Seat Occupancy Detection Based on a Low-Power Microcontroller and a Single FSR †

This paper proposes a microcontroller-based measurement system to detect and confirm the presence of a subject in a chair. The system relies on a single Force Sensing Resistor (FSR), which is arranged in the seat of the chair, that undergoes a sudden resistance change when a subject/object is seated/placed over the chair. In order to distinguish between a subject and an inanimate object, the system also monitors small-signal variations of the FSR resistance caused by respiration. These resistance variations are then directly measured by a low-cost general-purpose microcontroller unit (MCU) without using either an analogue processing stage or an analogue-to-digital converter. Two versions of such a MCU-based circuit are presented: one to prove the concept of the measurement, and another with a smart wake-up (generated by the sudden resistance change) intended to reduce the energy consumption. The feasibility of the proposed measurement system is experimentally demonstrated with subjects of different weight sitting at different postures, and also with objects of different weight. The MCU-based circuit with a smart wake-up shows a standby current consumption of 800 nA, and requires an energy of 125 µJ to carry out the measurement after the wake-up.


Introduction
Seat-occupancy monitoring systems generally use mechanical sensors that detect weight, pressure, force or acceleration over the seat. These mechanical sensors can be resistive [1], capacitive [2] or inductive [3], but the former are the most common. Different types of resistive mechanical sensors can be employed, for instance: metallic strain gauges, semiconductor strain gauges and force sensing resistors (FSR), the latter being the cheapest and the most easily integrated into the chair structure. Mechanical sensors, however, have difficulties in distinguishing between subjects and objects seated/placed over the seat. This is usually solved by comparing the set of data to known patterns [4] or by combining the information of different types of sensor. For example, in [5], the information of the resistive mechanical sensor is combined with that provided by two additional sensors: a thermal sensor and a capacitive proximity sensor. The combination of proximity sensors, instead of mechanical sensors, have also been reported for seat-occupancy monitoring systems, as suggested in [6] by combining inductive and capacitive proximity sensors.

Operating Principle
The proposed method for the seat occupancy detection and classification relies on a single FSR attached at the center of the seat. A FSR is a sensor whose resistance (R x ) decreases with increasing the force applied to it; in the application of interest, the force will be exerted by the subject/object seated/placed over the chair. This is a very low-cost, thin-size resistive mechanical sensor that can be easily integrated into the chair structure; however, it has limitations in terms of accuracy, linearity, interchangeability and time drifts. The repeatability due to time drifts can be quite critical when the FSR is subjected to a static loading for a long time interval (say, some hours), but it is quite acceptable (at least for FSRs from Interlink Electronics) when the FSR is subjected to a dynamic loading for a short time interval (say, a few seconds) [24]. The previous limitations are not expected to be critical here since the aim is to monitor resistance variations due to respiration in a few seconds, and not to determine the exact value of the subject/object weight nor the respiratory rate. The operating principle of the proposed measurement method is shown in Figure 1, where three scenarios are possible: (1) The seat is vacant: R x remains constant with respect to R x,0 , which is the nominal resistance under zero-force conditions. (2) A subject is seated over the chair: R x suddenly decreases (e.g., from R x,0 to R x,1 ) due to the subject's weight. Then, as the subject stays there, his/her respiration causes small-signal variations of R x (∆R x in Figure 1a) that enables us to confirm the presence of a subject. (3) An inanimate object is placed over the chair: R x undergoes a sudden decrease due to the object's weight, but remains constant at R x,2 , as shown in Figure 1b. Therefore, according to Figure 1, if the measurement system is able to detect such small-signal variations caused by respiration, the subject-object classification can be carried out using a single resistive mechanical sensor. nominal resistance under zero-force conditions. (2) A subject is seated over the chair: Rx suddenly decreases (e.g., from Rx,0 to Rx,1) due to the subject's weight. Then, as the subject stays there, his/her respiration causes small-signal variations of Rx (∆Rx in Figure 1a) that enables us to confirm the presence of a subject. (3) An inanimate object is placed over the chair: Rx undergoes a sudden decrease due to the object's weight, but remains constant at Rx,2, as shown in Figure 1b. Therefore, according to Figure 1, if the measurement system is able to detect such small-signal variations caused by respiration, the subject-object classification can be carried out using a single resistive mechanical sensor.

