DE19733920A1 - Device for suppressing interference signals with frequency-dependent adaptation - Google Patents

Abstract

The device has a measurement sensor (MS) for detecting the signals and a further sensor (RS) designed so that the ratio of the signal of the first sensor to that of the second sensor differs by the ratio of the ambient interferences originating from the source. The frequency is presented as a function of time for a series of both sensor signals in a given period. This is then converted into a source signal cleansed of the noise components. An Independent claim is also given for a gradiometer containing the device.

Description

The invention relates to a device for interference channel suppression according to the preamble of claim 1. Furthermore, the invention relates to such a front directional gradiometer according to the generic term of claim 9.

It relates to a method for determining the frequency dependent weighting factors of reference sensors for Interference signal suppression, which is based on the example of the Gra Diometry in HTS-SQUID systems for measuring the smallest Magnetic fields is demonstrated. The previously used electronic balancing with a frequency independent The gradiometer factor is reduced by the new process considerably improved.

Problem

When measuring very small signals (e.g. at biomagnetic measurements in brain research and Cardiology, but also in non-destructive testing using electromagnetic methods) must be natural and interference caused by civilization can be suppressed. Natural interference fields are in the case of magnetic field measurements solution the earth's constant field of about 40-50 µT as well higher-frequency noise components due to interference in the Io  nosphere. Civilizing fields of interference can be found ver divides over a wide frequency range. In particular at the mains frequency of 50 Hz and higher harmonics, when operating railways, e.g. B. 16 2/3 Hz, PC monitors etc. disturbances up to a few ten nT, maximum up to a few hundred nT. In low frequency The area is interference from moving ferromagneti cal objects, e.g. B. doors, vehicles etc. measurable.

In this environment, signals that are <1 pT (up to 100,000,000 times smaller than the earth's magnetic field) be measured in the appropriate bandwidth. At such small signals is also the noise of the sen sensors no longer negligible and prepares others Difficulties.

State of the art

To suppress interference fields or improve the Signal-to-noise ratio are different methods known, almost always in a combination of several Methods are applied.

Signal filtering

Disturbances in the network frequency and higher harmonics can be suppressed by comb filters (certain Frequencies are suppressed). This will, however also weakened the signal to be measured, so that to the information is also lost (Falsification of the measurement!).  

Mechanical shielding against interference

When using magnetic shielding chambers that are made of several layers of a highly permeable iron alloy and Aluminum can exist, depending on the use amount of material and chamber construction damping by a factor of 50-50,000 at 1 Hz or 10,000-100,000 can be reached at 50 Hz. This particularly ef fective solution, however, has the disadvantage of being very high Magnetic shielding chamber costs.

Active compensation

With active compensation, an interference field is superimposed on the interference field by coils so that the interference is compensated for. There are two ways to do this:
On the one hand, signals that come directly from the interference sources and only need to be amplified or shifted accordingly can be used. In most cases, this method can only effectively suppress interference at a few fixed frequencies. Mastering phase changes is another obstacle. In addition, this method is only applicable to known sources of interference.

Another option is to use Reference sensors and a suitable control loop To suppress interference. The advantage is that here due to unknown disturbances. The disadvantage here is that with less sensitive Re reference sensors the measurement signal the noise of the reference sensors and the coils is superimposed.  

Modulation of the measurement signal

By modulating the measurement signal with a fixed, interference-free frequency and lock-in detection at the Modu tion frequency, the measurement will be in a very narrow Frequency band made. This will cause disruptions frequencies other than the modulation frequency are not perceived. However, this procedure can be used at bioma genetic measurements that require a wide range not be carried out because the measurement signal is not can be modulated.

