Method and apparatus for improving the signal-to-noise ratio of a sensor output signal

Abstract

The objective of the present invention is to improve an SN ratio of a sensor output signal obtained by periodically switching and applying, to a sensor, a reference fluid and a sample fluid. According to an aspect of the present invention, a temporal interval that can be used for measurement is divided into a plurality of temporal intervals and, for each of these temporal intervals, the periodical switching described above is performed, thereby accumulating a sensor output signal obtained for each temporal interval. The drawing illustrated along with this abstract illustrates a measurement example of an output signal waveform in a case in which a number of divisions N is changed when the entire time interval that can be used for measurement is 120 seconds. According to the true frequency components of the output signal, which are components other than noise components in the sensor output signal, the SN ratio of the signal of the accumulation result (lower half in the drawing) can be improved in comparison with a sensor output signal obtained in an individual temporal interval (upper half in the drawing).

Description

[Technical Field]

[0001] The present invention relates to a method for improving the signal-to-noise ratio (hereinafter also referred to as S / N ratio) of a sensor output signal, and more particularly to a method and apparatus for improving the S / N ratio of an output signal from a sensor when measurements are performed in which a reference fluid and a sample fluid are periodically switched between them. [Background technology]

[0002] One of the major problems in measuring minute quantities is how to suppress noise. For example, but not limited to, when attempting to realize a measurement system that functions similarly to the human sense of smell, human olfaction can detect extremely dilute gases, depending on the target substance. However, even if one attempts to achieve sensitivity equivalent to that of humans for such gases, it is often impossible to obtain a highly sensitive sensor. Therefore, further improvement in the sensor's sensitivity is required, but this is usually quite difficult. Even if a highly sensitive sensor is obtained, it is often required to detect components at even lower concentrations. Furthermore, because many real-world measurements require the detection of multiple substances, it is often insufficient to develop a sensor that is highly sensitive to a small number of specific substances. Therefore, it would be extremely beneficial to provide a method for improving the signal-to-noise ratio that can be uniformly applied to many sensors. Summary of the Invention [Problem to be solved by the invention]

[0003] An object of the present invention is to provide a method for improving the signal-to-noise ratio of a sensor output signal obtained by periodically switching between a reference fluid and a sample fluid and applying the fluid to the sensor. [Means for solving the problem]

[0004] According to one aspect of the present invention, there is provided a method for improving the signal-to-noise ratio of a sensor output signal obtained from a sensor by periodically switching between a reference fluid and a sample fluid and supplying the fluid to the sensor, the method comprising integrating a time-varying signal for each period of the sensor output signal over multiple periods. Here, the integration over the multiple periods may be performed for the entire time interval of each period or for a partial time interval that is a part of the time interval. The sensor may also have a sensitive membrane capable of adsorbing and / or absorbing and desorbing at least some of the components in the sample fluid or the reference fluid, and may detect changes in the properties of the sensitive membrane due to the adsorption and / or absorption and desorption. The sensor may also be a surface stress sensor that detects changes in surface stress induced on the sensor by changes in the properties of the sensitive film. Furthermore, the length of the switching period for periodically switching between the reference fluid and the sample fluid may be the length of each of the small sections obtained by dividing a given time section into a plurality of small sections of equal length. Furthermore, when the length of the switching period when periodically switching between the reference fluid and the sample fluid is T and the number of accumulations is N, the signal-to-noise ratio of the signal resulting from accumulating the time-varying signal for each period within a time T×N may be better than or the same as the signal-to-noise ratio of the signal resulting from accumulating the time-varying signal obtained from the sensor by periodically switching between the reference fluid and the sample fluid over a time T' longer than T for a number of periods equal to or less than T×N / T'. Furthermore, when the length of the switching period when periodically switching between the reference fluid and the sample fluid is T and the number of accumulations is N, the signal-to-noise ratio of the signal resulting from accumulating the time-varying signal for each period within time T×N may be better than or the same as the signal-to-noise ratio of the signal resulting from accumulating the time-varying signal obtained from the sensor by periodically switching between the reference fluid and the sample fluid in time T' shorter than T over a number of periods equal to or less than T×N / T'. Furthermore, if the sensor output signal reaches a peak value and then decreases while the sample fluid is being supplied to the sensor, the time for supplying the sample fluid to the sensor in each switching period in which the reference fluid and the sample fluid are periodically switched and supplied to the sensor may be greater than or equal to the time until the sensor output signal reaches a peak value. According to another aspect of the present invention, there is provided a measuring device which includes a means for supplying a reference fluid, a means for supplying a sample fluid, a sensor, a means for switching between the reference fluid and the sample fluid and supplying the fluid to the sensor, and an information processing means which receives a sensor output signal obtained from the sensor and performs a calculation, and which performs a method for improving the signal-to-noise ratio of any of the sensor output signals. [Effects of the Invention]

[0005] In the present invention, by utilizing the characteristics of the output signal of a measurement system that alternates between reference fluid and sample fluid, the ratio of noise components to signal components in the output signal can be reduced without extremely lengthening the measurement time. [Brief explanation of the drawings]