Read-out Interface Circuit
The FSR variations shown in Figure 1 are proposed to be measured by a low-cost generalpurpose MCU applying the DIC concept [12]. The MCU only needs to have an embedded digital timer and a few input/output (I/O) digital port pins; no analogue (e.g., comparator) or mixed (e.g., ADC) embedded peripherals are required. Next, two proposals for the MCU-based circuit are presented. The first one is a basic topology that will enable us to prove the measurement method introduced in Section 2, whereas the second one is an energy-efficient topology with a smart wakeup.

Basic Read-out Circuit
The basic topology of the DIC to read the FSR is shown in Figure 2a. First of all, note that the FSR has a resistor (Rp) in parallel so as to linearize its response and also to avoid long measurements in zero-force conditions. The operating principle of this circuit is as follows. Initially, Pin 1 provides a digital '1' and Pin 2 is in high-impedance (HZ) state and, consequently, the well-known capacitor C is quickly charged to the supply voltage (VDD). Next, Pin 1 is in HZ and Pin 2 provides a digital '0' so that C is discharged towards ground through the equivalent resistance (Req = Rp‖Rx). In the meantime, an embedded timer (which uses a high-frequency clock signal as a reference) measures the time interval required to do the discharge. When the exponential discharging voltage crosses the threshold voltage (VT) of the digital buffer embedded into Pin 1, the timer is stopped, as shown in Figure 2b. The resulting digital number stored in the timer is proportional to the discharging time (Td) and also to Req, since Td = Req·C·ln(VDD/VT) [12]; this is valid provided that Req is much higher than the parasitic output resistance of the I/O pins involved in the measurement. For a given Td and assuming that C, VDD and VT are known, Req can be estimated as Unlike the circuits proposed in [12,13], the circuit in Figure 2a does not require any reference resistor because the information to be monitored is on the change, not in the absolute value.

Read-out Interface Circuit
The FSR variations shown in Figure 1 are proposed to be measured by a low-cost general-purpose MCU applying the DIC concept [12]. The MCU only needs to have an embedded digital timer and a few input/output (I/O) digital port pins; no analogue (e.g., comparator) or mixed (e.g., ADC) embedded peripherals are required. Next, two proposals for the MCU-based circuit are presented. The first one is a basic topology that will enable us to prove the measurement method introduced in Section 2, whereas the second one is an energy-efficient topology with a smart wake-up.

Basic Read-out Circuit
The basic topology of the DIC to read the FSR is shown in Figure 2a. First of all, note that the FSR has a resistor (R p ) in parallel so as to linearize its response and also to avoid long measurements in zero-force conditions. The operating principle of this circuit is as follows. Initially, Pin 1 provides a digital '1' and Pin 2 is in high-impedance (HZ) state and, consequently, the well-known capacitor C is quickly charged to the supply voltage (V DD ). Next, Pin 1 is in HZ and Pin 2 provides a digital '0' so that C is discharged towards ground through the equivalent resistance (R eq = R p R x ). In the meantime, an embedded timer (which uses a high-frequency clock signal as a reference) measures the time interval required to do the discharge. When the exponential discharging voltage crosses the threshold voltage (V T ) of the digital buffer embedded into Pin 1, the timer is stopped, as shown in Figure 2b. The resulting digital number stored in the timer is proportional to the discharging time (T d ) and also to R eq , since T d = R eq ·C·ln(V DD /V T ) [12]; this is valid provided that R eq is much higher than the parasitic output resistance of the I/O pins involved in the measurement. For a given T d and assuming that C, V DD and V T are known, R eq can be estimated as Unlike the circuits proposed in [12,13], the circuit in Figure 2a does not require any reference resistor because the information to be monitored is on the change, not in the absolute value.