Gradiometry

If the interference cannot be compensated for, a sensor for signal detection is not sufficient to obtain the undistorted measurement signal, since in this case two variables must be distinguished: measurement signal and interference. Therefore additional sensors are used to suppress interference signals. In the simplest case, this is e.g. B. an additional sensor that partially or not at all sees the measurement signal. If the disturbing influences are sufficiently far away (in relation to the distance between the two sensors), the disturbances influence both sensors in a similar way (small gradient), so that the disturbance can be approximately eliminated by a simple subtraction of the two sensor signals. Ideally, if only the lower sensor sees the measurement signal and the disturbance has no gradients, the following even applies exactly:

a = m + s b = s → m = a-b

Here, a is the sensor signal from sensor A, b is the sensor signal from sensor B, m is the signal to be measured and s is the interference signal. Since the disturbances due to field gradients do not normally have the same effect on both sensors, assuming that the gradient of the disturbance is sufficiently constant at the location of the sensors (source of disturbance further away, second-order gradient small), better interference suppression can be achieved if goes to 2nd order gradiometers. However, more sensors are required (in this case 3). In this case, the following applies approximately:

m = a-2.b + c

Basically, there is a difference between hardware gradioms tern, where the calculation of the sensor signals by a direct corresponding wiring of the sensors ge and software gradiometers, in which the ver calculation of the signals carried out by computer programs will be distinguished.

Compared to hardware gradiometers, software gradiometers can usually not be built so perfectly, since in this case the sensors are usually separate assemblies that have to be aligned by hand, so to speak. In the case of hardware gradiometers, both sensors are usually integrated on one chip, which automatically ensures exact parallelism. Because geometrical axis and angular misalignment as well as SQUIDs with somewhat different properties create irregularities that lead to incomplete suppression of the sources of interference. This can, however, be countered by modifying the weighting factors in the formula for calculating the measurement signal, so that at least the greatest disturbance is eliminated. In the case of 2 sensors:

m = ak _b .b. With the 2nd order gradiometer (3 sensors) m = ak _b .b + k _c .c.

Software radiometer with additional reference sensors

There are two ways of eliminating faults that are generated by several different sources of interference:
It is assumed that the gradient of the measurement signal is greater than the gradient of the disturbances at the location of the sensors. With the help of a Taylor development and the measured values of the sensors, which must then be oriented in very specific directions, the actual gradient of the measured signal is approximately determined. The number of sensors is determined by the order up to which the higher terms in the Taylor expansion should be suppressed (how well the gradient is calculated!). This method is also used if you are actually only interested in the spatial gradient of the signal and not in its absolute value.

One calculates: m = a-Σk _i .b _i and adjusts the factors k _i so that when the measurement signal is not active, the arithmetically resulting signal becomes minimal (least square fit!).

These two methods can of course be combined ren. None of these methods eliminates the interference exactly.

Adaptive compensation

With an adaptive compensation, the compen fits sation dynamically to the disturbances. In the above case the active compensation, for example, for a  Control loop given. In the case of compensation with help the corresponding Trans adapted to the malfunctions.

Averaging calculation

With periodic signals, the signal-to-noise Ratio can be increased by averaging the signal, where the signal-to-noise ratio by the factor √N (N = number of averages) is improved. This Lö solution has the disadvantage that for a clear verb must be averaged for a very long time and in this Time the signal should be sufficiently constant.

It is therefore an object of the invention to provide a device to create interference suppression, in which a improved interference suppression is achieved.

The object is achieved by a device for Interference signal suppression according to the entirety of the mem male according to claim 1. The object is further achieved by a gradiometer according to the entirety of the mem male according to claim 9. Further expedient or advantage adhesive embodiments or variants can be found in referring back to one of these claims Subclaims.

It was recognized that a new approach to suppressing interference was achieved in which the following requirements were met:
Determination of frequency-dependent k factors.

Use of the sensitivity of the measuring sensor in Fre frequency ranges in which no suppression of interference is required. (Noise minimization).  

No influence on the measurement signal (Information retention).

Calculation of the k factors during the measurement on the object (Optimal k factors!).

The following is still detailed:
Software radiometer with frequency-dependent weighting factors.

