WO2009053342A1

WO2009053342A1 - Compression and reconstruction of a pseudosinusoidal digital signal

Info

Publication number: WO2009053342A1
Application number: PCT/EP2008/064161
Authority: WO
Inventors: Silvère BARUT; Nicolas Mayer
Original assignee: European Aeronautic Defence And Space Company Eads France
Priority date: 2007-10-26
Filing date: 2008-10-21
Publication date: 2009-04-30
Also published as: FR2923104B1; FR2923104A1

Abstract

To compress a digital acquisition signal (12) representing a sampled centred pseudosinusoidal analogue signal, a data compression method (10) includes a lossy compression step (30) optionally followed by a lossless compression step (40). The lossy compression step (30) is such that each sample of the digital acquisition signal (12) corresponding to a local extremum is compared with two thresholds S1 and S2 and is preserved in a synthetic digital signal (13) without modification if the amplitude thereof is greater than S1 or less than S2 or, failing this, after modification if the amplitude of a sample corresponding to the immediately following and/or immediately preceding local extremum is greater than S1 or less than S2. The other samples are not preserved in the synthetic digital signal. The invention also relates to a reconstruction method (50), for reconstructing an approximation of the acquisition signal (12), and to a system for implementing the two methods.

Description

COMPRESSION AND RECONSTRUCTION OF A PSEUDO-SINUSOIDAL DIGITAL SIGNAL

The present invention relates to a method and a system for compressing and reconstructing a digital signal.

More particularly, the present invention relates to a method and a system for compressing and reconstructing a digital signal representing a pseudo-sinusoidal analog signal centered such as a signal delivered by a non-destructive ultrasonic control system used in the control of structures.

The methods of non-destructive ultrasonic testing are used in many industrial fields. Such methods make it possible to control a structure for detecting defects on the surface and / or in the depth of said structure without degrading it. These methods are used in a production phase of the structure to check the quality of manufacture and / or in a maintenance phase to check if defects have appeared in said structure. The methods of non-destructive ultrasonic testing use acoustic waves of frequencies chosen for example between 0.1 MHz and 25 MHz. A control system using ultrasound, shown in FIG. 1, generally comprises:

a probe 1 converting an electrical analog signal into an acoustic wave, and vice versa,

a generator / receiver 2 of electrical pulses connected to the probe 1,

a device 4 for acquiring an analog signal received from the generator / receiver 2, and performing in the current improved devices at least one conversion, called sampling, of the analog signal into a digital signal,

a storage device 5 for storing the digital signal in for further processing,

an analysis device 6, which can be confused with the acquisition device 4, for analyzing the digital signal.

A coupler 7 is placed between the probe 1 and a structure 3 under control in order to promote the diffusion of the acoustic wave from one to the other. This couplant 7 is most often a liquid, for example water or sometimes a gel.

A system such as that of FIG. 1 is based on the measurement of echoes which are reflections of the acoustic wave on interfaces between parts having different acoustic impedances, for example between the coupler 7 and the structure 3 and / or at a fault internal to said structure.

In most existing systems the electrical pulses provided by the generator 2 to the probe 1 are square pulses. In response to these square pulses, the probe generates ultrasonic acoustic pulses.

Echoes received at probe 1 are converted into electrical pulses that have the following characteristics:

- substantially single-frequency, - finite durations,

- centered around a generally zero average value.

These pulses are called pseudo-sinusoidal pulses centered or PSC pulses in the rest of the presentation.

The electrical analog signal received from the probe for a given position and during a control period, called A-scan in the context of non-destructive ultrasonic testing, is a time dependent amplitude function which is the sum of a structure noise and / or inherent to the measurement equipment, and PSC pulses shifted in time (echoes).

Such a signal, analog or digital, is called PSC signal. An example of a PSC signal is shown in FIG. To control the entire structure 3, A-scans are obtained with a plurality of positions of the probe 1 relative to said structure. The sampling by the acquisition device 4 of the A-scans thus obtained produces digital signals PSC which are stored in the storage device 5 for a subsequent detailed analysis by the analysis device 6.

