Tesis_anc

  • Upload
    fakmilo

  • View
    222

  • Download
    0

Embed Size (px)

Citation preview

  • 8/6/2019 Tesis_anc

    1/65

    Institutionen fr systemteknikDepartment of Electrical Engineering

    Masters thesis

    Active Noise Control of a Forest Machine Cabin

    Masters thesis in Automatic Controlat Linkping university by

    Patrik GrylinMrten Hedborg

    LITH-ISY-EX--07/4014--SELinkping 2007

    Department of Electrical Engineering Linkpings tekniska hgskolaLinkpings universitet Linkpings universitetSE-581 83 Linkping, Sweden 581 83 Linkping

  • 8/6/2019 Tesis_anc

    2/65

  • 8/6/2019 Tesis_anc

    3/65

    Active Noise Control of a Forest Machine Cabin

    Masters thesis in Automatic Controlat Linkping university by

    Patrik GrylinMrten Hedborg

    LITH-ISY-EX--07/4014--SE

    Supervisor: Johan Sjbergisy

    , Linkpings universitetKjell RnnholmKomatsu Forest AB

    Examinator: Fredrik Gustafssonisy, Linkpings universitet

    Linkping, 22 March, 2007

  • 8/6/2019 Tesis_anc

    4/65

  • 8/6/2019 Tesis_anc

    5/65

    Avdelning, InstitutionDivision, Department

    Division of Automatic ControlDepartment of Electrical EngineeringLinkpings universitetSE-581 83 Linkping, Sweden

    DatumDate

    2007-03-22

    SprkLanguage

    Svenska/Swedish

    Engelska/English

    RapporttypReport category

    Licentiatavhandling

    Examensarbete

    C-uppsats

    D-uppsats

    vrig rapport

    URL fr elektronisk version

    http://www.control.isy.liu.sehttp://www.ep.liu.se/2007/4014

    ISBN

    ISRN

    LITH-ISY-EX--07/4014--SE

    Serietitel o ch serienummerTitle of series, numbering

    ISSN

    TitelTitle

    Aktiv Bullerdmpning av Frarhytt p Skogsmaskin

    Active Noise Control of a Forest Machine Cabin

    FrfattareAuthor

    Patrik Grylin Mrten Hedborg

    SammanfattningAbstract

    Today, a high noise level is considered a problem in many working environments.The main reason is that it contributes to stress and fatigue. Traditional methodsusing passive noise control is only practicable for high frequencies. As a comple-ment to passive noise control, active noise control (ANC) can be used to reducelow frequency noise. The main idea of ANC is to use destructive interference ofwaves to cancel disturbing noises.

    The purpose of this thesis is to design and implement an ANC system in

    the drivers cabin of a Valmet 890 forest machine. The engine boom is one of themost disturbing noises and therefore the main subjective for the ANC system tosuppress.

    The ANC system is implemented on a Texas Instrument DSP developmentstarter kit. Different FxLMS algorithms are evaluated with feedback and feedfor-ward configurations.

    The results indicate that an ANC system significantly reduces the soundpressure level (SPL) in the cabin. Best performance of the evaluated systemsis achieved for the feedforward FxLMS system. For a commonly used enginespeed of 1500 rpm, the SPL is reduced with 17 dB. The results show fast enoughconvergence and global suppression of low frequency noise.

    NyckelordKeywords Active Noise Cancellation, Adaptive Filtering, Active Noise Control, Feedforward,

    Feedback, Active Noise Reduction

  • 8/6/2019 Tesis_anc

    6/65

  • 8/6/2019 Tesis_anc

    7/65

    Abstract

    Today, a high noise level is considered a problem in many working environments.The main reason is that it contributes to stress and fatigue. Traditional methodsusing passive noise control is only practicable for high frequencies. As a comple-ment to passive noise control, active noise control (ANC) can be used to reducelow frequency noise. The main idea of ANC is to use destructive interference ofwaves to cancel disturbing noises.

    The purpose of this thesis is to design and implement an ANC system in thedrivers cabin of a Valmet 890 forest machine. The engine boom is one of themost disturbing noises and therefore the main subjective for the ANC system tosuppress.

    The ANC system is implemented on a Texas Instrument DSP development starterkit. Different FxLMS algorithms are evaluated with feedback and feedforwardconfigurations.

    The results indicate that an ANC system significantly reduces the sound pressurelevel (SPL) in the cabin. Best performance of the evaluated systems is achievedfor the feedforward FxLMS system. For a commonly used engine speed of 1500rpm, the SPL is reduced with 17 dB. The results show fast enough convergenceand global suppression of low frequency noise.

    v

  • 8/6/2019 Tesis_anc

    8/65

  • 8/6/2019 Tesis_anc

    9/65

    Acknowledgments

    First of all, we would like to express our gratitude to Fredrik Gustafsson and Jo-han Sjberg at Linkping university. Of course, for their help during our mastersthesis but most of all for being excellent and inspiring teachers.

    Kjell Rnnholm, Per Holmberg, Per Hedstrm and others - building the worldsbest forest machines - for help and support during our time at Komatsu Forest.

    Mrten Nygren for kickstarting our project with valuable comments and sugges-tions.

    Josse, grandpa Lasse, Fredrik & Co, Oscar G. & Kent P. at the division of Elec-tronic Systems, Y3c, Kuratorvgen and Uno-x in Bjurholm deserves to be men-tioned as well.

    Patrik Grylin and Mrten Hedborg

    vii

  • 8/6/2019 Tesis_anc

    10/65

  • 8/6/2019 Tesis_anc

    11/65

    Contents

    1 Introduction 3

    1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

    1.2 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

    1.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

    1.4 The concept of ANC . . . . . . . . . . . . . . . . . . . . . . . . . . 4

    1.5 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

    2 ANC principles 7

    2.1 Acoustics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

    2.1.1 Wave equation . . . . . . . . . . . . . . . . . . . . . . . . . 7

    2.1.2 Principle of superposition . . . . . . . . . . . . . . . . . . . 7

    2.2 Feedforward system overview . . . . . . . . . . . . . . . . . . . . . 8

    2.2.1 Primary path . . . . . . . . . . . . . . . . . . . . . . . . . . 9

    2.2.2 Secondary path . . . . . . . . . . . . . . . . . . . . . . . . . 9

    2.2.3 Digital filter . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

    2.3 Feedback system overview . . . . . . . . . . . . . . . . . . . . . . . 10

    2.3.1 Secondary path . . . . . . . . . . . . . . . . . . . . . . . . . 10

    2.3.2 Digital filter . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

    3 ANC adaptive filter theory 13

    3.1 Finite Impulse Response . . . . . . . . . . . . . . . . . . . . . . . . 13

    3.2 Least Mean Square . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

    3.2.1 Normalized LMS . . . . . . . . . . . . . . . . . . . . . . . . 14

    3.2.2 Leaky LMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

    3.3 Feedforward filtered-x LMS . . . . . . . . . . . . . . . . . . . . . . 16

    3.3.1 Feedforward Multi Channel FxLMS . . . . . . . . . . . . . 17

    3.4 Feedback FxLMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

    3.4.1 Feedback MC-FxLMS . . . . . . . . . . . . . . . . . . . . . 20

    3.5 Secondary path identification . . . . . . . . . . . . . . . . . . . . . 213.5.1 Choosing an input signal u(n) . . . . . . . . . . . . . . . . . 21

    3.5.2 System latency . . . . . . . . . . . . . . . . . . . . . . . . . 22

    ix

  • 8/6/2019 Tesis_anc

    12/65

    4 Implementation 234.1 DSP hardware and equipment . . . . . . . . . . . . . . . . . . . . . 23

    4.1.1 Development Starter Kit . . . . . . . . . . . . . . . . . . . . 244.1.2 DSP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