This factor can now be used instead of the fixed value k frequency-dependent, so that in the case of m (f) = a (f) -k (f) .b (f) the disturbances are eliminated exactly the. (You then just pull that at every frequency due to the interference!). This rech voltage is of course only accurate if the disturbance not change during the measurement and the k-factors gates were determined exactly. In addition, with this simple calculation, no signal component in the reference sensor to be available. But if you take the proportion of Meßsi gnals in the reference sensor knows, either by a before previous measurement or by calculation during the measurement solution with the help of additional reference sensors a more accurate calculation again the correct measurement signal. You would be able to see the k factors at any time to determine during the measurement, one could actually Eliminate all faults completely.

This method is of course also on gradient signals applicable as measurement signals or could be in front of the gradients calculation can be applied to all sensors, see that the gradient of the disturbance is eliminated exactly. Here, the sensors with frequency independent  gen factors are used to calculate the gradient, also at the same time for frequency-dependent elimination serve the disturbances.

A fundamental disadvantage of additional sensors is however, that the noise power of the sensors added according to the k factors and the number of sensors and so longer must be averaged to a certain Si signal-to-noise ratio.

Noise reduction

The resulting noise is minimized by three measures:
when eliminating the disturbances, only those frequencies are taken into account at which a disturbance can be demonstrated during the measurement. All other frequencies of the measuring sensor are not corrected (corresponding k factors are set to zero).

It is averaged over time until the desired one Signal-to-noise ratio is achieved.

After averaging, the noise can be specifically reduced if the noise amplitude a _r ^{2 of} the time series is known depending on the frequency. Thus, z. B.

as a new noise-reduced amplitude, where N is the number which represents averages (everything depends on frequency).  

While the last two points are well known in the literature, the first point is new (an interpolation between the values is also conceivable, so that dynamically changing sources of interference are also taken into account). Since the amount square of the noise amplitude a _r ^{2 of} the time series is also calculated at point 4, a simple comparison with the amount square of the signal amplitude a _M ² gives the accuracy R of the calculation of the averaged quantity a _M ² : R ² = a _r ² / N a _M ² . Correspondingly, the accuracy of the calculated k factors can also be determined using error propagation methods. Too imprecise k values are set to zero.

If the interference contains only a few frequencies, the first point has the following advantages:
Since the contribution to the noise of the corrected time series by the reference sensor is small and is therefore measured with the sensitivity of the measuring sensor, the noise amplitude is almost smaller by a factor of √2.

To get a signal-to-noise ratio similar to that of the conventional gradiometry with fixed k values hold is only a half as long as nope tig. Since the period of the measurement signal is usually not Is 100% known, no averaging is ideal, and a shorter measuring time has additional advantages.

The shortening of the measuring time has the pleasant consequence that the k factors within a shorter period of time change less, d. H. are more stable and therefore can be determined more easily.

A significantly bad reference sensor can also be used rer sensor serve as the measuring sensor. Because the noise  of the reference sensor only at a few frequencies comes, can still be a significant advantage over the measurement consist only with the measuring sensor.

Preservation of information

If the measuring sensor only picks up the measuring signal, the information of the measuring signal is retained by simple calculation m (f) = a (f) -k (f) .b (f), since the interference is only subtracted. If the reference sensors also see part of the measurement signal, a second k-factor k '(f) is necessary, which represents the ratio of the measurement signal in the measurement sensor to the measurement signal in the reference sensor. If this is known, an exact calculation will be provided:

This is then the correct proportion of the measurement signal in the lower sensor. If there are several reference sensors are used, one uses for this calculation those whose k and k 'factors are best known and on are smallest, since then a correspondingly smaller one Share of noise is added (this can be for ver different frequencies also different sensors be!).

Optimal k factors

A big problem in determining frequency ab pending k factors is the fact that one can order one to get sufficient accuracy of the factors about average several Fourier transformations (time window) must and not even the 50 Hz mains interference phase sync  is chronological to the individual time slots. You mediate now the complex Fourier amplitudes over the different those time slots, all frequencies are averaged have no fixed phase, zero. A simple middle So lung leads to the need to advance the phases to correct. However, this is difficult to react to lize and usually leads to the fact that not all disorders be recognized.

