Self-heating in pulsed mode for signal quality 
improvement: application to carbon nanostructures-based 
sensors 
 
O. Monereo*, O. Casals, J. D. Prades and A. Cirera 
MIND-IN2UB, Department of Electronics, University of Barcelona, C/ Martí i Franquès 1, 
E-08028 Barcelona, Spain 
*Corresponding author e-mail: omonereo@el.ub.edu. 
Other authors e-mails: O. Casals (ocasals@el.ub.edu), J.D. Prades (dprades@el.ub.edu), A.Cirera 
(acirera@el.ub.edu). 
 
Abstract. 
Sensor signal instability and drift are still unresolved challenges in conductometric gas sensors. 
Here, the use of self-heating effect to operate a gas sensor in a pulsed temperature modulation 
mode (pulsed self-heating operation) is presented as an effective method to enhance signal 
stability and reduce consumption figures down to a few W. The sensor operation temperature 
was pulsed periodically between two levels, obtaining two different sensing states from one single 
device driven with self-heating, i.e. free of heater. The signal differences between both operating 
points correlated well with gas concentrations and displayed no drift. This methodology is 
exemplified with a thorough study of the response of carbon nanofibers to humidity. Specifically, 
after analyzing the influence of the pulse characteristics (i.e. temperature variation, pulse period 
and pulse duty cycle) on the sensor performance, thumb rules to select suitable pulsing conditions 
are provided. The methodology is successfully extended to other target gases, such as NO2 and 
NH3. Finally, its implementation in a real-time sensing system with low computational 
requirements is demonstrated and discussed in detail.  
Keywords: self-heating, low power consumption, high stability, pulsed sensing operation, 
conductometric gas sensor, nanomaterials.  
Highlights:  
 Pulsed self-heating operation is a simple and powerful methodology to improve the signal 
quality of a conductometric sensor. 
 Pulsed self-heating operation reduces power consumption to the range of W, without 
the need of incorporating heating elements. 
 Sensor signal can be systematically improved by understanding the role of pulsing 
temperature, period and duty cycle, and then optimizing these parameters. 
 For long term operation, the sensor response was found to be stable after 20 hours of 
continuous operation. 
 The required signal processing is simple enough to be implemented in real time systems 
with low computing power.  
*Manuscript
Click here to view linked References
1. Introduction 
Sensor instability is an usual, still unsolved, issue in conductometric gas sensing devices [1]. 
Usually, the transduction mechanisms of conductometric sensors rely on complex chemical 
interactions that can modify the sensing material properties. Such chemical modifications [5] (that 
include formation of surface complexes [2], molecule migration [3], surface poisoning with strong 
analyte adsorption [4–6], and gas diffusion inside the material [7]) can cause non-reversible 
changes of the sensor resistance. Additionally, the device structure itself can evolve during 
operation, suffering  structural [8,9] and phase [10] transformations of the sensing layer, or 
degradation of the contacts [11] and heating elements [12], leading to permanent sensor 
modifications. All these effects can deteriorate the sensing capacity causing signal drift, baseline 
offsets, decay of sensing response or conductivity loses. Typically, in order to compensate these 
issues and prevent false readings, tedious periodic sensor calibration is required.  
Improvements in baseline stability and reversibility can also be obtained by modifying the 
operation conditions. Traditionally, this means (1) modulating the sensor temperature [13–15] or 
(2) applying other energy sources as ultraviolet light [16] or visible light [17]. However, these 
techniques increase the power budget (typically 100 mW to 1 W in commercial devices [18]) and 
the device complexity and cost. Today, the most efficient designs can reduce power consumption 
down to tens of mW (with state-of-the-art micro heaters [19] or LEDs [20]). 
A cost-effective alternative [21] is the use of the Joule effect or the so-called self-heating effect 
[6,20,22]. In this mode of operation, the power dissipated by the probing current/voltage applied 
to measure the resistance heats up the sensor material. The self-heating effect was first observed 
in one single SnO2 nanowire [23,24], obtaining devices operated with less than 10 W. During 
the last years, self-heating was used in devices formed by one single nanostructure [25,26] or a 
few ordered nanostructures [23]. The complexity of these configurations hampered the 
widespread use of this approach. Recent reports demonstrate that the self-heating effect also 
occurs in large arrays of randomly deposited nanofibers [21], achieving power savings similar to 
those observed in highly ordered nanostructures (~ W) [23,24] with the advantage of a less 
complex fabrication.  
Temperature modulation was used in the past to enhance the selectivity towards the desired target 
gases [22,27] or to (partially) regenerate the sensor surface to the initial state by means of 
temperature cycling operation [28]. In addition, the application of temperature cycles is 
compatible with multidimensional data processing techniques (e.g. principal component analysis, 
PCA or/and linear discriminant analysis, LDA [29,30]) as the different spectra of temperatures 
applied to the sensor could be considered as a multidimensional stream of data for a particular 
odor.     
In this article, self-heating and pulsed temperature modulation are combined to obtain two signals 
corresponding to two operation states (i.e. low and high temperature). The results demonstrate, 
that the combination of this two operation points is enough to detect and correct signal instabilities 
and drift. 
 Here, the advantages of pulsed self-heating operation are exemplified using carbon nanofibers 
[31,32] (from now on CNFs) as the active sensing material. CNFs are a carbon allotrope with 
sensing characteristics similar to carbon nanotubes and graphene [32–36] (e.g. low temperature 
gas sensing, below 100 °C). 
In the following article, the role of the pulsing characteristics on the sensor performance is 
analyzed in detail for humidity sensing. The method is then extended to other gases and 
implemented in a real-time sensing system. All this demonstrates that pulsed self-heating 
operation is a convenient strategy to improve signal quality while keeping low the power and cost 
budget in simple, heater-free, devices. 
 
