1. Introduction

Millimeter-wave imaging (MMW) at 95GHz presents unique challenges and opportunities for remote sensing applications. A motivation for choosing 95GHz is the atmospheric attenuation [1] is 0.6dB km^-1 at 94GHz, but increases to 2dB km^-1 at 140GHz, and 8dB km^-1 at 220GHz. Thus, it can be argued that 95GHz is in the range of the highest RF frequencies low enough to avoid significant atmospheric attenuation. A desirable advantage of MMW imaging is that millimeter-waves penetrate through dust and fog [2, 3]. MMW imaging is also used for security screening, because many textile materials are transparent in the MMW regime [4]. A downside to MMW imaging is the poor resolution, and detector limitations including sensitivity and cost. The fabrication of MMW detectors with sufficient sensitivity to detect minute levels of passive radiation is difficult, and there are few MMW detectors commercially available.

Several researchers have demonstrated MMW cameras. For example, a refractive MMW camera with a diameter of 0.45m has been demonstrated [5], as well as a similar design with diameter of 0.76m [6]. MMW polarimetric imaging has also been investigated [7], with the conclusion that polarimetric imaging may be particularly useful in eliminating water as a false target. Other researchers have studied active and passive mm-wave imaging of objects such as guns and clothing [8]. Recently, researchers [9] have reported excellent passive 95GHz imaging results, using an optical upconversion detection method.

Generally, MMW imaging is still in the research stages, and the literature does not show many examples of MMW images. Such images are necessary for researchers to determine the utility of MMW imagery, and the system requirements for MMW imaging. In previous work [10], we developed a MMW simulation that uses material properties and optical phenomenology to generate expected scene images. Many common materials, such as water, wood and concrete, have a high reflectivity in the MMW regime. Therefore, our simulation uses non-sequential ray-tracing, allowing us to accurately model multiple reflections. The simulation generates photorealistic images that can be used to ascertain whether objects can be identified within a scene, given scene temperatures and the imager parameters such as the NEDT (noise equivalent detector temperature) and the size of the imager aperture.

In this paper, we report on the development of a passive MMW imaging system for remote sensing. The passive MMW diode detector [11] used in this experiment was manufactured by HRL [12]. The HRL chipset consists of two chips: a 5 stage InP HEMT low-noise 95GHz amplifier (LNA), followed by a zero-bias Sb-heterojunction backward tunnel diode detector. We refer to the combination of the input feedhorn, LNA, and diode detector as the radiometer. The radiometer has polarization dependence, as the waveguide and the detector are polarization sensitive. We refer to the integrated system, consisting of the radiometer and 0.6m diameter reflector, as the imager.

MMW images are acquired by mechanically raster scanning the antenna over the field of view. The results show atmospheric conditions play a very significant role in the image appearance and quality. Images were acquired at the two different linear polarizations, which gave quite different results, primarily due to the polarization dependence of the water reflectivity. Image spatial resolution was measured to determine the MTF (modulation transfer function), and ascertain that the system is diffraction-limited. We quantify the maximum noise per pixel acceptable in a MMW imager. Noise was added to experimentally obtained images, and the resultant images were studied to find where the image quality was significantly reduced.

We compare the simulated data from the MMW imager with the experimental results. Based on our imager aperture and sensitivity, we generated simulated images [10] for comparison with the experimental data. In comparing the simulated data with the experimental data, the MMW system has not yet been radiometrically calibrated, and comparison of real scenes with simulated scenes requires many assumptions in the simulated data about the material properties of the scene. Qualitatively, the MMW simulation was consistent with the experiment.

2. MMW imaging simulations

2.1 MMW remote sensing

The performance of a MMW imager is far worse than an infrared camera [13] (with the same aperture size) in regards to spatial resolution and sensitivity. The wavelength of 95GHz radiation is 3.2mm, which is a factor of 2133 greater than infrared 1.5µm radiation. This means an infrared camera with aperture diameter of 0.3mm would have the same angular resolution as the MMW camera with an aperture size of 0.6m. It is extremely challenging to build passive MMW radiometers, as the energy corresponding to 1K at 95GHz is only 100fW [14]; also it is difficult to build semiconductor devices to operate at high frequencies such as 95GHz [15].

