EP0507829A1

EP0507829A1 - Active vibration reducing system.

Info

Publication number: EP0507829A1
Application number: EP91901848A
Authority: EP
Inventors: Colin Fraser Ross
Original assignee: Noise Cancellation Technologies Inc
Current assignee: Noise Cancellation Technologies Inc
Priority date: 1989-12-30
Filing date: 1990-12-24
Publication date: 1992-10-14
Anticipated expiration: 2010-12-24
Also published as: JPH05508233A; DE69029044D1; EP0507829B1; ES2095311T3; WO1991010226A1; DE69029044T2; GB8929358D0; ATE144854T1

Abstract

An active vibration reducing system for controlling the vibration generated by a source of periodic or quasi-periodic vibration is characterised in that a weighted sum of the output of at least two sinusoidal waveform generators each synchronised to the frequency of the source of original vibration to generate harmonics of the frequency of the source, is supplied to at least one actuator (5) which generates additional vibration, sensor means (6) is used to monitor the effect of the additional vibration on the original vibration and the weighting of the sum of the outputs is adjusted so that within a defined region the amplitude of the combined vibration is reduced.

Description

Title; Active vibration reducing system

Field of the Invention

The invention relates to the control of vibration by active means. The basic principle of active vibration control is to introduce an additional vibration of the same size but in anti-phase to the original so that when the original and its antidote combine the result is no vibration. This principle and thus the invention can equally be applied to sound, noise and many other wave¬ like disturbances. For sound, the anti-phase sound would normally be produced by conventional loudspeakers positioned appropriately to provide some region where the sound and anti-sound cancelled each other.

Background to the Invention

The control of noise and vibration by conventional passive means tends to be ineffective at low frequencies where bulk and weight limit the amount of treatment than can be applied. Active vibration control has been suggested as having the potential to provide good vibration reductions at low frequencies and many patents have been taken out to cover various ideas. The earliest patent being in 1936 by Lueg, with increasing numbers from around 1976. One area which has received much interest is the control of repetitive or periodic noise. This appears to be because in many industries and passenger transport systems, noise or vibration, an unwanted disturbance, is created by rotating machines and because the technical difficulties in controlling periodic vibration are less than for random vibration. One of the earlier publications describing a successful experiment in this field was by KIDO in 1973 (KIDO in "Automatic Control of Acoustic Noise Emitted from Power Transformers by Synthesising Directivity" published as a report of the research institute of Electrical Communications, Tohoku University.), and one of the earlier patents was filed by Connover in 1957 US2776020. Since then much work has been done but very few commercial systems have been made available. This appears to be because there are so many technical choices to be made and so many possible ways of achieving some form of control but as yet the limitations of systems that have been selected have restricted their success.

One area which has turned out to be of major importance is the method by which the cancelling vibration produced by the control system can be adapted quickly to changing vibrations from the source or to changing speeds of the source. In order to be able to make the appropriate changes sensors are positioned in the region where quiet is required and information from their output is used to update the cancelling vibration.

This present invention is believed to overcome many of the limitations of previously described systems, and is particularly effective at responding to rapid changes.

General Aspect of the invention

According to the invention, an active vibration reducing system for controlling the vibration generated by a source of periodic or quasi-periodic vibration is characterised in that a weighted sum of the output of one or more sinusoidal waveform generators each synchronised to the frequency of the source of original vibration and generating harmonics of the source frequency is supplied to actuators which generate additional vibration and that sensor means are used to monitor the effect of the additional vibration on the original vibration such that within a region the amplitude of the combined vibration is reduced. Preferably the sinusoidal waveform generators are synchronised to the source of vibration by a synchronising signal which comprises an integral number of pulses per fundamental period of the vibration to be controlled, and preferably the sinusoidal waveform generators comprise computer memory which stores fixed data corresponding to the sampled oscillator output. The output means which form the output signals from a weighted sum of the outputs of the sinusoidal waveform generators do this by multiplying the individual outputs by a different variable output coefficient and adding the result of all the multiplications together. For each actuator receiving a signal from the output means the variable coefficients are different and independently variable. Variation of these output coefficients allows the vibration from the actuators to be controlled.

A set of sensors is used to monitor the effect of the actuator generated vibration on the original vibration and input means are provided to process the signals from these sensors to generate input coefficients which relate to the amplitude of the vibration at the harmonics produced by the sinusoidal waveform generators. The values of the output coefficients are adapted in accordance with an algorithm in response to the measured input coefficients by an adaption means. The adaption algorithm aims to minimise a weighted sum of the squares of the input coefficients whilst constraining the amplitude of the output coefficients. The weightings for and the positions of the sensors being chosen to allow the system to reduce the vibration in the desired region. Factors in the algorithm depend upon the frequency of the fundamental frequency of the source of vibration and so the adaption means is provided with a frequency signal. Advantageously the sinusoidal waveform generators, output means and input means are implemented in one digital computational unit with associated memory for data and programme. The adaption means can be implemented in a separate computational unit sharing the data memory or in the same unit.. Alternatively, the input means, output means and sinusoidal waveform generators can be analogue circuitry being controlled by the adaption means implemented in a digital computational unit with associated memory for data and programme.

