Automotive Catalyst State Diagnosis Using Microwaves

The state of catalysts plays a key role in automotive exhaust gas aftertreatment. The soot or ash loading of Diesel particulate filters, the oxygen loading degree in three-way catalysts, the amount of stored ammonia in SCR catalysts, or the NOx loading degree in NOx storage catalysts are important parameters that are today determined indirectly and in a model-based manner with gas sensors installed upstream and/or downstream of the catalysts. This contribution gives an overview on a novel approach to determine the catalyst state directly by a microwave-based technique. The method exploits the fact that the catalyst housing acts as a microwave cavity resonator. As “sensing” elements, one or two simple antennas are mounted inside the catalyst canning. The electrical properties of the catalyst device (ceramic honeycomb plus coating and storage material) can be measured. Preferably, the resonance characteristics, e.g., the resonance frequencies, of selected cavity modes are observed. The information on the catalyst interior obtained in such a contactless manner is very well correlated with the catalyst state as will be demonstrated for different exhaust gas aftertreatment systems.


INTRODUCTION Technical Background
Steadily increasing costs for fuel and the pressure on automotive manufacturers from costumers and legislation to reduce CO 2 emissions lead to booming market shares for Diesel passenger cars. Since Diesel engines are operated leanly and Nitrogen Oxides (NO x ) cannot be removed with conventional Three-Way Catalysts, or TWC (Alkemade and Schumann, 2006;Shelef and McCabe, 2000), novel exhaust gas aftertreatment concepts have emerged.
The ammonia selective catalytic reduction process (NH 3 -SCR) has been transferred from power plant applications to automotive requirements. For both heavy duty vehicles and passenger cars, systems have already been serialized in the past few years (Koebel et al., 2004;Johnson, 2012). In NH 3 -SCR systems, a urea water solution is injected into the exhaust. Ammonia is formed by hydrolysis and serves as a selective reduction agent for NO x . According to the reaction mechanism, NH 3 is initially adsorbed (stored) in the SCR catalyst.
The NO x reduction (conversion rate) depends strongly on the amount of stored NH 3 , especially at low temperatures (Busca et al., 1998;Kro¨cher et al., 2006;Johnson, 2012).
In NO x Storage Catalysts (NSC), also known as Lean NO x Traps (LNT), NO x is adsorbed and stored in the form of nitrates during a lean phase. Nitrate reduction occurs in a subsequent short rich phase, right before the storage capacity of the NSC is exhausted and NO x slip occurs (Takeuchi and Matsumoto, 2004).
For both NO x abatement technologies, NH 3 -SCR and NSC, novel exhaust gas sensors that provide information on the state of the catalyst (amount of stored NO x or NH 3 ) to supplement the well-known and mature lambda probe would be beneficial for catalyst control.
In order to remove the emitted particulate mass and number, Diesel Particulate Filters (DPF) have been serialized. To avoid clogging, they have to be regenerated regularly to burn off sorbed soot (Twigg and Phillips, 2009). Since regenerations consume fuel, the number of regenerations has to be kept to the minimum. Hence, detailed knowledge of the actual soot loading would be beneficial.
The exhausts of most gasoline-fuelled internal combustion engines are treated by TWC. Only under stoichiometric conditions, i.e. with an air-fuel equivalence ratio (also denoted as normalized air-to-fuel ratio, k) of k = 1, all limited exhaust components, like NO x , hydrocarbons, or carbon monoxide, are converted best. Lambda probes, which are installed upstream and downstream of each catalyst, determine the actual values of k in the gas phase (Riegel et al., 2002;Twigg, 2007). To buffer richto-lean fluctuations, the washcoats of the TWC (honeycomb-like substrate with catalytically active coating) contain large quantities of doped ceria-zirconia solutions as a component for oxygen storage, making use of cerium's two different oxidation states at exhaust temperatures. The Oxygen Storage Capacity (OSC) of a TWC is directly related to the amount of available ceria (Mo¨ller et al., 2009). By k measurements up-and downstream of a TWC, the oxidation degree is calculated by OSC models. Such an indirect characterization of the TWC oxygen loading degree is today's best available technology.

