A New PCB Substrate Used for Patch Antenna Fabrication

This article describes the modeling, fabrication and analysis of microstrip patch antennas. Extracting material parameters for a new PCB substrate as well as determining the suitability of this material for the fabrication of patch antennas are described. Microstrip test patterns are used in conjunction with the antennas to facilitate the parameter extraction. Also, procedures for modeling and particularly for producing well-matched antennas that resonate close to the design frequency are presented. These procedures include consideration of the physical characteristics of microstrip antennas and their behavior relative to the given material parameters. The principal antenna modeling tool consists of microstrip line models implemented in the LIBRA® software package. Comparisons are made with and conclusions drawn from a Maxwell equation solver that uses a finite-difference time domain (FDTD) analysis. Antennas produced from other materials are also discussed. Antenna fabrications include comparisons between chemically etched antennas and milled antennas produced using a PCB prototyping machine. Efficiency of 0.80 is achieved using the new material.

Christian Bean, Patrick Roblin, Vakur Ertürk and Roberto Rojas
The Ohio State University,
Department of Electrical Engineering
Columbus, OH

The extraction of material parameters, specifically dielectric constant and loss tangent for GML1000, a new PCB substrate, is desired. The parameters are determined using test patterns fabricated with the new material. After parameter extraction, the suitability of this substrate with regard to microstrip patch antenna fabrication is explored. The performance of the antennas fabricated with the material provides insight as to the validity of the er and tand values determined from the test patterns. The patch antennas are modeled using microstrip line models implemented in LIBRA, a circuit simulation software application. The design procedures used to achieve well-matched antennas resonating close to the design frequency as well as the performance of the resultant antennas are discussed. An alternative modeling approach using an FDTD technique is also described, and comparisons are made between antennas milled with the QuickCircuit prototyping machine and antennas that are chemically etched.

Test Patterns

To extract material parameters independent of antenna analysis, test circuits, as shown in Figure 1 , are utilized. The test bed, which is utilized to extract the S parameters from each pattern, uses the same SMA connectors for all the tests so the measurements are comparable. The three pairs of open stubs and the thru line are used to generate a circuit model of the SMA connectors in LIBRA for a known substrate. The three band-pass resonators (2, 2.5 and 3 GHz) then are used to measure the dielectric constant and loss tangent for each resonant frequency.

The first test pattern is fabricated from a known substrate, TLC32 (er = 3.19, tand = 0.003), using a PCB prototyping machine. The initial step is to generate an accurate model in LIBRA for the SMA connectors. The software's optimization abilities are used to vary the parameters of the connector model until a good match to the |S11 | data collected from the three pairs of open circuit stubs and the thru line is achieved. The connector model then is implemented in the LIBRA models for the three microstrip resonators. The 2, 2.5 and 3 GHz resonators then are compared to data in LIBRA. The |S21 | model-to-data comparison shows a tand value close to 0.003 (the value certified by the material's manufacturer).

Next, the GML1000 substrate is examined. The same connector model has demonstrated high accuracy when modeling the open circuit stubs from this material's pattern so it is left unaltered. Note that in the new material optimizations, one er value is used for all three resonators but separate tand values are assigned to each resonator. Therefore, three separate loss tangents may be extracted. Substrate parameters derived from these fits are listed in Table 1 . Obviously, some scattering is present in the tand values. However, a value of approximately 0.003 can be assumed for the GML1000 material.

Table I
60 Mil GML1000 Material Parameter Extraction

Resonator Frequency (GHz)

e r

tan d










Patch Antenna Fabrication

After using the test patterns to extract the new material's dielectric constant and loss tangent, a patch antenna using GML1000 material (which is well matched and resonates at the 2.45 GHz design frequency) is constructed. Comparisons are made using antennas fabricated with TLC32 and Duroid materials. Sixty-mil substrates are used for antenna fabrication, and the antennas utilize a microstrip-fed design, as shown in Figure 2 .

In addition, results of antenna simulations from a three-dimensional FDTD method are considered. This numerical method is formulated via discretization of the differential form of Maxwell's equations over a volume by approximating derivatives with a central-difference approximation. A computer program that implements this method1 is used to extract the S parameters of passive structures. Results of this program provide antenna performance characteristics such as directivity, efficiency and radiation pattern. These results are compared with the LIBRA simulation results and provide insight into the behavior of various antenna characteristics.

