In the late 1970s, various gas sensors were developed based on a variety of operational principles. Compared with other gas sensors, SAW-based gas sensors provide a high degree of sensitivity, reproducibility, and stability [22]. The sensing mechanism is based on the fact that volatile organic compounds are adsorbed on the surface of a substrate and an increase in mass loading causes a shift in frequency. The mechanism behind SAW sensors involves the input of voltage inducing an electric field between the IDTs. The piezoelectric effect induces dynamic strain on the substrate to launch the SAW, propagating across the surface of the substrate. At the receiving end, the IDT converts the mechanical signals to electric signals. Finally, fluctuations in the output frequency are measured. According to the design parameters from the simulation data of delta function model and the cross field model shown in Table 1, LiNbO₃-based. SAW devices with IDTs of Cr/Au (20 nm/100 nm) were fabricated in a photolithography process with a center frequency of 117.4 MHz. The size of the SAW sensors was 3 mm × 2 mm.

A SAW array was employed to fabricate a gas sensor with the advantages of small size, low cost, high sensitivity, and rapid response. As shown in Figure 3, the 2 × 2 non-continuous working oscillators were controlled by a multiplexing technique as a switching element [23].

The film of the sensor operated like a smart skin, responsible for generating chemical signals from interactions between molecules and the film [24]. The sensing polymer film acted as the mucosa of the nasal cavity, playing an important role in the detection of gases. To enhance sensitivity and selectivity, seven polymers were selected as materials to be used in the sensitive film, spin-coated on the surface of the resonators of the array (Figure 4). These polymers included poly-N-vinylpyrrolidone (PNVP), poly-4-vinylphenol (P4VP), polyvinyl acetate (PVAc), polyethylene-glycol (PEG), polystyrene (PS), polystyrene-co-maleic anhydride (PSMA) and polysulfone (Psu).

In the E-Nose system, a sensor array was used to form patterns for use in the recognition of gases. In this study, the sensor array comprised four SAW sensor devices. These four sensors did not operate at the same time by controlling the oscillators non-continuously to minimize power consumption and interference from crosstalk [23]. In addition, by operating the SAW array non-continuously, all sensor outputs could be connected, thereby only one interface readout circuit was needed, saving chip area and total cost. In this study, the non-continuous SAW sensor reached a stable output within 0.2 s after being switched on. Therefore, each sensor was switched on for 1s to ensure stable sensor output.

There are many kinds of circuit implementations to detect SAW frequencies and/or phases. One way is to convert the SAW frequency into a voltage signal. This approach wastes a large portion of the voltage range because the shift in the frequency of the sensor does not exceed a certain amount [25]. Another way measures phase differences, but the resolution is only at the level of MHz [26]. In this study, the center frequency was at 117.4 MHz, and the maximum frequency shift did not exceed 1 MHz, representing as little less than 1% that of the center frequency. Therefore, if a direct frequency to voltage converter was adopted, assuming a linear conversion from frequency to voltage, approximately 99% of the voltage range was wasted, making it difficult to improve resolution.

In this study, we implemented a mixed-signal interface chip, as shown in Figure 5. The first stage of the chip was analog, including a mixer, low-pass filter, and comparator. The function of the analog stage was to modulate, filter, and convert the signal from the sensor into a square wave. The second stage of the chip was a digital readout to detect changes in sensor frequency. To read out the sensor signal (frequency change), the frequency of the sensor f_in was subtracted from that of a reference sensor f_ref. This resulted in a decrease in the frequency change f_ref–f_in compared to f_ref. Because the amplitude of the signal requiring conversion was smaller, resolution could be improved and power consumption could be reduced. The reference sensor was not coated with a polymer membrane; therefore, the variations in its frequency due to input gas were very small. Nevertheless, the reference sensor was still influenced by the same environment parameters as the sensing sensor, such as temperature and humidity. Subtracting f_in from f_ref helped to eliminate the background effect of the sensor.

We integrated a mixer into the frontend of the analog section, the output of which had a high frequency term f_ref + f_in and a low frequency term f_ref–f_in. After the mixer, a low-pass filter passed the low frequency term f_ref–f_in, to a comparator to generate a square wave output to the digital stage.

A digital frequency readout circuit received the output from the comparator and reported the frequency data. The main idea was to use three counters with D flip-flops (DFF) for storage, as shown in Figure 6.

