WO1994027525A1

WO1994027525A1 - Active ear defender

Info

Publication number: WO1994027525A1
Application number: PCT/US1994/005271
Authority: WO
Inventors: Graham P. Eatwell; Joe Kulikauskas; Roger Geere
Original assignee: Noise Cancellation Technologies, Inc.
Priority date: 1993-05-21
Filing date: 1994-05-18
Publication date: 1994-12-08

Abstract

This invention relates to an active ear defender which uses an improved adaptive control system. The control is modified to include a parameter adjustment module (3) and a modified structure (1, 2) so as to provide selective cancellation of rapidly changing disturbances.

Description

Active Ear Defender

Background

The principle of using a controlled sound to cancel an unwanted noise is disclosed in U.S. Patent No. 2,043,416 (Lueg, 1932). The canceling sound must have the same amplitude but opposite phase compared to the unwanted noise. The use of a fixed analog feedback loop to maintain the desired relationship between the canceling sound and the unwanted sound is disclosed in U.S. Patent No. 2,983,790 (Olsen and

May, 1956). Early disclosures on the application of active noise control to ear defenders or headsets include U.S. Patent No. 2,966,549 (Fogel, 1954). In an active headset a loudspeaker is used to generate the canceling sound and a microphone is used to sense the sound. The microphone signal is used by a control system which generates the signal to drive the loudspeaker. A more practical headset using feedforward control is disclosed in German Patent

No. G 2,401,523 (Hesselman) and headsets using feedback control system are disclosed in British Patent No. GB 1,530,814 (Goom and Smith, 1976) and U.S. Patent No.

4,644,581 (Sapiejewski, 1985). An example of a feedback headset is shown in Figure 1

(Figure 3 of British Patent No. GB 1,530,814). It comprises a closed ear cup (1) which includes a loudspeaker (2) with a diaphragm (3). The signal from the microphone (5) is passed through an inverting amplifier (6) and a filter (7) to a speaker drive circuit (8).

The active part of the headset is only effective at low frequencies, at higher frequencies the attenuation is provided by the ear-cup.

One of the major potential problems with an active headset is that the feedback of the loudspeaker to the microphone can cause an instability. Avoiding this instability often compromises the performance of the headset, so active headsets usually work only at low frequencies (below 500Hz).

More recently digital control systems have been introduced for active noise control. A digital feedforward controller is disclosed in U.S. Patent No. 4,122,303 (Chaplin & Smith). This system is stabilized by estimating the effect of the loudspeaker on the microphone and then subtracting this signal from the input microphone signal. The same technique is used to stabilize a feedback control in the Digital Virtual

Earth (DVE) control system disclosed in U.S. Patent No. 5,105,377 (Ziegler). This is an adaptive control system in that the relationship between the input microphone signal and the loudspeaker drive signal is constantly adjusted so as to maintain optimal performance. This adaptive control system is shown in Figure 2 and is hereby incorporated by reference herein. An alternative adaptive control system applied to an ear defender is disclosed in U.S. Patent No. 4,654,871 (Chaplin et al, 1982). This system is for canceling periodic or tonal noise and uses a trigger signal as the input to the control system. This avoids the problem of instability, but the requirement for a trigger signal limits the usefulness of the system.

Adaptive feedforward and feedback control systems applied to 'in the ear' ear defenders are disclosed in patent application WO89/12432 (Langberg, Caruso al, 1988).

One disadvantage of passive ear defenders is that they prevent even desired signals such as speech or warning signals reaching the ear. The system of Chaplin has the advantage that it is selective so that only the noise is canceled and speech can still be heard.

Techniques for enabling other adaptive headsets to be made selective are disclosed in patent application PCT/GB91/00265 (Ross, Langley and Eatwell).

The known techniques are primarily designed for steady or slowly varying noises. The known control systems are not effective for noises which are both high frequency and rapidly varying. The noise from the siren of an emergency vehicle is an example of a noise which could not be controlled by the known techniques.