A design algorithm is implemented using LIBRA to produce a well-matched antenna resonating at the desired design frequency. The algorithm is based upon results of antenna fabrications and fitting the microstrip line model to data from these fabrications. Thus, the algorithm is essentially empirical and requires two or more fabrication iterations to achieve matching and the proper resonant frequency. For instance, in the GML1000 material case, a total of three antennas are fabricated. The first antenna design is based on equations derived from a transmission line antenna model.2 The second design is based on fitting a microstrip line antenna model implemented in LIBRA to data from the first antenna. (Antenna parameters such as radiation resistance and fringe capacitance are updated to allow the model to fit the measured data.) The third antenna design is based on simultaneously fitting the model to data from the first two antennas.

The design methodology is first implemented using the 60-mil GML1000 substrate. Note that all antennas are fabricated using the prototyping machine. The results of each fabrication iteration are listed in Table 2 .

Table II
Design Iterations for GML1000 Antennas


GML #1

GML #2

GML #3

|S11 | (dB)




Fr (GHz)




A large improvement in matching is observed between the first (GML #1) and second (GML #2) antennas, while only a small improvement is observed between the second and third (GML #3) antennas. Also, minimal improvement in resonant frequency results from GML #3 compared to GML #2. This result is curious since the design for GML #3 was based on the simultaneous fitting of both GML #1 and #2. It is believed that the fabrication limits of the prototype milling machine were reached in the third antenna iteration. (The precision of the machine is not sufficient to produce physical dimensions that result in an antenna resonating at exactly 2.45 GHz.)

The design algorithm is also implemented using 60-mil TLC32 substrate material. In this case, the initial antenna is fabricated using the artwork from the GML #3 antenna since the material parameters for TLC32 are close to those of GML1000. Again, a total of three antennas are produced. The results for each antenna iteration are listed in Table 3 .

Table III
Design Iterations for TLC32 Antennas


TLC32 #1

TLC32 #2

TLC32 #3

|S11 | (dB)




Fr (GHz)




Note that the matching for the second TLC32 antenna actually is inferior to the initial antenna's matching (although an improvement in resonant frequency is seen). However, a large matching improvement is observed for the third antenna compared to the first two TLC32 antennas. This result illustrates that the results are improved significantly when the fitting involves simultaneous consideration of two previous antennas.

Finally, the design methodology is implemented using 60-mil Duroid as the substrate material ( er = 2.33, tand = 0.0012). As with the GML1000 antennas, the first Duroid antenna design is performed using the transmission line model equations.2 Then, a second antenna is designed and fabricated based on these results. The results for the two iterations are listed in Table 4 .

Table IV
Design Iterations For Duroid Antennas


Duroid #1

Duroid #2

|S11 | (dB)



Fr (GHz)



Gain Measurements and Results

Gain measurements are taken for the final two GML1000 antennas, final TLC32 antenna and final Duroid antenna, and the results are shown in Figure 3 . These results are compared with a horn antenna of known gain to generate gain values for each patch antenna.

Efficiency values for each antenna can be extracted using the relationship between gain, efficiency and directivity, or

G = hD (1)

G = antenna gain
h = efficiency
D = directivity

The efficiency values listed in Table 5 were determined using 7.75 dB as the directivity value. This directivity was determined using the FDTD Maxwell equation solver1 and represents the most reliable directivity value available.

Table V
Antenna Efficiencies


Experimental: h = G/D

Simulated: h (LIBRA)

Simulated: h (FDTD)

GML #1

0.83± 0.03



GML #2

0.79± 0.03




0.82± 0.03




0.92± 0.04



Figure 4 shows an example E-plane directivity plot for a 0.003 loss tangent produced by the Maxwell equation-solving program. The efficiency values are based on the assumption that this directivity is realistic and constant for all four antennas. Also listed in the table are theoretical efficiencies from LIBRA simulations as well as efficiencies predicted by the FDTD program (GML1000 antennas only). Note that the errors in the experimental h values, which result from uncertainties in gain and directivity, show that all empirical and simulated efficiencies agree within tolerance.

The antenna efficiency analysis provides solid justification for the tand values used in the antenna designs. The tand value for TLC32 (0.003) is assumed correct because it is certified by the manufacturer. If the tand for GML1000 was significantly higher or lower than the 0.003 design value, this result certainly would manifest itself in the empirical efficiency data obtained. Note that one GML1000 antenna produces a higher efficiency than the TLC32 antenna while the other GML1000 antenna produces a lower efficiency value. This discrepancy is difficult to explain quantitatively. However, differences in fabrication processes (especially SMA soldering) as well as experimental measurement error may provide the explanation.

