RF Filter Design

2019, Jun 07    


This post is about the final project for my ECE 485: Microwave design class. I was partned with one other student, Rob Koll. We designed and tested a lowpass RF filter.

Design Specifications

Design a 5 section stub-loaded lowpass filter in microstrip configuration with Chebyshev response, 0.2 dB ripple with a cutoff frequency at 2 GHz.

Design Procedure/Physical Realization

To begin the design, the following generating equations for element values were used for fifth-order Chebyshev response:


The following schematic was created using these equations.


S(2,1) was then plotted to ensure correctness of the prototype.


Next, the following Kuroda identities were used to remove inductances from the design. Then, Richard’s Transformations were used to realize open-circuited transmission line stubs in place of the capacitors in the design. To design the filter for a 50Ω reference impedance instead of 1Ω, all characteristic impedances of the lines were multiplied by 50.


Next, moving towards physical realization, the TLIN ADS components were replaced with MLIN components, designed in LineCalc for the specified lengths and characteristic impedances. For more accurate simulations, an MTEE component was added for each stub. The following simulation results for this schematic were returned.


Due to the loss tangent of the substrate used for the components, the ripple of the filter begins to suffer. When the loss tangent is removed or ideal components are used instead, the design provides the expected 0.2dB ripple. Lastly, the ADS optimizer was used to achieve slightly better ripple and attenuation. The optimizer was run with two goals: keep less than -0.5 dB drop in the pass band and have a greater than -20 dB in the stop band. Optimizations were first run with changing stub lengths, and then with changing stub widths. This produced the following simulation:

The layout for the design was then generated. For EM simulations, the substrate was defined with the same definitions as in the schematics. EM simulations were then run with a frequency plan of 0.5 GHz to 5 GHz, 10 MHz step (451 points). After EM simulations were completed, a symbol was created and the results were plotted alongside the previous simulations:


Measured Response

After fabrication, the filter was connected to a network analyzer. S(2,1) and S(1,1) were measured and imported into ADS to be compared with the previously completed simulations.

pcb image





The final EM simulations of this filter design had fairly different passband ripple than that of an ideal Chebyshev filter. This is due to the high loss tangent (0.014 at 1 GHz) of the substrate used. The design procedure was completely successful, as the schematics for each step were simulated before the design was changed. When the loss tangent of the substrate is brought down to zero in the final design, the simulated insertion and return loss of the filter appears exactly as it should.

The measured results appear similar to the EM simulation results. However, because of physical effects such as measurement connectors and substrate inconsistencies, there are differences that appear when comparing such as frequency shift and phase shift of the measured S-parameters.