WO2002080618A1

WO2002080618A1 - Noise suppression in measurement of a repetitive signal

Info

Publication number: WO2002080618A1
Application number: PCT/DK2002/000198
Authority: WO
Inventors: Jan Petersen; Stefan Gram
Original assignee: Leonhard Research A/S
Priority date: 2001-03-30
Filing date: 2002-03-25
Publication date: 2002-10-10

Abstract

The method of suppressing noise in a signal from a device or from a living being captures a plurality of signals from the device or from the living being, where each of the captured signals is in a fixed temporal relationship to a synchronising signal. The synchronising signal can be derived from the device, eg a vibration signal, or from the living being, eg an ECG signal. Alternatively, the method uses a test signal with a repeated sequence which is supplied to the device under test. A signal is captured from the device representing the response of the device to the test signal. Characteristic parameters such as transient characteristics are derived from the captured signal, and a maximum value of the characteristic parameters is determined for each repeated response to the repeated sequence in the test signal. A minimum value is determined among the maximum values of the characteristic parameters. This method gives a fast convergence.

Description

Noise suppression in measurement of a repetitive signal

Technical field of the invention

The invention relates to testing and to quality control of mechanical, electrical, acoustic, electroacoustic and other systems, using excitation signals, measuring, and evaluating the response of the system to the excitation signals.

Background of the invention

Traditionally, electrical systems and in particular electroacoustic transducers such as loudspeakers and speaker transducers used in telecommunications equipment have been tested using swept sine wave signals, and the re- sponse of the equipment has been analysed for eg amplitude and phase characteristics and total harmonic distortion (THD) .

Non-linear characteristics such as rub-and-buzz are often short duration transient signals, which may be audible but difficult to measure. Recently, methods have been introduced for detecting non-linear and impulse characteristics such as rub-and-buzz. Such methods are described in EP 737 351 and in the references [1], [2] and [3]. Measuring rub-and-buzz by means of transient analysis has been found to be much more robust in noise than traditional techniques based on harmonic analysis or total harmonic distortion.

In environments with high background noise levels the transient analysis can be improved with known noise reducing techniques. Known methods of reducing noise in signals measured in response to an excitation signal involve repeated excitation of the device under test and averaging several noise infected responses from the device, which is time-consuming.

It is an object of the invention to improve noise suppression in methods using deterministic excitation signals .

Summary of the invention

The noise suppression method of the invention is based on the fact that the response signals, and in particular transients such as rub-and-buzz in the response, are caused by the excitation signal, and the response signals will therefore also be synchronized to the excitation signal. In principle, this applies to all system analysis where a signal is measured in response to an excitation signal. This means that if the same excitation signal, eg a swept sine wave, is applied two or more times to the same device under test such as a transducer, then eg transients caused by rub-and-buzz will occur in the same place in the time signal relative to the excitation signal. Background noise, whether steady state or tran- sients, will in most cases be stochastic, ie unrelated to the excitation signal, and background noise transients will therefore not occur in the same place in time in a repeated rub-and-buzz measurement but have a random distribution in time.

In one embodiment the method of the invention uses repeated test signals, or a test signal including a repeated sequence, to excite a device under test. The measured response, ie the time signal, from the device under test is analysed for characteristic parameters such as transients and divided into time intervals or bins . For each time interval the maximum values of the characteristic parameters are stored, and for each time interval the minimum of the stored maximum values is taken. This method is referred to as the max/min method. The method is fast converging. The invention will be described using examples on how to improve the noise suppression capability in rub-and-buzz measurements based on transient analysis. However, the fast converging method of the in- vention is useful for many other analyses, eg where the noise suppression would otherwise be performed by well known averaging methods. With the invention no time con- suming statistical calculations are made but rather only simple calculations.

The preferred test signals include a repeated swept sine wave signal and a stepped sine wave signal, where each test frequency includes a plurality of periods of the sine wave. Other test signals may be used, eg repeated random and pseudo-random sequences and repeated impulsive signals .