Objects An object of the current invention is to provide an adaptive noise control system for rapidly changing noises.

Another object is to provide a headset for use in emergency vehicles having a high noise environment.

A further object of the invention is to provide an adaptive noise control system for selective control of unwanted noise.

A further object of the invention is to provide an adaptive noise control system with improved robustness.

A further object of the invention is to provide an adaptive noise control system with a reduced computational requirement compared to known systems. A still further object of the invention is to provide an improved active noise reducing headset.

These and other objects will become apparent when reference is had to the accompanying drawings in which

Figure 1 is a diagrammatic view of a known active headset, Figure 2 is a diagrammatic view of a known adaptive feedback control system,

Figure 3 is a diagrammatic view of a modified adaptive feedback control system, Figure 4 is a diagrammatic view of an active noise reducing headset, Figure 5 is a frequency-time graph of a typical siren waveform, Figure 6 is a diagrammatic flow chart showing one embodiment using digital controller hardware, and

Figure 7 shows the measured results of a headset of the invention in a vehicle cabin.

Detailed Description of a Preferred Embodiment

By way of example a modified feedback controller will now be described. The basic structure of the digital controller is shown in Figure 3. The controller comprises three main components: the cancellation module (1), the adaption module (2) and a parameter adjustment module (3).

Cancellation Module

The cancellation module follows that disclosed in U.S. Patent No. 5,105,377 (Ziegler) and is shown in Figure 2 and incorporated in Figure 3. It comprises two digital filters. The first filter is the cancellation filter (4), which may be a Finite Impulse Response (FIR) filter with coefficients A(k) , or a lattice filter, or an infinite impulse response (HR) filter or other known filter. This is applied to a reference signal (5) with values x(i) and produces the output signal (6) with values y(i) . The second filter (7) is a feedback compensation filter, which may be an Finite Impulse Response (FIR) filter with coefficients C(k) for example, or may be another type of filter. This filter models the physical response of the system, including the actuator means (14), which comprises a digital to analog converter (DAC), analog filter, signal amplifier and sound generating means, and the sensor means (15), which includes analog filter, signal amplifier and analog to digital converter (ADC). The feedback compensation filter is applied to the output signal (6) and generates a compensation signal (8) with values z(i) related to the feedback from the controller output to the controller input. This compensation signal is subtracted from the input or error signal (12) so as to remove the effect of the output signal and produce a reference signal (5). The error signal has values e(i). In this embodiment the error signal (12) is used as the input signal and the reference signal is generated from this signal, but in other embodiments the reference signal is derived from the input signals from additional sensor as in U.S. Patent No. 4,122,303 (Chaplin & Smith).

One of the improvements of this invention is the inclusion of an additional filter placed in series with the physical system. A filter (9) may be applied to the input signal or a filter (10) may be applied to the output signal. These filters may, for example, be high pass filters which can be efficiently implemented as a low order Infinite Impulse Response (HR) filters . These filters can be considered part of the physical system. A high pass filter has two benefits. Firstly, it removes any D.C. components in the systems - these may be introduced by non-zero offsets in the ADC for example and are undesirable. Secondly it reduces the low frequency response of the system. When the feedback compensation filter (7) is an FIR filter, a large number of filter coefficients are required to model the low frequency response. The inclusion of a high pass filter reduces the number of coefficients required for a sufficiently accurate model. This reduces the processing requirements of the controller.

A further improvement is in the choice of the number of coefficients for the cancellation filter (4). In the known control filters, which are mainly used for controlling broadband signals, the cancellation filter is related to the inverse model of the physical system. This typically requires a large number of filter coefficients in order to provide the required frequency resolution. However, for tonal noise the cancellation filter response is only important at particular frequencies. This means that a small number of coefficients can be used. Typically two or more coefficients are used for each tone to be canceled. This is only possible when a feedback compensation filter is used. In many of the known techniques a single filter is used to both stabilize the system and to produce the cancellation signal. These techniques require the filter to have a larger number of coefficients in order to prevent the system from becoming unstable.