2. Materials and methods 
Sensor platforms formed by interdigitated platinum electrodes (IDE) over a ceramic substrate 
fabricated by Francisco Albero S.A. were used [37]. The sensor platform included an embedded 
thermoresistance and a heater. This elements were only emplyed for temperature calibration of 
self-heating operation (detailed information can be found elsewhere [21]) and for comparison  
between sensor operation at constant temperature and pulsed self-heating methodology. 
Commercially available CNFs, synthesized by means of CVD floating catalyst technique by 
Grupo Antolín S.A., were used. The fibers presented diameters ranging from 30 to 80 nm and 
lengths up to 75 m. Additional details of this material can be found elsewhere [38]. CNFs were 
dispersed in 2-propanol at a concentration of 1 mg/ml. Then, the dispersion was drop-casted over 
the IDE. Electrical resistance was monitored during the deposition process with a digital 
multimeter Agilent 34401A to assess the continuity of the film.  
The electrical measurements of sensor resistance were recorded using a Keithley 2401 source 
meter. The same unit served as a measurement system and as a power source to activate the self-
heating effects. Gas sensing experiments were conducted in a customized chamber of 15 ml in 
volume. The gas flow was maintained at 200 ml/min during all the experiments. In these 
conditions, the gas chamber can be fully flushed five times in less than 20 seconds. Reference 
gaseous atmospheres were provided by several independent mass flow controllers blending dry 
N2, synthetic air (SA, 79 % of N2 and 21 % of O2), NH3 (100 ppm in SA) and NO2 (10 ppm in 
SA). Constant relative humidity (from now on RH) levels were obtained by bubbling dry SA into 
deionized water (relative humidity obtained at a constant temperature of 25 °C at a pressure of 1 
atm).  
2.1. Signal response definition 
In the presented work, the sensors were operated using two methodologies: at constant operation 
temperature (for comparison purposes) and pulsed self-heating mode (see Fig. 1). The first 
method implies using a single operation temperature during all the experiment (see Fig. 1(a)). In 
this case, the sensor response (from now on, classic sensor response or Sclassic) was defined as 
Sclassic (%) = 100 · (R - Ro) / Ro, where R is the steady resistance signal in each exposure to gas 
and Ro is the baseline value of sensor resistance. For the second method, the pulsed self-heating 
mode, the sensor temperature was pulsed between two states by applying two DC voltage levels 
with a constant period (from now on Tp) (see Fig. 1(b)). Different measurements were also 
performed varying the pulse duty cycle (from now on, D). D was defined with respect to the high 
temperature state (D = 10 % corresponds to 10 % of the period at high temperature and 90 % at 
low temperature). Using the pulsed methodology, the sensing response was no longer defined 
directly in terms of the sensor resistances but in relation to the difference of resistance between 
both sensing states (from now on R), extracted as the difference of the upper envelope of the 
resistance signal (corresponding to the lower temperature state) and the lower envelope of the 
resistance signal (corresponding to the higher temperature state): 
R = Upper Envelope - Lower Envelope (see Fig. 1(b)). Therefore, the sensor response was 
defined as SR (%) = 100 · (R - Ro) / Ro, where R is the steady difference of resistance in 
presence of gas between the high and low temperature states and Ro is the baseline difference of 
resistance between the high and low temperature states.  
 