Exhaust Gas Sensors
Many efforts were being made in the past years to develop novel automotive exhaust gas sensors that are sensitive, selective, long-term stable, and cost effective (Riegel et al., 2002;Moos, 2005;Alkemade and Schumann, 2006;Zhuiykov and Miura, 2007;Fergus, 2007;Moos and Scho¨nauer, 2008). However, per today, only sensors based on yttria stabilized zirconia like the binary lambda probe, the wide-band lambda probe (also Universal Exhaust Gas Oxygen sensor, UEGO sensor, or linear lambda probe), and the amperometric NO x sensor have been serialized. In addition to direct engine control, it is the purpose of these selective exhaust gas sensors to detect the state of the catalyst indirectly, supported by models. "State" in this respect may mean, for instance, the oxygen loading of three-way catalysts, the NO x loading of NO x storage catalysts, the NH 3 loading of ammonia-SCR catalysts, the soot loading of Diesel particulate filters, the conversion efficiency, or the sulfur poisoning.

Direct Catalyst State Observation
In the past few years, it was appreciated more deeply that the catalyst state can be determined directly by monitoring the electrical properties of the catalyst coatings themselves. One may classify all works on this topic in two categories: -a direct but not contactless approach, and -a direct and contactless, microwave-based approach.
In the first approach, sections of the catalyst coating are typically investigated by impedance spectroscopy. This has been successfully demonstrated to determine in situ the oxygen loading degree in TWC formulations based on ceria-zirconia (Reiß et al., , 2011a, to detect in situ the NO x loading state, the regeneration state, the degree of sulfurization, and the thermal aging state of a NO x storage catalyst based on earth-alkaline oxides (Moos et al., 2008a), or to determine the ammonia loading in Fe-SCR zeolites with electrical ac measurements (Kubinski and Visser, 2008;Rodrı´guez-Gonza´lez and Simon, 2010). The soot loading of DPF can be measured as well, even with a local resolution (Hagen et al., 2011). However, the above-described devices are still additional sensors that need to be mounted into the exhaust pipe.
Even more sophisticated and promising is the recent microwave-based approach. It offers the chance to determine the catalyst state directly without the workaround via the concentration of a single compound in the exhaust. The microwave-based automotive catalyst state diagnosis technique is based on the dependence of the electrical properties of the catalyst materials on their loading states (Moos, 2010). To a course approximation, the catalyst housing may be considered a PEC (Perfect Electric Conductor) wall enclosing a cavity. Owing to the typical geometrical dimensions, the lowest-order modes in the cavity resonate at low GHz frequencies.
The cavity resonator may be coupled to a source and load via simple "antennas" such as short stubs or loops which are mounted inside the catalyst canning. Thus, the electrical properties of the catalyst device (ceramic honeycomb plus coating and storage material) can be measured from the outside. Therefore, the measurement method belongs to the large class of Non-Destructive Evaluation (NDE) methods which allow the investigation of the interior of solid bodies by waves propagating in the solid. Here, preferably, the resonance characteristics, e.g., the resonance frequencies, of selected cavity modes are observed. The information on the catalyst interior obtained in such a contactless manner correlates very well with the catalyst state as will be shown for the detection of the oxygen loading of a TWC, the NO x loading of an NSC, the soot loading of a DPF, and the amount of stored ammonia in an SCR catalyst.
1 MICROWAVE-BASED CATALYST MONITORING

Principle
As already described, the electrically conducting canning of a catalyst converter (or of a DPF) defines a cavity resonator. Typically, it is coupled to its environment via one or two coaxial waveguide feeds (antennas) which are mounted inside the canning (Fig. 1a). At microwave frequencies, voltages, currents, and impedances can be neither defined unambiguously nor measured directly. What can be measured are the amplitudes and phase angles of the electric and magnetic field components associated with the electromagnetic waves propagating along a waveguide. For a propagating sinusoidal mode in a uniform waveguide such as a coaxial line, the ratio of the amplitudes of any two field components and the phase difference between any two field components are constant along the waveguide. Hence, the mode can be described unambiguously by a complex-valued scalar wave amplitude: where k is the wave number and z denotes the coordinate along the waveguide axis (z = 0 corresponds to the location of a reference plane which may be chosen freely). It is a common practice to normalize the magnitude of a such that it is linked to the time-averaged power transported by the mode by: The cavity shown in Figure 1a has two interfaces, or ports, that separate it from the environment. A coaxial Setup of the microwave-based automotive catalyst state diagnosis. a) Schematic diagram. For research purposes, steel meshes may be inserted as indicated to exactly define the cavity length. The wideband lambda-probes or other exhaust gas sensors are for research purposes only. b) Representation as a microwave two-port with incident and scattered wave amplitudes. c) Equivalent circuit for a narrow frequency band near the resonance frequency of a single cavity mode.
