APPLICATION NOTE 052. A Design Flow for Rapid and Accurate Filter Prototyping

APPLICATION NOTE 052 A Design Flow for Rapid and Accurate Filter Prototyping Introduction Filter designers for RF/microwave requirements are challenged with meeting an often-conflicting set of performance demands. Often a final RF/microwave filter design is the result of a tedious, iterative process. Fortunately, as electronic design automation (EDA) software continues to improve, new semi-automated filter design and layout procedures are now possible to ease the journey to the final product. It is the purpose of this application note to outline a robust yet flexible method for the design of lumped element filters using Modelithics Microwave Global Models, Nuhertz Technologies FilterSolutions software, and NI/AWR Design Environment software, specifically AXIEM electromagnetic (EM) simulation. The procedure involves the integrated use of three commercial software tools and separately addresses filter synthesis, working with component models, and performing accurate circuit, yield, and EM analysis on the design. Four example filters designed and tested with this procedure are presented, with measured results impressively close to the predictions in a single design pass. The Challenges of Filter Design Filter synthesis is rarely easy or automatic, even for an experienced filter designer. Starting with just basic filter types, such as low-pass, high-pass, band-pass, or band-stop filters, and then trying to achieve a particular set of performance parameters for one of these filter types can turn into an almost endless iterative design process. For designers new to filter design, the simplest and most straightforward design approach builds upon classic filter architectures; however, classical filter topologies can suffer from any number of drawbacks. These include an excessive number of inductors and the need to account for circuit element losses, component and pad parasitics, substrate effects, and interconnect effects. Failure to take any of these into account during the design process can result in poor performance in the fabricated filter and failure to meet the target performance requirements. This is especially true at high frequencies above 200 MHz, but can also be true at lower frequencies. 1 of 16

Fortunately, modern EDA tools continue to improve, enabling the creation of filter designs in which computer predictions come consistently close to the measured results when those designs are finally fabricated in microstrip, stripline, or other high-frequency circuit technology. For example, such tools as FilterSolutions from Nuhertz Technology (www.nuhertz.com), coupled with accurate component models, such as Microwave Global Models from Modelithics (www.modelithics.com), can greatly assist filter designers who need to cut time and effort from the traditional tedious and iterative filter design process. Figure 1: Screenshot of Nuhertz FilterSolutions software. FilterSolutions (Advanced Panel shown in Figure 1) contains design functions and routines aimed at helping experienced and novice filter designers alike, with a FilterQuick (FQ) learner s panel for more novice users. The software also makes it possible to semi-automate the design process and drastically reduce the time required to take a filter from an initial design to a practical layout and fabricated circuit. What makes this design software even more effective is the use of advanced measurement-based scalable equivalent circuits represented by the Microwave Global Models available from Modelithics. These models accurately account for the parasitic effects of the components and circuit elements representing the filter design. By means of collaborative efforts between Nuhertz and Modelithics, excellent wideband results have been achieved in the design, fabrication, and testing of four different filter examples. 2 of 16

Overview of Design Procedure The design process starts by establishing and entering design requirements into the FilterSolutions software, such as filter center frequency, pass-band bandwidth, and stop-band attenuation. A designer must select a filter type (band-pass, band-stop, etc.), topology, and other design options, and the software will suggest a filter design based on ideal components. A desired substrate material and the Modelithics model part families are then selected. The design schematic diagram and layout can now be exported to NI/AWR Design Environment (Microwave Office circuit design software) for optimization and further analysis. The impact of parasitic effects that were omitted earlier in the design process can now be seen with the inclusion of Modelithics models. Figure 2 shows the change in filter response due to the inclusion of parasitics. Figure 2: Ideal filter response viewed within FilterSolutions (left). Filter response with parasitic models, interconnects, and substrate effects included as viewed within NI/AWR Design Environment (right). The most dramatic changes in frequency response that occur due to the inclusion of parasitics in this example are shifting of the entire pass-band to a lower frequency range from the ideal filter response and degradation of out-of-band rejection. The NI/AWR Design Environment schematic is used in combination with Modelithics Global Models to optimize part values and microstrip dimensions to compensate for these changes. The first round of optimization will be for component part values. The second optimization is performed on the interconnect line lengths and widths to further adjust performance. When the AWR project is created, in addition to a default microstrip layout, FilterSolutions automatically creates an EM extraction block in the AWR schematic. This permits easy EM 3 of 16

