The measurement setup, shown in Figure 1, consists of a PC, five-port circuit, a 16 bits Analog/Digital converter PICO ADC16 and a continuous wave signal source (NovaSource M2 NS2-1700104) with a constant 10 dBm output power at 2 GHz. The signals reflected from the sensor are measured using diode detectors (D1, D2, D3) at each output port P1, P2 and P3. The detectors were MIDISCO MDC1087-S Zero Bias Schottky diodes having ±0.5 dB flatness, max VSWR 1.2:1 with an SMA input and a BNC output. The sensors used in this work were the open ended coaxial and monopole sensor [7,8] constructed from an RG 402 semi-rigid cable.

The five-port circuit shown in Figure 2 was designed at 2 GHz using Microwave Office software version 4.3. The permittivity and thickness of the substrate were ɛ_r = 2.2 and 1.5748 mm, respectively.

The widths of each arm and ring of the five-port were w₁ = 1.008 mm and w₂ = 1.272 mm. The radius of the ring was r = 16.968 mm. The components SLIN, SSUB, STEE and SCURVE in Figure 1(b) represent the microstrip line, substrate, T junction and bend, respectively.

The relationship between the complex reflection coefficient Γ of the unknown load and the power ratios w_i can be written in the form [5,6]: (1) | Γ − q i | 2 = | w i | 2where: (2) Γ = x + j y (3) q i = u i + jv i (4) | w i | 2 = P i / k iwith i = 1, 2 and 3. P_i are the powers measured by the three detectors, k_i are the unknown constants to be determined from the calibration procedure and q_i are the values of the calibration standards. The four calibration standards used in this work to determine the unknown constant k_i, x_i, and y_i were a precision 50 Ω load, standard open standard, offset 120° and offset 240° open calibration standards. Assuming perfect matched load and open standards, we obtained from Equations (1–4): (5) P i k i = u i 2 − 2 u i x i + x i 2 + v i 2 − 2 v i y i + y i 2

At port 1 (I = 1): (6) P 1 k 1 = u 1 2 − 2 u 1 x 1 + x 1 2 + v 1 2 − 2 v 1 y 1 + y 1 2

After some algebraic manipulation of Equation (6), the unknown values k₁, x₁, and y₁ for port 1 can be determined from the linear equations system of Equation (7) by measuring the the four calibration standards: (7) ( P 1 open ( 0 ) − P 1 m 2 ( u 1 open ( 0 ) − u 1 m ) 2 ( v 1 open ( 0 ) − v 1 m ) P 1 open ( 120 ) − P 1 m 2 ( u 1 open ( 120 ) − u 1 m ) 2 ( v 1 open ( 120 ) − v 1 m ) P 1 open ( 240 ) − P 1 m 2 ( u 1 open ( 240 ) − u 1 m ) 2 ( v 1 open ( 240 ) − v 1 m ) ) ( 1 k 1 x 1 y 1 ) = ( u 1 open ( 0 ) 2 + v 1 open ( 0 ) 2 − u 1 m 2 − v 1 m 2 u 1 open ( 120 ) 2 + v 1 open ( 120 ) 2 − u 1 m 2 − v 1 m 2 u 1 open ( 240 ) 2 + v 1 open ( 240 ) 2 − u 1 m 2 − v 1 m 2 )

The same procedure can similarly be used to determine other k_i, x_i, and y_i values for ports 2 and 3.

From Equation (5), the power ratios at port 1, 2 and 3 can be written in the form: (8) P 1 k 1 = u 1 2 − 2 u 1 x 1 + x 1 2 + v 1 2 − 2 v 1 y 1 + y 1 2 (9) P 2 k 2 = u 2 2 − 2 u 2 x 2 + x 2 2 + v 2 2 − 2 v 2 y 2 + y 2 2 (10) P 3 k 3 = u 3 2 − 2 u 3 x 3 + x 3 2 + v 3 2 − 2 v 3 y 3 + y 3 2

For the device under test, u₁ = u₂ = u₃ = u and v₁ = v₂ = v₃ = v can be obtained from the following matrix: (11) ( − 2 x 1 + 2 x 2 − 2 y 1 + 2 y 2 − 2 x 2 + 2 x 3 − 2 y 2 + 2 y 3 ) ( u v ) = ( P 1 k 1 − P 2 k 2 − x 1 2 + x 2 2 − y 1 2 + y 2 2 P 2 k 2 − P 3 k 3 − x 2 2 + x 3 2 − y 2 2 + y 3 2 )

A computer program has been developed to control, acquire, and save data using the Agilent VEE version 7.0 graphical programming software. The program was also used to implement all the calibrations and calculations of the reflection coefficients. As an example, the ADC 16 data acquisition module is illustrated in Figure 3(a). In the final form, all the VEE objects are hidden and only the panel view of the Five-Port Reflectometer in run-time mode as shown in Figure 3(b) is accessible to the end user.