cable connected to port i supports an incident wave a i and a scattered wave b i (it is assumed that the cable supports only its fundamental transverse electromagnetic mode in the frequency range of interest, which is certainly true in the lower GHz range; Fig. 1b).
As any linear two-port, the housed catalyst is uniquely described by a linear system of equations between the incident and scattered wave amplitudes: The scattering parameters S ij can be easily measured in the laboratory by commercial Vector Network Analyzers (VNA). As a consequence of the normalization (2), the magnitudes squared of the S-parameters are ratios of time-averaged powers. For instance, the magnitude squared of the input reflection coefficient S 11 is the ratio of the average power impressed on port 1 to the average power scattered back at the same port when port 2 is matched (a 2 = 0). Similarly, the forward and backward transmission coefficients S 12 and S 21 describe the transmission of power from one port to the other through the interior of the cavity resonator. Figure 1c shows an equivalent circuit which describes the terminal characteristics of the microwave cavity resonator in a narrow frequency band around the resonance frequency of a single cavity mode. (A broadband equivalent circuit consists of many analogous RLC circuits.) There is an unambiguous relationship between the generalized voltages and currents of such an equivalent circuit and the complex wave amplitudes at the interfacing terminals. When the material parameters of the housed catalyst change as a result of electrochemical reactions, this shows up in a change of the equivalent-circuit element values R, L, C, k 1 , and k 2 or, equivalently, in a change of the S-parameters S ij .
Further details of the measurement technique can be found in the literature (Fischerauer et al., 2008(Fischerauer et al., , 2010a. Microwave concepts such as electromagnetic waveguides, cavity resonators, and S-parameters are treated in many standard texts (Harrington, 1961). There exists a vast literature on the use of cavity resonators for the measurement of material properties of small samples. This deviates from the present application as the catalyst-filled cavities contain large samples. Still, many basic features of the method can be understood from the standard theory of cavities perturbed by small samples (Sucher and Fox, 1963;Klein et al., 1993;Carter, 2001).

Application to Three-Way Catalysts
As mentioned above, today's gasoline-operated automobiles are equipped with TWC for best conversion of the limited components hydrocarbons, CO, and NO x . They have to be operated stoichiometrically (at k = 1). To buffer air-to-fuel fluctuations, ceria-zirconia solutions are added as oxygen storing materials. Using the signals of lambda probes upstream and downstream of the TWC, the oxidation states of the catalysts are modeled.
Since the electronic conductivity of ceria-zirconia solutions depends on the equilibrium oxygen partial pressure (Boaro et al., 2002), pO 2 , which is a measure of the oxygen loading degree (Mo¨ller et al., 2009), a direct catalyst state control determining the overall oxygen loading may be possible (Moos et al., 2008b).
A typical resonance spectrum of a TWC operated at steady state conditions in lean and rich exhausts is shown in Figure 2 (Reiß et al., 2011b). Both the resonance frequency and the magnitude of the reflection coefficient S 11 at the resonance frequency may be an appropriate signal feature from which the oxygen loading of the TWC can be inferred.
Using either the shift of the resonance frequency f res of a suitable cavity mode or the attenuation at this frequency, one may determine the amount of loaded oxygen. This has been shown by careful titrations in the engine dynamometer as well as in the lab test bench (Reiß et al., 2011a, b). The system is not sensitive to other exhaust components such as H 2 , CO 2 , or CO (Reiß et al., 2011c). However, as both the size of the cavity and Return loss |S 11 | of a canned TWC in the frequency range in which the two lowest-order cavity modes resonate and at approximately 450°C. Reiß et al. (2011b).