[0006] [Figure 1] FIG. 1 is a diagram showing an outline of a configuration example of a measurement device that can be used in the present invention. [Figure 2] FIG. 10 is a diagram showing an outline of another example of the configuration of a measurement device that can be used in the present invention. [Figure 3] FIG. 10 is a diagram for explaining that a noise signal is superimposed on an actual output signal. [Figure 4] This figure explains the difference in signal-to-noise ratio between measuring one cycle and measuring multiple cycles within a given time range for an output signal whose true output signal changes as a linear function of time. [Figure 5] 10A and 10B are diagrams illustrating examples of output signals in which the true output signal varies according to a step function of time. [Figure 6]This figure explains the difference in signal-to-noise ratio between measuring one cycle and measuring multiple cycles within a given measurement time range for an output signal whose true output signal changes according to a step function of time. [Figure 7] FIG. 10 is a diagram showing an example of an output signal in which the true output signal varies according to an exponential function of time. [Figure 8] FIG. 10 is a diagram showing that the level of the output signal at the time of switching between the reference fluid and the sample fluid fluctuates each time when the switching is repeated. [Figure 9] Schematic diagrams showing how the output signal asymptotically approaches a constant value over time after switching between the reference fluid and the sample fluid. (a) Schematic diagram showing how the output signal V(t) asymptotically approaches the saturated output value Vsat as the sample fluid continues to flow from this point on, when the output signal is Vo at the time of switching. (b) Schematic diagram showing how the output signal V(t) asymptotically approaches the baseline as the reference fluid continues to flow from this point on, when the output signal is Vo at the time of switching. [Figure 10] FIG. 2 is a diagram showing a schematic diagram of an output signal waveform when an introduction cycle is repeated in which a sample fluid is introduced into a sensor for a time T (sample fluid introduction phase) and then a reference fluid is introduced for a time T (reference fluid introduction phase). [Figure 11] The vertical axis of the graph shows the SNR improvement index F(N) / √N, which is the effect of improving the SNR by integrating N times, and the horizontal axis shows T / τ0 (where τ0 is the time constant of the sensor response). [Figure 12] A graph plotting the signal-to-noise ratio improvement index F(N) as a function of N, with τ / τ0 as a parameter. [Figure 13] FIG. 3 is a graph showing the results of the method of the present invention carried out using the measurement device configured as shown in FIG. 2. [Figure 14] 14 shows graphs obtained by adjusting the time axis of each graph shown in FIG. 13 so that one measurement period is approximately the same as the length of the horizontal width of the plot area of ​​the graph. [Figure 15] 10 is a graph of the SNR improvement index F(N) when the parameter τ / τ0 is set to 20, which is the value in the example. DETAILED DESCRIPTION OF THE INVENTION

[0007] By periodically switching between a reference fluid (also called a purge fluid (or purge gas in the case of a gas)) and a sample fluid and applying various analyses to the sensor output signal obtained, it is possible to identify the sample fluid, determine its composition, quantify its components, and so on. Figure 1 shows a conceptual diagram of an example configuration of an apparatus for performing such measurements. Here, the apparatus configuration for a gas is shown, but the configuration is basically similar when the fluid is a liquid. Non-Patent Document 1 briefly describes how a sensor signal that changes on the time axis is averaged 25 times. However, it does not describe in detail how the sensor signals obtained by a specific measurement sequence are averaged.

[0008] In the device shown in Figure 1, a reference fluid (reference gas) is supplied to two mass flow controllers (MFCs) from the left side of the figure. The reference gas that passes through the upper MFC is sent to the upper gas flow path at a set flow rate. Meanwhile, the reference gas that passes through the lower MFC is also sent to the lower gas flow path at a set flow rate. However, the lower gas flow path has an upstream vial containing a sample. As the reference gas passes through this vial, the sample vapor is mixed there and becomes a sample fluid (sample gas). The reference gas and sample gas sent through the two gas flow paths are mixed in the downstream vial and then sent to the sensor module containing a sensor (not shown). The two MFCs cyclically start and stop gas delivery in opposite phases to each other. That is, a first time period in which the upper MFC allows gas to flow at a set flow rate while the lower MFC stops the gas flow, and a second time period in which the upper MFC stops the gas flow while the lower MFC allows gas to flow at a set flow rate are alternately repeated. This allows the aforementioned operation of periodically switching between the reference fluid and the sample fluid and supplying them to the sensor housed in the sensor module to be realized. The output of this sensor is provided to an information processing device, which performs various necessary processing and calculations, transfers the output to a host device, and so on. The information processing device also controls the operation of various elements within the MFC and other measuring devices, controls the man-machine interface of the measuring device, transmits and receives control information to and from a host device, and performs various other operations. Furthermore, as will be apparent from the following description, the present invention does not limit the operating principle of the sensor to a specific one.Non-limiting examples of sensors that can be used include sensors that detect one or a combination of physical parameters selected from the following: surface stress, stress, force, surface tension, pressure, mass, elasticity, Young's modulus, Poisson's ratio, resonant frequency, frequency, volume, thickness, viscosity, density, magnetic force, magnetic quantity, magnetic field, magnetic flux, magnetic flux density, electrical resistance, electrical quantity, permittivity, power, electric field, charge, current, voltage, potential, mobility, electrostatic energy, capacitance, inductance, reactance, susceptance, admittance, impedance, conductance, plasmon, refractive index, luminous intensity, and temperature, among other physical parameters. The examples presented herein show the results of experiments using a membrane-type surface stress sensor (MSS), one such sensor and an invention of the present inventor. For details about MSSs, see, for example, Non-Patent Document 2 or Patent Document 1.

[0009] In the configuration of the measurement device shown in Figure 1, the sample is a volatile liquid and is contained in an upstream vial on the lower flow path. If the sample is provided as a solid containing volatile components, the solid can be directly contained in the upstream vial and a reference gas can be flowed from the upstream side to obtain sample gas. Alternatively, the solid sample can be dissolved in a solvent and contained in the upstream vial, and handled in the same way as a liquid sample. On the other hand, if the sample is provided in the form of a gas (sample gas) from the beginning, instead of the configuration shown in Figure 1, a configuration can be used in which the sample gas is supplied directly to the lower MFC and no upstream vial is used.

[0010] Furthermore, instead of providing two MFCs, it is also possible to provide a valve mechanism that selects one of two inlets connected to each of the two gas flow paths and connects it to the outlet, and a pump that controls the flow rate of the reference gas / sample gas sent to the sensor module through this valve mechanism. This pump can be installed anywhere on the gas flow path, but if it is installed downstream of the valve mechanism (between the valve mechanism and the sensor module, or on the exhaust side of the sensor module), both the reference fluid and the sample fluid can be controlled by a single pump.

[0011] The reference fluid typically contains only components that are not the target of measurement. When the fluid is a gas, measurements are often performed using air or nitrogen gas as the reference fluid (reference gas). However, there is no restriction to using such general-purpose fluids as the reference fluid; any fluid can be used as the reference fluid as long as it does not adversely affect the measurement. Furthermore, when investigating the difference between two samples with slightly different components, one sample can be used as the reference fluid and the other as the sample fluid. In this case, the sensor response reflects the difference between the two samples.

