A Compact Broadband Analog Complex Correlator with High Correlation Efficiency for Passive Millimeter-Wave Imaging System

Abstract

In this paper, the design, fabrication, and measurement of a compact broadband (4–8 GHz) analog complex correlator for a passive millimeter-wave imaging system are presented. To achieve high sensitivity and high integration of the imaging system, the wideband and miniaturization of the correlator are required. The correlator achieves wide bandwidth by using the add-and-square method, which is composed of a six-port circuit and a detection circuit. In order to realize the miniaturization of the correlator, the six-port circuit is realized on the chip base on the 0.15-μm gallium arsenide (GaAs) process. The influence of mismatch of the detection circuit that employs zero-bias Schottky diodes on the correlator is also analyzed to guide the design of the correlator. The measurement results of the designed chips and detector are consistent with the simulation result. Finally, a Sweep-frequency test is applied to the designed correlator, and the measurement results show that, within the frequency range of 4–8 GHz, the correlation amplitude fluctuation is less than 1.9 dB and the correlation efficiency is larger than 99%, which reveal that the correlator is suited for interferometric passive millimeter-wave imaging applications.

1. Introduction

Passive millimeter-wave imaging technology has been studied and applied in the field of security screening [1,2,3]. Since the thermal radiation power of a scene is extremely low (about −174 dBm/Hz) [4], the passive imaging systems must have the characteristic of high sensitivity to uncover the threats hidden beneath clothing. In the passive millimeter-wave imaging system, the complex correlator is one of the key components of the system by performing correlation processing on two signals [5], and the performance of the imaging system is significantly influenced by the correlator. For the passive imaging system, high sensitivity requires a large system bandwidth [6], which means that a broadband correlator is required to meet the system requirements.

Moreover, in our previous work, a Ka-band passive imaging system was researched and reported [7]; the imaging system achieves better spatial resolution by using 1024 receiving channels, then a large number of correlators is needed for the imaging system. Therefore, compact size is also required for the analog correlator to reduce the size of the passive imaging system.

Generally, the correlator can be implemented using either analog or digital technology. Compared with the digital correlator, the analog correlator has the advantages of large bandwidth, high sensitivity, and low cost [8,9,10]; therefore, it is adopted in our passive millimeter-wave security imager. In addition, the analog correlator can be implemented using either the direct multiplication technology [11,12] or the add-and-square technology [13,14]. The direct multiplication analog correlator is difficult to achieve via broadband due to the limitation of multiplier chips. The add-and-square analog correlator is normally based on the six-port technology [15], and it can achieve wide bandwidth and high sensitivity at the expense of a larger volume and complex structure [16,17]. Therefore, the add-and-square analog correlator is selected in this paper to realize the broadband characteristic.

In this paper, a compact broadband (4–8 GHz) analog correlator based on the integrated six-port circuit for the passive millimeter-wave imaging system is designed, fabricated, and measured. The theory of the complex cross-correlation circuits is described, and the influence of mismatch of the detection diode on the correlator is also analyzed to guide the analog correlator design. The block diagram of the proposed analog correlator is represented in Figure 1.

It can be seen that the designed analog correlator consists of a six-port circuit, detection circuits, and amplifiers. The additional amplifiers in the proposed analog correlator provide isolation between the six-port circuit and the detection circuit, reducing the influence of the mismatch of the detection circuit on the deterioration of the correlator. In addition, the designed correlator is divided into three parts for design and processing; part 1 is a chip integrated with a $0°$ divider and a $90°$ divider, part 2 is a chip integrated with two $90°$ couplers and four amplifiers, and part 3 is a detection circuit. Both part 1 and part 2 are fabricated with the 0.15-μm GaAs process to reduce the size of the six-port circuit, and part 3 is made of a microstrip circuit with the zero-bias Schottky diode to achieve broadband characteristics of the detector. In addition, part 1 and part 2 can form a compact six-port network by using Multi-Chip Module (MCM) technology, and the reason for designing and machining the first and second parts separately is that part 2 is also used in the receiver front-end in a passive imaging system [18].

The rest of this paper is organized as follows. In Section 2, we describe the theory of the complex cross-correlation circuits, after which the influence of mismatch of the detection diode on the correlator is also analyzed to guide the analog correlator design. Section 3 presents the design and measurement of each part of the correlator, including the 0° and 90° divider chips, coupler and amplifier chip, and the detection circuit. In Section 4, the integrated six-port circuit using the designed chip is measured, after which the Sweep-frequency test is applied to the fabricated analog correlator. Finally, the conclusion is provided in Section 5.