the electrical conductivity of the catalyst materials depend on temperature, temperature effects occur and may need to be compensated (Reiß, 2012). Very recently is has been shown in laboratory experiments that one can realize a low-emission engine control without lambda probes solely by using the resonance frequency of one cavity mode of the housed TWC operated as microwave one-port, i.e., coupled by a single antenna, which means that one can only measure one S-parameter, viz., S 11 (Beulertz et al., 2012). In this scheme, the resonance frequency serves as the controlled variable of the closed-loop control system, but, of course, it is a direct measure of the oxygen loading degree of the catalyst. Figure 3 shows how the oxidation degree of ceria, the resonance frequency, and k in equilibrium depend on each other. It is noteworthy to mention that, with this method, the model of Mo¨ller et al. (2009), which predicted the oxidation state of ceria in TWC in much more detail than earlier models, could have been validated for the first time directly, without titration.

Application to NO X Storage Catalysts
Nitrogen oxide removal (NO x = NO and NO 2 ) from exhausts of leanly operated engines is not possible by TWC, since the reducing components CO, H 2 , and hydrocarbons are oxidized by O 2 instead of being used for NO x reduction (Alkemade and Schumann, 2006). To overcome this, the NSC has been originally developed for direct injection gasoline engines that are operated in the lean mode (Takeuchi and Matsumoto, 2004) and has later been applied also for Diesel exhaust gas aftertreatment (Kasˇpar et al., 2003). During a lean phase, NO x is oxidized, sorbed, and stored in the form of nitrates (some minutes). Just before the NO x storage capacity is exhausted and the catalyst begins to let NO x pass, the engine is switched to the rich operation mode (some seconds) and the formed nitrates are reduced to N 2 (Epling et al., 2004). Usually, NSC-based exhaust gas aftertreatment catalysts cannot be operated in an open loop. A NO x sensor is mounted downstream of the NSC to detect NO x breakthroughs, acts as part of a closed-loop control system and thus helps to prevent overloading of the NSC. A lambda probe ensures that the regeneration period is only as long as necessary to avoid CO or hydrocarbon breakthroughs.
The microwave-based approach may detect the NO x loading degree directly. Initial tests were conducted in a lab test bench . Microwave spectra were taken from 1 to 4 GHz for fully loaded and depleted NSC. Compared to the TWC, in this case, a less pronounced shift in the resonance frequencies and a slight broadening of the resonance peaks were observed. The relative resonance frequency changes due to NO x loading, Df res /f res , were compared for all resonances and the one with the highest effect of NO x was used for further studies. The transient behavior of the resonance peak at ca. 2.615 GHz is shown in Figure 4 by way of an example. The following section is a summary of an extensive study (Fremerey et al., 2011). At the beginning of the test run, a NO x -free lean base gas was applied to the NSC (5% CO 2 , 1% O 2 , and 8% H 2 O in N 2 ) at t = t 1 after a previous rich regeneration phase (1% H 2 , 100 ppm C 3 H 8 , 5 000 ppm CO and 8% H 2 O in N 2 ) to fully remove stored NO x . Starting at t = t 2 , 2 000 ppm of NO was added to the base gas for 2 h. The concentration of NO in the feed gas was increased to 4 000 ppm at t = t 3 . At t = t 4 , the NO concentration was switched back to 2 000 ppm, and after t = t 5 , the feed gas consisted of only base gas without NO.
It is evident from Figure 4a, that the wideband lambda sensor upstream of the catalyst quickly reacts to the switch from rich to lean feed gas composition at t = t 1 , whereas the sensor downstream has a slightly delayed response. Its signal remains around k % 1 for several minutes before it reaches the same value as the sensor upstream of the catalyst. This delay is caused by the oxygen storage component in the NSC. As soon as the NSC is completely loaded with oxygen, both wideband lambda sensor signals coincide. After the addition of NO at t = t 2 , the k of the feed gas increases a little as indicated by the upstream lambda probe (highlighted in the inset). At the same time, the k downstream of the NSC drops a bit. This is not surprising since oxygen is required for the storage reaction and, therefore, as long as the NSC stores NO x , the downstream k should be lower than without NO in the feed. At t % 3.5 h, the catalyst is saturated (for 2 000 ppm NO in the feed), and no NO x can be stored anymore. A similar behavior occurs after t = t 3 , when the NO concentration is further increased. Owing to the chemical equilibrium of the storage reaction, the storage capacity of the NSC is greater at higher NO concentrations and more NO x can be stored. Analogously, the equilibrium changes back to lower loading states when NO is decreased to 2 000 ppm again at t = t 4 as well as to 0 ppm at t = t 5 . In both cases NO x , desorbs from the NSC.
