The original specifications of our mini-station were as follows:

• portable device (dimensions ≤15 cm × 10 cm × 5 cm and weight ≤500 g);

∘ Temperature (−10; 40) °C;

∘ Relative humidity (0; 100) %;

∘ Location;

∘ Acceleration detection (≥ 3 g) ms⁻²;

∘ NO (0; 1000) ppb approximately (0; 800) µgm⁻³;

∘ NO₂ (0; 1000) ppb approximately (0; 532) µgm⁻³;

∘ O₃ (0; 1000) ppb approximately (0; 500) µgm⁻³;

∘ PM_2.5 (0; 500) µgm⁻³;

• gas sensors based on the electrochemical technology;

These technical requirements were set to obtain a versatile device that could cover the typical indoor and outdoor concentration range for these pollutants, and be used both for stationary or itinerary applications, considering that one is rarely commuting more than 2 h.

Based on the requirements listed above, our partner HEPL (Haute École de la Province de Liège) chose to use:

• the Adafruit BME680 breakout for relative humidity, temperature and barometric pressure, which was chosen over the BME280 in the course of the project;

Both PM sensors are laser-based particle counters with a time response less than 10 s. We are not aware of a PM low-cost sensor based on another functioning principle than light scattering. The latter allows one to determine 1) the particle number concentration by counting the pulses of scattered light reaching the detector, 2) their size, which depends on the intensity of scattered light and 3) their shape, which depends on the spatial pattern of scattered light. The last two properties are usually only roughly estimated by PM low-cost sensors and remain the prerogative of higher end instruments. The particle diameter range goes typically from 300 nm to 10 µm and the number concentration range from 0 to 10⁶ particles/L. The Antilope can work either with the HPMA115S0 or the SPS30 sensor.

For the considered gases (NO, NO₂ and O₃) we wanted electrochemical rather than metal oxide sensors for they have a lower power consumption, typically 2 mW against 500 mW, they display a better selectivity and they are more stable; their biggest disadvantage is their life expectancy, typically 18 months against more than 10 years, and they are somewhat more expensive (Romain 2018). The working principle of the electrochemical gas sensors is an oxidation-reduction reaction that takes place at the working and counter electrodes of the sensor, which produces a limited current that is proportional to the concentration of the detected gas (Chou, 1999; Yi et al., 2015). For example, a NO₂ sensor might be the siege of the following reactions:

N O 2 + 2 H + + 2 e − ⇌ N O + H 2 O (reduction at the working electrode E = +1.15 V).

1 / 2   O 2 + 2 H + + 2 e − ⇌   H 2 O (oxidation at the counter electrode E = +1.25 V).

N O 2 ⇌ N O + 1 / 2   O 2 (overall reaction E = −0.10 V).

These equations were drawn from the Alphasense Application Note 107-06 (2009).

For common air quality applications, i.e. not in specific industrial environments, it would also be interesting to monitor NH₃ and VOCs, the first for its role in acidification, eutrophication and secondary particle formation, the latter because some of these species are carcinogenic, e.g. benzene, formaldehyde, and some of these species also play a role in new particle formation, e.g. isoprene, α-pinene.

For the Antilope all selected electrochemical sensors work in the (0; 20) ppm range with a warranty of performance. Each of them is delivered with its intrinsic sensitivity, zero current, auxiliary zero current and time response values. The sensitivity (or gain) is the factor that transforms pollutant concentrations into variations of the output current, it is expressed in nAppm⁻¹. The zero current (or offset) is the non-zero output current response in a pollutant-depleted environment, it is expressed in nA; the auxiliary zero current is basically the same but relates to the auxiliary electrode. Although this information is provided by the manufacturer on the basis of his calibration tests, field tests may highlight the necessity of adapting these parameters. Regarding the time response, all sensors received so far displayed a value less than 30 s.

