METHOD FOR DATA COMPRESSION DURING FILL LEVEL MEASUREMENT

Abstract

The present disclosure relates to a method for compressing an evaluation curve, which is recorded during a radar-based fill level measurement of a filling material located in a container, and to a corresponding fill level measurement device for carrying out the method. Corresponding to the compression method according to the present disclosure, the present disclosure comprises a corresponding method for decompressing the compressed evaluation curve. The compression method is characterized in that the compression occurs using linear prediction, by corresponding estimation coefficients and an error curve being created. This makes use of the finding according to the present disclosure that evaluation curves can be compressed for diagnostic purposes, in particular in the case of FMCW-based fill level measurement, efficiently and without data loss using the model of linear prediction.

The present disclosure relates to a method for compressing an evaluation curve, which is recorded during a radar-based fill level measurement of a filling material located in a container, and to a corresponding fill level measurement device for carrying out the method. Corresponding to the compression method according to the present disclosure, the present disclosure comprises a corresponding method for decompressing the compressed evaluation curve. The compression method is characterized in that the compression occurs using linear prediction, by corresponding estimation coefficients and an error curve being created. This makes use of the finding according to the present disclosure that evaluation curves can be compressed for diagnostic purposes, in particular in the case of FMCW-based fill level measurement, efficiently and without data loss using the model of linear prediction.

Claims

1-10. (canceled)

11. A method for compressing an evaluation curve, which is recorded during a radar-based fill level measurement of a filling material located in a container, comprising the following method steps: emitting a radar signal in the direction of the filling material, receiving a receive signal after reflection of the radar signal inside the container, recording an evaluation curve on the basis of at least the receive signal, generating an approximated evaluation curve and associated estimation coefficients using a linear prediction of the evaluation curve.

12. The method according to claim 11, wherein an error curve is generated by subtracting the evaluation curve from the approximated evaluation curve.

13. The method according to claim 12, wherein the estimation coefficients are iteratively calculated using the linear prediction in such a manner that the iteration is terminated as soon as a data sum of the estimation coefficients and the error curve is minimized.

14. The method according to claim 12, wherein a compressed error curve is generated using entropy coding based on the error curve and its corresponding statistical properties.

15. The method according to claim 14, wherein a logistic distribution is used as the distribution model for the entropy coding, such that a standard deviation of the error curve is calculated as a statistical property.

16. The method according to claim 14, wherein range coding or Golomb coding is used as the data compression method for the entropy coding.

17. A radar-based fill level measurement device for performing the method according to claim 14, comprising the following components: a signal-generating unit designed to emit a radar signal in the direction of the filling material, a receiving unit designed to receive a corresponding receive signal after reflection of the radar signal inside the container, an evaluation unit designed to generate an evaluation curve on the basis of at least the receive signal, generate an approximated evaluation curve and its estimation coefficients, using at least one linear prediction of the evaluation curve, generate an error curve by subtracting the evaluation curve from the approximated evaluation curve, and determine the fill level on the basis of the evaluation curve or the estimation coefficients, and an interface using at least the estimation coefficients and the error curve can be transmitted, and/or a memory designed to store at least the estimation coefficients and the error curve.

18. The fill level measurement device according to claim 17, wherein the evaluation unit is designed to perform the linear prediction on the basis of the lattice filter model.

19. The fill level measurement device according to claim 17, wherein the signal-generating unit, the receiving unit and the evaluation unit are designed to determine the evaluation curve on the basis of the FMCW principle.

20. A method for decompressing the evaluation curve according to claim 14, comprising the following method steps: generating the decompressed error curve by using entropy decoding of the compressed error curve and of the corresponding statistical properties, restoring the approximated evaluation curve by using the estimation coefficients on the basis of the model of linear prediction, and restoring the evaluation curve using addition of the decompressed error curve and the approximated evaluation curve.

Description

[0038] The invention is explained in more detail with reference to the following figures. The following is shown:

[0039] FIG. 1: A typical arrangement of a radar-based fill level measurement device on a container,

[0040] FIG. 2: The method sequence according to the invention for compressing the evaluation curve, and

[0041] FIG. 3: The structure of an FIR filter in lattice form as a model of linear prediction.

[0042] For a basic understanding of the invention, FIG. 1 shows a typical arrangement of a freely radiating, radar-based fill level measurement device 1 on a container 2. In the container 2 is a filling material 3, whose fill level L is to be determined by the fill level measurement device 1. For this purpose, the fill level measurement device 1 is mounted on the container 2 above the maximum permissible fill level L. Depending on the field of application, the installation height h of the fill level measurement device 1 above the container bottom can be up to more than 100 m.

[0043] As a rule, the fill level measurement device 1 can be connected via an interface 11, which is based on a corresponding bus system such as “Ethernet,” “PROFIBUS,” “HART” or “Wireless HART,” to a superordinate unit 4, for example a process control system, a decentralized database or a handheld device such as a mobile radio device. On the one hand, information about the operating status of the fill level measurement device 1 can thus be communicated. However, further information with regard to the fill level L can also be transmitted via the interface 11 in order to control any inflows or outflows that may be present at the container 2.

