DK201470196A1

DK201470196A1 - Algorithm for calculating Amplitudes, Phase- or time differences

Info

Publication number: DK201470196A1
Application number: DK201470196A
Authority: DK
Inventors: Frands Wulff Voss; Arne Stig Peters
Original assignee: Flonidan As
Priority date: 2013-10-14
Filing date: 2014-04-09
Publication date: 2015-04-20
Also published as: DK3058319T3; LT3058319T; HUE051573T2; PT3058319T; WO2015055209A1; ES2805837T3; WO2015055212A1; PL3058319T3; EP3058319A1; EP3058319B1

Description

Algorithm for calculating Amplitudes, Phase- or time differences.

Field of the Invention

The present invention relates to a system and a method for calculating the envelope, the phase- or time difference of a set of signals in a metering system. The phase- or time difference may typically he between a reference signal and a sensor signal or between two sensor signals or a transmitted and received sensor signal and the envelope of any of these signals.

Background of the Invention A non-published Danish patent application PA 2012 70241 discloses a system or a method for measuring flow in a flow duct, comprising at least two ultra sound transducers. It is the object of the pending application is to measure the flow of air in a duct by one or more transducers transmitting beams of ultra sound controlled by a microcontroller based electronic system. The object can be achieved if the microcontroller stores a vector of data samples for each direction of transmission, which vector comprises an appropriate number of N samples forming a frame, which microcontroller multiply each value of the frame which a complex number, which microcontroller based on the result calculates the flow in the duct. By the invention according to the present patent application an efficient flow measurement of air flowing in a duct can be achie ved by calculating the phase- or time difference between the two sampled set of signals.

Object of the Invention

It is the object of the pending patent application to achieve a very fast calculation of the envelope and/or phase- or time difference between a set of signals as for example for measuring the fluid flow, gas or liquid, in a flow channel. A further object is to achieve calculation of such an envelope or phase- or time difference with extremely low power consumption of the system calculating the phase- or time difference.

Description of the Invention

The object can be nil filled by a system or a method as disclosed in the preamble to claim 1 or 2 in that a reference frequency Fr is replacing the actual frequency Ft in the calculation of the envelope and/or the phase or time difference between two signals and where the frequency Fr is related to the sample frequency Fs given by the expression Fr = Fs/c where c is a real number.

This fast and power optimized calculation can be achieved by the following sequence of steps:

a. perform condensing and summarizing of the data by use of a mathematic model in order to fit data to the following general model function g(t) = hit)* sin(co *t + φ ) +C b: calculate the phase- or time difference by Least Square Fit LSQF for achieving the phase- or time difference (DT) between the two signals.

The metering system could typically be a distance measurement system or a flow' meter such as a Coriolis Mass flow' meter or an ultrasonic flow meter but not limited to these.

When calculating envelopes and/or phase- or time differences one often uses a technique of multiplying the sampled data with a Cosine and a Sine of the actual frequency in the system.

The envelope may basically be obtained by multiplying the same set of data with a p=cos(co*t + φ) and s=sm(oo*t+<p) respectively and then take the square root (may often be omitted to speed up the calculation further) of the sum of the squares p and s. Often a filter (weight-) function is also added in the calculation of the envelope in order to make it more robust to frequency variations as well as immune to noise.

Often such calculations are quite cumbersome and thus time- and power consuming because the Cosine and Sine are fairly complicated to calculate in embedded systems and thus making it difficult to handle battery operated equipment.

There are often different ways of handling this problem and one way is if the frequency of the system is varying very slowly and therefore the variables need not be calculated every time a new measurement is made but calculated values are to be stored and stored values to be renewed when the frequency changes.

The present invention discloses a different way to overcome the problem and may be used with even quite large variations in frequency and making calculations rather simple.

The trick is to choose the sample frequency in a certain relation to the system frequency but also to use a suitable weight function in the calculations.

The basic idea is to use a reference frequency which is one quarter of the sample frequency and then replace the system frequency with this reference frequency and do ail calculations by the reference frequency. The demand is that the system frequency is within certain intervals in relation to the reference frequency. With this choice the Sine and Cosine becomes a specific sequence of zeros, ones and minus ones elimination the need of calculating the values!

In order to make the calculation more robust and immune to frequency variations but also noise it is an advantage to use a Hanning type weight function of a certain length.

