1.INTRODUCTION
The Gigahertz TEM Cell (GTEM) is a special TEM device, that nowadays is used in about 300 test laboratories worldwide. This facility is utilized for both emission and immunity up to 1 GHz and above. Based on previous comments and recommendations in IEEE EMC sessions and including suggestions from several members of international standardization groups a very practical as well as challenging experiment was conducted. The GTEM cell is much smaller than for example an anechoic chamber. Therefor it is expected to find in the GTEM cell a stronger interaction of the EUT with the shielding structure as in an anechoic chamber. The empty GTEM cell 1750 (8m long, 4m wide, 3m high) has been investigated in the past, however further probing is necessary in order to understand the EUT interaction with the cell.

EUT INTERACTION
A metallic box (0.3 m cube) has been positioned in the center of the typical test volume, which is down the center line along the floor of the GTEM at 5.85 m from the apex at a height of 0.80 m above the floor. The septum height in this point is 1.6 m in total. The vertical electrical field strength has been measured 30 cm perpendicular to all 6 surfaces of the cube using a small mini-dipole and an analog fiber optic link system as well as a spectrum analyzer. Figures 2 to 4 show these test results normalized to the non perturbed field without the cube in the previously mentioned coordinates. Figure 1 shows the field variation without the cube in place.

Fig. 1:   Variation of the E-field 30 cm before and behind the center in relation to the field in the center point of the test volume Figure 2:   Variation of the E-field 30 cm before and behind the center with cube in relation to the field in the center point of the test volume Figure 3:   Variation of the E-field 30 cm above and below the center with cube in relation to the field in the center point of the test volume Fig. 4:   Variation of the E-field 30 cm left and right of the center with cube in relation to the field in the center point of the test volume

Fig. 1 reveals the typical spread in field strength variation of about ± 3 dB from 30 to 1000 MHz. Fig. 2 proves as expected the shadow area behind the box. The most evident explanation is defraction by the cube in the range below 300 MHz. In fig. 3 we find almost perfect symmetry. The curve labeled right is closer to the main access door, causing a slight asymmetry to the surface currents.

EMPTY CELL BEHAVIOR
Due to the fine resonant structures in the loaded cell a more detailed analysis was made. In the position 5.85 m length and 0.8 m height we took a reference measurement of the E field components above the center line (see fig. 5). The vertical field serves as the reference for the previous pictures and indicates 2 peaks in the critical transition area of the absorber section. The most affected frequency range of the working field component is somewhere in between 60 and 100 MHz.

In this range we also find the maximum of the mismatch in the VSWR. In contrast to the expected perfect TEM wave we find a strong longitudinal component at 123 MHz. Generally one can state that the overall decoupling of the longitudinal component is fairly weak. The transverse component is never a real problem and only limited by our measurement dynamics of the FM 2000 Holaday probe.

Volume Oriented Analysis at 123 MHz
Naturally one point is not sufficient, consequently we expanded into a triangular area over the center line towards the apex and a cross section perpendicular to the center line through the reference point. The increments in all cases were 10 cm steps, defined by the sensor cube dimensions. In each measurement we took the vertical and longitudinal component. Fig. 7 to 10 show the test results, Fig. 6 the used magnitude scale.

Figure 5:   E-field components in the reference point Figure 6:   Used E-field magnitude scale Figure 7:   E-field contour for the vertical component in the section towards the apex Figure 8:   E-field contour for the longitudinal component in the section towards the apex Figure 9:   E-field contour for the vertical component in the cross section Figure 10:   E-field contour for the longitudinal component in the cross section

Mapping the fields in figures 7 to 10 displays a non TEM wave structure in some areas. Measuring this field effects is extremely time and data consuming. Therefore a computer simulation, using a time difference code, was performed and the resulting cross sections can be depicted from figure 11.

Changing the Reference Point
The new point at 4.80 m length and 0.65 m height, where the septum height above ground is 1.30 m, shows a better decoupling than at 1.6 m height. Additionally the longitudinal peak is shifted higher in frequency, namely to 135 MHz (see figure 12).

Moving in the other direction towards the absorber wall we have taken one more point (length 6.20 m, height 0.85 m). In figure 13 we detect below 80 MHz a very bad decoupling. The longitudinal component supersedes the vertical working component.

SOURCES OF FIELD PROBLEMS AND RECOMMENDED SOLUTIONS
Although the GTEM cell has proven to be a very useful tool for pre-compliance and compliance measurements in EMC, there is still the need for further improvements. Better than trying to improve the emission correlation to open area test sites by statistics for example is it to find the sources of deviation inside the cell. For immunity tests as one can see there are some fairly sharp resonances. Comparing this fine structure of the fields, we find well designed absorber chambers still to be about 10 dB better in polarization decoupling. It becomes evident from the above mentioned and the figures presented that one of the main sources of problems in the GTEM cell is the "broad band" terminator.

Fig. 11:   Computed E-field contour in dB for the vertical a) and the longitudinal b) component in the cross section

Figure 12:   E-field components in the point 4.80 m length and 0.65 m height

Figure 13:   E-field components in the point 6.20 m length and 0.85 m height

1. Looking at the typical 5 dB reflectivity values for the 60 cm PU cone absorbers at about 100 MHz one immediate improvement would be adding ferrite tiles. This could gain another 5 dB typically. Foam and tiles, however, have to be impedance matched.
2. Reducing the length of the foam absorbers will certainly reduce the capacitance. The resistor plates would be shorter and consequently the inductance of the plates will decrease, preventing resonances.
3. Presently those resistors are installed on epoxy PCB with a typical dielectric constant of 4.8. The conductive junction islands on the boards between each single resistor are about 10 mm wide. This does introduce too much capacitance. A further step could be, to cut out the PCB material below the resistor bodies, leading to reduction of capacitance. Reducing the lead length of the resistors with several 10th of nH will further improve the RF performance of the boards.
4. Suppressing unwanted surface currents in the corners of the cross sections in the GTEM cell is going to improve the field quality [3].
5. As already known from EMP simulators since many years a tilted resistive termination is very useful. The electric field lines would than be in parallel to the resistors, aligning the natural flow of field lines better.
   Experiments indicate that these measures should improve the field quality in some areas of almost 10 dB.

1. D. Hansen, D. Ristau et. al., "Analysis of the Measured Field Structure in a GTEM 1750", in Proceedings of the IEEE EMC Sym. 1994, Chicago, Aug. 1994, p. 144-149.
2. D. Hansen, D. Ristau et. al., "Expansions on the GTEM Field Structure Problem", in Proceedings of the IEEE EMC Sym. 1995, Atlanta, Aug. 1995, p. 538-542.
3. L. Jendernalik, Zur Feldqualität von TEM-Zellen (translated: From the Field Quality in TEM Cells). Dortmund: University Dortmund, Faculty of Electrical Engineering, dissertation, 1995