By balancing the nitrogen oxide concentrations upstream and downstream of the NSC during the loading period, one should be able to obtain information on the amount of NO x stored in the NSC (Fig. 4b). The same picture is given by the resonance frequency, f res , in Figure 4c. One can clearly see the difference of the signal between the rich and lean gas atmospheres soon after t = t 1 , which is caused by the oxygen storage capability of the NSC formulation. As known from TWC (see above), the oxidation state of the oxygen storage component has a major influence on the microwave absorption. At the beginning of the NO x storage at t = t 2 , f res decreases continuously until t % 3.3 h. From then on, it remains constant. The curve strongly resembles the NO mass loading curve in Figure 4b. A further increased NO concentration in the feed gas (t 3 < t < t 4 ) also affects the microwave signal, indicating an increasing mass of stored NO in the NSC.
In summary, these results give a first hint that the NO x loading of the NSC can be measured directly by the contactless microwave-based monitoring technique. However, the effect is by far smaller than for TWC. Cross effects of the oxygen loading may be high enough to thwart the entire approach.

Application to NH 3 -SCR Catalysts
In SCR systems, an aqueous urea solution is injected into the exhaust. It decomposes to NH 3 which then serves as a reducing agent. In the SCR-catalyst, NO x is selectively reduced by NH 3 to N 2 and H 2 O. All major SCR mechanisms are based on an initial NH 3 storage step before NH 3 reacts with NO or NO 2 (Nova et al., 2006;Busca et al., 1998). The NO x conversion depends on temperature and on the amount of NH 3 stored in the SCR catalyst (Busca et al., 1998;Ciardelli et al., 2007). To avoid NH 3 breakthroughs, today's control systems use a model-based estimation of the NH 3 loading (Schuler et al., 2009).
Since zeolite materials change their electrical impedance when NH 3 is stored (Kubinski and Visser, 2008), it can be expected that the microwave response of zeolite-based SCR catalysts depends on their NH 3 loading degree. This was investigated in literature (Reiß et al., 2011d) and will be summarized in the following section.
A similar test as described above for NSC was conducted with a zeolite containing iron. In a key experiment (Fig. 5), a fully NH 3 -free catalyst was loaded with NH 3 . Then, NO was added and the feed ratio a ¼ c NH 3 =c NO was reduced stepwise. Between t = t 1 and t = t 2 , only NH 3 was added to the base gas. It loaded the SCR catalyst. At t = t a , the catalyst became fully NH 3 -loaded as indicated by an NH 3 breakthrough. The outlet NH 3 concentration decreased with increasing NO concentration until the catalyst was NH 3 -free at t = t b . The mass of stored ammonia, m NH 3 , was balanced (Fig. 5c). The plot of the resonance frequency of a mode resonating at about 2.92 GHz demonstrates the good correlation between m NH 3 and f res (Fig. 5d).
The influence of sorbed water was investigated as well. It was found that some resonance modes depend strongly on the water concentration, whereas others are mainly affected by the NH 3 loading. One has to keep in mind that the various resonance modes are standing waves which differ from each other by the locations of the nodes and maxima of the instantaneous field patterns. If the catalyst conductivity increases in a region where a mode has an electric field node, this mode will not respond to the conductivity change. In contrast, a mode with an electric field maximum in the region considered will respond strongly. The resonance modes that depend selectively on the NH 3 loading are suited best for future engine tests. In summary, one can determine the NH 3 loading of an SCR catalyst, but one has to keep in mind that relative resonance frequency shifts between the fully loaded and the unloaded state are smaller by a factor of ten in comparison to TWC.

Application to DPF Soot Loading Determination
Currently, a model based on the pressure drop over the DPF is utilized to estimate the soot loading (Rose and Boger, 2009). However, a more precise tool would be helpful. While the microwave-based catalyst state diagnosis is in an initial research state, the same principle has been applied for soot loading determination of Diesel particulate filters and is now in a serial development stage. Besides institutional research (Fischerauer et al., 2010b;Feulner et al., 2012), at least two suppliers (Sappok et al., 2010;General Electric, 2011) and some automotive manufacturers published recent work or applied for patents (Knitt and DeCou, 2007;Walton, 2007;Gonze et al., 2010;Hansson and Ingestro¨m, 2012;Kulkarni et al., 2012) on the microwave-based soot load detection.
