Design and Development of an Electronic Board for Supporting the Operation of Electrochemical Gas Sensors

Abstract

Air quality monitoring is performed by agencies using instrumentation based on extremely reliable technologies but characterized by high costs. An alternative gas sensing technology is the electrochemical gas sensor which, even though having a lower accuracy, offers some advantages, such as low costs and high miniaturization. Among the gas sensors designed for air quality monitoring, the most interesting are the ones based on electrochemical cells. To operate such sensors, it is necessary to have an electronic circuit typically implemented on electronic boards provided by the sensor manufacturer. The research described in this document regards the design and implementation of an electronic board to support the operation of the “B” series of the electrochemical gas sensors produced by Alphasense. This brand provides electronic boards that, on one side, are capable of offering excellent performances, but on the other side, are characterized by some limitations, such as the possibility of using only one sensor at a time. The experimental activities of our laboratory in the field of real-time air quality monitoring by using low-cost devices and technologies demand electronic boards to support the operation of such sensors having a higher grade of flexibility. To overcome this and other limitations, a new electronic board has been designed and implemented. In this document, its design and the implementation details are described.

1. Introduction

1.1. Background and Motivations

It is well proven that air pollution represents one of the main threats to public health in almost every country of the globe. As stated by the World Health Organization (WHO), almost the entire global population (99%) breathes air that exceeds the WHO air quality limits and threatens their health [1]. For these reasons, air quality monitoring is crucial for adopting policies aimed at reducing the risks to human health related to the air pollution issue. Currently, air quality monitoring is mainly performed by government authorities that generally use instruments like chemical analyzers for this purpose. Although such instruments are capable of providing reliable data concerning the concentration levels of the main pollutants, they are characterized by several disadvantages: bulkiness, maintenance requirements, and lack of portability are the most evident. In addition to this, they are quite expensive (their prices range between EUR 5000 and EUR 30,000) as documented in several previous studies [2,3,4,5,6,7].

To address these issues, in recent years, the adoption of technologies alternative to those featuring chemical analyzers has allowed to development of Low-Cost Air Quality Monitors (LCAQMs) [2,3,4,5,6,7,8,9]. These instruments are based on Low-Cost miniaturized gas Sensors (LCSs); for this reason, they are characterized by high portability grade and low power consumption, and also, they are capable of providing real-time measures of the air pollutant concentrations [4,5,6,7]. The flip side of the LCSs, and consequentially of the LCAQMs, is given by their low accuracy, if compared with the performance provided by the chemical analyzers [2,3,4,5,6,7,8,9,10,11,12,13]. To reduce the gap existing between LCAQMs and the chemical analyzers in terms of accuracy, several techniques have been tested [2,3,4,6,8,10], and this issue still represents an area of investigation for the scientific community [3,8,10].

Among the various types of LCSs available on the worldwide market, the most promising are represented by electrochemical sensors. In particular, the ones produced by the Alphasense brand results to be among the most used and investigated [3,6,7,8,9,10,12,14,15,16,17,18,19,20]. Any LCS needs an Electronic Support Board (ESB) for its use. As a matter of fact, to properly operate it, the LCS must be mounted on the dedicated ESB, whose main tasks can be summarized as properly powering the LCS and interfacing the electronic signals provided by the sensor with the main electronic board of the LCAQM.

In general, any LCS manufacturer provides the proper ESB for each LCS model produced; therefore, Alphasense provides various ESB models specific for each series of LCS produced [21]. As stated by the manufacturer, various electrochemical sensors belonging to the “B series” are capable of measuring low concentrations of air pollutant gas, if used with the related ESB (called ISB by Alphasense) [21]. The research activities carried out by our laboratory are focused on the investigation of the use of LCSs for the measurement of low levels of various air pollutants; thus, the “B series” of Alphasense sensors falls in the area of our research activities. Even though the ISB provided by Alphasense guarantees the use of the dedicated sensors with excellent performances, we found that our specific activities need an ESB for the Alphasense “B series” sensors having additional features not provided by the ISB. For this reason, we decided to develop a new ESB, called Alphasense B4 multisensor board (AB4mb), whose design and implementation are described in this article.

