1. 3D sound reproduction with “OPSODIS” and its commercial applications
Takashi Takeuchi, PhD
Chief Technical Officer OPSODIS Limited
Institute of Sound and Vibration Research, University of Southampton Kajima Technical Research Institute

2. University of Southampton University of Southampton
History
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
Student
University of Southampton
Acoustic
Consultant
(Kajima)
Staff
ISVR
University of Southampton
3D sound
Business
MSc
PhD
(Invention)
(OPSODIS)
Auralisation
OPSODIS
Joint R&D
Joint R&D
Stereo Dipole
OPSODIS (UK Ltd.)

3. Optimal Source Distribution
What is OPSODIS™?
Optimal Source Distribution
Principle
- 3D sound reproduction
- Binaural
- Loudspeaker reproduction
- 2 channel system (3ch option)
- Stereo & Surround
- reproduction compatible
What is special?
- Lossless crosstalk cancellation!
- Signal processing by loudspeaker
- Good & natural sound quality
- Robust control
- Elevation (height) reproduction
- Multiple listeners
Picture above; acoustic image created with Stereophony.
panning between two loudspeakers could generate a segment of sound images.
could feel depth
Picture below; acoustic image created with OPSODIS.
true 3D sound image nothing to do with loudspeaker direction or distance.
with high quality sound.

4. Spatial hearing and binaural signal
Sound source signal s
Characteristics of space (physics)
（relative position，response of space）
Spatial hearing (human)
（source position，movement，room size，reverberation）
stereo
surround A2
binaural A1
Binaural signals d
any number of sources …
Physics
　source signal　　　　propagation　　　　two different signal at ears
Perception
　Separates signal and spatial information　　　　binaural signals

5. Inverse filtering (crosstalk cancellation)
Loudspeaker reproduction interweaves loudspeaker position
In order to reproduce spatial information arbitrarily, independent transmission to each ear is necessary.
Minimum of two independent (real) sound source necessary
C22
Inverse filtering H
C12
binaural
Signals d
Synthesised
binaural signals
C21 w
C11

6. Conventional cross-talk cancellation
Independent control at both ears
Cancelling each other
Right ear t
Left ear t
Loudspeaker output t R t
Right ear L t t
Left ear

7. Conventional cross-talk cancellation
Independent control at both ears
Right ear t
Left ear t
Loudspeaker output t R t
Right ear L t t
Left ear

8. Conventional cross-talk cancellation
Independent control at both ears
Massive destructive
interference
Right ear t
Left ear t
Loudspeaker output t R t
Right ear L t t
Repetitive delay = Large output
(Frequency dependent)
Left ear

9. Crosstalk cancellation (frequency)
phase
amplitude
Destructive interference at both ears.
Inefficient, quality loss, sensitive to errors.

10. Frequency dependency Conventional frequency frequency Left ear
from Left SP
from Right SP
reproduced
frequency
frequency
Left ear
Right ear

11. Why have 3D systems not performed well so far?
Frequency response to listener
6dB
(two point sources)
(_____ ) 　　　　　
0dB
In-phase
signal
level
(_ _ _ )
Out-of-phase
signal
-40dB
20Hz
Frequency (linear)
20kHz
Sound emitted by loudspeakers do not reach listener’s ears at certain frequencies.
Large output required to compensate it.

12. Loss of performance through cross-talk cancellation process
Typical frequency response of the inverse filters
for 3D reproduction.
0dB
Out-of-phase
signal
level
In-phase
signal
-40dB
20Hz
Frequency
20kHz

13. Loss of performance through cross-talk cancellation process
0dB
Original
Performance
Dynamic range loss = Low Quality
≈ 5 bit
Synthesized sound
-30dB
Distortion
Noise
level
Exact matching
20Hz
Frequency
20kHz
Cross-talk cancellation degrades performance severely.