In some systems, eg mechanical, electrical and physiological systems, the system itself generates a signal, to which the signal to be measured and analysed is synchronised. In case of a rotating machine most signals repeat themselves with a period, which is derived eg as a multiple or sub-multiple of the period of a triggering basic event in the machine. In an internal combustion engine a basic event could be eg an electrical ignition pulse to a spark plug in a cylinder of the engine, or the triggering event could be derived eg from a noise signal captured with a microphone or from a vibration signal captured with an accelerometer or a laser based system.

In auscultation the physician listens to sounds from heart and lungs. It takes a trained professional to evaluate heart sounds, which are most often noise-infected. In human and animal physiological systems electrical pulses trigger physical activity such as heart activity and other muscle activity. In medical examinations electro-cardiograms (ECG) are routinely taken. QRS complexes reflect events triggering the heart muscles to contract . Thus, in another embodiment of the invention the measurement equipment emits no excitation signal or test signal. Rather, a triggering signal captured from a machine or a human or animal body is used to synchronise repeated capturing of signals from the machine or the living being.

The literature references [1], [2] and [3] give an overview on rub-and-buzz and the underlying theory and meth- ods. In [1] some examples are given, in [2] the transient analysis technique is described and in [3] a software program HARMONI™ with the technique implemented is described.

In the following some methods of improving noise suppression in rub-and-buzz measurements based on transient analysis are described. One method, which is referred to as the max/min method, is described in more detail, and some other methods are only outlined. The max/min method may be implemented using the transient analyser HARMONI™ [3] .

In general rub-and-buzz in transducers can be detected as transients with a transient analysis. See [1] and [2].

Brief description of the drawings

Figure 1 shows a typical electroacoustic measurement setup with only the basic components shown,

Figure 2 schematically shows the results of two measurements with the max/min method of the invention, Figure 3 shows a transient analysis of a 12 mm transducer with rub-and-buzz and no background noise,

Figure 4 shows a transient analysis of the transducer in figure 3 with background noise,

Figure 5 shows the result of the max/min method of the invention using only two measurements with a S/N ratio of 10 dB, and

Figure 6 shows the same as in figure 5, but using five measurements .

Detailed description ofthe invention Figure 1 shows a typical measurement set-up with only the basic components shown. In this example the device under test is eg a loudspeaker or a speaker transducer for use in communications equipment such as a mobile telephone. The device under test can also be a more complex elec- troacoustic or mechanical system.

A signal generator generates electrical test signals which are fed to the device or transducer under test. A microphone or other suitable transducer receives sound emitted from the device under test and outputs electrical signals representing the received sound signal. Other quantities such as vibration may also be measured and analysed using other suitable transducers. The output signals from the microphone are properly conditioned and amplified and fed to an analyser performing the desired analysis, eg the HARMONI™ transient analyser [3] per- forming a transient analysis, or an analysis for other characteristic parameters of the response signal.

In one embodiment of the invention a swept sine wave sig- nal is used as the excitation signal. In another embodiment of the invention to be described later a stepped sine wave signal is used. Other signals may be used and the noise suppression techniques may have to be adapted accordingly. The device under test is excited with a test signal. The test signal is a swept sine wave signal with either a linear or a logarithmic frequency scale, ie the rate of change of the frequency depends either linearly or logarithmically on time. The lower and upper frequency limits for the excitation signal are chosen in dependence on the system or device under test, so that the characteristic quality parameters are properly measured and possible or suspected flaws and defects will be revealed. A signal generator outputs the test signal which is amplified to a suitable amplitude for exciting the device under test and is fed to the device.

Max/min method

In the method according to the invention the device under test is excited two or more times using the same excita- tion signal, which is preferably either a swept or a stepped sine wave signal. The measured response signal is analysed for transients, and the time signal of the transient analysis is divided into time intervals or bins, where each bin contains measured signal values. For each of the repeated measurements the maximum transient value within each time interval is found and stored in the corresponding bin. Next, with the repeated measurements us- ing the same excitation signal, the minimum, ie the lowest one of the stored maximum values, within each bin is taken.

In repeated measurements using the same excitation signal, possible transients in the response to the excitation signal will occur at the same time relative to the excitation signal and in particular in the same time interval. Such transients occurring in the same time inter- val will therefore be stored in the same bin.

