III. Limitations of Previous Research
3.5 Fixed performance
[38] [39] [40] [41]
Architecture R to I +TDC
Resistance Divider + CT Delta Sigma ADC
CT Delta Sigma ADC
VCO +CT Delta Sigma ADC
Bandwidth 5 MHz 5 Hz 12.5 Hz 2.5 kHz
Power 50 uA - - 95 uA
Resistance Range 0.025~3.1M Flexible 1~600 0.02~20k
Resolution(bit) 5 11 10 14.5
Fig. 101 Conceptual block diagram of triple-mode incremental ADC.
Table Ⅵ
Reported ADCs and systems for resistive sensors.
6b DAC VIN
I A D C
Fine DAC
EC Logic VRESIDUE
I A D C
SAR ADC Extended Counting
Conventional 1 Conventional 2
Incremental ADC
∫
SAR Logic
∫
VIN
SAR-EC ADC SAR Logic
∫
6b &
Fine DAC EC
Logic
Proposed ADC Triple-Mode
High-resolution (Alternate+ Pipelined)
& 3 OTA IADC (Medium-
Fine) SAR
(Coarse) EC (Fine) Alternate
SAR-EC (Proposed)
Pipelined IADC-EC (Proposed)
IADC (Medium-
Fine) SAR
(Coarse) EC (Fine)
Medium-resolution (Conventional)
& 2 OTA
IADC (Medium-
Fine) SAR
(Coarse) EC (Fine)
Low-power (Alternate)
& 1 OTA Alternate
SAR-EC (Proposed)
25
To measure gas sensors and resistance, various ADCs are used in sensor systems. Reported ADCs and systems have one fixed performance as shown in Table Ⅵ. However, recent sensor applications are required to adjust performance such as resolution depending on the circumstances for efficient sensing. To overcome this problem, ADCs capable of power optimization have been reported as shown in Fig. 25 [42]. In the case of gas sensor application, the power for the ADC is inevitably required because the gas sensing operation must be continuously performed. To improve this, a gas sensing operation was performed with a high-resolution ADC only when gas leaks and gas sensor response increases significantly. This can save the power of the sensor interface, which usually has little possibility of gas leakage. To support multi-channel sensor, a wide bandwidth incremental ADC is designed. When the ADC is operated in an interleaved method, multi-channel gas sensing is possible with only one ADC, thereby reducing the chip area of the gas sensor ROIC. This work suggested three modes of ADC to allow the user to select the appropriate mode in various situations. To increase the resolution of the ADC, a delta-sigma ADC structure is used, which causes a low bandwidth. However, the work provides high bandwidth without reducing the resolution by utilizing a pipelined structure.
Nevertheless, the ADC has several drawbacks that are not suitable for gas sensor interfaces.
The first drawback is that ADC consumes more power than other them with the same bandwidth because it has 3 integrators. When operating in high-resolution mode, three integrators operate and consume about 1.6 mW of power. This power tends to consume higher power than conventional ADCs.
Another problem is that the ADC for gas sensor application is not required to have a too high bandwidth Fig. 102 Block diagram of triple-mode reconfigurable IADC with pipelined high-resolution
operation.
𝟎. 𝟓 𝟏 − 𝒛
𝟎. 𝟐𝟓 𝟏 − 𝒛 𝟎. 𝟓 𝟏 − 𝒛
26
due to the slow response speed of gas sensor. Of course, it is true that ADC with high bandwidth is advantageous for multi-channel operation. However, the bandwidth of the ADC is too high for gas sensor operation, which only increases power consumption.
SAR ADC Δ-Σ modulator Extended counting
VIN 1
0 0.5
-VREF_SAR
+VREF_SAR
0
···
+VREF_SAR
-VREF_SAR
···
0
Fig. 103 Detail implementation of proposed reconfigurable high-resolution mode.
27
Chapter Ⅳ
Proposed Gas Sensor Interface
This chapter describes the proposed gas sensor calibration and sensor interface configuration. Each circuit constituting the ROIC will be described in detail, and the advantages compared to previous works will be described in each sub-chapter.