    4.2 DSP software concepts and techniques . . . . . . . . . . . . . . . . 25

    4.2.1 Code composer studio . . . . . . . . . . . . . . . . . . . . . 254.2.2 DSP/BIOS . . . . . . . . . . . . . . . . . . . . . . . . . . . 254.2.3 PING-PONG buffering . . . . . . . . . . . . . . . . . . . . . 25

    4.3 DSP program design . . . . . . . . . . . . . . . . . . . . . . . . . . 264.3.1 Main function . . . . . . . . . . . . . . . . . . . . . . . . . . 264.3.2 DSP/BIOS threads . . . . . . . . . . . . . . . . . . . . . . . 274.3.3 ANC system states . . . . . . . . . . . . . . . . . . . . . . . 27

    4.4 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . . . 27

    5 Results 295.1 Feedforward FxLMS . . . . . . . . . . . . . . . . . . . . . . . . . . 29

    5.1.1 Primary path . . . . . . . . . . . . . . . . . . . . . . . . . . 295.1.2 Secondary path . . . . . . . . . . . . . . . . . . . . . . . . . 31

    5.1.3 Constant engine speeds . . . . . . . . . . . . . . . . . . . . 315.1.4 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

    5.2 Feedback FxLMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385.2.1 Secondary path . . . . . . . . . . . . . . . . . . . . . . . . . 385.2.2 Constant engine speeds . . . . . . . . . . . . . . . . . . . . 385.2.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

    5.3 Global noise reduction in the cabin . . . . . . . . . . . . . . . . . . 435.4 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

    6 Concluding Remarks 476.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

    Bibliography 49

    A Equipment 51

  • 8/6/2019 Tesis_anc

    13/65

    Nomenclature

    Signals

    x(n) Reference signald(n) Primary noise at the error sensor positiony(n) Adaptive filter outputy(n) Adaptive filter output after passing through a secondary path

    e(n) Error signalx(n) Reference signal after passing through a secondary path estimatione0(n) Gaussian random noisev(n) Generated Gaussian random noiseu(n) Low pass filtered input for system identificationy(n) u(n) after passing through a secondary path including unwanted noiseX(z) Z-transform of reference signalE(z) Z-transform of error signal

    Transfer functions

    P(z) Primary pathF(z) General filter

    S(z) Secondary pathS(z) Secondary path estimation

    S(z|s) Secondary path estimation during its identificationW(z|w) FIR filterH(z) Transfer function representing unwanted noise, driven by Gaussian random noiseLP(z) Low-pass filter

    Impulse responses, filter parameters

    w, w(n) Filter parameters ofW(z|w)s(n) Impulse response ofS(z)p(n) Impulse response ofP(z)

    s(n) Impulse response of S(z)s Filter parameters of S(z)

    s Filter parameters of S(z|s)

    1

  • 8/6/2019 Tesis_anc

    14/65

    2 Contents

    Constants

    Step size Very small number Leaky constant

    1 N Number of parameters in w(n)J Number of reference sensorsM Number of error sensorsK Number of secondary sourcesn Time delay of secondary pathM Number of delays for calculating cross correlationN Number of samples when calculating cross correlation

    Abbreviations

    AD Analog DigitalANC Active Noise Control

    BIOS Basic Input Output SystemCCStudio Code Composer StudioCPU Central Processing UnitDA Digital AnalogDRAM Dynamic Random Access MemoryDSK Development Starter KitDSP Digital Signal ProcessorEDMA Enhanced Direct Memory AccessFIR Finite Impulse ResponseFxLMS Filtered x Least Mean SquareLMS Least Mean SquareMcBSP Multi Channel Buffered Serial PortMC-FxLMS Multi Channel Filtered x Least Mean Square

    MSE Mean Square ErrorNLMS Normalized Least Mean Squarerpm Revolutions Per MinuteSPL Sound Pressure LevelTI Texas InstrumentsUSB Universal Serial Bus

  • 8/6/2019 Tesis_anc

    15/65

    Chapter 1

    Introduction

    1.1 Background

    Today, a high noise level is considered a problem in many working environments.The main reason is that it contributes to stress and fatigue [3]. Traditional meth-ods using passive noise control are only practicable for high frequencies. Thereason is that low frequency noise has long wavelengths compared to a typicalacoustic absorber [4]. As a complement to passive noise control, active noise con-trol (ANC) can be used to reduce low frequency noise. The main idea of ANC isto use destructive interference of waves to cancel disturbing noises. The theoryof ANC has been known for a long time but it is the development of fast digitalsignal processors (DSP:s) that have made practical applications feasible.

    1.2 Purpose

    The purpose of this thesis is to reduce the noise level in the drivers cabin of a

    Valmet 890 forest machine, see Figure 1.1, by using an ANC system. One of themost disturbing noises is the engine boom which is the main noise to be suppressedby the ANC system.

    1.3 Method

    The project was divided into four main parts:

    Literature studies: Old reports and articles concerning ANC were examined forideas and knowledge.

    Design of a prototype: To be able to build a prototype, system hardware hadto be specified.

    Implementation: ANC algorithms were implemented in a DSP.

    Evaluation: The algorithms were evaluated online in the cabin.

    3

  • 8/6/2019 Tesis_anc

    16/65

    4 Introduction

    Figure 1.1. Valmet 890.

    1.4 The concept of ANC

    The concept of ANC is to cancel a disturbing primary noise by generating ananti noise of equal amplitude and opposite phase. The primary noise combinedwith the anti noise results in a cancellation of both noises. To generate the antinoise, two main configurations of ANC systems are often used, feedforward andfeedback control, both using adaptive filters [4, 9]. For a feedforward ANC system

    two sensors are used. First a reference sensor is used to get a reference signaland second an error sensor is used to update the adaptive filter by measuring theresidual of the anti noise and primary noise, see Figure 1.2. The feedback ANCsystem uses only one sensor, namely an error sensor. This sensor is used both tocreate the reference signal and to update the adaptive filter, see Figure 1.3. Bothfeedforward and feedback configurations will be evaluated in this thesis.

  • 8/6/2019 Tesis_anc

    17/65

    1.5 Thesis outline 5

    ANC

    Algorithm

    Error SensorReference Sensor

    Primary Noise Anti-Noise

    Figure 1.2. An ANC feedforward system.

    ANC

    Algorithm

    Error Sensor

    Primary Noise Anti-Noise

    Figure 1.3. An ANC feedback system.

    1.5 Thesis outline

    Chapter 1 An introduction to the thesis where the background and purpose are

    presented.Chapter 2 Fundamental principles for active noise control are introduced.

    Chapter 3 A presentation of ANC adaptive theory is given.

    Chapter 4 The implementation of the system in a DSP is discussed.

    Chapter 5 The results of the work are given.

    Chapter 6 Conclusions and suggestions for future work are presented.

  • 8/6/2019 Tesis_anc

    18/65

    6 Introduction

  • 8/6/2019 Tesis_anc

    19/65

    Chapter 2

    ANC principles

    In this chapter fundamental ANC principles and phenomenas are introduced.

    2.1 Acoustics2.1.1 Wave equation

    A three dimensional harmonic wave can be expressed as

    (r, t) = Aekrt (2.1)

    with angle velocity , time t, and

    k r = kxx + kyy + kzz (2.2)

    where (kx, ky, kz) are the components of the propagation direction and (x,y,z) arethe components of the point in space where the displacement is evaluated.

    The differential wave equation satisfied by the harmonic wave (2.1) is given by

    2 1

    c20

    2

    t2= 0 (2.3)

    where the Laplacian operator is defined as

    2 =2

    x2+

    2

    y2+

    2

    z2(2.4)

    and c0 is the propagation speed in air. More information about wave equationscan for example be found in [12].