structure creation and EM analysis for the filter to provide further insight into expected filter performance. AXIEM is used to analyze the filters in this paper. Once the filter has been optimized and EM results have been examined, it is ready to be fabricated. This straightforward design process helps to eliminate guesswork and greatly simplifies a normally quite iterative filter design process. Design Example 135 MHz Elliptic Band-Pass Filter Figure 3: 135 MHz minimum-inductor zigzag elliptic band-pass filter ideal schematic (top) and layout (bottom). To demonstrate the effectiveness of this filter design process, a 135 MHz minimum-inductor zigzag elliptic band-pass filter is designed in FilterSolutions and optimized in NI AWR Design Environment using microstrip models and Modelithics models (Figure 3). Prior to assembly, yield analysis is performed to see the effect of part value tolerance on filter performance. 4 of 16

Measurement data is compared to simulation results using ideal models, Modelithics models, and EM analysis performed in AXIEM. Band-pass filters with topologies such as chebyshev II, hourglass, and elliptic, may have their inductor count reduced using a zigzag filter implementation. Odd order zigzag filters are more efficient than even order in that they require fewer inductors. Equal inductor zigzag filters have an additional advantage in that all the nodes may contain a capacitor ground, making it easy to absorb any parasitic node capacitance [1]. Yield Analysis Results Prior to fabrication, it is good practice to run a yield analysis to see how sensitive a design is to changes in part value due to tolerance. The results of a yield analysis performed with 2 percent tolerance on all capacitors, 5 percent on all inductors, and a normal (Gaussian) distribution for all parts can be seen in Figure 4. Part value tolerances seem to have a greater effect on passband performance causing frequency shift and changes in return loss. Figure 4: Yield analysis performed on 135 MHz elliptic band-pass filter. S11 (left) and S21 (right). Magenta simulation using Modelithics global models, grey yield analysis results after 50 iterations using 2 percent tolerance for capacitors and 5 percent tolerance for inductors. AXIEM Co-simulation Results Running an EM analysis is also an important step prior to fabrication. Schematic level simulations fail to capture line-to-line coupling or step effects at the junctions between lines of 5 of 16

different widths in the layout. In the case of this filter, the design frequency was relatively low and it can be seen in Figure 5 that there is little difference between the schematic level simulation using Modelithics models with microstrip line models and the co-simulation results using AXIEM and Modelithics models. However, this will not be the case with the higherfrequency filter examples discussed later in this paper. The inclusion of layout effects will cause a much more visible change in performance across the high-frequency range. For more information about using AXIEM with Modelithics models, see [3]. Figure 5: S11 (left) and S21 (right) for 135 MHz elliptic band-pass filter. Magenta circuit-level simulation using Modelithics global models and microstrip line models, red simulation using Modelithics global models and AXIEM EM simulation of the layout. Measurement Results Figure 6: Fabricated 135 MHz elliptic band-pass filter. Reference planes of the measurement are indicated by red dashed lines. The 135 MHz elliptic band-pass filter was assembled on 20 mil Rogers 4003C. S-parameters were measured to 3 GHz. The end-launch coaxial connectors were deembedded from all 6 of 16

measurement data shown in this application note. The reference planes were set to the inside of the coaxial connectors as seen in Figure 6. Testing was performed well beyond the filter s nominal pass-band to determine how accurately the simulation predicted not only the pass-band response but also the stop-band response. Throughout the frequency range evaluated, the measured filter response agreed quite closely with the simulated filter response (Figure 7 and Figure 8). Figure 7: Narrowband S11 (left) and S21 (right) for 135 MHz elliptic band-pass filter. Red simulation using Modelithics global models and AXIEM, blue measurement data. Figure 8: Broadband S11 (left) and S21 (right) for 135 MHz elliptic band-pass filter. Black simulation with ideal components and microstrip lines, magenta simulation using Modelithics global models, red simulation using Modelithics global models and AXIEM, blue measurement data. 7 of 16