[0012] Figure 2 shows an example of the configuration of a measurement device that can use any fluid as the reference fluid. In the measurement device illustrated in Figure 2, gases are assumed as the reference fluid and the sample fluid. The lower flow path for supplying the sample gas has the same configuration as that shown in Figure 1. Meanwhile, the upper flow path for supplying the reference gas includes a vial for adding a reference sample to the reference gas supplied from the upstream side. Therefore, a new reference gas, in which the components of the reference sample are added to the reference gas (e.g., nitrogen gas or air), is supplied to the sensor module. The output signal from the sensor module reflects the difference between the new reference gas and the sample gas. Naturally, a vial is not required to generate a new reference gas. For example, if the reference sample is initially gaseous or highly vaporizable, a parallel reference sample introduction device may be used, such as injecting the reference sample gas in parallel into the reference gas flow path coming from the upstream side. Furthermore, instead of supplying the same reference gas to both the upper and lower flow paths, a reference gas containing the components of the reference sample may be supplied to the reference gas supply path from the beginning. In the configuration shown in Figure 2, particularly in the reference gas flow path, the term "reference gas" is used to refer to both the gas supplied to the inlet of the reference gas flow path and the gas exiting the reference gas flow path and supplied to the sensor module (sensor), which may lead to confusion. Therefore, in the following, in the configuration shown in Figure 2, the gas supplied to the inlet of each flow path is sometimes referred to as an "inert gas," which refers to the fact that it is typically an odorless gas, such as nitrogen or air, and is composed only of components that are not typically the target of detection. Needless to say, the reference gas and sample gas referred to here are more generally referred to as a reference fluid and a sample fluid, respectively, but will be represented by the term "gas." Similarly, in the following explanations, "gas" may be used to represent "fluid," but such explanations also naturally apply to fluids other than gases unless clearly unreasonable.

[0013] By using the above-described measurement device configuration, it is possible to provide a fluid having a specific composition (also referred to as a reference composition) as a reference fluid in order to easily detect, for example, whether or not the composition of the sample fluid to be measured (the amount of each component and / or the type of component contained therein) deviates from that specific composition, and to what extent it deviates from the reference composition, although this is not intended to be limiting.

[0014] Alternatively, a component of the sample fluid may be known to be present in a significant amount, but the component itself is not of interest for measurement (i.e., the amount of the component is not included as a required measurement parameter). In such cases, including the uninteresting component in the reference fluid can essentially cancel out the contribution of the uninteresting component in the sensor output signal pattern. Removing large-amplitude but undesired signal components from the sensor output signal is effective for optimizing the dynamic range required for the measurement signal processing system and reducing noise that can arise from large-amplitude signals. However, noise is likely to be introduced when processing the sensor output signal. Therefore, by performing the above-described cancellation in the output signal from the sensor itself, the signal-to-noise ratio of the entire measurement system can be improved. Even if the exact amount of the uninteresting component (i.e., the magnitude of the signal component in the sensor output signal due to the uninteresting component) is unknown or subject to change, the amount of the uninteresting component in the reference fluid can be appropriately adjusted to increase the dynamic range of the sensor output signal and reduce noise, as long as the change in the sensor output signal corresponding to the uninteresting component is within a relatively narrow range.

[0015] Although various other structural and operational aspects of such a measuring device are possible, these are well known to those skilled in the art and will not be described further herein.

[0016] As is well known, actual sensor output signals contain not only signals under ideal conditions (true output signals) but also random noise signals due to various factors. When the sensor sensitivity to a component in a sample fluid is sufficiently high and the component is present in sufficient concentration in the sample fluid, the true output signal is sufficiently large compared to the noise signal, allowing for highly accurate evaluation of the sensor output signal by visual inspection or various signal analysis techniques. However, in reality, the component concentration in the sample fluid is often near the detection limit of the sensor used. In such cases, a large amount of noise signal is superimposed on the true output signal, and it is not uncommon for the noise signal level to be greater than the true output signal.

[0017] It is generally known that the signal-to-noise ratio (SNR) of a signal can be improved by integrating a signal in which a periodic true signal is superimposed with a random noise signal over multiple periods. However, such integration requires a measurement time that is equal to the measurement time for a single period multiplied by the number of periods in which the integration is performed. This makes this integration method unsuitable for applications requiring short measurement times. Furthermore, even if the measurement time can be extended, this integration method assumes the constancy of the measurement target and measurement device throughout the entire measurement period, which is often not the case in practice. As is well known, in many chemical sensors, an analyte adsorbed to the sensor's sensitive site during sensor measurement desorbs after the introduction of the sample fluid is stopped, taking a time equal to or longer than the time required for adsorption. Therefore, when a reference fluid and a sample fluid are periodically switched and applied to a sensor, the introduction of the next sample fluid occurs before the desorption of the sample due to the introduction of the reference fluid is complete. As a result, the signal obtained upon introduction of the sample fluid is superimposed with a response from the sample that has not yet been desorbed. This is mentioned, for example, in Non-Patent Document 3. Furthermore, Non-Patent Document 4 theoretically analyzes the gas adsorption and desorption behavior in nanomechanical sensors, a type of sensor to which the MSS used in the present examples belongs, and the resulting sensor response. Please refer to this document as needed. Furthermore, Non-Patent Document 5 analyzes the sensor response when a reference fluid and a sample fluid are periodically switched and applied to the sensor, taking into account the viscoelastic properties of the sensor's sensitive region. Please refer to this document as needed. Therefore, when performing this type of measurement, a simple integration method of multiple measurement results, as used in spectroscopic analysis, cannot be applied. Regarding Non-Patent Document 1, due to the operating principle of the sensor used there (photoacoustic spectroscopic gas sensor), analyte adsorption on the sensor as described above does not occur or is almost negligible. Therefore, past measurement history has no substantial effect on the signal obtained from the current measurement.Therefore, Non-Patent Document 1 does not provide any motivation to consider how to address the problems associated with simple integration described above. From another perspective, for example, when a measurement device is used continuously, it is impossible to completely prevent fluctuations in the responsiveness of the sensor itself, such as its sensitivity, or in the characteristics of the electrical circuit that inputs and amplifies minute signals from the sensor. Furthermore, various parameters of the sample being measured, such as its component composition and temperature, naturally fluctuate. Furthermore, even when the amount of sample available is very small or when analyzing a sample obtained at a specific point in time from a system that may fluctuate over time, long-term measurements are not possible.