    2.1.2 Principle of superpositionThe fundamental physical phenomena making an ANC system possible is theprinciple of superposition. The principle of superposition states that a linear

    7

  • 8/6/2019 Tesis_anc

    20/65

    8 ANC principles

    combination of solutions is also a solution to the same linear system. To be morelegible, if 1 and 2 are independent solutions of the wave equation (2.3), thenthe linear combination

    = a1 + b2 (2.5)

    where a and b are constants, is also a solution [12].

    Since the propagation of an acoustic wave is described fairly well by (2.3) alsoin practise, it is possible to use an anti noise of equal amplitude and oppositephase, see Figure 2.1.

    Hence the objective for an ANC system is to measure the primary noise andgenerate the anti noise.

    0 1 2 3 4 5 6 7 8 9 101

    0.8

    0.6

    0.4

    0.2

    0

    0.2

    0.4

    0.6

    0.8

    1

    Primary noise

    Secondary noise

    Primary noise + Secondary noise

    Figure 2.1. Principle of superposition.

    2.2 Feedforward system overview

    In Figure 2.2, an overview of an ANC feedforward system is illustrated. Thesummation junction in the block diagram represents an acoustic superposition ofsound waves.

  • 8/6/2019 Tesis_anc

    21/65

    2.2 Feedforward system overview 9

    P(z)

    F(z) S(z)

    x(n)

    y(n)

    d(n)

    y(n)

    e(n)

    Electrical domain Acoustical domain

    Figure 2.2. Overview of an ANC feedforward system.

    2.2.1 Primary path

    The primary path P(z) is the acoustic transfer function from the reference sensorclose to the noise source, to the error sensor at the point where the ANC system isoperating. The primary noise being measured in an upstream position is referredto as x(n) and the primary noise at the error sensors is referred to as d(n).

    2.2.2 Secondary path

    The secondary path S(z) describes the transfer function from the filter outputy(n) to y(n), the acoustic point where the error sensor is placed. This transferfunction includes computer system delays, amplifier and speaker dynamics and theacoustic path between speaker and error sensor. The speaker generating the antinoise is often referred to as the secondary source.

    2.2.3 Digital filter

    The noise x(n) is usually measured with a microphone and used by the digital filterF(z) as a reference signal. As an output, the digital filter produces the signal y(n)which is phase shifted and driven through the secondary source to become y(n)at the point where to cancel out d(n). The acoustic signal y(n) is often referredto as the anti noise or secondary noise. The error signal e(n) = d(n) y(n) atthis point is measured with a microphone and used by the digital filter to improvethe estimation y(n).

    In general, the digital filter F(z) contains an estimation W(z|w) of the trans-fer function P(z)S1(z) and an adaptive algorithm which derives the estimated

    parameters w. The estimation W(z|w) is normally a finite impulse response (FIR)filter, that is a standard filter in acoustic path modeling [6], see Section 3.1.

  • 8/6/2019 Tesis_anc

    22/65

    10 ANC principles

    To improve F(z), other subsystems such as an estimation of S(z) are often in-cluded and some are described in Section 3.

    A fundamental system limitation

    For a feedforward system, the time delay of P(z) has to be longer then the timedelay ofF(z) and S(z). If this causality condition is met, the ANC system is capa-ble of cancelling random broad-band noise. Otherwise, only predictable periodicnoise can be cancelled out.

    2.3 Feedback system overview

    In Figure 2.3, an overview of an ANC feedback system is illustrated. The differencefrom a feedforward system is that no reference sensor is used and thus no primarypath exists. Instead an internal synthesized reference signal is used to generatey(n), see Section 3.4.

    F(z) S(z)

    d(n)

    y(n)

    y(n)

    e(n)

    Electrical domain Acoustical domain

    Figure 2.3. Overview of an ANC feedback system.

    2.3.1 Secondary path

    As for the feedforward system, a secondary path exists for the feedback system.For more information, see Section 2.2.2

    2.3.2 Digital filter

    The digital filter works the same way as in a feedforward system, see Section 2.2.3.The main difference is that the reference signal is synthesized instead of measuredfrom a reference microphone.

  • 8/6/2019 Tesis_anc

    23/65

    2.3 Feedback system overview 11

    A fundamental system limitation

    Since no reference signal from the primary noise exists in a feedback system, goodperformance is only achieved for periodic noise.

  • 8/6/2019 Tesis_anc

    24/65

    12 ANC principles

  • 8/6/2019 Tesis_anc

    25/65

    Chapter 3

    ANC adaptive filter theory

    Here different adaptive filter algorithms for ANC applications are presented.

    3.1 Finite Impulse ResponseThe most commonly used filter for ANC applications is the FIR filter. The as-sumption of a FIR filter is that the output y(n) depends on a weighted combinationof a finite number N of past input values x(n), expressed as

    y(n) =

    Ni=0

    wix(n i) (3.1)

    A FIR filter has a number of useful properties, e.g., all poles are at the originwhich means that it is bounded input bounded output (BIBO) stable [5].

    3.2 Least Mean SquareIn this section the least mean square (LMS) algorithm for a system using a singlereference input, a single error sensor and a single secondary source is derived. AFIR filter W(z|w) of order N and the signal definitions according to Figure 3.1are used. Assume that the output of the filter is given as

    y(n) = x(n)Tw(n) (3.2)

    wherex(n) =

    x(n 1),...,x(nN)

    T(3.3)

    andw(n) = w1(n), w2(n),...,wN(n)

    T. (3.4)

    Using (3.3) and (3.4), an expression for the error signal can be found as

    e(n) = d(n) y(n) = d(n) s(n) [x(n)Tw(n)] (3.5)

    13

  • 8/6/2019 Tesis_anc

    26/65

    14 ANC adaptive filter theory

    P(z)

    W(z|w)

    LMS

    S(z)

    x(n)

    y(n)

    d(n)

    y(n)

    e(n)

    Electrical domain Acoustic domain

    Figure 3.1. Block diagram of a feedforward LMS system.

    where s(n) is the impulse response of the transfer function S(z). The primary dis-turbance signal is given by d(n) = p(n) x(n), where p(n) is the impulse responseofP(z), and the secondary noise is given by y(n) = s(n) y(n).

    Assuming a mean square error (MSE) as a theoretical cost function

    V(w) = E{e2(n)} (3.6)

    the LMS-algorithm minimizes an estimation of (3.6), that is,

    V(w) = e2(n). (3.7)

    Using the steepest descent algorithm the LMS-algorithm is finally given by

    w(n + 1) = w(n)

    2

    dV(w)

    dw(3.8)

    where is the step size, see Section 3.2.1. An example of the steepest descentalgorithm is illustrated in Figure 3.2.

    3.2.1 Normalized LMS

    The step size, in (3.8), is a design parameter that affects the convergence rate

    and the MSE. Large step sizes increase the convergence rate but will also result inan increased MSE, while small step sizes lead to small MSE but slow convergencerate. To decrease the influence of the amplitudes of x(n) and d(n), is normalized

  • 8/6/2019 Tesis_anc

    27/65

    3.2 Least Mean Square 15

    50

    0

    50

    50

    0

    50

    0

    1000

    2000

    3000

    4000

    5000

    w1

    w2

    V(w)

    Figure 3.2. Convergence of V(w) towards the global optimum using LMS for a FIRfilter of order 2.

    by the energy of x(n) i.e.

    (n) =

    + xT(n)x(n)(3.9)

    often referred to as normalized LMS (NLMS). To prevent singularities when xT(n)x(n)is very small, a small number is introduced.

    3.2.2 Leaky LMS

    ANC systems are often overloading their secondary sources, making the systemnonlinear. Instead of introducing output power constraints, a modified cost func-tion

    Vl(w) = e2(n) + wT(n)w(n) (3.10)

    including the weighted filter parameters can be used. This approach has severaladvantages. It has a stabilizing effect on the adaptive filter, and in finite-precisionimplementations it reduces numerical errors [9]. Using (3.8) and (3.10), the up-dating algorithm of the leaky LMS filter parameters become

    w(n + 1) = w(n)

    2

    dV(w)

    dw(3.11)

    where

    = 1 . (3.12)A disadvantage with introducing the leaky parameter, , is that it adds a biaserror to the residual.