Discussion of selected prior art

GB2054999 proposes a signal processing system which is a bank of filters centred on multiples of a fundamental frequency. Each filter has two outputs, an in-phase and a quadrature one. The output of each filter is amplified by an appropriate gain and then the output summed. Whilst it is mentioned that, for repetitive sources, the fundamental frequency can be synchronised to the repetition rate in order to match the peaks of the filter to the peaks in the vibration spectrum and whilst it is also mentioned that the input can be derived directly from the source the signal processing system described requires an input to be filtered by the parallel bank of filters. It is thus fundamentally different in nature to the present invention which is combining the output of sinusoidal waveform generators whose frequencies are synchronised to the fundamental frequency or harmonics of the source.

PCT/GB87/00706 proposes an active vibration control system which is different from the present invention in two fundamental respects 1) the control system operates on a real time base and thus where internal digital oscillators are used as sinusoidal waveform generators their frequency needs to be constantly adjusted to match the source of vibration frequency and 2) the system operates by filtering the input signal.

GB1577322 proposes a noise control system which operates on a time .base synchronised to the fundamental frequency of the vibration but the whole of the output signal corresponding to one cycle is stored and output repetitively. This method of storage is not suitable for rapid changes in the output signal.

Applications and further aspects of the invention

The present invention is concerned with an active vibration reduction system which can be used to reduce vibration produced by a source which generates the vibration in a periodic or quasi-periodic manner.

Sources of this type include i) propellers on aircraft which generate noise inside the cabin, ii) propellers in ships and submarines which generate sound in the water iii) diesel generators which produce vibration of the supporting structure iv) exhaust systems of petrol and diesel engines which produce sound v) centrifugal and axial fans which produce sound vi) car engines which generate sound inside the passenger cabin vii) transformers

The characteristics of all of these sources is that the vibration (which is called sound when air is vibrating) is produced at one or more harmonics of a fundamental frequency. Here the term harmonic is defined so that the first harmonic is the fundamental, consequently the term 'harmonic* includes the fundamental. If the fundamental frequency is f then sound is produced at one or more of the frequencies corresponding to f, 2f, 3f, 4f, 5f, 6f, 7f, 8f, etc. The fundamental frequency is normally a multiple or sub-multiple of the rotation rate of the source where the source is a rotating machine.

In the case of propellers, where there may be 4 identical blades on one shaft, most of the noise is produced at the blade passing frequency (ie 4 times the shaft rate) and its harmonics. In the case of a car engine most of the noise is produced at the firing rate which is at twice the crankshaft rotation rate in a four cylinder, four-stroke engine.

In most of the cases described above the speed of rotation of the source will change with time. This change in speed changes the frequency of the vibration produced. The active vibration reduction system which is the subject of this invention derives a speed signal from the source which synchronises the frequencies of the cancelling vibrations produced by the active vibration reduction system to the frequencies of the vibrations from the source.

One embodiment of the invention comprises

i) a speed and position reference signal from the source of vibration. Where the source of vibration is a rotating machine, for example a car engine, the reference signal may be a tachometer giving one or more pulses per revolution of the engine crankshaft.

ii) a pulse conditioning system which receives the speed and position reference signal as an input and generates a train of pulses as output. There are, say, N pulses generated per fundamental period of the vibration. For example, the number of pulses per revolution of a four cylinder engine might be chosen to be 2N, say. Typically N, an integer, would be 8 or 16 but it is not restricted to a multiple of two.

iii) One or more sinusoidal waveform generators. Each oscillator will receive the pulses from the pulse conditioning system and generate a signal corresponding to a single harmonic. If they use sampled digital hardware, then each time the digital oscillator receives a pulse it produces the next sample of the single harmonic signal. Normally the oscillators are arranged in pairs, both producing the same harmonic, one being the in-phase (or sine) component and the other being the quadrature (or cosine) component. There would normally be one pair of oscillators for each harmonic of the vibration that is to be controlled. iv) An output module which generates one or more continuous output signals. Each output signal is formed from a weighted sum of the current outputs of all of the sinusoidal waveform generators. When the output of the sinusoidal waveform generators is a sampled signal these weighted sums will also be sampled signals. Each of the digital sampled signals is then converted by a 'digital to analogue converter and sample and hold' circuit into a continuous analogue output signal. The digital form of the output signal, X, at time instant i is

X(ι) = 2__,ajWSj(i) + bjWCj(i) (1)

Where aj is the weighting coefficient for the j th frequency in-phase generator bj is the weighting coefficient for the j th frequency quadrature generator

WSj(i) is the i th sample of the output of the j th frequency in-phase generator

WCj (i) is the i th sample of the output of the j th frequency quadrature generator.