 
Fig. 1 Comparison between (a) classic sensor operation working at constant temperature (60 °C, 
12 mW) and (b) pulsed self-heating operation, where two temperatures are used (25 °C, 10 W; 
60 °C, 10 mW) with Tp = 60 seconds and D = 50 %. Both sensors were exposed to pulses of 50 
% of RH during 10 minutes. From the upper and lower envelopes, a new signal (R) can be 
extracted. 
3. Results & discussion 
3.1. Intrinsic sensor signal instability 
Conductometric sensor signals usually present signal instabilities due to several phenomena 
occurring at the sensor active material or in the device structure. Fig. 2(a) shows the typical 
resistance record of CNFs at a constant temperature. At room temperature, CNFs display 
significant drift and poor reversibility. In fact, the signal is extremely sensitive to the previous 
sensor history: exposure to high humidity contents leads to signal instabilities that are not fully 
recovered after removing the moisture. This situation improves slightly at a higher temperature 
(60 ºC), but Sclassic is reduced. Fig. 2(b) shows a summary of this trade off: temperature is an 
effective method to reduce drift but reducing systematically the response and increasing the power 
consumption. Anyhow, the drift improvement is not complete until the sensor response is almost 
vanished (T > 120 °C). Therefore, constant temperature excitation does not solve in full the signal 
instability issues. 
 Fig. 2 (a) CNFs sensor signals obtained for a pseudo-random sequence of relative humidity 
concentrations at temperatures (T) of 25 °C and 60 °C. (b) Sensor response (Sclassic) for a pulse of 
100 % of relative humidity (squares), its signal drift normalized to the initial value of sensor 
resistance (Ro) in each experiment (circles) and the power consumed by the embedded heater 
(triangles). 
3.2. Instability correction 
Usually, the classical signal response to gases is modulated by the operation temperature of the 
sensor. The adsorption of gas molecules on the sensor active material surface is dependent on the 
temperature (e.g. Langmuir adsorption model [39], Freundlich and Temkin isotherms [40], etc.). 
Following the methods described previously, sensor temperature was pulsed periodically between 
two levels by means of the self-heating effect (see Fig. 1 and Fig. 3(a)) in order to obtain 
information from the high and low temperature operation states. Since CNFs display 
semiconductor behavior, temperature changes lead to two sensor resistance values: lower 
resistance at high temperature and higher resistance at low temperature. Therefore, an upper and 
a lower envelope of the pulsed resistance signal could be extracted (e.g. maximum and minimum 
resistance values reached in each pulse step). The envelopes corresponding to both states were 
subtracted to obtain R, a signal presenting an amplitude modulation proportional to gas 
concentration (see Fig. 3(b)). This dependence is related to the differences in surface adsorption 
due to temperature.  
Moreover, as R is a combination of two sensing states originated in the same sensor, both 
baseline signals are completely correlated: simple subtraction seems to be enough to remove 
instability effects as undesired changes of the active film properties (i.e. poisoning, chemical 
states modification, degradation, etc.), which are simultaneous to both operation conditions. At 
the same time, R shares the good characteristics of both envelopes: good response at low 
temperature and fast response at high temperature.  
In essence, with a pulsed self-heating operation, two signals can be extracted from the same sensor 
device helping to improve the sensing characteristics dramatically (see Fig. 3(a) vs. Fig. 3(b)), 
without increasing the amount of system components. In contrast with sensor arrays, all signals 
are generated in the same device. Therefore, all of them are submitted to the same instabilities, in 
full correlation. 
 
Fig. 3 (a) Pulsed CNFs sensor between 25 °C and 60 °C by means of self-heating. Signal 
envelopes corresponding to a high and low sensor operation point are plotted. In the upper-left 
inset: pulse characteristics scheme for the probing voltage, equivalent temperature is also shown. 
In the upper-right inset: detail of the transient CNFs signal. (b) Difference of the extracted 
envelopes: R. Inset: Response obtained from R signal (SR). 
3.3.  Thermal influence vs. gas effects 
In self-heated systems, the temperature variation can be much faster than in their heater-driven 
counterparts [41]. For comparison, Fig. 4(a) shows the resistance variation in equivalent pulsed 
heating conditions (between 25 °C and 70 °C), driven with self-heating and with the substrate 
heater. A pulse with Tp = 120 s was used to ensure full thermal stabilization. In all the tested 
devices, the self-heating response was almost four times faster. This is due to the much higher 
thermal inertia in the heater-driven system, in which the dissipated power heats not only the CNFs 
array, but also, the heater/substrate mass. Consequently, in heater-driven systems more power 
dissipation is needed for equivalent temperatures, and longer thermal stabilization times are 
expected. 
A detailed analysis of the resistance records in self-heating operation, showed two dynamic 
regimes during the heating/cooling process (Fig. 4(a)): a fast initial variation (in the scale of 
milliseconds) followed by a slower stabilization of the signal (in the scale of seconds). The 
phenomena occur symmetrically both in heating and cooling cycles. 
In atmospheres that are typically non-reactive for the CNFs, such as low vacuum (3·10-5 hPa), N2, 
and SA atmospheres (see Fig. 4(b)), both regimes are insensitive to the presence of one or another 
gas. 
In contrast, in presence of humidity (25 %, 50 % and 100 %, Fig. 5(a)), significant differences are 
observed in the resistance change during the slow regime (slow processes labeled as I and III in 
Fig. 5). The fast processes, in turn, were insensitive to humidity variations with a nearly constant 
resistance change (fast processes II and IV).  
Therefore, the fast processes (cooling II and heating IV) are exclusively related to the temperature, 
and the slow processes (cooling I and heating III) correlate with the presence of gases (Fig. 5(b)). 
This explains why the magnitude R (which is the accumulated resistance change occurred after 
I, II, III and IV) reflects the signal variation only dependent on the gas concentration. 
 Fig. 4 (a) Relative difference of the CNFs sensor signal pulsed in equivalent conditions using the 
external heater and self-heating operation, the time needed to reach the 90 % of the total relative 
change is shown (90%). (b) Transient resistance signal of a pulsed CNFs sensor between 0.1 V 
(25 °C) and 3 V (60 °C) in N2 and SA atmospheres, no change in the sensor signal is appreciable. 
Upper-left inset: pulse characteristics for the probing voltage, equivalent temperature is also 
displayed. Upper-right inset: comparison of the pulsed resistance signal (25 °C – 60 °C) in 
vacuum (3·10-2 hPa), N2 and SA atmosphere. 
 Fig. 5 (a) Transient resistance signal of a pulsed CNFs sensor between 0.1 V (25 °C) and 3 V (60 
°C) in SA atmospheres with 0 %, 25 %, 50 % and 100 % of relative humidity. Signal envelopes 
are displayed and normalized to the resistance corresponding to the lower envelope (RLower Envelope). 
(b) R corresponding to the different processes occurring during pulse: slow cooling (I) and 
heating (III), fast cooling (II) and heating (IV). 
3.4.  Effect of the pulsing conditions 
In order to elucidate the dependence of R magnitude on the self-heating pulse conditions, the 
pulsed temperatures, period (TP) and duty cycle (DC) were studied in detail for humidity sensing. 
3.4.1. Pulsed temperatures 
Fig. 6(a) shows the response to different humidity contents of a sensor pulsed between a reference 
temperature (25 ºC set with 0.1 V) and different high temperature states (200 °C, 100 °C, 50 °C 
and 35 °C, set with 8 V, 5 V, 3 V and 1 V, respectively). After extracting the envelopes and their 
difference, the R signal (Fig. 6(b)) was fully stable throughout the experiment and scaled up 
with the relative humidity content. In general, larger Ro values were obtained increasing the 
temperature difference in the pulses but also similar SR. Thus, for this particular target gas, larger 
temperature spans do not lead to any particular benefit. In low temperature span measurements, a 
large response only to pulses of saturated humidity was detected. This behavior is attributed to a 
partial wetting of the sensor surface.  
 