The following section summarizes the findings and the differences to the catalyst state diagnosis.
Since soot loading in DPF increases the filter conductivity by orders of magnitude (Hagen et al., 2010), one observes effects as great as in TWC. In order to suppress the possible influence of soot deposition at the coupling antenna, one may use a single waveguide feed located downstream of the DPF where the exhaust is almost soot-free. The spectra of the reflection parameter look very similar to those encountered in the TWC case (Fig. 6). The higher the soot load, the more the resonance frequency is shifted towards lower values and the more the resonance peaks broaden, i.e. the Full Width at Half Maximum (FWHM) increases (Fischerauer et al., 2010b). In other words, the quality factors, or Q, of the resonance modes decrease, which is always a consequence of higher losses. It has been clearly demonstrated that uncoated and filters coated with an oxidizing material behave equally (Fischerauer et al., 2010b).
The results in Fischerauer et al. (2010b) were obtained only at room temperature but they could be reproduced for several different DPF devices. Very recently, dynamic tests were conducted in a dynamometer test bench. By tracing one resonance peak, it should be possible to estimate the soot mass continuously from the magnitude of the reflection coefficient S 11 at a resonance, the resonance frequency, or the resonance Q. In the following, special attention will be paid to the frequency shift, as this parameter appears to be the most promising one for an application in the engine exhaust.  In a typical experiment, the DPF was continuously soot-loaded in a test bench at constant speed and load (Fig. 7). During pauses every hour, the DPF was dismounted, weighed, and mounted again in order to determine the actual soot mass inside the filter. As can be seen, a very good agreement between differential pressure (Dp), resonance frequency, and relative soot mass (in g/L) can be observed.
In another test, soot mass increase during loading and regeneration was observed. Here, regeneration was carried out at engine idle by increasing the exhaust gas temperature with a Diesel-fueled burner upstream of the DPF. The results in Figure 8 confirm the good correlation of differential pressure (upper curve) and resonance frequency (lower curve) not only during soot accumulation in the DPF, but also during filter regeneration. After each regeneration (in Fig. 8 at 1.1 h and at 2.2 h), both measuring methods indicate a soot-free DPF. Hansson and Ingestro¨m (2012), report on the General Electric system that uses two antennas and considers the attenuation losses. It has been shown that an average value of |S 21 | over a distinct frequency range is also a good indicator for the soot loading. Cross-effects of temperature and mass flow are also studied. Similar to the TWC, the temperature dependence cannot be neglected and needs to be corrected. The space velocity seems to affect the average |S 21 | value only marginally. These results have been independently confirmed by Feulner et al. (2013b) on a separate system.
The feasibility of the microwave-based approach for DPF soot loading has been demonstrated by several groups. Future work needs to address interfering effects like the influence of humidity, which modifies the soot, or the dependence on the exhaust temperature. It has also been investigated whether soot and ash loading can be distinguished. It would be beneficial if substrates of electrically conducting silicon carbide could be monitored as well.

Comparison of the Magnitude of the Resonance Frequency Change for Different Systems
To compare the influences of the loading states on the resonance frequency changes for different exhaust gas aftertreatment systems, one may define a relative measure (or figure of merit) by: here, Df res is the resonance frequency difference between the fully loaded state (100% loaded, "full") and the unloaded state (0%, "empty"). f res denotes the resonance frequency in the fully loaded state. Therefore the relative measure F, which may also be considered as a relative sensitivity, expresses the resonance frequency shift in ppm per % (relative) load of NO x , NH 3 , O 2 , or soot in the respective catalyst or filter devices. The most prominent effect occurs in DPF. Feulner et al. (2013b) has shown very recently that the frequency of the first resonance mode changes at an exhaust temperature of 235°C from 1 235 to 1 178 MHz when the aluminum titanate DPF gets loaded with 4.5 g/L soot (uncoated DPF, ;5.66" 9 6"). If one assumes a maximum acceptable soot load of 10 g/L one would obtain F = 1 025 ppm/% soot load. If only 5 g/L can be tolerated, the resonance frequency still changes by about 500 ppm/%  Differential pressure (upper graph) and resonance frequency of a selected cavity mode (lower graph) during two cycles of loading at constant speed and load (T % 350°C) and during regeneration (idle phase with external burner). From Feulner et al. (2013a), with kind permission from Springer Science + Business Media.
