That methodology is what is in dispute. I was hoping for another authoritative source to back it up.
I've seen that formula for adding reflection coefficients and converting to return loss before, outside the Andrew catalog. It shows up in some application notes from HP and, of all places, Aeroflex. But I don't have time to search and dig up links.
I think the fact that some sites measure fine and some don't is the root of the suspicion. They don't seem to understand the meaning of "worst case" performance, or just do not want to believe the results.
The sites that measure ok... are they all identical?
You left out the insertion loss of the waveguide feedline, but I made an educated guess, then ran the numbers, per the formula on page 228 of the Andrew Catalog. The number I arrived at is... wait for it... 14 db.
We were surprised as well, until we found the Andrew design guide which backs up the result. I think the fact that some sites measure fine and some don't is the root of the suspicion. They don't seem to understand the meaning of "worst case".
What you need to do is convert the formula to a spread sheet, and run some "what if" scenarios. Remove a couple of sections of rigid, or the sampling port. That right there is a sizable hit to RL. See what you can do to redesign the waveguide system to reduce the number of bumps and mismatches. More on that later... *
Can you elaborate on this?
You mention that you swept 5.9-7.1 GHz, but you didn't specify your actual operating frequency, or the specific type of waveguide. So, I'll assume you're using EW63 eliptical, and some rigid equivalent. If your operating frequency is actually 5.8 or 5.9, for example, and you're close to cutoff, any bends or joints or twists can, for a short distance, increase the cutoff frequency to something above your operating frequency. You'd see that as a lower RL, and additional insertion loss. The same can happen if you're too close to the upper frequency limit. Discontinuities inside the waveguide can excite higher order modes, and have the same end result.
So, make sure that frequency you're operating at is as close to the center of the waveguide's frequency range as you can get.
Thanks for the response. We are using non-tunable elliptical connectors which measure great (see below). We are calibrating out the waveguide transitions during measurements. The tester (Anritsu 820D) we are using has measurement uncertainty on the order of 3dB based on the directivity of the coupler in the instrument.
Sounds like you're doing everything right, although my first recommendation would be to only used tuned connectors.
According to the Andrew design guide, switching our rigid sections to 1.02 VSWR (i.e. narrowband components) should provide a system worst case of 18dB return loss.
That's still pretty dismal for a waveguide system. I guess the system engineer should be able to specify an acceptable figure. My worst case spec was always 23 db RL for a 30 MHz bw digital radio. Analog was even more stringent. Multiline routes with 1:n protection need a much better RL than that.
There are various reasons that drove us to using the rigid waveguide.
For one, it takes up less space inside the shelter than elliptical, and you can use tighter bends. What they don't tell you in the catalog is that those tighter bends cost you in terms RL. So do all the flanges bolting those rigid section together.
The system degredation was unexpected and we are trying to ensure we understand why now, before spending more money to address the issue. Thanks again for any ideas.
I'm pretty confident that, by now, we have identified the cause of the degradation. It's all those little things adding up.
* The "more" from above...
What to do? You have 5 sections of rigid bolted together, along with a sample port, a pressure window, and a piece of flexiguide. Every flange bolted together is a major cause of reflections inside the waveguide. It wants to be glassy smooth, and every joint is a physical bump. You have two choices... reduce the bumps, or tune them out. Reducing the bumps would entail doing something other than bolting 5 chunks of waveguide together. One long one? Are there any bends?
You could try loosening the joints one at a time, and aligning them, and retightening, and see of you can tune the bumps out that way. Or, you could put a tuning section (like in the tuned connectors) and tune the bumps out that way. See Andrew or Pasternack for waveguide matching sections. Or... got a drill press? Make your own. A 3 or 5 screw tuner is easy to make. Threaded bolt holes in the center of the waveguide spaced 1/4 guide wavelength apart. Tune with a network analyzer until you have that eureka moment when all the bumps go away, and your return loss hits 30 db. They make those tuned elliptical waveguide connectors for a reason!
More than anything, though, make the formula into a spread sheet, and start playing with it and see how all those little things add up.