1.2. Similar Devices

To the best of our knowledge, only two boards are available in the worldwide market to support the operation of the Alphasense “B series”: the ISB board manufactured and distributed by Alphasense [22] (see Figure 1), and the “Gases Pro Sensor board” developed by the “Iscape project”, and a part of the “Smart Citizen Kit 2.0” (see Figure 2) [23,24].

As can be found in the ISB board manual [22], this device is capable of operating just one sensor at a time. Unfortunately, the electric schematic of the board is not published by the manufacturer, but it is delivered to the users along with the board when it is purchased. As indicated by the sensor manufacturer in the application note AAN 105-03 [25], when an electronic board is designed to operate the “B series” sensors, it is preferable to put a shorting JFET transistor between the “working” and the “reference” electrodes of the sensor in the electric circuit for implementation. The reason for this design choice is that the transistor automatically will electronically short the two electrodes when the circuit is switched off. In this way, the two electrodes of the sensor will be always at the same electric potential, allowing a ready use of the sensor when the power is switched on. Without the presence of this transistor, the sensor will take a few hours to stabilize, as explained in the AAN 105-03 application note provided by Alphasense. If we inspect the electric diagram of the ISB board, the JFET transistor is not present. Another feature characterizing the ISB board is given by the amplification of the sensor output electronic signals that is fixed by the manufacturer, and it cannot be modified by the user for various purposes.

Regarding the “Gases Pro Sensor Board”, it can be used with three sensors at a time, but, as it can be found by viewing the electric schematic of the board [24], it is not provided by the JFET shorting transistor. Moreover, the level of the amplification of the sensor output analogic signal is fixed, although the A/D (analogic to digital) converter present in the circuit allows their digital conversion and amplification.

Both the two boards provide excellent performances, but in order to fulfill some laboratory requirements and add more flexibility, a new board (the AB4mb board) was designed and implemented.

2. Hardware Design

The AB4mb is capable of being used with four sensors at a time, it is provided by the shorting JFET transistor for the ready use of the sensors when the device is powered, and also, the modification of the amplification levels of the sensor output signals is allowed, thanks to the resistive trimmers present in the amplification blocks, and their core constitutes of the operational amplifiers in trans-impedance configuration (see Figure A1, Figure A2, Figure A3 and Figure A4 in the Appendix A). To implement these features and functionalities, the board design was performed by using the “Orcad 10.5” CAD (Computer-Aided Design).

Another feature characterizing this board is represented by the possibility of fine-adjusting the “zero” levels of the output signals through the precision resistive trimmers, as shown in Figure A1, Figure A2, Figure A3 and Figure A4. Adjusting the amplification of the sensor signals and their “zero” levels allows the use of the board in both heavily polluted environments and in all those situations featured by an averagely less polluted environment. As a matter of fact, when the board is used for measuring low gas concentrations, a greater amplification is beneficial to increase the resolution of the device; on the contrary, when it is used to measure high pollutant gas concentrations, it is favorable to limit the amplification level and properly adjust the electronic “zero” of the sensor signals in order to prevent the clamping of the device outputs, which brings to erroneous measurements.

