitivity is something else, though I suppose you could insert an attenuator berween two cards wired with coax to measure sensitivity, once you know the power output of the reference card.
I think I have enough gear to do the power measurement, but I don't own any attenuators rated at 2.5GHz.
That's correct. Everybody lies, but that's ok because nobody listens.
I presume you've found some subjective tests. Any particular measurements you're trying to determine?
Yes. The problem is that I signed an NDA that limits what I can disclose.
If you go through the FCC type certification test reports, you'll see page and after page of xmit power tests and measurement methods. Start at:
search for whatever product you find interesting. The test report is quite detailed. If you need test limits and general proceedures, there's quite a bit in FCC Part 15 rules-n-regs.
Incidentally, many people measure xmit power with a spectrum analyzer and wonder why they're yielding far too low in power output. The problem is that the spectrum analyzer shows the direct sequence spread spectrum signal splattered over a 22MHz bandwidth. In order for the spectrum analyzer to measure the de-spread xmit power, you need to add a correction for the spreading. For example, if the main lobe is
10Mhz wide, and the IF bandwidth of the SA is 1 MHz wide, you add: 10 log (10 / 1) = 10 log 10 = 10dB So if the spectrum analyzer reads 5dBm xmit power, then the real CW power is 10 + 5 = 15dBm.
Receiver sensitivity is a different mess. The problem is that there's no FCC specification so EVERYONE lies (and I do mean everyone). Some vendors use the chipset vendors sensitivity specs. Others conveniently forget to include the coax cable pigtail loss, the diversity switch loss, the matching network loss, the long trace across the circuit board loss, etc. In almost all routers I've played with, the sensitivity on each of the diversity antenna ports are different.
Measuring receiver sensitivity is actually fairly easy if you happen to have a BER (bit error rate) tester, a large bank accounty, and all day to run the test. You setup the generator (about $18,000 hardware plus $3,000 in software), connect to the access point ($50), and connect a 100mbit/sec BER to the ethernet port ($8,000). Set the access point to a fixed wireless speed. Increase the signal level until the BER = 10E-5 (one error in every 100,000 bits). That's the sensitivity. You can also measure PER (packet error rate) which is quite a bit easier on the test equipment budget. Repeat the test for each of the dozen speeds and modulation modes. Figure on about 15 minutes per speed to get a stable value.
Range is yet another mess. The problem is that speed, range, and reliability are all interchangeable. I can go as far as you want, but the link is not going to be very fast or very reliable. There are estimators available that do a credible job of predicting the range given the antennas, coax loses, path loss, Fresnel zone attentuation, and of course equipment used. That's the ideal and everything else reduces the range from the ideal. For example, rain, inversion layers, interference, corrosion, moisture, bird shit, coax adapters, and rotten coax will all reduce the range.
There are plenty of path and link calculators online. See the FAQ at:
how to do a simple link calculation. You might also be interested in the theoretical maximum speeds at:
Some gear has a button to compute the power in a band as all the information is present in the analyser. In fact, this feature is present in some audio gear as well since telecom works in power rather than peak to peak voltage.
Jeff Liebermann wrote in news: email@example.com:
Be nice to know what a product is capable of BEFORE you buy it.
Already said: transmit power and receive sensitivity
Are you referring to:
? If so, then you have to get alot of other info to fill in all the blanks first.
To use these reliably you still need ACCURATE power and receive figures, or else garbage in garbage out.
To know what a product is capable of BEFORE buying. Example there are at least 3 "high power" pcmcia cards out there with similar claims in their data sheets. Which is telling the truth and how to compare? Ditto, outdoor units.
All you need is the FCC ID number on the serial number sticker. You can also use the vendor search to get the 3 letter FCC ID prefixes used by the vendor, and then just browse through their list of hardware. The FCC ID search does sorta work, but not consistently or reliably. Pick a maker and model number and I'll demonstrate how it works.
True. It really depends on what level of accuracy you're trying to achieve. 1dB is about 9% accuracy. Since the range calculations are based on a rather wide fade margin of 20dB, the results are at best a ballpark guess. That's the nature of the beast for range calculations as there is no abrupt transition between good/bad for range. There are also a mess of minor variables that will add additional errors. It makes no sense to nail the power and sensitivity to within perhaps 1dB and then use these to guess the range with perhaps a 10dB variation in fade margin plus another 10dB in minor path impediments.