Read-out Circuit with a Smart Wake-up
The main limitation of the circuit shown in Figure 2a is that the presence of a subject/object over the chair is detected by polling, thus involving unnecessary energy consumption especially when the seat is vacant. This can be improved using the DIC-based topology shown in Figure 3, which has a smart wake-up generated by the FSR itself when a sudden decrease of resistance occurs. In other words: the MCU is in a deep sleep mode by default, but when a sudden decrease of resistance happens, the MCU wakes up and measures the FSR in the same way indicated in Section 3.1. This kind of smart wake-up in MCU-based circuits was also proposed in [5,25], but employing a sensor for the wake-up and another for the measurement of interest. Here, the FSR is employed for both purposes.
The operating principle of the circuit shown in Figure 3 is the following. Initially, the MCU is in a deep sleep mode and the circuit is configured as a voltage divider using a series resistance (Rs). The pins are configured as: Pin 3 provides a digital '1', Pin 2 is set as external interruption with falling edge attention, Pin 1 provides a digital '0', and Pin 4 is in HZ. Note that the current consumption of this voltage divider is very low since the FSR offers a very high resistance in zero-force conditions. When a subject/object is seated/placed over the FSR, its resistance suddenly decreases and, consequently, the output (Vint) of the voltage divider also decreases, thus generating an external interruption attended by Pin 2. In such a case, the MCU wakes up and then measures the FSR as described in Section 3.1. In comparison with the circuit shown in Figure 2a, there are two differences: (1) Pin 3 is set in HZ and Pin 4 provides a digital '0' during the charge-discharge process shown in Figure 2b, and (2) Rp is not required here since the sensor resistance is quite low when a subject/object is seated/placed over the chair and, hence, the discharging time is not expected to be so long. Therefore, in such a topology, we have Req = Rx during the discharge process.
Once the MCU is awake, the measurement of Td is repeated n times with a sampling frequency of fs during an acquisition time of Tacq (= n/fs) that should be long enough to appropriately monitor the respiratory signal. The main energy consumption occurs when charging C to VDD and when measuring Td with a timer running at high frequency so as to reduce the effects of quantization. Accordingly, the energy required to carry out the complete measurement can be approximated as where Itimer is the current consumption of the timer in charge of measuring Td. This is valid provided that the current consumption of the MCU while controlling both the charging stage and the sampling frequency is negligible. Note that these two tasks can easily be controlled by a timer running at low frequency (e.g., 32 kHz) [26] that generally involves a current consumption smaller than 1 µA. The consumption related to the data processing is not considered in Equation (2).