The performance of the five-port reflectometer was tested by comparing reflection coefficient S₁₁ values of eight different offset shorts using both the reflectometer and a commercial VNA. The results are listed in Table 1.

The mean error in magnitude between the reflectometer and VNA measurements was 0.0202, whilst the phase mean error was 1.91°. The accuracy of the reflectometer was further tested by comparing the reflection coefficient S₁₁ measurement results between the reflectometer and the VNA for several well known materials using both the monopole and open ended coaxial sensors. Again good agreement between the reflectometer and VNA results were obtained for all the materials listed in Tables 2 and 3 for the monopole and open ended coaxial sensors, respectively. The maximum errors between the VNA and reflectometer when using the open ended coaxial sensor were 0.0318 and 3.89° (or 3.18% and 1.08%) for magnitude and phase, respectively. The corresponding errors were 0.0304 and 3.14° (or 3.04% and 0.87%) for the monopole sensor.

More than 100 fruits in various stages of fruit ripeness were measured using both the coaxial and monopole sensors in conjunction with the five-port reflectometer. Abnormal, not fully developed, dry, and rotten fruits were not considered. The surface of the fruit was wiped dry to free excess surface moisture. All the reflection coefficient S₁₁ measurements of the fruit samples using the five-port reflectometer were done at 26 °C. The samples were then dried in a forced-air oven for four days at 105 °C for moisture content determination on a wet basis [7].

The variations in the magnitude and phase of S₁₁ with moisture content in oil palm fruits are shown in Figures 4 and 5 for the open ended coaxial and monopole sensors, respectively. It can be clearly seen from Figure 4 that both magnitude and phase of the reflection coefficient generally decrease with increasing moisture content in the range between 21% and 75% for the open ended coaxial probe.

However, for the monopole sensor shown in Figure 5, only the phase showed a tendency to decrease with increasing moisture content, whilst the magnitude values were randomly scattered with variation in moisture content. The equation and the determination coefficient R² shown in each graph were obtained by applying regression analysis. The determination coefficients for both magnitude and phase methods of the open ended coaxial sensor were almost similar. As expected, the highest correlation was obtained from phase measurements using monopole sensor. The phase of S₁₁ is highly sensitive not only to the length of the extended the inner conductor of the monopole sensor but also sensitive to small variation in the complex permittivity of the samples due to different percentages of moisture content. The weak correlation for the magnitude measurement using monopole sensor was due to multiple wave reflection between the extended inner conductor and fruit.

A computer program was developed using the Agilent VEE software to predict moisture content in oil palm fruits by applying inverse relationships to the equations in Figures 4 and 5. A panel view of the program is illustrated in Figure 6.

The results for a different batch of 100 samples of oil palm fruits using open ended coaxial and monopole sensors are compared with the actual values of moisture content obtained using standard oven drying method in Figures 7 and 8, respectively. The phase measurement of the monopole is the most accurate technique to predict the moisture content in oil palm fruits. In contrast, the magnitude of monopole is saturated and has a complex relationship with moisture content. The strong saturation effect in Figure 8(a) was due to fringing fields created by the interaction between the monopole back lobe radiation and the sample. A ground-plane flange is required to minimize the influence of this external noise and can also be used to increase the directivity of the monopole sensor. However, because of its size, a ground-plane flange monopole sensor is not suitable for small samples such as the oil palm fruits. The mean absolute errors were 6.78% and 5.95% when using the magnitude and phase equations of the open ended coaxial sensor, respectively. The lowest absolute mean error was 3.73% when using phase method with the monopole sensor. However, as expected from Figure 5(a), the monopole sensor also recorded the highest absolute mean error (11.32%) when using the magnitude method. This was due to multiple reflection along the interface between the extended inner conductor and fruit.