  • 8/6/2019 Tesis_anc

    28/65

    16 ANC adaptive filter theory

    3.3 Feedforward filtered-x LMS

    To be able to use the LMS algorithm (3.8) in an ANC application the system hasto compensate for the effect of the secondary path transfer function. Otherwise,when updating the filter parameters the error signal is not aligned in time with

    the reference signal and this will lead to instability [9]. An algorithm that com-pensates for this effect is derived below.

    From (3.7) we obtain

    dV(w)

    dw=

    de2(n)

    dw= 2[

    de(n)

    dw]e(n) (3.13)

    and the differentiation of (3.5) yields

    de(n)

    dw= s(n) x(n). (3.14)

    Combining (3.8), (3.13) and (3.14) yields

    w(n + 1) = w(n)

    2

    dV(w)

    dw= w(n)

    22de(n)

    dwe(n)

    = w(n) + [s(n) x(n)]e(n). (3.15)

    Here, the secondary path impulse response s(n) is not known and has to be esti-mated by s(n).

    By substituting s(n) with s(n) and defining

    x(n) s(n) x(n) = x(n),...,x(nN+ 1)

    T(3.16)

    the expression (3.15) can finally be written as

    w(n + 1) = w(n) + x(n)e(n) (3.17)

    also known as the feedforward filtered-x LMS (FxLMS) algorithm [11], illustratedin Figure 3.3. The z-transform of the error signal is

    E(z) = [P(z) S(z)W(z|w)]X(z) (3.18)

    and to achieve a residual error equal to zero it is necessary that

    W(z|w) =

    P(z)

    S(z) , (3.19)

    which is possible if the causality condition described in Section 2.2.3 is satisfied.

  • 8/6/2019 Tesis_anc

    29/65

    3.3 Feedforward filtered-x LMS 17

    P(z)

    W(z|w)

    LMSS(z)

    S(z)

    x(n)

    y(n)

    d(n)

    y(n)

    x(n)

    e(n)

    Electrical domain Acoustic domain

    Figure 3.3. Block diagram of a feedforward FxLMS system.

    3.3.1 Feedforward Multi Channel FxLMS

    A commonly used notation for multi channel systems is JMK, meaning thatJ reference inputs, M error sensors and K secondary sources are used. In general,the theory for the single channel system can be applied on the multi dimensionalcase, but some functions have to be redefined.

    For a JMK system the FIR filter can be expressed as

    yk(n) =

    Jj=1

    xj(n)T

    w(n)kj k = 1, 2,...,K (3.20)

    and the cost function is approximated by

    V(w) =Mm=1

    e2m(n). (3.21)

    The parameters of the FIR filters will be updated according to

    wkj(n+1) = wkj(n)+Mm=1

    xjkm(n)em(n) k = 1, 2,...,K; j = 1, 2,...,J (3.22)

    where

    xjkm(n) smk(n) xj(n) k = 1, 2,...,K; m = 1, 2,...,M. (3.23)

  • 8/6/2019 Tesis_anc

    30/65

    18 ANC adaptive filter theory

    The feedforward multi channel FxLMS (MC-FxLMS) system is illustrated in Fig-ure 3.4. The block S represents MK secondary path transfer functions Smk(z).The matrix P represents M J primary path transfer functions Pmj(z) and theK J separate adaptive FIR filters are represented by W.

    P

    W

    LMSS

    S

    x(n)J

    Jy(n)

    K

    d(n)M

    y(n)M

    Mx(n)

    J

    e(n)

    Figure 3.4. Block diagram of a feedforward MC-FxLMS system.

    3.4 Feedback FxLMS

    A feedback FxLMS system is viewed as a feedforward FxLMS system that synte-sizes its own reference signal based on the adaptive filter output and error signal.The reference signal x(n) is synthesized as

    x(n) = e(n) + s(n) y(n) (3.24)

    where s(n) is the impulse response of S(z), e(n) is the signal obtained from theerror sensor and y(n) is the secondary signal generated from the adaptive filter.In Figure 3.5, a feedback FxLMS system is illustrated.

    If S(z) = S(z) the z-transform of the error signal is

    E(z) = [1 S(z)W(z|w)]D(z) (3.25)

    and the block diagram representing this equation is shown in Figure 3.6. In thisway, the feedback problem has been transformed into a feedforward problem withW(z|w) acting as a predictor for the primary noise d(n).

    By viewing the feedback FxLMS system as in Figure 3.6 the importance of two

    statements made earlier is emphasized; the importance of a quality estimate ofS(z) and the need for d(n) to be predictable, see Section 3.3 and 2.3.2, respec-tively.

  • 8/6/2019 Tesis_anc

    31/65

    3.4 Feedback FxLMS 19

    W(z|w)

    LMSS(z)

    S(z)

    S(z)

    d(n)

    y(n)

    y(n)

    x(n)

    x(n) e(n)

    Electrical domain Acoustic domain

    Figure 3.5. Block diagram of a feedback FxLMS system.

    S(z) W(z)

    LMS

    -

    d(n)

    e(n)

    Figure 3.6. Equivalent block diagram to the system in Figure 3.5, under the assumption

    that S(z) = S(z).

  • 8/6/2019 Tesis_anc

    32/65

    20 ANC adaptive filter theory

    3.4.1 Feedback MC-FxLMS

    Extending the earlier presented single channel FxLMS feedback methode to a mul-tiple channel feedback FxLMS system is fairly easy, see Figure (3.7).

    A KM feedback MC-FxLMS system consists ofKM secondary path trans-fer functions Smk(z) from the k:th secondary source to the m:th error sensor,estimated by the corresponding filter Smk(z). To synthesize M reference signals

    W

    LMSS

    S

    S

    d(n)M

    y(n)K

    y(n)M

    Mx(n)

    x(n)M

    e(n)

    Figure 3.7. Block diagram of a feedback MC-FxLMS system.

    xm(n), M error signals em(n), MK secondary path estimates Smk(z) and Ksecondary signals y

    k(n) are used according to

    xm(n) = em(n) +

    Kk=0

    smk(n) yk(n) (3.26)

    where m = 1, 2,...,M.

    The coefficients of the KM adaptive filters Wkm(z) are updated by the MC-FxLMS algorithm similarly as in Section 3.3.1, for the special case J = M.

  • 8/6/2019 Tesis_anc

    33/65

    3.5 Secondary path identification 21

    3.5 Secondary path identification

    Finding an estimate of the secondary path transfer function S(z) is done by systemidentification. As seen below, adaptive filtering is closely related to the widersubject of system identification.

    LP(z)White NoiseGenerator

    H(z)

    LMS

    S(z)

    S(z|s)

    v(n) u(n)

    e0(n)

    y(n)

    e(n)

    Electrical domain Acoustic domain

    Figure 3.8. Block diagram of a system estimating S(z) with S(z|s).

    There are several recursive identification algorithms to choose between. A blockdiagram of a recursive LMS algorithm estimating S(z) with a FIR filter S(z|s) is

    shown in Figure 3.8.

    3.5.1 Choosing an input signal u(n)

    To identify the secondary path low-pass filtered white noise is generated and driventhrough the secondary source. White noise is a random noise signal that has thesame sound energy level at all frequencies, i.e., its spectrum can be written as

    () = C, [0,] (3.27)

    where C is a constant. To assure that a proper model of the secondary path isderived for the lower frequencies, a low-pass filter is used to concentrate the energy

    to these frequencies [10]. The reason for this is explained below.

