Technical Field

This invention relates to an "active attenuation" method of cancelling noise (or other vibration) in which a primary vibration entering a selected location is nulled or can- celled by means of a secondary vibration which is fed into the location to reduce (and at best eliminate), the influence of the primary vibration. An accurate matching of phase and amplitude of the secondary vibration to the primary vibration is necessary if effective nulling is to result. In particular this invention is concerned with repetitive primary vibrations such as arise, for example, with an internal combustion engine. Background Art

In the specification of United States Patent No. 4,153,815 there is described a method for the reduction of unwanted vibrations received at a selected location from a source of recurring vibrations which method comprised feeding a specially synthesised secondary vibration to said location, the generation of the secondary vibration being triggered by a signal derived from the source, and the waveform of the synthesised secondary vibration being adapted with respect to time to improve the extent to which the secondary vibra¬ tion nulls the primary vibration at the said location. Disclosure of Invention The present invention relates to an adaptive method similar to that described in the preceding paragraph but which is characterised in that the time taken to evolve a successful _' adaption of the secondary vibration is reduced by monitoring the contemporary residual vibration at the said location which results from interaction between the primary vibration and the contemporary secondary vibration, and effecting the adaption of the secondary vibration on the basis of the amplitude and/or the polarity of at least a part of the contemporary residual vibration. Suitably the amplitude and/or the polarity of each pulse of the residual vibration is cyclically sensed in a

plurality of successive time intervals, each separate sensing being separately used in the correct sequence to control the adaption of a different part of the secondary vibration. Brief Description of Drawings

The invention will now be further described, by way of example, with reference to the accompanying drawings, in which

Figure 1 is a schematic illustration of the broad princi- pies behind the method of the invention,

Figure 2 is a.block diagram illustrating one embodiment of waveform generator as shown in Figure 1,

Figure 3 is a schematic indication of the principle for the fast adaption achieved by the method of the invention, Figure 4 shows three graphs of actual and simulated waveforms,

Figure 5 is a schematic illustration of a first embodi¬ ment of apparatus for carrying out the method of the inven¬ tion, Figure 6 is a graph illustrating how delay is compen¬ sated for in the method of the invention,

Figures 7 and 8 are schematic illustrations of second and third embodiments of apparatus for carrying out the method of the invention, Figures 9 and 10 are graphs illustrating the performance obtainable with the method of the invention, and

Figure 11 is a schematic illustration of a fourth embodi¬ ment of apparatus for carrying out the method of the invention. Description of Preferred Embodiments In the following description reference will be made to the cancellation of noise (i.e. acoustic vibrations in air) but it should be understood that the invention is equally applicable to the cancellation of vibration in gases, liquids or solids. Figure 1 shows a source 1 of a recurring noise which is nulled in a protected area 4 by a secondary vibration gener¬ ated by a transducer (e.g. a loudspeaker) 3. A waveform generator 5 powers the transducer 3 and the output of this

generator 5 is synchronised with the noise generation by a transducer 2 on the source 1. A microphone 6 in the area 4 senses the residual sound and feeds this to a control unit 7 which controls an adaptation unit 8 used to modify the output waveform of the generator 5 and thus seek for total cancell¬ ation in the area 4.

One form of generator 5 and adaptation unit 8 which can be used in the system of Figure 1 is illustrated in Figure 2. The synchronisation signal(s) from the transducer 2 will be at a repetitive frequency f. If only one synchronising pulse per cycle is available, this will be fed to a frequency multiplier 9a. (e.g. incorporating a phase locked loop) which feeds a frequency which is an integer multiple of the frequency f to a frequency divider chain 9b which sequentially addresses locations of the memory 9 . Preferably the syn¬ chronisation signal is derived from a toothed wheel or the like driven by the engine, and in this way a train of per¬ haps a hundred equally spaced pulses will be generated in each cycle of the source so that the integer 9a. would not then be required.