2. Principle of the Proposed Analog Correlator

An ideal analog correlator is usually used to measure the phase difference between two input signals. In our previous work [19,20], the analog complex correlator based on the six-port technology has been applied in the passive imaging system for security applications. Assuming the inputs are two single-frequency signals $S_1\left(t\right)$ and $S_2\left(t\right)$, which can be presented as

$$\begin{array}{l}{S}_1\left(t\right)=a\cos \left({\omega }_{0}t+{\theta }_1\right)\\ S_2\left(t\right)=b\cos \left({\omega }_{0}t+{\theta }_2\right)\end{array}$$

(1)

After the signal distribution network, zero-biased square-law detector and LPF, assuming the dc voltage resulting from the second-order transconductance of the square law-detector is 1, then the useful output dc voltage can be expressed [20,21] as

$$\begin{array}{l}{V}_3=\frac{ab}{2}\left[1-\sin \left({\theta }_1-{\theta }_2\right)\right]\\ V_4=\frac{ab}{2}\left[1+\sin \left({\theta }_1-{\theta }_2\right)\right]\\ V_5=\frac{ab}{2}\left[1+\cos \left({\theta }_1-{\theta }_2\right)\right]\\ V_6=\frac{ab}{2}\left[1-\cos \left({\theta }_1-{\theta }_2\right)\right]\end{array}$$

(2)

Then, the real and imaginary parts of the cross-correlation function can be obtained as

$$\begin{array}{l}{V}_{real}=V_5-V_6=ab\cos \left({\theta }_1-{\theta }_2\right)\\ V_{imag}=V_4-V_3=ab\sin \left({\theta }_1-{\theta }_2\right)\end{array}$$

(3)

Equations (1)–(3) indicate that both the real and imaginary parts of the cross-correlation function could be measured by the complex correlator [16]. In addition, the phase difference between the two input signals needs to be swept from $0°$ to $360°$ to get the complete correlation circle.

As shown in Figure 1, in the detection circuit, a low-pass filter (LPF) is connected behind the diode, which means that the RF signal will be reflected by the LPF. Therefore, the entire detection circuit in the wideband range is mismatched, and it is necessary to consider the influence of the mismatch of the detection circuit on the correlator. Assuming that each device in the six-port circuit is completely ideal, the reflected signal will be consumed by the isolation resistance or the internal resistance of the signal source, and it will not affect the correlation results. To evaluate the influence of the reflected signal on the correlation results, we assume that the isolation of each device in the six-port circuit is 20 dB, the return loss of each port is −20 dB, and the amplitude and phase consistency are ideal. Then the simulation of the influence of reflected signal on correlation results is carried out in Advanced Design System (ADS), which is represented in Figure 2.

The return loss of the diode is greater than −1 dB, which means the detection circuit is mismatched. The reverse isolators are added between the diode and the six-port circuit to adjust the amplitude of the reflected signal into the six-port network. On the other hand, we can get the complete correlation circle by sweeping the phase difference between the two input signals from $0°$ to $360°$. Therefore, each reverse isolation will get the corresponding correlation circle. Generally speaking, the correlation coefficient of the analog correlator is required to be greater than 0.9 [22,23]. To estimate the demand of the isolation between two input signals for the correlator quickly, we can specify that the variation range of the correlator circle radius is less than 2%. The simulation result of the influence of reflected signal on correlation results is represented in Figure 3. It can be seen that the requirement of the reverse isolation between the six-port circuit and detection circuit is larger than 17 dB.

3. Implementation of the Analog Correlator Based on Integrated Six-Port Circuit

As shown in Figure 1, the designed C-band (4–8 GHz) correlator is divided into three parts for design and processing, including the 0° and 90° divider chips, the coupler and amplifier chips, and the detection circuit. The designed chip of part 1 and part 2 is based on the GaAs process, and the Skyworks SMSA7630-061 diodes are used for the detection circuit design in part 3.

3.1. The 0° and 90° Integrated Divider Chip

To reduce the area of the designed chip, the $0°$ and $90°$ divider adopts a lumped Wilkinson power divider, and the design focus is to ensure that the phase difference is $90°$ between the two output ports of the $90°$ divider in the operating frequency range of 4–8 GHz. Therefore, the $\pm 45°$ phase-shifting structure [24] is applied at the output of the divider to meet the requirement of the divider. It is worth noting that two +$45°$ phase-shifting structures are also applied at the output of the $0°$ divider to keep the consistency of phase and amplitude. Then the schematic, layout design, and simulation result of the designed integrated divider chip are shown in Figure 4.