    First, rewrite the MSE cost function estimate (3.7) using the inverse Fourier trans-

  • 8/6/2019 Tesis_anc

    34/65

    22 ANC adaptive filter theory

    form

    V(s) = e2(n) =1

    2

    e()d (3.28)

    where e() is the spectrum of the prediction errors {e(n)}. The system is givenby

    y(n) = S(z)u(n) + H(z)e0(n) (3.29)

    where S(z) is the true system, u(n) are system inputs and unwanted noise isrepresented by H(z), a inversely stable monic1 filter driven by white noise {e0(n)}.Then, ifu and e0 are independent, the following expressions are valid

    s = arg mins

    V() = arg mins

    S(ei) S(ei|s)2Q()d (3.30)

    Q() =u()

    |H(ei)|2

    (3.31)

    where s are the FIR filter parameters of S(z) identified by the process.For all frequencies, the difference between S(z) and S(z) is weighted with Q(w)which is dependent of the input energy for all frequencies. Hence the lower fre-quencies will be dominating in the minimization (3.30) when using a low-pass filter[10].

    3.5.2 System latency

    The secondary path S(z) has a latency that needs to be included in the estimationand it is measured in amount of samples n. Cross correlation is a method forestimating the similarity between signals, in this case u(n) and y(n). A highcross correlation value corresponds to high similarity between signals.

    By calculating the cross correlation between u(n) and y(n + )

    Ruy() = limN

    1

    N

    Nn=1

    u(n)y(n + ) (3.32)

    for different delays = 0, 1, 2...M. The time delay, n, is chosen as the whereRuy() is significantly greater than zero.

    1A filter whose direct term is equal to 1

  • 8/6/2019 Tesis_anc

    35/65

    Chapter 4

    Implementation

    In this chapter the implementation of the ANC system is presented.

    4.1 DSP hardware and equipment

    The computer platform for the ANC system is a Development Starter Kit TMS320C6713DSP from Texas Instruments. Other equipments such as amplifiers, microphonesand speakers are also used in different setups. More information about the equip-ment can be found in Appendix A.

    Figure 4.1. TMS320C6713 DSP DSK.

    23

  • 8/6/2019 Tesis_anc

    36/65

    24 Implementation

    4.1.1 Development Starter Kit

    The Development Starter Kit (DSK) is a card built to test and evaluate theTMS320C6713 DSP, see Figure 4.1. Some key features of the DSK include:

    A Texas Instrument floating point digital signal processor TMS320C6713operating at 225MHz.

    An AIC23 audio codec (AD/DA converter) with stereo input and output.

    16 Mbytes of synchronous DRAM.

    512 Kbytes of non-volatile Flash memory.

    USB communication with a host computer used for software development.

    4.1.2 DSP

    In Figure 4.2, a block diagram of the TMS320C6713 DSP is shown. The processorhas many features, two of importance for the implementation of the ANC system

    are described below.

    EMIF

    McBSPs

    Timers

    HPI

    Interrupt

    Selector

    Power Down

    Logic

    Boot

    Configuration

    Enhanced

    DMA

    Controller

    PLL

    L2 Memory

    L1P Cache

    C6000 DSP Core

    A Reg. File

    Data Path A

    L 1 S1 M1 D1

    B Reg. File

    Data Path B

    D2 M2 S2 L 2

    L1D Cache

    Instruction Fetch

    Instruction Dispatch

    Instruction Decode

    Interrupt

    Control

    In-Circuit

    Emulation

    Test

    Control

    Logic

    Control

    Registers

    Figure 4.2. A block diagram of a TMS320C6713 DSP.

    Multichannel Buffered Serial Port

    The Multichannel Buffered Serial Port (McBSP) is a serial port that buffer samplesautomatically with the aid of the EDMA controller, described below. It consistsof a data path and a control path, which connect to external devices, and four

  • 8/6/2019 Tesis_anc

    37/65

    4.2 DSP software concepts and techniques 25

    other pins that control clocking and frame synchronization. The DSP consists oftwo McBSPs that are used in the ANC system to communicate with the AD/DAconverter. The first one, McBSP0, is used for control and configuration and thesecond one, McBSP1, is used to transfer data between the AD/DA converter andthe CPU. [16]

    Enhanced Direct Memory Access

    Enhanced Direct Memory Access (EDMA) allows devices to transfer data inde-pendently of the CPU. Typically, block data transfers and transfer requests fromperipherals, such as the AIC23 audio codec, are performed by the EDMA thusrelieving the CPU to do performance-intensive operations. In this system, theEDMA controller is configured to take every 16-bit audio sample arriving on theMcBSP1 and store it in a buffer in the memory until it can be processed. Once thedata have been processed, the EDMA controller sends it back to the McBSP1. [15]

    4.2 DSP software concepts and techniques

    4.2.1 Code composer studio

    Code composer studio (CCStudio) is a development tool used for programmingthe TMS320C6713 DSP. It is part of Texas Instruments real-time eXpressDSPTM

    software and development tool strategy. The program is run on a host PC andcommunicates with the target DSP through a USB interface.

    4.2.2 DSP/BIOS

    DSP/BIOS is a real-time kernel that is used on the TMS320C6713 [17]. It isdesigned for real-time applications and supports scheduling, synchronization, real-time instrumentation and host-to-target communication. It is designed to min-imize memory and CPU requirements on the DSP. There are many DSP/BIOSoptions for program development and some are listed below:

    Several thread types are supported such as hardware interrupts, softwareinterrupts, tasks, periodic tasks and idle functions.

    Semaphores, mailboxes and resource locks are provided to support commu-nication and synchronization between threads.

    It is possible to use both dynamically and statically created objects.

    Writing programs using the DSP/BIOS kernel is done by using CCStudio.

    4.2.3 PING-PONG buffering

    PING-PONG buffering is a technique where two buffers (referred to as the PINGbuffer and the PONG buffer) are used for a data transfer instead of only one. Ifonly a single buffer is used, problems with new data overwriting the data being

  • 8/6/2019 Tesis_anc

    38/65

    26 Implementation

    transmitted will occur. Using the PING-PONG technique, data in PING buffer isbeing processed while the next set of data is read into the PONG buffer, and viceversa. This ANC system is using PING-PONG buffers both when transmittingand receiving, for a total of four buffers.

    4.3 DSP program design

    The ANC program is implemented using the language C and an overview of thedesign solution is given in this section. In Figure 4.3 the design solution is illus-trated.

    ANC-off

    STATES

    ANC-on

    Identification

    Main

    function

    DSP/BIOS

    Idle-loopedmaHwi

    processBufferSwi

    THREADS

    return

    Bu

    fferingcompleted

    EDMA Interrupt

    Buffering not completed

    retu

    rn

    DSP/BIOS

    Kernel

    Figure 4.3. An overview of the ANC program design solution.

    4.3.1 Main function

    When first starting the program the main function is run. This is done only onceand outside the DSP/BIOS real-time kernel. It initiates all functions, hardwaresetups and libraries the ANC program needs when entering the DSP/BIOS kernel.Important properties of the main function include:

    Initializing of the audio AIC23 codec on the DSK. The codec sample fre-quency is set to 8000 Hz and returns an integer of 16 bits.

    Initializing of the McBSP interface towards the AIC23.

  • 8/6/2019 Tesis_anc

    39/65

    4.4 Summary and conclusion 27

    The EDMA controller is set to transfer samples from the McBSP to the CPUcore using PING-PONG buffering.

    Memory space used by arrays in the ANC algorithm is cleared.

    After the program has returned from the main function it enters the DSP/BIOSkernel and the DSP/BIOS idle loop.

