Nitrogen dioxide (NO₂) is one of the major environmental pollutants, which belongs to the family of nitrogen oxides (NO_x). The major source of NO₂ in the air comes from combustion of fossil fuels [1] . NO₂ gas forms when fuel is burned at high temperatures, primarily from the exhaust of motor vehicles and power plants. NO₂ is a strong oxidizing agent that may form highly toxic nitric acid and organic nitrate. Furthermore, NO₂ leads to the production of ground level ozone, which is also recognized as smog [2] . Studies have shown that NO₂ and related derivatives (NO_x) will increase the risk of lung cancer [3] [4] . The International Agency for Research on Cancer recently classified outdoor air pollution and particulate matter as carcinogenic (Group 1) [5] . In the United States, EPA (Environmental Protection Agency) has set primary standards in compliant of the Clean Air Act of NO₂ to not exceed 53 ppb as annual mean (Figure 1) [6] . Therefore, fast, sensitive and reliable analysis of NO₂ is an essential need for the public health, as well as for personal protection. In this review, we will summarize some of the key techniques reported to detect and quantitate NO₂ in the past two decades, and give advice for further research.

Widely used NO₂ analysis technologies can be divided into the following four categories based on the mechanism of detection: traditional chemical analysis, optical detection, field-effect transistor detection and other detection techniques.

Analysis of nitrogen content in organic samples started in the late 19^th century. Kjeldahl [8] published the work on digesting organic matters with sulfuric and phosphoric acids in 1888. The nitrogen content was converted to ammonium sulfate and then to ammonia, which was distilled into a measured volume of standard acid. Further analysis by acid-base titration of ammonia completed this indirect analysis of any form of nitrogen contents in the sample. Many researchers have worked on optimization of the classic Kjeldahl method to improve its digestion efficiency, while the essence of chemistry behind this method stays the same. This method has its own drawbacks on speed, accuracy, and most importantly, specificity. Almost all forms of nitrogen, including NO₂, NH₃, and other organic nitrogen, are analyzed as combined together.

Andres Ferrari proposed a system for continuous digestion and analysis based on the Kjeldahl method [9] . Automation of sample handling was performed with a grooved quartz tube, which was rotated by motor-driven wheels and allowed the sample solution to flow from one end of the tube to the other. The tube was heated inside a tube furnace to 400˚C - 600˚C. The high temperature and presence of oxidants converted nitrogen content into ammonium salts. The salts were then treated with excess of base. Afterwards, the liberated ammonia was collected in measured volume of standard acid. The residual acid was titrated with standard base, which quantified the total nitrogen content [9] .

This automated method has produced similar results as the classic manual method with significantly reduced workload requirement. The rate of analysis using this automated system is up to 10 samples per hour. However, the classic methods are only capable of determining nitrogen content in single digit percentage. Limit of detection of both manual and automatic titration is about 0.1% of the sample weight.

Chemiluminescence was the most prevalent and dominating method for the analysis of NO₂ in the air. The United States Environmental Protection Agency (EPA) has published the official guidance of the standard chemiluminescence method for NO₂ analysis [10] . This technique was first discussed in detail by Fontijn et al. in 1964 [13] . Since then, it has been intensively studied since then. The resulted reliable, simple and accurate method was then recommended by EPA and other regulatory agencies around the world as the first choice for routine NO₂ analysis.

The standard method regulated by EPA employs the creation of chemiluminescence using ozone. In this reaction, NO₂ is first reduced to nitric oxide (NO). Then NO reacts with O₃ to form the exited state of nitrogen dioxide, , which emits light while returning back to the ground state of NO₂.

Intensity of light is proportional to the concentration of NO₂. Commercially available instruments deliver a fully validated analysis and conform to EPA specifications. The dynamic range used for daily NO₂ monitoring is from 0 to 500 ppb. Traceable NO or NO₂ standards are also readily available for the calibration of the instruments. Instrument using the luminol method is currently the first and only method available in pursue of the regulated environmental monitoring of NO₂ in air. NO and other nitrogen oxides are common interferences for this method. In addition, free radicals in the air may also reduce the amount of exited NO₂, leading to a lower result.

Wendel described a continuous NO₂ monitoring system capable of detecting NO₂ in the sub ppb range [14] . This instrument employed several pumps to circulate air and the reagent solutions. PMT was used to detect the emitting light. Due to very high amplification gains of PMT, the detection limit is down to 30 pptr. Interference of O₃ was avoided using gas washing prior to entering the reaction chamber. However, there were still interferences including other nitrogen oxides (NO_x). An alternative method is recommended for extreme cases in cluding field study, instead of using chemiluminescence. Collins et al. [15] [16] introduced an improved Luminol-based NO₂ analysis (Figure 2). The solution circulation system was replaced by a thin film of hydrogel, which immobilized luminol with necessary catalysts.