When the sinusoidal waveform genertors are analogue then X(i) represents the continuous output at time t=i. The weighting is achieved with a variable gain and *a digital to analogue converter and sample and hold' circuit is not required. For each output signal there will be a different set of weighting coefficients aj and bj j=l,M. v) One or more actuators, connected via suitable filters and power amplifiers, to the output module. One actuator receives one of the analogue output signals. The actuators, being responsive to the output signals, generate a vibration which combines with the existing vibration from the repetitive source of vibration so that within some region the vibration is reduced in amplitude.

vi) One or more sensors, responsive to the vibration in the region where attenuation is required. The output from the sensors being fed via suitable signal conditioning and filtering to an input module.

vii) An input module, receiving one signal from each sensor. The input module processes each input signal to derive, for each one, a pair of harmonic coefficients for each of the M harmonics to be controlled. This processing is done by heterodyning the input signals with the output of each sinusoidal waveform generator and integrating the resultant output. When the input module is a sampled digital system it samples each sensor signal synchronously with the train of pulses from the pulse conditioning system, (ie each time a pulse is received a sample is taken of each of the sensor signals.) These sampled signals are then converted, by an analogue to digital converter, into digital form. The resultant in-phase coefficient, c, and quadrature coefficient, d, are then given by the expression

'3 y(i).WSj(i) , y(i)WCj(i_. (2) viii) An adaption module which varies the output coefficients, aj and bj j=l, M, for each output signal in order to reduce the vibration in the region to be controlled. This variation is done in response to the changing input coefficients, cj, and dj, j=l,M, of each input channel.

Most commonly the aim is to reduce the level of vibration in a particular region. The input sensors may be distributed within that region and the system set up to reduce the level of vibration measured by the sensors. In that case, the objective would be to minimise a weighted sum of the squares of the input coefficients, ie

minimise Tj_e (cj² + dj²)_e (3)

by variation of aj and bj for all harmonics and outputs, where e in the index for the input channel number.

The weighting factor T is included because different input sensor signals (at different harmonics) may need to be given particular emphasis in order to achieve the aim of reducing the vibration generally in the region.

Alternatively, the input sensors may be distributed on one part of the boundary of the region if, for example, it is known that the vibration is transmitted into the region through that part of the boundary. Again it may be important to weight the input sensor signals differently in order to achieve a uniform attenuation in the region required. One example of this would be the reduction of radiated noise from a fishing vessel to minimise the disturbance to a shoal of fish, where it is known that the vibration is transmitted to the water through the hull surface. Sensors distributed on the hull surface would be an effective monitor of the vibration entering the water. If the vibration at the hull were reduced, the noise in the water would also be reduced.

There are two general ways of achieving the minimum, either, a) it is achieved using an algorithm which attempts to get there in one step or, b) by an algorithm which iterates towards the minimum by a gradient descent method. The derivation of the equations for these algorithms is well known (see, for example, WIDROW and STEARNS, 'Adaptive signal processing' published 1985 by Prentice Hall) .

For a) X_k+1 = Xk " (B^HQB)^"VQY (4)

For b) X_k+1 = x_k - _/xB^HQY_k (5)

where Q is a diagonal matrix whose diagonal elements are the elements Tj_e and the other terms are defined below. At this stage, it is important to note that both algorithms can be generalised into one form.

Thus, for one harmonic

X_k+1 = αX_k - iAY_k (6) where X is a complex vector containing P elements. Each complex element being a+jb for the respective output channel, there being P output channels.

Y is a complex vector containing Q elements. Each complex element being c+jd for the respective input channel, there being Q input channels. is a real weighting factor in the range 0-1. μ is a small real update factor which determines the rate of convergence of the algorithm.

A is a P x Q complex matrix, whose elements are based upon measured parameters of the controlled system and weighting factors.

In applications where the speed of the rotating machine is changing then a speed signal is used to select the values of the elements of matrix A appropriate to the frequency of the harmonic being adapted.

The update matrix. A

The update matrix is primarily determined from the response of the input sensor's input coefficients to changes in the output signal's output coefficients. Before control is attempted the control system may be programmed to set one output signal's set of output coefficients in turn to some test level. The response of the input coefficients to each change in the output coefficients being measured so as to generate a matrix, B, relating the input coefficients to the output coefficients.