Transients caused eg by rub-and-buzz are synchronized to the input signal, and the values contained in bins in which the transients are detected will therefore not drop below the level of the rub-and-buzz. In bins with no detected rub-and-buzz transients the values will vary randomly, and both "low" and "high" values can be expected, and with repeated measurements eventually a "low" value can be expected in each bin. Increasing the number of re- peated measurements will result in lower minimum values in each bin.

For the repeated measurements the minimum value in each bin will be taken as the result. In bins with detected transient responses to the excitation signal, the minimum value will represent the transients, or more precisely, the lowest one of the transients, whereas in bins with no detected transient responses the minimum value will represent the lowest noise transient value detected.

In Figure 2 the max/min method is illustrated schematically for two consecutive transient measurements marked x and D, respectively, representing the transients found in the respective measurements using the same excitation signal. One particularly high transient is indicated x and is here assumed to be caused by background noise and is therefore only represented in one of the measurement series as indicated by the x with a high value. In the other measurement series indicated D the transient value in the same bin is low, because no transient but only noise is detected. This is taken to signify that the high transient value in that bin is not a response related to the excitation signal and therefore represents a random transient signal due to background noise. In the max/min method the minimum of the stored maximum values within each bin is chosen as the result of the transient analy- sis, and the transient marked x, which is derived from background noise is therefore suppressed.

If the transducer suffers from rub-and-buzz and there is no background noise, then the detected transient or tran- sients caused by rub-and-buzz will be detected in the measurement. Figure 3 shows an example of a rub-and-buzz measurement for a 12 mm transducer for a mobile phone. A swept sine wave signal as an excitation signal is used, ranging from 100 Hz to 800 Hz with a duration of 1 s . The bin size or width is about 1 ms . Figure 3 shows the transients found in the measurement. A simple explanation is that transients above a certain level are indicative of either rub-and-buzz or background noise. In this example the transducer has rub-and-buzz around 0.24 s and 0.66 s after the start of the excitation signal corresponding to two narrow frequency ranges around 270 Hz and 560 Hz. In references [1] and [2] a more thorough explanation of the analysis method is given.

If the transducer suffers from rub-and-buzz and there is background noise then the maximum within each bin could be derived either from rub-and-buzz or from noise. Figure 4 shows an example of a measurement on the same transducer as above, but measured in background noise. The signal-to-noise ratio is around 10 dB with the signal be- ing the swept sine with rub-and-buzz and noise being a speech signal (male voice) .

In the max/min method the maximum in each bin is found for each measurement . For two or more measurements the minimum in each bin is then taken across the measurements. As the rub-and-buzz is synchronized to the sweep and the background noise is stochastic, a repetition will suppress the noise and keep the rub-and-buzz. Figures 5 and 6 show the result after two respectively five meas- urements in noise (S/N about 10 dB) after the max/min method has been applied.

It can be seen that after only 5 measurements or iterations the result resembles the result for the measurement without background noise.

The method of the invention will preferably have a stop criterion. A usable stop criterion could be that either the measurements stop when the accumulated result drops below a certain pre-set pass/fail limit, or it continues until the pre-set maximum number of iteration is reached. A more refined stop criterion could be applied if the measurement is started shortly before the excitation signal is applied and continues a short time after the excitation signal. If transients are detected before or after the excitation signal, then those transients are bound to be caused by the background noise, and the measurement will have to be repeated if the pass/reject limit has been exceeded. If no transients from background noise are detected before and after the excitation signal has been applied, then there is a high probability that any transient detected has been caused by rub-and-buzz and not background noise.

The usual approach when background noise is present is to take the average of two or more measurements. The average would also work with this method but much slower.

The measurement is repeated two or more times. If transients within the same bin are close in value then the average is taken. If one transient is far from the rest then it is rejected. A pre-set constant is set in order to decide how far in value a transient can be in order to be considered.

As an alternative, a stepped sine wave signal is used as a test signal instead of a swept signal. A stepped sine wave signal uses a number of discrete frequencies one at a time, and the test signal is stepped through a predetermined range of frequencies with each frequency having an individual, predetermined duration. The duration of each frequency will typically be determined as a number of full cycles of the frequency with due regard to the impulse response of the device under test. For each of the discrete frequencies the test signal repeats itself for each cycle of 2π of the signal. Each cycle of the measured response signal is divided into time intervals or bins each representing only a fraction of a full cycle of 2π.