Besides chemiluminescence, Cavity Attenuated Phase Shift Spectroscopy (CAPS) developed by Aerodyne Inc. was also used for NO₂ analysis (Figure 3) [17] [18] [22] . CAPS is related to cavity ring-down spectroscopy (CRDS). In CAPS and CRDS, the measurements of light absorption are made possible by using an extremely long effective path length. A beam of laser is introduced into a cavity with two highly reflective mirrors. The light can “bounce” back and forth with path length over one kilometer. The resulting signal measured at the exit mirror will show exponential decay over time (Ring-down). The decay time is related to the concentration of the

light-absorbing species in the cavity [23] [24] .

The main oscillation chamber used in CAPS was constructed using two highly reflective mirrors facing each other, similar to CRDS. A beam of modulated broadband light created by LED is directed into the chamber. Presence of NO₂ causes a phase shift in the signal output, which is proportional to the concentration of NO₂. 430 nm was chosen as the light wavelength because it matches the maximum absorption of NO₂. With the use of an interference filter, the limit of detection is 0.5 ppb. This sensor is capable of measuring 0 - 200 ppb of NO₂ in the air. However, the noticeable curvature seen in the calibration towards 200 ppb is caused by loss of light during absorption, which is comparable to the phase shift effect at higher concentrations [22] . Table 1 summarized chemiluminescence and non-chemiluminescence techniques used in optical NO₂ detection.

Though chemical and optical analysis techniques are good options for NO₂ analysis, researchers are continuously seeking for new sensors that are in solid state, easy to operate, maintenance-free and small in size. Without the requirement of any chemical solutions, gas sensors have been made perfect candidates for varies of applications both in the lab and the field. Its simplicity also enables the mass production with relatively low cost for each unit. Integration with modern electrical engineering allows direct control and digital signal transmission to a data collection device. Interest for the FET gas sensors is increasing, accompanying hundreds of research articles and communication papers in this field. The study of sensor design and material has been divided into two ways. One is the optimization of the sensing performance. The other is the additional feature and utilizations on top of the good performance.

Early work on FET gas sensors can be tracked back to 1980s. In a study by Kolesar et al. [25] in 1989, IGE- FET was coupled to a thin film of copper phthalocyanine (CuPc), which selectively reacted with IGE-FET at ppb levels. Measurement was based on the sensitivity of the field effect transistor’s output current to changes in the molecular structure or chemical composition of the thin film which covers the interdigitated gate electrode.

A versatile sensor platform of NO₂ in the air was developed by Oprea et al. in 2007 (Figure 4) [26] . This

platform was based on capacitive coupled field effect transistors (CC-FET). Unlike other common gas sensors, the sensing material in this study was suspended on top of the substrate, which is called suspended gates (SG) [26] . There is a small gap of 1 - 5 μm distance between SG and the FET channel. A schematic illustration is shown in Figure 4. Therefore, the sensing material can be separated from the platform so that a large variety of materials and substrates can be tested. Sensing thin film can be easily replaced and compared in parallel under exactly the same testing conditions.

In the past two decades, a great deal of research work has been focused on resistance and conductivity gas sensors [10] [25] [27] - [32] . In this type of sensors, a thin film usually made of semiconductor oxide serves as the main sensing material. Concentration of NO₂ is directly related to resistance and conductivity of the material and is measurable with peripheral circuits. The center of many current research targets at discovering and developing better sensing materials.

NO₂ is an oxidizing gas so it induces an increase of resistance and impedance of n-type semiconductors. Sensors based on semiconducting metal oxides have been developed and reviewed by a number of researchers [29] . The base material includes tin oxide (SnO₂), tungsten oxide (WO₃), indium oxide (In₂O₃), titanium oxide (TiO₂), zinc oxide (ZnO) and other non-metal materials such as polymers and grapheme [28] - [30] [33] - [35] . In addition, doping and use of different additive materials has opened another pathway of developing new sensing materials. Common preparation methods are chemical vapor deposition (CVD), electrophoretic deposition, electrostatic spray, solution precursor plasma-spray and other similar approach. Characterization of the microscopic structure of these materials usually utilizes technologies such as X-ray powder diffraction (XRD), scanning electron microscope (SEM), Transmission electron microscopy (TEM), atomic force microscope (AFM) and X-ray photoelectron spectroscopy (XPS).

ZnO has been proved to be a stable, low cost and non-toxic sensing material. However, the sheet resistance of pure ZnO thin film is substantially high. If impedance of the sensor becomes higher, the resulting weak signal of current change will considerably increase the complexity of the associated controlling and measuring circuitry of instruments, though sensing of NO₂ is based on the increase of resistance. Data may also become distorted at such week signal levels. Ferro et al. [28] reported indium and cadmium doped ZnO thin films with a lowered resistance and improved performance. Spray pyrolysis was used for its simplicity, low cost and ability for mass production. Undoped ZnO sensor was tested with different film thickness and grain size in order to find an optimized combination. The best amount of doping was 3% w/w. Fast response and recovery was observed at an operating temperature of 275˚C. The effect of gas sensing with the additive was considered not related to electronic catalytic mechanism because increased carrier concentrations by doping deteriorate performance of the sensor. The reaction mechanism was proved to be the increase of the surface active area and the number of active absorbing sites, by using SEM and AFM to examine the surface structure of the doped sensing film. The detectable range of doped ZnO film is 0 - 5 ppm [28] .