Most objects reflect specularly in the millimeter wave regime. Every material is described by its emissivity, reflectivity, and transmissivity, which sum to 1. The emissivity, reflectivity, and transmissivity parameters can be derived from the complex index of refraction [16] and the material thickness. We measured [10, 17] the complex refractive index of a variety of materials including fabric and water. Fabrics, fog, and dust are relatively transparent in the MMW. Concrete has a reflectivity of 0.2 [18], and the reflectivity of wood is also approximately 0.2 [19]. The emissivity of water depends on temperature, and is 0.6 at room temperature [20]. The complex index of refraction of water [21] is approximately n=3.95+2i at 95GHz, which would correspond to a reflectivity of R=0.4 [20, 22]. The literature did not have millimeter-wave bidirectional reflectance distribution function (BRDF) data, so in these simulations we made informed guesses as to whether the materials were specular or Lambertian, based on the surface roughness. We assumed rough materials such as grass, trees, and bricks were Lambertian, as the blades of grass could be pointing randomly. Smoother materials such as water, concrete, glass, and wood were assumed to be specular. Metals are highly reflective, and usually specular. Whether a material is specular or Lambertian has significant implications for both active and passive MMW imaging. Specular targets are more difficult to actively image because they require the target be at the correct angle to reflect the incident beam back to the imaging system.

Atmospheric conditions play a significant role in the appearance of MMW imaging because of two factors. First, the ground (or sea) often has a considerably higher apparent temperature than the sky. Secondly, as described above, many common materials, such as concrete or water, have a significant MMW specular reflectivity. For example, consider the MMW image of a metal boat on the ocean near the shore, on a clear day when the ground is at temperature of 300K, and the water at temperature of 270K. The reflectivity of the ground is 0.1 [23] and the reflectivity of the sea is 0.4 [20], for normally incident rays. Then, the apparent temperature of sky, sea, and ground could be 50K [24], 217K, and 275K, respectively. The metal boat has a high specular reflectivity. Thus, if we ray trace the rays from the MMW imager to the boat, many of these rays will be reflected to the sky, depending on the angle of the boat’s facets. So the boat’s apparent temperature will be the same as the sky, at least in many regions of the boat, and the apparent temperature difference between the boat and the sea will be over 150K, making the boat clearly visible if the imager has a sensitivity of 1K per pixel.

We observed indoor scenes demand higher temperature sensitivity, because the temperature differences inside may be only 5K. In outdoor scenes, trees and bushes are more difficult to recognize than manmade objects such as cars and buildings. Natural features are often smaller than the imager resolution, and their temperatures and reflectivities can be very similar to the background with random patterns. Man-made objects such as buildings, boats and helicopters typically present sharp geometrical lines for the eyes to detect, allowing the mind to process a blurry image. The images shown in this paper are outdoor scenes of boats.

2.2 Scene Simulation

In previous work [25], a MMW scene simulator was developed. Figure 1 shows the theoretical and experimental images of a boat; the experimental results will be discussed in a later section. For the MMW simulation, the air temperature of objects in the scene, such as the boat and trees, was assumed to be 289K (60°F). The water temperature was assumed to be 283K(50°F). In our initial simulation, we set the boat surfaces to be high-reflectivity metal and the imaging system diameter to 0.6m.

Currently, our scene simulator does not incorporate the effects of polarization. We approximate the appropriate indices for water in the two different polarizations, and an additional approximation will be that the incident polarization angle does not change in the case of multiple reflections. The Fresnel formulae [22] describe the reflectivity as a function of angle for the s- and p- polarizations, and in this case the s-polarization is horizontal and the p-polarized light lies in the plane perpendicular to the water. As the boat is 30m away from the imager, and the imager height above the water is 3m, we calculate an incidence angle (measured from the normal to the water) of 84°. Then the s-plane reflectivity of the water is found to be R_s=0.89, and the p-plane reflectivity is R_p=0.16. Figure 1(c) shows the simulated MMW image of the boat for the case when the detector is vertically polarized, and Fig. 1(e) shows the simulated image when the detector is horizontally polarized. Note that for the horizontally polarized image, the sea appears darker, as compared to the vertically polarized image. This is because for the horizontally polarized image, the sea reflects the cold sky, while for the vertically polarized image, the sea is more emissive.