The quantities of data that need to be recorded in the storage device 5 can be very important, because it is not unusual in the industry to have to look for any submillimetric size defects in structures of several square meters.

In order to limit the amount of recorded data to reasonable values, the data is usually compressed.

For the compression of a digital signal PSC, it is known to keep for each PSC pulse of said digital signal only the sample (amplitude and time position) having the maximum amplitude.

It is also known from a so-called "multi-peak" method to keep only the samples of the digital signal PSC which correspond to vertices and whose amplitudes exceed a given threshold.

Both methods achieve high compression rates, but are very destructive. In particular, it is not possible with these methods to reconstruct the PSC digital signals in order to perform a more detailed analysis of the results, for example a correlation and / or a frequency analysis for reconstructing an image of the damage of the controlled structure. The present invention relates to a method of data compression adapted to a digital signal PSC which authorizes the reconstruction by a reconstruction method of an approximation of said signal adapted to the detailed analysis, and a system for the implementation of said methods.

The data compression method of the invention, implemented to compress a digital acquisition signal representing a pseudo-sinusoidal analog signal sampled, the digital acquisition signal being composed of samples and each sample being defined by a sampling instant and a measured amplitude, comprising a lossy compression step producing a synthetic digital signal, and characterized in that in lossy compression the samples of the acquisition digital signal corresponding to local extrema are identified, each sample identified as local extremum is compared with two thresholds S1 and S2, S2 being less than S1, and: 1) is kept identically in the synthetic digital signal if the measured amplitude of the sample in question is greater than S1 or less than S2,

2) is retained in the synthetic digital signal after modification if the measured amplitude of the sample in question is less than or equal to S1 and greater than or equal to S2, and a sample corresponding to an immediately following local extremum and / or a corresponding sample at an immediately preceding local extremum corresponds to 1).

In lossy compression, the samples of the digital acquisition signal are not retained in the synthetic digital signal if they do not correspond to 1) or 2).

The modification in case 2) is preferably a suppression of the measured amplitude of the sample in question, or a replacement of the measured amplitude of the sample in question by a predefined amplitude Vd between S1 and S2.

In a particular mode of implementation of the data compression method, the thresholds S1 and S2 are equal in absolute value.

To reduce the power of a noise present in the pseudo-sinusoidal analog signal centered, a raw digital signal obtained by sampling said analog signal is preferably filtered low-pass to produce the digital acquisition signal.

In order to increase the overall compression rate of the data compression method, the synthetic digital signal obtained by compression with loss of the acquisition signal is preferably compressed without loss.

In a particular embodiment of the data compression method, the lossless compression of the synthetic digital signal implements a Huffman-type coding, advantageous when a signal to be compressed contains values that are repeated and particularly adapted to the signal. synthetic in the case where it has many amplitudes forced to the value Vd.

The digital acquisition signal is most often centered around a substantially zero average value; in this case, the thresholds S1 and S2 are respectively positive and negative, and the predefined value Vd is preferably equal to zero.

The reconstruction method of the invention is particularly suitable for reconstructing a digital signal representing a pseudo-sinusoidal centered signal. The reconstruction method produces a digital analysis signal from a synthetic digital signal consisting of samples, each sample being defined by a sampling instant and an amplitude. The reconstruction method according to the invention comprises at least one interpolation step in which an interpolation function is forced to pass through each sample with a horizontal tangent, and in which the amplitude of the digital analysis signal is forced. to a predefined value Vm between two consecutive samples of magnitude amplitude Vm of the synthetic digital signal.

Preferably, an amplitude of value Vm is associated beforehand in the interpolation step with each sample of the synthetic digital signal having no associated amplitude, or before each amplitude of value equal to a predefined Vd value is replaced beforehand. by an amplitude of value Vm, most often equal to zero.

