Achieving First Pass Success in PCB-Based Filter And Matching Circuit Designs

Parent Category: 2015 HFE

By  Scott Muir, Larry Dunleavy, and Tom Weller 

Introduction

This paper provides a comparison of three fabrication techniques used to generate a simple three element 250 MHz low pass filter (LPF) design measured and modeled to address in-band and out-of-band response through 5GHz. For each of the three scenarios, the filter will be fabricated using the values of the passive lumped elements obtained in the ideal LPF design process; it is expected that the measured data will differ from the ideal LPF design due to the parasitic effects present at high frequencies. Parasitic elements are added incrementally to the design in order to illustrate the progression towards better measured to model agreement; the final full parasitic LPF model design will contain transmission line (TL) elements and Modelithics® CLR Library models for the passive lumped elements. 

The purpose of this work is to provide insight into a “good, better, best” progression through fabrication techniques used to generate and measure a LPF.  One objective of this work is to highlight how the progressive modeling of the parasitic effects present at high frequencies will provide better measured to model agreement.  The best approach also utilizes on-board RF-probing and related thru-reflect-line calibration, whereas the other two approaches utilize standard coaxial connectorized measurements using a commercial short-open-load-thru calibration along with subsequent connector modeling/de-embedding. 

The same techniques discussed herein for a simple filter design apply equally for PCB-based matching circuit designs and should be of interest to anyone committed to improved simulation-based design success. 

Filter Design Approach 

The LPF shown in this document is one of three constant-k Butterworth¹ tee junction filter designs that belong to the 5th laboratory assignment of the Wireless Circuits & Systems Lab course offered at USF². A Butterworth filter is a device designed to have as flat of a frequency response as possible in the pass band; for a LPF, the pass band exists prior to the cut-off frequency. The cut-off frequency of a LPF is the frequency where the signal power has been reduced by half (the frequency where the signal experiences 3dB of loss). This 3dB loss is seen in the forward transmission coefficient (S21) of the LPF S-Parameters. The S-Parameters of the LPF can be calculated using the ABCD parameter equations derived for a general tee junction formed by three elements with given impedances. Also shown below is a table listing the capacitor and inductor values calculated to have LPFs with cut-off frequencies of 100, 250, and 500 MHz; also shown in the table are the passive values selected from the vendor datasheet. Note: the passive lumped element values used in the simulations and fabrication are the values obtained from the vendor datasheets for the ideal design. 

Table 1 • Calculated and vendor capacitor and inductor values used for three LPF designs.

Once the ideal design has been obtained, the next step would be to begin accurately modeling the parasitic affects of the physical device present at high frequencies in order to better predict the device behavior. These effects are caused by the passive devices, the pads each device is mounted on, the grounding and transmission line elements and the coaxial connectors (where applicable). 

In some cases, a design engineer can obtain S2P data for passive devices directly from the vendor. While these data files represent actual measured data for the passive device being used in the design, there is a degree of uncertainty as to how that data was obtained. The fixture used by the vendor to obtain the S2P data may be quite different from the fixture used by the designer. Differences can exist in the pad dimensions and substrate selection that can drastically affect device performance. The frequency range of interest to the designer also may not be included in the S2P data.

The steps of iteratively modeling the parasitic affects of the physical device will be demonstrated later in the paper.

Fabrication

The first fabrication technique used a milling machine, with the procedure performed by a University of South Florida (USF) Center for Wireless and Microwave Information Systems (WAMI Center) teaching assistant. This is the least accurate approach to fabrication and is labeled the “good” approach. This technique is the least accurate since it utilizes soldered slip-on coaxial connectors as a means of measurement and is grounded with a wire soldered between the TL and the ground plane. This rudimentary grounding is not optimal and is used in the USF WAMI lab to reduce course costs. The substrate used in this fabrication is 59 mil FR4 (Er = 4.3). 

Figure 1 • Fabricated 250 MHz LPFs. “Good” approach (top image), “Better” approach (middle image), “Best” approach (bottom image); measurement reference planes indicated by the red arrows.

The second fabrication technique was performed professionally by an outside board manufacturer with chemical etching as well as plated through via holes; however soldered slip-on coaxial connectors are still used as a means of measurement. This grounding technique is the preferred method at high frequencies since it is provides reliable and repeatable fabrication as well as minimizing the parasitic inductance present between the TL and ground plane. This approach is labeled as the “better” approach. The substrate used in this fabrication is 60 mil Rogers 4003C (Er = 3.65).  

Figure 2 • Ideal LPF.

The third fabrication technique uses the same board and plated through via hole technology as the “better” assemblies. This technique is labeled the “best” due to the use of probe pads allowing ground-signal-ground RF wafer probes to be used as a means of measurement. The use of probe tip structures provides yet another degree of reliable and repeatable measurements by removing the uncertainties that can exist when using soldered slip-on coaxial connectors.