Y = BX (7) Where Y is that part of Y determined by changes in X (ie it does not include the part related to the original source of vibration) .

The matrix A can have many forms and two examples are 1) A = B^HQ where the superscript H means Hermitian transpose and 2) A = (B^HQB)^-1B^HQ.

The elements of matrix B, and thus those of matrix A, are likely to depend upon the frequency at which it is measured. Consequently each harmonic will have a different A, and as the speed of the engine changes the A for each harmonic will change. The particular value for A would be selected from memory to correspond to the appropriate frequency. If a value of A was not available for the particular frequency then either the matrix corresponding to the closest frequency would be selected or one would be generated by interpolation.

More detailed description of some elements of the invention

Speed and position reference signal

This signal is derived from the source of vibration and gives information about the frequency, and position of the source of vibration in its periodic cycle. It would normally be in the form of a pulse train where the frequency of the pulses is proportional to the speed of rotation of the source and one pulse is generated each time the machine reaches a fixed point in its cycle. It is advantageous to have a tachometer which produces many pulses per revolution of the source so that the active vibration reduction system can follow speed changes effectively. The pulses should preferably be evenly spaced in the revolution cycle and will often be produced by a shaft encoder. Alternatively the reference signal can be derived from the sinusoidal output of an electrical generator connected to the shaft of the machine generating the vibration or a sensor which responds to the passage of one point on the shaft (eg a crankshaft angle sensor on a car engine) . When the source is not a rotating machine, a transformer, for example, the speed and position signal can be derived from the mains as this gives the repetition rate of the vibration production process.

The frequency of the pulses may be a sub-multiple or multiple of the fundamental frequency of the vibration.

Pulse conditioning system

The pulse conditioning system (PCS) is not essential because the speed and position reference signal may already be of the right form. However this is not always the case. The output of the pulse conditioning system should be a set of N pulses per fundamental vibration cycle, evenly spaced in the cycle. When the input to the pulse conditioning system is a single pulse per fundamental cycle then the task of the Pulse conditioning system is to multiply the frequency of the pulses by a factor of N. This can be done using a phase-locked loop or by a digital equivalent. When the input to the pulse conditioning system is a single pulse per shaft rotation of a car engine, for example, where the fundamental vibration frequency is twice the crankshaft rotation frequency the task of the PCS is to multiply the frequency of the pulses by 2N. When the input to the pulse conditioning system is, say, 5 pulses per crankshaft revolution of a car engine then the PCS must produce 2N for every 5 pulses it receives. This could be done by multiplication of the input pulses by 2N and subsequent selection of one in every 5 pulses.

Sinusoidal Waveform Generators

Each sinusoidal waveform generator is synchronised to the pulse train from the PCS and generates a signal corresponding to one of the harmonics produced by the vibration source. One embodiment of the sinusoidal waveform generator comprises computer memory which stores the individual samples of the digital signal WS(i), i=l, N, and each time a pulse is received from the PCS the relevant sample WS(i) is recalled from memory. Since all of the signals produced by the oscillator are either the fundamental or harmonics of the fundamental cycle, which is N pulses long, only N samples need to be stored. However, when N is a power of two some of the values of different samples are the same as each other and thus a saving of memory can be made.

When the sinusoidal waveform generator is producing the third harmonic (3f) the stored samples WS(i) are sin (67ri/N) . Normally the sinusoidal waveform generators are arranged in pairs, both oscillators of the pair producing the same harmonic, one being the in-phase component, WS(i)=Sin (2πi/N) and one being the quadrature component WC(i)=Cos(27ri/N) . There being one pair of oscillators per harmonic to be controlled. The advantage of using sinusoidal waveform generators of this form is that the frequency that they produce is exactly synchronised to the frequency of the source of vibration and the signal is generated purely by recalling samples from memory. Other sinusoidal waveform generators have been described in PCT/GB87/00706 which are digital oscillators in the form of recursive digital filters. There are two disadvantages of the waveform generators described there; 1) they require multiplications to be performed which is more complicated, and 2) either a) they are part of the same processor as the controller in which case the real time base of the controller, which is necessarily the sampling period of the oscillator, will not coincide with the time base of the vibration period. Consequently, the ratio of frequencies is controlled by a stored numerical factor which requires measurement, constant updating, and may not be accurate. Or b) the digital oscillator is on the same time base as the vibration period and a separate processor is required leading to extra complications and a potential problem of passing information between the processors operating on different time bases.