In the case where the reconstruction method is used on a lossless compressed digital signal, the compressed digital signal is decompressed without loss to produce the synthetic digital signal. Advantageously, the interpolation function is obtained between two consecutive samples, not both of magnitude amplitude Vm, by implementing a cubic interpolation spline, more precisely an interpolation by Powell-Sabin elements.

The ultrasonic non-destructive testing system of the invention includes a probe connected to an electrical pulse generator / receiver, an acquisition device and an analyzing device.

The non-destructive ultrasonic testing system is characterized in that the acquisition device comprises means for sampling a pseudo-sinusoidal analog signal supplied by the receiver in a digital acquisition signal, compression means with loss of the digital acquisition signal, which lossy compression means:

- identify the samples of the digital acquisition signal corresponding to local extrema,

compare each sample identified as local extremum to two thresholds S1 and S2, S2 being less than S1, and keep the sample considered in a synthetic digital signal:

1) identically if the measured amplitude of the sample considered is greater than S1 or less than S2,

2) after modification if the measured amplitude of the sample considered is less than or equal to S1 and greater than or equal to S2, and a sample corresponding to an immediately following local extremum and / or a sample corresponding to an immediately preceding local extremum corresponds to to 1). - and do not keep the samples of the digital signal in the synthetic digital signal if they do not correspond to 1) or 2).

The non-destructive ultrasonic testing system is also characterized in that the analysis device comprises means for interpolation of the synthetic digital signal producing an approximation of the digital acquisition signal, which interpolation means impose a function of interpolating to go through all points of said synthetic digital signal with a horizontal tangent.

In a particular embodiment of the ultrasound non-destructive testing system of the invention aimed at increasing the compression ratio obtained by the lossy compression means, the acquisition device comprises compression means without loss of the digital signal. synthetic forming a compressed digital signal. The analysis device also comprises in this embodiment means of decompression without loss of said compressed digital signal.

Preferably, the interpolation means of the ultrasonic non-destructive control system implement a cubic interpolation spline, more specifically a Powell-Sabin element interpolation, which is particularly suited to the reconstruction of centered pseudosinusoidal signals.

In a particular embodiment of the non-destructive ultrasonic testing system, the acquisition device and the analysis device are connected by a communication network.

The following description of embodiments of the invention is made with reference to the figures which show in a nonlimiting manner:

FIG. 1, a non-destructive ultrasound control system, FIG. 2, an example of a digital signal PSC obtained with a system such as that of FIG. 1,

FIG. 3, the steps of the data compression method according to the present invention,

FIG. 4, an example of a synthetic digital signal obtained using the data compression method,

FIG. 5, the steps of the reconstruction method adapted to the data compression method,

FIG. 6, an example of a digital analysis signal obtained by using the reconstruction method. In the following description, the invention will be detailed in the case of a non-destructive ultrasonic testing system of a structure 3, as shown in Figure 1, comprising a probe 1, a generator / receiver 2 of electrical pulses, an acquisition device 4, a storage device 5 and an analysis device 6 (possibly confused with the acquisition device 4). The acquisition devices 4, storage 5 and analysis 6 are for example connected by a communication network 8 such as an Ethernet-based local area network.

This exemplary architecture of a non-destructive ultrasonic testing system is not restrictive and methods according to the invention are applicable to any system of acquisition / analysis of pseudo-sinusoidal centered (PSC) analog signals having at least less a PSC pulse, that is to say substantially mono-frequency of finite duration and centered around a mean value generally substantially zero, and possibly a structure noise, for example in case of porosity of said structure and / or a thermal noise inherent in the measuring equipment. As illustrated in FIG. 3, a data compression method 10 implemented by the acquisition device 4 takes as input a so-called raw digital signal 11 which is the result of a sampling of an analog signal. PSC provided by the receiver 2 connected to the probe 1. The raw signal 11 is a set of N samples (a _n , t _n ) of index n, 0 ≤ n ≤ N-1, such that a _n is representative of the amplitude of the analog signal PSC measured at a sampling instant t _n = t ₀ + nT _e , where t ₀ is a start time of the acquisition and T _e is a sampling period. Equivalently, a sample of index n is represented as (a _n , n). The sampling is carried out by the acquisition device 4 or by another device not implementing the data compression method 10, in which case the raw signal 11 is read by the acquisition device 4 from the data acquisition device. storage 5.