4.1 Proposed gas sensor calibration
The gas sensor has different reactivity depending on the concentration. In the gas sensor system, the relationship equation between the gas sensor response and concentration is embedded, and the gas concentration is inferred through this. However, the response of the gas sensor is changed by factors such as aging. Fig. 28 shows a conceptual diagram of gas sensor calibration. The left figure depicts that initial response is changed due to variables such as aging, where the offset and slope of the gas response are different. This change causes the sensor system to have different responsiveness to the same gas concentration, resulting in a difference between the inferred concentration and the actual concentration.
To overcome these problems, this work corrects the response offset and slope through Ro update and indirect response slope calibration, respectively.
Fig. 104 Conceptual diagram of gas sensor calibration.
28
4.1.1 Methodology of response self-calibration
(a) (b)
The conventional gas sensor calibration method was to expose the gas to the sensor and calibrate it through the changing resistance and current values of the sensor. Eventually, there was gas as input, and the change of gas response could be inferred through the resistance or current value corresponding to the output. This work is aimed at reducing labor and cost to directly expose the gas to the sensor, and automatically calibrating the gas sensor, which means a new input to replace the gas was required.
Various works related to aging have been reported, and Fig. 29(a) shows a study on conductivity versus aging time for different heating temperature aging [43]. As the aging progresses, conductivity of the sensor is decreased. In addition, it shows that the reduction rate of the conductivity of the sensor varies depending on the degree of aging through heating. Through this, it is possible to adjust the aging rate through heating, and it can be assumed that it is related to aging. Fig. 29(b) illustrates the maximum NO & H2 conversion versus aging temperature[44]. The degree of aging can be controlled through the heating temperature, and the conversion degree for each sensor's heating temperature can be known.
The conversion of H2 and NO is the ratio of how much it binds to oxygen and is related to the change in resistance of the sensor. Each sensor has a different conversion slope as the heating temperature increases in a specific section, which means that the change rete of the sensor resistance is also different.
In other words, if the sensor resistance value for each heater temperature is obtained through heater temperature control in the gas sensor system, the response characteristics of the sensor can be derived by calculating the resistance change rate. This means that the sensor response can be obtained indirectly by entering the heater temperature as an input without gas.
Fig. 105 Conductivity versus aging time for different temperature aging (a), maximum NO & H2 conversion versus aging temperature (b).
0 100 200 300 400 Aging time (h) 0.25
0.5 0.75 1
143°C
134°C 100°C
114°C
0 150 300 450 600 Temp (°C) 25
50 75
100
900800 600 500
Aging temp
Different slope
29
Fig. 30 shows a conceptual diagram of logarithmic resistance versus heater temperature for different gas response. For original and aged response, the logarithmic resistance at heater temperature T1 is Ro1, Ra1 and the logarithmic resistance at T2 is Ro2, Ra2. Using four resistance values, the Ro slope for the original and aged response can be obtained. The sensor's response is expressed as equation (1) by [45], and the Ro resistance is represented as equation (2) based on the work Yulong Xu et al [46].
𝑅/𝑅′ = 𝑒𝑥𝑝 𝑡/(𝜏 𝑒𝑥𝑝 𝐸
𝑘𝑇 ) (1)
𝑅 =𝑒𝑥𝑝 𝐸
𝑘𝑇
𝐴𝑋 (2)
Where, EA is activation energy, A is a constant, X = 𝑃 / , N is parameter determined by type of carrier, and PO2 is oxygen pressure. Obtained ln(R) by combining equations (1) and (2) is
𝑙𝑛(𝑅) = 𝐸
𝑘𝑇− 𝑙𝑛(𝐴𝑋) − 𝑡/(𝜏 𝑒𝑥𝑝 𝐸
𝑘𝑇 ) (3)
To get the Ro slope of the original response shown in Fig. 33, ln(Ro1) and ln(Ro2) can be obtained using Equation (3) as follows.
