Hi
I designed an microstrip edge coupled line filter at 5 GHz ,the bandwidth is 300 MHz,with 5 section on RT/duroid 5880, I used filter wizard in Microwave office and then tuned the dimention(lengths, widths and spacing )
it seems that my filter is sensitive for small variation in the dimientions of the filter
my qustion is how to design the filter to be less sensitive to the etching tolerances
Thanks
1. Make the bandwidth wider so that your desired frequency range is within the passband at the limits of etching errors.
2. Use a lower dielectric constant material. This will increase the dimensions and the etching errors will be a smaller fraction of the dimensions.
3. Try some other filter topology that might be less sensitive to the dimensions. I cannot think of any, but other board members may know of them.
4. Use a more expensive fabrication method that has lower etching tolerances or use thinner metal (which will decrease the chemical etching time and undercutting) and then plate the thickness up.
You have one fore important Option apart from flatulent mentioned
complete the design by allowing production tolerances on
the track and the board material, and analyse using the Yield analysis...
Then finally get the Component Sensitivity Histogram plots..
use the Yield optimizer (Design center the parameter) to make less sensitive...
Sensitivity analysis is obtained by adding a Ysens measurement to a graph.
For an example go through the AWR installation folder...
AWR\AWR2004\Examples\Circuit Features\Yield Analysis\Ku_Band_ECM.emp
Sometimes you can increase the gap spacings by screwing around with the filters in/out impedance. Use a λ/4 transformer before the filter, and another one after the filter.
Thank you all for your ideas,
another thing, when I measured the filter the lower passband has more insertion loss than the higher passband,eventhough the return loss still good "below -15dB"
is that normal for this kind of filters,
Attached is an example on Coupled Line Filter Synthesis and Analysis using AWR Microwave Office...
WHich shows a typical example on Simulated vs Measured response...
I think your results looks OK!...