Read-out Circuit with a Smart Wake-up
The main limitation of the circuit shown in Figure 2a is that the presence of a subject/object over the chair is detected by polling, thus involving unnecessary energy consumption especially when the seat is vacant. This can be improved using the DIC-based topology shown in Figure 3, which has a smart wake-up generated by the FSR itself when a sudden decrease of resistance occurs. In other words: the MCU is in a deep sleep mode by default, but when a sudden decrease of resistance happens, the MCU wakes up and measures the FSR in the same way indicated in Section 3.1. This kind of smart wake-up in MCU-based circuits was also proposed in [5,25], but employing a sensor for the wake-up and another for the measurement of interest. Here, the FSR is employed for both purposes.  An algorithm to be executed by the MCU for the subject-object classification is proposed in Figure 4. When the MCU wakes up from the deep sleep mode, it measures Td up to n times and the result is stored in a 16-bit variable Ni. From the set of measurements, the maximum (Nmax) and minimum (Nmin) values are found; a low-pass filter processing of the data could be required before determine Nmax and Nmin so as to avoid the effects of aberrant measurements. Afterwards, the difference (ΔN) between Nmax and Nmin is calculated. If such a difference is higher than a given threshold level, the presence of a subject is confirmed. As explained later in Section 5.2, the circuit in Figure 3 is expected to operate at fs = 2 Sa/s during Tacq = 10 s and, therefore, 20 samples of Td (of 16 bits) will be stored. This involves 40 bytes of RAM, which are available in most MCUs.  The operating principle of the circuit shown in Figure 3 is the following. Initially, the MCU is in a deep sleep mode and the circuit is configured as a voltage divider using a series resistance (R s ). The pins are configured as: Pin 3 provides a digital '1', Pin 2 is set as external interruption with falling edge attention, Pin 1 provides a digital '0', and Pin 4 is in HZ. Note that the current consumption of this voltage divider is very low since the FSR offers a very high resistance in zero-force conditions. When a subject/object is seated/placed over the FSR, its resistance suddenly decreases and, consequently, the output (V int ) of the voltage divider also decreases, thus generating an external interruption attended by Pin 2. In such a case, the MCU wakes up and then measures the FSR as described in Section 3.1. In comparison with the circuit shown in Figure 2a, there are two differences: (1) Pin 3 is set in HZ and Pin 4 provides a digital '0' during the charge-discharge process shown in Figure 2b, and (2) R p is not required here since the sensor resistance is quite low when a subject/object is seated/placed over the chair and, hence, the discharging time is not expected to be so long. Therefore, in such a topology, we have R eq = R x during the discharge process.
Once the MCU is awake, the measurement of T d is repeated n times with a sampling frequency of f s during an acquisition time of T acq (= n/f s ) that should be long enough to appropriately monitor the respiratory signal. The main energy consumption occurs when charging C to V DD and when measuring T d with a timer running at high frequency so as to reduce the effects of quantization. Accordingly, the energy required to carry out the complete measurement can be approximated as where I timer is the current consumption of the timer in charge of measuring T d . This is valid provided that the current consumption of the MCU while controlling both the charging stage and the sampling frequency is negligible. Note that these two tasks can easily be controlled by a timer running at low frequency (e.g., 32 kHz) [26] that generally involves a current consumption smaller than 1 µA. The consumption related to the data processing is not considered in Equation (2). An algorithm to be executed by the MCU for the subject-object classification is proposed in Figure 4. When the MCU wakes up from the deep sleep mode, it measures T d up to n times and the result is stored in a 16-bit variable N i . From the set of measurements, the maximum (N max ) and minimum (N min ) values are found; a low-pass filter processing of the data could be required before determine N max and N min so as to avoid the effects of aberrant measurements. Afterwards, the difference (∆N) between N max and N min is calculated. If such a difference is higher than a given threshold level, the presence of a subject is confirmed. As explained later in Section 5.2, the circuit in Figure 3 is expected to operate at f s = 2 Sa/s during T acq = 10 s and, therefore, 20 samples of T d (of 16 bits) will be stored. This involves 40 bytes of RAM, which are available in most MCUs. An algorithm to be executed by the MCU for the subject-object classification is proposed in Figure 4. When the MCU wakes up from the deep sleep mode, it measures Td up to n times and the result is stored in a 16-bit variable Ni. From the set of measurements, the maximum (Nmax) and minimum (Nmin) values are found; a low-pass filter processing of the data could be required before determine Nmax and Nmin so as to avoid the effects of aberrant measurements. Afterwards, the difference (ΔN) between Nmax and Nmin is calculated. If such a difference is higher than a given threshold level, the presence of a subject is confirmed. As explained later in Section 5.2, the circuit in Figure 3 is expected to operate at fs = 2 Sa/s during Tacq = 10 s and, therefore, 20 samples of Td (of 16 bits) will be stored. This involves 40 bytes of RAM, which are available in most MCUs.

General
A commercial FSR (FSR 406 from Interlink Electronics) was arranged in the center of the seat of a conventional chair. This FSR has R x,0 > 10 MΩ, an active area of 4 × 4 cm 2 and a rise time lower than 3 µs. The sensor was then connected to a DIC implemented by a low-cost, low-power MCU (MSP430F123 from Texas Instruments, Dallas, TX, USA) operating at 8 MHz and powered at V DD = 3.3 V. This MCU has several Low-Power Modes (LPM) to control which on-chip circuitry, such as the central processing unit (CPU) and peripherals, is active. It also has 256 bytes of RAM.
An embedded 16-bit timer (running at 8 MHz and operating in LPM3) measured T d as in Figure 2b. Several samples of T d were taken at a sampling frequency of f s , which was controlled by another embedded timer. The frequency of the respiratory signal is expected to be very low (lower than 1 Hz) and, hence, f s ≥ 2 Sa/s following the Nyquist criterion. Each sample of T d was transmitted in real time to a personal computer and, then, converted to the corresponding value of R eq applying Equation (1) through a control program implemented in LabVIEW TM . The proposed measurement system was experimentally tested with three healthy subjects of different weight and age (S1: 45 kg/14 years; S2: 64 kg/39 years; S3: 91 kg/41 years), and also with three objects of different weight (O1: 5 kg; O2: 10 kg; O3: 30 kg). The volunteers were asked to position themselves over the chair and breathe freely, but to keep quiet during the measurement so as to avoid movement artifacts. Four different sitting postures of the subject were also tested, as shown in Figure 5. The idea of placing the FSR at the backrest was not considered because the signal detected would be insignificant in certain sitting postures, such as P3 in Figure 5.
The measurements of current consumption were carried out by a digital electrometer (Keithley 6514) following the procedures indicated in [26].