The simulations of Fig. 1(c) and Fig. 1(e) include blurring, as determined by the aperture size, wavelength, and target distance. The blur corresponds to an angular resolution of 0.37° (or 6.4mrad), giving a spatial resolution of 19cm at a distance of 30m. For this boat scene, the blur does not significantly reduce the capability of the observer to recognize the boat or even objects on the boat. The noise was set to 1K per pixel, which is on the order of magnitude of the noise temperature in experimental imagers.

The simulated visible image (Fig. 1(a)) was designed to match the experimental visible image (Fig. 1(b)). In generating the MMW simulations, there are numerous parameters that the user may choose, such as object temperatures, material parameters, and ray-tracing parameters. These parameters were all chosen based on the best estimates of the true physical parameters, not chosen to make the simulated MMW scene appear similar to the experimental MMW images. We used material parameters from the literature and our measurements. These material parameters play a significant role in determining the MMW scene appearance. Scene temperatures in the simulation were assigned based on temperatures measured with a thermometer, at the time that the experimental images of Fig. 1 were acquired. We set the sky temperature to 60K, based on experimental measurements under similar atmospheric conditions [24]. The ray-tracing parameters include the maximum number of bounces that an optical ray can take before it is eliminated from the simulation, and we set this number of bounces equal to 20, as the image appearance converges with this bounce value.

In Fig. 1(c), the primary image degradation arises from the blurring and not from the noise, which was verified by observing the image appearance at 0K, and at 1K noise per pixel, was practically indistinguishable. In Fig. 1(g), we increased the noise per pixel of Fig. 1(c) to 12K. At 12K noise temperature, the image starts to degrade, although most of the features seen in Fig. 1(c) can still be distinguished in Fig. 1(g). This approach of adding noise allows us to specify the noise temperature for an imager.

3. The single-pixel passive millimeter-wave imager

3.1 Detector

The passive MMW radiometer [11] used in this experiment was fabricated from an HRL chipset [12]. The HRL chipset consists of two chips: first, a 5 stage InP HEMT low-noise amplifier (LNA), and then a zero-bias Sb-heterojunction backward tunnel diode detector. Using a zero-bias diode is advantageous because the electrical power requirement is low, and the 1/f noise is lower, as compared to the biased Schottky diode. Quinstar [26] worked with us to wirebond and package the HRL chipset, with a waveguide-to-microstrip transition. The Quinstar package is shown in Fig. 2(a); the amplifier biases are controlled externally and the output signal is from the SMA connector. Figure 2(b) shows the radiometer unit, and Fig. 2(c) shows the block diagram for the radiometer. Signal from the feedhorn is coupled to a waveguide-to-microstrip transition, through the 95GHz LNA, and then the diode. An output capacitor helps to prevent the sensitive diode from voltage spikes in the line. Microstrip lines couple the output diode signal into the SNA cable. This design has no 95GHz cables between the feedhorn and the detector, which is optimal because 95GHz cables are quite lossy.

 

Fig. 1. Simulated and experimental images of a trawler boat, at a distance of 30m. The color bar at right shows the temperature(Kelvin) grayscale for the MMW images. (a) Simulated visible image of boat at a distance of 30m. (b) Experimental visible image of boat. (c) Simulated MMW image of boat in vertical polarization, including the effects of finite resolution and 1K noise temperature. (d) Experimental passive MMW image of boat with imager vertically polarized. Boat motion during the imager acquisition time is apparent in this image from the slanted appearance of the boat cabin. (e) Simulated mm-wave image in horizontal polarization. (f) Experimental passive MMW image with imager horizontally polarized. (g) As in Fig. 1(c), this simulated MMW image shows the boat in vertical polarization, including the effects of finite resolution and noise temperature increased to 12K (h) This is the same image as Fig. 1(d), with 12K noise computationally added.