From Figure 4a, the lumped Wilkinson divider is a two-stage divider to increase the bandwidth of the divider chip. The layout design of the integrated divider chip is represented in Figure 4b; numbers 1 to 6 represent the port number of the chip, ports 1 and 2 are input ports, and ports 3 to 6 are output ports. Moreover, the coplanar waveguide is used as the transmission line at each output port to increase the isolation between the transmission lines and prevent signal crosstalk inside the chip.

It can be seen that the return loss ($S_{11}–S_{66}$) of each port is below −17 dB, the insertion loss ($S_{31}, S_{51}, S_{42}, S_{62}$) varies from −3.8 to −4.5 dB within the operating frequency of 4–8 GHz, and the fluctuation of the insertion loss is less than 0.7 dB. As shown in Figure 4d, the amplitude imbalance is less than $\pm 0.3$ dB and the phase imbalance is less than $\pm 3.6°$. Moreover, the size of the proposed integrated chip is 2 × 2.3 mm, and the fabricated chip and measurement result are represented in Figure 5.

The measurement of the fabricated divider chip is carried out on the probe station, and the Keysight N5227B vector network analyzer is also used. It can be seen that the measurement results are consistent with the simulation results. From 4–8 GHz, the return loss ($S_{11} \mathrm{and} S_{22}$) of the input port is better than −18 dB, and the insertion loss ($S_{51} \mathrm{and} S_{62}$) of the $0°$ and $90°$ divider is better than −4.7 dB, with the in-band fluctuation less than 0.5 dB. From Figure 5c,d, the amplitude imbalance is less than $\pm 0.4$ dB and the phase imbalance is less than $\pm 4°$.

3.2. The Coupler and Amplifier Integrated Chip

As discussed in Section 1, the coupler and amplifier integrated chip is also used in the receiver front-end in the interferometric imaging system. Therefore, the design details of the chip will not be introduced in this section, and only the layout design, the fabricated chip, and simulation result will be introduced, which is shown in Figure 6. The size of the proposed integrated chip is 2.3 × 4 mm.

It can be seen that the designed integrated chip includes two couplers and four amplifiers, which can form a six-port circuit with the 0° and 90° divider chips. The amplifier in the six-port circuit can not only amplify the signal to compensate for the loss caused by the signal distribution network but also isolate the six-port circuit from the detection circuit. As shown in Figure 6b, in the operating frequency of 4–8 GHz, the return loss ($S_{11}$) is better than −14 dB, the gain ($S_{51}$) is larger than 13 dB, and the reverse isolation ($S_{15}$) is larger than 40 dB, which meets the requirement for the correlator. In addition, the amplitude imbalance is less than ±0.5 dB, and the phase imbalance is less than ±2°.

3.3. Detection Circuit

The detection diodes in the detection circuit are the zero-bias Schottky diodes (SMSA7630-061) in terms of their high stability and high dynamic range. The block diagram of the detector circuit is illustrated in Figure 7.

In the detection circuit, three parallel filter capacitors are used in the design of LPF, and their capacitance values are 1.3, 2.0, and 4.7 pF, respectively, to prevent the output of the high-frequency signal. The measurement result of the designed LPF is shown in Figure 8a. It can be seen that the suppression of the RF signal by the LPF in the working frequency band (4–8 GHz) is greater than 19 dB. In addition, the analog integration circuit, which consists of a parallel $100 \mathrm{k}\mathsf{\Omega }$ resistor and a parallel 100 pF capacitor, is located at the output port of the detector to meet the integration time requirements of the passive millimeter-wave imaging system [19,21,25].

In terms of detector input matching, common matching methods include impedance matching and microstrip stub matching [26,27,28,29]. The microstrip stub matching structure has the advantage of a small loss of signal, so the sensitivity of the detector will be relatively high, but the bandwidth of this matching method is limited, and the detection sensitivity fluctuates greatly in the working frequency band [29,30]. The impedance matching network usually uses a parallel 50-ohm resistor for matching. The advantage is that the matching state of the detector in the broadband range is very good, and the sensitivity in the working frequency band is relatively flat. However, this matching method will cause most of the energy of the input signal to be consumed by the resistor, so the sensitivity of the detector will deteriorate sharply [20,21].