Fig. 6 Transient sensor response and R obtained for a series of relative humidity pulses (100 %, 
50 %, 25 % and 50 %). Sensor resistance pulsed between 25 °C and 200 °C, 100 °C, 50 °C and 
35 °C. Inset: Responses obtained from R signal.  
3.4.2. Pulse period (Tp) 
Pulsed temperature operation keeps the sensor surface swinging between two gas/surface 
interaction states. Such periodic transition involves the adsorption and subsequent desorption of 
gaseous molecules from the surface. Only for long Tp, the adsorption and desorption processes 
are completed reaching steady resistance values. In fast pulsing periods (with low Tp) adsorption 
and desorption are suddenly interrupted by a temperature change. Therefore, the resistance 
readings always correspond to surface transient states in which the sensor surface sites are not 
completely fill or empty (Fig. 7 (a)). 
In this context, the concepts of response and recovery times are artificial and essentially related 
to the Tp value of choice. According to our definition of R (i.e. the difference between the 
envelopes, or the maximum and minimum resistance values during the hot and cold semiperiods 
(Tp/2)), the R signal is sampled once in every pulsing period and this defines the effective 
“response and recovery” times of the sensor system. Therefore, we could consider that the R 
signal changes “immediately” after the temperature change, provided that there has been a change 
in the gas concentration.  
Therefore, the period of the pulsed signal should be enough to ensure, at least, a partial gas 
adsorption-desorption process. In addition, it is possible to lower the pulsed period to improve the 
response and recovery times almost at will. 
A summary of the dependence of R with the pulsing period is shown in Fig. 7(b). All curves 
follow a behavior similar to that of a low-pass frequency filter modulated by the gas 
concentration: as water molecules have more time to interact with CNFs surface during each TP/2, 
R increases with both the amount of humidity and with Tp. In this sense, very short periods lead 
to very low R variations. Interestingly, the pulsed response SR is constant for pulse periods 
longer than the typical response time to humidity in constant temperature operation (e.g. 60 s for 
100 % of RH at 25 °C in constant operation [21]).  
Summarizing, the pulsing period Tp determines the response and recovery times of the pulsed 
signal. At the same time, the pulsing period serves to adjust the R variation with humidity, 
obtaining larger responses for longer periods, until reaching the steady response limit. All this 
leads to a compromise between response dynamics and response magnitude that could be used to 
provide some specificity over different target gases. 
 Fig. 7 (a) Detail of the transient resistance signal for pulse period of 5, 15, 30, 60 and 120 seconds 
in an atmosphere of RH = 100 %. (b) Dependence of R with the pulsed signal period. The 
response time for continuous sensing mode at 25 °C is marked with a vertical dashed line. 
3.4.3 Pulse duty cycle (D) 
Different pulse duty cycles D (10 %, 30 %, 50 %, 70 % and 90 %) were tested in similar conditions 
(see Fig. 8).  Essentially, the lower the D the lesser the time the sensor remains at the higher 
temperature state with the benefit of reducing the power consumption. From the gas/surface 
interaction perspective, D tailors the adsorption/desorption balance.  
For a given period configuration and provided that the D is large enough to ensure a quantifiable 
gas adsorption-desorption process between both temperature states, the shape of R does not vary 
significantly with D (see Fig. 8(b)).  
Therefore, D modulation offers chances to further reduce the power consumption. The average 
power (Pavg) consumed in pulsed operation can be calculated as Pavg = D · Phigh + (1 - D)·Plow, 
where D is the duty cycle of the pulsed signal, Phigh and Plow is the average power consumed at 
the high and low temperature sensing condition, respectively. Since Plow << Phigh, the dependence 
in Pavg approximates to Pavg ≈ D · Phigh, in accordance with the results shown in  Fig. 9. 
Consequently, Pavg can be reduced up to two orders of magnitude varying the DC from 90 % to 
10 % (the lowest Pavg value obtained was (9.58 ± 0.02) W, Fig. 9), without losing any sensing 
capability.  
 
