Radiometer Aids Fire Detection

The receiver is basically a Dicke radiometer, a configuration preferred to a total power radiometer architecture for its improved performance in terms of enhanced achievable resolution.⁴ Additional features have also been added to the basic Dicke configuration, such as an amplified detector. Compared to a superheterodyne receiver (or standard Dicke architecture), this direct-detection receiver does not require a downconversion mixer or local oscillator (LO) for signal translation. As a result, the receiver design exhibits lower noise temperature and power consumption, simpler structure, and reduced costs compared to a superheterodyne receiver.

Moreover, with respect to a Dicke radiometer architecture, an RF isolator is connected between the output of RF switch SW1 and the input port of low-noise-amplifier (LNA) AMP1. This minimizes variations in the noise figure of amplifier AMP1 due to the effects of the input load (the receiver connected to the antenna or to a reference load), for improved radiometer performance in terms of measurement accuracy. Active filters are used in the video section of the receiver, allowing for low-cost amplification of received modulated input signals. Based on this architecture, a Ku-band radiometer prototype operating at a center frequency of 13 GHz was been designed and assembled. The switching rate, f_s, is set to 80 Hz.

The radiometer was assembled entirely with standard commercial RF/microwave components operating at Ku-band frequencies, except for bandpass filter BP1. The main features of these components are listed in the table. All of the standard components have been selected based on a previous performance analysis of the radiometer receiver as a whole. Of primary importance are the low-loss input switch and the low-noise amplifier (LNA). Moreover, the amplifier gain has been selected to provide the required radiometer resolution with a feasible radiometer integration time.

The radiometer’s amplified bandpass filter (BP1) consists of a band-limited amplifier (AMP2) cascaded with a bandpass filter (BP2); both are centered at 80 Hz, the receiver switching frequency (f_s). The amplifier [Fig. 2(a)] has a 160-Hz 3-dB bandwidth centered at f_s. Active bandpass filter BP2 is a second-order Chebyshev circuit [Fig. 2(b)], realized by cascading two multiple-feedback bandpass cells. The filter is designed to provide 0.5-dB pass band ripple, 21.5-Hz equiripple bandwidth, and 60-dB peak gain at the center frequency. Figure 3 shows the measured transfer function of this amplified BPF, with experimental measured results compared with the filter’s computer-simulated response. The measured amplitude gain is 97 dB at the 80-Hz center frequency.

The radiometer’s synchronous AC/DC demodulator is obtained by cascading a unity-gain inverting amplifier to a quadruple single-pole single-throw (SPST) switch (Fig. 4). The four switch ports are connected and controlled in such a way that the device output voltage for one-half period of the control signal is equal to input voltage V_IN and, for the following one-half period of the control signal, is equal to the output of the inverting amplifier. The switches are controlled by two complementary transistor-transistor-logic (TTL) signals at 80 Hz. 

Lowpass filter LP1 is an active Butterworth third-order filter with 10-Hz cutoff frequency. It employs one second-order Sallen & Key cell cascaded with a first-order passive cell (Fig. 5). The second-order cell has equal R and C. Thus, the resulting filter presents nonunitary DC gain—2 V/V, in the current design. Figure 6 shows the measured and simulated filter amplitude gain; the filter exhibits roughly 6-dB amplitude gain and 10-Hz cutoff frequency.

The complete radiometer prototype (Fig. 7) comprises all RF and video devices already described. Additional components complete the receiver: an ADC with sampling frequency of 50 Hz connected at the output of the LPF, as well as the circuitry for the generation of modulating square-wave LO1 that controls both the input RF switch and the synchronous demodulator. A matched load is used as reference source.

As a first step, the sensor was characterized in terms of transfer function (linear relation between the input noise power/ antenna noise temperature and the output voltage mean value) and the associated resolution (standard deviation of the radiometer output voltage—i.e., standard deviation of the measured antenna noise temperature or radiometric resolution). This was done by connecting the antenna input port to a calibrated noise source while the radiometer output voltage, V_OUT has been acquired with the ADC. Several acquisitions were performed for different values of input power (P_RF) in the range from -102 to -82 dBm (in this context, the input power is calculated as the product of the input power noise density times the noise bandwidth). The lowpass filter output is acquired by an ADC (with total dimensions of 50 × 30 cm).