Download Full Size | PDF

We used a aqueous blackbody calibration source [27] to study the radiometer NEDT (noise equivalent temperature difference). The output of the diode is a DC voltage, and the radiometer responsivity is approximately 3µV/K. Our experiments indicated that the drift in the radiometer responsivity was minimal, even in outdoor environments. There is a DC offset in the radiometer response, due to capacitance in the radiometer unit as well as capacitance in the data acquisition system, and variation in this radiometer DC offset leads to 1/f noise. When the system warms up for several hours, the drift in the radiometer DC offset decreases substantially. Still, there is a remaining drift associated with temperature variations in the radiometer unit, with a magnitude of 1µV over 4 minutes, corresponding to 0.3K. Although such drift is acceptable over short image acquisition times, the drift in the radiometer response does make it necessary to calibrate the imager in the field, which is non-trivial. Also the drift makes it more difficult to work with multiple detectors, as different detectors may drift in different directions.

When a chopper was used, we measured a radiometer NEDT of 0.2K, with a bandwidth of 1kHz. When no chopper was used, the radiometer NEDT was measured at 0.6K over a time of 4 minutes, at an acquisition rate of 1kHz. The increase in NEDT for the case without the chopper is due to the drift in the radiometer DC offset. For the case of no chopper and a 4 minute acquisition time, the radiometer is uncalibrated, and the measurement is relative.

HRL reported the radiometer has NETD=0.3K for a video bandwidth of 1kHz, with 10Hz drift calibration rate. Their measurement parameters were different than ours, as we did not calibrate drift at a rate of 10Hz, but the HRL results are not inconsistent with our results.

The radiometer is mounted on the motorized image platform, as described below. The moving platform creates additional challenges for the data acquisition system, as we must measure µV signal levels, and avoid picking up noise from the gimbal motors, and from motions of the cables connecting the radiometer to the data acquisition system. To minimize noise from the motor system and cable motion, we located the data acquisition digitization electronics on the imaging platform. Locating the data acquisition electronics on the gimbal platform leads to the additional constraint that the data acquisition system be physically small and lightweight.

Our data acquisition system was the National Instruments [28] NI USB-6289, with a specified sensitivity of 760nV at the sampling rate of 125kHz. Each pixel represents the average of 1000 points from the data acquisition system. The random noise per pixel is 1.5µV, and the pixel sampling rate is 125Hz, so that the integration time per pixel is 8ms. Using this sensitive data acquisition system allowed us to acquire the signal without an additional video amplifier. This data acquisition system was lightweight (less than 1lb) and connected with a USB cable to the computer, so that it was suitable for mounting on the gimbal platform.

3.2 Antenna Design

Zemax optical design software [29, 30] was used to design the Cassegrain telescope. Figure 3(a) shows the primary. In Fig. 3(b) the secondary is seen, as well as the data acquisition electronics behind the primary, and the motorized telescope mount. The telescope primary is a 0.6m diameter spun aluminum dish with an f-number of F/0.45. This primary dish was designed for use at 2.4GHz, so the surface precision may be insufficient for use at 95GHz, leading to scattering which reduces the signal intensity. The aluminum secondary has a diameter of 10.5cm, and was designed and machined with a hyperboidal surface. At the focus of the secondary is the 15° FWHM (full-width half-maximum) scalar feedhorn, which is then attached to the mixer diode. The feedhorn angle corresponds to a -10dB edge taper on the secondary, which according to [25] is a good selection to minimize spillover and minimize the diffraction spot size.

The Zemax simulation, shown in Fig. 3(c), predicted that the imager would be diffraction-limited, giving a resolution of 0.37°. The secondary design was optimized for an imaging distance of 30m, and the secondary position could be adjusted to give close to diffraction-limited imaging at a distance of 100m. As seen in Fig. 3(c), the simulation showed that rays from the primary will not strike the central 2cm diameter region of the secondary, because these rays are obscured by the secondary. Therefore the central region of the secondary will only contribute stray light to the image. To reduce this stray light, a 2cm diameter absorber was placed at the center of the secondary.

 

Fig. 2. (a) The HRL chipset is mounted in the Quinstar package. (b) The radiometer, showing the scalar feedhorn connected to the Quinstar package. The braided wires are used to bias the HRL components, and the radiometer output is from the SMA connector on the Quinstar package. Ruler marks are in inches.(c) Block diagram of the radiometer. A scalar feedhorn couples to a 95GHz amplifier, and then to a zero-bias diode. The resultant output is coupled into the SMA connector.