 
Fig. 8 (a) Pulsed self-heating measurements using voltages of 0.1 V and 2 V (corresponding to 
25 °C and 45 °C) with a pulse period of 1 minute in different concentrations of humidity (100 %, 
50 %, 25 % and 50 %). On the right, detail of the last humidity pulse at 50 % of relative humidity 
(shaded time frame). (b) Extracted signal R from the data showed at (a).  
 Fig. 9 Average power, Pavg, vs. duty cycle (D) dependence for three sensors with different baseline 
resistance Ro (the value Ro was obtained in steady state conditions, being the sensor in a SA 
atmosphere during two hours). Inset: Pavg dependence with the baseline resistance,. Ro. 
 
3.5.  Long term stability 
The baseline stability and the sensing response repeatability were analyzed. A CNFs sensor was 
maintained at stable atmosphere of pure SA for 78 hours and at 50 % of RH for 72 hours (Fig. 
10(a)). The resistance sensor signal drifted along the measurement due to variations of the ambient 
temperature. In contrast, the signal R was found to reduce the deviation of the sensor signal by 
a 10 factor for both the SA and with 50 % of RH. A humidity test with RH concentrations of 15 
%, 50 %, 75 % and again 50 % was repeated during 66 hours. The R magnitude showed an initial 
stabilization period of about 15 hours. After that, the responses were sufficiently stable to 
discriminate each concentration. In summary, the pulsed self-heating methodology presents a 
highly repeatable behavior without losing sensing capabilities. 
 
 Fig. 10 (a) Long term stability study for a pulsed self-heated CNFs sensor for 150 hours (78 hours 
at pure SA atmosphere and 72 hours at 50 % of relative humidity). Resistance and R signal 
represented with same span. (b) Repeatability of the Ro (0 % of RH) and R magnitude to pulses 
of humidity (25 %, 50 %, 75 % and 50 % of RH), the gas test sequence was repeated 
uninterruptedly during 67 hours. On the right: Gaussian distributions fitted to the histograms of 
R measurements at each concentration. 
3.6.  NO2 and NH3 pulsed sensing 
Carbon materials, such as CNFs, are also sensitive to other gases such as NO2 and NH3 [32]. In 
order to prove that the pulsed self-heated methodology is suitable to sense gases other than water 
vapor, experiments with NO2 and NH3 were conducted at different pulsing temperatures 
(reference temperature at 25 ºC and high temperature states at 30 °C, 35 °C, 45 °C, 70 °C and 80 
°C). Fig. 11 shows that the benefits in signal stability are also obtained in presence of NO2 and 
NH3. Moreover, the differences in R response to NO2 and NH3 are much larger at higher 
temperature variation during the pulse. Therefore, the use of different pulsing conditions partially 
modifies the selectivity towards different gases, opening the door to further investigations on the 
role of pulsed self-heating in gas selective-sensing. 
 
 
Fig. 11 Resistance signal and R for five tests of 100 ppm of NH3, 50 % of RH and 2.5 ppm of 
NO2. Test temperatures between room temperature (25 °C) and 30 °C, 35 °C, 45 °C, 70 °C and 
80 °C.  
 3.6 Implementation in real environments 
The results presented so far relay on the processing of the pulsed resistance values to extract the 
R signal. In this section we show that the R calculation is a very simple procedure that could 
be implemented in an embedded microcontroller, with low computing power.  
A system to extract the R magnitude should carry out the following tasks:  
1. set the probing voltage to the high temperature operation point,  
2. wait for a settling time, defined by Td ·D, to reach the high temperature state 
3. acquire one measurement corresponding to one data point of the lower envelope signal 
(Rh)  
4. set the probing voltage to the low temperature operation point,  
5. wait for the setting time, defined by Td · (1 - D), to reach the low temperature state 
6. acquire one measurement corresponding to one data point of the upper envelope signal 
(Rl)  
7. Calculate one signal sample as R = Rl - Rh with the last pair of data values acquired. 
8. Repeat from 1.  
With this algorithm, no filtering is needed to extract the envelopes and no large data buffers are 
required: only the difference between two consecutive data points is calculated.  
In order to exemplify the feasibility of implementing such approach in a real-time system, a 
LabVIEW® software was designed to pulse the sensor temperature with a source-meter, to acquire 
the sensor resistance values and to compute the R signal. The software also allowed us to change 
the pulsed conditions (high and low pulsing voltage levels, pulse period and pulse duty cycle) in 
real time during the measurements. Fig. 12 (a)-(c) shows the effect of the pulse conditions of the 
real time signal extracted from the sensor, which are essentially the same issues discussed in the 
previous sections. To prove the capability of the system to detect ambient gas variations in real 
time, the sensor was kept in an ambient atmosphere and exposed to human breath Fig. 12 (d), 
displaying fast responses (less than 1 s) full signal recovery, no drift with power consumption of 
less than 1 mW. 
In summary, the computing power needed to perform the described tasks should be compatible 
with sensor systems operating at low energy, which could be assured because of the low power, 
heater-free, operation of the sensor device. 
 