An alternative to using a pair of sinusoidal waveform generators per harmonic is to use only one sinusoidal waveform generator per harmonic with phase adjustment. For this one generator at the ith instant the relevant sample, recalled from the sinusoidal waveform generator memory and passed on to the output module, may be advanced or retarded to adjust the phase of the generator. (ie at the ith instant the sample passed on may be WS(i-P).) The value of P may be different for each output channel of the output module. It may be advantageous to store intermediate elements to allow finer control of the phase. Output Module

The output module generates one signal for each actuator. In the sampled digital implementation of the system each element of the output signal is generated at the instant that it is needed by multiplication of the output coefficients by the samples recalled from the sinusoidal waveform generator.

The advantage of this digital method of generating the output signals over previously reported methods is now described.

In GB1577322 Chaplin described the use of a waveform generator which stores the output waveform for a whole cycle in sampled form and repetitively outputs the waveform in response to a synchronisation signal. This waveform is previously stored and updated periodically in response to the changing sound to be cancelled. Either individual components are adjusted as is described in GB1577322 by an adaptive process which operates on a different time scale from the output of the signal or the whole waveform is adjusted periodically as is described in GB2107960. The crucial advantage of the present invention over systems of the type described by Chaplin is that the output signal is generated at the instant that it is needed and thus it is immediately responsive to any changes in the output coefficients made by the adaptive algorithm. In systems where the sound to be cancelled is changing rapidly, for example, where the speed is changing rapidly this is very important. In PCT/GB87/00706 the output signals are generated by filtering original signals from the source of vibration. These original signals are generated by a separate means and must contain the harmonics of the signals to be cancelled. They are then sampled on a real time base and filtered in a digital filter and finally output. The advantage of the present invention is that it is not necessary to create a signal with the harmonics to be cancelled which has then to be sampled and filtered. One advantage comes from the use of a time base that is synchronised to the fundamental frequency. This may seem a small difference but it provides a significant reduction in the systems comlexity and an improvement in its reliability. The second advantage lies in the fact that the variation of the output signals is achieved by variation of output coefficients which are controlling orthogonal elements (in-phase and quatrature signals being orthogonal) . This leads to a well conditioned and efficient adaption. The variation of sequential weighting coefficients of a digital filter is less well conditioned and thus less robust.

Where only a single sinusoidal waveform generator is used per harmonic, the output module operates in a modified way, the digital, sampled form of output signal is:

X(i) - ; gjWSj (i-Pj) (8)

where Pj is the phase factor. Instead of a and b defining the output waveform g and P are used. The relationship between them being:

a + jb = g exp(2π-P/N) (9) Input Module

The purpose of the input module is to sample the input signals and identify the amplitude of each input component, CJ and dj, j=l,M. The input components are estimated by heterodyning the input signals with the outputs from the sinusoidal waveform generator and time averaging the resultant output. This time averaging can be taken over a fixed number of samples, R, (see equation (2)) where preferably R is a multiple of N, giving the samples equal weight, or by an exponential averaging process:

Cji » βCji_ι + y(i).WSj(i) (10)

where β is a variable from 0-1 which determines the averaging time.

Alternative forms of input module have been described which fall into two types.

Type a) described in GB2107960 samples the waveform from one or many whole cycle(s) and then takes a Fourier transform of the waveform to generate harmonic coefficients. This process is slower than that of the present invention where the information about the value of the current input coefficients is available as soon as the most recent input sample has been received. 2Q Type b) described in one part of PCT/GB87/00706 multiplies the input signals by sine and cosine signals and forms the coefficients by integration. This process is more complicated than the present invention because the system described uses a real time base for the sampled input signals and cannot make use of the stored sinusoidal waveform generator signals.

Adaption Module

One of the advantages of the present invention is that the vibrations to be cancelled at different harmonics can be controlled independently (ie the output coefficients at one harmonic are only dependent upon the input coefficients at the same harmonic.) This enhances the speed of adaption of the control system.

The other advantage of the present invention is that the output signal generation process and the input coefficient calculation process both operate independently from the adaption process, and yet, as soon as change in the output coefficients are made by the adaption process they are used by the output module. Additionally, as soon as the inputs respond to changing vibration the effect on the input coefficients is immediate and available for use in the adaption process. The time scale for the adaption process is determined by the transport delays in the particular system being controlled whereas the input and output process time scales are determined by the fundamental vibration frequency thus it is desirable to have these processes independent and yet able to pass information between each other. In some circumstances when two or more actuators are positioned close together, or two or more sensors are close together, the matrix of interactions B can be almost singular. This can result in large output signals which may exceed the range of the electronic system and which may be undesirable as it is likely to cause non-linear distortion. In this circumstance, it is desirable to limit the amplitude of the output signals and this is most conveniently done by increasing the 'leak' applied to the old value of the output coefficient. This corresponds to reducing the size of in equation no. 6. Consequently, it is appropriate to modify during the adaption process in response to the size of the output coefficients.