General
A commercial FSR (FSR 406 from Interlink Electronics) was arranged in the center of the seat of a conventional chair. This FSR has Rx,0 > 10 MΩ, an active area of 4 × 4 cm 2 and a rise time lower than 3 µs. The sensor was then connected to a DIC implemented by a low-cost, low-power MCU (MSP430F123 from Texas Instruments, Dallas, TX, USA) operating at 8 MHz and powered at VDD = 3.3 V. This MCU has several Low-Power Modes (LPM) to control which on-chip circuitry, such as the central processing unit (CPU) and peripherals, is active. It also has 256 bytes of RAM.
An embedded 16-bit timer (running at 8 MHz and operating in LPM3) measured Td as in Figure 2b. Several samples of Td were taken at a sampling frequency of fs, which was controlled by another embedded timer. The frequency of the respiratory signal is expected to be very low (lower than 1 Hz) and, hence, fs ≥ 2 Sa/s following the Nyquist criterion. Each sample of Td was transmitted in real time to a personal computer and, then, converted to the corresponding value of Req applying Equation (1) through a control program implemented in LabVIEW TM . The proposed measurement system was experimentally tested with three healthy subjects of different weight and age (S1: 45 kg/14 years; S2: 64 kg/39 years; S3: 91 kg/41 years), and also with three objects of different weight (O1: 5 kg; O2: 10 kg; O3: 30 kg). The volunteers were asked to position themselves over the chair and breathe freely, but to keep quiet during the measurement so as to avoid movement artifacts. Four different sitting postures of the subject were also tested, as shown in Figure  5. The idea of placing the FSR at the backrest was not considered because the signal detected would be insignificant in certain sitting postures, such as P3 in Figure 5.
The measurements of current consumption were carried out by a digital electrometer (Keithley 6514) following the procedures indicated in [26].

Basic Read-out Circuit
The circuit in Figure 2a was tested using Rp = 3570 Ω, and different values of C (470 nF, 1µF and 2.2 µF) and fs (2, 20 and 60 Sa/s). The values of C were high enough to have a good resolution in the measurement of Td and, hence, of Req [27]. The selected value of Rp (together with the maximum value of C) generated Td ≤ 8 ms, thus avoiding the overflow of the timer in zero-force conditions.
The function of Pin 1 and Pin 2 in Figure 2a was implemented by pins P1.1 and P1.2, respectively, which are general-purpose I/O digital pins. The former has a Schmitt-Trigger buffer with VT = 1.2 V and is associated with a capture module that automatically captures the value of the timer when the external signal crosses VT.

Read-out Circuit with a Smart-Wake-up
In the circuit shown in Figure 3, the MCU operated by default in LPM4, where the CPU and all clocks are disabled and only the external interruption generated by Pin 2 is enabled; the wake-up time from LPM4 is 6 µs, which is fast enough for the application considered herein. Moreover, the circuit employed Rs = 1 MΩ, which is more than ten times smaller than the FSR resistance in zero-

Basic Read-out Circuit
The circuit in Figure 2a was tested using R p = 3570 Ω, and different values of C (470 nF, 1 µF and 2.2 µF) and f s (2, 20 and 60 Sa/s). The values of C were high enough to have a good resolution in the measurement of T d and, hence, of R eq [27]. The selected value of R p (together with the maximum value of C) generated T d ≤ 8 ms, thus avoiding the overflow of the timer in zero-force conditions.
The function of Pin 1 and Pin 2 in Figure 2a was implemented by pins P1.1 and P1.2, respectively, which are general-purpose I/O digital pins. The former has a Schmitt-Trigger buffer with V T = 1.2 V and is associated with a capture module that automatically captures the value of the timer when the external signal crosses V T .