Although designed and fabricated for a relatively low center frequency of 135 MHz, the filter s response in the stop-band was correctly predicted across a very wide frequency range using this design process. Figure 7 shows a narrowband view of the 3 db cut-off frequency of the filter and the excellent measurement to simulation agreement achieved. Figure 8 shows the progression of agreement between measured data and simulation results. The ideal circuit-level simulation shows reasonable agreement with measured data in the pass-band of the filter but shows a large frequency shift after 1 GHz. The parasitic circuit-level simulation (using Modelithics models), as well as the EM co-simulation, show excellent agreement across the frequency range with measured data. The Effect of Layout Success in filter synthesis depends on a large number of variables, during the initial design stages but also at the layout stages. FilterSolutions offers options within the NI/AWR Design Environment Export Utility to control the initial length and the widths of the microstrip interconnects (Figure 9). This gives the designer control over the physical size of the final filter. To demonstrate, the 135 MHz elliptic band-pass filter previously demonstrated was fabricated with a more compact layout (referred to as Layout B) and compared to the measured response of the original layout (referred to as Layout A) (Figure 10). Figure 9: Screenshot of the NI AWR Design Environment Export utility in FilterSolutions. Layout options are highlighted in red. 8 of 16

Layout A (original design) Layout B Figure 10: 135 MHz minimum-inductor zigzag elliptic band-pass filter layouts. Top - Layout A (original layout), bottom Layout B (compact layout) Figure 11 compares the simulated responses of the different layouts for the 135 MHZ bandpass filter. The pass-band is somewhat broader and shifted lower in frequency for Layout B compared to Layout A. Although the stop-band performance of both layouts shows a similar trend, the response of Layout B is shifted. Figure 11: Simulation comparison of two different layouts for the 135MHz band-pass filter. S11 (left) and S21 (right). Red simulation of filter with Layout A, blue simulation of filter with the more compact Layout B. 9 of 16

Figure 11 and Figure 12 show the measurement-to-simulation results of the 135 MHz bandpass filter using Layout B. The simulated response shows a similar trend to the measured data but is shifted across the band. Figure 12: Narrowband S11 (left) and S21 (right) for 135MHz elliptic band-pass filter with Layout B. Red simulation using Modelithics global models and AXIEM, blue measurement data. Figure 13: Broadband S11 (left) and S21 (right) for 135MHz elliptic band-pass filter with Layout B. Black simulation with ideal components and microstrip lines, magenta simulation using Modelithics global models, red simulation using Modelithics global models and AXIEM, blue measurement data. 10 of 16

Additional Filter Examples As seen in the previous filter examples, simply substituting ideal component models for Modelithics global models in the schematic significantly aids in achieving first-pass design success. The inclusion of EM analysis in the cases shown previously did not alter the results greatly. However, this is because of the low design frequency. At higher design frequencies, as will be shown in the following filter examples, EM analysis becomes a critical tool in predicting performance. 3GHz Chebyshev Type I Band-Pass Filter Example Figure 14: Ideal schematic (left) and layout (right) for 3GHz Chebyshev type I band-pass filter with coupled resonators. The filter shown above is a Chebyshev type I band-pass filter with coupled resonators. Chebyshev I filters are defined by their pass-band ripple, possess a sharp pass-band cut-off characteristic, and large group delay. Coupled resonators are narrow-band approximations of band-pass filters with the advantage of more practical element values at high frequencies and flexible element value selection [1]-[2]. Figure 15 and Figure 16 show measurement versus simulation results for the 3 GHz Chebyshev type I band-pass filter in a narrowband and broadband view, respectively. The co-simulation results seem to predict filter response quite well in the broadband view. However upon closer inspection of the pass-band, a shift can be seen. 11 of 16