The rate at which the adaptive algorithm converges is controlled by the amount of notice that is taken of the new information from the sensor inputs. The factor μ determines this in conjunction with the matrix A. It is desirable to maximise the rate of convergence of the algorithm and thus to keep μ large however if it is too large then the algorithm can become unstable. The limits of stability depend upon the actual value of the matrix of interactions B and the value chosen for μ and A. If A is set to (B^HQB)^-1B^HQ, where B is the measured value of B, and this measurement is accurate then the optimum value of μ is 1. If μ is increased beyond 2, then the algorithm becomes unstable. When the estimate of B, B is poor then it is no longer possible to say, with confidence, what the limiting value is for μ and it is advantageous to reduce its size. If A is set to B^HQ and the estimate is good, then the maximum size of μ is proportional to the reciprocal of the largest eigenvalue of B^HB. Again when the estimate is poor, it is advisable to reduce the value of μ . For those reasons. it is advantageous to adjust the size of μ during the adaption process.

It has been said previously that in the beginning, B is measured. If the active vibration reduction system is to control the vibration from a device that is changing the frequencies of the vibration that it is producing, then B should be measured at a few different frequencies to cover the range. During operation of the active vibration reduction system, the system under control may change, for example, a passenger may move inside the cabin of a car or aeroplane. This may cause the actual value of B and its estimate, B, to be significantly different and may lead to instability or at least slower convergence. It is thus desirable to re-estimate the value of B on-line. This can be done by correlating the changes made to the output coefficients with the response of the input coefficients during the adaption process. However, when the system is working well, the input coefficients will be small, and perturbed by noise, and the resultant estimation procedure will be ill conditioned. In this case, it is desirable to introduce an extra low-level perturbation to the output coefficients and correlate the changes in the output coefficients with the resultant change in the input coefficients. Whilst this will introduce a little extra vibration, it can be at sufficiently low a level not to cause a significant loss of performance. One least squares estimation procedure that is appropriate to estimate B is given below. Rows of the matrix are estimated separately. Each row contains the response of one input sensor's coefficients to all of the output coefficients and is represented as b^fc. The equation for one input sensor's coefficients is:

y - b^-X + v where v is the vibration measured (11) without control and y is the complex number representing the input sensor's coefficients.

This equation can be rewritten as

y = (X^l) [b] (12)

[v]

and if many equations for sequential measurements are put together as a column then the resultant matrix equation can be written as

L = HK+e, (13)

where, L is a column vector containing sequential elements of y, H is a matrix containing sequential rows of (X^-l) and K is the vector (b^t:v)^t.

The best estimate of K and thus the row of b is:

K = (H^HH)^_1H^HL (14) Actuators

The actuators are fed with an output signal from the output module. In order to make the output signal suitable for the actuators it will normally be low-pass filtered (in order to smooth the output and to minimise the introduction of higher frequency quantisation noise) and amplified to a suitable drive level. The actuators convert the electrical signal they receive into a mechanical action and can be of many different types, for example, electromagnetic, piezo-electric, magnetostrictive, pneumatic, hydraulic, electrothermal, electrostatic.

Sensors

The sensors convert a mechanical action (eg pressure, vibration, displacement, etc) into an electrical signal and can be of many different types, for example, electromagnetic, capacitive, piezo-electric, magnetostrictive, pneumatic, hydraulic, thermoelectric. In order to make the direct signal from a sensor suitable for the input module it would normally be fed through a preamplifier and a bond-pass filler, the lower turn-over frequency being set to eliminate any very-low frequency interference, and the upper turn-over frequency set to prevent aliasing in the sampling process of the input module when it is a sampled data system. The turn-over frequency required to prevent aliasing is normally chosen at about one third of the sampling frequency. Where the frequency of the source is likely to change significantly it may be desirable to alter the turn-over frequency accordingly. One well- known method doing this is using switched capacitor filters. The invention will now be described by way of example with reference to the accompanying drawings.

In the drawings

Reference will be made to the accompanying drawings, in which:-

Figure l is a block schematic diagram of an active vibration reduction system associated with one particular enclosed space.

Figure 2 is a block diagram of one arrangement of an active vibration reduction system.

Figure 3 is a block diagram of an output module.

Figure 4 is a block diagram of an input module.

Figure 5 is a schematic diagram illustrating aspect of the input module.

Figure 6 is a block diagram of one form of pulse conditioning system.