Download Full Size | PDF

The Zemax simulation showed two primary sources of loss in this imager. The first loss, which occurs in all Cassegrain telescopes, arises because the secondary obscures the central region of the primary. The secondary diameter is 19% that of the primary diameter, resulting in a power loss of about 4%.

Another source of loss is caused when the diffraction-limited beam size at the feedhorn has energy outside the diameter of the feedhorn. The design was optimized to minimize this loss, which was predicted to be 2dB. The reflector optics introduced a loss of 3–5dB, which was somewhat higher than expected from the Zemax simulation. Initially, we attributed the difference between experiment and simulation to the surface tolerance on the primary. However, we machined the primary mirror to a smooth surface, and the optical loss remained unchanged, indicating that the primary surface tolerance did not lead to optical loss. Our Zemax simulation did not accurately model the Gaussian feedhorn beam, which could explain the difference between the simulation and the experimental results.

The optical loss was measured by comparing the radiometer readings when an extended absorber, dipped in liquid nitrogen, was placed directly in front of the radiometer, and when the same target was placed at the radiometer focus. Looking at water at a distance 7m away, we measured an imager responsivity of 2.3µV/K and an offset voltage of 3.37mV.

In our imaging system, no chopper was used and we did not compensate for radiometer drift. From a system perspective, it is advantageous to omit the optical chopper from the system, especially for multiple detectors, as the chopper may impose constraints on the mechanical design, and the data acquisition rate. The imager takes 4 minutes to acquire an image, and the noise associated with the image is 0.9K. The image noise is higher than the radiometer noise, due to loss in the optical system. The noise sources are random noise in the LNA, as well as drift in the detector and data acquisition electronics. The random noise and the drift noise are of similar magnitude, over the 4 minute data acquisition time. At various stages of system development, we used different detectors in the radiometer, with the radiometer NEDT varying by 0.3K. The imager NEDT reported here was measured with a different detector than was used for the radiometer NEDT measurement, and so the imager and radiometer NEDTs are not exactly related by the optical loss.

 

Fig. 3. Single-pixel millimeter-wave imager and associated electronics. (a) Oblique view of imaging system, showing the 0.6m primary dish. The detector is behind the central aperture of the primary reflector. (b) Side view of the system., showing the feedhorn and electronics behind the primary. (c) Zemax optical simulation.

Download Full Size | PDF

3.2 System

To simplify the development process, optical components were mounted on a breadboard, as shown in Fig. 3(a). Accessories such as a video camera can be easily mounted on the breadboard, as well as a translation stage so that the secondary could be precisely focused. Analog-to-digital acquisition electronics, as well as radiometer power supplies, were mounted on the breadboard, behind the feedhorn, shown in Fig. 3(b). Mounting the electronics on the breadboard allowed us to keep all cables short, and avoid relative motion between the electronics and the radiometer. Reducing electrical noise was critical because the detector signal is on the order of µV. Special care was taken to isolate the radiometers from electrical noise associated with the motorized mount movement.

A significant issue with the single-pixel imager is data acquisition time. The HRL radiometer was sensitive enough so our system was not limited by the radiometer, but by the speed of the motorized mounts mechanically rastering over the scene. The imager noise temperature was measured at approximately 1K. The motorized mount [31] can slew 15° in 3s, so that acquisition of a 15×15° image with a sampling resolution of 0.2° took 4 minutes. The sampling resolution is chosen at approximately half the diffraction-limit which is considered appropriate for an incoherent system. To minimize data acquisition time, we designed our system so data could be acquired under continuous motion.

The imager was designed to be lightweight, portable, and economical. The entire system including electronics easily fits into a car, and we transported the system to various outdoor locations to acquire images, using a battery for power.