The new approach now consists in using only information that is independent of the absolute phase of the sensor signals in order to determine the ratio of the frequency components that occur pha-correlated in the sensors. The information about the relative phases is retained. To do this, products are formed from the complex amplitudes and then these products are averaged. Let ax be the real part and ay the imaginary part of the complex amplitude vector of the reference sensor A at any frequency f and bx, by the corresponding one of the measuring sensor B:

ax.ax, ay.ay, bx.bx, by.by ax.by, ay.bx, ax.bx, ay.by

Now be

a ² : = ax.ax + ay.ay

b ² : = bx.bx + by.by

axb: = ax.by-ay.bx

a.b: = ax.bx + ay.by

The products represent correlations between Fourier spectra and are to be regarded as averaged. These can be easily calculated from the individual Fourier transforms. a ² denotes the amount square of the complex number a. Here, the projection of the signals onto each other is just now and thus represents the common interference signal component that occurs in phase (regardless of the absolute phase of the sensors, since only the phase difference is important here!). While from represents the interfering signal component, which is just 90 degrees out of phase in both sensors. Now be:

a.b: = a.b-i a.b.

If there is no measurement signal, a ^{2 is} determined by adding the squares of the amounts of external disturbance a _St and intrinsic noise ar. The same applies to b ² . There are no mixed terms because noise is uncorrelated to all signals, so that corresponding products are averaged to zero:

a ² = a _St ² + a _r ²

b ² = b _st ² + b _r ²

where k _b = (a _st .b _st ) / b _st ²

So you have 4 unknowns. In order to clearly determine these, 4 acquaintances are necessary. Now one might believe that a ² , b ² , from time to time also supply four values. But this is not the case! Since you are dealing with complex amplitudes, the last two values are in reality only a complex value with which the phase of the complex k-factor can be determined (in a first approximation, however, the k-factor is always real!)

a.b = a.b-i a.b.

The angle of this complex number corresponds exactly to the phase shift of the disturbance between the measured time series of the two sensors. The third equation is then:

| ab | ² = a _st ² .b _st ² .

So you have to in the case of a reference sensor make additional assumption to the equation system to make clear, e.g. B. that the noise ratio of Measuring to reference sensor is known (the absolute Rau sensor also depends on the current condition magnetic field!). Then you can use the equation systems and the noise of the sensors as well the proportion of disturbances in both sensors (and thus their ratio, which is just the k factor you are looking for is) determine.

k-factor calculation with 2 reference sensors

For the unambiguous determination of all quantities, a second reference sensor is necessary for the case of no measuring signal present, which again brings with it two new unknowns (noise and interference), but supplies three new products: c ² , ca and cb In this case it is obtained the following system of equations with 6 unknowns:

a ² = a _St ² + a _r ²

b ² = b _st ² + b _r ²

c ² = c _st ² + c _r ²

| ab | ² = a _st ² .b _st ²

| ac | ² = a _st ² .c _st ²

| cb | ² = c _st ² .b _st ²

The solution to the noise is:

a _r ² = a ² - [| ab |. | ca |] / | cb |

b _r ² = b ² - [| ab |. | cb |] / | ac |

c _r ² = c ² - [| ac |. | cb |] / | ab |

If B and C are the reference sensors, the following k-factors are kept:

k _b : = (a _st .b _st ) / b _st ² = ac / bc

and

k _c : = (a _st .c _st ) / c _st ² = ab / cb

Since it is assumed in this case that the Sen sensors see no measurement signal is such a determin measurement of the k-factors only before the actual measurement possible if there is no measurement signal. Or you're stuck some sizes as known into the calculation. If the reference measurement without a measurement signal Ratios of the noise amplitudes depending on frequency is correct and one of the reference sensors is sufficient Not aware of the measurement signal, everything is on again clearly identifiable during the actual measurement, and the k factors can be calculated online. Especially the arrangement with 2 reference sensors is also suitable for pure noise measurements to determine the noise partly in all three sensors.