Detailed description of the preferred Embodiments

In figure 1 an active vibration reduction system, 1, according to the invention is shown associated with the passenger cabin, 2, of a propeller-driven aircraft, 3. The propellers, 4, create noise and vibration which is heard inside the cabin 2. In this example, the system 1 uses two actuators 5 comprising two loudspeakers fitted into the aircraft cabin and four sensors 6, comprising microphones fitted to the seat backs 7. The loudspeakers are driven by the control unit, 8. Control unit 8 receives a speed and position reference signal, 9, from the electrical generators connected to one of the propellers, 4. This speed and position reference signal, 9, is fed into the control unit, 8, where the pulse conditioning system, 10 received it and generates a set of pulses. In this example the propellers have six blades and the speed and position reference signal,

9, is a sinusoidal signal with a frequency four times the rotation rate of the shaft of the propellers. The pulse conditioning system, 10, shown in figure 6, is configured to multiply the frequency of the speed and position reference signal by twenty four. This will then generate 16 pulses per blade passage. In this particular embodiment of a pulse conditioning system,

10, the speed and position reference signal is shaped in a pulse shaper circuit, 11 to produce a suitable level square wave pulse train for the phase comparitor, 12,. The phase comparitor compares the signal from circuit 11 with the output of divider circuit 13 and adjusts the frequency of oscillator 14 in order to maintain a fixed phase relationship between the two signals 15 and 16. The output of oscillator 14 feeds both the divider circuit 13 and a buffer amplifier 17. The output of the buffer amplifier is the output of the pulse conditioning system and this is fed to the digital oscillators 18.

Figure 2 shows more details of the control unit, 8, including the input module, 19, output module 20, adaption module 21, pre amplifiers 22, band-pass filters, 23, filters, 24 and power amplifiers, 25. Figure 3 shows more details of the output module, 20.

The output from the sinusoidal waveform generators, 18 is amplified (or multiplied) by gain elements, 26, whose gains, aj and bj (which are different for each output channel) are determined by the adaption module, 21. The signals for each output channel are individually summed in summers, 27, before being fed to digital to analogue converters and sample and hold circuits, 28. Finally the output signals 29 are fed to the loudspeakers, 5, via filters, 24 and power amplifiers 25. The purpose of the active vibration reduction system is to reduce the noise in the cabin of the propeller-driven aircraft which has been produced by the propellers, 4 by generating noise with the loudspeakers, 5. The noise generated by the loudspeakers being adjusted generally to cancel the propeller noise.

As can be seen from the above description the adjustment of the loudspeaker sound is achieved by variation of the output coefficients, a and bj, by the adaption module, 21. The adaption module is responsive to information that it receives from the input module 19 which is in- turn responsive to signals derived from the sensors 6. The adaption module changes the output coefficients aj and bj automatically in order to minimise a weighted sum of squares of the input coefficients. The adjustment being according to an adaptive algorithm,

X_k+1= β X_k - μ AY_k.

The number of sinusoidal waveform generators 18, and the harmonic that they produce is matched to the number of harmonics of the propeller sound that are to be reduced. The number may vary from aircraft 3 to aircraft but is only likely to be a few of the lower harmonics. The number of loudspeakers 5 to be used and their position in the cabin, 2, is chosen in order to achieve sufficient attenuation of the sound at reasonable cost. Generally the more loudspeakers 5 that are used the greater the attenuation. However, care must be taken to position the loudspeakers, 5 in the cabin 2 as significant variations can be found from different distributions. Generally the positions are chosen from those that satisfy other constraints to maximise the attenuation achievable. The number and position of the sensors 6 is chosen so that the sound that reaches the region where quiet is required (ie around the heads of the passengers) is monitored effectively. In this example, where the sound is coming from the walls of the cabin 2, in all directions, it is advantageous to have the sensors close to the heads of the passengers. It is likely, therefore, that most of the passenger seats will have sensors. The number of sensors is preferably greater than the number of loudspeakers, 5, as this tends to achieve a more even distribution of attenuation and improves the adaption process.

Figure 4 shows the input module, 19 which receives the signals from the sensors, 6, and provides some of the information required by the adaption module, 21. The signals from the sensors pass through pre-amplifiers, 22 to boost their level and band-pass filters, 23, before reaching the sampling and digitisation circuit 30. The analogue input signals 31 are sampled synchronously with the pulses from the pulse conditioning systems 10 and in this example this is achieved by clocking each analogue to digital converter 30 with the pulse signal 32. Alternatively one analogue to digital converter can be used in combination with a multiplexer as there is no necessity to have all the input signals 31 sampled simultaneously. Once the signals have been digitised by circuit 30 they are heterodyned 33 with the outputs of the sinusoidal waveform generators 18. Details of the heterodyning process are shown in figure 5. Each digital signal representing one of the input signals 31 is hetrodyned 33 with each one of the outputs of the generators 18. The resultant data from each multiplication (heterodyning) process is averaged by an averager 34 functioning according to equation (10) in order to produce the input coefficients, CJ and dj for each input channel.