The waveform memory stores a plurality of samples each having a unique address in the memory 9c_. The samples repre¬ sent portions of a precursor of the required waveform to be generated and are presented sequentially to a digital-to- analog converter 9ά ( which includes a power amplifier and a low pass filter to remove sampling frequencies) to gener¬ ate the actual waveform to be fed to the transducer 3. It is because each of the samples must be presented once per repetition of the acoustic waveform to generate the required secondary wave that the need arises for a frequency multi¬ plier if only a single synchronising pulse is available from the source. The degree of multiplication depends on the number of samples required and in a typical case this could be 32. The samples stored in the memory 9 can be derived in a variety of different ways but since the memory is modi¬ fied by the unit 8 to minimise the output from the unit 7_ it is not generally too important•what the starting samples are,

since eventually if each burst of recurring primary sound energy is like each other burst, the correct samples will appear in the memory Sc_ and the pattern of samples one starts with merely affects how long it takes to produce the correct cancelling signal.

Although the foregoing description referred to 32 dis¬ crete elements, there could of course be a larger or a smaller number. Alternatively, one can interpolate between time elements in such a way as to simulate the effect of a much larger number of elements than are actually present. This interpolation can be performed either in the residual signals line, in order to provide the information for adjust¬ ing the appropriate cancelling elements, or can be performed between elements of the cancelling waveform. Referring again to Figure 1, the control unit 7 serves to process the residual signal picked up by the residual noise microphone 6, this signal being the sum of the sound from the recurring sound source 1 and the cancelling sound being generated by the transducer > . In the system described in U.S. Patent No. 4,153,185, this signal is processed by the control unit 7 to give an output related to the overall power of the residual signal. The cancelling waveform is adjusted so as to reduce this overall power level.

The present invention involves deriving information from at least one part of the residual signal to reduce the adaption time. For example parts of the residual waveform may be masked such that the residual power signal includes only those parts of the residual waveform which are consider¬ ably affected by a modification to the cancellation wave- form.

A further embodiment involves sampling the residual sig¬ nal at an appropriate point after an alteration to the cancell¬ ing waveform and using the -polarity and/or magnitude of the change in residual signal as a correcting parameter to the corresponding element of the next cycle of the cancelling waveform. Such an operation can be performed on a single time element or on a plurality of time elements per repeat cycle of the waveform.

However, a more effective system would be to perform the operation the other way round, i.e. to observe the residual signal and make changes to the subsequent cycle cancellation waveform based on information from the previous residual waveform. Again a single modification or a plural¬ ity of modifications could be performed during each repeat cycle.

A particular element of the residual noise signal can be related to a cause in the cancellation signal at integer 3 at an earlier time. There may be a single significant causing element or a series of significant causing elements at various earlier times. It is necessary therefore to compensate for the delay or delays between the causing element(s) and the residual effect. The compensation is applied after sensing the residual signal and prior to applying a correction to the next cyclic cancellation signal. Instead of applying a con¬ trolled modification exactly one repeat cycle later, it is applied at a time period (Δt) less than one repeat cycle later such that 'the difference Δt compensates for the acoustic delay.

This delay could be manually adjusted but in practice it may vary. For example, in the case of cancellation in a cab, the opening or closing of a cab window would affect the delay. It will therefore probably be advantageous to auto- mate the delay by, for example, relating it to the impulse response of the system, or by sensing one or more maxima in the cross-correlation function between the cancelling wave¬ form and the residual waveform.

It will be seen therefore that in order to reduce the time taken for the waveform generator 5 to build up the cancellation signal, it is necessary to retain more infor¬ mation about the nature of the reisudal noise signal. In one simple arrangement it is possible to obtain information about the polarity of the reisudal noise signal at a given point during a noise cycle, and thus use this information to ensure that at the same point in the noise cycle the incre¬ mental modification to the cancellation waveform has the opposite polarity and will hence improve cancellation at least

at the said point.

A more sophisticated system allows not only the polarity, but also the magnitude of the residual noise signal to be measured at a particular instant in time. This information can then be used directly to rapidly build up the cancella¬ tion signal and reduce the adaption time for substantially complete nulling to occur.

