Microstrip Filter Tolerances

[rf997]

I designed a bandstop filter with a bandwidth of 60 MHz at a frequency of 1950 MHz. However, the filter is quite sensitive. Even when I enter the tolerances as 1%, the insertion loss shifts left and right by 50 Mhz. The production technology I use is not good. Is it possible to stabilize the filter?

 

[BigBoss]

Therefore manufacturing technology is important.
Another option is to use "adjustable filter" topology. But this type cannot be realized for Microstrip Planar Filter. They can be implemented with Cavity 3D Filter Topology.
But it's hard to simulate and manufacture.

 

[rf997]

Therefore manufacturing technology is important.
Another option is to use "adjustable filter" topology. But this type cannot be realized for Microstrip Planar Filter. They can be implemented with Cavity 3D Filter Topology.
But it's hard to simulate and manufacture.

Can the filter that meets these demands be produced with a defected ground structure topology?

 

[volker@muehlhaus]

Even when I enter the tolerances as 1%

I would check what the physical error sources are.

Etching precision issue with gaps in coupled line: You can increase substrate height, so that for same line impedance the absolute dimensions for line width and gap width increase.

Etching precidion issues with resonator length: You can use a lower permittivity, for longer resonator dimensions.

Shift from substrate permittivity tolerances: you obviously need to use a substrate with less tolerances.

 

[BigBoss]

Can the filter that meets these demands be produced with a defected ground structure topology?

I don't know the answer.
But don't forget that the Microstrip Line Models are approximated in schematic and you have to implement this filter on a substrate and simulate with EM simulator.
If you want to get a stable and consistant result, you should work with tight tolerance substrates. FR-4 or similar cheapo substrates make you sick.
To my experiences, a good substrate for these kind of applications is Rogers 5008. But it's salty..
3800 USD /square-meter (double sided, 0.762mm height, Gold Plated ) in China.

 

[mtwieg]

I designed a bandstop filter with a bandwidth of 60 MHz at a frequency of 1950 MHz. However, the filter is quite sensitive. Even when I enter the tolerances as 1%, the insertion loss shifts left and right by 50 Mhz. The production technology I use is not good. Is it possible to stabilize the filter?

When you say "tolerances as 1%", what do you mean specifically? Does it mean each component can vary independently of the others? That would be unrealistic. For example the Z0 of a line might vary by 5% from board to board due to substrate variability (especially if they're from different fab lots), but the variation between lines on the same layout should be much smaller.

 

[D.A.(Tony)Stewart]

When you have a -3dB BW that is 3% of center made from 4 notches each narrower than this, you must consider all that it takes to reduce your conductor tolerances and dielectric tolerances by considerably more. Consider ceramic.

 

[Andrew77]

In this case you should introduce some fine tuning and adjust the filter by a workbench

 

[biff44]

It IS a real issue. the substrate you use has a dielectric constant that is specified IN A RANGE of values. There are also etch tolerances one has to live with.

For truly critical filter designs, i have seen people come up with perhaps four different mask sets (each with elements a little longer than the previous one). They make sure all the substrates are from the SAME MANUFACTURING LOT. They print on a filter on one substrate, and if the corner frequency is off, they chose another of the mask sets. When they finally find the correct mask set for that one lot of substrates, they then manufacture the entire production run.

It was explained to me once this way: "Kyocera takes a kiln the size of a train car, puts it into the oven, and fires all the substrates at the same time. Some get more heat, some get less heat. They throw out the ones that are way off for dimension or dielectric constant. but they sell you the rest!"

A better method is to do a proper SYSTEM DESIGN so you do not rely on such high tolerance filters. Maybe you can make up for variations in the IF section of a receiver with RF filters, for instance. Or at baseband using FFT filtering. One might need to overdesign compression points and IP3 in a reciever to allow for such spread out filtering to work.