signal strength, link quality; interference

try Netstumbler - if it supports your cards chipset - has a nice moving graph for each SSID that it detects.

Agreed. However, the current version of Netstumbler (0.4.0) does not exactly work with Windoze 98se. It works only if you have the original Hermes chipset based Orinoco Gold or Silver PCMCIA card, also known as Orinoco "classic". However, the previous version (0.3.30) works just fine but has a few bugs that were fixed in 0.4.0. Avoid

0.3.23.

Since you're into antique hardware and operating systems, this might be of interest:

You can use signal strength to aim the antenna. The response time on most wireless manager graphs aren't really fast enough to do it properly, but with patience, it can be done. The problem is that the wide variations in signal strength you're observing is for real.

2.4GHz bounces around quite a bit. You're seeing cancelations and reinforcements of multiple paths between endpoints. Just pointing the antenna in some random direction, and walking around nearby, will yield substantial variations. In an uncontrolled environment, this is what you'll get. If you had an RF anechoic chamber, it will be far more stable.

Just to make life difficult, you'll find that the RF output of a Wi-Fi xmitter various somewhat between 1Mbits/sec 802.11b thru 54Mbits/sec

802.11g. Additional modes will also have different power levels. You might want to lock your xmit mode to some favorite speed before the variations drive you nuts.

Interference definately shows up in "link quality". The problem is that chipsets have different ways to measure link quality. Some use the noise level between transmission to calculate SNR (signal to noise ratio). Others use the bit error rate to estimate the SNR. I think there are at least two other methods in use. All of them will show the effects of interference.

For testing antenna, you need a (drum roll) antenna test range, a signal source, and a reference antenna with a known gain. It doesn't have to be fancy, but it should make an effort to avoid interference and reflections. I have a convenient WISP system located on a mountain top approximately 5 miles away. I'm on a hill side, surrounded by trees. The trees block most reflections and sources of interference. The valley in between prevents ground bounce from becoming a problem. I can point a directional antenna at the WISP system, and get a fairly stable signal. Even so, I like to plot the signal levels with a data logger PC, throw out the extremes, and use the average values. I always use the reference antenna to remove the effects of calibration, drift, and coax cables.

I *enormously* respect Jeff's views and this is not intended to contradict them.

One approach that I have used successfully to engineer a few wireless networks is to test the link with fping.exe.

This allows you to sent pings at the fastest rate possible (i.e. send the next one as soon as you have a reply to the last one) and provides a sensitive indicator of good communications. It is best in my view to specify long (say 1400 byte) packets. I cannot at the moment recall the exact parameters that I like but post back if you require further assistance with it.

Thanks. However, don't over rate my competence. I do make mistakes.

Please note that the original question was for testing antennas, not testing the link. There's a difference. With antenna testing, you're really interested in:

1. The antenna pattern, both vertical and horizontal. Example:
2. VSWR plot versus frequency (or usable bandwidth). Same example. Ping isn't going to show those.

Yep. fping is good for testing links. Under idea conditions, what you're looking for is a stable and consistent latency for every ping, with no timeouts. If you've got any manner of link impairment, you're going to see variations and timeouts. These represent retransmissions, retries, and in extreme cases, packet loss. In effect, you're measuring the quality and stability of the link with fping. Unfortunately, fping does not yield a quantified measure of quality and stability.

For link testing, I prefer iperf and jperf. Jperf (iperf with a Java front end) is generally easier to use and operate: (tutorial)

These program flood the wireless pipe with as much traffic as it can handle. Wireless slows down when it hits errors, interference, reflection, impairments, etc, which will be indicated by the thruput benchmarks. If thruput is slow, there's something wrong. It can also be used to test maximum thruput through a router, which is sometimes a limiting factor on cable modem and FIOS high speed connections.