    4.3.2 DSP/BIOS threads

    Three threads are used

    The DSP/BIOS idle-loop is the thread with lowest priority. This thread isused for communication between the target (DSP) and the host (PC runningCCStudio).

    edmaHwi is a hardware interrupt and is the highest prioritized thread. It istrigged when a complete DMA frame has been received by the EDMA. If boththe receive and transmit transfers have been completed, the processBufferSwi

    is trigged.

    processBufferSwi is a software interrupt that is trigged by the edmaHwi.This task executes the signal processing and identification algorithms de-scribed in Chapter 3 depending on the present state of the the system.

    4.3.3 ANC system states

    The ANC system has three different states, ANC-on, ANC-off and identifica-tion implemented in the processBufferSwi task. Switching between the differentstates is done by editing two variables using a feature in CCStudio called watchwindow.

    Variable ident_onoff is set to 1/0 for on/off. Initially 0.

    Variable anc_onoff is set to 1/0 for on/off. Initially 0.

    In identification mode, the system identifies the secondary path according to Sec-tion 3.5, when completed ident_onoff is automatically set to 0.

    4.4 Summary and conclusion

    Implementing the ANC system using the TMS320C6713 DSK has both advantagesand disadvantages. It is fairly easy to use and there are lots of open code writtenalready. On the other hand, without adding extra hardware the ANC system islimited to use only two microphones and two speakers. Adding hardware is costly

    and extra code has to be written.

    Comparing a DSP platform to a PC platform, the DSP is by far more faster and

  • 8/6/2019 Tesis_anc

    40/65

    28 Implementation

    efficient. As an example, the loop through a system including sound card, speaker,air, microphone and back again takes 80 ms on a Matlab/Simulink implementa-tion using Windows XP. The same loop with the laptop replaced by a DSP takesonly 3.6 ms. Hence, there is a big difference and for real time implementation aDSP implementation is preferable.

  • 8/6/2019 Tesis_anc

    41/65

    Chapter 5

    Results

    In this chapter, detailed results for the feedforward and feedback system are pre-sented in Section 5.1 and Section 5.2, respectively. Due to the way the ANCsystems are implemented, the DSP is halted for an instant when turning on thesystems. This is causing a noise impulse, visible in some time plots. To calculatethe reduction of sound pressure level (SPL), following formula is being used

    Lp = 20log10poffpon

    dB (5.1)

    where Lp is the SPL reduction, poff is the 8000 samples root mean square errorbeing measured when the ANC system is off and pon when the ANC system is on.

    5.1 Feedforward FxLMS

    An overview of a feedforward system was presented in Section 2.2 and theory isavailable in Section 3.3. During the evaluation of the feedforward system twomicrophones are used, the reference microphone is put close to the engine and theerror microphone is put close to the ceiling of the cabin.

    5.1.1 Primary path

    To achieve a good result, a strong dependence between noise in the referencemicrophone and noise in the error microphone is necessary. The dependence ishere illustrated by coherence estimate plots. Coherence is a function of frequencywith values between 0 and 1 that indicate how well two signals corresponds to eachother at each frequency. A value of 1 inicates a strong dependence. In Figure 5.1,

    Figure 5.2 and Figure 5.3, coherence estimate plots are used to illustrate thedependence between error and reference singal for the engine speeds of 840, 1500and 2150 revolutions per minute (rpm).

    29

  • 8/6/2019 Tesis_anc

    42/65

    30 Results

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.20

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

    1

    Frequency (kHz)

    Magnitude(dB)

    Coherence Estimate via Welch

    Figure 5.1. Coherence estimate between reference michrophone and error microphone

    at an engine speed of 840 rpm.

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.20

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

    1

    Frequency (kHz)

    Magnitude(dB)

    Coherence Estimate via Welch

    Figure 5.2. Coherence estimate between reference michrophone and error microphone

    at an engine speed of 1500 rpm.

  • 8/6/2019 Tesis_anc

    43/65

    5.1 Feedforward FxLMS 31

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.20

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

    1

    Frequency (kHz)

    Magnitude(dB)

    Coherence Estimate via Welch

    Figure 5.3. Coherence estimate between reference michrophone and error microphone

    at an engine speed of 2150 rpm.

    5.1.2 Secondary path

    In Section 3.5 the theory of secondary path identification was presented. Low-passfiltered Gaussian random noise is used to estimate the secondary path and by usingcross correlation the system delay is estimated. In Figure 5.4 the cross correlationbetween the output noise and the measured noise is shown at different time delays.The result of the secondary path identification is presented in Figure 5.5 where

    the parameters of the FIR filter S(z) are plotted. Since the delay of the systemwas 57 samples during the identification, the same amount of samples are set tozero in s, see Figure 5.5.

    5.1.3 Constant engine speeds

    To get a feeling of how well the ANC system is working for different engine speedsthree cases are studied: idling engine speed, maximum engine speed and an enginespeed of 1500 rpm. The engine speed 1500 rpm is a commonly used engine speedfor maximum performance and economical driving and will in the rest of thisreport be referred to as working enginge speed.

    Idling engine speed (840 rpm)In Figure 5.6, a distinct difference is visible for the operating ANC system andthe convergence of the system is seen during the time interval 2 2.3 seconds.

  • 8/6/2019 Tesis_anc

    44/65

    32 Results

    0 100 200 300 400 500

    0.4

    0.3

    0.2

    0.1

    0

    0.1

    0.2

    0.3

    0.4

    Lag

    SampleCrossCorrelation

    Sample Cross Correlation Function (XCF)

    Figure 5.4. Cross correlation between time delays.

    0 100 200 300 400 5005

    4

    3

    2

    1

    0

    1

    2

    3

    4

    5x 10

    3

    i

    si

    Figure 5.5. Filter parameters si, i = 1,...,560, of S(z).

  • 8/6/2019 Tesis_anc

    45/65

    5.1 Feedforward FxLMS 33

    The SPL is reduced with 15.5 dB. A plot of the power spectral density for theANC system off and on is shown in Figure 5.7. Significant differences are visibleespecially for the peaks at 45 Hz and 90 Hz, but other frequencies are suppressedas well, while a few are enhanced.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.01

    0.008

    0.006

    0.004

    0.002

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    Time (s)

    Figure 5.6. Time plot of error signal. The feedforward ANC system is turned on at

    time 2 s. Engine speed 840 rpm.

    Working engine speed (1500 rpm)

    A significant difference is visible for the operating ANC system, see Figure 5.8.

    Looking at the power spectral density plot, Figure 5.9, the peak of 84 Hz is sup-pressed well. The SPL is reduced with 17.5 dB.

    Maximum engine speed (2150 rpm)

    For an engine running full speed, the ANC system is not making that big differenceas for the previous cases, see Figure 5.10. The low frequenices do not have asmuch energy as in the previous cases, but the peak of 55 Hz is suppressed well,see Figure 5.11. The SPL is reduced with 2.5 dB.

    5.1.4 Robustness

    Robustness is crucial for a usable system. To evaluate the robustness two simpletests are done: ANC turned on with open door and ANC running while closingthe door.

  • 8/6/2019 Tesis_anc

    46/65

    34 Results

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2120

    110

    100

    90

    80

    70

    60

    50

    40

    30

    Frequency (kHz)

    Power/frequency(dB/Hz)

    (Rel.arbitraryreference)

    Periodogram Power Spectral Density Estimate

    ANC off

    ANC on

    Figure 5.7. Power spectral density of error signal for the feedforward ANC system.

    Engine speed 840 rpm.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.01

    0.008

    0.006

    0.004

    0.002

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    Time (s)

    Figure 5.8. Time plot of error signal. The feedforward ANC system is turned on attime 2 s. Engine speed 1500 rpm.

  • 8/6/2019 Tesis_anc

    47/65

    5.1 Feedforward FxLMS 35

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2120

    110

    100

    90

    80

    70

    60

    50

    40

    30

    Frequency (kHz)

    Power/frequency(dB/Hz)

    (Rel.arbitraryreference)

    Periodogram Power Spectral Density Estimate

    ANC off

    ANC on

    Figure 5.9. Power spectral density of error signal for the feedforward ANC system.