Figure 15: Narrowband S11 (left) and S21 (right) for 3 GHz Chebyshev type I band-pass filter. Red simulation using Modelithics global models and AXIEM, blue measurement data. Figure 16: Broadband S11 (left) and S21 (right) for 3 GHz Chebyshev type I band-pass filter. Black simulation with ideal components and microstrip lines, magenta simulation using Modelithics global models, red simulation using Modelithics global models and AXIEM, blue measurement data. As seen in the previous example, yield analysis of part value tolerance can help the designer gain insight into possible causes of frequency shift in the fabricated filter. The fabricated filter used Coilcraft 0302CS inductors and Murata GJM15 capacitors. Based on the Coilcraft datasheet, a single tolerance (±5%) is specified for all part values in the family and there do not appear to be tight tolerance versions of this part. Therefore, the statistical analysis settings used in the yield analysis were a 5 percent tolerance on all inductors and a normal (Gaussian) distribution. By contrast, the Murata capacitors used have multiple tolerances available for each part value in the family (±0.05pF and ±0.1pF). As a result, the normal-tol distribution (Gaussian 12 of 16

distribution with center values removed) was selected for the capacitors. Assuming the capacitors used in the fabricated filter had a tolerance of ±0.1pF, and the tighter tolerance parts (±0.05pF) have been removed from the distribution, the yield analysis results in Figure 17 were obtained. The measured s-parameters for this filter fall within the variation in performance predicted by the yield analysis. (For more information about yield analysis see [4].) Figure 17: Yield analysis performed on 3 GHz Chebyshev type I band-pass filter. S11 (left) and S21 (right). Red simulation using Modelithics global models and AXIEM, blue measurement data, grey yield analysis results after 50 iterations. 5GHz Elliptic High-Pass Filter Example Figure 18: 5GHz elliptic high-pass filter deal schematic (left) and layout (right) The filter shown above is a 5GHz elliptic high-pass filter. This design example is unique in that it uses a combination of lumped and distributed elements. FilterSolutions allows the substitution of 13 of 16

low-q lumped elements with distributed stubs, and uses lumped elements when high-q parts are available [1]-[2]. AXIEM Co-simulation Results In the 135 MHz band-pass filter example (using Layout A), the results from AXIEM co-simulation were very similar to the schematic-level simulation using Modelithics models and microstrip line models. This is likely due to the spread-out nature of the circuit and relatively low design frequency. In contrast, the high-pass filter layout is more closely spaced, has a much higher design frequency, and also uses radial shunt stubs. Figures 18 and 19 show measurement versus simulation results for the 5 GHz elliptic high-pass filter. As can be seen in Fig. 19, the AXIEM co-simulation gives drastically different results from the schematic simulation. It is likely that the microstrip line models do not capture the behavior of these elements and their interactions with the rest of the layout as fully as an EM analysis. Figure 19: Narrowband view of S11 (left) and S21 (right) for 5GHz elliptic high-pass filter showing a comparison of measured to simulated 3dB cut-off frequency. Red simulation using Modelithics global models and AXIEM, blue measurement data. 14 of 16

Figure 20: Broadband S11 (left) and S21 (right) for 5GHz elliptic high-pass filter. Black simulation with ideal components and microstrip lines, magenta simulation using Modelithics global models, red simulation using Modelithics global models and AXIEM, blue measurement data. Conclusion By using software design tools such as FilterSolutions and NI AWR Design Environment, specifically Microwave Office and AXIEM, with accurate Global Models from Modelithics, the effects of parasitic circuit elements, substrates, and even layouts can be handled and accounted for in a single design cycle, helping to greatly simplify the entire RF/microwave filter design process. References [1] "Passive Filters." Lumped LC FilterSolutions. Nuhertz Technologies. http://www.filtersolutions.com/passive.html. 06 Jan. 2015. [2] "FilterSolutions User s Guide". Nuhertz Technologies. [3] I. Delgado, APPLICATION NOTE 051: AXIEM Co-Simulation with Modelithics Models, Modelithics, Inc. www.modelithics.com [4] L. Dunleavy, L. van der Klooster, APPLICATION NOTE 046: Improved Microwave Circuit design Flow Through Passive Model Yield and Sensitivity Analysis, Modelithics Literature, Modelithics, Inc., 02 Nov. 2012, Web. www.modelithics.com 15 of 16

About this note This note was authored by Isabella Delgado (student intern at Modelithics) with assistance from Jeff Kahler (Nuhertz Technologies) and Scott Skidmore (Modelithics). Contact Information For more information about Modelithics products and services, please contact Modelithics, Inc., Email: sales@modelithics.com / Web: www.modelithics.com For more information about Nuhertz products please see www.nuhertz.com For more information about NI AWR software products please see www.ni.com/awr All other trademarks are property of their respective holder. 16 of 16