Another important thing to keep in mind when working with electrochemical sensors is the necessity of maintaining a specific bias voltage between the working and the reference electrodes, e.g. +200 mV for the NO sensor, and 0 mV for the NO₂ and O₃ sensors. This bias voltage should be maintained even when the system is off, at the risk of degrading the sensor and delivering wrong measurements during the first hours of use after the system is switched on again.

Since the levels of current variation at the output of the sensor are extremely low, i.e. a few nA, the interfacing electronics must be as little noisy as possible to avoid degrading the SNR (signal to noise ratio). Beyond the use of low noise electronics dedicated to this type of measurement, the routing of the printed circuit tracks must also be carried out with care. The tracks must be as short as possible and avoid loops, which would capture the neighboring magnetic field and increase the noise captured by the track, hence lowering the SNR.

The Texas Instrument AFE was tested and compared against the Alphasense AFE during lab experiments. A desiccator served as a controlled environment where various steps and ramps of gas pollutant concentrations were applied. As shown in Figure 1, the signals obtained with the Texas Instrument AFE are much noisier than those obtained with the Alphasense AFE. However, when a 1-min running mean is applied, the different steps are much more visible in all signals, and eventually when calibration factors are applied, we can observe a pretty good match between both output signals. The gas sensor itself was switched several times from one AFE to the other to prevent its potential effect on the quality of the signal.

The embedded system is based on the Microchip ATmega2560 8-bits microcontroller because it allows for very fast hardware and software prototyping with the Arduino Mega development kit and includes the interfaces required for this project: I²C for the AFE, the accelerometer and the environmental sensor, UART for the GPS and the PM sensor, and SPI for the SD card. The general architecture of the Antilope’s electronics is shown in Figure 2. Further developments have led to the inclusion of a BLE (Bluetooth Low Energy) module that establishes a wireless link between the Antilope and any compatible device e.g. smartphone.

The enclosure was designed to be 3D printed, allowing for a quick and cheap prototyping. It needed to protect as much as possible the electronics and the sensors from dust and rain but also to ensure ventilation in the direct vicinity of the gas sensors. The choice made was to have numerous slots, tilted at an angle of 45°, on the three sides surrounding the electrochemical sensors, allowing for air inflow and preventing, at least partly, rain to enter the housing when the device keeps its regular vertical orientation. The SD card slot and the micro USB type B port used for power supply are both located on one side of the box, with a little extension protecting them from direct rain.

A 3-part design was considered to facilitate the assembly of the Antilope; its components are shown in Figure 3. The bottom part contains the main PCB, on which are mounted the accelerometer, the PM and the gas sensors. The middle part acts as a ring onto which the other parts, namely the environmental condition sensor and the GPS, are attached. The lid is bolted into the bottom part and keeps the ring in place. A FDM (Fused Deposition Modeling) printer was used to make this housing for it is easy, cheap and quick, and offers custom ways of prototyping. PLA (Polylactic acid) filament was chosen as the base material for the enclosure.

To avoid coding in bare metal C, which can quickly become time-consuming, Sparkfun’s Arduino-based libraries were used to control the different computer buses with sensors. They simplify the programming and allow one to spend his time on the application code layer.

The data acquisition is done every second. The sampling period is based on a timer/counter overflow, which is matched, once a day at midnight, with the GPS time for logging. Records are stored in a FAT-formatted SD card within a daily CSV file that contains the date, time, location, PM_2.5, NO, NO₂, O₃+NO₂ levels, shock detection, temperature, barometric pressure and relative humidity. The gas concentrations are provided as raw signals, they need a transformation to be expressed in ppb or µgm⁻³.

One of the most common applications with low-cost sensors is the densification or the creation of a complementary monitoring network. In order to evaluate the performance of our device, we set two of them up side by side with reference instruments at the urban traffic station of Antwerp-Borgerhout (Belgium) run by VMM (Vlaamse Milieumaatschappij), and at the urban background station of Liège-Val-Benoît (Belgium) run by ISSeP (Institut Scientifique de Service Public).