The data compression method 10 has several steps that are executed sequentially or simultaneously, for example on separate data sets.

In a first step, performed in a particular mode of implementation of the data compression method 10, the raw signal 11 is low-pass filtered to reduce the power of the noise present in said raw signal. A cut-off frequency 21 of the low-pass filtering 20 is chosen so that the power of the PSC pulses contained in the raw signal 11 is substantially the same before and after filtering. For example, the cut-off frequency 21 is greater than or equal to three times a maximum frequency of the PSC pulses received by the generator 2. The low-pass filtering produces an acquisition signal 12 which is a set of N samples (f _n , t _n ), 0 ≤ n ≤ N-1. In the case where no low-pass filtering is applied to the raw signal 11, said raw signal and the acquisition signal 12 are the same.

In a second step of the data compression method 10, the acquisition signal 12 is lossy compressed to produce a so-called synthetic digital signal 13. By lossy compression it is to be understood that the acquisition signal 12 can not be reconstructed exactly from the synthetic signal 13.

The acquisition signal 12 has a number M, M <N, of samples which are local extrema (f _{g (m)} , t _{g (m)} ), 0 <m <M-1 and 0 <g (m ) <N-1, where g (m) gives the index of the local extremum number m.

The local M extrema of the acquisition signal 12 are first identified and then evaluated with respect to two thresholds S1 and S2, S2 being less than S1, which are parameters 31 of the lossy compression 30. The thresholds S1 and S2 are chosen so that S1 is greater than a mean value Vm of the pseudo-sinusoidal raw signal 11 centered, known a priori and generally substantially zero, and so that S2 is of value less than said average value Vm. It is considered initially that the average value Vm is substantially zero, and that the thresholds S1 and S2 are respectively positive and negative.

A local extremum (f _g ( _m ), t _{g (m} )) of index g (m) is preserved in the synthetic signal 13, possibly after modification, if one of the following two conditions is satisfied:

1) the amplitude f _{g (m} ) of the local extremum is greater than S1, or less than S2,

2) the amplitude f _{g (m)} of the local extremum is less than or equal to S1 and greater than or equal to S2, but condition 1) is satisfied by the local extremum of index g (m-1) ( if it exists, that is, if m-1> 0) and / or the local extremum of index g (m + 1) (if it exists, that is, if m + 1 <M-1) of the acquisition signal 12.

In the case 1), the local extremum is kept without modification and in the case 2), the local extremum is preserved after modification of the amplitude f _{g (m} ), which is replaced by the value zero.

Samples that do not correspond to local extrema and samples for which conditions 1) and 2) are not satisfied are deleted.

The synthetic signal 13 is for example obtained from a so-called intermediate digital signal, which is a set of M samples (fg (m), tg (m)), 0 ≤ m ≤ M-1, obtained by modification of the amplitude of the local M extrema of the acquisition signal 12. The M samples (f ' _g (m), tg _(m )), 0 <m <M-

1, are such that:

- f'g (m) = fg (m) if the amplitude fg ( _m ) is greater than S1, or less than S2,

- f'g _(m) = 0 if the amplitude f _{g (m)} of the local extremum of index g (m) is less than or equal to S1 and greater than or equal to S2.

In such a case, a sample (f _g (m), t _{g (m} )), 0 <m <M-1, is retained without modification in the synthetic signal 13 if it satisfies one of the two following conditions:

1 ') the amplitude fg _(m) is non-zero, 2') the amplitude f _g ( _m ) is zero but condition 1 ') is satisfied by the sample of index g (m-1) (s) it exists ie if m-1> 0) and / or the sample of index g (m + 1) (if it exists ie if m + 1 ≤ M -1) of the intermediate signal.