(or 600 kHz/%) soot load. At the moment, interfering parameters like adsorbed water are under study. In TWC, a marked resonance frequency change can be found as well. For a platinum-coated, ceria-zirconiacontaining TWC (diameter 4.66" 9 5") at 450°C, F can reach values of almost 300 ppm/% oxygen load (Reiß et al., 2011c). Since the resonance frequency change strongly depends on the effective conductivity change of the catalyst or filter material, it becomes clear that a system like a DPF, in which the conductivity changes by orders of magnitude from almost zero to a very high value because of soot loading, shows a much more pronounced response than a ceria-zirconia coated cordierite substrate (TWC). After all, the conductivity of ceria-zirconia changes only by approximately three orders of magnitude between an oxidized and reduced state (Izu et al., 2008). It has been known for many years that zeolites change their conductivity with ammonia loading (Simon et al., 1998;Rodrı´guez-Gonza´lez and Simon, 2010), but these effects are in the order of only half a magnitude. Therefore, it is astonishing that for SCR catalysts effects in the range of 50 ppm/% ammonia load occur at 300°C (Reiß et al., 2011d). In recent but yet unpublished work, we even found 80 ppm/% for serial type Cu-Chabazite zeolites at 200°C. However, it is questionable, whether such a sensitivity is high enough to measure the ammonia load accurately, especially if one considers that water adsorption as well as temperature also affect the electrical properties of zeolites. If one calculates the sensitivity for NSC from the data of Fremerey 2011, F % 11 ppm/% can be found, a value which seems by far too low for real-world applications but which can be explained if one keeps in mind that the conductivity of the NSC coating changes only by a few tens % from the NO x -depleted to the fully NO x -loaded state (Moos et al., 2008a).

CONCLUSION
We have given an overview of the state of the art in the microwave-based catalyst state observation technique.
Although the application of the method to the in situ observation of catalysts and other electrochemical systems is relatively young, it has already attracted much attention. In contrast to the conventional lambda probes, NO x sensors, or ammonia sensors, the key parameter "catalyst state" can be determined directly.
The state of the art as presented in this paper can be summarized as follows: firstly, it has been demonstrated for various systems (TWC, LNT, SCR catalyst, DPF) that the method is applicable in principle. Both test bench and dynamometer test runs have proven that suitable features of the microwave signals (S-parameters) correspond very well with catalyst states obtained indirectly by physical quantities such as lambda values upstream and downstream of a TWC or differential pressure and soot mass in a DPF. Secondly, system-level simulations have indicated that the microwave-based look into the catalysts allows closed-loop control schemes with better characteristics than are possible with the current indirect (model-based) measurement methods. Thirdly, the demonstrator setups rely heavily on laboratory equipment and as such cannot be implemented in actual vehicles. And finally, some influencing effects such as interfering analyte concentrations or temperature have been studied and quantified. However, neither can this be called exhaustive nor is it appropriate to state definitively that one can cope with the influencing effects in a manner satisfactory for practical applications. Further investigations are required. Therefore, if one intends to bring such a system into serial production, many issues need to be solved: the influence of temperature, the effect of noise factors, the long-term stability of the coupling antenna, and, of course, the cost aspects of the entire system including electronics. As to the latter, it may be remarked that, fortunately, the low GHz range is already used for communications purposes (cellular phones, Wi-Fi, bluetooth, wireless sensor networks, etc.). As a result, the required circuitry has become quite inexpensive. There can be no doubt, therefore, that it is possible to replace the laboratory equipment used so far (in particular, the VNA) by far less expensive customized circuitry. It remains to be seen which increase in measurement uncertainty this step will incur. Also, one has to carefully calculate whether the possible advantages of such a system, like reduced catalyst volume, reduced amount of noble metals, omission of one exhaust gas sensor, and improved accuracy of the catalyst state model, will outweigh the additional cost.