The electronic circuit of the board can be schematized in four sections comprising all the electronic components and their wirings for correctly operating each of the four sensors that can be mounted on the board. Each of these sections is in turn composed of two main blocks whose cores are, respectively, represented by the ADA4051 operational amplifier and by the AD8609 operational amplifier. The first block is the potentiostatic circuit designed by following the indications provided in the AAN 105-03 application note. This block is necessary to ensure that the proper electronic currents flow through the sensor electrodes, and it is also characterized by the JFET transistor devoted to automatically shortening the two electrodes of the sensor. The second block is composed of two amplification chains, each of them formed by two operational amplifiers integrated into the AD8609 chip. Particular attention should be paid to the operational amplifier selection: as highlighted in the AAN 105-03 application note, the potentiostatic circuit must use a device having a low input bias current (<5 nA), a low offset voltage (<100 µA), and at least 2.2 V of voltage output swing. These requirements are fulfilled by the ADA4051 device, which is characterized by a top input bias current equal to 50 pA, a maximum offset voltage equal to 15 µA, and a rail-to-rail input/output capability, which means an output swing equal to 3.5 V. Regarding the amplification chains, it is recommended to use an operational amplifier having a low electronic noise and a stable unity gain. Both these characteristics were found to be present in the AD8609 device, which was therefore selected for the board design. This operational amplifier is dedicated to amplifying the weak currents coming out from the “working” and the “auxiliary” electrodes of the sensors. The functions of these two electrodes can be, respectively, summarized as providing a current proportional to the concentration of the gas experienced by the sensing element (the “working” electrode), and compensating the variations in the “zero” level of the sensor (the “auxiliary” electrode).

Although, in the previous figures, the symbols of the NO₂-B43F, OX-B431, CO-B4, and SO₂-B4 sensors are displayed, the AB4mb can be used with other sensors belonging to the “B-series” of the Alphasense manufacturer. The complete list of sensors belonging to the “B-series” usable with the AB4mb board is described in

Table 1. The list of the Alphasense sensors usable with the AB4mb board.

Sensor Model	Target Gas
NO2-B43F; NO2-B1	NO2
OXB-431	O3
CO-B4; CO-BX; CO-B1; CO-BF	CO
SO2-B4; SO2-BF; SO2-BE	SO2
VOC-B4	VOC
H2S-B1; H2S-B4; H2S-BH; H2S-BE	H2S
H2O2-B1	H2O2
HCN-B1	HCN
PH3-B1; PH3-BE	PH3
HCL-B1	HCl
H2-BF	H2
CL2-B1	Cl2
NO2-B43F; NO2-B1	NO2
OXB-431	O3
CO-B4; CO-BX; CO-B1; CO-BF	CO
SO2-B4; SO2-BF; SO2-BE	SO2

.

The sensors listed in Table 1 are a part of the “B-series” sensors produced by Alphasense. As a matter of fact, there are some sensor models that need a bias potential on the “working” electrode to correctly operate, which cannot be used with the AB4mb board, because its electronic circuitry is not designed to provide such feature. This is the case, for example, of the NO-B4 sensor, designed to measure NO concentrations.

A minimum voltage of 3.5 V is required for powering the board, while the maximum voltage is 5 V. This last value is due to the maximum voltage required by the AD8609 operational amplifier used in the circuitry of the board. The total current drawn by the device is less than 3 mA. The powering voltage range and the total current consumption make this board suitable to be integrated into portable monitors or battery-powered devices.

Another goal pursued in designing this board is represented by the minimization of its size. A suitable level of miniaturization can enhance the level of portability of the air quality monitors or instruments using this board. For this reason, the design of the Printed Circuit Board (PCB) is based on the Surface-Mounted Technology (SMT), and the electronic parts used to implement the AB4mb are Surface-Mounted Devices (SMD) (see Figure 3). As shown in Figure 3, the PCB implemented for the board is a double-layer PCB, whose top layer hosts the sensor sockets and the regulation trimmers for better accessibility.

The “auxiliary” and “working” currents of the four sensors are amplified and translated into eight analogic voltage levels that constitute the output signals of the board. The location of the pins related to these signals, along with the power and ground pins are shown in Figure 4. The output signals reflect the concentrations of the gases experienced by the sensors mounted on the board.

By referring to Figure 4, the meaning of the labels indicating the output signals has to be intended as follows: the minus symbol (−) refers to the “auxiliary” electrode of the sensor, while the plus symbol (+) refers to the “working” electrode of the sensor. So, for example, the pin S4+ is the output voltage related to the “working” electrode of the sensor mounted in the S4 socket.

The capability of regulating the amplification levels of the sensor output currents is ensured by the resistive trimmers, whose position on the board is shown in Figure 5.