    Engine speed 1500 rpm.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.01

    0.008

    0.006

    0.004

    0.002

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    Time (s)

    Figure 5.10. Time plot of error signal. The feedforward ANC system is turned on attime 2 s. Engine speed 2150 rpm.

  • 8/6/2019 Tesis_anc

    48/65

    36 Results

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2120

    110

    100

    90

    80

    70

    60

    50

    40

    30

    Frequency (kHz)

    Power/frequency(dB/Hz)

    (Rel.arbitraryreference)

    Periodogram Power Spectral Density Estimate

    ANC off

    ANC on

    Figure 5.11. Power spectral density of error signal for the feedforward ANC system.

    Engine speed 2150 rpm.

    ANC turned on with open door

    In the first case the secondary path is changed since identification of S(z) wasdone with closed door. When the ANC system is turned on it is easy to see howthe system converges despite a model error in S(z), see Figure 5.12.

    ANC running while closing the door

    In the second case the door is open and the ANC system is running with a modelerror in S(z). By closing the door, S(z) is changed back to normal. In Figure 5.13it is visible that despite the change of the secondary path and a violent noise fromthe closing of the door, the system is working fine.

  • 8/6/2019 Tesis_anc

    49/65

    5.1 Feedforward FxLMS 37

    0 1 2 3 4 5 6 7 8 9 100.03

    0.02

    0.01

    0

    0.01

    0.02

    0.03

    Time (s)

    Figure 5.12. Time plot of error signal with open door. The feedforward ANC system

    is turned on at time 5.8 s. Engine speed 840 rpm.

    0 1 2 3 4 5 6 7 8 9 10

    0.1

    0.05

    0

    0.05

    0.1

    0.15

    Time (s)

    Figure 5.13. Time plot of error signal. The feedforward ANC system is running and

    the door is closed at time 7 s. Engine speed 840 rpm.

  • 8/6/2019 Tesis_anc

    50/65

    38 Results

    5.2 Feedback FxLMS

    An overview of a feedback system was presented in Section 2.3 and the theory isbriefly described in Section 3.4. During the evaluation of the feedback system onlyone error microphone is used and it is put close to the ceiling of the cabin.

    5.2.1 Secondary path

    The identification of the secondary path for the feedback system is identical as forthe feedforward system, see Section 5.1.2.

    5.2.2 Constant engine speeds

    As for the feedforward system, three different engine speeds are studied for thefeedback system: idling engine speed, maximum engine speed and working enginespeed.

    Idling engine speed (840 rpm)

    In Figure 5.14, a distinct difference is visible for the operating ANC system andthe convergence of the system is seen during the time interval 2 2.3 seconds.The SPL is reduced with 10.7 dB. A plot of the power spectral density is shownin Figure 5.15.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.01

    0.008

    0.006

    0.004

    0.002

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    Time (s)

    Figure 5.14. Time plot of error signal. The feedback ANC system is turned on at time2 s. Engine speed 840 rpm.

  • 8/6/2019 Tesis_anc

    51/65

    5.2 Feedback FxLMS 39

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2120

    110

    100

    90

    80

    70

    60

    50

    40

    30

    Frequency (kHz)

    Power/frequency(dB/Hz)

    (Rel.arbitraryreference)

    Periodogram Power Spectral Density Estimate

    ANC off

    ANC on

    Figure 5.15. Power spectral density of error signal for the feedback ANC system.

    Engine speed 840 rpm.

  • 8/6/2019 Tesis_anc

    52/65

    40 Results

    Working engine speed (1500 rpm)

    The time plot for the operating ANC system is shown in Figure 5.16 and the powerspectral density plot is presented in Figure 5.17. The SPL is reduced with 14.9 dB.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.01

    0.008

    0.006

    0.004

    0.002

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    Time (s)

    Figure 5.16. Time plot of error signal. The feedback ANC system is turned on at time

    2 s. Engine speed 1500 rpm.

    Maximum engine speed (2150 rpm)

    For an engine running full speed, the ANC system is not making a big difference,see Figure 5.18. The peak of 118 Hz is suppressed, see Figure 5.19, and the SPL

    is reduced with 4.8 dB.

    5.2.3 Robustness

    As for the feedforward system, two simple tests are done to evaluate the robustnessfor the feedback system: ANC turned on with open door and ANC running whileclosing the door.

    ANC turned on with open door

    The feedback system does not become unstable with a significant model error inS(z) caused by an open door, see Figure 5.20.

    ANC running while closing the doorAn impulse provocation of the system, closing the door, does not make the systemunstable, see Figure 5.21.

  • 8/6/2019 Tesis_anc

    53/65

    5.2 Feedback FxLMS 41

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2120

    110

    100

    90

    80

    70

    60

    50

    40

    30

    Frequency (kHz)

    Power/frequency(dB/Hz)

    (Rel.arbitraryreference)

    Periodogram Power Spectral Density Estimate

    ANC off

    ANC on

    Figure 5.17. Power spectral density of error signal for the feedback ANC system.

    Engine speed 1500 rpm.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.01

    0.008

    0.006

    0.004

    0.002

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    Time (s)

    Figure 5.18. Time plot of error signal. The feedback ANC system is turned on at time

    2 s. Engine speed 2150 rpm.

  • 8/6/2019 Tesis_anc

    54/65

    42 Results

    0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2110

    100

    90

    80

    70

    60

    50

    40

    30

    Frequency (kHz)

    Power/frequency(dB/Hz)

    (Rel.arbitraryreference)

    Periodogram Power Spectral Density Estimate

    ANC off

    ANC on

    Figure 5.19. Power spectral density of error signal for the feedback ANC system.

    Engine speed 2150 rpm.

    0 1 2 3 4 5 6 7 8 9 100.025

    0.02

    0.015

    0.01

    0.005

    0

    0.005

    0.01

    0.015

    0.02

    0.025

    Time (s)

    Figure 5.20. Time plot of error signal with open door. Feedback ANC system turned

    on at time 5.8 s. Engine speed 840 rpm.

  • 8/6/2019 Tesis_anc

    55/65

    5.3 Global noise reduction in the cabin 43

    0 1 2 3 4 5 6 7 8 9 100.25

    0.2

    0.15

    0.1

    0.05

    0

    0.05

    0.1

    0.15

    0.2

    0.25

    Time (s)

    Figure 5.21. Time plot of error signal. The feedback ANC system is running and thedoor is closed at time 5 s. Engine speed 840 rpm.

    5.3 Global noise reduction in the cabin

    All results in this section are measured from the error microphone. But sincethe noise is minimized at this position, the noise level does not necessary have tobe decreased elsewhere in the cabin. To visualize what the driver is hearing, amonitor microphone is put close to the ear of the driver. In Figure 5.22 and 5.23,the error microphone and monitor microphone are presented for an idling engine.In this case, the feedback ANC system is used but the results are similar to thefeedforward ANC system. The plots show that suppression is achieved globally in

    the cabin and not just locally at the position of the error microphone.

  • 8/6/2019 Tesis_anc

    56/65

    44 Results

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.2

    0.15

    0.1

    0.05

    0

    0.05

    0.1

    0.15

    0.2

    Time (s)

    Figure 5.22. Time plot of error signal. The feedback ANC system is turned on at time

    2 s. Engine speed 840 rpm.

    0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 50.2

    0.15

    0.1

    0.05

    0

    0.05

    0.1

    0.15

    0.2

    Time (s)

    Figure 5.23. Time plot of monitor signal. The feedback ANC system is turned on at

    time 2 s. Engine speed 840 rpm.

  • 8/6/2019 Tesis_anc

    57/65

    5.4 Summary of results 45

    5.4 Summary of results

    A summary of the results is presented in Table 5.1.