 
 Fig. 12 Test performed with a source-meter unit controlled by a LabVIEW® software. The 
program compute the R with a pair of points, both acquired just before the change of sensor 
operation point and allow in-situ modification of the pulsing conditions to the user. (a) 
Modification of the high and low voltage levels. (b) Modification of the period. (c) Modification 
of the signal duty cycle. (d) Detection of human breath in ambient atmosphere. Resistance signal 
and the computed R are shown.  
 
 
4. Conclusions 
Pulsed self-heating is a reliable method to extract a highly stable signal only dependent on the gas 
presence. The methodology successfully removes drifts due to the sensor instabilities, such as 
chemical surface modification or structural degradation. 
With this approach, it is also possible to reduce the power consumption needs down to a few W, 
in an extremely simple, and cost effective, heater-free configuration. 
The pulsed temperatures , the period of the pulse and the duty cycle of the probing signal can be 
used to modulate the resulting gas sensing response, optimizing response time, response 
magnitude and power consumption respectively. Whether sensing conditions such as the pulse 
period could also provide specificity is still an open question. 
The presented methodology was used to sense humidity, NO2 and NH3 with carbon based sensors 
(CNFs sensors), suggesting that the same approach could be extended to other gases and 
materials, such as metal oxides.  
 
Acknowledgments 
The research leading to these results has received funding from the European Research Council 
under the European Union's Seventh Framework Programme (FP/2007-2013) / ERC Grant 
Agreement n. 336917.  
J.D. Prades acknowledges the support from the Serra Húnter Programme. 
A. Cirera acknowledges support from ICREA Academia program. 
 