Using an ultrasonic meter as a more detailed example such a metering unit is typically comprising at least a first (A) and a second (B) transducer, which method is concerns a plurality of mathematical calculations, generate and transmit a burst of n periods of ultrasound from the first transducer (A) towards the second transducer (B), and receive the burst of ultrasound at the transducer (B) defined as signal a, generate and transmit a burst of n periods of ultrasound from the second transducer (B) towards the first transducer (A), and receive the burst of ultrasound at the transducer (A), defined as signal b, c, perform analog to digital conversion of the received burst at the receiving transducer (A,B), and calculate from the sampled data a direct measurement of the time difference between the signals a and b by using a Least Square Fit method, LSQF, by fitting the sampled data to a certain model function describing the received signals and to calculate arrival times Ta and Th respectively by using a Digital CFD making use of the envelope of the received signals

The use of a mathematical model for the function is highly effective for getting rid of any noise that occurs, may be at relative high frequencies, but calculating further in the system using the model will give a phase- or time difference, DT using Least Square Fit to achieve the best fit of the measured data to a model function can lead to an expression of the phase- or time difference DT, which phase or time difference DT is representing the actual needed parameter in order to calculate for example the fluid flow in a flow duct. In that way a relative fast and effective flow calculation can be performed.

In a preferred embodiment of the invention the system performing Least Square Fit to a model function will directly calculate the phase- or time difference DT between two received signals a and b with transmission times (Ta, Tb) by fitting data to the given model function previously disclosed. Given a set of observations, one often wants to condense and summarize the data by fitting it to a "model" that depends on adjustable parameters. One has to choose or design a merit function that measures the agreement between the data and the model with a particular choice of parameters. The parameters of the model are then adjusted to achieve a minimum in the merit function, yielding best-fit parameters. The adjustment process is thus a problem in minimization in many dimensions.

Hereby it can be achieved that as soon as the physical system is designed many of the parameters will be the same for the system in operation. Many parameters are to be calculated only once and therefore, a calculation of phase- or time difference can be performed by using a number of constants which constants are depending on the different already controlled parameters.

In a further preferred embodiment of the invention can the system define same envelope function (h(t)) for both signals where ha is equal to hb. Hereby a relative simple calculation can be achieved because same envelope function is used for both of the two signals. This can be done because most of the different parameters having influence of the en velope function are equal in the two situations. The only difference which also will be expressed in the envelope function is the time delay between the transmission time in the flow direction and the transmission time in the direction against the flow direction. Therefore, it is logical that if the transducers are mostly equal it is possible to use the same envelop function without having added any critical failures to the further calculation of the flow. Actually the envelopes may differ by a scale factor without affecting the calculation.

In a further preferred embodiment the envelope function h(t) may be considered a constant.

In a further preferred embodiment of the invention can the system define a reference frequency (fr) as the sampling frequency (fs) divided by a constant ( Q ), which constant ( Q ) can be an integer. Hereby can a much more simple calculation of for example sin and cos functions be performed simply because the triangular functions have to be calculated for constant values. Calculating sin, cos and tangent values to defined values because there is a common relation between two variables where the variable are depending on each other, it is possible instead of performing sin and cos calculations simply to pick up a value in a table. This value within the same value each time a calculation is performed. This will reduce the processing time in a computer system because calculation of sin and cos functions always are time consuming measured in the number of steps that have to be performed in the processor. Simply picking up a constant in a memory will be may be 100 or 200 times faster than calculating a sin and cos function. Because all calculations in a processor also are power consuming will the highly effective calculations with less computer steps result in a much smaller power consumption of the processor.

Defining the IJSQF

In a further preferred embodiment of the invention can the system calculate the phase-or time difference by a Least Square Fit, LSQF, to a model function and thus obtain especially the time difference DT between the two earlier mentioned signals but also the transmission times (Ta, Tb) may be obtained this way. DT is obtained by a mathematical analyse resulting in the following formulas: gij) = g(Jl a1; a2, a3,..., aM) (10)

The model function g can either be linear or non-linear. d(j) are measured data (numbers from 1 to Ns) and a's are parameters in the model function to be fitted to best fit to the data.

As can be seen from figure 1 an obvious model would be the following: git) = hit: a3, a4,..., aM) sinCc^ t + α2) + C (12)

The Least Square Fit method can result in a relative simple calculation, especially if some of the values are constant values. This will be the case as previous described in many situations simply because the flow duct is well-known as well as the transducers and if the media is well-known and not have any changes it is possible by the Least Square system to perform a relative fast and highly effective calculation of a flow.