3. Build Instructions

The number of electronic parts constituting the AB4mb is 148, as can be seen in the bill of materials. If we consider that the PCB is a double-layer type, it appears very clear that the best way to build the AB4mb is to resort to the services offered by various PCB manufacturers and assemblers available on the internet.

The AB4mb files necessary for ordering the PCB from a manufacturer can be downloaded from a repository [26], and they are also available in the Supplementary Materials provided with this manuscript. At the time of writing this article, the estimated cost to assemble a copy of the AB4mb board can be quantified as EUR 85.

4. Operating Instructions

4.1. Preliminary Operations for the First Use of the Sensors

For the proper use of the device, we must distinguish two different cases: the case concerning the use of the board when the sensors present on it are going to be used for the first time, and the case when the sensors present on the board were already used other times in the past. In the second circumstance, it is sufficient to power the board through a stable source whose voltage is between 3.5 V and 5 V, minding to not exceed the latter value in order to prevent the damaging of the operational amplifiers present on the board. After powering the AB4mb, it is advisable to wait for a few minutes due to the stabilization period needed by the sensors. Once the stabilization period finishes, we can consider the readings of the device outputs as reliable and effectively use the device.

In the first case, namely, when we are going to mount on the board a new sensor, some operations are necessary for effectively using the device, before powering it. The first thing to do is setting of the amplification levels of the output currents of the sensors selected to be mounted on the board. Setting the amplification levels means fixing the resistive values of the resistive trimmers shown in Figure 5. The purpose of the various resistors described in Figure 5 is schematically summarized in

Table 2. The list of the variable resistors for setting the amplification levels of the sensor currents.

Trimmer Name and Alias	Function
R17 (G1−)	Setting the gain of the S1 auxiliary signal
R10 (G1+)	Setting the gain of the S1 working signal
R35 (G2−)	Setting the gain of the S2 auxiliary signal
R28 (G2+)	Setting the gain of the S2 working signal
R53 (G3−)	Setting the gain of the S3 auxiliary signal
R46 (G3+)	Setting the gain of the S3 working signal
R71 (G4−)	Setting the gain of the S4 auxiliary signal
R65 (G4+)	Setting the gain of the S4 working signal

.

By using a screwdriver and an ohmmeter, the user can set the resistor to the desired value and check its real value by putting the ohmmeter probes on the points indicated in Figure 5. To obtain an indicative value of the resistance to set, the user can calculate it by Equation (1):

$$R=\frac{\mathrm{\Delta }{V}_{\max }}{5S_s{C}_{\max }}-{10}^5,$$

(1)

where R is the resistive value to set expressed in ohms, S_S is the sensitivity of the sensor, C_max is the maximum expected concentration of the gas to measure, and ΔV_max is the output voltage maximum swing of the AD8609 operational amplifier, which is equal to the power voltage in good approximation, with AD8609 being a rail-to-rail device. As an example for better clarifying Equation (1), we can use the CO-B4 sensor designed to measure the CO concentrations. By reading the CO-B4 datasheet published by Alphasense, we find that the mean sensitivity (S_S) of the CO-B4 sensor can be considered equal to 500 × 10⁻⁹ A/ppm, and also, we can suppose that the power voltage provided to the AB4mb is 3.5 V. If the maximum expected concentration of CO gas in the environment where the AB4mb and the CO-B4 sensor will be deployed is equal to 10 ppm, therefore, Equation (1) gives us a value equal to 40 Kohm. Thus, the user will set the R28 and R35 resistances to such values (see Table 2, Figure A2 and Figure 5). As mentioned earlier, the value given by Equation (1) represents just an indication of the final value to set. The reason is due to the fact that the sensitivity of the electrochemical sensors can significantly vary from piece to piece. As a matter of fact, the manufacturer provides a range of possible sensitivity values in the datasheets of each sensor model. The final resistance can be found by fine adjusting the initial value after empirical evaluations.