4. Experimental results and analysis

In Fig. 1(b), we show the visible image of a boat at a distance of 30m. The angular width of the boat is about 8°, so the boat breadth was approximately 4m. The air temperature was approximately 289K and the water temperature was approximately 270K, as used in the simulation of Fig. 1. The MMW imager resolution at this distance is (theoretically) 19cm. Figure 1(d) shows the MMW image taken when the radiometer is horizontally polarized. Figure 1(f) shows the image when the radiometer is vertically polarized.

To present the experimental images of Fig. 1, the output voltages from the radiometer were converted into temperature, using a linear mapping. The radiometer responsivity is constant,, but the radiometer offset voltage varies. We are still studying imager calibration in the field; the calibration is non-trivial. We could use a target of hot water at a known distance from the imager, but this calibration is complicated because water has a non-zero reflection coefficient. At the time these images were acquired, the imager was not radiometrically calibrated (especially with regards to radiometer offset), so the MMW scene is relatively, but not absolutely, accurate. In this paper, we estimated the offset voltage value by setting the lowest sky temperature equal to 68K. The offset voltage varies for the different images.

The MMW images of Fig. 3 show skewing, which resulted from boat movement during image acquisition. Such movement effects motivate future work to develop high temporal resolution MMW imagers.

The image of Fig. 1(h) was generated by computationally adding 12K of additive Gaussian noise to the image of Fig. 1(d). This was possible because the main calibration issue was the detector offset drift, while the detector responsivity did not significantly change. Using this technique, we found the maximum noise temperature that could be added to the image, and still have most of the features recognizable, was 12K. This type of analysis allows us to specify system noise temperatures for future systems.

We now compare the simulation and experimental results. The comparison will be primarily qualitative, because the imager is not yet radiometrically calibrated. Also, for a real physical scene, comparing the theoretical and experimental values pixel-by-pixel is impractical, because the assumed material properties will be somewhat different than the real values. The theoretical and experimental images have a similar appearance, with respect to their resolution, and the reflectivity of the boat in the water. Table 1 presents some values from Fig. 1; these values are averages over regions in the scene. The trends of the temperatures in Table 1 with respect to vertical and horizontal polarizations agree; the temperatures for the boat and for water drop considerably in the horizontal polarization as compared to the vertical polarization. This drop occurs because of increased reflectivity in the horizontal polarization, which moves temperatures closer to that of the cold sky.

Table 1. Scene temperatures (K) from Fig. 1, at different locations and polarizations.

View Table

The sky temperatures are considerably different in the simulation and experiment. The experimental sky temperatures show a gradient as we scan from the horizon towards the vertical, and in the simulation, we did not assume such a sky gradient. As the sky temperature has a significant effect on the scene appearance, the incomplete model of the sky in the simulation could be a contributing factor to the wide quantitative difference between the simulated and experimental images. In Fig. 4, we study the sky temperature. Figure 4(a) shows a visible picture of the sky, and Fig. 4(b) shows the MMW image that spatially corresponds to the bounding box in Fig. 4(b). In Fig. 4(c) we plot the temperature along the red line of Fig. 4(b). The result of Fig. 4(c) shows that the temperature decreases linearly with angular elevation.

 

Fig. 4. (a) Visible image of sky (b) MMW image of sky. The region of sky corresponds to the bounding box in Fig. 4(a). (c) Plot of the temperature as a function of angle, along the red line of Fig. 4(b).

Download Full Size | PDF

We compared the experimental and theoretical modulation transfer function (MTF) [32]. Using Zemax, we calculated the geometric MTF and the Huygens diffractive MTF for the Cassegrain design (Fig. 5(a)). The point spread function (PSF) was rotationally symmetrical. The MTF is computed at the image plane, and then this result is multiplied by the effective focal length (EFL) of 109cm, as computed by Zemax, to obtain the MTF in angular units. Taking the Fourier transform of the MTF, we found that the PSF has a FWHM width of 5.3mrad. We calculated the FWHM of this curve is related to the Airy criterion by a factor of 0.84, so the diffractive curve would correspond to an angular resolution of 6.3mrad. This agrees with the Airy calculation of 1.22λ/D=6.5mrad. The diffractive curve is slightly higher than the geometric curve, indicating the imager is not diffraction-limited, which may be improved in future designs.