[0018] The present invention utilizes the fact that the switching period (also called the measurement period) when periodically switching between a reference fluid and a sample fluid and applying them to a sensor to perform one measurement cycle can be set freely within a wide range. In other words, the inventors of the present invention have discovered that by taking advantage of this degree of freedom in setting the switching period, instead of fixing the length of the switching period when performing integration and repeating such a fixed measurement unit multiple times, it is possible to shorten the switching period length to repeat a desired number of measurements within a given measurement period and integrate the sensor output signals from each measurement, thereby improving the S / N ratio in many cases, and have completed the present invention.

[0019] In one embodiment of the present invention, the measurement period length is determined so that measurements for a desired number of periods are completed within a predetermined measurement period, and the measurement results obtained in this manner for multiple periods are integrated. This may improve the signal-to-noise ratio, depending on the characteristics of the true output signal within the measurement period. If the time rate of change of the true output signal is relatively small, integration may actually worsen the signal-to-noise ratio. However, attempting to reduce noise using the above method is useful in cases where the true output signal is buried in noise signals and it is difficult to determine the shape of the true output signal, or where it is difficult to predict how the true output signal will change due to the difficulty of predicting the components in the sample.

[0020] In another aspect of the present invention, a noise reduction method is provided that can be applied when it is possible to estimate in advance that the functional form of the true output signal pattern is predetermined, or when it is known that the functional form of the true output signal pattern is predetermined when a noise signal is superimposed, but it is still desired to further reduce the noise signal.

[0021] Specifically, in a measurement system in which a reference fluid and a sample fluid are periodically switched and applied to a sensor to obtain an output signal from the sensor, a membrane (sensitive membrane) that selectively adsorbs and / or absorbs various components in the fluid is provided on the surface of the sensor body, and a sensor output signal corresponding to the amount of adsorption by the sensitive membrane is obtained from the sensor body. To avoid complication, the term "adsorption" will be used below to encompass the selective adsorption and / or selective absorption of various components by the sensitive membrane, as well as both actions. Here, when considering the rate at which a component in a fluid is adsorbed by the sensitive membrane, the adsorption rate at a given point in time can be considered to be determined by the difference between the component concentration near the fluid-sensitive membrane interface and the component concentration near the sensitive membrane interface. In particular, over a narrow concentration range, the adsorption rate can be considered proportional to the concentration difference, and when the change in adsorption amount is within a narrow range, the concentration of the component in the sensitive film can be considered proportional to the sensor output signal (more generally, when the sensor output signal strongly increases or decreases relative to the component concentration in the sensitive film, the change in this sensor output signal can be approximated by a linear change, so the following explanation does not lose generality to this extent). Furthermore, when the concentration of the component in the surrounding fluid decreases, the component adsorbed to the sensitive film moves into the fluid by a process opposite to adsorption, i.e., desorption. This causes the sensor output signal to change in the direction opposite to the adsorption process.

[0022] Here, since the sensitive film is usually very thin, it can be assumed that the concentration distribution in the sensitive film is uniform. Furthermore, since the fluid applied to the sensor is naturally not stationary but flows at a certain speed, and the rate of adsorption by the sensitive film can be considered not to be very high in most cases, it can be assumed that the concentration of the component in the fluid applied to the sensor is usually constant regardless of location, even in a situation where the component in the fluid is being adsorbed by the sensitive film.

[0023] If the above conditions are met, the true sensor output signal V at any time t (where t ≥ 0) is calculated as follows: T (t) is calculated by solving the first-order linear differential equation V sat (1-e^(-t / τ0)), where V sat is the output signal V after a sufficient time has elapsed. T τ0 is a constant that indicates the saturation value of (t), and is determined by the type of material, temperature, flow rate, etc. of the fluid, the type and concentration of components in the fluid, as well as the characteristics of the sensitive membrane, and is also called the time constant. In this case, if the sample fluid continues to be supplied to the sensor, the component concentration in the sensitive membrane will asymptotically approach the saturation concentration exponentially, and therefore the true sensor output signal V T It can be seen that (t) also asymptotically approaches the saturation value V0 in the same manner. In the following, the actually observed output signal (the true output signal with the noise signal n(t) superimposed) will be represented as V(t).

[0024] Below, the true sensor output signal V T Depending on the form of (t), integration may improve the S / N ratio or worsen it. Also, the conditions under which integration improves the S / N ratio, especially V T We show this condition when (t) is expressed as above: where the noise level is σ and there is no baseline drift.

[0025] First, the true output signal V T Let us consider the case where the function form of (t) is different from that explained above and is a first-order polynomial. That is, VT Assume that (t) = at (a is a constant). This is the true output signal V T This is an example of a case where the rise of V(t) is slow. The actually observed output signal V(t) is superimposed with a noise signal n(t). This is shown in Figure 3. T If (t)=at, the true output signal V T The value of can be as large as you want without saturating, so this may seem like a physically unrealistic assumption, but this is not a problem. The reason is that time t is bounded because we are only considering a point in time within one period, and therefore V T because it never diverges.