Once the resistance values of the resistors listed in Table 2 have been set, the sensors can be mounted on the board sockets (see Figure 3). It must be noted that each sensor model listed in Table 1 can be mounted in any of the sockets shown in Figure 3, even though each socket is labeled by a specific sensor model. After inserting the sensors in their socket, the user can power the device by applying a voltage between 3.5 V and 5 V to the “Vcc” and “gnd” pins shown in Figure 4.

After powering the AB4mb, it is necessary to wait for the stabilization of the sensor signals. If one or more sensors have been just mounted, or used for the first time, the stabilization period can take a couple of hours, or something more. Unfortunately, the sensor manufacturer does not specify this parameter, but based on our experience, in general, it can take up to four hours. It is advisable that during this period, the device is placed in environments characterized by a low concentration of the “target” gases (namely, the gases for which the sensors in use are designed). Once the output signals have been stabilized, the “zero” levels can be set by acting through a screwdriver on the precision resistive trimmers listed in

Table 3. The list of the precision resistive trimmers dedicated to set the “zero” levels of the output signals. To facilitate their use, the effect of the clockwise screw turning is reported.

Trimmer Name and Alias	Function	Screw Clockwise Effect
R19 (B1−)	Setting the zero level of the S1 sensor auxiliary signal (denoted as S1−)	Decreasing the S1− voltage level
R14 (B1+)	Setting the zero level of the S1 sensor working signal (denoted as S1+)	Increasing the S1+ voltage level
R37 (B2−)	Setting the zero level of the S2 sensor auxiliary signal (denoted as S2−)	Decreasing the S2− voltage level
R31 (B2+)	Setting the zero level of the S2 sensor working signal (denoted as S2+)	Increasing the S2+ voltage level
R55 (B3−)	Setting the zero level of the S3 sensor auxiliary signal (denoted as S3−)	Decreasing the S3− voltage level
R49 (B3+)	Setting the zero level of the S3 sensor working signal (denoted as S3+)	Increasing the S3+ voltage level
R31 (B4−)	Setting the zero level of the S4 sensor auxiliary signal (denoted as S4−)	Decreasing the S4− voltage level
R67 (B4+)	Setting the zero level of the S4 sensor working signal (denoted as S4+)	Increasing the S4+ voltage level

and shown in Figure 4.

The “zero” level of an output signal is represented by the output voltage given when the sensor is described to a very low concentration of the gas, or near zero. Its exact value is at the user’s discretion, but some considerations must be made concerning this aspect. It has to be noted that some Alphasense sensor models, such as the NO₂-B43F designed for measuring the NO₂ gas, provide output currents proportional to the gas concentration, while, other models, such as the CO-B4 designed for measuring the CO gas, provide output currents inversely proportional to the gas concentration. The output signals of the AB4mb reflect the trends of the sensor currents; therefore, it is worthwhile to set the zero close to the 0 V (or ground) in the first case, while it is advisable to set it close to the power voltage in the second case.

4.2. Calibration of the Device

As earlier explained, the output signals of the AB4mb are related to the concentrations of the gases to which the sensors are described, but an appropriate algorithm, mathematical law, or calculation is necessary to translate the output signals into data expressing gas concentrations; in other words, it is necessary to calibrate the system composed of the AB4mb and the sensors mounted on it.

Various procedures and methods are viable for calibrating the device with various levels of complexity. As a matter of fact, we must consider that the currents provided by the sensors are influenced not only by the “target” gas but also by interfering factors, such as temperature and humidity variations, sensor drift, or the presence of interfering gases, namely, other gas types to which the sensing element of the sensor is also sensitive [3,8,10]. All these factors constitute the source of the inaccuracies affecting the technology of electrochemical gas sensors. As mentioned in the Introduction Section, the scientific community has investigated and still continues to investigate [2,3,4,6,8,10] the techniques to mitigate the negative effects of such interfering factors. One possible approach can be given by the use of advanced algorithms, such as multivariate linear regressions, or the use of artificial neural networks, that take into account as inputs, temperature, humidity, and the concentrations of the interfering gases in addition to the signals provided by the sensors [6,10,15,16,17].