k-factor determination with 4 reference sensors

If measurement signals (a _M , b _M , c _M ,...) Are also present at the sensors at the same time, a correspondingly large number of unknowns are added. In order to be able to clearly determine all unknowns, 4 reference sensors are required (a total of 15 unknowns and each sensor supplies the square of the amount ( 5 ) and the products with the other sensors ( 10 )). The advantage is now that all necessary sizes are available online during the measurement. The following system of equations results (disturbance and measurement signal are also uncorrelated with each other, so that mixed products average to zero):

a ² = a _St ² + a _M ² + a _r ²

b ² = b _st ² + b _M ² + b _r ²

c ² = c _st ² + c _M ² + c _r ²

d ² = d _St ² + d _M ² + d _r ²

e ² = e _st ² + e _M ² + e _r ²

ab = a _st .b _st + a _M .b _M

ac = a _st .c _st + a _M .c _M

ad = a _st .d _st + a _M .d _M

ae = a _st .e _st + a _M .e _M

bc = b _st .c _st + b _M .c _M

bd = b _st .d _st + b _M .d _M

be = b _st .e _st + b _M .e _M

cd = c _st .d _st + c _M .d _M

ce = c _st .e _st + c _M .e _M

de = d _st .e _st + d _M .e _M

It should be noted here that this system of equations cannot differentiate between a _St and a _M. It is only known that there are two different signals that are not correlated with one another. A mathematical confusion of a _St and a _M is therefore not excluded. In order to be able to make a distinction, one calculates the corresponding k factors and then decides which size corresponds to which signal by comparing the two k factors. At this point you have to z. B. make the assumption that k _{M is} significantly larger than k _St , since the measurement signal source is usually much closer to the sensors than the interference sources.

With corresponding simplifications and estimates can also work with just one reference sensor become. In this case we have the superiority this method compared to conventional gradient ver drive on a measurement of the biomagnetic signal of the human heart in our shielding chamber demon strictly.

General procedure for k factor determination

Measurement of time series to determine the k factors with a measuring sensor and one or more reference sensors.

Conversion of the time series into time-dependent frequency representations.

Product formation (phase independent!) Of the complex ampli and averaging over time.

With the help of these products from the pre put the frequency-dependent equation systems Determine k factors and noise amplitudes of the sensors.

All k factors that have a certain uncertainty limit exceed are set to zero.

Correction of the time series to be evaluated using the calculated k factors. The time series to be evaluated should match the time series for which k-factor calculation was used. For this, the Time series in time-dependent frequency representations con vertiert, made the correction in the frequency domain and  the corrected representation back in the time series converted.

Averaging the corrected measurement signal time series over the periodic measurement signal until the desired signal Noise ratio is reached.

Subsequent noise reduction.

The invention is further based on figures and Embodiment explained in more detail. It shows:

Fig. 1 calculated frequency-dependent k factors;

Fig. 2 time series of the gradiometer signal and fre frequency-corrected gradiometer signal;

Fig. 3 Fourier transform to the signal of FIG. 2;

Fig. 4 Comparison of the Fourier transform of he inventive gradiometer ter for Magnetome (corresponding to the lower sensor)

Embodiment Possible applications

In general, the device according to the invention and the method according to the invention offer advantages when measuring the smallest signals, e.g. B .:
biomagnetic measurements with and without additional shielding.

Measurements in non-destructive material testing.

Measurements of geomagnetic fields.

Measurement of the smallest signals in acoustics or telecommunications communication.

Embodiments Determination of the k factors with only one reference sensor

If you only have one reference sensor, only 3 variables are available during the measurement to actually determine 6 variables: signal component, interference and noise in both channels. Additional assumptions are therefore necessary to obtain a clear solution. The following estimates were made:
The noise ratio w of the sensors is frequency-independent and known. (Could also be determined depending on the frequency by a special reference measurement with a 3rd sensor before the actual measurement).

