If you think your link is odd, consider this image for a 1.5km city link with perfect visual LOS:
As you can see, it regularly goes from between -50 to -70.
Hello,
Please tell me where do You generate that nice signal graphs?
I use mojiro’s cacti template from http://forum.mikrotik.com/t/cacti-mikrotik-routeros-statistics-update-11-19-2013/18506/1
RK, thanx…
Maybe try looking at the problem upside down, in so far as, your signal of -80 was the norm (all be it less than expected)
And that the periods of -60 are from times of signal enhancement.
The terrain looks like a candidate for temperature inversions. You can spot the phenonema crudely when smoke from a stack ceases to rise but drifts horizontally.
Temperature Inversion also preludes ducting another phenonema affecting VHF, UHF and less commonly Microwaves.
Maybe it was raining somewhere in the path those days, or there was high humity, causing refraction and then multipath.
What about a car that was parked somewhere in that link, that may have caused a mulipath affect, causing the signals arriving partly in phase to add.
Anyhow, interesting thread.
Best of luck… Simon
http://www.ips.gov.au/IPSHosted/NCRS/wars/wars2002/proceedings/comm-f/screen/kerans1.pdf
http://sirkova.com/documents/ICEST_2003_Short_Distances.pdf
http://www.its.bldrdoc.gov/pub/ntia-rpt/79-24/79-24_chap1-3.pdf
http://www.dxinfocentre.com/propagation/tr-modes.htm
Some of the above links are tiring reading, especially the third. For 30 years I played around with VHF communications, exploiting the uses of all known propagation phenonema, including meteor scatter and moonbounce.
Maybe as I suggested, its not equipment related in any way, just that the freznel zone has given you less than your first expectations, and on the days of 20dB extra, your signal is actually enhanced.
The last link is the simplest explanation from us radio ham chappies
73’s EA5/G0MGA
Those power lines in the image you posted can also create anomalies.
They will also expand and contract with temp fluctuations, etc.
My beleif is that there is not one or two things, but several things, all working against you on this shot.
THE MACHINE IS AGAINST YOU!
biger dishes - beter signal 
simle as that … with two 90cm solid state, and some LOS … there is no way of having any problems on 20km
have you tryed moving your link o the front of the roof so that you cut out the noise on the roof?
so your 20km link becomes 19.7km
i always try keep my anntennas close
and another thing whay power are you using ups power and 12v or 48?
just a question i did not see any answers to
Just an update. A few weeks ago I placed at the data center side a pacwireless 32db solid dish, single polarity as a test to see what happens. This increased the signal to -66 or so, a huge improvement. I also went up about 5 more feet. This weekend I am installing on both ends the pacwireless 32db dual polarity solids ($49 each on ebay!) and hope to be in the low 60’s. Will post pictures and other info shortly. Thanks for the help everyone!
Nice. Very nice.
They got more?
nope, the guy said he had 3. they were listed as satellite dishes, in the wrong category. he didn’t know they were $280+ new 
these 3 were all new. cost me $300 total with shipping for 3 of them, yeee haw. The surplus seller said he got them from a lost shipping palette, so someone here is probably missing them : )
Just to cast my vote, the effects you are seeing sound just like what I have on my long links, which is the thermal ducting that was mentioned earlier. I see it most after a hot day, late at night when things start cooling off, but usually only when there is almost no wind.
The effects of thermal ducting can help or hurt you depending on which “layer” you are in. Think of it this way, when the layers develop, if both of your antennas are within the same layer, your signal gets “stuck” in the layer and it helps increase the signal since less of it is getting lost in the fresnel zone. If your antennas happen to fall into different layers, it can hurt your signal as much as it helped.
I can watch my 35 mile 5ghz link. Normally it hovers in the -75 area. Late at night, I can watch it go into the -90’s and start to disconnect… In a matter of minutes, it can be back to the -70’s, and actually increase into the -50’s. As the layers disappear, the signal levels back out around -75.
Happens like clockwork. About the only solutions I know of are to throw more signal at it (bigger dishes) or add a repeater in the middle.
Joe
This has been a very long thread. Please update us with your new antennas. I am sitting on the edge of my chair waiting to see a resolution…
In my experience, fixing a very difficult problem will teach you more than you ever expected to learn. That is the saving grace of these kinds of problems. Everyone following this will benefit from your frustrations.
If your Fresnel is really 0.3F1, then in my opinion that is your problem.
It is my opinon that for a reliable link you should have 0.8F1. I consider (again my opinion) 0.6F1 to be the absolute minimum.
Given your description, I would both increase the height of the antenna at the data center and also if possible at the water tower.
Also if possible move the antenna at the data center so that you are not pointed across the metal roof, I have had many problems in the field going over metal roofs in that manner.
I have replaced both ends with PacWireless 32db dual polarity dishes. I also increased height at the data center by about 4-5 feet, and on the water tower about 2-3 feet. Signal went from -80’s to -50’s. I have a feeling those PacWireless 29db grids aren’t really 29db. Going to the data center over the next week to flush out installation and will take more pictures and post the final solution. Will be testing nstreme2 today with the dual polarity dishes, hopefully there is enough isolation to not cause me problems. BUT - WOOO HOOO, finally a link that is working. I’m hoping to get 80-120mbps from the data center into the wireless mesh, could be a pipe dream : )
Oh yeah, when ducting happens the signal goes down into the -40s now…
Thanks everyone for the help. I really appreciate it.
I have used plenty of the pac 29’s (probably 50 of 'em), they are perfectly good antennas. Improving your fresnel is what actually helped.
The importance of good fresnel cannot possibly be understated. It is absolutely the single most important property of any link!
I vote for sticky 
This thread is very helpfull and usefull.
Assuming different antennas but exact same alignment and heights, that improves fresnel because of the degree of beamwidth correct? Going from 4 degree to 3 degree increases fresnel simply because its a narrower path?
Now I have the problem of RB433 w/ 2 radios and getting half speed when the other associates, but I’ll keep that to the original thread if I can find it. Pictures coming shortly on the link…