The synthetic signal 13 is composed of a set of L ≤ M samples (si, t _h (i)), 0 ≤ I ≤ L-1, and 0 ≤ h (l) ≤ N-1, the amplitudes si, 0 ≤ I ≤ L-

1, are equal to f _{h (} i ₎ or zero depending on the case and h (l) gives the index of a local extremum of the acquisition signal 12 satisfying one of the conditions 1) or 2) above. An example of synthetic signal 13 obtained by lossy compression of the signal of FIG. 2 is shown in FIG.

According to a preferred embodiment of the lossy compression

30, the thresholds S1 and S2 are modifiable and are for example chosen to adjust a compression ratio of the lossy compression as a function of the capacities of the storage device 5. In a particular embodiment of the compression with loss 30, the thresholds S1 and S2 are equal in absolute value.

In a third step, performed in a particular embodiment of the data compression method 10, the synthetic signal 13 is losslessly compressed 40 to produce a so-called compressed digital signal 14. The lossless compression 40 is a reversible compression such as that the synthetic signal 13 can be reconstructed identically after an appropriate lossless decompression, in the absence of disturbances on the compressed signal 14.

In the case where no lossless compression is applied to the synthetic signal 13, said synthetic signal and the compressed signal 14 are the same.

Advantageously, the lossless compression 40 implements a Huffman type coding. The Huffman coding, which makes it possible to obtain high compression ratios when a signal to be compressed contains values that are repeated, is particularly well suited to the synthetic signal 13 because of the presence in the latter of numerous amplitudes forced to zero.

The compressed signal 14 is for example stored in the storage device 5 for subsequent processing by the analysis device 6 using a reconstruction method 50 adapted to the data compression method 10, or directly processed by said device analysis 6.

As illustrated in FIG. 5, the reconstruction method 50, applied to the possibly corrupted compressed signal 14, comprises several steps that are executed sequentially or simultaneously, for example on separate sets of data.

In a first step of the reconstruction method 50, performed only if the lossless compression step 40 has been executed, the compressed signal 14 is decompressed without loss 60 to produce an uncompressed signal 15, which is the possibly corrupt synthetic signal 13. In the remainder of the presentation, it is assumed that the synthetic signal 13 and the decompressed signal are the same, and the notations of said synthetic signal are used to describe said decompressed signal.

In the case where the lossless compression step 40 of the compression process 10 has not been performed, the decompressed signal 15 and the compressed signal 14 are the same.

In the case where the lossless compression 40 implements a Huffman type coding, the appropriate lossless decompression 60 performs a Huffman type decoding. In a second step of the reconstruction method 50, the decompressed signal 15 is interpolated 70 to produce a so-called digital analysis signal 16, which is an approximation of the acquisition signal 12.

The interpolation 70 is performed by imposing on an interpolation function R (t) of the decompressed signal 15 to go through each sample (si, t _{h (} i)), 0 ≤ I ≤ L-1, of the decompressed signal 15 ( that is, R (t _{h (} i)) = Si, 0 ≤ I ≤ L-1) with a horizontal tangent.

The analysis signal 16 is the set of N samples (R (t _n ), t _n ), 0 ≤ n ≤ N-1.

An interpolation particularly adapted to the pseudo-sinusoidal centered characteristics of the analog signal represented by the acquisition signal 12 is a cubic interpolation spline, for example an interpolation by Powell-Sabin elements. In such an interpolation, for each interval [t _{h (} i), t _{h (} ι ₊ i)], 0 ≤ I ≤ L-2, we search for two cubic curves C [l, +] (t) and C [l +1, -] (t) such that: - C [l, +] (t) passes through (si, t _{h (} i ₎ ) with a horizontal tangent,

- C [l + 1, -] (t) passes through (sι + i, t _{h (} i ₊₁₎ ) with a horizontal tangent,

- C [l, +] (t) and C [l + 1, -] (t) intersect at a point at a moment (t _{h (} ι _r t _h (i + i)) / 2 and their tangents in this point are confused.

