The use of a laser Doppler technique in vibrometers for measuring movement of a target is well known. The laser Doppler technique is based on using interferometric methods of observing the frequency shift of a laser beam, as it is scattered by a target comprising a moving surface. The frequency shift is imposed on the scattered component of the light by virtue of the movement of the moving surface relative to the incident laser beam.

There are a variety of interferometer configurations, which can be used in vibrometers; commonly available commercial instruments are based on either the Michelson or Mach-Zender arrangements.

In the Michelson interferometer illustrated in Figure 1, there is provided an input arm 130, a reference arm 131, a measurement arm 132 and an analysis arm 133. The laser 101 in the input arm 130 projects a beam (BEAM) of single frequency (which may be in the visible domain, or may be of infra-red or ultra-violet frequency) into a polarising beamsplitter 102. The beam is divided between the transmitted reference beam (REF) in reference arm 131 and the reflected measurement beam (TEST) in the measurement arm 132. (Note that the alternative arrangement is equally valid, that is the reference beam (REF) is reflected by the polarising beamsplitter 102 and the measurement beam (TEST) is transmitted. The illustrated arrangement has been selected as an illustration of the technique). The reference beam (REF) passes through a quarter waveplate 103, which is aligned so as circularly polarise the transmitted beam. The reference beam then strikes a reflector 104, which sends the reference

beam back along its original path. As the reference beam passes through waveplate 103 a second time, it is repolarised in the orthogonal direction to the original beam.

The reference beam is then reflected efficiently by beamsplitter 102 to the analysis arm 133.

In the measurement arm 132 of the interferometer the measurement beam (TEST) passes through a telescope arrangement 105, which expands the beam and focuses it on a target 106, via a quarter waveplate 107. Where the polarisation is preserved in the backscattered beam, the waveplate 107 acts in exactly the same way as waveplate 103. Some of the light that is scattered by the target 106 passes back through the waveplate 107, is collected by the telescope 105, and is then transmitted by beamsplitter 102 to the analysis arm 133.

In the analysis arm 133 the measurement and reference beams (TEST and REF) are recombined by beamsplitter 103 to provide a combined beam (COMB) of which the measurement (TEST) and reference (REF) beams are component beams. The combined beam (COMB) is output from the beamsplitter. The component parts (REF) and (TEST) of the combined beam are coaxial and have the same beam profile, but with orthogonal planes of polarisation.

For vibrometry we require to be able to determine the direction of motion (towards or away) of the target 106. With the Michelson design, a well-known technique is to generate two interference signals in quadrature whereby to determine the direction of motion of the target.

Consequently, in the analysis arm the combined beam (COMB) from beamsplitter 102 is split into identical parts in two channels A and B using non-polarising beamsplitter 108.

A phase shift is introduced on channel A (could be either channel). A quarter wave- plate 109 is placed in channel A, aligned with the polarization axes of the combined beam (COMB) so as to retard the measurement beam component (TEST) relative to the reference beam component (REF) (or vice versa) in that channel. This approach

has been referred to as passive phase shifting. Additional polarising components 110 and 111 are required in the subsequent optical path of channels A and B respectively to resolve common polarisation components of the reference (REF) and measurement (TEST) beam components so that interference can be generated. Where, as shown, a polarising beamsplitter is used, both the reflected and transmitted components will be able to generate interference patterns (fringes) at detectors 112,113,114 and 115. The passage of fringes with respect to the detectors 112-115 provides a measure of movement of the target in terms of the wavelength of laser (i. e. coherent) light.

Comparing between the transmitted and reflected beams from each polarising beamsplitter, the interference patterns have a phase difference of 180° to each other; electronic subtraction is used to suppress common mode noise, such as laser noise, while doubling the size of the signal.

It is possible to dispense with the phase shifting waveplate, and generate directional information by superimposing a fixed frequency shift (sometimes referred to as active phase shifting) on the reference channel. While the Michelson interferometer could be used, a Mach-Zender interferometer is a more convenient arrangement and is illustrated in Figure 2.

The beam from the laser 201 is split by beamsplitter 202. The reference beam (REF) is reflected by beamsplitter 202, and then by mirror 203. It passes through a Bragg cell 204, and this adds a frequency shift, typically between 40 and 120 MHz for practical reasons.

The measurement beam (TEST) is transmitted by beamsplitter 202 and beamsplitter 205, before being expanded and focussed on the target 206 by the telescope 207.

Some of the light scattered by the target is collected by telescope 207, re-collimated, and then reflected by beamsplitter 205.

The reference beam and measurement beam are then re-combined at the beamsplitter 208 to provide a combined beam (COMB) ; two interfering beams (which have components of both measurement and reference beams) are output from this

beamsplitter onto detectors 208 and 209. Clearly, polarisation-preserving components and wave-plates or similar may be used to increase efficiency.

