JP2629511B2 - How to remove noise from signals - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise removing method capable of removing a noise signal mixed in a signal with high accuracy and in real time using a simple device.

[0002]

2. Description of the Related Art For example, in an electric measuring device which converts physical variables such as pressure, density, humidity, flow rate, temperature and the like into electric variables and measures them, a spike-like projection signal (analog) is included in an analog signal to be measured. Hereinafter, referred to as pulse noise). Since the pulse noise causes a measurement error or the like, various methods for removing the noise pulse have been proposed. As the noise removal method, conventionally,
A smoothing method of inserting a capacitor, a resistor, or an operational amplifier into an analog electronic circuit and performing smoothing is often used.

On the other hand, in recent digital signal processing apparatuses and computer software, noise removal is performed by a simple averaging method or an exponential smoothing method that is functionally equivalent to the above-described smoothing method. Further, with the development of electronic computers capable of high-speed arithmetic processing, advanced numerical analysis methods such as a fast Fourier transform, an inverse Fourier transform, and a maximum entropy method have been performed.

[0004]

However, when the noise signal is removed by the above-mentioned smoothing method, the distortion with respect to the original waveform tends to increase. It is not practical in the field due to the large error. Further, in the case of a numerical analysis method in the frequency domain such as the fast Fourier transform, a certain amount of data is required, and it takes a long time to perform processing due to many repetitive calculations. Is unsuitable and also expensive in terms of equipment.

Accordingly, a main object of the present invention is to provide a noise removing method capable of performing processing with high accuracy and in real time with a simple device.

[0006]

The object of the present invention is to remove a noise superimposed on a measured signal waveform by first differentiating the measured signal waveform, and the absolute value of the first derivative is equal to or larger than a set threshold. In the case of, it is judged that there is noise, and the position of the noise occurrence is specified by the position of the sign change point at which the sign of the first derivative changes, and the average value of each measurement signal before and after the noise occurrence position is interpolated. The noise can be solved by calculating as a value and replacing the interpolated value with the noise signal to remove the noise.

[0007]

According to the present invention, a complicated repetitive arithmetic processing is not required, and the noise removal processing is basically performed only by subtraction, division, identification of a noise position by sign judgment, and numerical replacement. (personal computer)
However, processing can be performed in real time. Also,
As is clear from the embodiments, high accuracy can be obtained.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail based on embodiments. For example, as a result of the measurement, a signal waveform 1 as shown in FIG.
Assuming that the (original waveform) is obtained, the method of the present invention is suitably applied to remove the pulse noises 2 and 3 present in the waveform 1. The original waveform 1 is represented as a continuous line having a numerical intensity for each sampling period by AD conversion,
In the method of the present invention, the pulse noises 2 and 3 are removed by an arithmetic processing between RawWaves (names of stored memories) of the signals for each sampling period. When the signal of each cycle of the original waveform 1 is defined as Raw Wave (I) (I = 1, 2, 3,...), The primary differential value Def Wave (I + 1) of the original waveform 1
Is expressed as Def Wave (I + 1) = Raw Wave (I + 1) −Raw Wave (I) (I = 1,2...), And the continuous value is expressed as a waveform shown in FIG. Is done.

The primary differential signal value Def Wave (I + 1) for each sampling period has a sign of ±, but gradually changes to + to-or-to + for a signal without noise. It falls within a certain range of values. This range is set as a noise determination threshold, and when the absolute value of the primary differential signal value exceeds the set noise determination threshold (BPF), it is determined that there is noise, and when it is within the noise determination threshold (BPF), Since the presence of pulse noise is not recognized, no noise removal processing is performed. In the arithmetic processing, all the first derivative signal values Def Wave (I + 1) are calculated by a formula using a noise determination threshold (B
PF), and the absolute value | De |
By determining whether De | = 0 or | De | ≧ 1, it is determined whether or not the value is within the threshold. De = Def Wave (I + 1) / BPF (I = 1,2...) −−−−−−− The target of the absolute value | De | of the division value De is 1 or more, and Def Wave (I + Focusing on the sign of ± in 1), this sign is (0 → + → − → 0) or (0 → − → + → 0)
The sign change points 5 and 5 (horizontal coordinate positions at which extreme values are taken) in FIG. 2 in which the sign changes from + → − or − → +
Is specified as the position where the pulse noise 2 (3) in FIG. 1 occurs.

Next, if the pulse noise position can be specified, an interpolation value (INS.V) is obtained using signals before and after the pulse noise generation position. This interpolated value (INS.V)
Can be obtained as an arithmetic average of signals before and after the position where the pulse noise 2 (3) is generated from the following equation. Interpolated value (INS.V) = 1/2 (Raw Wave (I + 1) + Raw Wave (I-1)) --- If the interpolated value (INS.V) is obtained by the above equation,
By replacing this numerical value with the pulse noise signal, the pulse noises 2 and 3 are removed.