In algebraic terms, the precursor of the cancelling noise is stored as a series of samples in a memory and may be repre- sented as an ensemble of samples |~y _n _]j where n is the number of samples for one cycle of the waveform. The residual noise signal e_(_t) picked up by the microphone 6, may be sampled and also represented as an ensemble [ ; J having the same number of samples as | __- _n _3 _' ^{Jt s} possible ^" to relate a change in any sample of C ^ to a corresponding change in the sampled error signal C_~ _n _]» thus enabling information conveyed by e J adjust [y _n ]. ^~

Figure 3 shows how the microphone 6 generates an error signal e(t) from the sum of the original noise n(;t) from the source 1 and the cancellation noise y(t from the source 3 and how this is converted into an ensemble ( ^" e l in the device 7. The device 7 could be a sample-and-hold circuit or an analog- o-digital convertor, its purpose being to determine a parameter of e_(t at a repeatable position on the repetit- ive waveform.

The upper graph a in Figure 4 shows a waveform repre¬ senting a single cycle of the error signal (;t). This wave¬ form is also shown in sampled form Ce.] in the middle graph b_ of Figure 4. The lower graph ci of Figure 4 shows the cancellation waveform C∑ _n !]* If the j_ element of E _n 3 ^s adjusted by an amount depicted by the hatched area in graph c_ it will ha-ye an effect on the error signal e_(_t) and its sampled form Ce l _> also as shown by the hatched areas in graphs a and b. Conversely it is therefore possible to determine what adjustment must be made to the j_ element of [ ^y ] in order to reduce the size of the j_ element of C _n _I*

The analysis so far has ignored the effect arising because there will be a time delay between changing the value

of the j_ element of [y-]| and seeing its effect on _. l _ή -1 _' This time delay is, among other things, due to the acoustic delay between the microphone 6 and the source 1. Since in many cases this delay will remain substantially constant at any particular firing rate, it will correspond approximately to a constant number of waveform samples, and it is a simple matter to arrange that it is the (j+k) , element of the error signal that is reviewed when seeking to adjust the j_th element of the cancelling waveform, "k" being the number of elements by which the signal has been delayed.

Since this process may be executed rapidly enough, it becomes possible to adjust all n samples of C _n _] f° ^r each cycle of the noise, thus obtaining a marked increase in speed of adaption. Figure 5 shows how one such type of control function may be realised for a practical system.

The control unit 7 within the dashed line box, here com¬ prises a sample/hold circuit 7a_, and a comparator, 7b. The error signal from the microphone 6 is fed to the sample and . hold circuit 7a., which is synchronised by the timing signals received from the transducer 2 which is also controlling the initiation of each cancelling waveform pulse generated by the generator 5. Thus the error signal £(t) will be sampled in synchronism with the cancelling signal. The sampled error signal [_£ _n _] i ^s then compared with the desired error signal, which wilT normally be chosen as zero, result¬ ing in a binary level signal representing the polarity of the sampled error signal at each element n. This control function called G( ) may be represented as

^G ( ) _n = ≤g-n _n This may be used in the following algorithm to adjust the values of the cancelling waveform £y ^" j (assuming zero acoustic delay) ^—

where j_ represents th ⁿ ^ k i the k iteration of the process Δ represents the increment by which each sample of the waveform is adjusted. Δ may have a single value or may take on different values at different times to enable coarse and fine adjustment of [_y_ ]. The

value of Δ may be changed either as a function of time or of amplitude of the residual noise, depending on the adaption requirements for a particular application.

The circuit 7a can be dispensed with and the error 5 signal fed directly to the comparator provided the unit 8 can accept data from the comparator at the correct times.

To obtain more rapid adaption, even more information about the error signal may be retained. For example, a scheme, illustrated in Figure 7, may be employed, in which 1 ₀ information about the amplitude of the error signal is employed in the adaption.