As the detector is used in the analog correlator for interferometric passive imaging applications, the voltage sensitivity of the detector needs to be flat in the operating band to ensure the equivalent noise bandwidth of the correlator. On the other hand, the voltage sensitivity of the detector also needs to be as high as possible to meet the requirements of the high sensitivity of the correlator in interferometric imaging applications [3]. Based on the analysis of the requirement, the matching network of the detector adopts a parallel 100-ohm resistor for matching to ensure the flatness of sensitivity in the operating bandwidth. Moreover, an additional microstrip stub matching structure is added to improve the matching state to improve the voltage sensitivity of the detector, which is illustrated in Figure 8b. The measurement result of the DC output and voltage sensitivity of the detector is shown in Figure 8c–e.

It can be seen that the return loss of the detector in which only a 100-ohm resistor is used for matching is better than −4 dB within the operating bandwidth (4–8 GHz), and when a 100-ohm resistor and microstrip stub is used for matching the return loss of the detector is better than −10 dB. Although the return loss of the detector is still insufficient in the broadband range, considering that the amplifier between the six-port circuit and the detector can play the role of reverse isolation, the designed detector can still be used in the circuit of the correlator.

From Figure 8d, the square-law detection region of the designed detector is less than −15 dBm (decibel relative to one milliwatt). A comparison of detector voltage sensitivity under 100-ohm matching and 100-ohm resistor and microstrip stub matching is represented in Figure 8e; it can be seen that the improvement of voltage sensitivity by an additional microstrip stub is limited in the frequency range of 4–6 GHz, and the improvement of voltage sensitivity by an additional microstrip stub is about 20% in the frequency range of 6–8 GHz. Since the detector is in a good matching state at low frequency and a poor matching state at high frequency when only a 100-ohm resistor is used for matching, the additional microstrip stub matching improves the sensitivity of the detector better at high frequencies. In addition, the calculated voltage sensitivity flatness is better than 1.9 dB within the operating frequency band.

4. Measurements

As shown in Figure 1, the designed analog correlator is based on the six-port technology, then the performance of the integrated six-port circuit is measured. Next, a Sweep-frequency test is applied to the correlator to evaluate the frequency-dependent characteristic of the correlator.

4.1. Measurement of the Six-Port Circuit

To evaluate the performance of the six-port circuit, a printed-circuit-board test fixture, which is designed using a 5-mil thickness Rogers 3003 substrate with 50 Ω microstrip, is taken for the measurement of the integrated six-port chip. The Keysight N5227B vector network analyzer and Rohde and Schwarz HMP 2030 programmable power supply are used for the measurement, which is represented in Figure 9.

As required by the amplifiers in the coupler and amplifier integrated chip, the DC voltage of the amplifiers is 3 V, and the total consumed current of the four amplifiers is 156 mA. Moreover, the size of the six-port circuit is 7.1 × 4 mm; compared to the previous work [19,20,21], the size of the six-port circuit has been significantly improved. The measurement result of the six-port circuit is represented in Figure 10.

The port number of the six-port circuit is shown in Figure 9a. It can be seen that, within the working frequency band of 4–8 GHz, the return loss of each port is better than −14 dB and the average gain is about 5 dB, which is illustrated in Figure 10a,b. From Figure 10c,d, the measurement result of the amplitude imbalance of the six-port circuit is less than $\pm 1.1$ dB, and the phase imbalance is less than $\pm 9°$.

4.2. Sweep-Frequency Test of the Correlator

As shown in Figure 1, an analog correlator consists of a six-port circuit and four detection circuits, then the final correlator has been realized on a 5-mil thickness Rogers 3003 substrate, and the substrate is bonded to a metal carrier for fixing. In addition, two types of analog correlators have been designed for measurement, which is represented in Figure 11. The difference between the two types of correlators is the matching of the detectors, and the other circuits of the correlator are the same type I used 100-ohm and microwave stub for matching, and type II only used 100-ohm for matching. Considering that the amplifier plays the role of reverse isolation, the purpose of designing two types of correlators is to further investigate the influence of the matching state of the detector to the correlator. Moreover, the size of type I is 49 × 36 mm, and the size of type II is 39 × 36 mm.

The Sweep-frequency test is focused on the Sweep-frequency correlation performance of the correlator, including its amplitude variation within 4–8 GHz and amplitude variation versus RF input power at the center frequency of 6 GHz. The setup for Sweep-frequency measurement is shown in Figure 12, which has been described in our team’s previous publications [19,31].

