Notes Project Output Archive Links
Index table
+Intro
+Prephase
+TimePlan
+Outline
+Research Phase
+-28 Feb.-
+-5 Mar.-
+-12 Mar.-
+-Odysee
+-19 Mar.-
+-26 Mar.-
+-2 Apr.-
+Develope Tests Spec.
+- 9 Apr.-
+Test Phase
+- 16 Apr.-
+- > 23 Apr.-
+- 30 Apr.-
+Review Results
+- 7 May.-
+Prep. Report
+- 14 May.-
+- 21 May.-
+After Project
 The easter break and birthdays at home - needed the break, beside this I finished nearly the first part of the report. At the end of the week (return to Dublin), I done the first real test - found, that the range indoor is really bad. I rethought this and agree, that the purpose of this equipment is for one room only, the idea is to make several small cells instead of one big - would be also better for the capacity. As handover and interlinking is possible it still can be set up, that a continues network connection is hold, while moving through the house.

More Thoughts about the first test

It is possible, while beeing assoziated to move further away, than the place were a reassoziation can take place. I believe this is caused by the employed thresholds, which are for deassoziation and assoziation different, to guarantee a safe handover.
A test to size the actual distance possible to cover (with clear line of sight). I went first to the back parking place and then to the park (far end wall, about 200 meter). The conditions were good - it was possible to use the connection to remote control the server PC (this was not possible during a RTS threshold test...see belowe). However, the test data showed high delays and the ping even higher delays than usual. (Usually the wireless link has around 10 or 20 ms delay, which depends one a retransmission or not. In the park delays of 30,40 and even 60 were possible. I connect this with the fact, that I needed to connect the long cable on the access point, which needed to stay at the window. This means also, that I needed to route the test traffic wire the switch in 407, which also delivers other traffic (namely the IPX File and Print Server traffic) to the access point. As IPX will be transperently forwarded, while TCP/IP will be only forwarded if the destination is on the other side of the wireless link, the increased traffic explains the longer (and not only full times of 20ms) delay.
I planed for the next week tests at 3 places on floor 4. Directly in the room 407 next to the access point (corresponding to good conditions, -47dBm), opposite the elevator (corresponding to normal conditions, -65dBm) and at the last row of 405 (corresponding to bad conditions, -80dBm).

Thoughts about other tests

As my supervisor would like to know about the capacity, coverage and quality. I thought about ways to test this and more important to present the data in a good way. First idea was a 3d plane diagram, with the 3 important parameters on the axes. This would need to define a threshold, at which time a connection can be seen as usable and at which not. Then it would be possible to set a fixed distance, chose a quality (by using different packet sizes) and increase the number of simultaneous channels. If at 5 channels all values inside the thresholds a point could be made in the diagram, if at 6 channels one of them didn t have the needed bandwidth the point stays at 5, otherwise it can move up to 6. This can draw a plane, which by theory would be falling from lowest distance and lowest quality (as capacity would be the z axes) toward the outer regions.
I done 2 small programs to test how far it is easy possible to let this plane generate automaticall out of the test results, but as I found, that even defining the thresholds is quite tricky, I refrain from using this diagram type. Another point was, that this would need to increase the number of clients continously, which would mean if a test takes 3 minutes for each point in the distance-quality frame I would need about 3 to say 18 times 3 minutes, which means 45 minutes. Taken, that the distance are changed in steps of 2 meter (or maybe 5 dBm), which is around 10 tests. Also the quality should have at lest 5 different steps. That would make all together 5*10*45 minutes=2250 minutes or if assumed 8 hours per day nearly 5 days testing time.
© and created by Jörg Abendroth