The reduced number of coefficients of the system of this invention has several advantages. Firstly, the processing requirements are reduced. Secondly, the adaption rate is improved. Thirdly, the system is more selective.

The frequency resolution of the cancellation filter is determined by the sample rate divided by the number of filter coefficients. For example, if the sampling rate of the digital process is 12KHz and six coefficients are used, then the frequency resolution is 2KHz. This means that the controller cannot cancel broadband signals such as speech. Further, the filter has only six degrees of freedom, so when it is used with a Least Mean Square algorithm and applied to a mixed signal it will tend to cancel the loudest components first. The net result is that the control system will tend to be selective in that it will cancel the noise rather than the speech, and in addition there will be limited distortion of the speech because of the low frequency resolution of the controller.

When FIR filters are used, the filter equations are

nc-\ z(i) = ∑C(i-j)yU),

_•^"=<>

na-l y(i) = ∑A i-j)^χϋ)

J=<3 where nc -is the number of coefficients in the feedback compensation filter and na is the number of coefficients in the cancellation filter.

Adaption Module The adaption module, (2) in Figures 2 and 3, adjusts the coefficients of the cancellation filter (4). There are many algorithms which can be used to adjust these coefficients. The simplest is a Filtered-x Least Mean Square (LMS) type of adaption algorithm which is one of the many algorithms described in B. Widrow and S.D. Stearns "Adaptive Signal Processing", Prentice Hall, 1985. The use of this type of adaption for a feedback control system is disclosed in U.S. Patent No. 5,105,377 (Ziegler). In this algorithm the reference signal (5), x(i), is filtered by a model of the system response (11) before being used together with the error signal (12) to adjust the coefficients of the cancellation filter. In the LMS adaption, the correlation of the two signals over one or more samples is estimated. In multi-channel applications (see patent applications PCT/US92/07650 (Ziegler) or PCT/GB91/01850 (Ross and Eatwell)) the adjustment uses an estimate of the cross-correlation matrix of the filtered reference signals and the error signals. In more sophisticated algorithms the inverse cross-correlation is estimated. The filters (7) and (11) may have the same characteristics.

The inclusion of the high pass filters (9) or (10) can reduce the number of coefficients in the filter (11). This reduces the computational requirements for the digital processor.

The adaption rate of the LMS algorithm is dependent upon the convergence step size, this in turn must be limited in amplitude so that the adaption process remains stable. Unless a complicated Recursion Algorithm is used, the step size must be scaled to be inversely proportional to the number of coefficients (see for example S.D. Sommerfeldt, "Adaptive Vibration Control of Vibration Isolation Mounts Using an LMS-based Control Algorithm", (Ph.D. Thesis, The Pennsylvania State University, University Park, PA, August 1989, pp 84-88)). Hence, the adaption rate of the system is improved by limiting the number of coefficients in the cancellation filter (4). The noisy, filtered-x, LMS algorithm and the n-th update gives the new cancellation coefficients in terms of the old cancellation coefficients, the filtered reference signal r(i) , and the error signal e(i). The operation of the adaption module is described by the update equations

nc-\ r(ⁱ) = ∑C(ⁱ- ^"λ =0

A" (k) = (1 - μλ )A"^~l (lc) - μ r(i - k)e(i) where nc is the number of coefficients in the FIR model of the system response, λ is a leakage parameter and μ is a convergence parameter. This method of adaption is referred to as a 'noisy, filtered-x LMS method with coefficient leakage'. Either or both of the signals r(i) and e(i) may be pre-filtered prior to use in the update equation. This pre- filtering can be used to improve convergence as described in J. Triechler, C.R. Johnson and M. Larimore, "Theory and Design of Adaptive Filters" Wiley, New York, 1987, or it can result from weighting the cost function used to derive the update equations.