Where active phase shifting is used as shown in Figure 2, the frequency of the measurement beam reflected by the target will increase and decrease as the target moves according to the Doppler equations in the same way as for passive shifting.

However, since the Bragg cell has shifted the frequency of the reference beam, an interference between the two beams will result in a beat frequency when the test object is at rest that equates to the frequency shift imposed on the reference beam. As the test object moves towards and away from the instrument, the beat frequency will rise and fall about this central point. The frequency shift applied by the Bragg cell to the reference beam must be set at a level that is higher than the maximum frequency shift on the measurement beam due to test object movement for the velocity range of the instrument. As a result, the beat frequency does not fall to zero, and only a single channel is required to determine direction of movement.

In both types of instrument, a measurement is made at a single point on the target.

Variants exist that move the point to be measured via a pair of scan mirrors, so that over a period of time, a representation of a vibrating surface can be built up. This technique is appropriate for analysing continuous cyclic motions, where phase may be measured relative to the driving waveform.

A limitation of scanning vibrometers is seen in cases where the motion of the test surface is non-repetitive. This is found in many situations, for instance in self-excited structures (e. g. brake callipers, civil engineering components such as bridge structures), in structures exhibiting non-linear differential response to a driving waveform (e. g. movements across rubber seals), or in visualizing the propagation of the transient response to an impulse. Such a surface may be monitored by employing a number of single point laser Doppler instruments ; this however, becomes unattractive when a large number of instruments must be set up to record the information. There may also be difficulties in capturing the vibration information from many points concurrently in a useful form.

According to a first aspect, the present invention provides a method for processing cyclic interferometer phase information into target displacement information comprising sampling the cyclic interferometer phase information at a first clock frequency (note that the first clock frequency determines the maximum surface velocity of the target that the apparatus can handle), and encoding the said information for analysis at a second lower clock frequency (note that where the displacement is a vibration, the second lower clock frequency is determined by the maximum vibration frequency to be measured).

Said phase information is preferably gathered from a vibrometer measuring movement of a target. Preferably the encoding is performed by dedicated electronics, and preferably the analysis of the encoded information is carried out in a computer.

Said phase information is preferably gathered from. a vibrometer measuring movement of many points of a target simultaneously.

Preferably said phase information is gathered by carrying out a plurality of comparisons of pairs of signals in quadrature (e. g. where one signal has an instantaneous amplitude of A and the second B, providing an indication of the results of comparisons A > B, A > 0, A >-B, B > 0), and from these comparisons determining discrete phase values by logical combination.

Preferably said encoding includes the step of subtracting successive discrete phase values to give transitions, and summing the transitions for a predetermined sampling period, and the step of analysing the encoded information includes the step of analysing the summed transitions.

Preferably, said encoding includes recording the time of the last transition within each sampling period, and the step of analysing the summed transitions includes correcting for the corresponding transition times.

The present invention also provides according to a second aspect of the invention a method of detecting phase values from two signals including the phase value to be

determined, said two signals being in quadrature, said method comprising the steps of carrying out a plurality of comparisons of the pair of signals in quadrature and from these comparisons determining discrete phase values by logical combination.

Preferably one signal of a pair has an instantaneous amplitude of A and the second signal of a pair has an instantaneous amplitude of B, the method including the steps of providing an indication of the results of comparisons A > B, A > 0, A >-B, B > 0.

Preferably said phase information is gathered from a vibrometer measuring movement of a target surface and in a yet preferred arrangement, said phase information is gathered from a vibrometer measuring movement of many points of a target surface simultaneously. In the latter case, each point of the target surface may provide a respective pair of said signals in quadrature.

Preferably the method includes the further step of subtracting successive discrete determined phase values to give transitions, and summing the transitions for a predetermined sampling period. It may also include the step of recording the time of the last transition within each sampling period, and analysing the summed transitions by including correcting for the corresponding transition times.

Preferred embodiments of the invention will now be described by way of example only and with reference to the accompanying drawings in which: Figure 3 is a diagrammatic representation of a laser Doppler interferometer in accordance with a preferred embodiment of the invention, Figure 4 is an alternative of part of Figure 3, Figure 5 is a diagram showing a method of providing a digital representation of the phase of a pair of analogue signals, Figure 6 is a diagram showing the analysis of the digital signals provided by Figure 5,

Figure 7A and 7B show diagrammatically (Figure 7A in plan view and Figure 7B in side view) a laser Doppler interferometer comprising a second embodiment of the invention, Figure 8A and 8B show diagrammatically (Figure 8A in plan view and Figure 8B in side view) a laser Doppler interferometer comprising a third embodiment of the invention, and Figure 9 is a diagram of sampling periods and transitions In the instrument of Figure 4 onwards, the limitations of the prior art described above are reduced or overcome, as the motion of a number of points on the test object are measured concurrently. The example illustrated features 16 points arranged in a single line. In theory, any reasonable number of measurement points could be generated in any spatial arrangement required; practical limitations would arise from the number of channels that could be processed.