The Keysight N5227B vector network analyzer in this measurement is used to generate two RF signals that have identical power levels and frequencies and controllable phase differences. When the phase difference of two input signals is swept between $0°$ and $360°$, then the outputs of the analog correlator can describe a correlation circle [16,29]. The outputs of the correlator are recorded by the digital multimeter.

In practice, due to the non-ideal behavior of the actual devices, there will be errors in the output of the correlator [31,32]. From Equation (3), the error model of the correlator output is proposed as

$$\begin{array}{l}{V}_{real}=ab\cos \left({\theta }_1-{\theta }_2\right)+C_r\\ V_{imag}=abg\sin \left({\theta }_1-{\theta }_2+\varphi \right)+C_i\end{array}$$

(4)

where $g$ and $\varphi$ are the quadrature amplitude and phase error of the correlator, respectively, $C_{\ast }$ ($C_r$ and $C_i$) is the output DC offset error of the correlator. We can calculate these error parameters through the least-squares method to calibrate the measured correlation circle. In addition, the correlation efficiency of the analog correlator can be evaluated by the frequency-dependent amplitude characteristic of the correlator. It should be pointed out that the reference [11] uses the deterioration factor to describe the characteristics of the correlator, and this paper uses the correlation efficiency to describe it. But their calculation formula is the same, which is formulated as [11,33]:

$${\eta }^2=\frac{{\left|{\displaystyle \int \left|W\left(f\right)\right|\cos \left[\delta \left(f\right)\right]df}\right|}^2}{{\displaystyle \int B·{\left|W\left(f\right)\right|}^{2}df}}$$

(5)

where $\eta$ is the correlation efficiency of the correlator, $\left|W\left(f\right)\right|$ and $\cos \left[\delta \left(f\right)\right]$ are the amplitude response and phase response of the correlator, respectively, and $B$ is the working bandwidth of the correlator. To estimate the correlation efficiency of the correlator at the point frequency, it can be calculated by coincidence between the measured correlation circle and the fitting circle at the single frequency point of the correlator [34]. Then the correlation efficiency can still be calculated using Equation (5), but the meaning of its parameters has changed. Where $\left|W\left(f\right)\right|$ and $\cos \left[\delta \left(f\right)\right]$ are the amplitude response and phase-detection error of each test point on the measured correlation circle, respectively.

4.2.1. The Sweep-Frequency Test of the Correlator with Input Power Variation

In this measurement, the frequency of the input signal is fixed at the center frequency (6 GHz), and the input power of the two input signals changes from −31 to −20 dBm. Moreover, under each input power, the phase difference of the two input signals is also swept from $0°$ to $360°$. Then, the measurement result of the analog correlator is represented in Figure 13.

It can be seen that the correlation circle of the type II correlator is similar to an ellipse; and the correlation circle of the type I correlator is close to the ideal circle, which is reflected in the quadrature amplitude error of the correlator in Figure 13c. Moreover, from Figure 13d, there is a linear relationship between the correlation amplitude (dB). In addition, by fitting the amplitude curve of the correlator, when the input power is greater than −28 dBm the correlator is out of the square-law detection range, resulting in the correlation circle gradually becoming a diamond, which can be seen in Figure 13a,b. Therefore, the input power of the correlator is recommended to be less than −28 dBm. Then the input power (dB) and the average correlation amplitude of the type I correlator are about 25% larger than that of the type II correlator, while the input power is less than −28 dBm, which can be calculated from Figure 13d.

As the phase difference of two input signals is swept from 0° to 360° by using the Keysight N5227B vector network analyzer to form a correlation circle, we can calculate the phase difference of two input signals by using the measured correlation circle, which has been calibrated. Then, the calculated phase difference will be different from the actual phase difference, which is called the phase-detection error. The maximum phase detection error refers to the maximum absolute value of phase-detection error when the phase difference is swept from 0° to 360°. Figure 13e shows the maximum phase detection error versus the input power; we can see that the maximum phase detection error increases with the increase in input power, and it is less than 1.4° with an input power less than −28 dBm, which satisfies the 3° angular accuracy requirement for the interferometric imaging system applications [35].

By using Equation (5), the calculated correlation efficiency of the correlator is shown in Figure 13f. The correlation efficiency of both types of correlators is greater than 99%, which meets the index requirements of the analog correlators, and the correlation efficiency of the type I correlator is larger than that of the type II correlator.