However, there are some cases in which temperature and humidity variations are very limited, and the expected concentrations of the interfering gases are very low, or near zero. This situation could happen when we are monitoring the CO gas concentration (for example) in apartments or flats. In these cases, the calibration of the device can follow a simpler scheme, avoiding the use of advanced algorithms, and considering just the AB4mb outputs as input variables as expressed by Equation (2):

$$C=G_A(S_{i+}-S_{i-})+C_0,$$

(2)

where C is the gas concentration, S_i+ is i_th (i = 1, 2, 3, 4) output “working” signal, S_i− is the i_th output “auxiliary” signal, C₀ is the “zero concentration” offset, and G_A is a coefficient that in first approximation can be calculated through Equation (3):

$$G_A=\frac{1}{S_S{R}_i}.$$

(3)

In this Equation, S_S represents the mean sensor sensitivity obtainable by reading the sensor datasheet, and R_i is the value of the variable resistance listed in Table 2. The condition under which Equation (2) can be used is that the value of the resistance concerning the “working” signal of a sensor on the board must be roughly equal to the one related to the “auxiliary” signal. To clarify this aspect, as an example, we can consider a sensor mounted in the socket S1 (see Figure 5); in this case, the user must set R17(G1-) and R10(G1+) (see Table 2) roughly at the same value. It could happen that the difference between “working” and “auxiliary” signals in Equation (2) is not zero when there is an absence of the “target” gas. In this case, the coefficient C₀ is necessary to correct this situation, and its value can be found by Equation (4):

$$C_0=-G_A(S_{i+}(0)-S_{i-}(0)),$$

(4)

where S_i+(0) and S_i−(0) are the output signals of the board when there is absence of the “target” gas.

Equations (3) and (4) can represent a theoretical approach for the calibration issue, but from a practical point of view, if we consider that the sensitivity parameters of the sensors can significantly vary from one piece to another one of the same sensor model, it appears clear that an experimental, or empirical, way is necessary to improve the accuracy of this calibration method.

Under the light of these considerations, the “three-point” calibration method represents a valid option, providing a relatively simple way to effectively calibrate the device. This procedure consists of exposing the sensors to three distinct known gas concentrations: C₂, C₁, and zero. If the difference between “working” and “auxiliary” output signals of the board is ΔS₂ and ΔS₁ at the concentrations C₂ and C₁, then, the coefficient G_A can be given by Equation (5):

$$G_A=\frac{C_2-C_1}{\mathrm{\Delta }{S}_2-\mathrm{\Delta }{S}_1}.$$

(5)

Obviously, the C₀ coefficient can be still calculated by using Equation (4).

5. Validation

The AB4mb device was tested both in the laboratory and during an experiment designed to measure the concentrations of O₃, CO, and NO₂ in an apartment occupied by a family composed of four individuals during their everyday lives [10]. In this experiment, the use of the AB4mb allowed us to understand the potentialities of different calibration algorithms, and at the same time, it acted as a test bed for the AB4mb device.

The sensors mounted on the board for this experiment were the OX-B431, the CO-B4, and the NO₂-B43F; moreover, the AB4mb was installed in the “SentinAir” portable air quality monitor designed and developed in our laboratories [9]. The monitor hardware was necessary to convert the AB4mb analogic outputs into digital data to store in a dataset useful to study the performance provided by four different calibration models: a multivariate linear regression, a random forest regression, a support vector machine, and an artificial neural network. The SentinAir monitor was also used to capture the data coming out from the reference instruments co-located with the sensors that were useful to measure the real O₃, CO, and NO₂ concentrations. The chemical analyzers used for measuring NO₂, CO, and O₃ were, respectively, the “405nm” by 2Btech [27], the “CO12m” by Envea [28], and the “106-m” by 2Btech [27].

The inputs selected for the calibration models were the gas sensor outputs, the temperature, the relative humidity, and the real gas concentrations provided by the reference instruments. The temperature and the relative humidity measurements were performed by using, respectively, the TC1047A sensor by Microchip, and the HIH5031 sensor by Honeywell [10]. These sensors are a part of a module previously installed in the SentinAir monitor [6].