Read-out Circuit with a Smart-Wake-up
In the circuit shown in Figure 3, the MCU operated by default in LPM4, where the CPU and all clocks are disabled and only the external interruption generated by Pin 2 is enabled; the wake-up time from LPM4 is 6 µs, which is fast enough for the application considered herein. Moreover, the circuit employed R s = 1 MΩ, which is more than ten times smaller than the FSR resistance in zero-force conditions, thus generating a digital '1' at the input of Pin 2 when the seat was vacant. The circuit was optimized in terms of energy consumption using the minimum values of C (470 nF) and f s (2 Sa/s) from those indicated in Section 4.2, as suggested by Equation (2).
The function of pins 1, 2, 3 and 4 in Figure 3 was implemented by pins P1.1, P1.2, P1.3 and P1.4, respectively. In addition to the features indicated in the second paragraph of Section 4.2, P1.2 was configured as an external interruption to automatically wake the MCU from LPM4 when the FSR underwent a sudden change.

Basic Read-out Circuit
The experimental results of the circuit in Figure 2a for different subjects are shown in Figure 6a. When a subject sat down (at t ≈ 10 s), a large-signal variation (between 600 and 2400 Ω) was observed, and this depended on the subject's weight. For the heaviest subject (S3), the resistance dropped to an average value of about 1250 Ω. While seated, small-signal resistance variations of around ±125 Ω with a respiratory rate between 12 (for S3) and 24 (for S1) breaths per minute, which correspond to a frequency between 0.2 and 0.4 Hz, were monitored. This well-defined respiratory signal clearly confirms the presence of a subject over the chair without incorporating any other sensor.

Basic Read-out Circuit
The experimental results of the circuit in Figure 2a for different subjects are shown in Figure 6a. When a subject sat down (at t ≈ 10 s), a large-signal variation (between 600 and 2400 Ω) was observed, and this depended on the subject's weight. For the heaviest subject (S3), the resistance dropped to an average value of about 1250 Ω. While seated, small-signal resistance variations of around ±125 Ω with a respiratory rate between 12 (for S3) and 24 (for S1) breaths per minute, which correspond to a frequency between 0.2 and 0.4 Hz, were monitored. This well-defined respiratory signal clearly confirms the presence of a subject over the chair without incorporating any other sensor.
The results when placing different objects over the chair are represented in Figure 6b, which only shows a large-signal variation due to the object's weight but not the small-signal variations. It is worth mentioning that in Figure 6b the large-signal variation is similar and even higher than that in Figure 6a although the weight is lower. This is because the objects under test had small dimensions and, consequently, their weight caused a "point" force on the FSR that was higher than the corresponding component of distributed force generated by the subject. Figures 7-9 show the small-signal resistance variations monitored for different values of fs, C and sitting posture, respectively. According to Figure 7, a sampling frequency of 2 Sa/s seems to be enough to recover the frequency of the respiratory signal, with the corresponding benefits in terms of energy consumption. From Figure 8, a low-value capacitor (470 nF) seems valid to detect the resistance variations involved in the measurement, thus reducing the energy consumption of the circuit even more. According to Figure 9, the amplitude of the small-signal resistance variation was quite similar for the four sitting postures under test. Posture P3 generated the highest large-signal variation at the instant at which the seat was occupied, whereas posture P4 provided a more unstable signal probably because of the crossed leg. The results when placing different objects over the chair are represented in Figure 6b, which only shows a large-signal variation due to the object's weight but not the small-signal variations. It is worth mentioning that in Figure 6b the large-signal variation is similar and even higher than that in Figure 6a although the weight is lower. This is because the objects under test had small dimensions and, consequently, their weight caused a "point" force on the FSR that was higher than the corresponding component of distributed force generated by the subject. Figures 7-9 show the small-signal resistance variations monitored for different values of f s , C and sitting posture, respectively. According to Figure 7, a sampling frequency of 2 Sa/s seems to be enough to recover the frequency of the respiratory signal, with the corresponding benefits in terms of energy consumption. From Figure 8, a low-value capacitor (470 nF) seems valid to detect the resistance variations involved in the measurement, thus reducing the energy consumption of the circuit even more. According to Figure 9, the amplitude of the small-signal resistance variation was quite similar for the four sitting postures under test. Posture P3 generated the highest large-signal variation at the    Figure 5). The test was carried with subject S2 at fs = 20 Sa/s and with C = 1 µF.