The k 'factor of the measurement signal is also frequency independent and known. (Could also be beforehand be true. In our case, the measurement signal component  in the upper sensor very small, so this factor is none of great importance).

The measurement signal component in the upper sensor is essential less than the interference component.

The measurement signals are not out of phase with each other.

With these assumptions one obtains the noise amplitude a _r ²

For the k-factors of the disturbances:

This can easily be deduced from (ab) / (b ² -b _r ² ).

If k 'is significantly larger than k (reference sensor sees almost no measurement signal), the simpler formula results:

k = (ab) / (b ² -b _r ² )

In this case, there are k values that are up to 1% Ge accuracy is also calculated using the simpler formula can be, so that the corrected time series practically no differentiate from each other.  

Example measurements

In the case of only one reference sensor, corresponding measurements in a shielded environment were processed under the assumption of a known noise ratio of both sensors. There were no significant differences in the result when the noise ratio was varied between 0.5 and 2.0. In Fig. 1, the specific measurement sequence on the k-factors are to be seen, which has been adopted for frequencies up to 10 Hz by hand, the value of 50 Hz. This was done to eliminate slow dynamic disturbances that would otherwise produce a significant offset. For the remaining frequencies, only those were used whose k-value could be determined more than 3%. This is also the deviation that the 50 Hz mains disturbance shows in the course of 2 hours (in a measurement of 1 min, the k-factor can be determined from 50 Hz to 0.5%!). In a shielded environment, the detectable disturbances are at 50 Hz and harmonics (100 Hz and 150 Hz). As can be seen in Fig. 2, the absolute noise amplitude in the corrected time series is significantly lower than when using a fe most k-factor. This is because only a few frequencies have been corrected.

This is confirmed once again in the Fourier transform of both signals in Fig. 3. Fig. 4 shows for comparison the Fourier transform of the measuring sensor, in which the 50 Hz disturbance including harmonics can be clearly seen, but at the same time the lower noise component increases is observed (similar to the corrected measurement!).

Claims (10)

1. Device for recording and further processing of signals, in which a measuring sensor (MS) is provided for recording the signals,
and at least one further sensor (RS) which is arranged such that the ratio of the measurement signals of the first sensor (MS) to the measurement signals of the further sensor (RS) differs from the ratio of the ambient interference not originating from the source, and
Means for time-dependent frequency display from a time series in a defined period of time recorded signals from the sensors (MS, RS), and
Means that provides a conversion of the time-dependent frequency representation of measured signals into a source signal adjusted for the interference components. 2. Device according to claim 1, characterized by means of temporally adaptive, depending on the respective disturbance agreed compensation, especially with the help of Fourier transform, as a means of time-dependent Frequency representation. 3. Device according to claim 1 or 2,  characterized in that with tel are provided, which is a time-dependent frequency representation independent of the absolute phase of the Allow sensor signals so that a subsequent Averaging or smoothing is made possible. 4. The device according to claim 3, characterized in that white tere reference sensors are provided, for each because appropriate signal acquisitions take place. 5. Device according to one of the preceding claims, characterized in that with tel are provided that cause the frequencies of the RS signals in which there are no interference components are neglected. 6. Device according to one of the preceding claims, characterized in that the first sensor compared to the other or the further Sensors closer to the source to be examined is not. 7. Device according to one of the preceding claims, characterized in that after interference signal suppression by means of a or several reference sensors no influence or There is falsification of the measurement signal, which achieves is represented by using the time-dependent frequency position of the sensors and correlations in each of the Sensors of the measurement signal component and the interference signal  are designed to be partially distinguishable Assign a signal adjusted for the respective disturbance receive. 8. Device according to one of the preceding claims, characterized in that for Compensation of disturbances on the respective disturbances different frequencies ren (RS) are used. 9. Device according to one of the preceding claims, featured by means of compensation for the interference with Help the Fourier transform, being complex Amplitude products are formed. 10. Gradiometer as a device according to one of the preceding claims.