Related information highlighting the importance of the measurement of noninsertable devices can be found in the product note “Agilent 8510-13”³.

Figure 3 • Ideal “good” LPF containing TL and ground models.

“Good” Approach: Measured vs Model

Figures 2 through 4 show the transition from ideal design to full parasitic design. The dimensions of the TL elements were obtained from the CAD design of the “good” LPF.

Plots of the ideal LPF model performance can be seen in Figure 5; Figures 6 and 7 show plots of measured data vs modeled performance of the “good” LPF.

Figure 4 • Full parasitic “good” LPF containing TL and ground models as well as Modelithics CLR models for the passive lumped elements (CLR models indicated by the red arrows) and connector models (connector models indicated by the green arrows).

As the parasitic elements are added incrementally, the measured to model agreement improves. However, even in the final full parasitic model, there are still discrepancies between the measured data and model performance. These discrepancies can be attributed to the less than ideal fabrication techniques used: in house milling on the USF campus, use of coaxial connectors as a means of measurement, and using a wire soldered between the TL and ground planes. 

Figure 5 • Ideal LPF model performance; S21 (blue line) and S11 (red line). Broad-band (left) and narrow-band (right). 

Figure 6 • Ideal “good” LPF with TL elements; measured data (symbols) vs model performance (solid line). S21 (blue) and S11 (red); broad-band (left) and narrow-band (right). 

Figure 7 • Full parasitic “good” LPF; measured data (symbols) vs model performance (solid line). S21 (blue) and S11 (red); broad-band (left) and narrow-band (right). 

“Better” Approach: Measured vs Model

Figure 8 shows the full parasitic design; the ideal design can be seen in Figure 2. The dimensions of the TL elements were obtained from the CAD design of the “better” LPF.

Figure 8 • Full parasitic “better” LPF containing TL and ground models as well as Modelithics CLR models for the passive lumped elements (CLR models indicated by the red arrows) and connector models (connector models indicated by the green arrows).

Plots of the “better” LPF measured data vs modeled performance can be seen in Figure 9; compare this to the ideal model performance in Figure 5.   For the “better” approach, there has been a drastic reduction in the discrepancies between the full parasitic LPF model performance and the measured data when compared to the “good” approach.

Figure 10 • Full parasitic “best” LPF containing TL and grounding effects as well as Modelithics CLR models for the passive lumped elements (CLR models indicated by the red arrows).

“Best” Approach: Measured vs Model

Figure 10 shows the full parasitic design; the ideal design can be seen in Figure 2. The dimensions of the TL elements were obtained from the CAD design of the “best” LPF.

Plots of the “best” LPF measured data vs modeled performance can be seen in Figure 11; once again, compare this to the ideal model performance seen in Figure 5.  For the “best” approach, the agreement between the full parasitic LPF model performance and the measured data has improved over the “better” approach and is much higher than the “good” approach.

Figure 11 • Full parasitic “best” LPF; measured data (symbols) vs model performance (solid line). S21 (blue) and S11 (red); broad-band (left) and narrow-band (right). 

Closing Remarks and Conclusions

In this document, a comparison was performed of three fabrication techniques used to generate a simple three element 250 MHz low pass filter (LPF) design. It was shown that using less than ideal calibration methods and fabrication techniques can lead to major discrepancies between model performance and measured data. This document also provided insight into how parasitic effects will cause the measured data to differ from the ideal model performance, especially when viewed at high frequencies. To obtain improved measured to model agreement, proper modeling of the parasitic effects in subsequent design iterations is required. 

Once a full parasitic model has been obtained, it is possible to tune the design in order to achieve the ideal LPF performance in the desired band.

About the Authors

Scott Muir was previously with Modelithics, Inc. He is now with TriQuint Semiconductor in Apopka, Florida. Larry Dunleavy and Tom Weller are with Modelithics.

Acknowledgements

The milling of the “good” LPF was performed by Anand Kumar Santhanakrishna, a USF PhD student.  The authors would like to thank Isabella Delgado and Alberto Rodriguez for their helpful editing comments.  

The ADS and CAD files associated with the filters shown here are available upon request. 

References

¹ David M. Pozar, “Microwave Engineering”, Second Edition, pp. 433-450.

²  T. Weller and L. Dunleavy, “Wireless and Microwave Education: From Circuits to Systems,” Proceedings of the 1999 European Microwave Conference - Invited Paper, pp. 93-97.

³ Keysight (formerly Agilent) Technologies, “Measuring Noninsertable Devices”, A new technique for measuring components using the 8510C Network Analyzer – Product Note.