Read-out Circuit with a Smart Wake-up
The experimental results of the circuit in Figure 3 for different subjects and objects are shown in Figure 10a,b, respectively, assuming that t = 0 is the instant at which the subject/object is seated/placed over the chair, thus waking up the MCU. Although the values of C and fs were low so as to reduce    Figure 5). The test was carried with subject S2 at fs = 20 Sa/s and with C = 1 µF.

Read-out Circuit with a Smart Wake-up
The experimental results of the circuit in Figure 3 for different subjects and objects are shown in Figure 10a,b, respectively, assuming that t = 0 is the instant at which the subject/object is seated/placed over the chair, thus waking up the MCU. Although the values of C and fs were low so as to reduce    Figure 5). The test was carried with subject S2 at fs = 20 Sa/s and with C = 1 µF.

Read-out Circuit with a Smart Wake-up
The experimental results of the circuit in Figure 3 for different subjects and objects are shown in Figure 10a,b, respectively, assuming that t = 0 is the instant at which the subject/object is seated/placed over the chair, thus waking up the MCU. Although the values of C and fs were low so as to reduce  Figure 5). The test was carried with subject S2 at f s = 20 Sa/s and with C = 1 µF.

Read-out Circuit with a Smart Wake-up
The experimental results of the circuit in Figure 3 for different subjects and objects are shown in Figure 10a,b, respectively, assuming that t = 0 is the instant at which the subject/object is seated/placed over the chair, thus waking up the MCU. Although the values of C and f s were low so as to reduce the energy consumption, the circuit was able to recover the frequency of the respiratory signal and to detect the resistance variations caused by respiration, as clearly shown in Figure 10a. If an object was placed over the chair instead of a subject, the resulting R eq was almost constant over time after the wake-up, as shown in Figure 10b. Accordingly, the subject-object classification is also feasible with the circuit in Figure 3. An appropriate threshold level (see Figure 4) for the classification could be, for instance, 50 Ω, which corresponds to ∆N = 188 counts under the operating conditions indicated before. From Figure 10a, the acquisition time should be around 10 s in order to have enough samples (even for very low respiratory rates) that enable the detection of the respiratory signal. The effects of the sitting posture when using the circuit in Figure 3 are represented in Figure 11, which shows a performance quite similar to that represented before in Figure 9. the energy consumption, the circuit was able to recover the frequency of the respiratory signal and to detect the resistance variations caused by respiration, as clearly shown in Figure 10a. If an object was placed over the chair instead of a subject, the resulting Req was almost constant over time after the wake-up, as shown in Figure 10b. Accordingly, the subject-object classification is also feasible with the circuit in Figure 3. An appropriate threshold level (see Figure 4) for the classification could be, for instance, 50 Ω, which corresponds to ΔN = 188 counts under the operating conditions indicated before. From Figure 10a, the acquisition time should be around 10 s in order to have enough samples (even for very low respiratory rates) that enable the detection of the respiratory signal. The effects of the sitting posture when using the circuit in Figure 3 are represented in Figure 11, which shows a performance quite similar to that represented before in Figure 9. The test was carried out at fs = 2 Sa/s with C = 470 nF. Figure 11. Small-signal resistance variations monitored by the DIC in Figure 3 for different sitting postures (see Figure 5). The test was carried with subject S2 at fs = 2 Sa/s and with C = 470 nF.