Comparison with Etched Antennas

The fact that the third and final GML1000 antenna produced no improvement in resonance frequency compared to design frequency suggests that the fabrication limitations of the prototype milling machine were reached. Beyond a certain point the machine is not capable of producing a more accurately resonating antenna due to the mechanical tolerances of the actual milling process. The absolute milling tolerances of the machine are not known exactly but, from experience, the tolerance is thought to be on the order of a few mils at best. To test the hypothesis, the architectures for the final two GML1000 antennas are chemically etched. A comparison is then made between the antennas produced from etching and the antennas produced from the milled fabrication.

From the results, it is obvious that etching and milling using identical layout architectures producesomewhat different results. It is interesting to note that in both cases the milled antenna exhibits matching superior to the etched antenna. Of course, this result can be explained by the fact that the designs for each antenna are based on empirical data from previous antennas. It is believed that if the design algorithm was repeated using antennas fabricated from chemical etching rather than mechanical milling, improvement in antenna performance would result. The results for the final GML1000 antenna are illustrated in Smith chart format, as shown in Figure 5 . The milled/etched comparison for GML #2 also shows differences in |S11 |.


Material parameters for the new PCB substrate have been extracted successfully using test patterns. Specifically, er was determined to be 3.24 and tand was 0.003. Empirical efficiency measurements and comparisons with antennas fabricated from the industry-standard TLC32 and Duroid substrates provide confidence with regard to the validity of these values. Also, an empirical design algorithm that provides well-matched antennas close to the design frequency has been implemented successfully. Note that the TLC32 and GML1000 antennas achieve 0.80 efficiency. Finally, comparisons between chemically etched and milled antennas show a definite difference between the two fabrication approaches.


The authors would like to thank Chuck Ludwig, John DeLockery, Kenny Davis and Phillip Johnson of Glasteel Industrial Laminates (GIL), Collierville, TN, for their fruitful discussions, and the GIL Corp. for providing logistic and financial support for this research project. The LIBRA software package is a product of HP EEsof, Santa Rosa, CA. The QuickCircuit PCB milling machine is a product of T-Tech Inc., Atlanta, GA.


1. Vakur Ertürk, Design and Analysis of an Active Integrated Transmitting Antenna, Masters Thesis, 1996, The Ohio State University.

2. I.J. Bahl and P. Bhartia, Microstrip Antennas, Artech House, 1980, pp. 31-82.

Christian Bean received his BS in engineering physics in 1995 from the University of Pittsburgh. He also received his MS in electrical engineering in 1997 from The Ohio State University where he was a graduate research assistant in the department of electrical engineering from 1996 to 1997. Bean's master's thesis, conducted under the advisement of Patrick Roblin, is entitled Evaluation of GML1000 for Microstrip Antenna Design. Currently, he is working with Lucent Technologies in Columbus, OH.

Patrick Roblin received his maitrise de physics degree from the Louis Pasteur University, Strasbourg, France, in 1980 and his MS and DSc degrees in electrical engineering from Washington University, St. Louis, MO, in 1982 and 1984, respectively. He joined the department of electrical engineering at The Ohio State University as an assistant professor in August 1984 and became an associate professor in June 1990. During the 1993-1994 academic year, Roblin was on professional leave at the Microelectronics and Technology Center at Allied Signal Aerospace in Maryland. His research interests include nonlinear semiconductor device modeling and circuit modeling for microwave and wireless applications.

Vakur B. Ertürk received his BS degree in electrical engineering from Middle East Technical University, Ankara, Turkey, in 1993 and his MS degree from The Ohio State University (OSU), Columbus, OH, in 1996. He is currently pursuing his PhD degree at OSU in the Electroscience Laboratory where he has been a graduate research associate since 1993. Ertürk's research interests include microstrip and active integrated antennas and numerical solutions of antenna problems.

Roberto G. Rojas received his BSEE from New Mexico State University and his MS and PhD degrees in electrical engineering from The Ohio State University in 1979, 1981 and 1985, respectively. He currently is associate professor in the department of electrical engineering at The Ohio State University. Rojas' research interests include the development of analysis and design tools for conformal antenna structures and active integrated antennas as well as the analysis of electromagnetic radiation and scattering phenomena in complex environments. He is a senior member of the IEEE and an elected member of US Commission B of URSI.