The whole experiment duration lasted roughly eight days, during which, the measurements of the first four days, namely, the calibration period, were used to calculate the calibration model parameters, and therefore, to find the algorithms capable of providing the gas concentrations starting from the inputs earlier specified; while, in the last four days, namely, the validation period, the algorithms just found were used to calculate the real gas concentrations measured by the reference instruments. The main indicators used to understand the performance quality of the system given by the AB4mb, and the gas sensors calibrated by the different algorithms were the Mean Absolute Error (MAE), the Root Mean Squared Error (RMSE), and the coefficient of determination R² defined as follows:

$$MAE=\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}\left|m_i-r_i\right|,$$

(6)

$$RMSE=\sqrt{\frac{1}{N}{\displaystyle {\sum }_{i=1}^N}{\left(m_i-r_i\right)}^2},$$

(7)

$${\mathrm{R}}^2=\frac{{\left(\sum _1^N\left(m_i-\overline{m}\right)(r_i-\overline{r})\right)}^2}{\sum _1^N{\left(m_i-\overline{m}\right)}^2{(r_i-\overline{r})}^2}$$

(8)

In Equations (6)–(8), N represents the number of measurements, r_i is the i_th measure of the reference instrument, m_i is the i_th gas concentration predicted by the device under test, $\overline{r}$ is the average of the reference measures, and $\overline{m}$ is the average of the predicted values. The indicators just introduced were selected because they give us a complete idea about the potentialities of the device under test composed of the AB4mb and the sensors installed on it, to accurately follow the measurements of the reference instrument providing the real gas concentrations. The MAE and the RMSE are indicators providing the overall gap between the measures of the references, considered reliable and accurate, and the measures performed by the device under test. A good performance means MAE and RMSE values close to zero. The coefficient of determination (R²) improves the understanding of the performance of the device under test because it gives us a quantification of the capability of the AB4mb output signals to follow the trends of the reference measures. It expresses a dimensionless quantity that can vary from 0 to 1. Values ranging from 0.7 to 1 denote a good performance, while values ranging from 0.4 to 0.7 represent a moderate capacity to follow the accurate measures of the reference, and lastly, values less than 0.4 indicate a poor performance.

The results provided by this experiment showed that the different calibration methods produced different results, depending on the considered performance indicator, and the considered period of the experiment (calibration or validation period); therefore, it was not possible to find the best calibration method in absolute terms. For this reason, and for the sake of conciseness, the best results related to the validation period are summarized in

Table 4. Some of the performance indicator values related to the validation period of the test carried out to validate the hardware. MAE and RMSE are expressed in ppm for CO measures, otherwise in ppb for NO₂ and O₃.

Measured Gas	Calibration Method	R2	MAE	RMSE
CO	artificial neural network	0.924	0.377	0.366
NO2	multivariate linear regression	0.890	8.381	10.618
O3	artificial neural network	0.137	9.552	16.340

.

The performance indicators shown in Table 4 provide quantitative information about the final performance of the system under evaluation. Anyway, for a more complete picture of the achievable quality of data, it is useful to inspect Figure 6, Figure 7 and Figure 8. Regarding the temperature and the relative humidity measurements, they ranged, respectively, from 20.5 °C to 30 °C and from 23% to 60%.

In these figures, the plots of the time series concerning the measurements performed during the validation period by both the reference instruments and the device under test are described to add more completeness for understanding how close the measures of the system under evaluation are to the real gas concentrations.