In Figure 3 light beam (BEAM) from a single source laser 301 is expanded by telescope 302 and directed onto a diffractive optical element 304 via a half-wave retarder 303. The retarder 303 is mounted such that it can be rotated about the incident beam axis (i. e. the axis of the beam received from telescope 302); this enables the polarisation vector to be set to any required orientation. The diffractive optical element 304 splits the incident beam into 16 beams (BEAM 1-16), each of approximately equal strength, and with a constant angle of separation between each successive beam. In this way, the output beams form a fan in a single plane. Other arrangements are possible; this physical embodiment is chosen as the fan pattern then maps to a commercially available detector unit. The fan pattern may also be deflected as a whole relative to the previous optical axis so that the zero order beam can be conveniently blocked.

The fan of beams (BEAM 1-16) propagates forward, and is directed into a lens system 305. The lens system 305 collimates the axis of each of the beams, whilst bringing each individual beam to a focus at intermediate image planes 307 and 310.

Perfect collimation of the axes of the beams is not essential to the operation of the instrument, however it simplifies other aspects of the system, including obtaining an accurate quadrature relationship between the channels.

A polarising beamsplitter 306 is placed between the lens system 305 and the transmitted intermediate image plane 307. The orientation of the half-wave retarder 303 described earlier enables the relative strengths of the output beams to be varied.

In the arrangement illustrated, the transmitted path through the beamsplitter 306 is used for the reference arm 331, and the reflected path is used for the measurement arm 332. The choice is made for convenience of physical layout; the opposite arrangement would be equally valid.

The reference arm 331 is a conventional arrangement with quarter-wave retarder 308 and reflecting surface 309. The retarder 308 is set such that the beams (REF 1-16) in the reference arm propagating through and onto the reflector are circularly polarised.

The reflector 309 directs the beams (REF 1-16) back along their incoming path, while the action of the retarder 308 on the return path returns the beams (REF 1-16) to linear polarisation, but with the polarisation axis orthogonal to the incoming beams. This arrangement ensures that the reference beams (REF 1-16) are then efficiently reflected into the analysis arm 333 of the interferometer.

The beams (TEST 1-16) in the measurement arm 332 reach a reflected intermediate focal plane 310 and then propagate forward through a quarter waveplate 311 and an objective lens 312. The order of these components can be interchanged. The objective lens 312 relays the 16 measurement beams (TEST 1-16), so that a new image of the intermediate image plane is formed on the test object 313. (It will be understood that the 16 beams strike the test object 313 in a predetermined pattern of 16 points 333 across the object. The points 333 are arranged in a straight line but may be arranged in other formations, for example, in a formation of 4 x 4). If the objective lens has a zoom capability, the spacing between the individual points 333 on the target can be varied. The waveplate 311 is arranged so that the measurement beams (TEST 1-16) are circularly polarised. A proportion of the light from the each of the measurement

beams (TEST 1-16) is backscattered, returning through objective lens 312, and waveplate 311, and the pattern is refocused at the intermediate image plane 310. The reflected portion of the measurement beams (TEST 1-16) continue to propagate through the polarising beamsplitter 306, where they are transmitted, and continue along the analysis arm of the interferometer. It should be noted that the corresponding measurement and reference beams in the analysis arm 303 are co-axial, but of orthogonal polarisations.

Where the measurement beams (TEST 1-16) are arranged side-by-side in a single plane, some measurements may require that the plane of projected fan of measurement beams be rotated, so that the vibration pattern along a different section can be analysed. Since the instrument may be fairly bulky, an image rotator can be interposed between the polarising beamsplitter 306 and the retarder 311. One arrangement makes use of a pechan prism 401, shown as a detail of figure 3, in figure 4. The prism 401 is mounted such that the propagating light beams pass through the centre, and the prism rotates about the optical axis. The projection array will rotate at twice the rate of the prism, so a 90'rotation of the prism will result in a t180° rotation of the plane. To function correctly in this context, it is necessary to take polarisation into account. The plane of polarisation must be held constant relative to the reflecting planes of the prism 401, so a half-waveplate 402 is placed between the polarising beamsplitter 306 and pechan prism 401, and rotated at half the rate of the prism 401. This arrangement is chosen because the plane of polarisation will rotate at twice the rate of the half-waveplate as the light passes through it. Given that the polarisation is held at a constant angle relative to the prism 401, the quarter waveplate 311 earlier is fixed to the pechan prism 401, such that circularly polarised light is maintained in the measurement region. Note that as the light is reflected back from the target, the processes will be reversed, so that the returned array will be de-rotated, and so that the image is correctly mapped to the detectors..