5. Bibliography 
[1] G. Korotcenkov, B.K. Cho, Instability of metal oxide-based conductometric gas 
sensors and approaches to stability improvement (short survey), Sensors Actuators 
B Chem. 156 (2011) 527–538. doi:10.1016/j.snb.2011.02.024. 
[2] J.L. Figueiredo, M.F.R. Pereira, M.M.A. Freitas, J.J.M. Órfão, Modification of the 
surface chemistry of activated carbons, Carbon N. Y. 37 (1999) 1379–1389. 
doi:http://dx.doi.org/10.1016/S0008-6223(98)00333-9. 
[3] M. Liess, Electric-field-induced migration of chemisorbed gas molecules on a 
sensitive film—a new chemical sensor, Thin Solid Films. 410 (2002) 183–187. 
doi:10.1016/S0040-6090(02)00209-2. 
[4] X. Liu, S. Cheng, H. Liu, S. Hu, D. Zhang, H. Ning, A survey on gas sensing 
technology, Sensors. 12 (2012) 9635–9665. doi:10.3390/s120709635. 
[5] D. Vincenzi, M. Butturi, V. Guidi, M. Carotta, G. Martinelli, V. Guarnieri, et al., 
Development of a low-power thick-film gas sensor deposited by screen-printing 
technique onto a micromachined hotplate, Sensors Actuators B Chem. 77 (2001) 
95–99. 
[6] D.R. Kauffman, A. Star, Carbon nanotube gas and vapor sensors, Angew. Chemie 
- Int. Ed. 47 (2008) 6550–6570. doi:10.1002/anie.200704488. 
[7] B. Kamp, R. Merkle, R. Lauck, J. Maier, Chemical diffusion of oxygen in tin 
dioxide: Effects of dopants and oxygen partial pressure, J. Solid State Chem. 178 
(2005) 3027–3039. doi:10.1016/j.jssc.2005.07.019. 
[8] J.G. McCarty, M. Gusman, D.M. Lowe, D.L. Hildenbrand, K.N. Lau, Stability of 
supported metal and supported metal oxide combustion catalysts, Catal. Today. 47 
(1999) 5–17. doi:10.1016/S0920-5861(98)00279-X. 
[9] G. Korotcenkov, V. Brinzari, M. Ivanov, A. Cerneavschi, J. Rodriguez, A. Cirera, 
et al., Structural stability of indium oxide films deposited by spray pyrolysis during 
thermal annealing, Thin Solid Films. 479 (2005) 38–51. 
doi:10.1016/j.tsf.2004.11.107. 
[10] P. Wynblatt, G.S. Rohrer, F. Papillon, Grain boundary segregation in oxide 
ceramics, J. Eur. Ceram. Soc. 23 (2003) 2841–2848. doi:10.1016/S0955-
2219(03)00308-X. 
[11] H. Meixner, U. Lampe, Metal oxide sensors, Sensors Actuators B Chem. 33 (1996) 
198–202. doi:10.1016/0925-4005(96)80098-0. 
[12] H. Esch, G. Huyberechts, R. Mertens, G. Maes, J. Manca, W. De Ceuninck, et al., 
The stability of Pt heater and temperature sensing elements for silicon integrated 
tin oxide gas sensors, Sensors Actuators B Chem. 65 (2000) 190–192. 
doi:10.1016/S0925-4005(99)00301-9. 
[13] T.A. Kunt, T.J. McAvoy, R.E. Cavicchi, S. Semancik, Optimization of temperature 
programmed sensing for gas identification using micro-hotplate sensors, Sensors 
Actuators B Chem. 53 (1998) 24–43. doi:10.1016/S0925-4005(98)00244-5. 
[14] J.D. Fowler, M.J. Allen, V.C. Tung, Y. Yang, R.B. Kaner, B.H. Weiller, Practical 
chemical sensors from chemically derived graphene., ACS Nano. 3 (2009) 301–
306. doi:10.1021/nn800593m. 
[15] M. Roth, R. Hartinger, R. Faul, H.-E. Endres, Drift reduction of organic coated 
gas-sensors by temperature modulation, Sensors Actuators B Chem. 36 (1996) 
358–362. doi:10.1016/S0925-4005(97)80096-2. 
[16] O. Monereo, S. Claramunt, G. Vescio, H. Lahlou, R. Leghrib, J.D. Prades, et al., 
Carbon nanofiber flexible gas sensor modulated by UV light, in: Solid-State 
Sensors, Actuator, 2013: pp. 1154–1157. 
[17] M.W.G. Hoffmann, A.E. Gad, J.D. Prades, F. Hernandez-Ramirez, R. Fiz, H. 
Shen, et al., Solar diode sensor: Sensing mechanism and applications, Nano 
Energy. 2 (2013) 514–522. doi:10.1016/j.nanoen.2012.12.003. 
[18] I. Simon, N. Bârsan, M. Bauer, U. Weimar, Micromachined metal oxide gas 
sensors: opportunities to improve sensor performance, Sensors Actuators B …. 73 
(2001) 1–26. 
[19] I.-S. Hwang, E.-B. Lee, S.-J. Kim, J.-K. Choi, J.-H. Cha, H.-J. Lee, et al., Gas 
sensing properties of SnO2 nanowires on micro-heater, Sensors Actuators B Chem. 
154 (2011) 295–300. doi:10.1016/j.snb.2009.11.012. 
[20] J.D. Prades, R. Jimenez-Diaz, F. Hernandez-Ramirez, S. Barth, a. Cirera, a. 
Romano-Rodriguez, et al., Equivalence between thermal and room temperature 
UV light-modulated responses of gas sensors based on individual SnO2 nanowires, 
Sensors Actuators B Chem. 140 (2009) 337–341. doi:10.1016/j.snb.2009.04.070. 
[21] O. Monereo, J.D. Prades, a. Cirera, Self-heating effects in large arrangements of 
randomly oriented carbon nanofibers: application to gas sensors, Sensors Actuators 
B Chem. 211 (2015) 489–497. doi:10.1016/j.snb.2015.01.095. 
[22] J.D. Prades, R. Jimenez-Diaz, F. Hernandez-Ramirez, J. Pan, A. Romano-
Rodriguez, S. Mathur, et al., Direct observation of the gas-surface interaction 
kinetics in nanowires through pulsed self-heating assisted conductometric 
measurements, Appl. Phys. Lett. 95 (2009) 053101. doi:10.1063/1.3192358. 
[23] E. Strelcov, S. Dmitriev, B. Button, J. Cothren, V. Sysoev, A. Kolmakov, Evidence 
of the self-heating effect on surface reactivity and gas sensing of metal oxide 
nanowire chemiresistors, Nanotechnology. 19 (2008) 355502. doi:10.1088/0957-
4484/19/35/355502. 
[24] J.D. Prades, R. Jimenez-Diaz, F. Hernandez-Ramirez, S. Barth, a. Cirera, a. 
Romano-Rodriguez, et al., Ultralow power consumption gas sensors based on self-
heated individual nanowires, Appl. Phys. Lett. 93 (2008) 123110. 
doi:10.1063/1.2988265. 
[25] J.D. Prades, F. Hernández-Ramírez, T. Fischer, M. Hoffmann, R. Müller, N. 
López, et al., Quantitative analysis of CO-humidity gas mixtures with self-heated 
nanowires operated in pulsed mode, Appl. Phys. Lett. 97 (2010) 243105. 
doi:10.1063/1.3515918. 
[26] J.D. Prades, R. Jimenez-Diaz, F. Hernandez-Ramirez, J. Pan, a. Romano-
Rodriguez, S. Mathur, et al., Direct observation of the gas-surface interaction 
kinetics in nanowires through pulsed self-heating assisted conductometric 
measurements, Appl. Phys. Lett. 95 (2009) 93–96. doi:10.1063/1.3192358. 
[27] P.H. Rogers, S. Semancik, Feedback-enabled discrimination enhancement for 
temperature-programmed chemiresistive microsensors, Sensors Actuators, B 
Chem. 158 (2011) 111–116. doi:10.1016/j.snb.2011.05.051. 
[28] A.P. Lee, B.J. Reedy, Temperature modulation in semiconductor gas sensing, 
Sensors Actuators, B Chem. 60 (1999) 35–42. doi:10.1016/S0925-4005(99)00241-
5. 
[29] L. Fernandez, S. Marco, a. Gutierrez-Galvez, Robustness to sensor damage of a 
highly redundant gas sensor array, Sensors Actuators B Chem. 218 (2015) 296–
302. doi:10.1016/j.snb.2015.04.096. 
[30] J. Fonollosa, L. Fernández, R. Huerta, A. Gutiérrez-Gálvez, S. Marco, 
Temperature optimization of metal oxide sensor arrays using Mutual Information, 
Sensors Actuators, B Chem. 187 (2013) 331–339. doi:10.1016/j.snb.2012.12.026. 
[31] Y.A. Kim, T. Hayashi, M. Endo, M.S. Dresselhaus, Carbon Nanofibers, in: 
Springer Handb. Nanomater., 2013. 
[32] O. Monereo, S. Claramunt, M.M. De Marigorta, M. Boix, R. Leghrib, J.D. Prades, 
et al., Flexible sensor based on carbon nanofibers with multifunctional sensing 
features., Talanta. 107 (2013) 239–47. 
http://www.ncbi.nlm.nih.gov/pubmed/23598218. 
[33] S. Claramunt, O. Monereo, M. Boix, R. Leghrib, J.D. Prades, A. Cornet, et al., 
Flexible gas sensor array with an embedded heater based on metal decorated 
carbon nanofibres, Sensors Actuators, B Chem. 187 (2013) 401–406. 
[34] E. Llobet, Gas sensors using carbon nanomaterials: A review, Sensors Actuators 
B Chem. 179 (2013) 32–45. doi:10.1016/j.snb.2012.11.014. 
[35] G. Lu, L.E. Ocola, J. Chen, Gas detection using low-temperature reduced graphene 
oxide sheets, Appl. Phys. Lett. 94 (2009) 083111. doi:10.1063/1.3086896. 
[36] G. Lu, L.E. Ocola, J. Chen, Reduced graphene oxide for room-temperature gas 
sensors., Nanotechnology. 20 (2009) 445502. doi:10.1088/0957-
4484/20/44/445502. 
[37] F.M. Ramos, C. López-Gándara, A. Cirera, J.R. Morante, Monolithic Ceramic 
Technology for Sensing Devices, in: Proc. 2009 Spanish Conf. Electron Devices, 
2009: pp. 293–296. 
[38] T.R. M. Weisenberger, I. Martín-Gullon, J. Vera-Agullo, H. Varela-Rizo, C. 
Merino, R. Andrews, D. Qian, The effect of graphitization temperature on the 
structure of helical-ribbon carbon nanofibers, Carbon N. Y. 47 (2009) 2211–2218. 
[39] S. Gomri, J.-L. Seguin, J. Guerin, K. Aguir, Adsorption–desorption noise in gas 
sensors: Modelling using Langmuir and Wolkenstein models for adsorption, 
Sensors Actuators B Chem. 114 (2006) 451–459. doi:10.1016/j.snb.2005.05.033. 
[40] G. Limousin, J.P. Gaudet, L. Charlet, S. Szenknect, V. Barthès, M. Krimissa, 
Sorption isotherms: A review on physical bases, modeling and measurement, Appl. 
Geochemistry. 22 (2007) 249–275. doi:10.1016/j.apgeochem.2006.09.010. 
[41] J.D. Prades, R. Jimenez-Diaz, F. Hernandez-Ramirez, a. Cirera, a. Romano-
Rodriguez, J.R. Morante, Harnessing self-heating in nanowires for energy 
efficient, fully autonomous and ultra-fast gas sensors, Sensors Actuators B Chem. 
144 (2010) 1–5. doi:10.1016/j.snb.2009.09.040.  
 
