Symposium on Nonlinear Compensation of Loudspeakers

Thursday December 2, 2004, 13:00-17:00, room 053, building 321,
Informatics and Mathematical Modelling, Technical University of Denmark

The development in nonlinear signal processing algorithms and hardware platforms provides an interesting challenge of compensation of nonlinearities in loudspeakers. Successful electronic compensation will enable high sound quality in small transducers even at high sound pressure levels. The applications are multitude ranging from hearings aids, mobile telephones and head sets to home theater and PA equipment. The symposium focuses on application of digital signal processing for optimization of transducer performance. Further a panel will discuss future expectations and the challenges of applying the techniques in real-world devices.

Program

13:00-14:00 Active Compensation of Transducer Nonlinearities
Dr. Wolfgang Klippel, Klippel GmbH, Dresden, Germany

Download presentation handout (pdf)
Nonlinearities inherent in electromechanical and electroacoustical transducers produce signal distortion and limit the maximal amplitude of the output signal. Assessing the large signal performance has been a subject of acoustical research for many years providing nonlinear models and new methods for the measurement of the large signal parameters. The identified model allows prediction and simulation of the nonlinear behavior and direct comparison with measured symptoms. The good agreement between model and reality is the basis for developing novel digital controllers dedicated to transducers that compensate actively for nonlinear distortion by inverse preprocessing of the electrical input signal. This talk gives a summary on the activities in the last 15 years and new challenges of the future.

14:00-14:20 Coffee break

14:20-15:20 Deliberate Nonlinear Design and Control of Loudspeaker
Dr. Andrew Bright, Nokia A/S

Download presentation handout (pdf)
Some electroacoustic systems can benefit by a deliberate introduction of nonlinearity. In the design of an electrodynamic motor, a certain trade-off exists between sensitivity and linearity. Some evidence suggests there is a net benefit to introducing some nonlinearity, as the increase sensitivity can provide more value than the cost of adding a control system for compensating the nonlinearity introduced. Within some limits, a deliberately nonlinear system, supported by a nonlinear control system, may be more optimal than a passively linear system. This approach will be discussed in terms of three different loudspeaker transduction principles, with a focus on electrodynamic loudspeakers. A brief explanation of nonlinear controller design suitable for digital signal processors will also be discussed.

15:20-15:30 Break

15:30-16:15 Learning Based Signal Processing for Nonlinear Compensation
Associate Professors Jan Larsen and Ole Winther, IMM, DTU

Download presentation handout (pdf)
The use of learning based signal processing techniques makes it possible to model nonlinear loudspeaker nonlinearities and perform active nonlinear compensation. The talk will focus of using nonlinear state space models for the loudspeaker and the compensation device. Further we discuss the application of modern machine learning techniques for learning the models beyond classical least squares.

16:15-16:20 Break

16:20-17:00 Panel discussion: Challenges in Commercial Compensation Systems
Dr. Wolfgang Klippel, Klippel GmbH
Dr. Andrew Bright, Nokia A/S
Assoc. Prof. Finn Agerkvist, Ørsted.DTU
Dr. Søren Riis, Oticon A/S
M.Sc. Jan Abildgaard Pedersen, AM:3D A/S
M.Sc. Thomas Kyhn, Bang & Olufsen A/S

Topics to be addressed by the panel:

How far are we from commercial implementations?
Which solutions will be feasible in the near future - online compensation or fixed pre-compensation?
What are the challenges for successful commercial systems?

Program

13:00-14:00	Active Compensation of Transducer Nonlinearities Dr. Wolfgang Klippel, Klippel GmbH, Dresden, Germany Download presentation handout (pdf) Nonlinearities inherent in electromechanical and electroacoustical transducers produce signal distortion and limit the maximal amplitude of the output signal. Assessing the large signal performance has been a subject of acoustical research for many years providing nonlinear models and new methods for the measurement of the large signal parameters. The identified model allows prediction and simulation of the nonlinear behavior and direct comparison with measured symptoms. The good agreement between model and reality is the basis for developing novel digital controllers dedicated to transducers that compensate actively for nonlinear distortion by inverse preprocessing of the electrical input signal. This talk gives a summary on the activities in the last 15 years and new challenges of the future.

14:00-14:20	Coffee break

14:20-15:20	Deliberate Nonlinear Design and Control of Loudspeaker Dr. Andrew Bright, Nokia A/S Download presentation handout (pdf) Some electroacoustic systems can benefit by a deliberate introduction of nonlinearity. In the design of an electrodynamic motor, a certain trade-off exists between sensitivity and linearity. Some evidence suggests there is a net benefit to introducing some nonlinearity, as the increase sensitivity can provide more value than the cost of adding a control system for compensating the nonlinearity introduced. Within some limits, a deliberately nonlinear system, supported by a nonlinear control system, may be more optimal than a passively linear system. This approach will be discussed in terms of three different loudspeaker transduction principles, with a focus on electrodynamic loudspeakers. A brief explanation of nonlinear controller design suitable for digital signal processors will also be discussed.

15:20-15:30	Break

15:30-16:15	Learning Based Signal Processing for Nonlinear Compensation Associate Professors Jan Larsen and Ole Winther, IMM, DTU Download presentation handout (pdf) The use of learning based signal processing techniques makes it possible to model nonlinear loudspeaker nonlinearities and perform active nonlinear compensation. The talk will focus of using nonlinear state space models for the loudspeaker and the compensation device. Further we discuss the application of modern machine learning techniques for learning the models beyond classical least squares.

16:15-16:20	Break

16:20-17:00	Panel discussion: Challenges in Commercial Compensation Systems Dr. Wolfgang Klippel, Klippel GmbH Dr. Andrew Bright, Nokia A/S Assoc. Prof. Finn Agerkvist, Ørsted.DTU Dr. Søren Riis, Oticon A/S M.Sc. Jan Abildgaard Pedersen, AM:3D A/S M.Sc. Thomas Kyhn, Bang & Olufsen A/S Topics to be addressed by the panel: How far are we from commercial implementations? Which solutions will be feasible in the near future - online compensation or fixed pre-compensation? What are the challenges for successful commercial systems?

Organized by
Informatics and Mathematical Modelling Ørsted.DTU
IEEE Denmark Section
AES Denmark Section

Organized by
Informatics and Mathematical Modelling Ørsted.DTU	IEEE Denmark Section	AES Denmark Section