After reflection, both measurement (TEST 1-16) and reference (REF 1-16) beams propagate back through the polarising beamsplitter 306. In this implementation, passive phase shifting is used, so each of the sixteen combined beams (COMB 1-16, where COMB1 is a combination of TEST1 and REF1 ; COMB2 is a combination of

TEST2 and REF2; etc) in the analyser arm are split into two channels A and B using a non-polarising beamsplitter 314. A phase shift between the two interference signals is applied by passing one channel A through a retarder 315, aligned with one of the polarisations. The retarder 315 may be a quarter waveplate, although the value of retardance required will be dependant on any phase differences imposed by reflection or transmission of either of the polarisations by the non-polarising beamsplitter 314.

Passive phase shifting is clearly preferable here, as it is more practical to pass a number of beams through a waveplate than through a Bragg cell.

Because the light in each of the channels A and B is diverging on the return path from the intermediate image points, it is necessary to re-image the points 333 onto the respective elements of the detector arrays, 316 and 317. Each detector array comprises sixteen photo detectors set out in a line. Conventional lens systems 318 and 319 are used for this purpose. It is preferable to place the lenses after the non-polarising beamsplitter 314, because after passing through the lenses 316/317, the ray families from the different measurement points are at different angles one to the next. If the light paths were to pass through the beamsplitter 314 after the lens, a variety of phase differences between the measurement points would be seen, as a result of the different angles. This would compromise the quadrature between the channels, and lead to difficulties with the processing of the signals. For the same reason, dielectric surfaces are to be avoided on any mirror surfaces be used to fold the optical path after the relay lens 318,319.

The combined beams (COMB 1-16) in each of channels A and B pass through respective polarising components 320 and 321; these are arranged to resolve a measure of both the measurement (TEST 1-16) and reference (REF 1-16) components of the combined beams (COMB 1-16) into a common polarisation so that interference can occur on the surface of the detectors.

There is an additional detector 322, which receives a small part of the direct laser output beam (BEAM) that is reflected by a beamsplitter 323. This enables the laser power and noise to be continuously monitored, and serves as an input to the processing electronics.

In the arrangements of Figures 7 and 8, the components and operation of the system are identical to those of Figures 3 and 4 up to the point where the measurement (TEST 1-16) and reference (REF 1-16) beams remerge through the beamsplitter 706 as combined beams (COMB A 1-16) and (COMB B 1-16).

In a first arrangement, instead of a non-polarising cube beamsplitter 314 of Figures 3 and 4 dividing the beam in the analyser arm into two orthogonal channels A and B, a dim-active optical element (DOE) 703 comprising a block divides the input combined beam (COMB 1-16) which is incident thereon into two separated beams (COMB A 1- 16) and (COMB B 1-16). Said block carries a diffraction grating on an external ? surface through which the combined beam (COMB 1-16) passes which can be either the surface through which the beam passes into the block or the surface through which the beam leaves the block. Clearly, a diffraction grating formed on the surface of a component is a more physically stable and robust structure than a multilayer coating sandwiched between two relatively large pieces of glass, and the interaction between light and grating will be more stable.

Diffractive components on the surface of blocks are well known for providing a variety of optical effects including providing a fan of separated beam patterns. The diffraction grating may be provided by embossing a thermoplastic material at temperature using a master mould, the thermoplastic material being attached to the surface of the block, or may be provided by ultraviolet embossing of the block itself where it comprises a fused silica or glass substrate or may be carried out by means of an etching process.

After exiting polarising beamsplitter 301, the combined beams (COMB 1-16) strike half wave retarder 702. The orientation of the retarder allows the polarisations of the combined beams (COMB 1-16) propagating onwards to be rotated; in this arrangement it would be convenient to set the retarder such that the polarisations are rotated by nominally 45°.

After passing through the retarder 702, the combined beams (COMB 1-16) are incident on to the diffractive optical element (DOE) 703 where they are split into two separated combined beams (COMB-A 1-16 and COMB-B 1-16) in channels A and B respectively, each comprising equal intensities of both reference and measurement beam components (REF 1-16) and (TEST 1-16) respectively, and with unchanged polarisation. The split is arranged so that the separated combined beams (COMB-A 1- 16 and COMB-B 1-16) diverge from one another above and below the original axis of propagation, as shown in the side view of Figure 7B.

The action of splitting the combined beam using a diffractive optical element results in a smaller divergence (5° to 10°, typically 7°) between the two output beams (COMB A 1-16) and (COMB B 1-16) than the 90° that results from a beamsplitting cube 114 as shown in Figures 3 & 4. The smaller divergence allows a single lens 704 to be used to re-collimate the two separated beams (COMB A 1-16) and (COMB B 1-16). Furthermore, with both sets of separated beams (COMB A 1-16) and (COMB B 1'-16) (i. e. channels A and B) at a controlled separation, a single, larger polarising beamsplitter cube, 708, may be used rather than one for each channel. The design allows sufficient space between the channels A and B, and between the lens and the polarising cube to allow the insertion of the phase-shifting quarter waveplate 709 in one channel (A) only.