14. Analogy to picture quality
Reduce;
Constant resolution
Magnify;
Pixel size
Quality Loss

15. Loudspeaker span 180° 10° (Stereo Dipole) ≈ - 3 bit frequency (log)
Original Performance
0 dB
≈ - 3 bit
≈ - 7 bit
-16dB
Smaller loss
More complexity
Less complexity
-43dB
Larger loss
Synthesized sound
20Hz
20kHz
20Hz
20kHz
frequency (log)
frequency

16. Optimal Source Distribution
Conventional
Transducer position vs Performance
(excess amplification)
poor
20kHz
OPSODIS
Conventional
Stereo Dipole
2kHz
frequency
200Hz
high
frequency
low
OPSODIS
20Hz
good 0°
60°
120°
180°
transducer position

17. Conventional OPSODIS (Stereo Dipole) The worst conditioned frequency
dictates performance
for all frequencies!!
(Stereo Dipole)
OPSODIS

18. OPSODIS loudspeaker Frequency response to listener with OPSODIS.
No frequency dependency on
amplitude response and
phase response
OPSODIS
0dB
level
-40dB
frequency

19. OPSODIS principle Frequency response of the OPSODIS inverse filters.
No frequency dependency on
amplitude response and
phase response
0dB
robust
Simple process
No loss resulting from signal manipulations
High quality 3D reproduction
Natural & pure sound quality
Elevation reproduction
level
-40dB
frequency

20. Crosstalk cancellation of OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear
Cancellation R
Right ear
½ amplitude, -90º phase
Addition
½ amplitude L
Left ear
Loudspeaker output

21. Crosstalk cancellation of OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear R
Right ear
½ amplitude, -90º phase
½ amplitude L
Left ear
Loudspeaker output

22. Crosstalk cancellation of OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear
Destructive interference R
Right ear
½ amplitude, -90º phase
Constructive interference
½ amplitude L
Left ear
Loudspeaker output
Efficient, Lossless, Robust to errors.

23. Reproduction quality Example: 16bit hardware Stereo Dipole frequency
Reproduced
dynamic range
≈ +3 dB
Reproduced
dynamic range
≈ -43 dB
16 bit
16 bit
OPSODIS
16 bit
16 bit
≈9 bit
L channel
R channel
reproduced

24. Electro-acoustic transducer for OPSODIS
Flat panel loudspeakers
plate
high frequency low
shaker
driving plate
narrow width wide
Acoustic waveguide
large stiffness small
This principle requires a pair of monopole transducers whose position from which sound is radiated varies continuously as frequency varies.
This might be realised by exciting a plate at each position individually.
Or, exciting a triangular shaped plate whose width and stiffness varies along its length.
Alternatively, changing the width of a slot along an acoustic waveguide might do the job.
slot
driver
high frequency low
narrow slot wide
high frequency low
Variation in excited frequency and radiation impedance

25. Discrete system Width of n enables discretisation
frequency(Hz)
It is also possible to realise a practical system even when a continuous variable span transducer pair is not available.
By allowing n to have small width nu, it is possible to discretise into a number of transducer pairs with a limited frequency range.
This principle is extremely useful and practical since a single transducer which can cover the whole audible frequency range is not practically available anyway.
Discretisation also expands the operational frequency range by lowering fl.
Please note that this is still a simple 2-channel sound system where only two independent signals are necessary to synthesise full 3D spatial sound environment.
transducer span
Width of n enables discretisation
2 channel, multi-way system
Use of conventional transducers
Wide frequency range
U-shaped valley
Small performance loss

26. Effects of room reflection (sound radiation)
Small radiation
by OPSODIS
Not affected by reflection
- dB
0 dB
- dB
0 dB
0 dB
Reference to radiation
by mono loudspeaker
Very large sound radiation other than receiver directions
Needs anechoic room
conventional
OPSODIS

27. Radiation pattern of Conventional
0dB
-∞dB
Radiation pattern is different depending on the frequency.
Single listener．
Very small sweet area.