From the experimental images we determined the imager resolution. We took a horizontal cross section at a boat edge of Fig. 1(f), at the vertical angle of 3°. The derivative of the line gives the LSF (line spread function), and then the magnitude of the Fourier transform of the LSF gives the MTF [32].

Figure 5(a) shows that the theoretical diffractive MTF is a factor of 4 broader than the experimental MTF, and also the geometrical MTF is broader than the experimental MTF. In regards to the geometrical MTF, it is possible the adjustment on the secondary distance was sufficiently accurate. However, the width of the experimental PSF (point spread function), shown in Fig. 5(b) results in an Airy resolution of 7.4mrad, which is comparable to the theoretical diffraction limit.

 

Fig. 5. MTF curves.(a) (Red) experimental MTF along the x-axis (Black) Theoretical MTF (Green) Theoretical diffractive MTF. (b) The PSF (point spread function) is calculated by taking the Fourier transform of the MTF

Download Full Size | PDF

 

Fig. 6. (a) Visible image of the trawler boat (length: 22m), side angle. (b) Passive millimeter-wave image corresponding to 6(a); detector vertically polarized. (c) Visible image of men in a rigid-hull inflatable boat (RHIB)- boat length is 3.7m. (d) Passive MMW image corresponding to 6(c). Distance to target is 30m for images 6(a)–6(d).

Download Full Size | PDF

Figure 6(a) shows the visible image of the side view of a trawler, and Fig. 6(b) shows the corresponding MMW image, in the vertical polarization. The trawler length is 22m. Boat details such as the metal railing and windows can be clearly seen. Figures 6(c) and 6(d) show the visible and (vertically polarized) MMW image of men in a rigid-hull inflatable boat (RHIB); the men can be clearly seen in the MMW image. The RHIB length is 3.7m.

5. Conclusions

In this paper we report on a portable passive 95GHz single-pixel imager for remote sensing,. The imager was designed in a Cassegrain configuration, and acquired a 15×15° image in 4 minutes. The limitation on the image acquisition time was the mechanical rastering speed of the gimbal work.

The imager uses a radiometer assembled from the HRL chipset. The sensitivity of the radiometer was 0.6K, which was very suitable for imaging outdoor scenes, which have temperature differences exceeding 10K. Still, drift was an issue with the radiometer, as well as electronic noise from sources such as the motorized mount. The low output voltage from the radiometer required the use of a high-sensitivity data acquisition system. The images presented in this paper show relative temperature, rather than absolute temperatures.

We used the imager to acquire images of boats. The imager was found to have a NEDT of 0.9K, over the image acquisition time of 4 minutes. From the experimental data, we analyzed the resolution and the MTF, and found that the imager resolution was 7.4mrad (0.4degrees), which is slightly worse than the theoretical diffraction-limited resolution of 6mrad (0.34 degrees). Image appearance as a function of NEDT was studied to find the necessary noise performance of an imager. For the various scenes, the image could still be recognized with a noise per pixel ranging from 10–20K. This suggests that 1–2K is a reasonable value for a remote sensing millimeter-wave imager.

Future plans for the MMW imager include decreasing the image acquisition time, through the use of multiple detectors and modified opto-mechanical designs. We are investigating methods to absolutely calibrate the radiometer during image acquisition. We are developing a video amp after the radiometer, which will reduce drift and thereby allow us to use lower-sensitivity data acquisition electronics.

In previous work [10] we simulated MMW images, to explore the utility of MMW imagery and develop specifications on MMW imagers. Since the radiometer was not radiometrically calibrated and also real scenes have variations in material properties and scene temperatures, the comparison between simulation and experiment was primarily qualitative rather than quantitative. The simulated images and the experimental images did have a very similar appearance, thereby supporting the use of the simulation to represent MMW imagery.

The impact of atmospheric conditions plays a significant role in determining the appearance of MMW images. In our simulation, we assumed a constant temperature for the sky. But our experimental data showed that the sky has a MMW temperature gradient, decreasing linearly with angular elevation. It is likely that such data on the sky temperature must be obtained through experimental observation at MMW frequencies, rather than from simulations or analysis of visual images. In future work, we will incorporate experimental data on the MMW sky temperature, as a function of angle, into our MMW scene simulator.