[0026] Figure 4(a) shows a single measurement cycle within a given time interval (i.e., the measurement time interval (time length 2τ)). The sample fluid is first introduced for time τ, then switched to the reference fluid and introduced for the same time τ. The signal-to-noise ratio (S / N) of this measurement is S / N = aτ / σ, since the true signal maximum and noise level within this cycle are aτ and σ, respectively. Figure 4(b) shows a measurement cycle performed over N cycles within the same measurement time interval (time length 2τ). If the output signal V(t) for these N cycles is integrated (averaged), the true signal maximum S within the resulting signal pattern (time interval t = 0 to 2τ / N) is aτ / N. In other words, as the number of cycles increases, the value of S decreases in proportion to the duration of one cycle. Furthermore, the noise level within the same time interval divided into N cycles by integration is σ / √(N). Therefore, the signal-to-noise ratio of the output signal measured and integrated over the N periods is aτ / (√(N)σ). T If the function form of (t) is a first-order polynomial and the signal pattern has a slow rise, dividing the fixed measurement time interval into multiple parts and performing measurement and integration will actually worsen the SNR compared to when such division is not performed.

[0027] Next, the true output signal V TWe will consider the case where the rise of (t) is very steep and saturates in a short time. Specifically, we will take as an example the case where the output signal V(t) is close to a step function, as shown in Figure 5. This figure shows the change in the output signal V(t) when the fluid sent to the sensor is switched from the reference fluid to the sample fluid at t = 0, but we will assume that a similar change occurs when the fluid is switched from the sample fluid to the reference fluid (the signal pattern in Figure 5 is upside down). We will also assume that other conditions are the same as in the above example.

[0028] FIG. 6(a) is a diagram illustrating the result of measuring only one period during a fixed measurement time interval 2τ, i.e., the case where the output signal in FIG. 4(a) is changed to that shown in FIG. 5. As in the case of FIG. 4(a), the maximum value of the true output signal during one period of time 2τ is the saturation value V of the true output signal. sat and the noise level is σ. Therefore, in this case, S / N=V sat / σ.

[0029] In contrast, Figure 6(b) shows a case where, like Figure 4(b), N cycles of switching between the sample fluid and the reference fluid are performed during the same measurement time interval 2τ (one cycle of switching refers to switching from the sample fluid to the reference fluid and then switching back to the sample fluid, or switching from the reference fluid to the sample fluid and then switching back to the reference fluid, completing one cycle of switching). As can be seen from Figure 6(b), unlike the case shown in Figure 4(b), here the true output signal V T Since (t) has a pattern with a shape similar to a step function and reaches its maximum value V0 immediately after switching between the reference fluid and the sample fluid, the true output signal V T The maximum value of (t) is V regardless of the number of periods (number of divisions of the fixed measurement time interval) N. sat In the end, the true output signal V during one period that has been shortened by division is TThe maximum value of (t) is the time (2τ / N) of one period resulting from this division, in other words, it is unrelated to the number of divisions N. On the other hand, the noise level due to integration is σ / √(N), as in the case discussed in Figure 4(b), so the SN ratio in this case is √(N)V sat That is, even if the length of the measurement time interval is fixed at 2τ, by dividing this interval into multiple intervals, N, switching between the sample fluid and the reference fluid in each interval, and integrating these results, the S / N ratio can be improved by a factor of √(N).

[0030] In the following, we will calculate the SN ratio when the measurement time interval of length 2τ is divided into N and measurement and integration are performed, based on a more realistic model of the adsorption of the sensitive film than the example mentioned above, and use the true output signal expressed as an exponential function obtained above, that is, V T (t)=V sat In contrast, the sensor used in Non-Patent Document 1 (photoacoustic spectroscopic gas sensor) has, due to its operating principle, an extremely small time constant for the sensor response compared to the time constant for the dynamic behavior of the gas supply system to the sensor, and therefore its output signal takes the form of a step function as shown in FIG. 5. Therefore, even if the sensor in Non-Patent Document 1 is adopted in a measurement system that periodically switches between a sample fluid and a reference fluid, there is no need to discuss the case where the response signal takes a pattern intermediate between those in FIG. 5 and FIG. 3, as will be discussed below.

[0031] In the case of the realistic adsorption model of the sensitive film described above, the true output signal V T The output signal V(t) with the noise signal superimposed on (t) has a pattern as shown in Figure 7 (only the patterns before and after switching from the reference fluid to the sample fluid are shown). Here, the pattern shown in Figure 7 is called the true output signal V T Compared with the pattern in Figure 3, where the function form of (t) is a first-order polynomial, and the pattern in Figure 5, where the function form is close to a step function, the first-order polynomial V in Figure 3 T(t) is the case where T / τ0<<1 (where T is the length of one period) for the true output signal expressed as the exponential function above, i.e., either of the following is true: (i) τ0 → ∞, i.e., the time constant is very large (slow response), or (ii) T << τ0, i.e., the measurement time for each period is very short. On the other hand, the V T (t) is the case where T / τ0>>1 for the true output signal expressed as the exponential function above, i.e., either of the following holds: (iii) τ0<< 1, i.e., the time constant is very small (fast response), or (iv) T → ∞, i.e., the measurement time for each period is very long.

[0032] From the above description, the present invention is more advantageously implemented in the following manner.

[0033] When the sample fluid is very dilute or the difference between the sample fluid and the reference fluid is small, the expected true output signal is very small, or the pattern of the true output signal is unclear from the actual measurement results: In this case, integration should be performed as many times as possible. However, since τ0 is often unknown here, in such cases, it is necessary to take measures such as setting an initial value for the time of one cycle in advance, or preparing multiple initial values ​​and switching between them to perform integration processing. Furthermore, if the S / N ratio worsens further due to integration, or if integration improves the S / N ratio to some extent but the effect is small, trial and error may be performed, such as extending the time of one cycle, or increasing the number of cycles instead of shortening the time of one cycle.

[0034] When the components that may be detected are known in advance, when the noise signal is relatively small and τ0 can be estimated with sufficient accuracy from the output signal pattern before integration processing, or when τ0 is unknown but as a result of trial and error integration using a preset time of one period, as described above, the SNR is improved to a certain extent and the true output signal pattern becomes visible, and τ0 becomes known: In this case, noise can be further reduced by determining an appropriate time for one period based on the estimated τ0 and performing multiple measurements and integrations.