A single lens system 704 which comprises an optical component common to both channels A and B is placed near to the diffractive optical element 703. The lens 704 receives both separated combined beams (COMB-A 1-16 and COMB-B 1-16) and forms a second image of the intermediate image plane 705 on two sets of detectors 706 and 707, through a polarising component 708. The divergence caused by diffractive optical element 703 leads to the formation of two images, vertically displaced (see Figure 7B); these are the two channels that are required for the signal processing. At a point where the separated combined beam (COMB-A 1-16 and COMB-B 1-16) are well separated vertically, a quarter wave retarder 709 is placed in one separated combined beams (COMB-A 1-16) so as to achieve interference patterns in the two separated combined beams (COMB-A 1-16 and COMB-B 1-16) that are in quadrature. If required, a compensating block 710, of the same thickness as the

quarter wave retarder can be placed in the unretarded separated combined beam (COMB-B 1-16) so that a common focal plane can be achieved. A single polarising beamsplitter cube 708 is used to resolve the common components of polarisation of the measurement and reference components of the beams in each channel. To increase the signal levels we could either add detectors to receive the reflected beams Whereupon it would also necessary to double some of the electronic circuitry and modify the laser power monitoring and noise offset..

Clearly, it is possible to eliminate the half wave retarder 702, and rotate the polarising beamsplitter 709 so as to resolve components of both measurement and reference beams. An illustration of this arrangement is given in Figure 8A and 8B, where the rotated polarising component is shown at 801 (the remaining components retaining the reference numerals from Figure 7). This is a simpler arrangement, although if a second detector array is used, the mounting of the detector at an angle may be more problematic than the installation of the half wave retarder.

The benefits of this arrangement over Figures 3 and 4 are as follows. The layout of the multipoint vibrometer is simplified. Only one lens system and polariser is used between the beamsplitter 701 and detectors. The non-polarising beamsplitter 114 is eliminated; this can be a problematic component due to its inherent complexity.

Furthermore, certain designs of non-polarising beamsplitter can cause phase errors between the phase of the transmitted and reflected S and P polarisations. This can lead to a situation where the phase-shifting waveplate 115 must be specifically selected to match the characteristics of the non-polarising beamsplitter. The diffractive optical element allows better control over this factor.

We will now describe a technique for compensating for laser noise. In a simple interferometer, a measurement beam and a reference beam interfere at a detector. The detector responds to the light intensity (I), which is the square of the combined amplitude (a): a = R cos (ot + M cos (mt + (p)

I = a2 = R2 cos2 cat + 2RM cos cat cos (wt + (p) + M2 cos2 (wt + cp) mean (I) = (R2 + M2)/2 + RM cos p where R is the amplitude of the reference beam, M is the amplitude of the measurement beam, w is the angular frequency of the light and (p is the phase of the measurement beam (which changes with target position).

If the reference and measurement beams can be arranged to have similar amplitudes, the fringe contrast is high, that is, the ac component is not much smaller than the dc component.

However, if the target is at long range and is not particularly reflective, the measurement beam intensity may be greatly reduced. Corresponding reduction in the reference intensity does not help, because as the optical gain reduces the signal disappears in noise (due to thermal noise in resistors and amplifier input noise).

If the light source were completely stable, a small ac component could be extracted by using a high-pass filter to block the large dc component. Unfortunately, only a small amount of laser noise in the frequency band of interest will swamp the signal.

Current single-channel vibrometers (e. g. the Ometron VH300) achieve high sensitivity by subtracting signals from a pair of detectors, which sense interference patterns in orthogonal polarizations. The signal component is arranged to have opposite phase at the two detectors so that the signal reinforces as the background cancels. The vibrometers therefore require 4 detectors, two for each Doppler signal.

This arrangement allows satisfactory operation with a measurement beam amplitude of less than 10-3 of the reference beam (intensity ratio 104).

Scaling up this technique to a multi-channel system would require 4 arrays of detectors, which would be expensive and cause problems in optical/mechanical design and alignment.

A preferred aspect of the present invention deals with these problems by using a single element reference beam detector to back off the part of the dc component on each of the channels due to the reference beam (R2/2). The part of the dc component due to the measurement beam (nu/2) can be ignored for M << R. Consequently only one detector array is required for each Doppler signal. The basic assumption is that the reference beam illumination of the detector arrays is uniform; however, even a 5% non-uniformity still reduces the effect of laser noise by a factor of 20. This method gives sufficient optical sensitivity to allow vibration measurement of unprepared surfaces at moderate distances.