Measured results are shown in Fig. 8, where the measured output voltage mean values are plotted versus test input P_RF. By interpolating the measured data, a linear (as expected) relation in between the input power and the output DC voltage has then been extrapolated. The obtained radiometer transfer function is given in Eq. 1:

V_DC(mV) = γP_RF(mW) + V₀ = 1.13 × 10¹¹ × P_RF(mW) + 556.6    (1)  

Given the radiometer calibration data obtained, it is possible to determine a linear function relating the antenna temperature T_A and the measured receiver output, V_OUT, when the radiometer input is connected to the antenna to observe the surrounding scene. In particular, V_OUT is proportional to T_A through the Boltzmann constant, the receiver bandwidth, and the calibration factor, γ of Eq. 1. From Eq. 1, one can conclude that, for the realized prototype, a 1 K increase in the antenna temperature corresponds to a 3.9-mV increase in the radiometer output signal.

Based on the same measurement setup, the radiometric resolution (standard deviation associated with the estimated antenna temperature) has also been measured by first calculating the standard deviation of V_OUT and then by considering the linear relation between the measured output signal and the input antenna noise temperature (T_A, as in Eq. 1). With the current approach, a 1.2 K radiometric resolution has been measured for 300 K antenna noise temperature. With digital filtering strategies, the obtained radiometric resolution can be further reduced: by applying a mobile average filter (with filter order N = 50) the radiometer sensitivity has been decreased down to 0.3 K.  The ADC sampling frequency is 50 Hz; a 50-order filter corresponds to a 1-s integration time.

The radiometer was then tested in an open-space environment, using the measurement setup of Fig. 9. A 20-dBi Ku-band horn antenna was connected to the radiometer input. The antenna was placed at roughly a 1-m height and pointed toward the fire spot, with fire area of 0.09 m² (30-cm-side square). The horizontal distance between the fire and the instrument was 3 m. The experimental results are shown in Fig. 10. Raw data were acquired with the ADC and then filtered with a 50-order mobile average digital filter (both raw data and filtered acquisitions are plotted in Fig. 10).

A maximum increment in radiometer output voltage of 80 mV was observed when the fire was at maximum (or 20 K in terms of radiometric contrast based on Eq. 1). This value is compatible with the predicted radiometric contrast for the current measurement setup,¹ confirming the feasibility of the proposed approach for fire-detection applications. Fire extinguishing was recorded during a subsequent acquisition. As Fig. 10 shows, as the fire goes out, the output signal (starting from a maximum value) slowly decreases to the value corresponding to the absence of fire. Recorded local increments correspond to temporary increase in the fire area due to the presence of wind.

In summary, the Ku-band radiometer prototype exhibits radiometric resolution of 0.3 K for a 300-K antenna temperature, comparable with the performance of similar devices.¹ Improved resolution may also be possible with increased integration time. Preliminary experimental results have indicated the usefulness of the instrument as a fire sensor: a fire spot of 0.09 m² at a 3-m distance has been detected with a 20-dBi horn antenna, with corresponding increment in the radiometer output signal of 80 mV.^2-5

The radiometer can be improved and additional features need to be implemented in the sensor. First of all, the dimensions of the device may be reduced. For the RF circuitry, rather than using commercial components (isolator, switch, or RF amplifier), in-house developed devices may be considered: this would allow for miniaturization as well as a cost reduction. Moreover, temperature monitoring still needs to be implemented in the instrument.

Finally, the absolute accuracy of the instrument as temperature sensor has not been evaluated. For fire-detection applications, however, this parameter is of minor importance as far as the increase in the antenna noise temperature rather than its absolute value should be properly measured. In short, these experimental results confirm the feasibility of the proposed design approach and represent a first step toward the realization of low-cost, high-performance remote temperature sensor suitable for surveillance and industrial applications.

G. Bianchi, R&D Engineer