4.2.2. The Sweep-Frequency Test of the Correlator with Frequency Variation

In this measurement, the input power of the input signal is fixed at −30 dBm, which works in the square-law detection range; the frequency of the two input signals changes from 3 to 9 GHz and the phase difference between the two input signals is also swept from $0°$ to $360°$. Then, the measurement result of the analog correlator is represented in Figure 14.

It can be seen that the quadrature amplitude error of the type I correlator is better than the type II correlator, and the correlation circle of the type II correlator is closer to the ellipse, which is consistent with the conclusion of the measurement in Section 4.2.1. In addition, in the frequency range of 4–8 GHz, the average correlation amplitude of the type I correlator is about 22% larger than that of the type II correlator, which is represented in Figure 14d, and the normalized correlation amplitude fluctuation of both types of correlators is less than 1.9 dB, which is also represented in Figure 14d. Meanwhile, by comparing with the correlator, which has been published [31], the proposed correlator exhibits a more stable frequency-dependent property. As shown in Figure 14e, the maximum phase detection error of both types of correlators is less than 1.4° within the operating frequency range of 4–8 GHz, which meets the requirement of the interferometric imaging system. The calculated correlation efficiency of both types of correlators is greater than 99%, and the correlation efficiency of the type I correlator is larger than that of the type II correlator, which is shown in Figure 14f. From these measurements, we can see that the proposed analog correlator could be used to implement correlation measurements, and its amplitude flatness satisfies the requirements of the imaging system.

From the Sweep-frequency test of the two types of correlators, the quadrature amplitude error of the type I correlator is better than type II, which leads to the correlation circle of the type II correlator being closer to the ellipse, but this error can be eliminated by the calibration process. On the other hand, the correlation efficiency of both types of correlators is greater than 99%, which meets the requirements of the imaging system.

A comparison table between two types of correlators is represented in

Table 1. Comparison between two types of correlators.

Type	Type I	Type II
Frequency	4–8 GHz
Input Power 1	−30 dBm
Correlation Circle Radius 2	26 mV	21 mV
Quadrature Amplitude Error	$<\pm 0.5$ dB	$<\pm 1$ dB
Max Phase Detection Error	<1.2°	<1.4°
Correlation Efficiency	>99.6%	>99.3%
Size	49 × 36 mm	39 × 36 mm

¹ The input power during measurement. ² The correlation circle radius at the center frequency.

. The average correlation amplitude of the type I correlator is about 22% larger than that of the type II correlator, and the size of the type I correlator is about 25% larger than that of the type II correlator. Therefore, the improvement of the detector matching state can improve the output amplitude of the correlator, but this improvement is at the expense of volume. The type I correlator is suitable for the system with higher sensitivity requirements, the type II correlator is suitable for the system requiring higher integration, and both types of correlators meet the requirements of the current imaging system. In addition, the performance of the designed analog complex correlator is compared with other relevant work, which is shown in

Table 2. Comparison table of the analog complex correlator.

Reference	This Work	[19]	[21]
Freq (GHz)	4–8	3.5–8	1.5–2.5
Input Power * (dBm)	−30	−20	−17
Normalized Correlation Amplitude Fluctuation (dB)	<1.9	<8.2	<1
Correlation Efficiency	99.3%	96.6%	95.9%

* The input power during measurement.

. It can be noted that the designed correlator achieves lower normalized correlation amplitude fluctuation and higher correlation efficiency in the broadband range (4–8 GHz).

5. Conclusions

In this paper, a compact broadband (4–8 GHz) analog complex correlator is presented for the passive millimeter-wave imaging system. The designed correlator adopts the add-and-square structure to obtain broadband characteristics, and the six-port circuit in the correlator adopts the 0.15-μm GaAs process to reduce the circuit size. By using MCM technology, 0° and 90° divider chips and coupler and amplifier chips can form a compact six-port network with a size of only 7.1 × 4 mm. The matching network of the detection circuit adopts a parallel 100-ohm resistor to ensure the flatness of sensitivity in the operating bandwidth. Then the analog correlator is designed and fabricated, and the measurement results show that the input power of the correlator should be less than −28 dBm. When the input power is fixed at −30 dBm, the phase detection error is less than 1.4°, the normalized correlation amplitude fluctuation is less than 1.9 dB, and the correlation efficiency is larger than 99% within the frequency range of 4–8 GHz, which reveal that the correlator is suited for the passive millimeter-wave imaging applications.

In the next step, to further reduce the size of the analog correlator, it is necessary to implement the detection circuit with a chip. The ultimate goal is to integrate the six-port circuit and detection circuit into a single chip to minimize the size of the correlator.