Each of the discrete test frequencies in the stepped sine wave signal can be regarded as a time window of a pure sine wave signal. For one full cycle of the sine wave signal from 0 to 2π the measurement could be divided into n bins . The first bin represents the angular interval 0 to x, where x is a fraction 1/n of 2π. This bin will be processed together with the bin from 2π to 2π + x, which contains the measured response signal from the next period, which is a repetition of the corresponding interval x in the previous cycle. In the same way all the bins that are processed together are placed 2π apart and will thus represent repeated measurements of the same fraction x of the full cycle. Like in the max/min method described above the maximum values of the repeated response signal are stored in each bin, and the minimum of the stored maximum values is taken as the result. Alternatively, a running minimum or running average over m bins could be applied.

Transients in bins representing the same part of the signal are thus processed together. Alternatively, the method could be to take the minimum or average possibly combined with a running process. If the excitation signal is a sweep, then the bins have to be transformed in a proper way so that each full cycle is split up into exactly the same number of bins.

In another embodiment of the invention a repeating triggering signal is obtained from the device or living being under test. In eg a rotating machine the triggering signal could be derived from a vibration signal representing vibration of critical parts, and in a living being the triggering signal could be a physiological signal such as a QRS complex in an ECG signal triggering the heart. In the analyser each captured triggering signal triggers a recording of the signal to be analysed, eg a vibration signal or heart sounds . Repeated recordings are made and analysed as described above.

Literature

[1] Application note, "Measuring rub-and-buzz in

Transducers, Quality control of loudspeakers", can be found on the Internet address www.leonhardresearch.dk

[2] Application note, "Measuring the instantaneous energy in signals", can be found on the Internet address www. leonhardresearch . dk

[3] Product specification sheet "Transient Analyser - HARMONI™Lab", can be found on the Internet address www. leonhardresearch . dk

Claims

1. A method of suppressing noise in a signal from a device or from a living being, the method comprising

- obtaining a plurality of synchronising signals, - capturing, from the device, a plurality of signals each of which having a fixed temporal relationship to one of the synchronising signals,

- deriving characteristic parameters from the captured signals, - determining, for each repeated response to the repeated sequence, a maximum value of the characteristic parameters,

- determining, among the maximum values of the characteristic parameters, a minimum value.

2. A method according to claim 1, wherein the device is a periodically operating machine, and the plurality of synchronising signals are derived from the periodical operation of the machine.

3. A method according to claim 2, wherein the synchronising signals are derived from a signal representing vibration of the machine.

4. A method according to claim 2, wherein the synchronising signals are derived from a signal representing sound emitted from the machine.

5. A method according to claim 2, wherein the captured signals represent vibrations of the machine.

6. A method according to claim 2, wherein the captured signals represent sound emitted from the machine.

7. A method according to claim 1, wherein the plurality of synchronising signals are derived from a periodical physiological process in a living being.

8. A method according to claim 7, wherein the synchronising signals are electro-cardiogram (ECG) signals, and the captured signals relate to heart activity in the living being.

9. A method according to claim 8, wherein the signals relating to heart activity are heart sounds captured with a microphone.

10. A method according to claim 1, wherein the device is a device under test, the method further comprising

- generating a test signal, the test signal including a repeated sequence,

- supplying the test signal to the device, and

- deriving the plurality of synchronising signals from the repeated sequence in the test signal, and wherein the captured signals represent the response of the device to the test signal.

11. A method according to claim 10, wherein the repeated sequence is a swept sine wave signal.

12. A method according to claim 10, wherein the repeated sequence is a stepped sine wave signal.

13. A method according to claim 10, wherein the repeated sequence is a random or pseudo-random sequence.

14. A method according to claim 10, wherein the repeated sequence includes impulses .

15. A method according to claim 10, wherein the characteristic parameters derived from the captured signal represent possible transients in the captured signal.

16. A method according to claim 10, wherein the characteristic parameters are derived in a plurality of time intervals.