Parameter Adjustment Module The main parameters used in the LMS adaption algorithm are the coefficient leakage parameter, λ , and the convergence step size, μ. These are described in Chapter 4 of J. Triechler, C.R. Johnson and M. Larimore, "Theory and Design of Adaptive Filters" Wiley, New York, 1987. In the known DVE control system these parameters are fixed. However, improved performance can be achieved if these parameters are varied continuously.

For feedforward signal processing systems some forms of parameter variation are known. For example, a normalized LMS algorithm is described in A.E. Albert and L. A. Gardener, "Stochastic Approximation and Non-Linear Regression", MIT Press, Cambridge, 1967. A Normalized LMS algorithm which normalizes the convergence step size by a recursive estimate of the power of the input signal is described on pages 83-84 of Triechler et al.

For use with active noise control systems the LMS algorithm must be modified to allow for the modification of the output signal by the physical system. The modification results in the filtered-x LMS algorithm described in B. Widrow and S.D. Stearns "Adaptive Signal Processing", Prentice Hall, 1985. A similar normalization using the power of the input signal can be achieved for this algorithm. This approach has been used for feedforward active vibration control (see for example L.B. Bischoff, "Multichannel Adaptive Vibration Control of a Mounted Plate", M.Sc. Thesis, The Pennsylvania State University, University Park, PA, May 1991, pp 39-40). All of these modifications have been used with feedforward control systems, however, equivalent approaches could be used for feedback control.

The known normalization method is primarily designed for stationary signals and the normalization compensates for the level of the input signal. For transient tonal systems this is not sufficient and a further modification is required. The convergence of the adaption algorithm is determined by the level of the noise and also by the loop gain of the system. This loop gain is different at different frequencies, but it is only the gain at the frequencies of the noise that is important. This can be recognized if the input signal is filtered through the model of the physical system before the power measurement is made. Thus , in the preferred embodiment, the step size is normalized by an estimate of the power in the filtered input (filtered-x) signal.

In one embodiment the convergence step size is adjusted according to

μ-- μ_ I power (r)

where the power of the filtered reference signal, r , can be estimated from a suitable norm of r , such as

- -_n . 2/π

where the angled brackets denote some expected or averaged value, and n can take different values. «=/ is useful for implementation on fixed point signal processors. «=2 gives the most accurate estimate.

The leakage parameter can be varied according to an estimate of the power in the output signal.

The leakage parameter can be varied according to an estimate of the power in the input signal, so as, for example, to make the system more robust in the presence of sensor failure.

The level of the leakage parameter can be used to check the stability of the system. This improves the robustness of the control system.

Known on-line system identification can be included to maintain a good estimate of the physical system response. These involve adding a test signal to the controller output. This same signal is passed through the model of the system. The difference between the resulting signal and the sensor response is used, together with the test signal to adapt the model of the system. Examples of this technique are disclosed in U.S. patent application no. 08/015,195 (Eatwell).

Active Headset for Emergency vehicles

The modified feedback control system described above can be used with a headset to provide an active ear defender. One embodiment of an active noise reducing headset incorporating the current invention is shown in Figure 4. The figure shows the control system for one ear only, in practice two independent systems are used, one for each ear. A sensor mounted on the headset close to the ear of the wearer provides the residual signal (12), and a loudspeaker or other actuator is used to generate the canceling noise. The actuator is driven by the output signal (6). The DAC and ADC and associated signal conditioning are omitted from the diagram for the sake of clarity. The headset can be open backed or closed backed. The active headset will provide protection from low frequency random noise or from tonal or narrowband noise.

There are many environments where noise levels are high enough to cause hearing damage and yet communication is required. An example is in the cab of an emergency vehicle such as police car, fire engine or ambulance. Here the source of the noise is the siren which has to generate very high sound levels in order to be effective. The existing solution is to use passive ear defenders together with an electrical communications system. This is expensive, uncomfortable and all of the headsets must be wired together for communication. A further disadvantage is that all other noise - such as the noise from other sirens for example - is blocked out.