In a further preferred embodiment for the invention can the envelop function have the following formula: gA(t) = hA sin (cot + φΑ) (13) gB(l) = /iB sin (cut + φΒ) (14)

The gA and gB can be rewritten to the following: gA (t) = hA cos (ip,5) sin (cut) + hA sin(tp^) cos (cut) = CQ sin(cot) -f cos (cot) (15) gB(t) = hB cos(<pB) sin (cut) + hB sm(<pB) cos(mt) - D0 sin(cnt) + D1cos(cut) (16)

Where C0 = hA cosi<pA) ; C\ = hA sm((pA) ; D0 ----- hB cos((pB) ; D1 == hB sin(cpg) or φΑ = atanCCo/Ci) (17) φΒ = atan(D0/D1) (18)

The time difference DT = rfD between the received signals at transducer A and B is given by:

where f is the transducer frequency.and gB are linear function with respect to the parameters C0, C1, D0 and Dr. The least square for linear function can be solved analytical and is given by the following [1]: ATAC = ATdA (20) ATAD = ÅrdB (21) Where

By replacing f with fr where ff is a chosen reference frequency and choosing fr as the sample frequency divided by an integer number, calculations will be extremely simplified and thus very fast to implement in a microprocessor.

When replacing f with fr the calculated Td will not be the correct time difference for the real signals but this will just be a scale factor but may also be adjusted for by multiplying calculated Td with fr/f.

The model parameter is only applied to a finite part of the received signal. The interval for observation is kept constant for the two set of observations. The two signals have a small phase difference, such that the subset of the two envelop function is not the same. This means that phase difference implies that the start and end of the two envelop functions are not the same and only the overlapping part in the middle will be the same. To remove the edge part at the start and end has a windows weight function been multiplied to each observations, which is called the winNiS function. ATA and Å is in-depended by the actual measurements and can therefore be calculated in advance.

In the particular implementation has a>r been forced to a fixed value:

The Hanning (also known as Hann) function has been selected for the particular implementation. Other windows function could also be used if needed.

One of the important property of the weight function is be continues. As an example would a Hamming function be discontinues at the start and the end. Discontinuity will result in systematic errors.

Another important property is to let as many as possible observation have a high weight. As another example the Blackman windows function is continues, but it will favour the middle section of observations more than the edge section of the observations in comparison w'ith the Hanning function. Hence, the Hanning function will give a better robustness for noise, because more samples are with a higher weight. ωτ might not reflect the actual transducer frequency a>t and could result in an error. However, this is not the case, because the same error is made for both signals and due to the least square minimization will the phase difference still be valid.

By using the envelop function, the so-called Hanning function, it is possible to limit all calculations to the very important part of the received curvature. By the Hanning function it is possible to achieve the actual time delay measurement at mostly the same position of an oscillating sin curve. Because only it is a time different that is important for the measurement and therefore only the 0 crossing of the actual sin function that is important most of the rest of the sin curves has for the measurement nearly no influence. Therefore, can a highly effective calculation of flow be performed. A ' A and A are in-depended by the actual measurements and can therefore be calculated in advance.

Actually, the selected cor and Nls is even makes sure that the least square minimization is made over complete number of sin cycles of the signal. This property prevents an error, which would be made for in-complete sin cycle of the signal.

Defining the DCFD (Digital Constant Fraction Discriminator),

In a further preferred embodiment of the invention can the system calculate the transmission times by a DCFD defined by the following formulas:

Where

Ns is the number of samples.

Nf is the number of samples for one period of transducer frequency, tv is the window function with size Nw. A is the complex reference function, f is the transducer frequency, fr is the reference frequency and fs is the sample frequency. d is sampled data received either at Transducer A or B. LP is a low pass filter.

Dcfd and KCFD are constants.

The DCFD basically defines a specific position in the received signal relative to the start of the signal DCFD "POS. This position will stay the same relative to the start of the signal no matter the time shift in the signal and no matter the amplitude of the signal! In obtaining Ta and Tb we actually use at least one zero crossing of the received signal specified relative to DCFD__POS!

In the formulas defining the DCFD a relative simple and fast calculation of the flow may be achieved if f is replaced by if in the formulas, because sin and cos functions as previously described can be reduced to picking up constant values in a memory.

The pending patent application further concerns use of a system as disclosed in claim 1 or in a method as disclosed in the claim 2-21, whereby the system and the method is used for very fast calculating of fluid flow in gas or liquid in a flow channel. Herby can be achieved calculation of flow with extremely low' power consumption of the system calculating the fluid flow.

Description of the Drawing

Fig. 1 discloses one possible embodiment according to the invention.

Fig, 2 discloses one possible example of the burst and the received signal.

Fig. 3 discloses one possible method to verify the algorithm behaviour

Detailed Description of the Invention

Fig. 1 show's one possible embodiment related to the invention, namely an ultrasonic flow meter where the fluid flow is calculated by calculating the phase difference between two received signals.