[0035] Below, we will more specifically determine the conditions under which noise can be reduced by dividing the single cycle originally intended for the reference fluid-sample fluid switching measurement, or the period allocated for the switching measurement, into multiple cycles, performing measurements in each cycle, and accumulating these results.

[0036] In a system where the output signal, which represents the sensor response to a given fluid, changes exponentially, we consider the behavior of the system when the reference fluid and the sample fluid are alternately introduced at intervals of T. The output signal rises when the sample fluid is introduced for time T, and the reference fluid is introduced for time T immediately after the sample fluid introduction. At the end of this 2T introduction period, the output signal is approaching zero (the baseline) but has not yet returned to zero. Considering a cycle of [introduction of sample fluid at time T] followed by [introduction of reference fluid at time T], the output signal does not return to zero at the end of the 2T period. Therefore, the actual sensor response waveform is not consistent across periods, as shown in Figure 4(b) or 6(b), but rather fluctuates from the value of the output signal at the beginning of the period, as shown in Figure 8. Here, we assume that the sensor response, or output signal V(t), is given by the following equations when the sample fluid and the reference fluid are introduced:

number

[0037] Let us now consider a case where the initial state of the output signal is 0, and from this state, a cycle consisting of two introduction phases, namely, an introduction cycle, is repeated, in which a sample fluid is introduced for time T (sample fluid introduction phase) and then a reference fluid is introduced for time T (reference fluid introduction phase). Figure 10 shows a schematic diagram of the output signal waveform when such an introduction cycle is repeated. Here, if the first introduction phase, which is the sample fluid introduction phase, is numbered 1 and these introduction phases are numbered with sequence number n (if n is odd, it is the sample fluid introduction phase, and if n is even, it is the reference fluid introduction phase), the sensor output V in the nth introduction phase is n (t) is

number

number

[0038] V n (t) and V s,n is written in the form of a recurrence formula as above, but V s,n can be written as the following general solution:

number

number

number

number

number

[0039] Here, we consider that the measurement time interval length (also called total measurement time) 2τ, which is the length of the measurement period, which is the entire time interval allowed for measurement, is fixed, and this is divided into N (multiple) divided periods for measurement and integration. In this case, since T=τ / N, F(N) is given by

number

[0040] In the above description, we have discussed a case where the time interval of one measurement cycle is further divided. However, there are often cases where measurements are permitted for periods longer than a normal measurement cycle, and the constancy of the measurement system and the measurement target is maintained over such a long period. In such cases, a measurement cycle consisting of the divided time intervals can be repeated for a longer period, i.e., within the permitted time range, rather than repeating the measurement cycle only as many times as fits within the time interval before division, to achieve even greater noise reduction. For example, in the embodiment described below, an initial measurement cycle of 10 seconds (5 seconds for the sample fluid introduction phase and 5 seconds for the reference fluid introduction phase) is divided into 50 measurement cycles of 0.2 seconds each (0.1 seconds for the sample fluid introduction phase and 0.1 seconds for the reference fluid introduction phase). In this case, N=50. However, in the measurements performed repeatedly for this example, it was found that if the initial measurement cycle of 10 seconds was repeated 2000 times and the first 300 cycles were discarded, the constancy of the measurement system and the object being measured would be sufficient for the remaining 1700 cycles, i.e., 1700 x 10 seconds = 17000 seconds. In such a case, the number of repetitions of the divided cycles could be increased beyond the initial cycle of 10 seconds, i.e., 50 times, and could be accumulated up to 17000 seconds / 0.1 seconds = 170000 times.

[0041] Furthermore, noise reduction effects can be achieved not only by dividing one fixed period into integers (i.e., when N is an integer), but also when N includes a fractional part (i.e., when x, defined as 1 / N, is not an integer fraction). For example, in the above-mentioned embodiment of the present invention, instead of dividing the original period (10 seconds) into 50 parts, it can also be divided into a different number of parts with a fractional part, such as 50.1 parts. In this case, if division is performed within the original period, 50 divided periods will be included within this time interval, and a remainder of 10-10 / 50.1 x 50 ≒ 0.01996 seconds will be generated. This remaining time is not used in the integration, but if there is such remaining time, and it is advantageous to divide using a fractional part, i.e., a division number with a fractional part, for example, to improve the signal-to-noise ratio compared to an integer division number, it is of course possible to use a division number with a fractional part.

[0042] Even if there is a remainder within the initial cycle, by repeating the process over a longer time interval, the remainder can be eliminated or reduced to a negligible value.

[0043] Furthermore, if the time for one cycle, i.e., the time for one measurement cycle, 2τ, is made sufficiently long (for example, the maximum time that can fully ensure the constancy of the measurement system and the object being measured, or any arbitrary time within this maximum period that is allowable for one measurement based on other factors, or that is reasonable), then, as is clear from the formula for the SNR improvement index F(N) calculated above, F(N) = √Ntanh(τ / (2N τ0)), it is possible to determine the number of divisions required to obtain a good SNR, as described above, and by dividing the long time for one cycle using this number of divisions and switching between the sample fluid and reference fluid for each division, and performing the measurement and integration as described above, it is possible to obtain an integrated signal with an extremely good SNR.

[0044] Note that in the above explanation, the true output signal is expressed as a single exponential function, i.e., V T (t)=V satAlthough the case where (1-e^(-t / τ0)) is described above, the present invention can also be applied to other cases, such as when the true output signal can be expressed as the sum of multiple exponential functions with different time constants. In such cases, the above-mentioned period division, measurement, and integration can be performed by focusing on the exponential function with the largest time constant. In this case, even if the true output signal can be expressed as the sum of multiple exponential functions with different time constants, exponential functions with small coefficients can be ignored.

[0045] Furthermore, even if a true output signal component contains an exponential function with a very long time constant, it may be better to suppress the influence of such an exponential function with a long time constant on the integrated signal. In such cases, the above-mentioned influence can be suppressed by selecting the number of divisions when dividing so that the S / N ratio of the exponential function with a very long time constant worsens due to the division.

