Making a continuously variable attenuator, is it possible?

I have been puzzled for a while about existing variable attenuators. Affordable types are mostly stepped, if you want them continuously variable they seem to be more expensive. Stepped types are easy to make with some switches or a rotary switch and matching resistors, but how about making a continuously variable attenuator? Until now I have used a potentiometer connecting the input across the potentiometer and the output to the wiper, but you will have severe mismatch to the circuit either at the input or output or both!

Attenuators matched to 50 Ohms will have a T- or PI-type configuration. The configuration we use is a bridged-T attenuator, which seems to be the only type suitable for continuous operation. Because we want a decibel scale, a logarithmic potentiometer should be able to do the job.
bridged-T variable attenuator
bridged-T continuously variable attenuator

R1 and R2 have a 50 Ohm value. The potentiometer is a 2-gang logarithmic type which is connected in such a way that while one section increases in value, the other section decreases. The maximum attenuation you get depends on the potentiometer value, being about 30 dB for 1 kilo-Ohm. This was also the smallest value I could get from the shop.

So what can we expect? These type of potentiometers are intended for low frequency use, how will they behave at higher frequencies? But there is something else: are they really logarithmic?

logarithmic potentiometer setup and measurement
setup and measurement of logarithmic potentiometer values

As you can see from the graph: relationship between scale (0-10) and resistance value is not truly logarithmic. To reduce costs, these type of potentiometers are built with 2 different values of resistance with linear behavior, which produce a logarithmic approximation. In fact, we have 2 straight and combined lines in one graph. Next, what kind of losses can we expect?

attenuation and VSWR versus scale
attenuation (left) and VSWR (right) versus scale

Measurements were done at a frequency of 10 MHz. As you can see from the left graph, there is a nice relationship between attenuation and scale up to 90%, after that the value increases rapidly. Insertion loss measured is -0.36 dB, which is about 8%. The right graph shows a maximum VSWR above 2.2, another loss in power of 13%.

In conclusion, this type of attenuator is not intended for accurate setups, but it might still be useful at driver level or at the input of a receiver where loss is not critical and we have plenty of signal. Losses are caused by the potentiometer design: they are not suited for high frequencies and not exactly logarithmic. Thats why good variable attenuators are so expensive: they are difficult to make!

I have used an ancient Hewlett Packard (now Keysight) 8753A network analyzer to do my measurements.