The input coefficients are used by the adaption module operating with equation (6) to produce updated output coefficients. The update matrix A used for each harmonic is selected from memory according to the frequency of the pulses on pulse signal 32. In this example, where the fundamental frequency is nominally 100Hz but it varies over a range from 85Hz-110Hz and the first and second harmonic are being controlled. Values of matrix A are measured and stored at 85, 90, 95, 100, 105, 110, 170, 180, 190, 200, 210 and 220 Hz. The frequency of the pulses can be measured using a frequency to voltage converter circuit followed by an analogue to digital converter. The resultant digital value is proportional to the frequency of the pulses and thus the fundamental frequency. Once it is calibrated this value can be used to select the appropriate A from the store. For example if the value indicated that the fundamental frequency were 94Hz then either the value of A corresponding to the closest frequency (ie 95 for the first harmonic and 190 for the second harmonic) would be chosen or the value of A would be calculated by interpolation (eg for the first harmonic the calculated value of A would be 1/5 of the value of A for 90Hz and 4/5 of the value of A for 95Hz) . Generally A does not change very quickly with frequency and so slight inaccuracy in frequency measurement is acceptable. The control unit, 8, of the active vibration reduction system excluding the pulse conditioning system can be implemented using a computer with memory. Various types of computer can be used for example a Digital Signal processing unit type DSP 32C (AT & T) . The sinusoidal waveform generators would then be data stored in memory recalled by either the input or output module calculations. The timing of the calculations of the input and output modules would be provided by interrupts generated from the pulse signal 32. Alternatively, the adaption unit could be the only part implemented using a computer with memory. The sinusoidal waveform generators would be implemented as oscillators phase locked to the output of the pulse conditioning system. The output coefficients would be implemented as gains controlled by the digital adaption unit, for example, using multiplying digital to analogue circuits. The input coefficients would be voltage levels, produced by analogue heterodyning and integration circuitry, and would be sampled by the digital adaption unit using an analogue to digital converter. The sample rate of the adaption unit would then correspond to the adaption rate which will be much slower than the sample rate required by a wholly digital system.

Claims

1. An active vibration reducing system for controlling the vibration generated by a source of periodic or quasi- periodic vibration is characterised in that:

a weighted sum of the output of at least two sinusoidal waveform generators each synchronised to the frequency of the source of original vibration to generate harmonics of the frequency of the source, is supplied to at least one actuator which generates additional vibration,

sensor means is used to monitor the effect of the additional vibration on the original vibration

and the weighting of the sum of the outputs is adjusted so that within a defined region the amplitude of the combined vibration is reduced.

2. A system according to claim 1, wherein the sinusoidal waveform generators are synchronised to the source of vibration by a synchronising signal which comprises an integral number of pulses per fundamental period of the vibration to be controlled.

3. A system according to claim 1 or claim 2 , wherein the sinusoidal waveform generators comprise computer memory which stores fixed data corresponding to a sampled oscillator output.

4. A system according to any one of the preceding claims, wherein the weighted sum of the output of one or more sinusoidal waveform generators is formed by multiplying the outputs of the individual sinusoidal waveform generators by a different variable output coefficient and adding the result of all of the multiplications together.

5. A system according to claim 4, wherein the variable output coefficients are independently variable for each actuator.

6. A system according to claim 5, wherein the variable output coefficients are different for each actuator.

7. A system according to any one of claims 4, 5 or 6 , wherein the vibration from the actuators is controlled by variation of the output coefficients.

8. A system according to any one of the preceding claims, wherein the sensor means generate signals and input means are provided to process the signals from the said sensor means to generate input coefficients which relate to the amplitude of the vibration at the harmonic produced by the sinusoidal waveform generators.

9. A system according to claim 8, whrein the output coefficients are adapted in accordance with the algorithm in response to the input coefficients by an adaption means.

10. A system according to claim 9, wherein the algorithm aims to minimise a weighted sum of the squares of the input coefficients.

11. A system according to claim 8, wherein the algorithm aims to minimise a weighted sum of the squares of the input coefficients whilst constraining the amplitude of the outut coefficients.

12. A system according to claim 10 or 11, wherein the weighting for the squares of the input coefficients is chosen to allow the system to reduce the vibration in the region.

13. A system according to any one of the preceding claims, wherein the adaption means is provided with a frequency signal.

14. A system according to any one of claims 8 to 13, wherein the sinusoidal waveform generators and input means are implemented in one digital computational unit with associated memory for data and programme.

15. A system according to claim 14, wherein the adaption means is implemented in a separate computational unit sharing the data memory or in the same unit.

16. A system according to any one of claims 8 to 13, wherein the input means and sinusoidal waveform generators are implemented in analogue circuitry and the adaption means is implemented in a digital computational unit with associated memory for data and programme.