The system 2 comprises at least one flow duct 4 in which flow' duct at least two ultrasound transducers 6 and 8 are in operation. The two transducers are transmitting ultrasound 10, may be such as one for example 6 is operating as a transmitter where 8 is operating as a receiver. Very shortly after the transmission is ending the transmission is changed so it performed in the opposite direction. The transducer 6 is by communication line 7 connected to a control system 12, where the transducer 8 is connected by connection line 9 to the same processor system 12. The processor system 12 is connected by a communication line 14 for example to a display 16, Further is indicated one signal line 18 may be connected to a wireless communication system.

In operation the flow will be measured between the transducers 6 and 8 where in one situation, the transducer 6 acts as transmitter and the transducer 8 acts as receiver, and in the next situation the transmission occurs in the opposite direction where the transducer 6 is the receiver and the transducer 8 is the transmitter. On the basis of these signals and by the means provided inside the microcontroller 12 the system can calculate the flow in a highly efficient manner,

An ultrasonic flow meter consist of two transducers 6 and 8 as disclosed in the fig 1.

The following sequence of steps is performed: 1. A burst of n periods are sent from transducer 6 and received on transducer 8. 2. A burst of n periods are sent from transducer 8 and recei ved on transducer 6.

The flow is given by:

Where <3 is the flow' [m3/s], K is a correction factor, C is a nominal calibration constant given by 23.7 10~6 m3, TD is the time difference between TA and TB, TA is the time from sending the burst on transducer A until received on transducer B and TB is the time from sending the burst on transducer B until received on transducer A, in Figure 2 is an example of the burst and the received signal shown.

The main objective of the algorithm is to determine TD, TA and TB with the following properties: 1. High accuracy - standard deviation for TD below' 5 ns, TA and TB below 250 ns. 2. Adapt to the received signal amplitude, e,g. not dependent on a particular gain. 3. Robust in respect to noise and DC offset for the received signal. 4. Low power consumption, e.g. very efficient and fast execution time, so that consumed power is low.

The algorithm has been designed with the above features in mind.

The algorithm consists of two parts: 1. Determine TA and TB. 2. Use the found TA and TB to select NLS samples for received signal at Transducer A and B. Perform a Least Square fit to determine TD,

Fig, 3 discloses one possible method to verify the algorithm behaviour, which algorithm has a number of simulations be made. In the following are the main results of the simulations stated.

The simulation has been made in Scilab [1]. Two Scilab scripts have been made for the calculation of the CFD and the LS.

Fig. 4 shows a possible embodiment for an amplitude signal is used to determine the arrival times Tup and 7L by means of a digital constant fraction discriminator CFD 146.

The determined Ta and Tb are used to place the sampling frame so it starts where the pulse would have been if there has been no dead delay. The dead delay is the delay due to cables, solid parts of transducers and delay in the band pass filter. The start of the sampling frame shall ideally equal the transmission time in the fluid Tup and Tdwn. The frequency of the timer clock limits the resolution, but the precision is sufficient for the Tup and Tdwn values in the denominator, but not sufficient for the difference At since it requires more resolution than the sampling frames are adjusted with.

1 ALGORITHM

An ultrasonic flow meter consists of two transducers A and B,

The following sequence of steps is performed: 3. A burst of n periods are sent from transducer A and received on transducer B. 4. A burst of n periods are sent from transducer B and received on transducer A.

The flow is given by:

(28)

Where q is the flow [m3 fs j, K is a correction factor, Cal is a nominal calibration constant, TD is the time difference between the two received signals a and b with transmission times TA and TB, TA is the time from sending the burst on transducer A until received on transducer B and T8 is the time from sending the burst on transducer B until received on transducer A.

In Figure 1 is an example of the burst and the received signal shown.

Figure 1: Burst (Blue) is send on Transducer A and received (Red) on Transducer B.

The main objective of the algorithm is to determine TD, TA and TB with the following properties: 5. High accuracy - standard deviation for TD below 5 ns, TA and TB below 250 ns. 6. Adapt to the received signal amplitude, e.g. not dependent on a particular gain, 7. Robust in respect to noise and DC offset for the received signal. 8. Low power consumption, e.g. very efficient and fast execution time, so that consumed power is low.

The algorithm has been designed with the above features in mind.

The algorithm could consist of two parts: 3. Determine TA and TB. 4. Use the found TA and TB to select Nls samples for received signal at Transducer A and B. Perform a Least Square Fit, LSQF to determine TD.