Active Noise Control Architectures and Application Potentials Shawn Steenhagen - Applied Signal Processing, Inc. 3 Marsh Court Madison, WI 53718 Tele: 608-441-9921 Fax: 608-441-9924 Web: www.appliedsignalprocessing.com General Problem Definition: • A measurable signal, y, contains a desired signal component, d, and a noise component, n, which is to be removed from y. The measurable signal can be a humanly observable event such as a vibration or sound, or an electrically observable event such as radio frequency interference. y=d+n ANC Architectures - www.appliedsignalprocessing.com 2 The Active Noise Control Solution • Feed Forward Adaptive Active Noise Control uses an LMS Adaptive filter to create and introduce a control signal, ŷ which when subtracted from y, results in an error signal whose power is minimized in a mean square sense. y=d+n xref ŷ LMS Filter - ANC Architectures - www.appliedsignalprocessing.com e 3 LMS Adaptive Filter • Minimizes the error between the observation, y, and the estimate, ŷ . • Matches/Removes only the components y which are correlated to the reference signal. • Update EQ: A(k+1) = A(k) + mu*X(k)*e(k) y=d+n xref ŷ A - ANC Architectures - www.appliedsignalprocessing.com e 4 An Active Noise Control System needs: • A Reference Signal, x (either correlated to d or n). • A measurable observation y or a measurable Error Signal e= (y - ŷ ) • A method for adding/mixing the control signal ŷ into the system. (either in the digital or physical domain) y=d+n xref ŷ LMS Filter - ANC Architectures - www.appliedsignalprocessing.com e 5 ANC Feed Forward Architectures • System or Plant Identification. (For Random and/or Tonal noises) • Signal Identification (For Tonal noises) • “Filtered-X” variants of Plant Id or Signal Id Systems (for acoustic and vibration control.) ANC Architectures - www.appliedsignalprocessing.com 6 Filtered-X LMS • Typical Adaptive Filter– measures the error signal directly. • Typical Acoustic Active Noise Control Configuration – measures a filtered version (C) of the error signal. • Requires use of the “Filtered-X” LMS algorithm. • Update EQ: A(k+1) = A(k) + mu*{CX(k)}*eobs(k) y = Px xref P E S A ŷ C e(k)obs = SEe(k) C=SE ANC Architectures - www.appliedsignalprocessing.com 7 ANC - Plant Identification Architecture x(t) y(t) e(t) P E ^ -y’(t) F Acoustical S x(n) e(t) Signal Processing 0 e(n) ^ -y(n) A C when A B PE SE B PF ^C ANC Architectures - www.appliedsignalprocessing.com ^C N 8 Plant Id – Causality Requirements A is causal if Pdelay > Sdelay x(t) y(t) e(t) P E ^ -y’(t) B is causal if PFdelay > 0 F Acoustical S x(n) e(t) Signal Processing 0 e(n) ^ -y(n) A C when A B PE SE B PF ^C ANC Architectures - www.appliedsignalprocessing.com ^C N 9 ANC - Plant Identification Architecture • Advantages: – Can cancel random noise. – Once converged, no need to re-adapt to track changes in reference signal • Disadvantages: – Needs longer physical plant lengths to meet causality requirements. – Requires Persistence of Excitation for proper convergence. – Computationally intensive for higher filter order (# of taps) when more frequency resolution is needed. – Requires higher filter order for better low frequency tonal performance. ANC Architectures - www.appliedsignalprocessing.com 10 ANC – Signal Identification Architecture y y 0 y1 ...yn Phase, speed, or rotational info E - yˆ yˆ 0 yˆ1 ...yˆn sin( 0 ) N = # ers d r o D/A ref[0] A/D As0 Tone Generator cos( 0 ) C - ref[1] Ac0 c_mu galois Noise C C a_mu0 a_mu0 ANC Architectures - www.appliedsignalprocessing.com 11 ANC – Signal Identification Architecture • Reference Signal Generator (from a phase or frequency observation.) • Two Tap Quadrature Adaptive Filter (One Pair for each frequency to be controlled) matches phase and amplitude of frequency component, yn , within y. • Output and Update Equations: yˆn (k ) Aˆ cosn (k ) cosn (k ) Aˆ sinn (k ) sin n (k ) Aˆ (k 1) Aˆ (k ) cos (k )e(k ) cosn cosn n Aˆ sinn (k 1) Aˆ sinn (k ) sin n (k )e(k ) ANC Architectures - www.appliedsignalprocessing.com 12 ANC – Signal Identification Architecture • Advantages: – Fast Convergence. – Computational Simplicity. – Excellent frequency resolution. • Disadvantages: – Tonal or Periodic Noise Applications only. – In Filtered-X LMS applications, requires 2N separate filtering operations for each frequency component. ANC Architectures - www.appliedsignalprocessing.com 13 Why is Acoustic Active Noise Control Difficult? • Typical Active Noise Control Configuration requires use of the “Filtered-X” LMS algorithm. • The C path must be known and typically it changes, so an adaptive process for it is also required. • The most reliable way to model the C path is using an auxiliary noise source. This presents customer acceptance challenges. • Errors between C and SE effect convergence rates of the adaptive filter and stability requirements. • Complexity expands in MIMO cases. Number of C models = NUM_ACT * NUM_ERR ANC Architectures - www.appliedsignalprocessing.com 14 Crafting the ANC Solution: • Evaluate Viability of Active as an Approach. – Initial litmus tests – (physics & costs) – Noise Analysis – characterize the noise. – Market Analysis – cost & end user constraints. • Choose Configuration: – System ID, Signal ID or hybrid. – Filtered-X vs. Direct LMS update. • Simulation and Analysis • Real Time Implementation. ANC Architectures - www.appliedsignalprocessing.com 15 Viability Considerations of an ANC Solution • • • • Availability of a reference signal, error signal, and mixing method. Power Requirements. Cost relative to Target Application. Noise Spectrum Characteristics: – – – – – – Tonal, random, or mix. Dynamic or Stationary tonal characteristics. SPL or Vibration Levels. Frequency Range. Geometric Attributes – plane wave, point source, free space. Portion of which can be removed with active with respect to total noise spectrum. – Coherence between reference signal and observation or error signal. • • • • Dimensionality of the system. Size/geometry/packaging space. Operating environment (hot, cold, caustic) Complexity vs. Passive Methods. (Cost/Benefit) ANC Architectures - www.appliedsignalprocessing.com 16 Potential ANC Applications • • • • • • • • • • • • • • Communication Systems (cell phone, two way radios, intercom) within any noisy environment.** Audio - Post Production clean up. Aircraft – active engine mounts** HVAC, Industrial Blowers/Fans** Automotive – air induction* Aircraft – cabin interior* Vibration Isolation – sensitive manufacturing processes.* Computer Fan Noise.* Automotive - interior*, road noise. Automotive – exhaust*, Lawn mowers*, vacuum cleaners, dishwashers, refrigerators Factory Noise in free space (hard to beat ear plugs) Loud impulsive noise (jack hammers, punch presses) Snoring, Neighbor or Teenager’s Stereo, Politicians. ANC Architectures - www.appliedsignalprocessing.com Easy/Practical Hard/Impractical Impossible 17 Application Example • Clean up Outbound Cell Tx in Vehicle During Hands Free operation. – Engine noise is tonal. (two tap quadrature can be used) – A reference signal can be generated from readily available CAN signals. – The mixing environment is in the digital domain Direct LMS can be applied. ANC Architectures - www.appliedsignalprocessing.com 18 Application Example Hands free echo cancellor and engine noise cancellor Gadc sin( 0 ) Tone Generator A/D ref[0] D/A A0 cos( 0 ) Gdac ref[1] A1 AEC RPM Calculation a_mu0 a_mu1 CAN\ Converter Cell Receive Cell Transmit ANC Architectures - www.appliedsignalprocessing.com 19 Conclusion/Looking Forward • Increasing MIPS capacity of DSPs can make computationally impractical applications of the past more viable. • Recent research in highly directional acoustic sources via loud speaker arrays may help expand potential application areas. ANC Architectures - www.appliedsignalprocessing.com 20 References/Further Reading • “Active Noise Control Systems – Algorithms and DSP Implementations”; Sen M. Kuo, Dennis R. Morgan. • “Lectures on Adaptive Parameter Estimation”, C. Richard Johnson Jr. • “Active Control of Sound”; P.A. Nelson & S.J. Elliot. ANC Architectures - www.appliedsignalprocessing.com 21