[0046] It is also known that the output signal observed during the flow of the sample fluid does not increase monotonically but overshoots, i.e., the output signal increases initially when the sample fluid starts to flow, but then begins to decrease as the sample fluid continues to flow. In such cases, theoretically, by continuing to flow the sample fluid from the start of the sample fluid flow until the point at which the output signal reaches its maximum value (maximum value point), and then switching to the reference fluid at this point, it is possible to observe without losing any information contained in the sensor response to the sample fluid. Therefore, the period division described above requires that the division period be no shorter than the time interval from the start of the sample fluid flow to the maximum value point.

[0047] However, in many cases, it is not known in advance when the true output signal, excluding noise components from the output signal, will reach its maximum output. In such cases, it is necessary to observe the output signal from the sensor and determine the time when it reaches its maximum value as the maximum value time. However, it is not possible to know that the time-varying signal has reached its maximum value until the signal begins to decrease. Moreover, since the actual output signal from the sensor contains noise, it is only possible to determine that the output signal has begun to decrease once it has decreased to a certain extent, and the maximum value time determined in this way will naturally contain an error. Therefore, it should be noted that it may be preferable to set the maximum value time slightly later than the observed and determined maximum value time.

[0048] Furthermore, rather than cases where the true output signal is completely unpredictable, the components of the sample fluid are often known to some extent in advance. For example, there may be cases where the components are known in advance based on the source of the sample fluid, but the amounts or composition ratios of multiple components are unknown. Also, the sample fluid may be selected from a finite number of components. In such cases, since the components that may be contained in the sample are known, the set of exponential time constants that appear in the output signal corresponding to those components is also known. In such cases, it is easy to determine the appropriate number of divisions described above. Even in cases where this is not the case, the number of divisions can be determined by trial and error.

[0049] In other words, the formula for the SNR improvement index F(N) calculated above, F(N)=√Ntanh(τ / (2N τ0)), includes the time constant τ0 of the sensor response, so it cannot be argued that the above formula, which includes the time constant τ0 of the sensor response, which can only be determined once the true output signal is known, is meaningless for determining the number of divisions in actual measurements.In other words, if the components in the sample can be predicted to a certain extent, the optimal or close to optimal number of divisions can be determined in advance, and even if this is not the case, it is possible to determine the number of divisions through trial and error.

[0050] In the following, experimental results are shown for a case where a membrane surface stress sensor (MSS), which is a type of surface stress sensor and was invented by the inventors of the present application, is used as the sensor, the measurement device configuration is as shown in Fig. 2, and the reference fluid supplied to the sensor is not a fluid that is often considered a reference, such as nitrogen gas or air, but a gas containing gas generated from the same type of material as the sample to be measured. However, it is clear that this does not cause a loss of generality. [Example]

[0051] The present invention will be described in more detail below based on examples of the present invention. It should be noted that the following examples are not intended to limit the present invention, but are provided to aid in understanding the present invention.

[0052] In the examples, an experiment was conducted using the configuration shown in FIG. 2 , one of the two measurement device configuration examples mentioned above. Specifically, two types of chocolate (also referred to as sample chocolate and reference chocolate, respectively) were placed in vials in the sample gas flow path and the reference gas flow path, respectively, and nitrogen gas was introduced as an inert gas from the inlets of both flow paths to measure the difference (deviation) in the odor of the sample chocolate from that of the reference chocolate. Furthermore, as described above, the sensor module used an MSS, and the output signal from the MSS was sent to an information processing device for analysis. The information processing device was also configured to control the MFC or other components (not shown) to control the gas supply and other aspects of the entire measurement device. With this configuration, an experiment on noise reduction was conducted by integrating the sensor output signal according to the present invention.

[0053] The difference in odor emitted by the two types of samples that belong to the same class, chocolate, used in the experiment is usually quite small to the human sense of smell, so the time change pattern of the output signal from the sensor module also tends to be quite small compared to the output signal when the gas supplied to the sensor module from the reference gas flow path does not contain the odor emitted from the reference chocolate (for example, when the configuration shown in Figure 1 is used). In this example, we demonstrate that even slight differences in odor between two types of samples that belong to the same class, "chocolate," can be detected as a clear pattern in the output signal from the sensor.

[0054] In the experiment of this example, the measurement time interval 2τ was set to 120 seconds. The sample gas and reference gas were supplied to the sensor for 0.1, 0.3, 1.0, 3.0, and 5.0 seconds, respectively. When the gas supply times were set in this way, the number of divisions N was 600, 200, 60, 20, and 12, respectively.

[0055] The measurement results obtained in this manner are shown in Figures 13 and 14. The graphs in these figures, from left to right, are graphs obtained when measurements were made with the supply times of each gas set to 0.1, 0.3, 1.0, 3.0, and 5.0 seconds (i.e., when the number of divisions N was 600, 200, 60, 20, and 12). In Figures 13 and 14, the upper graphs show the raw output signals without integration (specifically, the output signals from the final measurement of the measurements repeated for the corresponding number of divisions), while the lower graphs show the results of integrating the output signals the number of times corresponding to the number of divisions, as described above. Figures 13 and 14 show graphs of the same data, but while the x-axis (i.e., the time axis) is fixed in Figure 13, the time axis of each graph has been adjusted in Figure 14 so that the signal pattern of one cycle represented by one supply of sample gas and one supply of reference gas fills the entire width of the graph. In each graph in FIG. 13 and FIG. 14, the vertical axis (value of the output signal) is fixed.

[0056] Since the output signal of the sensor naturally contains noise, it is not possible to know the true output signal after completely removing the noise. Therefore, it is not possible to know the time constant τ0 of the sensor response mentioned above accurately, but from the measurement results shown in Figures 13 and 14, it is estimated that the time constant τ0 is approximately 3 seconds. This is because, from the result of N=12 where N is the smallest (or T is the longest), the saturated value V of the sensor response sat is approximately 0.05mV, and the time it takes for this value to reach 0.032mV, which is (1-e^(-1)) ≒ 0.63, is the time constant τ0 of the sensor response. This time can be estimated from Figure 13 to be approximately 3 seconds. Therefore, τ / τ0 in this measurement is 20. Figure 15 shows a graph of the SNR improvement index F(N) calculated above when τ / τ0 is this value. From this graph, we can see that F(N) increases as N is increased up to N=9, but after that, F(N) decreases with each increase in N. Since the values ​​do not change significantly near the peak of the graph and because the estimate of τ0 is rough, we can say that the optimal number of divisions for this system is approximately 10.