In Figure 7, the error signal from the microphone 6 is fed to a sample-and-hold circuit 7a (which may be dispensed with) and then to an analog to digital convertor 9 which 15 modifies a combined waveform generator and adaptation unit 5/8 in the correct sequence controlled by the train of syn¬ chronising signals derived from the transducer 2.

In the embodiment of Figure 7 use is made of a linear control function

20 ^G ^n ⁼ -n in which a proportion α of the error signal is fed back to the waveform adaptor 5/8.

The algorithm used to adjust the values of r ] in this case is (assuming zero acoustic delay) ^— _o 2 _c 5 ^" _∑I k -_ yk-1 - «ejk-1

Thus each value of C _n !3 i ^s updated by an amount proportional to the error between the noise source and cancelling source. This system produces rapid convergence provided α is less than unity. 0 The invention will now be further described with refer¬ ence to Figures 5 to 10 by the following Example.

EXAMPLE A 1HP single cylinder petrol engine 20 was used as a noise source creating a repetitive noise at an exhaust out- 5 let 21. The noise was picked up at microphone 26 placed approximately one metre from the outlet 21 and was fed to a control circuit generally designated 27. Here the waveform information is processed before being fed to a

generally designated 30. The microprocessor 30 generates and adjusts a cancelling waveform which is fed to a loudspeaker 23, the waveform being synchronised to the speed of the engine by a train of signals received from a transducer 22. The cancelling waveform generated by the microprocessor 30 consists of a number of samples (between 32 and 256) whose values are stored in successive locations in a digital memory 301b. These are read out to a digital-to-analog converter 3θ at intervals dictated by the train of signals from the transducer 22, so that the cancelling signal -will be in strict synchronism with the firing strokes of the engine 20.

The residual noise signal from the microphone 26 is sampled into the same number of samples as the number of samples being generated for the cancelling signal by a sample-and-hold circuit 27a. This sampled signal can then be fed either to a comparator (e.g. such as 7b_ in Figure 5) to give a sensing of the polarity of each sample of the residual noise signal with respect to zero or to an analog-to-digital converter (e.g. such as 9 in Figure 7).

Dealing first with the polarity-sensing adaptive system, if each element of this error signal is made to correspond with a particular sample value of the cancelling waveform, then that sample value may be adjusted in the manner described by the following equation: k k-1 . j__-l ∑l = ∑l - Δ sgn e f ^* v_ where j_ is the j_ element of the waveform

_ι_τ_ k is the k iteration

Δ is the increment by which the element is to be adjusted sgn * _j represents the sign of the j_ element of the sampled error signal. In other words, if the error signal is positiva at a particular point j_ on the waveform, then the cancelling wave- form at that point is reduced by an amount Δ making it more negative and hence reducing the positive pressure of the noise waveform. If the error .signal is negative then the corresponding sample of the cancelling waveform is incre-

~r. V.

!* ^' ktiκ-ι\κ

mented by Δ. Δ represents an arbitrary increment by which each element is adjusted. Its value may be fixed or can be altered to facilitate coarse and fine adjustment of the wave¬ form. In practice the j_ element of the error signal does not correspond to the j_ element of the cancelling waveform, since for a practical system there is a delay of several milliseconds between the time the signal is fed to the speaker and the time it is received at the microphone. This delay (see Figure 6) can be accommodated by subtracting a time offset from the point in the waveform which is con¬ trolled by the error signal at ^ ή _^ , k-1 . k-1 ^1>e ' *(i-z) - ∑Ci-z) ^" ^ sgn £j

The magnitude of this offset, the time between B and C in Figure 6, is shown in Figure 6 expressed in samples of the waveform. The time delay between the application of a signal from the cancelling waveform generator and its maximum offset at the residual microphone must correspond to £ samples. i.e. Δt = z x T_ where Δt = delay, sensibly constant for a given speaker-microphone arrangement

£ = offset number of samples T_ = sampling period = Tf

- ^~ N ^" and T _f = periodicity of the noise

i.e. Δt T _£

This offset number can be updated from time to time under programme control for example ^' by measuring Ts, thus compen- sating for changing repeat cycle rates and changing periods for each of the samples.