We use either a separate sensor or a dedicated element at one end of the detector array (so, in Figure 3, for example, the end detector of the 16 detectors could be used for this purpose). In the latter case, light going to this reference detector element would be masked to block the measurement beam at the intermediate image plane (310) in the measurement beam; however this is not our preferred method because it sacrifices one measurement point. In a system which uses CCD arrays instead of photo-diode arrays, the reference pixel is at the read-out end, so that its voltage is held and subtracted from all subsequent pixels on the line as they are clocked out of the device.

Processing of detected signals We will now describe the processing of the detected signals (ie the signals from the thirty two detectors of Figures 3 and 4 or Figures 7A, 7B, 8A, 8B).

In a conventional single beam vibrometer such as Figure 1 using passive phase- shifting, a fixed optical phase shift of 7r/2 is established between the interference fringes reaching two detectors. This means that a target surface moving at constant velocity generates a simultaneous cosine/sine pair of input ("Doppler") signals.

Conversion of these signals to an analogue of the velocity of the surface is performed by mixing with a higher frequency carrier, followed by demodulation. In the analyser, a mixer implements the equation:

cos (C-D) = cos C cos D + sin C sin D where cos (D) and sin (D) are the Doppler (detected) signal pair and cos (C) and sin (C) are the carrier signal pair. Demodulation uses a frequency-to-voltage converter, which operates by transferring a fixed amount of charge at each zero crossing of the waveform.

The expected approach to signal processing in a multi-channel vibrometer (e. g. as shown in Figures 3,4,7A, 7B, 8A, 8B with 16 channels) would be to replicate the above analogue processing for each channel, requiring a large quantity of electronics.

Furthermore, to function as a single instrument, rather than merely as a collection of single-point instruments, all the vibration data must be fed to a digital computer for combined analysis and display. This would require fast high-resolution analogue-to- digital conversion synchronized for all the channels.

We have realized that it is necessary to provide the initial signal processing at a high clock frequency to keep up with the maximum rate of fringe passing, but it is only necessary to provide the later stages of signal processing at a lower clock frequency sufficient to represent the highest vibration frequency of interest. Given the reduction in clock frequency,-the signals can be handled by a computer such as a personal computer. We have developed a method of such signal processing.

Broadly the method comprises processing the cyclic interferometer phase information from the interferometer (which as has been described looks at many points across a target) into target displacement information giving information about the movement (usually but not always vibration) of the target, comprising sampling the cyclic interferometer phase information at a first clock frequency, and encoding the said information for analysis at a second lower clock frequency.

The encoding is performed by dedicated electronics, and the analysis of the encoded information is carried out in a computer.

Said phase information is gathered by carrying out a plurality of comparisons of pairs of signals in different phase, for example, in quadrature (e. g. where one signal has an instantaneous amplitude of A and the second B, providing an indication of the results of comparisons A > B, A > 0, A >-B, B > 0), and from these comparisons determining discrete phase values by logical combination.

The encoding includes the step of differencing (i. e. subtracting) successive discrete phase values to give transitions, and summing the transitions for a predetermined sampling period, and the step of analysing the encoded information includes the step of analysing the summed transitions.

The encoding includes recording the time of the last transition within each sampling period, and the step of analysing the summed transitions includes correcting for the corresponding transition times.

The step of gathering the cyclic interferometer phase information has been described above with reference to the various preferred types of interferometer.

Sampling the interferometer phase information Each pair of input signals from respective pairs of detectors is first converted to a stream of digital phase values at a sample rate sufficient to track the highest velocity (ie fringe rate) specified for the instrument (at a clock rate of 10 MHz, for example).

For each of the 16 channels (derived from the output of the sixteen pairs of detectors), phase changes are accumulated, and sampled at a rate sufficient to represent the highest vibration frequency specified for the instrument (at a clock rate of 100 kHz, for example). The encoded data is multiplexed onto a 16-bit wide link to a personal computer, which captures the data and reconstructs the surface vibrations. The technique essentially measures displacement, rather than velocity, though of course velocity may be derived by the computer.

Phase determination

The first step is to convert pairs of analogue input signals A and B (i. e. signal A from a detector in channel A and signal B from the corresponding detector in Channel B) (see Figure 5) into a digital representation of the phase of the interference fringes. The higher the phase resolution, the better the ability to cope with vibrations which have peak-to-peak amplitudes of less than the size of a fringe, but the greater the complexity and cost of the electronics. Also it is less likely to be affected by variation in reflectivity and distance of surfaces. We use a set of four comparators to compare the two signals against zero and against each other; these comparisons are robust against change of signal amplitude.