𝑙𝑛(𝑅𝑜1) = 𝐸
𝑘𝑇 − 𝑙𝑛(𝐴𝑋) − 𝑡/(𝜏 𝑒𝑥𝑝 𝐸
𝑘𝑇 ) (4)
𝑙𝑛(𝑅𝑜2) = 𝐸
𝑘𝑇 − 𝑙𝑛(𝐴𝑋) − 𝑡/(𝜏 𝑒𝑥𝑝 𝐸
𝑘𝑇 ) (5) Fig. 106 Conceptual diagram of logarithmic resistance versus heater
temperature for different gas response.
30 The Ro slope S1 using ln(Ro1) and ln(Ro2) is
𝑆1 =𝑅 − 𝑅
𝑇 − 𝑇 =𝐸
𝑘 1
𝑇 − 1
𝑇
𝑇 − 𝑇 = −𝐸
𝑘
(𝑇 − 𝑇 )
𝑇 𝑇 (6)
Since the original response's Ro slope is not aging, there is no change with time. However, it varies depending on the heater temperature T1 and T2.To obtain the Ro slope of the aged response, ln(Ra1) and ln(Ra2) using Equation (3) can be obtained as follows.
𝑙𝑛(𝑅 ) = 𝐸
𝑘𝑇 − 𝑙𝑛(𝐴𝑋) − 𝑡/(𝜏 𝑒𝑥𝑝 𝐸
𝑘𝑇 ) (7) 𝑙𝑛(𝑅 ) = 𝐸
𝑘𝑇 − 𝑙𝑛(𝐴𝑋) − 𝑡/(𝜏 𝑒𝑥𝑝 𝐸
𝑘𝑇 ) (8)
Ro slope S2 using ln(Ra1) and ln(Ra2) is
𝑆2 = 𝑆1 + 𝑡
(𝑇 − 𝑇 )𝜏 𝑒𝑥𝑝 𝐸 𝑘
1
𝑒𝑥𝑝 1
𝑇
− 1
𝑒𝑥𝑝 1
𝑇
(9)
This means that Ro slope of the aged response is proportional to the time.
31
4.1.2 Ro update
The gas sensor’s resistance and current are changed by exposure to the gas and goes through a recovery state that returns to its original value when the gas disappears. Ideally, it should return to the gas sensor value of the air state perfectly but, the difference will occur as shown in Fig. 31 [47]. Due to this phenomenon, the response of the gas sensor is expressed as a value other than '1', which causes an error in the gas sensor concentration. To improve the problem, this paper proposes a Ro update. If the sensor system determines that there is no gas, it updates the current Rs to RoN instead of preset Ro.
This make that the gas concentration can be kept at ‘0’ in the air state. To realize the Ro update, it is necessary to determine the presence or absence of gas, and this is possible through the proposed gas recognition algorithm.
Fig. 107 Measured gas response of CuO/Cu2O/Ag nanopattern sensor.
Fig. 108 Phase of measured H2 gas response.
32
Fig. 32 shows the phase of measured H2 gas response. When the gas sensor is exposed to H2 gas, it is divided into a varied phase in which the response changes rapidly and a steady phase in which the response becomes saturated after sufficient time. When gas injection is stopped, gas sensor recovery occurs in the varied phase. Finally, when the gas sensor response approaches 1, it can be determined as gas off. Eventually, to determine the presence or absence of gas, an algorithm that detects varied and steady phases is required.
Fig. 33 illustrates the conceptual flow chart of gas recognition algorithm. Steady phase is detected by gas pattern recognition based on edge computing. Varied phase is a section in which a sudden change in response appears. To detect this, a rough gas recognition algorithm using a moving average is proposed. If gas is not detected in both detection methods, gas recognition algorithm determines that there is no gas and the current gas resistance Rs is updated to Ro.
Fig. 109 Conceptual flow chart of gas recognition algorithm.
Fig. 110 Result of gas pattern recognition for steady phase.
33
Gas pattern recognition is mainly used to complement the selectivity of a gas sensor and detects the presence of gas even when the response reaches saturation. Fig. 34 shows that the steady phase is detected through the gas pattern recognition algorithm. When pattern recognition is made, gas response can be set, so even very low concentrations of gas can be detected. However, an error in determining the presence of a gas may occur due to the environmental effects in practical gas sensor system.
Therefore, it is important to set the appropriate response in pattern recognition.