The case where the pulse noises 2 and 3 are present independently has been described in detail above. However, as shown in FIG. 3, continuous pulse noises 7 and 8 are included in the original signal waveform 6. If present, the method of the present invention can also be applied. As in the case of the single pulse waveform, the first derivative of the obtained measurement signal waveform 6 is performed to obtain a continuous waveform 9 of the first derivative of Def Wave (I + 1) shown in FIG. If the absolute value of exceeds the noise determination threshold (BPF), it is determined that there is noise. This determination is made based on the division value De in the above equation, as in the case of the single pulse noise. The first derivative Def Wave
The sign of ± of (I + 1) changes from (0 → + → undefined value → − → 0) or (0 → − → undefined value → + → 0).
The position changing to → + or the position changing to 0 →-
In this case, it is specified as the position where the pulse noise is generated.
If the positions of the two consecutive pulse nozzles 7 and 8 that have been determined in this way can be specified, the interpolation value (INS.V) is obtained using the signals before and after the pulse noise 7 and 8 signals.
Ask for. This interpolated value (INS.V) can be obtained as an arithmetic average of the signals before and after the pulse noises 7 and 8 by the equation. Interpolated value (INS.V) = 1/2 (Raw Wave (I + 2) + Raw Wave (I-1)) Then, this interpolated value (INS.V) is replaced with pulse noise 7 and 8 signals. Thereby, the pulse noises 7 and 8 are removed.

(Embodiment) An embodiment in which the noise signal removing method according to the present invention is applied to temperature control in an actual steelmaking furnace will be described. FIG. 5 shows a processing block diagram in this case.
There are three types of measurement items: molten steel temperature, solidification temperature, and oxygen concentration, and a thermocouple 10 and an oxygen probe 11 are provided according to each measurement item. Each sensor 1
The analog signal obtained by 0, 11 is
Are sent to the A / D converters 13, 13,... Via the digital signal converter 2, and then input to the computer A from the data input unit 14. In the computer A, about 5 points per cycle are processed as one block unit. The noise criterion calculation unit 15 fetches five points of data into the fetching memory and calculates the difference [Raw Wave (I + 1) −Raw Wave (I)] of the signal value for each adjacent cycle.
(I = 1,2...)], The first derivative Def Wave (I + 1) is obtained. Then, the primary differential signal value Def Wave (I + 1) is integer-divided by the noise determination threshold (BPF), and the divided value (De) is sent to the noise determination unit 16. If the absolute value of the division value | De | = 0, the noise determination unit 16 processes it as 0 and excludes it from the noise removal target.
When De | ≧ 1, the sign of De is stored, and different parts of the sign are extracted as shown in FIG. 6, for example, and this sign change point is specified as a noise generating unit. In the case of FIG. 6, it is determined that there is noise in Def Wave (7). In addition, about the data of 1 block of the last time,
Displayed in the table is, for example, as shown in FIG.
If the sign of ± appears at the end position of the data 6 or 10, it is because the presence or absence of noise cannot be determined without referring to the immediately preceding and succeeding one block. Processing is performed by comparing on the same table. As described above, when the noise occurrence position can be specified, an interpolation value (INS.V), which is a replacement value of the noise signal, is obtained from the equation, and this interpolation value (INS.V) is used as the signal of the noise occurrence position. Replace.

FIG. 8 is a graph showing the temperature waveform of the molten steel obtained as described above. For comparison, FIG. 9 shows a temperature waveform without removing noise. In the case of FIG. 9, a noise signal is generated intermittently, and in such a case, a subtle change between the molten steel and the slag boundary may not be able to be determined. However, noise was removed by the method of the present invention. In the case of FIG. 8, a smooth temperature waveform curve is drawn, and the temperature change point can be specified without erroneous judgment.

Further, a constant voltage is intentionally applied from the input terminal of the amplifier 12, a noise signal is simulated, and a test is performed on the correlation between the temperature of the original signal and the temperature of the signal after noise removal. Was. The result is shown in FIG. As is apparent from FIG. 10, a line at θ = 45 ° is drawn, and it is found that the signal after noise removal is obtained as a signal faithful to the original signal. Numerically, the distortion ratio between the two is 0.003% or less, indicating extremely high accuracy. The noise elimination ratio equal to or higher than the noise determination threshold (BPF) was 99.3%.

[0015]

As described above, according to the present invention, a simple electronic computer (personal computer) can be used.
Since noise can be removed with high accuracy and no complicated arithmetic processing is required, noise can be removed in real time.

[Brief description of the drawings]

FIG. 1 is an example of an original signal waveform of single pulse noise.

FIG. 2 is a first derivative waveform of the waveform of FIG.

FIG. 3 is an example of an original signal waveform of continuous pulse noise.

FIG. 4 is a first derivative waveform of the waveform of FIG. 3;

FIG. 5 is a noise processing block diagram.

FIG. 6 is a diagram illustrating a process for specifying a noise position.

FIG. 7 is a diagram illustrating a process for specifying a noise position.

FIG. 8 is a molten steel temperature waveform chart when the method of the present invention is applied.

FIG. 9 is a molten steel temperature waveform diagram when noise removal processing is not performed.

FIG. 10 is a diagram illustrating a distortion rate measurement between a signal after noise removal processing and an original signal.

[Explanation of symbols]

1: Original signal waveform, 2.3: Pulse noise, 4: Primary differential waveform

Claims (1)

(57) [Claims] To remove noise superimposed on a measured signal waveform, the measured signal waveform is first-order differentiated, and when the absolute value of the first-order differential value is equal to or greater than a set threshold value, it is determined that there is noise. The noise occurrence position is specified by the position of the sign change point at which the sign of the first derivative value changes, and the average value of each measurement signal before and after the noise occurrence position is calculated as an interpolation value. And removing the noise by replacing the noise signal with the noise signal.