As illustrated in Figure 5, a 4-bit twisted-ring code is obtained from the signals A and B using the comparisons A > B, A > 0, A >-B, B > 0; after sampling, the valid states are converted by Boolean logic to a 3 bit phase value, resolving the phase to one eighth of a fringe. [A simpler method, which resolves phase to only one quarter of a fringe, is also possible, using comparisons A > 0 and B > 0.] twisted-ring code binary phase value decimal phase P3 P2 PI PO D2 Dl DO 1111 100-4<BR> 1110 101-3<BR> 0 0 1 1 0-2<BR> 1 0 0 0 1 1 1-1<BR> 0 0 0 0 0 0 0 0<BR> 0 0 0 1 0 0 1 1 0 0 1 1 0 1 0 2 0 1 1 1 0 1 1 3 derivation of binary phase value: D2 = P3 <BR> <BR> D1 =P3 ^P1<BR> DO = P3"P2"PI"PO where A represents the exclusive OR operation Phase sampling

For a steady movement of the target surface the phase cycles continuously around its range of 2 7r, so the phase needs to be"unwrapped"to determine displacement. To provide enough data for this process, the input must be sampled at a sufficiently high rate that the phase change between samples is always within the (non-inclusive) range - jar to +x, i. e. the sample rate must be more than twice the maximum fringe rate. In a discrete phase system the requirement is more stringent. To avoid ambiguity in the 8 phase method described above with reference to Figure 5, the maximum allowed change is 3 phase steps per sample period (since 4 steps could be interpreted as either a positive or a negative movement). Hence the minimum sample rate is 8/3 times the maximum fringe rate.

In the interferometer, the passage of a single fringe represents a movement of the target surface of half the wavelength of light. A single phase step, or sub-fringe, is eight times smaller, giving a basic displacement resolution of about 33 nm, using a 532 nm green laser source. The encoder circuit schematic, Figure 6, shows a front-end sample rate of 10 MHz, which is sufficient to cope with a transient maximum velocity of almost 1 m/s. In practice, the size of the sub-fringe accumulator (see below) sets a limit of around 425 mm/s, and the bandwidth of the analogue input circuitry is set for 250 mm/s.

A noise rejection scheme may be implemented if the input signals are initially sampled at a higher rate than the required minimum.

Data encoding The phase samples can be passed directly to a computer for further processing.

However, the speed of the communication link and the load on the computer would limit the maximum velocity which could be handled.

Alternatively, and in accordance with a preferred aspect of the invention, we use the power of programmable logic devices to carry out phase unwrapping in hardware.

The phase totals only need to be sampled at a rate of at least twice the maximum

vibration frequency (in practice four times is used to give better reconstruction, particularly of velocities).

Hardware processing provides extra information, which information allows more accurate, lower noise, reconstruction of the vibration waveforms. Simply adding up discrete sub-fringes gives a reasonable representation of displacement, particularly if the amplitude is at least a few fringes, but gives a poor determination of velocity. This is because the actual distance moved within any period of time is always subject to a one sub-fringe uncertainty. This encoding technique supplements the count of number of sub-fringes within a sampling period with a time value for the last transition, and a direction flag to indicate whether that transition was approached from above or below.

In the current implementation, each output sample period is sub-divided into 100 input time periods, giving an effective local velocity accuracy of 1%.

Figure 6 sets out a schematic of the encoder circuit, which shows circuit for use with a 16-channel system of Figures 3 and 4 or 7 and 8. The diagram shows the data flow and operations, but leaves out internal clocks.

The section 50 within dashed lines is replicated for each of the 16 channels; i. e. there is provided a relevant section 50 for each of the 16 detector pairs.

The four signals PO, PI, P2, P3, derived as set out above for a particular pair of the 16 pairs of signals A & B, are input to the section 50 by inputs 51-54 to a digital filter 56 and the four signals are sampled by input latch 57 at the high 10 MHz rate. The signal from the latch 57 is converted to binary phase at converter 59. The 3-bit output of converter 59 is passed to a latch 61 which stores the preceding output value. The output 59 is also passed to a subtractor 62 which also receives an output from the last value latch 61 and subtracts the two values. The output from the subtractor 62 will be a difference value being a measure of the change of phase.

The output of subtractor 62 is passed to sub-fringe accumulator 63, which totals both positive and negative values for a period of time until cleared by a 100 kHz clock

signal on line 70 (actually a clock signal derived from the 10 MHz clock signal 58 counted in by counter 60).

The output of subtractor 62 is also passed to comparators 64,65 and 66.

If the change obtained by the subtractor 62 is non-zero (determined by comparator 65), the direction of change (determined by comparator 66) is stored in latch 71.

If the change is non-zero, the current time count is stored in latch 69. Thus at any one time, the latch 69 carries the time of receipt of the last signal from comparator 65 (note that the previous time is overwritten). The usefulness of this will be explained later.