28. Radiation pattern of OPSODISB A B A
0.6m
0.6m
A: 1 or 3 listener(s)
B: 2 listeners
Radiation pattern is constant regardless the frequency; Large sweet area
No spectral change; elevation works everywhere.
Multiple listener listening with a little performance loss.
The interval of L-R cross-talk cancellation is about 0.3m; Alternative seating positions.

29. Reproduced sound level (lateral position）
Left ear signal
Right ear signal
Lateral position　（m）
3 seater or 2 seater

30. Aspects of the “OPSODIS”
No dynamic range loss
Quality (Lossless)
Good S/N
No distortion
Robust to errors in plants and filters
Robust to individual difference in HRTFs
Generic inverse filters
Flat amplitude response of inverse filters
Elevation (height etc) works at any location
No colouration at any location
Small sound radiation other than receiver directions
Robust to room reflections
Consistent radiation over frequency
Multiple listener listening
With this assumption, the expression for the inverse filter matrix H becomes very simple like this, in fact, g is a real positive number close to 1, so the binaural synthesis over loudspeakers is nearly achieved by simple addition with 90° phase change.
||H|| is 1 over square root of 2 for all frequencies, so there are no dynamic range loss, which means good S/N and no distortion.
The condition number is 1 which is smallest possible for all frequencies, so the system is robust to errors in plant and inverse filter matrix at all frequencies.
Inverse filters have flat frequency response so there are no colouration at any location in the listening space, even outside the sweet spot.
As illustrated here, the sound radiation by the OSD system is always smaller than by a single monopole transducer producing the same sound level at the ears.
So it is also robust against reflections of a listening room.

31. 3 channel OPSODIS

32. Inverse filter ~ 2 channel OSD
frequency
high
low
0dB
Out-of-phase
component
level
In-phase
component
2ch OSD
-40dB
Frequency/azimuth

33. Inverse filter ~ 3 channel OSD
frequency
high
low
high
low
low
0dB
Out-of-phase
component
level
In-phase
component
3ch OSD
-40dB
Frequency/azimuth

34. Extension to 3 channel 2ch OSD 3ch OSD 2ch OSD 3ch OSD Out-of-phase
component
2ch OSD
3ch OSD
2ch
2ch OSD
In-phase
component
3ch
3ch OSD
Singular values of the inverse filter matrix H

35. 3 channel OSD 2ch OSD peak 3ch OSD 2ch OSD 3ch OSD valley
excess amplification
peak (in phase)
= valley (out of phase)
3ch OSD
peak
20kHz
peak (out of phase)
= valley (in phase)
2kHz
frequency
2ch OSD
200Hz
3ch OSD
20Hz
valley 0°
60°
120°
180°
transducer azimuth

36. Crosstalk cancellation of 3ch OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear
Cancellation
¼ amplitude, -90º phase R
Right ear C
½ amplitude
Addition
Left ear L
¼ amplitude, +90º phase
Loudspeaker output

37. Crosstalk cancellation of 3ch OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear
¼ amplitude, -90º phase R
Right ear C
½ amplitude
Left ear L
¼ amplitude, +90º phase
Loudspeaker output

38. Crosstalk cancellation of 3ch OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear
Destructive interference
¼ amplitude, -90º phase R
Right ear C
½ amplitude
Constructive interference
Left ear L
¼ amplitude, +90º phase
Loudspeaker output
Efficient, Lossless, Quality, Robust to errors.

39. Crosstalk cancellation of 3ch OPSODIS
Independent control at both ears
Right ear
phase
amplitude
Left ear
Destructive interference
1/3 amplitude, -120º phase R
Right ear C
1/3 amplitude
Constructive interference
Left ear L
1/3 amplitude, +120º phase
Loudspeaker output
Efficient, Lossless, Quality, Robust to errors.

40. Summary
Combination of signal processing and the new type of loudspeaker based on a new principle
enables ideal 3D sound
delivered to ears.
OPSODIS
Enabling high definition 3D sound reproduction with 2 channel (or 3ch) loudspeakers
Natural sound quality without distortion
Natural sound at any location
Effective in any room
Compatible with surround reproduction