Pattern recognition is implemented through gas, temperature, and humidity data received from the server. However, since the gas sensor platform receives data from many gas sensor modules, the server for pattern recognition is overloaded. To overcome these shortcomings, edge computing system was used as shown in Fig. 35. Edge computing system reduces server overload because each gas sensor module performs a pattern recognition operation. This edge computing starts with creating a model of pattern recognition using ANN algorithm through tensor flow. The obtained model information is converted into C language-based code through STM32 Cube AI. The generated code is loaded into the MCU and pattern recognition is performed in the gas sensor module.
Fig. 111 Flow of implementing edge computing.
34
Fig. 36 shows the result of rough gas recognition for varied phase. To detect the varied phase corresponding to the rapidly changing response, the proposed rough gas recognition utilized a moving average.
Fig. 37 illustrates the flow chart of rough gas recognition algorithm. If the difference between the moving average of the gas response and the current gas response is more than a X value, it is considered a varied phase, otherwise it is judged as a gas off state. However, due to various noises or external environmental influences on gas sensors, the judgment of gas off is jumbled and inaccurate. To prevent errors due to the transition period and unstable conditions, the moving average was used once more. In this way, gas undetected stage can only be reached in a fully saturated gas off condition from Yn = 1, which determines that there is still no gas continuously.
Fig. 112 Result of rough gas recognition for varied phase.
Fig. 113 Flow chart of rough gas recognition algorithm.
35
Fig. 38 shows the Conceptual diagram of Ro update of gas response. Due to aging or recovery error, gas sensor may reach ROC rather than ROI that is the initial Ro. The difference between the two values eventually causes an error in gas concentration, and to solve this, the current ROC is set to Ro. Then, the gas response changed to ‘1’ and the concentration is maintained at ‘0’ as shown in the right-side figure.
Fig. 114 Conceptual diagram of Ro update of gas response.
36
4.1.3 Methodology of self-calibration for FET type sensor
(a) (b)
There are several types of current type gas sensors [48]. In addition, the operation method and structure are different depending on the type. This work proposes a methodology for calibration of FET- based current type gas sensors, which are frequently reported in recent years. Since the FET type gas sensor does not require a heater, power can be greatly reduced. Fig. 39 (a) shows the structure of FET- type gas sensor. Unlike the resistance type, the input port is composed of gate, source, and drain. The voltage required for operation is applied to the drain and source, and the gate voltage is set to have a favorable interface for sensor operation. Fig. 39 (b) shows the sensor current according to the gate voltage VCG and VDS. The current varies greatly depending on the gate voltage, and it can be seen that this is a FET type-based gas sensor. If VDS is fixed, the sensor response is adjusted through the gate voltage. For that reason, the characteristics of the gate voltage can be regarded as similar to that of the heater. Instead of controlling the heater temperature in the resistance type gas sensor self-calibration, in the current type of gas sensor, Io_slope is obtained through the control of the gate voltage and is as follows.
𝐼 _𝑠𝑙𝑜𝑝𝑒 = (10) When the gate voltage VCG is applied, IRES_slope, which is the Res_slope of the current type, is obtained based on the current measurement data according to the concentration.
𝐼 _𝑠𝑙𝑜𝑝𝑒 = 𝑓(𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛) (11)
The FET type sensor calibration will be achieved by using the correlation between Io_slope and current Res_slope.
Fig. 115 The structure of FET-type gas sensor (a), and sensor current according to VDS
and VCG (b).
37
4.2 Gas ROIC interface
Fig. 40 illustrates the schematic of proposed wide dynamic range gas ROIC with self-calibration. The gas sensor stage consists of six resistance type gas sensors, a resistance type temperature and humidity sensor, and four current type FET gas sensors. The gas sensor stage is connected to the proposed gas ROIC for resistance and current measurement. Gas information obtained through the ROIC is stored in the MCU and then transmitted to the server via LTE module for digital processing and monitoring. The proposed Gas ROIC is divided into heat control and gas detection front end. The proposed heater control supplies the heater current through a 12-bit current DAC, so there is no need for an ADC and an external circuit to measure the heater current. Gas detection front-end will be composed of a wide dynamic range gas detector, CDS, ADCs, digital temperature sensor and chip-calibration mode. The wide dynamic range coarse gas detector will be operated at low-power and show 8-bit performance. High-resolution
Fig. 116 Proposed wide Dynamic range Gas ROIC with self-calibration.