There are two overflow conditions which set an error flag in error accumulator (67), (a) phase change too large (ambiguous value of +/-4 as determined by comparator 64) and (b) sub-fringe accumulator overflow (exceeded 8 bit signed range of-128 to +127).

Each of the latches 69 and 71 and accumulators 67,68 are driven by the 10 MHz clock, but, importantly, they are emptied by the 100 kHz clock signal on line 70 which is connected to each of them. On emptying, the signals stored in latches 69,71 and accumulators 67,68 are passed to output latch 72, driven by the 100 kHz clock signal. The output signal from latch 72 provides the data packed into a 16-bit word plus the error bit.

It should be noted that there is a relevant latch 72 for each of the 16 detector pairs.

The outputs of the 16 latches 72 are passed to multiplexer 73 which provides a 16 bit output plus an error state output under the control of output control 74. The output control 74 provides an output strobe output 77.

The output from the multiplexer 73 are passed to a 3 channel multiplexer 76 which provides the output data in 16 bit form to output 78 and thence to a personal computer where it may be analysed.

The direction-of-last-transition bit and the last transition time occupy the high byte and the signed sub-fringe displacement occupies the low byte. Special values in the time field (beyond the normal range of 0 to 99) are used to indicate either no transitions or error during the period, or to identify the start word.

Encoding summary bits field transition (s) no transition error code start code D15 last direction 0 (+) or 1 (-) 0 1 1 D14-D8 last time 0 to 99 127 127 112 D7-DO displacement-128 to 127 0-128 to 127 16 channels (Ox7FOO) (OxFF ??) (OxFO 10) The output sampling rate is 100 kHz, adequate to handle vibration frequencies up to 25 kHz, which more than covers the audio range. Every 10 ps the output multiplexer generates a start word (which contains a field specifying the number of channels) followed by 16 data words; for convenience the 17 words are strobed out at 500 ns intervals. The mean data rate into the PC is only 3.4 M byte/s, so it is easy to store recordings of motions lasting for many seconds. Note the contrast with the total input data rate to the encoder, which is 80 M byte/s.

We now return to consideration of the latch 69. At the end of the sampling period, the latch 69 contains a count indicative of the time of the last change. Consider, for example, a target moving at a low constant velocity, so that the rate of transitions is of the same order or less than the rate of sampling: although the transitions are at regular intervals, they appear irregularly in the sampling periods. This is illustrated in Figure 9. Successive sampling periods are shown as Sl, S2, S3, S4, etc. The transitions are indicated as T1, T2, T3 etc. Transition T1 appears early in the first sampling period Sl and successive transitions appear later in each successive sampling period. Thus in the sampling period S4, transition T4 appears late on. Indeed, the next transition T5 appears not in the sampling period S5 but early in the sampling period S6. Thus the sampling period S5 will have no transition. Thus although as is clear from Figure 9 the transitions are at regular intervals, the sampling periods Sl to S7 will show the

following irregular sequence of number of transitions per sampling period :- 1, 1,1,1, 0, 1,1.

By retaining a count which records the time at which the last transition (indeed the only transition in this case) occurs during each sampling period, one can reconstruct the regular nature of the transitions shown in Figure 9.

The processing for a 16-channel instrument fits into a single commercially available programmable logic device. This encoding scheme may therefore be extended to handle more surface points, higher velocities or higher vibration frequencies, without imposing an excessive load on the receiving computer.

Vibration reconstruction The counter/timer encoding technique described above yields a succession of accurate points on the displacement time graph for each channel. However these points are in general irregularly positioned in time, which is an unsuitable data set for subsequent processing and display. We need to reconstruct the displacement waveform at sample points equally spaced in time.

In a real situation, the target is subject to finite acceleration, so it does not move discontinuously and the displacement graph is smooth. Where the velocity is high enough to give at least one transition within each sample period, simple curve fitting works well. At lower velocities, and particularly around turning points in the displacement graph, there may be large gaps with no sub-fringe transitions. What is needed in these regions is a smooth and physically realistic interpretation of the data, so that the displacement graph remains within the correct sub-fringe box, while using the simplest (lowest order) interpolation equations to achieve this. The aim is to get best appearance of reconstructed waveforms and lowest noise in the spectra.

Reconstruction of the waveform is carried out as follows : Gradient values are estimated at each data point from the quadratic which passes through it and the data point on either side (this assumes locally constant acceleration). Within each section

(interval between two data points) a cubic curve is fitted which matches the two end points and two end gradients. Displacement at each sample time within a section is calculated using the cubic, and checked to ensure the value lies within the relevant sub-fringe box. If any interpolated points go outside the box, further work is required.

First a new point is inserted at the last sample time in the section (with an adjusted gradient), to prevent"corner-cutting"by the graph. A new cubic fit is calculated for the section using this point. If there are still overshoot points, a quartic term is added to deal with them.

The invention is not restricted to the details of the foregoing example.