The interpolation function R (t) is such that R (t) = C [l, +] (t) for t in the interval [t _{h (} i ₎ , (t _{h (} ι ₊ i ₎ -t _{h (} i ₎ ) / 2] and R (t) = C [l + 1, -] (t) for t in the interval [(t _{h (} i ₊ i ₎ -t _{h (} i ₎ ) / 2, t _{h (} ι ₊₁₎ ], 0 <I <L-2.

An example of an analysis signal 16 obtained from the signal of FIG. 4 by a cubic interpolation spline by Powell-Sabin elements is shown in FIG. 6. In a mode of implementation of the advantageous interpolation 70 from the point of view of reducing the complexity of the system of equations to be solved, no equations are made between two consecutive samples of the decompressed signal having zero amplitudes. In this case the amplitudes of the intermediate samples to be interpolated are all forced to zero.

According to a preferred embodiment of the reconstruction method 50, if the decompressed signal 15 does not comprise samples at times of indices 0 and / or N, the interpolation 70 is performed from an uncompressed signal to which the samples (0, t ₀ ) and / or (0, t _N- i) have been added to produce an analysis signal 16 over the same time interval as that of the acquisition signal 12.

The analysis signal 16 obtained is an approximation of the acquisition signal 12. The approximation error mainly depends on the thresholds S1 and S2, which are chosen in order to minimize this approximation error while maintaining a global compression ratio high data compression method.

More generally, the data compression method 10 is also applicable to a raw signal 11 centered around a non-zero average value Vm known a priori. In this general case where the raw signal 11 is centered around a non-zero average value Vm, the thresholds S1 and S2 are chosen so that S1 is of value greater than the average value Vm, and so that S2 is of value less than said average value Vm. The modification of the local extrema corresponding to the preceding condition 2) is for example in this case a replacement of the amplitude f _{g (m)} by the value Vm. Even more generally, other types of modification are possible for local extrema preserved after modification in the lossy compression stage of the data compression method (i.e. local extrema corresponding to previous condition 2), for example:

the amplitude fg _(m) is suppressed;

the amplitude f _g ( _m ) is forced to a predefined value Vd between S1 and S2.

In the case where one of the preceding modifications is implemented, the interpolation step 70 of the reconstruction method 50 is modified as follows:

if the decompressed signal comprises samples without amplitude, a value amplitude Vm is associated beforehand with each sample without amplitude; if the decompressed signal comprises samples of amplitude of value Vd different from the average value Vm, we replace before each amplitude of value Vd by an amplitude of value Vm.

The data compression method 10 according to the present invention makes it possible to reduce the amount of data to be stored in the case of a system for acquiring a pseudo-sinusoidal analog centered signal. The compression method further allows reconstruction with the reconstruction method 50 of a signal approaching the acquisition signal 12 adapted to a detailed analysis.

Claims

- Data compression method (10) for compressing a digital acquisition signal (12) representing a sampled centered pseudo-sinusoidal analog signal, said digital acquisition signal (12) being composed of samples, each sample being defined by a sampling instant and a measured amplitude, said method comprising a lossy compression step (30) producing a synthetic digital signal (13), and being characterized in that in said lossy compression:

- the samples of the digital acquisition signal (12) corresponding to local extrema are identified,

- each sample identified as local extremum is compared to two thresholds S1 and S2, S2 being lower than S1, and:

1) is kept identical in the synthetic digital signal (13) if the measured amplitude of the sample considered is greater than S1 or less than S2,

2) is retained in the synthetic digital signal (13) after modification if the measured amplitude of the sample considered is less than or equal to S1 and greater than or equal to S2, and a sample corresponding to an immediately following local extremum and/or a sample corresponding to an immediately preceding local extremum corresponds to 1),

- the samples of the digital acquisition signal (12) are not preserved in the synthetic digital signal (13) if they do not correspond to 1) or 2). - Data compression method (10) according to claim 1, in which the modification in case 2) is a deletion of the measured amplitude of the sample considered.

