1
\$\begingroup\$

I'm a systems engineering student and I have taken a summer job at a company that manufactures various microwave filters, my role is to "try" improve and standardise their processes in order to improve production reliability. As the company manufactures over 13,000 filters, it is common for an experienced tuner to show a less experienced tuner how to tune a filter that they have no experience with and thus diverting the more experienced tuners time from the more trickier tasks.

I have read a few papers that use vector fitting and aggressive space mapping (VF-ASM) to tune a general two-port cross-coupled network A FAST TUNING METHOD FOR MICROWAVE FILTER USING VF-ASM TECHNOLOGY, I believe this is concerned with reducing time taken by a human tuner and not something that I can use. However, I am wondering if algorithms exist to tune any filter that will enable me to create a process that can be followed to reduce the amount of time being diverted by the more experienced guys.

I have no experience with tuning filters so if my question seems crazy to some of you I apologise, I would appreciate any tips, advice or direction you can give me.

\$\endgroup\$
6
  • 2
    \$\begingroup\$ It depends on the type of filter. For instance, a coupled resonator bandpass filter where the resonant frequencies are tunable, but the coupling is fixed, can be tuned by Dishal's method. Other topologies may have other methods. \$\endgroup\$
    – Neil_UK
    Jun 30, 2019 at 18:27
  • 1
    \$\begingroup\$ huh. If someone builds a machine to turn all the knobs automatically, this would actually be something that machine learning could be quite nice at solving (but it might take a long time to train per filter): The target function is the frequency response deviation, you have a strongly nonlinear but smooth function of tuning screw rotation effecting the frequency response... \$\endgroup\$ Jun 30, 2019 at 19:35
  • \$\begingroup\$ I could easily show you on a Falstad Simulator when I get I time ( BD party BBQ soon) with sliders for each reactor say using a 10 pole Chebychev filter at 2.4 Ghz and then show you how it’s done and you can interact with the inputs to see the output on your browser. \$\endgroup\$ Jun 30, 2019 at 19:57
  • \$\begingroup\$ you are asking how, for an arbitrary-shaped-search-space, to quickly find the steepest-decent path to the global optimum, yet NOT enter an false optima. Genetic algorithms, per a guy I assist in system development, can handle this search task. \$\endgroup\$ Jun 30, 2019 at 21:48
  • \$\begingroup\$ FYI: Keysight Technologies Network Analysis Solution Advanced Filter Tuning Using Time Domain Analysis: literature.cdn.keysight.com/litweb/pdf/5980-2785EN.pdf \$\endgroup\$
    – tomnexus
    Jul 1, 2019 at 5:21

1 Answer 1

0
\$\begingroup\$

You are asking how, for an arbitrary-shaped-search-space, to quickly find the steepest-decent path to the global optimum, yet NOT enter an false optima.

Genetic algorithms, per a guy I assist in system development, can handle this search task.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.