Incidentally, when I was measuring receiver sensitivity on the bench with big bux test equipment, I would frequently get 3dB variations in results just before lunch and after lunch. Some of these variations were obvious, but most of them were untraceable. In other words, accurate numbers are difficult to get.
Also, one fun exercise was to have everyone in the lab read the indicated power of a spread spectrum signal on a spectrum analyzer. Without touching anything, 5 different people gave 5 different results varying over about a 3dB range.
I'm with you there. The recent trend in data sheet product is to supply the absolute minimum amount of information possible. The reason is that users tend to sue if they don't get the advertised features or preformance. I too would like to see detailed specifications, especially receiver dynamic range, but that's not what sells commodity hardware. You'll get that on high end radios (Proxim, Alvarion, and Aironet/Cisco, but not on cheap junk.
If I could give you accurate numbers that I personally tested, how could you be sure I'm trustworthy? Would you settle for a range of numbers? If so, I'll give you 3 sigma variations, which accounts for
98% of the units shipped, but is such a wide range that the numbers would be useless.
Does the test equipment have a "referece" transmitter, or do you test two wifi boxes as WAP and client, checking BERT vs signal level. If it is the later, then there is the situation where you are doing BERT with your box talking to your box, then your box talking to another box, etc.
Why not use a power meter?
Technically, what a chip manufacturer insures with a datasheet is that if you find the part does not meet the specifications, they will refund your purchase or give you a new chip. Now the parts are factory tested, but the consumer grade chips may only get one test (probably hot). The manufacturer tries to infer from one test that the part will work over temperature and power supply range.
By test, I mean a suite of tests on ATE Most manufacturers will list their test flow in the catalog/website. The flow could be 100% hot and QA cold (test a small subset), or GBD (Guaranteed By Design) for temperature. Having designed chips for GDB, I personally don't have a lot of faith in it. However, market forces lead to such testing in consumer products. Generally there is a footnote and some verbage like "Guaranteed by design. Not subject to production testing", "Guaranteed by design, characterization or correlation to other tested parameters.", etc.
The signal generator section has an 802.11b "transmitter". Actually, it's a fairly generic QAM modulator, with overpriced software 802.11 emulation.
The BER does talk to itself. However, there's no loop as the pseudo random data generator is independent of the data receiver. It has to be in the same box because there's no other way to detect any errors without known what was sent.
The output of the BER random data generator goes to a different piece of equipment that acts as a BER tester. The BER sends data to the signal generator. The modulator does the QAM thing. The RF oscillator acts as a transmitter. The access point receives the RF through a coax cable. The output is ethernet, which goes to an ethernet to serial converter, and then to the BER tester. The tester compares the sent data, with the received data, and counts the number of discrepancies (errors).
Because I blew up the thermistor head for the HP 435A power meter. We were trying to figure out if the spectrum analyzer was accurate enough to substitute until the head came back from repair and calibration.
It's difficult to separate failure to meet component specifications on a production board. The RF chipset may be just fine, but digital noise on the board could ruin the sensitivity. One reason that many manufacturers use the chipset reference design and layout is so that they can point the finger at the vendor if something doesn't work.
Yeah, blowing up the sensor is a real life hazard. Any funny thing is, nobody knows how it happened. I like, er, just went to use the power meter and it didn't work. [The Tek current probe failure is another lab mystery. Well, not really a mystery as if you drop it, it stops working. Just noboy remembers dropping it.]
Technically, the HP meter is a bolometer. I'll admit I never saw the word until I was reading the manual on a radio I acquired, which had a spot to connect the bolometer.
Nope. I know exactly who blew it up and how. It was me screwing around with some ham radio equipment. For a 25 watt xmitter, I usually use a -30dB 100watt load/attenuator between the xmitter and the power meter. That will deliver +14dBm to the HP435A which is well within its operating range. Unfortunately, the lab had a spare "dead" attenuator that should have been either fixed or tossed long ago. Although labeled -30dB, it's really more like -10dB. That will deliver +34dBm (3 watts) to the thermistor, which will kill it. Since the good attenuator was being used, I grabbed the wrong attenuator. Don't ask what this cost me.
Fingerprint the probe. If you're clever, you can tell the age of the fingerprint and who handled it last.
Remember: Blame must be assigned before any problem can be solved.
Strictly speaking, a thermistor is a type of bolometer, which is the general term for an EM energy sensor. For the HP meter head, either is correct.