The current consumption of the circuit shown in Figure 3 in LPM4 (i.e., when waiting for an interruption generated by a subject/object) was 800 nA, including the current of the voltage divider formed by Rs and FSR. On the other hand, the current consumption of the embedded timer (at 8 MHz and in LPM3) while measuring Td was 500 µA. According to Equation (2) and assuming VDD = 3.3 V, VT = 1.2 V, C = 470 nF, Itimer = 500 µA, Req = 1600 Ω, fs = 2 Sa/s, and Tacq = 10 s, the energy required is 125 µJ. Therefore, in case of using a lithium battery of 3.6 V − 1Ah, the circuit in Figure 3  the energy consumption, the circuit was able to recover the frequency of the respiratory signal and to detect the resistance variations caused by respiration, as clearly shown in Figure 10a. If an object was placed over the chair instead of a subject, the resulting Req was almost constant over time after the wake-up, as shown in Figure 10b. Accordingly, the subject-object classification is also feasible with the circuit in Figure 3. An appropriate threshold level (see Figure 4) for the classification could be, for instance, 50 Ω, which corresponds to ΔN = 188 counts under the operating conditions indicated before. From Figure 10a, the acquisition time should be around 10 s in order to have enough samples (even for very low respiratory rates) that enable the detection of the respiratory signal. The effects of the sitting posture when using the circuit in Figure 3 are represented in Figure 11, which shows a performance quite similar to that represented before in Figure 9. The test was carried out at fs = 2 Sa/s with C = 470 nF. Figure 11. Small-signal resistance variations monitored by the DIC in Figure 3 for different sitting postures (see Figure 5). The test was carried with subject S2 at fs = 2 Sa/s and with C = 470 nF.
The current consumption of the circuit shown in Figure 3 in LPM4 (i.e., when waiting for an interruption generated by a subject/object) was 800 nA, including the current of the voltage divider formed by Rs and FSR. On the other hand, the current consumption of the embedded timer (at 8 MHz and in LPM3) while measuring Td was 500 µA. According to Equation (2) and assuming VDD = 3.3 V, VT = 1.2 V, C = 470 nF, Itimer = 500 µA, Req = 1600 Ω, fs = 2 Sa/s, and Tacq = 10 s, the energy required is 125 µJ. Therefore, in case of using a lithium battery of 3.6 V − 1Ah, the circuit in Figure 3 would have autonomy to check more than 100 million times if the interruption was generated by a  Figure 11. Small-signal resistance variations monitored by the DIC in Figure 3 for different sitting postures (see Figure 5). The test was carried with subject S2 at f s = 2 Sa/s and with C = 470 nF.
The current consumption of the circuit shown in Figure 3 in LPM4 (i.e., when waiting for an interruption generated by a subject/object) was 800 nA, including the current of the voltage divider formed by R s and FSR. On the other hand, the current consumption of the embedded timer (at 8 MHz and in LPM3) while measuring T d was 500 µA. According to Equation (2) and assuming V DD = 3.3 V, V T = 1.2 V, C = 470 nF, I timer = 500 µA, R eq = 1600 Ω, f s = 2 Sa/s, and T acq = 10 s, the energy required is 125 µJ. Therefore, in case of using a lithium battery of 3.6 V − 1 Ah, the circuit in Figure 3 would have autonomy to check more than 100 million times if the interruption was generated by a subject or an object. Taking into account that the circuit in Figure 3 was clearly able to detect the resistance variations caused by respiration, the discharging-time measurement could also be carried out at a lower operating frequency (e.g., 1 MHz instead of 8 MHz), thus reducing even more the energy required since I timer would be smaller. However, the previous estimation of the autonomy would be shorter if the system had a transceiver circuit whose energy consumption can be quite significant.

Conclusions
This work has proved that detecting and confirming the presence of a subject in a chair is feasible using a single FSR directly connected to a general-purpose MCU. The proposed system first detects the subject by monitoring his/her weight and then confirms his/her presence by monitoring the respiration. The proposed MCU-based circuit has also been improved in terms of energy consumption by incorporating a smart wake-up generated by the FSR itself. In such a way, the MCU is by default in a deep sleep mode. We believe the proposed system is suitable for applications related to autonomous sensors where it is important to detect and confirm the presence of people sitting in chairs, such as intelligent airbag deployment systems and aircraft boarding systems.