[0057] Considering the graphs of the results of applying the integration process of the present invention to actual output signals shown in Figure 14, in the graph on the far left, where the division number N is very large (600), the signal pattern of the integration result is sufficiently suppressed for short-period components, but it is quite different from the pattern of a normal MSS output signal, and the signal pattern is also quite different from that when the division number is smaller. This indicates that when the division number N is set to such a large number, the signal pattern of the integration result deviates significantly from the true signal pattern and is deemed inappropriate. In the three graphs on the right (with division numbers N = 60, 20, and 12), the signal pattern of the integration result resembles the pattern of a normally observed MSS output signal, and these three patterns (especially the two patterns on the right corresponding to division numbers N = 20 and 12) are quite similar. This suggests that when the division number mentioned above matches or is close to the optimal division number of 10, the signal-to-noise ratio in the integration result is improved, resulting in a signal pattern closer to the true output signal.

[0058] It should be noted that in actual measurements such as the experiments in this example, baseline drift and other factors can cause changes in the shape of the output signal in each division interval, and therefore the theoretical analysis described above may not necessarily be applicable as is. Furthermore, although the above description stated that a large division number N (i.e., a short gas supply time) would be deemed "inappropriate," it is possible to extract information different from that obtained when the division number N is small (i.e., a long gas supply time). Therefore, it should be noted that methods such as setting multiple gas supply times and flexibly adjusting the measurement conditions to extract more information are also possible. Furthermore, even if a signal appears "inappropriate" at first glance, if it can achieve its purpose, such as detecting a trace sample or distinguishing between different samples, shortening the gas supply time can significantly reduce the measurement time. [Industrial Applicability]

[0059] As described above in detail, according to the present invention, in a measurement system in which measurements are performed by alternately switching between a reference fluid and a sample fluid, the signal-to-noise ratio of the measurement results can be improved without increasing the overall time available for measurement. [Prior art documents] [Patent documents]

[0060] [Patent Document 1] International Publication WO2011 / 148774 [Non-patent literature]

[0061] [Non-Patent Document 1] J. Huber, A. Ambs, J. Woellenstein, Miniaturized Photoacoustic Carbon Dioxide Sensor with Integrated Temperature Compensation for Room Climate Monitoring, Procedia Engineering 120 (2015) 283-288.

Outdoor Tool2

Outdoor Tools3

Outdoor Tools 4

Direct Environment 5

Claims

1. 1. A method for improving the signal-to-noise ratio of a sensor output signal obtained by periodically switching between a reference fluid and a sample fluid applied to the sensor, comprising: integrating a time-varying signal for each period of the sensor output signal over a plurality of periods; A method for improving the signal-to-noise ratio of a sensor output signal, wherein the period is a time interval during which a sample fluid is switched to a reference fluid and then switched back to the sample fluid, or a time interval during which a reference fluid is switched to a sample fluid and then switched back to the reference fluid.

2. 2. The method of claim 1, wherein the integration over multiple periods is performed over an entire time interval of each period or a partial time interval that is a portion of the time interval.

3. 2. A method for improving the signal-to-noise ratio of a sensor output signal as described in claim 1, wherein the sensor has a sensitive film capable of adsorbing and / or absorbing and desorbing at least some of the components in the sample fluid or the reference fluid, and detects changes in the properties of the sensitive film due to the adsorption and / or absorption and desorption.

4. 4. The method for improving the signal-to-noise ratio of a sensor output signal according to claim 3, wherein the sensor is a surface stress sensor that detects changes in surface stress induced on the sensor by changes in the properties of the sensitive film.

5. 2. A method for improving the signal-to-noise ratio of a sensor output signal as described in claim 1, wherein the length of a switching period for periodically switching between the reference fluid and the sample fluid is the length of each sub-interval obtained by dividing a given time interval into a plurality of equal-length sub-intervals.

6. When the length of the switching cycle when the reference fluid and the sample fluid are periodically switched is T and the number of times of the integration is N, the signal-to-noise ratio of the signal resulting from integrating the time-varying signal for each period within a time period T×N is better than or equal to the signal-to-noise ratio of the signal resulting from integrating the time-varying signal obtained from the sensor by periodically switching between the reference fluid and the sample fluid for a time period T′ that is longer than T over a number of periods equal to or less than T×N / T′; 10. The method of claim 1, wherein the signal-to-noise ratio of a sensor output signal is increased.

7. When the length of the switching cycle when the reference fluid and the sample fluid are periodically switched is T and the number of times of the integration is N, the signal-to-noise ratio of the signal resulting from integrating the time-varying signal for each period within a time period T×N is better than or equal to the signal-to-noise ratio of the signal resulting from integrating the time-varying signal obtained from the sensor by periodically switching between the reference fluid and the sample fluid over a time period T′ that is shorter than T for a number of periods equal to or less than T×N / T′; 10. The method of claim 1, wherein the signal-to-noise ratio of a sensor output signal is increased.

8. If the sensor output signal peaks and then decreases while the sample fluid is being applied to the sensor, a time for supplying the sample fluid to the sensor in each switching period in which the reference fluid and the sample fluid are periodically switched and supplied to the sensor is equal to or longer than a time required for the sensor output signal to exhibit a peak value; 10. The method of claim 1, wherein the signal-to-noise ratio of a sensor output signal is increased.

9. a means for supplying a reference fluid, a means for supplying a sample fluid, a sensor, a means for switching between the reference fluid and the sample fluid and supplying the fluid to the sensor, and an information processing means for receiving a sensor output signal obtained from the sensor and performing a calculation; A measuring device that performs the method for improving the signal-to-noise ratio of a sensor output signal according to any one of claims 1 to 8.

Images