3 - Data compression method (10) according to claim 1, in which the modification in case 2) is a replacement of the measured amplitude of the sample considered by an amplitude of predefined value Vd between S1 and S2.

4 - Data compression method (10) according to one of the preceding claims in which a raw digital signal (11) is low-pass filtered (20) to produce the digital acquisition signal (12).

5 - Data compression method (10) according to one of the preceding claims, wherein the synthetic digital signal (13) is compressed without loss (40).

6 - Data compression method (10) according to claim 5, wherein the lossless compression (40) of the synthetic digital signal (13) implements Huffman type coding.

7 - Reconstruction method (50) for producing a digital analysis signal (16) from a synthetic digital signal (13) consisting of samples, each sample being defined by a sampling instant and an amplitude, said reconstruction method being characterized in that it comprises at least one interpolation step (70) in which an interpolation function is required to pass through each sample with a horizontal tangent, and in which the amplitude of the digital signal analysis (16) is forced to a predefined value Vm between two consecutive samples of amplitude of value Vm of the synthetic digital signal.

8 - Reconstruction method (50) according to claim 7, wherein in the interpolation step (70), an amplitude of value Vm is previously associated with each sample of the synthetic digital signal (13) not including amplitude associated, or we replace beforehand each amplitude of said synthetic digital signal of value equal to a predefined value Vd by a amplitude of value Vm.

9 - Reconstruction method (50) according to claim 7 or 8, wherein a compressed digital signal (14) is decompressed without loss to produce the synthetic digital signal (13).

10 - Reconstruction method (50) according to one of claims 7 to 9, in which the interpolation function is obtained between two consecutive samples not both of amplitude of value Vm by implementing a cubic interpolation spline by Powell-Sabin elements.

11 - Non-destructive ultrasonic control system comprising a probe (1) connected to a generator/receiver (2), an acquisition device (4) and an analysis device (6), said non-destructive control system destructive by ultrasound being characterized in that said acquisition device comprises: - means for sampling a centered pseudosinusoidal analog signal supplied by the receiver (2) into a digital acquisition signal (12),

- lossy compression means (30) of the digital acquisition signal (12), which lossy compression means: o identify the samples of the digital acquisition signal (12) corresponding to local extrema, o compare each sample identified as local extremum at two thresholds S1 and S2, S2 being less than S1, and retain the sample considered in a synthetic digital signal (13):

2) after modification if the measured amplitude of the sample considered is less than or equal to S1 and greater than or equal to S2, and a sample corresponding to an immediately following local extremum and/or a sample corresponding to an immediately preceding local extremum corresponds to 1), o do not retain the samples of the digital acquisition signal (12) in the synthetic digital signal (13) if they do not correspond to 1) or 2), and in that the analysis device (6) comprises means of interpolation (70) of the synthetic digital signal (13) producing a approximation of the digital acquisition signal

(12), which interpolation means require an interpolation function to pass through all the points of said synthetic digital signal with a horizontal tangent. - Ultrasonic non-destructive testing system according to claim 11, in which the modification in case 2) is a deletion of the measured amplitude of the sample considered. - Ultrasonic non-destructive testing system according to claim

11, in which the modification in case 2) is a replacement of the measured amplitude of the sample considered by an amplitude of predefined value Vd between S1 and S2. - Ultrasonic non-destructive testing system according to one of claims 11 to 13, in which the acquisition device (4) comprises lossless compression means (40) of the synthetic digital signal

(13) forming a compressed digital signal

(14) and the analysis device (6) comprises lossless decompression means (60) of said compressed digital signal. - Ultrasonic non-destructive testing system according to one of claims 11 to 14, in which the acquisition device (4) and the analysis device (6) are connected by a communication network (8).