Known active headsets are not effective in this environment. This is because the frequency of the siren signal is above the range of active headsets with fixed feedback control systems. In addition, the siren noise signal can vary very rapidly in both frequency and amplitude. A typical emergency vehicle siren can generate several different sound patterns.

A typical noise pattern from a siren manufactured in Japan is known as the "Fast Wail". In this siren the signal produced by the siren speaker is a high level square wave with a cyclically varying frequency as shown on Figure 5. The modulation cycle is over a four second period with a fast rise time and a slower fall time. The frequency varies from 400 Hz to 800 Hz during each cycle.

The noise heard in the vehicle cabin is more complex than the original drive signal due to the acoustics of the cabin. Resonances and multiple acoustic paths cause the amplitude and phase to vary rapidly as the frequency changes.

Known adaptive feedback systems, such as that described in U.S. Patent No. 5, 105,377 (Ziegler), are not effective for the rapid change in frequency or amplitude. The method of U.S. Patent No. 4,654,871 (Chaplin et al), which uses a frequency input, is also ineffective for the rapid amplitude variations.

The active headset of the current invention can be used in this environment since it has the ability to control rapid variations in both amplitude and frequency. The selective nature of the cancellation can eliminate the need for electrical communication channels between the occupants of the cab.

As described before the siren noise is a simple waveform (only two or three harmonics with the dominant energy at the fundamental rate) which has a rapidly changing fundamental rate and due to the cab acoustics, also has a rapidly changing amplitude at the ear. The siren noise is also the loudest noise in the cab. It is the combination of the simplicity of the noise and its dominance that allows a solution.

Figure 6 is a block diagram of one embodiment of the Digital Signal Processing (DSP) hardware capable of executing the cancellation algorithm. It is a one board system using two channels of processing (one for each ear) and is driven by the vehicle 12 volt power system (29). The hardware includes anti-aliasing filters (30, 31) which limit the upper frequency of the residual signals to that of the noise, about 2500 Hz for the described siren signal. The filter are designed to reject energy outside this band to avoid aliasing or the tendency for signals above one half of the sampling rate to get translated in frequency by the sampling process while minimizing the time delay incurred in the filtering process. The residual noise signals are converted to digital form by analog-to-digital converters (32, 33). They operate at a 10kHz sampling rate to minimize delay and have 12 bit accuracy to have sufficient range to handle changing noise conditions. The converters feed a 16 bit DSP processor (34) such as the TI TMS 320C25.

The algorithm is written in the assembly language of the DSP processor to maintain efficient use of the processing resources. The machine language program is stored in an external ROM chip (35). Fed from the processor (34), the signals pass to digital-to-analog converters (36, 37) which are used to generate the output anti-noise signals. Converters (36, 37) are also designed to minimize signal delay and have 12 bit accuracy at the 10 kHz sampling rate. The converter output is then fed to reconstruction filters (38, 39) which remove the copies of the desired anti-noise at frequencies that are multiples of the sampling rate. Power supply (29) may be a high efficiency switching supply that accepts the vehicle battery voltage (nominally 12 volts) and generates the required set of precision voltages for the analog and digital electronics in controller (18).

Figure 7 shows the measured results of using the system in Figure 4 in a vehicle cab. The upper half shows five seconds, the modulation cycle is four seconds, of the original noise at the ear with the active cancellation turned off. The chart shows the complexity and rate of change of this noise due to the complex acoustical path taken by the noise within the cab. The lower half of the chart shows the result when the active cancellation is turned on. The system reliably reduces the siren noise by 10 to 15 dB with little impact on the other external sounds. In this example six coefficients were used for the FIR cancellation filter and the filter was run at a sampling rate of 10kHz. The controller therefore has sufficient degrees of freedom to control three harmonic components of the noise. This is sufficient because the vehicle cabin acts as a low pass acoustic filter. The control systems does not have sufficient degrees of freedom to cancel other lower level noises such as communication signals.