Self-heating in pulsed mode for signal quality 
improvement: application to carbon nanostructures-based 
sensors 
 
O. Monereo*, O. Casals, J. D. Prades and A. Cirera 
MIND-IN2UB, Department of Electronics, University of Barcelona, C/ Martí i Franquès 1, E-
08028 Barcelona, Spain 
*Corresponding author e-mail: omonereo@el.ub.edu 
Other authors e-mails: O. Casals (ocasals@el.ub.edu), J.D. Prades (dprades@el.ub.edu), A. Cirera 
(acirera@el.ub.edu.) 
 
Author biographies 
 
Oriol Monereo was graduated in Physics at the University of Barcelona in 2010 and in Master 
in Electronic Engineering in 2011. Now he is Ph.D. student in Department of Electronics of the 
same university. His current research is focused in gas sensors. 
 
Olga Casals was born in Barcelona in 1973. She received her diploma in optics and optometry at 
the Polytechnic University of Catalonia in 1994, her degree in physics at the University of 
Barcelona in 2001 and her Ph.D. at the same institution in 2012. Her professional experience is 
set in development of new technologies on solid-state gas sensors.  
 
Joan Daniel Prades was born in Barcelona in 1982. He graduated in Physics and Electronic 
Engineering at the University of Barcelona and obtained his Ph.D. at the same institution in 2009. 
He has experience in modelling of the electronic and vibrational properties of nanostructured 
metal oxides and in their experimental validation. He is actively involved in the development of 
innovative device prototypes based on nanomaterials. He has published more than 40 papers in 
peer-reviewed journals and contributed to more than 100 international conferences. He has also 
contributed to five industrial patents. 
 
Albert Cirera is graduated in Physics (1996) and Ph.D. (2000) by the University of Barcelona. 
He is currently associate professor of the Electronics Department. He has published more than 60 
papers in international journals with more than 1000 cited and h = 19; as well as more than 100 
presentations, 5 book chapters and 5 more patents, he has directed more than 15 industrial and 
research projects. His current research group deals with computational and applied 
nanoelectronics, including sensors. 
Author Biographies