repairing an ALAN Industries 100dB attenuator

[worsthorse]

I recently picked up a small collection of test equipment that included a heavy duty switch attenuator from Alan Industries. There's no documentation to be found for it but from the labeling it is probably mil or aeronautical spec gear:



If you look closely, you can see that two of the switches are broken (the first and third from the right).  But the price was right (free) and I figured it was worth trying to fix it. When I got it home, I did a simple continuity check and, as expected, it was open between the input and output. 

Taking it apart was interesting.  There are twelve screws on the bottom, each of the switches has two nuts, and there is that weird row of screws.



I take the twelve screws out and though I can see a seam, the bottom isn't coming off as I had hoped. So I remove the nuts on the switches, thinking maybe they are attached to the bottom plate (I know, dumb idea, anyway that's what I did). Still no joy.  I look at the screws on the side, and thankfully, decide they can't be holding the base plate on so I leave them alone.  One of my friends on the TEA thread suggests jimmying it apart, a prospect I don't look forward to but am resigned to trying.

So I put the unit in a vise, put a shim on the seam, tapped it and the bottom came free in my hand. That was a lot easier than I thought it would be. It was glued in place and it was my good fortune that the glue had weakened over the years:



The body is cast aluminum (I think). That row of screws? They serve as soldering points for the grounded sides of the resistor pi network at each step. I am glad I didn't muck with them. 

A closer look at the switch arrangement:

     

The photos aren't great and I will post a couple when I start the next round of work. The attenuator uses double pole double throw switches. The top poles are joined by a copper plate, the bottom poles are attached to a resistive pi network, and the middle poles attach to the switch before and after, again with copper plate. The two wire measurement from end to end was less than 0.2 ohms.  The small set screws serve a solder points for the pi network. The resistors used are almost all 0.1% tolerance. The thing is built like a tank.  And good news... this is a 50 ohm attenuator.



More good news... after I bridged the two broken switches, the input and output were connected. I tested it up to 100MHz and there was no discernible loss (just using an oscilloscope) from input to output with the switches in the 0dB position.  I checked the resistance one step at a time on the good switches and all was as expected. The working switches all work.  Another quick check demonstrated that the switched-in positions for each working stage insert the expected loss.

That means if (a) I can remove the sections with the broken switches and (b) find replacement switches and (c) the resistive networks on those two sections aren't damaged and (d) I can put it back together, I will have a pretty nice attenuator for my bench.

I think I see a way to remove the switches without damaging the networks but it is late and I will leave it until tomorrow.

[worsthorse]

Tonight's task was to remove the two broken switches, hopefully without damaging the resistor networks attached to them.  The first switch was attached to the coax connector. Given the tight spacing, I unsoldered the copper plate on the switch side and removed the coax connector before trying to unsolder the two resistors from the ground posts and the copper plate from the second switch. Here's what it looked like when I started:



Unsoldering the copper plate wasn't too difficult. I did need to insert a blade between the switch lug and the copper plate to keep them separated once I got the solder melted as there was no way to get it all out from between the plate and lug with solder wick.  Here's the coax connector:



Getting the resistor legs free of the ground posts - they are tucked behind it and soldered on the side - proved to be a problem and the leg from one of the resistors broke free from the switch:



The second broken switch came out a little more easily, with the network intact:



Like I said, this thing is built like a tank. There is a lot of mass to it:



While I planned to use the existing resistor networks I certainly have to rebuild the first of the two. I may just order the parts to do both of them from scratch.  Luckily the switch is a C&K part that is still available though I may not be able to find the long toggle version. Now that I know what I have, time to order parts. Then I will move the copper plates from the old switches to the new ones, move (or rebuild) the resistor networks, and install the new sections.

[worsthorse]

Found a source for replacement switches.  I should have them next week.

0.1% axial precision resistors are a taller order, especially for the low values. The pi networks for the 1dB and 3dB sections are made up of 5.7 | 870 ohms and 17.6 | 292 ohms respectively.  After searching (thanks to the folks that replied to my post over on the metrology thread) for replacements it appears that the low values are as good as unobtanium, so I am going to sort through the pile of 5.6 and 18 ohm 1% resistors I have and pick the closest matches.  Digi-key has 294 and 866 ohm 1% resistors, which are very close to what I need, so I will order ten of each and pick the closest matches for those, too.   

I considered trying to parallel resistors to get the values I need but there's not really enough room to do that. Easier to go the selected part route. Will post again when I have parts in hand.

[capt bullshot]

Nice and thanks for the teardown / repair pictures.

Wonder what bandwidth this attenuator achieves,  since the switches are quite standard ones (except for their long lever), and the resistors look like standard metal film ones. Otherwise, someone put some effort into providing a 50 Ohms path through it, at least it looks that way.

For example, this is the internals of a 2.7GHz attenuator:

[worsthorse]

It is supposed to be DC to 1 GHz but yes, the C+K switches cost a few bucks but aren't aren't anything special (from their 7000 series). The resistors are appear to be 0.1% tolerance (and might be similar to the Vishay line designed for RF) but it uses standard BNCs and hand soldered copper plate connections. It isn't nearly as engineered as the 2.7 GHz attenuator you shared. 

I found another one on ebay and got a good deal on it, so there's another one coming my way. If it works I will check it out with a spectrum analyzer. Maybe before I replace the two damaged sections I will figure out how to put this one on the SA, too.

[Tryer]

Hi,
The typical RF-attenuator is built as the sketch below.
Each cell has 3 resistors. The two verticals (shunts) have the same value.
For 50 Ohm (most usual) and 10 dB the values are R1= 71.2 Ω, R2=R3=96.3 Ω
For 50 Ohm and 20 dB the values are R1=247.5 Ω, R2=R3=61.1 Ω

Online calculator e.g.
https://chemandy.com/calculators/matching-pi-attenuator-calculator.htm

What looks strange for me is that you have 4 resistors but also the strange values.
When a cell is overloaded (either  due to high power RF or DC) the values of the resistors change.
I could not read how many dB's that cell had. Can you tell that?
And the fourth resistor? Can you post a sketch of how it is connected?
And you have surely more identical cells. You can verify the values by measuring them. (no need to desolder them,
just calculate the composite resistor and verify).
And do not get fool with 0.1% resistors.
1% is more that enough. That would give you an attenuation error of much less than 0.1 dB
In RF it is something different compared to DC. The dominant effects are different (reflections (missmatch), crosstalk a.o.)
wish you success

[worsthorse]

The second attenuator arrived and has two broken switches.  Luckily they are different switches. It is also pretty beaten up, compared to the one I already had, so I went to plan B: removing the 1dB and 3dB switches from the second attenuator and putting them in the first one.



Getting the switches out went better this time. Not surprising since I've removed four of them at this point. The resistor networks on both are still good and the switches work pretty well. The next step is to install them and test the re-assembled unit.