A communications signal, for communications from other vehicles or dispatchers, can be added to the controller output. The selective nature of the controller ensures that the speech is not canceled by the action of the controller. Further, any tonal noise present in the communications signal will be canceled or reduced.

Claims

1. A control system for attenuating an unwanted disturbance, said system comprising input means for receiving a residual signal related to a residual disturbance, cancellation module means responsive to an input signal and adapted to produce an output signal, output means receiving said output signal and causing a counter disturbance, adaption module means responsive to said residual signal and a signal derived from said input signal, and a parameter adjustment module responsive to said output signal and/or a signal derived from said input signal and producing adaption parameters, said system characterized in that the operating characteristic of said adaption module is determined by said adaption parameters.

2. A control system as in claim 1 in which said input signal is derived from said residual signal.

3. A control system as in claim 1 in which said input signal is derived from an additional signal related in part to said unwanted disturbance.

4. A control system as in claim 1 in which the adaption module uses coefficient leakage with a leakage parameter adjusted according to the level of the controller output.

5. A system as in claim 1 in which the adaption module adjusts the cancellation filter coefficients according to the product of a convergence step size parameter and an estimate of the correlation between said residual signal and a signal derived from said input signal and in which the convergence step size parameter is adjusted according to the level of a signal derived from the input signal.

6. A control system as in claim 1 in which the signal derived from the input signal is produced by filtering the input signal by a model of the system response.

7. A control system as in claim 1 in which the adaption module uses a noisy, filtered-x LMS method with coefficient leakage.

8. A system as in claim 1 in which the cancellation module means includes an adaptive cancellation filter means and a feedback compensation filter means.

9. A system as in claim 8 in which the number of coefficients in the cancellation filter means is limited to twice the number of frequency components in the unwanted disturbance.

10. A system as in claim 8 and including a fixed filter means in series with said adaptive cancellation filter means.

11. An adaptive system as in claim 8 in which said cancellation filter means and said feedback compensation filter means are sampled data devices.

12. A system as in claim 1 in which said residual signal and said input signal are provided by sound sensing means and in which the output means is a sound producing means characterized in that the system is adapted to attenuate unwanted sound at a listener's ear.

13. A system as in claim 12 in which the sound sensing means and the sound producing means are mounted on a headset.

14. A control means for attenuating unwanted disturbances, said means comprising a pair of control systems commonly mounted on a digital signal processing chip, each said control systems comprising input means for receiving a residual signal related to a residual disturbance, cancellation module means responsive to an input signal and adapted to produce an output signal, output means receiving said output signal and causing a counter disturbance, adaption module means responsive to said residual signal and a signal derived from said input signal, and a parameter adjustment module responsive to said output signal and/or a signal derived from said input signal and adapted to produce adaption parameters, said system characterized in that the operating characteristic of said adaption module is determined by said adaption parameters.

15. A system as in claim 14 in which said input signals are derived from said residual signals.

16. A system as in claim 14 in which said input signals are derived from additional signals related in part to said disturbance.

17. A control means as in claim 14 in which the adaption module uses coefficient leakage with leakage parameters adjusted according to the level of the controller outputs.

18. A control means as in claim 14 in which each adaption module adjusts the cancellation filter coefficients according to the product of a convergence step size parameter and an estimate of the correlation between said residual signal and said signal derived from input reference signal and in which the convergence step size parameter is adjusted according to the level of a signal derived from the input signal.

19. A control means as in claim 14 in which the signals derived from the input signals are produced by filtering the reference signals by a model of the corresponding system response.

20. A control means as in claim 14 in which the adaption module uses a noisy, filtered-x LMS method with coefficient leakage.

21. A system as in claim 14 in which said residual signals and said input signals are provided by sound sensing means and in which the output means are sound producing means characterized in that the system is adapted to attenuate unwanted sound at a listener's ear.

22. A system as in claim 21 in which the sound sensing means and the sound producing means are mounted on a headset.

23. A control means as in claim 22 in which a communication signal is added to one or both of the output signals.