Theory of Operation

Before delving into the troubleshooting and repairs let's first lay the theory of operation for the device, say that you'd like to drop a fixed amount of voltage between your input and output, your first approximation might be to employ something like a voltage divider

However this simple solution comes with it's own set of problems, for instance you might notice that the output and input impedances are not matched

This might prove problematic if we want to attenuate a HF signal where cable impedance plays a big role, thus giving rise to an additional loss in power due to mismatched impedances and skewing our desired attenuation. One quick fix would be to set an additional resistor in series with the output such that equal impedances are seen from both sides

If we match both resistors at the input and output we would have achieved our goal of equal output and input impedance, say that we use 600Ω coax at our lab, we could employ a set of values R1, R2 such that

We can make further improvements on this design though, say that we want to attenuate our signal to a really low level (the HP4436A might produce outputs as low as 10µV for a 1V input at the highest attenuation!) we should then concern ourselves with signal quality, very few amounts of input noise could completely throw our Signal to Noise Ratio. One way of fighting coupled noise is to implement a balanced line, a balanced line carries 2 signals, one inverted and one normal, this means that the T-pad attenuator must be duplicated but with the added benefit that any common noise that might couple to both lines will be eliminated at the measuring end when the inverted line output is inverted once more and summed to the non inverted line. The reasoning behind this is signals at the end are twice inverted before summing them with their non inverted form which yields a coherent sum, however noise coupled during signal transfer is only inverted once before summing it against itself which effectively cancels it.

This kind of configuration is refered to as an H-pad attenuator and can be seen below, we can verify that by halving the resistances and duplicating the circuit we still preserve our input and output impedances. Just by looking at the circuit however we can see that the circuit accurate behaviour relies strongly in how matched the resistors are between each other. A total of 6 resistors need to be matched for it to work, this will be relevant later. 

Our last modification will consist in making this attenuation tunable, say that we find a set of R1 and R2 resistors that drop 100dB, we could build a different attenuator with resistances R3 and R4 such that it's output and input impedances are the same (600Ω) but it's attenuation is only 0.1dB. Setting them both in series would result in a total system gain of -100.1dB. We can finally choose whether to drop 100 or 100.1dB through a set of bypassing switches like so:

This is in essence how the device operates, a network of 12 H-pad attenuators in series (4 per decade) can be bypassed or not to configure a certain attenuation with great precision and noise immunity. If you wish to better understand the theory of operation for this device you may play with the LTSpice schematic I built attached on the files section for this project.


The Mechanical Problem

As we've explained in the previous section there are many switches to operate to set each decade of attenuation, 16 to be precise, we should then have a way to link a certain digit in our decade to 16 on/off positions. The mechanism behind this is mechanical, each rotation of the decade dial would rotate in turn a shaft with multiple cams or lobes positioned along its length. These cams engage with a series of levers, which in turn actuate the switches by sliding in or out a dielectric between two contact depending on the axial (or rotational) position of the shaft. 

Below...

Read more »