By inspecting Figure 6, Figure 7 and Figure 8 and Table 4, it is possible to obtain an idea about the performance of the device under test compared to the reference instrument. In the case of CO measurements, if we compare the MAE and RMSE low values (0.377 ppm and 0.366 ppm, respectively) with the range of concentrations detected by the reference instrument (which is about 2.5 ppm), we note a good performance. This consideration is confirmed by the value expressed by the coefficient of determination (R² = 0.924) being very close to 1, which indicates a good capacity of the device under test to follow the trends of the real gas concentrations measured by the reference instrument. The same considerations can be drawn in the case of the NO₂ gas, while, for the O₃ measurements, we can note a low R² value. This datum indicates poor capacity to follow the reference measurements, which can be explained by considering two concurrent elements: the simultaneous presence of NO₂ and O₃ gases and the lack of a filter for the NO₂ gas on the ozone sensor. Indeed, the OXB431 sensor is equally responsive to O₃ and NO₂ gases, and this fact can be a source of error in ozone measurements. On the contrary, the NO₂ sensor is equipped with a filter for the ozone gas, and even though it is equally responsive to both O₃ and NO₂, it does not show any performance degradation in the case of simultaneous presence of both gases.

Making a comparison with previous studies represents a challenging task due to several factors. In the first place, we must consider that the performance assessment of LCSs or LCAQMs strongly depends on the test environment: laboratory test chamber, indoor, or outdoor environment. Moreover, the evaluation of LCSs or LCAQMs in previous studies is often performed by considering different indicators, different types of LCSs, or different calibration methods [3,10,29]. In addition to this, the values of the performance indicators used to evaluate the same device can vary by considering the calibration or the validation data [3,10], and also, they may depend on the concentration ranges experienced by the LCSs or LCAQMs. Anyway, by considering the R² indicator and excluding the cases of evaluations carried out in the laboratory (which is well known that they provide performance indicators way better than the ones found in any other environment), it is possible to make an indicative comparison that is summarized in

Table 5. A comparison concerning the performance of LCSs or LCAQMs in terms of R² indicator.

Work or Study	Pollutant Gas	R2
Karagulian et al. [3]	CO	0.15–0.91
	NO2	0.17–0.94
	O3	0.53–0.91
Kang et al. [8]	CO	0.4–0.8
	NO2	0.1–0.8
	O3	0.1–0.9
Rodenas et al. [29]	CO	0.38–0.94
	NO2	0.02–0.97
	O3	0.02–0.99
This work	CO	0.924
	NO2	0.890
	O3	0.137

.

The data described in Table 5 are drawn from some review articles that addressed the issue related to the performance assessment and the comparison of LCSs and LCAQMs [3,29]. As can be seen, we can find a very wide range of values for each pollutant gas considered. This aspect can be explained through the earlier mentioned various conditions and situations affecting the LCS or LCAQM assessment process. The so far described results of the validation test proved to us that the goals of the AB4mb design were achieved, and all our expectations were met.

6. Conclusions

The electrochemical sensors represent an attractive option for air quality monitoring activities by using portable and easy-to-use LCAQMs. In particular, the “B series” of the electrochemical sensors produced by Alphasense are characterized by a good quality/price ratio. Their operation is supported by electronic boards that, even though providing good performance, are characterized by a flexibility grade not corresponding to the requirements of the experimental activities of our laboratory. For this reason, the AB4mb has been designed and developed in support of such types of sensors.

The hardware designed for the purpose of allowing to shorten the stabilization period of the sensor output signals by the introduction of suitable electronic parts as suggested by the sensor producer. Moreover, the capability of regulating the amplification levels of the sensor electric currents and the regulation of the “electronic zero” were implemented in the AB4mb device. Another important characteristic of the AB4mb is given by its capability of the simultaneous use of up to four sensors. All the features of the AB4mb have been implemented by designing hardware offering the highest miniaturization level possible subject to its building costs.

This device was tested in the laboratory, but the most interesting results were found during an on-field experiment performed in the indoor environment to measure the concentrations of CO, NO₂, and O₃ by using different calibration methods. During the test, the device provided useful data without any failures, proving that the aims of the AB4mb design were completely fulfilled.

However, it needs to be acknowledged that, if, on the one hand, the AB4mb can offer a higher flexibility grade in its use by allowing the regulation of the amplification of the sensor signals and their “zero” levels, on the other hand, the same flexibility adds one more step of complexity due to the calibration process left in charge of the user, but necessary for the proper use of this device.