Counter 2 provided a fixed sampling time T_S = 2¹⁷ × T_CLK2, where T_CLK2 was the clock period of CLK2. The input signal (Fin) went into Counter 3 as the clock. When Counter 3 counted for the time 2¹⁷ × T_CLK2, the MSB of Counter 2 triggered the DFF and stored the output data (D) of Counter 3 at this moment. Counter 1 generated control signals for the multiplexor (MUX), the reset signal of Counter 3, and the Read signal for the Octopus II. The multiplexor (MUX) output the data stored in the DFF. The input signal (Fin) could be calculated later by: (1) F in = D 2 17 × f CLK 2 = D T S

According to Equation (1), the resolution of the mixed-signal interface chip depends on the sampling time T_S. Theoretically, a finer resolution can be achieved by a longer sampling time. For example, a sample time of one second corresponds to a resolution of 1 Hz (each bit in Counter 3 represents 1 Hz), and a sampling time of 100 ms corresponds to a resolution of 10 Hz (each bit in Counter 3 represents 10 Hz). Since each sensor was switched on for 1 s, a sampling time of 1 s would be too long. Moreover, 10 Hz resolution was enough for detection. Consequently, a sampling time of 100 ms was chosen in this work.

The Octopus II [27,28] WSN platform was selected to transmit sensing data. Because the output of the frequency readout circuitry was not a fast-changing signal, it could be connected directly to the Octopus 50-pin extension connector. Octopus II includes MSP430F1611, a USB Interface, Inverted F, and SMA Type Antenna. The basic features of Octopus II are as follows: RF range is approximately 450 m, board size is 80 mm × 31 mm, maximum output power is approximately 10 dBm, compatible with IEEE 802.15.4 (ZigBee), and operates using 2 × AA batteries (3.3 V 2,700 mAh). Detailed information can be obtained from the website (http://www.wsnc.ntu.edu.tw/Files/Octopus--_0913_V1_2%20[----].pdf).

The sensor data was transmitted through a wireless module to a personal computer operating as a base station for data storage, processing, and analysis. To verify the E-Nose sensor node, three tests had to be performed: (1) a gas experiment to validate the SAW sensor array; (2) a configuration test to verify the accuracy of the mixed-signal readout chip; (3) a verification test to determine the correctness of the wireless transmission.

Figure 7 illustrates the gas experimental setup, in which seven different sensing membranes were used in the SAW sensor array. The array was tested using ethanol and acetone, and the complete system was tested in a 1L 4-neck bottle chamber. The four windows were for the vacuum pump, test gas input, wire connection, and air flow valve. Considering that the membrane may have absorbed excessive water vapors, we first baked the sensor array for 20 minutes at 90 °C, prior to measurement, to reduce the water vapor interference from inside the membrane. One cycle of gas testing involved four steps:

Pumping: The system was operated stably under vacuum pumping for 5 minutes.

The total time required for one cycle was approximately 15 minutes. The system responded very rapidly to the test gas causing a frequency shift within 3 minutes. To test the accuracy of the mixed-signal readout chip, the output was compared with a frequency counter.

The SAW array was tested using ethanol and acetone. Figure 8 shows a typical response of the sensors (coated with PMSA, PEG, and PNVP) exposed to ethanol. The experiment was repeated three times. The baseline seemed to drift due to temperature variation. The SAW devices are known to be sensitive to its environment parameters, especially temperature [29]. Currently, many SAW devices are manufactured with materials such as lithium niobate or lithium tantalite. The advantage of using these two materials is their high K² values, at the price of their high temperature coefficients. Our way to compensate the temperature effect is to use a reference SAW device without membrane.

To verify the accuracy of the interface chip, a signal generator was used to emulate the SAW signals. A spectrum analyzer measured the outputs of the mixer and the low-pass filter. The two input signals to the mixer were 117 MHz and 118 MHz sinusoidal wave with amplitude 0.5 V. As a result, the spectrum analyzer showed that the mixer output spectrum had two frequency components, 1 MHz and 235 MHz, and the low-pass filter output spectrum was 1 MHz. These results verified the function of the mixer and the low-pass filter.

The output of the low-pass filter was sent to a comparator to convert the signal into a square wave, which was passed to the digital frequency readout circuit. To test the frequency readout circuit, a square wave was input to the circuit and into a frequency counter at the same time. The test frequencies were 10, 100, 1 k, 10 k, 100 k, and 1 MHz. Each frequency was sampled 1,000 times. The total test time was 100 seconds. In all the test results, the frequency readout circuit outputs were the same as the frequency counter. The power consumption for the entire chip was 1.48 mW using a power supply of 3.3 V. Figure 9 shows a photo of the die used in this low-power mixed-signal SAW interface ASIC. Table 2 provides a benchmark with other studies of the SAW interface circuit [25,26].

Figure 10 shows the test setup of the sensor node. The SAW array was connected to the interface chip, and the output was connected to the wireless module. Both the interface chip and the wireless module were running on two 1.5 V batteries. The wireless module transmitted the sensor data to a base station PC. This data was compared with the frequency counter output according to Equation (2): (2) ( ∑ i = 1 n   A i − B i ) / n = errorwhere data list A is the frequency counter output, data list B is the data transmitted to the base station PC, n is the data number used for comparison. Table 3 summarizes the mean error and standard deviation of the two data lists. Both the mean error and standard deviation between the transmitted data from the sensor node and the frequency counter output are less than 4 Hz.