1.8V VDD HIGHEN
38
gas sensing mode is provided through CDS and ADCs, and a fully digital temperature sensor is designed to compensate for changed resistance and current sources with the temperature in gas detection circuit.
In addition, a calibration mode was designed to compensate for gas sensor aging.
39
4.3 Gas response slope self-calibration
Fig 41 shows the proposed Conceptual diagram of proposed gas response slope calibration. First, offset calibration proceeds through Ro update. Then, to return the changed response slope SA to the original response, the slope calibration coefficient α is used. The slope calibration coefficient α is the ratio of the changed response slope SA and the initial response slope SI as follows.
𝑆 = α𝑆 (12) Fig. 117 Conceptual diagram of proposed gas response slope calibration.
Fig. 118 Flow of proposed gas response slope calibration.
40
Where, SA is obtained through an equation between the Ro slope and the gas response slope. To apply the slope calibration coefficient α to the response slope, this work proposed a ROIC calibration mode.
Fig. 42 illustrates the flow of proposed gas response calibration. If Rs is obtained by conventional method, the gas sensor conversion proceeds only in steps 1 and 2. Step 1 is the operation of coarse determination of the gas sensor resistance Rs. Using RDAC and SAR logic, the value between the two resistors is close to the common mode voltage VCM. Then, Rs is set similar to RDAC roughly. Step2 flows current through the two resistors using IDAC.IDAC is set so that the node voltage with the gas sensor resistor Rs is close to VCM. The difference between the generated two voltages Vref and Vs is converted into digital values through correlated double sampling (CDS) and ADC. Where, CDS plays the role of amplifier as well, which is used to complement the resolution of sensors with high resistance [37]. The gas sensor resistance Rs finally obtained is as follows.
𝑅 = 𝑉
2 ∗ 𝐺 ∗ 𝐼 + 𝑅 (13) Where, VADC is a converted voltage through ADC, and GCDS is the amplification gain of CDS. The proposed gas response slope calibration adds a ROIC calibration mode between step 1 and step 2. ROIC calibration mode converts RDAC to αRDAC through MCU after step 1. In Step 2, a current as much as IDAC/α flows, and the finally obtained gas sensor resistance Rs is as follows.
𝑅_ = 𝛼𝑅𝑠 = ∗
∗ ∗ + 𝛼 ∗ 𝑅 (14) Through calibration mode, the resistance output from ROIC is changed as much as the response slope change coefficient α.
41
4.4 Heater controller
Many heater controls for gas sensors have been reported, but each has various problems. ADCs and built-in temperature sensors were either essential or the devices had to be changed manually to control the temperature. To improve these problems, the schematic of the proposed heater control is shown in Fig. 43.
The operation sequence is as follows. First, the MCU sends an 8-bit heating code to the R-2R DAC to set the desired heater voltage. The R-2R DAC receives the heating from the MCU and generates the heater voltage. The heater voltage is changed to match the reference through a series-connected resistor and compared with the R-2R DAC output voltage through a comparator. If the heater voltage is small, the comparator outputs “0” and this signal increases the current of the 12-bit current DAC. This process continues and eventually the adjusted heater voltage and the output voltage of the R-2R DAC become equal. The 12-bit current DAC information is transferred to and stored in the MCU. Unlike other heater controllers, the proposed heater control does not require an ADC to measure power. Therefore, It has the advantage of saving power consumption and chip area. Since the heater resistance of each gas sensor is different, when the same heater voltage is applied as before, a different gas sensor’s response occurs due to a different heating temperature. However, the proposed heater control supplies the same heater power or desired one to each different heater resistance. When the same heater power is supplied, sensing reliability increases. In addition, since desired heater power can be supplied to different sensors,
R
HeatSupply
V
Heat12b Current
DAC
R
HighR
HighR-2R DAC MCU
On Chip
Fig. 119 Schematic of proposed heater control.