Using this technique each waveform element is adjusted by an amount Δ during each cycle of the noise, hence the adap¬ tion time is ver much faster. Figure 9 shows a plot of sour.; pressure level as ordinate against tine as abscissa. Ini - ially the cancellation signal is zero and there is no atten-

uation. After about 4 seconds, the sound pressure level has been reduced by 20 dB.

Adaption time can be improved still further by using the control system described with reference to Figure 7. This is the arrangement shown in Figure 8. The error signal from the microphone 26 is sampled by the sample-and-hold circuit 27a in the same manner as described earlier. The sampled signal is then fed to an analog-to-digital convertor 29 which converts each sample value into an eight-bit digital representation. The waveform is generated in the manner previously described but now the sample values are adjusted by subtracting the digital representation of the error signal from the previously stored cancellation waveform sample values in a memory processing unit 30a. This can be described mathematically as

where α is a gain factor and £. is the digital represen- tation of the jth element of t—he error signal.

Additionally the error signal from the microphone 26 may be weighted prior to the subtraction process in order to compensate for the characteristics of the loudspeaker 23 and/or the acoustic conditions. This compensation may be performed by a filter whose characteristics may be calculated or measured for a given system. The filter may also be of an adaptive variety whose parameters are adjusted prior to or during the vibration cancellation process.

Figure 10 shows a graph of sound pressure level plotted as ordinate against time plotted as abscissa. It can be seen that the time taken for 20 dB attenuation to be achieved is less than one second. Because this time is so short it becomes possible for the cancellation system to track the wave shape of a recurrent noise source whose speed or load is constantly changing whilst still maintaining cancellation. An alternative implementation of this system is illustr¬ ated in Figure 11 which shows two signal delay paths which may be partially digital (e.g. programmed on a microprocessor) or analog (e.g. implemented in charge coupled device

ogy ) .

The first delay, integer 4l of Figure 11, provides the delay between arrows A and B of Figure 6. The period of this delay is T _f -Δ_t which can be arranged by a suitable circuit 44, synchronised to the repetition rate of a vibra¬ tion source 1, in systems which must tolerate rapid changes in repetition rate. Integer 44 provides timing pulses for the delay unit 4l of such a frequency that the delay is T _f -Δjt, where T _f is the slowly varying repetition rate of the waveform and Δjϊ is constant. For systems operating at con¬ stant repetition rate but with rapid waveshape changes, integer 4l may provide a simple delay function.

The second delay, integer 42, stores and cyclically pre¬ sents the repetitive cancelling wave, summed (in. an integer 43) with the controlling error signal from the integer 4l. The timing for the integer 42 may be synchronised by the source 1 by, for example, a tone wheel on rotating machin¬ ery producing the repetitive noise.

In a practical system (e.g. when using a secondary vibration to quieten the driving cab of a vehicle) it may be necessary to provide means to trim the delay during use (e.g. to compensate for a cab window being opened) and such trimming of the delay can be effected in a number of differ¬ ent ways. Further it may be necessary to use a plurality of delays 41 in parallel (e.g. as shown dotted at 4la in Figure 11) where the response of the microphone 6 to a delta function fed as the input to the transducer 3 gives more than one significant peak. It may also be necessary to trim the delay slightly to ensure it is accurately represented by an integral number of time periods £ (see Figure 6) and this can be done with a variable delay (such as 10 shown in Figure 5).

Figure 6 indicates how the delay can be accommodated. The waveform shown in the upper part of Figure 6 will be seen to repeat after the period T„ (i.e. from arrow A to arrow C represents one cycle of the repetitive waveform). This period is divided into n equal samples (as discussed

with reference to Figure 4). As the repetition rate of the source changes (e.g. T„ is reduced) each interval n will become shorter but there will still be n samples in the changed period T _f . The acoustic delay (unaffected by chang- ing the repetition rate) will now occupy more samples ( ) of 'the waveform, so that the electronic compensating delay A-B must be reduced.