In addition, sensors of oxidizing gases usually utilize high temperature to enhance the desorption kinetics. In studies of gas sensors, response time (T90) is defined as the time required for the signal to reach 90% of the final value upon exposure of the target gas, while the recovery time (T10) is the time for the signal to reduce to 10% of the highest level when the target gas is removed. Slow response and recovery within 10 - 60 min is commonly observed across this type of sensors (Figure 5) [30] . Several types of sensors that may function at room temperature have been discovered and investigated. However, the lowered temperature may slow down the response and recovery further more. A reported In₂O₃ nanowire sensor showed a response to sub-ppm level NO₂ in 15 min with a recovery time of 45 - 90 min. [31] Another study by Liu et al. [29] significantly lowered the requirement of high operating temperature, which is usually 300˚C - 500˚C and observed among resistive gas sensors. Kinetic studies have shown that absorption of NO₂ on the surface of sensing materials was much faster than the desorption process.

Graphene has recently been reported as a potent material for construction of sensors, small-scale electronics and solar cells. The unique wide-spread 2D structure of graphene reassembles a honeycomb lattice, which forms π bonds across the hexagonal skeleton. π bondis half filled and allows free-flowing electrons, which contributes to the most notable electronic properties of graphene, the quantum Hall effect [36] [37] . The selective absorption and desorption was demonstrated by Ko et al. [32] . Unlike other metal oxide thin film gas sensors, adsorption of NO₂ molecule increases of conductivity of graphene sensors. As Zhang et al. suggested there were three possible configurations of the adsorption, the nitro-, nitrite- and cyclo-addition, increase of conductivity of graphene is due to charge transfer between NO₂ molecules and graphene. Desorption has been found to be a very slow process. A possible way to expedite it is to expose the sensor under UV radiation [32] .

When reduced graphene oxide is sprayed withasoluble semiconducting polymer, a double-layered sensing film is formed. Xie et al. [33] reported that the use of poly-(3-hexylthiophene) (P3HT), which is a p type semiconducting polymer, can enhance the gas sensing properties (Figure 6). The double-layered material was constructed on an existing organic thin film transistor (OTFT) sensor platform. Pure P3HT thin film was prepared and compared to the double layer of P3HT/Graphene designs. P3HT and/or Graphene layers were applied to the OTFT substrate using an air brushing technique.

The response to exposure of NO₂ showed a very interesting three-stage curve, the rapid response, the slow response and the recovery, especially at higher concentrations of NO₂. The mechanism that caused the 3 stages can be explained that at higher concentrations (over 60 ppm), the rapid rise of signal (stage 1) is due to the adsorption of gas molecules onto low energy bonding sites such as sp2-bonded carbon. With gradual saturation of these sites, the gas molecules will interact with higher energy sites such as vacancies and structural defects, which is much slower (stage 2). Stage 3 showed an exponential decay of the signal when the analyte (NO₂) was purged out with air. However, at lower NO₂ concentrations, only very few NO₂ molecules can reach to the surface of graphene through the P3HT top layer. The fast responding low energy sites will not be saturated. Thus all the NO₂ molecules interact with the low energy sites, which lead to a relatively fast over-all response.

One of the major drawbacks of graphene based sensing material is the baseline shift. As showed also in the response curve over time, the recovery of the signal did not go back to the original level, especially after exposure to high levels of NO₂. Within 90 min, the signal will return to baseline close to the original value, but long recovery is not desirable in practice and needs further investigation for improvement.

Besides these dominating detection techniques mentioned above, there are other methods to analyze NO₂, such as Gas Chromatography/Mass Spectrometry (GC/MS), Residual Gas Analyzer (RGA) and immunochemical methods [38] - [40] . One example is the detection of NO₂ using piezoelectric quartz crystal [11] .Vibration frequency of an oscillating crystal may be decreased by substances that are absorbed on the surface of crystal. Selectivity is achieved by carefully choosing a coating material on the surface of the piezoelectric crystals. In the study by Karmarkar et al., two types of materials, Ucon 75-H-90000 and Ucon-LB-300X were chosen for the specified detection of NO₂ and NH₄, respectively [11] . Base frequency of the crystal was 9 MHz. A 40 Hz decrease was observed with exposure to about 1 ppm of NO₂ in the sample mixture. The response as change of frequency was linear in the range from 1 ppb to 1 ppm. The response time was about 5 minutes. In short, the piezoelectric crystal is a good detector with high sensitivity, without use of liquid chemicals or moving parts. However, there are drawbacks of this type of detector. First, it responses to atmospheric moisture. It would be difficult to distinguish between the effect of moisture and a true air pollutant. A dry wash device can be used to eliminate moisture content, before the air samples enter the detector. Another drawback is that after a long exposure to air, the detector loses all the detection ability to NO₂. This disability may be reversed by exposing the detector to NO₂ mixture again for a short time [11] .