    SPL

    rpm Feedforward Feedback840 15.5 dB 10.7 dB

    1500 17.5 dB 14.9 dB2150 2.5 dB 4.8 dB

    Table 5.1. Summary of results

  • 8/6/2019 Tesis_anc

    58/65

    46 Results

  • 8/6/2019 Tesis_anc

    59/65

    Chapter 6

    Concluding Remarks

    In this chapter conclusions and suggestions of future work are discussed.

    6.1 ConclusionsIn this thesis we have evaluated active noise control of a forest machine cabin.Two different sensor configurations have been studied, feedforward and feedback.The feedforward system uses a reference sensor to get a reference signal, while thefeedback system generates its own reference signal from the error signal and theadaptive filter output. In both cases the adaptive filter is updated by the errorsensor. Microphones are being used as sensors in the evaluated systems. TheFxLMS algorithm is being used for both feedforward and feedback to compensatefor different effects that comes up running the system in real time.

    The results show that the systems works best for idling engine speed and for anormal working engine speed, mainly because of the low frequency noise generated

    at these engine speeds. Both the feedforward and the feedback system show fastenough convergence and global suppression of low frequency noise in the cabin.

    6.2 Future work

    As a result of the time limit of the project, improvements of the system are pos-sible and some suggestions of future work follows below.

    For a feedforward system other reference sensors than microphones, such asan engine speed sensor, should be evaluated.

    Evaluation of reference sensors at other positions should be done to cancelout other disturbing noises, e.g., one possible position is close to the hydraulicpump.

    47

  • 8/6/2019 Tesis_anc

    60/65

    48 Concluding Remarks

    A combination of a feedforward ANC and a feedback ANC system, oftenreferred to as a hybrid ANC should be evaluated since good results havebeen achieved in other projects [14], [8].

    Place speakers and microphones at the drivers seat, close to the ears, to

    achieve better results at higher frequencies as a complement to the existingsubwoofer.

    To increase system performance, other step size algorithms should be eval-uated, e.g., different variable step size algorithms have been shown to beefficient [1], [7].

    Low pass filter reference microphone signals to get rid of unwanted noise.

    To keep S(z) updated continuously, evaluation of online secondary path mod-eling should be done [19].

    Implement a music compensator to improve system performance while usingan audio system simultaneously in the cabin [13].

  • 8/6/2019 Tesis_anc

    61/65

    Bibliography

    [1] Akhtar M.T., Abe M., Kawamata M. (2004). Modified-filtered-x LMS algo-rithm based active noise control system with improved on line secondary-pathmodeling. Proceedings of the IEEE International Midwest Symposium on Cir-cuits and Systems, 1:I-13I-16.

    [2] Akhtar M.T., Abe M., Kawamata M. (2005). A method for on line secondarypath modeling in active noise control systems. Proceedings of the IEEE In-

    ternational Midwest Symposium on Circuits and Systems, 1:264267.

    [3] Arbetsmiljverket. (2005) Buller. Arbetsmiljverkets frfattningssamling,AFS 2005:16.

    [4] Elliott S.J., Nelson P.A. (1993). Active Noise Control. IEEE Signal ProcessingMagazine, 10(4):1235.

    [5] Glad T., Ljung L. (2003). Reglerteori. Studentlitteratur.

    [6] Gustafsson F. (2000). Adaptive Filtering and Change Detection. John Wiley& Sons, Ltd.

    [7] Haweel T.I. (2004). A simple variable step size LMS adaptive algorithm. In-

    ternational Journal on Circuit Theory and Applications, 32:523536.[8] Kong X., Liu P., Kuo S.M. (1998). Multiple Channel Hybrid Active Noise Con-

    trol Systems. IEEE Transactions on Control Systems Technology, 6(6):719729.

    [9] Kuo S.M., Morgan D.R. (1999). Active Noise Control: A Tutorial Review.Proceedings of the IEEE, 87(6):943973.

    [10] Ljung L. (1999). System Identification, Theory for the User. 2nd edition Pren-tice Hall Information and System Sciences Series.

    [11] Morgan D.R. (1980). An analysis of multiple correlation cancellation loopswith a filter in the auxiliary path. IEEE Transactions on Acoustics, Speech,and Signal Processing, 5:454467.

    [12] Pedrotti F.L., Pedrotti L.S. (1993). Introduction to Optics, 2nd edition. Pren-tice Hall International Editions.

    49

  • 8/6/2019 Tesis_anc

    62/65

    50 Bibliography

    [13] Sano H., Inoue T., Takahashi A., Terai K., Nakamura Y. (2001). ActiveControl System for Low-Frequency Road Noise Combined With an AudioSystem. IEEE Transactions on Speech and Audio Processing, 9(7):755763.

    [14] Streeter A.D., Ray L.R., Collier R.D. (2004). Hybrid Feedforward-Feedback

    Active Noise Control. Proceedings of the 2004 American Control Conference,3:28762881.

    [15] Texas Instruments. (2003).TMS320C6000 DSP Enhanced Direct Memory Ac-cess (EDMA) Controller Reference. SPRU234.

    [16] Texas Instruments. (2004).TMS320C6000 DSP Multichannel Buffered SerialPort (McBSP) Reference Guide. SPRU580C.

    [17] Texas Instruments. (2004). TMS320 DSP/BIOS Users Guide. SPRU423D.

    [18] Texas Instruments. (2006).TMS320C67x DSP Library Programmers Refer-ence Guide. SPRU657B.

    [19] Zhang M., Lan H., Ser W. (2005). On comparison of online secondary pathmodeling methods with auxiliary noise. IEEE Transactions on Speech andAudio Processing, 13(4):618628.

  • 8/6/2019 Tesis_anc

    63/65

    Appendix A

    Equipment

    Development card

    One Texas instrument TMS320C6713 Development Starter Kit (DSK) is used as

    a signal processing hardware platform, see Figure 4.1. Key features of this cardinclude:

    A Texas Instrument floating point digital signal processor TMS320C6713operating at 225MHz.

    An AIC23 audio codec (AD/DA converter) with stereo input and output.

    16 Mbytes of synchronous DRAM.

    512 Kbytes of non-volatile Flash memory.

    USB communication with a host computer used for software development.

    Amplifier

    One four-channel Reflexion X500:

    Max output power 4 x 225 W.

    Signal to noise ratio >95 dB.

    Frequency response 10 Hz - 45 kHz.

    Subwoofer

    One 8 inch MDS subwoofer enclosure:

    Speaker with frequency response 24Hz - 3 kHz.

    Built in amplifier

    51

  • 8/6/2019 Tesis_anc

    64/65

    52 Equipment

    Preamplifier

    To power the microphones an Alto Mictube stereo preamplifier is used. Thispreamplifier is connected between microphones and the DSK and has specifica-tions:

    48 V phantom power supply to power high quality condenser microphones.

    Frequency response 20Hz - 22 kHz

    Microphones

    Two TW omni condenser lavaliere microphones are used.

    Frequency response 20Hz - 20 kHz.

    Omni directional, sensitive to sounds from all directionals.

    One Superlux CM-H8K microphone for use in very high sound pressure environ-ments.

    Frequency response 40Hz - 20 kHz.

    Sound pressure level 134 dB.

    Omni directional, sensitive to sounds in all directions.

    Mixer

    One Helix board 12 FireWire mixer is used.

    12-input small-format analog mixer with extremely low noise circuitry.

    96 kHz FireWire interface for streaming 10 independent channels of audioto computer with near-zero latency.

    Four mono Mic/Line channels

    Two stereo channels, two stereo AUX returns, two AUX sends.

    Inserts and phantom power on Mic channels.

    Mini stereo and RCA.

  • 8/6/2019 Tesis_anc

    65/65