Non-Linear Convolution: A New Approach For The Auralization of Distorting Systems

NON-LINEAR CONVOLUTION: A NEW
APPROACH FOR THE AURALIZATION

OF DISTORTING SYSTEMS
Angelo Farina, Alberto Bellini and Enrico Armelloni

Industrial Engineering Dept., University of Parma, Via delle Scienze 181/A
Parma, 43100 ITALY – HTTP://pcfarina.eng.unipr.it
Goals for Auralization
 Transform the results of objective electroacoustics
measurements to audible sound samples suitable
for listening tests
 Traditional auralization is based on linear
convolution: this does not replicates faithfully the
nonlinear behaviour of most transducers
 The new method presented here overcomes to this
strong limitation, providing a simplified treatment
of memory-less distortion
Methods
 We start from a measurement of the system
based on exponential sine sweep (Farina, 108th
AES, Paris 2000)
 Diagonal Volterra kernels are obtained by post-
processing the measurement results
 These kernels are employed as FIR filters in a
multiple-order convolution process (original
signal, its square, its cube, and so on are
convolved separately and the result is summed)
Exponential sweep measurement
 The excitation signal is a sine sweep with constant

amplitude and exponentially-increasing frequency
Raw response of the system
Many harmonic orders do appear as colour stripes

Deconvolution of system’s impulse response
The deconvolution is obtained by convolving the raw response with a suitable

inverse filter
Multiple impulse response obtained
1st
3rd
5th
2nd
The last peak is the linear impulse response, the preceding

ones are the harmonic distortion orders
Auralization by linear convolution
Convolving a suitable sound sample with the linear IR, the frequency
response and temporal transient effects of the system can be simulated
properly
What’s missing in linear convolution ?
 No harmonic distortion, nor other nonlinear

effects are being reproduced.
 From a perceptual point of view, the sound is
judged “cold” and “innatural”
 A comparative test between a strongly nonlinear
device and an almost linear one does not reveal
any audible difference, because the nonlinear
behavior is removed for both
Theory of nonlinear convolution
 The basic approach is to convolve separately, and
then add the result, the linear IR, the second order
IR, the third order IR, and so on.
 Each order IR is convolved with the input signal
raised at the corresponding power:
M 1 M 1 M 1
y( n )   h1  i   x  n  i    h 2  i   x  n  i    h 3  i   x 3  n  i   .....
2
i 0 i 0 i 0
The problem is that the required multiple IRs are not the
results of the measurements: they are instead the diagonal
terms of Volterra kernels
Volterra kernels and simplification
 The general Volterra series expansion is defined as:
M 1 M 1 M 1
y( n )   h1  i1   x  n  i1     h 2  i1 , i 2   x  n  i1   x  n  i 2  
i1  0 i1  0 i 2  0
M 1 M 1 M 1
    h 3  i1 , i 2 , i 3   x  n  i1   x  n  i 2   x  n  i 3   .....
i1  0 i 2 0 i3 0
This explains also nonlinear effect with memory, as the system

output contains also products of previous sample values with
different delays
Memoryless distortion followed by a linear
system with memory
Noise n(t)
distorted signal
input x(t) Not-linear output y(t)
w(t) linear system
system
w(t)h(t)
+
K[x(t)]
 The first nonlinear system is assumed to be memory-less,

so only the diagonal terms of the Volterra kernels need to
be taken into account.
 Furthermore, we neglect the noise, which is efficiently

rejected by the sine sweep measurement method.
Volterra kernels from the measurement results
The measured multiple IRs h’ can be defined as:
y( t )  h'1  sin var   h' 2  sin 2  var   h' 3  sin 3  var   ...
We need to relate them to the simplified Volterra kernels h:

y( t )  h1  sin var   h 2  sin 2  var   h 3  sin 3  var   ...
Trigonometry can be used to expand the powers of the sinusoidal terms:

1 1 3 1
sin2         cos 2      sin 3        sin       sin 3     
2 2 4 4
3 1 1
sin 4         cos 2        cos 4     
8 2 8
5 5 1
sin 5        sin        sin  3        sin  5     
8 16 16
Finding the connection point
Going to frequency domain by taking the FFT, the first
equation becomes:
Y()  H '1    X   H ' 2    X  / 2  H ' 3    X  / 3  ...
Doing the same in the second equation, and substituting the
trigonometric expressions for power of sines, we get:
 3 5   1 1 
Y( )   H1   H 3   H 5   X     H 2   H 4   j  X  / 2 
 4 8   2 2 
 1 5  1 1
   H 3   H 5   X  / 3   H 4  j  X  / 4   H 5  X  / 5  ...
 4 16  8 16
The terms in square brackets have to be equal to the

corresponding measured transfer functions H’ of the first
equation
Solution
 Thus we obtain a linear equation system:
 3 5
 1H '  H 1   H 3   H5
4 8
 We can easily solve it,
 2

 H '2   j  1  H 2  H 4  obtaining the required Volterra

 1
 H '3    H 3   H 5
5 kernels as a function of the
 4 16 measured multiple-order IRs:
 1
 H ' 4  j   H4
8
  H1  H '1 3  H '3  5  H '5
H '5  1  H 5 
 16  H 2  2  j  H ' 2  8  j  H '4

 H 3  4  H '3  20  H '5

 H 4  8  j  H ' 4
 H 5  16  H '5

Non-linear convolution
As we have got the Volterra kernels already in
frequency domain, we can efficiently use them in a
multiple convolution algorithm implemented by
overlap-and-save of the partitioned input signal:
input x(t)
x2 x3 x4
Normal Squared Cubic signal Quartic

signal signal signal
h1(t)  h2(t)  h3(t)  h4(t) 

output y(t)
+ + +
Software implementation
Although today the algorithm is working off-line (as a mix of
manual CoolEdit operations and some Matlab processing), a more
efficient implementation as a CoolEdit plugin is being worked out:
This will allow for real-time operation even with

a very large number of filter coefficients
Audible evaluation of the performance
Original signal Linear convolution
These last two were compared in

a formalized blind listening test
Live recording Non-linear multi convolution

Subjective listening test
 A/B comparison
 Live recording & non-linear
auralization
 12 selected subjects
 4 music samples
 9 questions
 5-dots horizontal scale
 Simple statistical analysis of
the results
 A was the live recording, B
was the auralization, but the
listener did not know this
95% confidence intervals

of the responses
Conclusion
Statistical parameters – more advanced statistical methods
would be advisable for getting more significant results
Question Number Average score 2.67 * Std. Dev.
1 (identical-different) 1.25 0.76
3 (better timber) 3.45 1.96
5 (more distorted) 2.05 1.34
9 (more pleasant) 3.30 2.16
Final remarks
- The CoolEdit plugin is planned to be released in two months – it
will be downloadable from HTTP://www.ramsete.com/aurora
- The sound samples employed for the subjective test are available
for download at HTTP://pcangelo.eng.unipr.it/public/AES110
- The new method will be employed for realistic reproduction in a
listening room of the behaviour of car sound systems

Non-Linear Convolution: A New Approach For The Auralization of Distorting Systems

Uploaded by

Copyright:

Available Formats

Non-Linear Convolution: A New Approach For The Auralization of Distorting Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Non-Linear Convolution: A New Approach For The Auralization of Distorting Systems

Uploaded by

Copyright:

Available Formats

NON-LINEAR CONVOLUTION: A NEW

APPROACH FOR THE AURALIZATION

Angelo Farina, Alberto Bellini and Enrico Armelloni

 The excitation signal is a sine sweep with constant

Many harmonic orders do appear as colour stripes

The deconvolution is obtained by convolving the raw response with a suitable

The last peak is the linear impulse response, the preceding

 No harmonic distortion, nor other nonlinear

This explains also nonlinear effect with memory, as the system

 The first nonlinear system is assumed to be memory-less,

 Furthermore, we neglect the noise, which is efficiently

We need to relate them to the simplified Volterra kernels h:

Trigonometry can be used to expand the powers of the sinusoidal terms:

The terms in square brackets have to be equal to the

Normal Squared Cubic signal Quartic

h1(t)  h2(t)  h3(t)  h4(t) 

This will allow for real-time operation even with

These last two were compared in

Live recording Non-linear multi convolution

95% confidence intervals

You might also like