Two systems were used in Antwerp alongside a certified fine dust measurement device from Palas, the FIDAS 200; this experiment was part of a longer and larger comparison exercise led in the framework of the VAQUUMS project (https://vaquums.eu/sensor-db/tests/life-vaquums_PMfieldtest.pdf/view). Figure 4 shows the PM_2.5 time series obtained before (top panel) and after correction (bottom panel). Both LCSS were in good agreement with the reference analyzer: the correlation coefficients were 0.95 and 0.93, and the RMSE (Root-Mean-Square Error) 5.14 and 6.22 µgm⁻³. After a simple calibration (linear regression), computed over the whole considered period, RMSE decrease to 2.25 and 2.58 µgm⁻³. No correction linked to the environmental conditions was applied.

To illustrate the performance evaluation of the electrochemical sensors, we present a 1-week comparison of ozone concentrations measured in Liège by 18 Antilope and a certified analyzer from Horiba, the APOA370. In practice, we usually make such a comparison exercise between two batches of subjects for the personal exposure experiment but on a shorter period of two to 3 days. Figure 5 shows the O₃ time series obtained by using the sensor’s lab-defined sensitivity and zero current values (top panel) and by applying a correcting bias and gain (bottom panel). Since a linear transformation is applied, correlation coefficients between the different instruments are in the same range in both cases (0.49; 0.92) with a mean value of 0.82. The range of the RMSE decreases from (20; 245) µgm⁻³ to (8; 18) µgm⁻³ with a mean value of 120 and 11 µgm⁻³ respectively.

In the personal exposure part of the project, every 2 weeks four subjects living in the cities of Liège and Namur were provided with an Antilope, an AethLabs AE51 aethalometer and a GlobalSat DG200 GPS for 7 days. Instructions about the instrument use and charging were given orally but some video tutorials were also available. All the measurement equipment was placed in the nest of a backpack to easily shadow the subjects in their daily activities.

At the beginning of their 1-week experiment, subjects had to fill in a questionnaire about their profile (age, gender, professional status, etc.), their health in general (allergy or asthma, frequency of physical activities, smoking exposure, etc.) and their environment with questions about their house (heating type, ventilation, kitchen and floor equipment, etc.), their habits (most occupied rooms, vacuum frequency, etc.) as well as their place of work if appropriated.

During the week, participants had to fill in a journey log-book with all their activities. Each activity had to be characterized by a start time, an end time, a type (work, shopping, staying at home, cooking, sport, leisure, etc.) and an environment (indoor or outdoor). Travels are considered as an activity with an indoor/outdoor type depending on the mode of transport (car, bus, train, walk, etc.). Finally, subjects also had to report every day any respiratory discomfort or crisis.

Such information is very useful to evaluate exposure to air pollution according to activities and modes of transport and to corroborate some of the measurements such as the location provided by the GPS or the lack thereof.

Figure 6 shows BC and PM_2.5 concentrations measured by one subject during 24 h of his 1-week experiment. In this case, one can clearly see peaks related to daily car journeys for both pollutants; indoor measurements of these pollutants seem to be much less correlated. Anecdotally, one can value the precision of this subject when reporting timestamps in the log-book.

A preliminary analysis of 72 candidates has shown that cooking was the activity where people are exposed to the highest PM_2.5 concentrations (28 µgm⁻³), followed by indoor sport (23 µgm⁻³) and outdoor leisure activities (excluding sport) (17 µgm⁻³). As shown by Shehab et al. (2021) and Hu et al. (2012) even cooking with an electric stove emits particulate matter, as a result of the burning of the food itself. At the bottom of the list is shopping in the city (7 µgm⁻³). We did not consider activities with less than 10 h of cumulated duration for this analysis. The variation range of relative humidity during these activities only spanned from 26–35%, it is thus difficult to consider it as a major factor to explain the differences. Regarding the transport, on average bus commuters were exposed to 21 µgm⁻³, pedestrians to 16 µgm⁻³, cyclists to 12 µgm⁻³ and car drivers to 8 µgm⁻³.