Demant A/s · OMXCOP:DEMANT
Demant A/s — William Demant Holding — Investor Day (December 2002)
24 slides from Demant A/s's investor day presentation, Dec 9, 2002
Slide 1: DSP development
DSP development 9 December 2002 1
Slide 2: Thomas Evers Christensen
Thomas Evers Christensen • MScEE with 17 years’ experience • Close to 13 years with Oticon • Pioneer in Digital Signal Processing in Oticon • Co-managing Team Technology • Experience from two other companies (very different industries, though) 2
Slide 3: Outline
Outline • Short overview of a hearing aid • What’s in it? • I will concentrate on the amplifier • Digital Signal Processing (DSP) basics • Add and multiply • Examples: Adapto • VoiceFinder and Dynamic Feedback Reduction 3
Slide 4: What's in a hearing aid?
What’s in a hearing aid? ITE BTE 4
Slide 5: Elements of hearing aid development
Elements of hearing aid development Interconnect No feedback IC technology Audiological No occlusion kernel Shell technology Speech Directionality understanding 5
Slide 6: What makes a good hearing aid?
What makes a good hearing aid? • True user benefits • Improved communication abilities • Comfort • The good hearing aid is a compromise • Performance (delay!) • Size • Power consumption 6
Slide 7: User benefits
User benefits Index •True user 100 benefits are 90 getting 80 expensive 70 •In MIPS 60 Complexity •In complexity 50 40 Development 30 time 20 10 0 1995 1999 2001 2005 1997 7
Slide 8: Signal processing basics
Signal processing basics • DSP algorithms are the mathematical formulation of the functionality • Digital Signal Processing is about adding and multipling … (linear system theory) • The basic digital filter equation: y(n) = Σ[x(n-i)*a(i)] + Σ[y(n-i-1)*b(i)] • … and any hardware will do 8
Slide 9: DSP in hearing aids
DSP in hearing aids • Hearing aids have very special requirements • Low power consumption (< 1 mW) • Non-linear signal processing • Real time audio • High audio quality • No standards apply • Algorithms tightly connected to audiology • Basic algorithm development needed 9
Slide 10: Example: Adapto
Example: Adapto • Signal processing overview • Concentrate on VoiceFinder and Dynamic Feedback Reduction 10
Slide 11: VoiceFinder history
VoiceFinder history • Based on research from the late 80’s • New findings in 1994- 1995 at Eriksholm • Refined and tested in collaboration between Eriksholm and Strandvejen 11
Slide 12: VoiceFinder background
VoiceFinder background • VoiceFinder detects speech, even under very difficult conditions • The principle: • Incoming signals are analyzed using an efficient narrow-band filter bank • Detects the presence of synchronous energy and harmonic structures of vowels • Very robust detection of speech even
Slide 13: VoiceFinder basics I
Capital Market Da « William Demant / Holding VoiceFinder basics! Speech Mode 4 band frequency analyzer Comfort x ~ Mode — f - 4 i f —— VoiceFinder a | (Detection of Hearing aid synchronous energy) microphone { \ . t 4 . Input Signal 13
Slide 14: VoiceFinder basics II
VoiceFinder basics II Input signal Low-pass 1500-2500 Hz 2500-3500 Hz 3500-4500 Hz 4500-5500 Hz 14
Slide 15: VoiceFinder: what was so difficult?
VoiceFinder: what was so difficult? • Distinguishing voice from noise! • Implementation considerations • Extensive math must be mapped to comply with hearing aid demands • Testing • Several rounds of user tests at Eriksholm and Strandvejen Concept Development Testing Implementation 15
Slide 16: VoiceFinder
VoiceFinder • Speech • Noise “Looks like speech” • Speech + noise 16
Slide 17: OpenEar Acoustics
OpenEar Acoustics • Dynamic Feedback Cancellation (DFC) is the enabling technology 17
Slide 18: Dynamic Feedback Cancellation
Dynamic Feedback Cancellation Feedback Hearing + - aid Feedback amplifier • The textbook solution Adaptive • … doesn’t work! LMS algorithm 18
Slide 19: DFC history
DFC history • Research/development project initiated in 1994 • Collaboration with Eriksholm and foreign universities • First user tests in 1998 • Launched in 2001 • Three patent applications 19
Slide 20: DFC: this works!
Lowers Lowerssensitivity sensitivitytowards towards DFC: this works! low-frequency sounds low-frequency sounds Change Changefrom fromslow slowto to fast adaptation fast adaptation Watchdog Watchdog Best Bestestimate estimateof ofthe the acoustic acoustic feedbackpath feedback path 20
Slide 21: DFC: testing
DFC: testing • Heimdal • Based on the most advanced floating-point DSP processor • Li-ion battery lasts 8-10 hours • Difficulties with implementation • Processor capacity • Speed • Precision 21
Slide 22: DFC: IC implementation
DFC: IC implementation • Power consumption reduction • Wearable DSP processor: > 300 mW (for research purposes) • Adapto IC version: < 0.1 mW • Reduction factor > 3000 • Performance improvement • Due to limitations in the wearable DSP 22
Slide 23: The DSP/IC development process
The DSP/IC development process • DSP and IC development must be in tune • Concurrent engineering is a must IC development Algorithm development 23
Slide 24: DSP in the future
DSP in the future • Growing • In importance for true user benefits • In complexity to accomplish true user benefits • Part of the development effort • Puts demands on implementation • Efficient implemtation of DSP algorithms will become increasingly important 24
William Demant Holding — Investor Day (December 2002) on 6ix