[0044] Since the fill level measurement device 1 shown in FIG. 1 is designed as freely radiating radar, it comprises a corresponding antenna. As indicated, the antenna can be designed as a horn antenna, for example. The fill level measurement device 1 is oriented in such a manner that corresponding radar signals S.sub.HF are emitted in the direction of the filling material 3. Thereby, the radar signals S.sub.HF are generated in a signal-generating unit of the fill level measurement device 1 designed for this purpose, depending on the measuring method (pulse time-of-flight or FMCVV).

[0045] The radar signals S.sub.HF are reflected at the surface of the filling material 3 and received after a corresponding signal propagation time by the antenna or a downstream receiving unit of the fill level measurement device 1 as receive signals E.sub.HF. The fill level L can be determined from the receive signals E.sub.HF, since the signal propagation time of the radar signals S.sub.HF, E.sub.HF depends on the distance d=h−L of the fill level measurement device 1 to the filling material surface.

[0046] In order to determine the fill level L, an evaluation unit of the fill level measurement device 1 generates a digital, time-discrete evaluation curve x[n] on the basis of the receive signal E.sub.HF. When implementing the FMCW method, the evaluation unit generates the evaluation curve x[n] in principle by mixing the currently received receive signal E.sub.HF with the just transmitted radar signal S.sub.HF, wherein the radar signal S.sub.HF is transmitted continuously for this purpose with a sawtooth-shaped frequency change. Thereby, the frequency of the evaluation curve x[n] has a proportional dependence on the distance d to the filling material surface, such that the fill level L can be determined on the basis of the measured frequency. In reality, the frequency is superimposed by additional frequency components due to interfering influences. Therefore, in the case of FMCW, the evaluation curve x[n] is subjected to a Fourier transform to determine the frequency.

[0047] In the case of the pulse time-of-flight method, the evaluation curve x[n] is generated by undersampling the pulse-shaped receive signal E.sub.HF, wherein the pulse frequency of the sampling signal for this purpose differs by less than one per mill from the pulse frequency of the transmitted radar signal S.sub.HF or of the receive signal E.sub.HF.

[0048] Depending on the functionality of the fill level measurement device 1, it can occur during the measuring operation that no fill level L can be determined from the recorded evaluation curve x[n]. The cause for this can be, for example, an antenna with the build-up of residue. However, there can also be a disturbance in the container itself, for example if reflective interfering bodies are present in the container 2. Thereby, the causes of such any disturbances can be, as a rule, detected by trained specialist personnel on the basis of the evaluation curve x[n]. However, since the evaluation curve x[n] are re-recorded with clock rates of 10 Hz or higher, a prophylactic storage of all recorded evaluation curves x[n] is not possible due to limited memory resources in the fill level measurement device 1. Moreover, any transmission of the evaluation curves x[n] via the interface 11 of the fill level measurement device 1, for example to the superordinate unit 4, in uncompressed form of the evaluation curves x[n] is also only possible at least very slowly, since the interface protocols of field devices are also primarily designed to be power-optimized and correspondingly slow.

[0049] The method according to the invention, with which the evaluation curve x[n] can be compressed, is therefore illustrated in more detail with reference to FIG. 2:

[0050] According to the method according to the invention, initially the quantized evaluation curve x[n] is approximated by means of the model of linear prediction, as described, for example, in “The Theory of Linear Prediction,” P. P. Vaidyanathan, Morgan&Claypool, 2008. This results, on the one hand, in the approximated evaluation curve {circumflex over (x)}[n] and corresponding estimated coefficients K.sub.i.

[0051] The model is based on autoregression, i.e., estimation of the upcoming values {circumflex over (x)}[n] of the evaluation curve x[n] based on previous values x[n−i] of the evaluation curve x[n]:

[00001]$\hat{x}\left[n\right]=-\underset{i=1}{\overset{N}{.Math.}}{p}_{i}x\left[n-i\right]$

Here, p.sub.i are the filter coefficients for the direct shape of an FIR filter. If, instead, an FIR filter in lattice form is present (see FIG. 3), the corresponding filter coefficients are referred to here as estimation coefficients K.sub.i. Filter coefficients p.sub.i and estimation coefficients K.sub.i can be converted into one another. N is the order of the FIR filter, which is increased by 1 with each iteration. This iterative calculation makes it possible to describe the original evaluation curve x[n] on the basis of the previously described mathematical model, which is based on linear prediction, and the corresponding estimation coefficient K.sub.i. Thus, to decompress the evaluation curve x[n], in the compressed case only the estimation coefficients K.sub.i and the error curve e[n] have to be stored or transmitted.

[0052] As can be derived from the above formula, the number of generated estimation coefficients K.sub.i depends on the filter order N and thus on the number of iterations, wherein the energy of the error curve e[n] between the actual values x[n] of the evaluation curve x[n] and the estimation value {circumflex over (x)}[n]

[00002]$e\left[n\right]=x\left[n\right]-\hat{x}\left[n\right]=x\left[n\right]+\underset{i=1}{\overset{N}{.Math.}}{K}_{i}x\left[n-i\right]$

is minimized with each further iteration.

[0053] The original evaluation curve x[n] can thus be restored without loss for diagnostic purposes, for example, by determining incrementally the approximated evaluation curve {circumflex over (x)}[n] from the estimation coefficients K.sub.i and the model of linear prediction, in order to generate the evaluation curve x[n] from it by addition with the error curve e[n]. Accordingly, the diagnostic unit can also comprise, for example, an FIR filter in lattice form for restoring the evaluation curve x[n].

[0054] The estimation coefficients K.sub.i can be determined on the basis of the least squares method. The optimum estimation coefficients K.sub.i are accordingly those with which the mean square deviation F of the error curve e[n] is minimized.

[00003]$F=\underset{n=-\infty }{\overset{\infty }{.Math.}}{e}^2\left[n\right]$

[0055] As a condition for minimizing F, it is required that all partial derivatives of F after the filter coefficients p.sub.i is zero:

[00004]$\frac{\partial F}{\partial p_i}=2\underset{n=-\infty }{\overset{\infty }{.Math.}}\left(x\left[n\right]+\underset{k=1}{\overset{N}{.Math.}}{p}_k*x\left[n-k\right]\right)*x\left[n-i\right]\stackrel{\Delta }{=}0$

[0056] On the basis of the derivatives, a linear equation system can be generated, the equations of which are known as “Wiener-Hopf equations.” Suitable iterative methods can be used for the solution thereof, e.g. the so-called “Levinson-Durbin recursion” (see “Implementing the Levinson-Durbin Algorithm on the StarCore™ SC140/SC1400 Cores”, C. Margina & B: Costinescu, Freescale Semiconductor, 2005). In terms of software or hardware, this form of linear prediction can be implemented in the evaluation unit of the fill level measurement device 1 (or also in any diagnostic unit), for example by means of an FIR filter in lattice form, as shown in FIG. 3. In this case, Z.sup.−1 are corresponding delay elements.

[0057] Through this implementation of linear prediction, the approximated evaluation curve {circumflex over (x)}[n] or the estimation coefficients K.sub.i are generated iteratively such that the iteration is terminated as soon as the required memory space for the coefficients K.sub.i and the estimated memory requirement for the error curve e[n] become minimal.

[0058] In order to save further memory space, the evaluation unit of the fill level measurement device 1 can also compress the error curve e[n]. For this purpose, the method of entropy coding is especially suitable, as described, for example, in “The Data Compression Book”, M. Nelson, J.-L. Gaily, Second Edition, M & T Books, 1996. Thereby, a logistic distribution is suitable as the underlying distribution model for error curve e[n], such that the original error curve e[n] can be represented by the compressed error curve ê[n] and the standard deviation a, which is characteristic for the logistic distribution (with mean value=0). Accordingly, in this case, in addition to the compressed error curve ê[n], only the standard deviation a of the error curve e[n] has to be transmitted or stored (besides the actual, approximated evaluation curve {circumflex over (x)}[n]). As a data compression method for entropy coding, so-called “range coding” or“Golomb coding” can be implemented in the evaluation unit.

[0059] The diagnostic unit can unpack the evaluation curve x[n] accordingly in that: [0060] The decompressed error curve e[n] is generated by means of entropy decoding of the compressed error curve e[n] and the corresponding statistical properties σ, [0061] The approximated evaluation curve {circumflex over (x)}[n] is restored by means of the estimation coefficients K.sub.i on the basis of the model of linear prediction, and [0062] The evaluation curve x[n] is restored by adding the decompressed error curve ê[n] and the approximated evaluation curve {circumflex over (x)}[n].

[0063] As a result, depending on the type of evaluation curve x[n], it is possible to achieve a compression factor of up to 2, without the compression resulting in a loss of data.

LIST OF REFERENCE SIGNS

[0064] 1 Fill level measurement device [0065] 2 Container [0066] 3 Filling material [0067] 4 Superordinate unit [0068] 11 Interface [0069] d Measuring distance [0070] E.sub.HF Receive signals [0071] e[n] Error curve [0072] ê[n] Compressed error curve [0073] F Area under the squared error curve [0074] h Installation height or measurement range [0075] L Fill level [0076] K.sub.i Estimation coefficients [0077] p.sub.i Filter coefficients [0078] S.sub.HF Radar signals [0079] x[n] Evaluation curve [0080] {circumflex over (x)}[n] Approximated evaluation curve [0081] σ Standard deviation of the error curve