Watermark G a Watermark Decoder, Met d for Providing a Watermarked
Based o Discrete Valued t a d Method for Providing Discrete Valued
Data Depende e o a Watermarked Sig al
Description
Technical Field
Embodiments according to the present invention are related to a watermark generator for
providing a watermark signal based on discrete valued data. Further embodiments
according to the present invention relate to a watermark decoder for providing discrete
valued data in dependence on a watermarked signal. Further embodiments according to the
present invention are related to a method for providing a watermark signal based on
discrete valued data. Further embodiments according to the present invention are related to
a method for providing discrete valued data in dependence on a watermarked signal.
Further embodiments are related to corresponding computer programs.
Background of the Invention
In the following different watermarking systems shall be reviewed in short.
A watermarking system can be viewed as a communication system. Let the bit- ise
information to be transmitted be represented by a watermark signal "wm", which is the
desired signal. The signal wm is 'embedded' into a host signal "a" by adding the two
signals (the watermark signal wm and the host signal a), obtaining a watermarked signal
"aw ". With respect to the watermark, the host signal can be seen as an additive
distort ion This means that awm deviates from its ideal value wm, corrupting the decoding
process (if the original host signal a is not known at the decoder). The signal awm is
further affected by a transmission channel, in that the channel introduces distortions.
Examples of transmission channels are the compression of the signal awm with an audio
codec such as A A C as well as the playback of the signal awm with a loudspeaker, its
propagation in air, and its pickup with a microphone.
A characteristic of watermark systems is that one part of the distortion, namely the host
signal, is known at the transmitter. If this information is exploited during embedding the
method is called informed embedding or watermarking with side information (see also
Ingemar J. Cox, Ed., Digital watermarking and steganography, The Morgan auf am
series in multimedia information and systems Morgan Kaufma , Burlington, 2 . ed.
edition, 2008). In principle, weighting the watermark wm according to power levels given
by a perceptual model is already a ease of informed embedding. However, this information
is used merely to scale the watermark in order to make it imperceptible whereas the host
signal is still seen as an unknown noise source for the generation of the watermark prior to
the weighting. In certain cases, it is possible to create the watermark signal in a way that
compensates for the host signal induced distortion so that only channel -induced distortion
corrupts the decoding. Such methods are called host-interference rejecting methods (see
a so Chen and Wornell, Quantization index modulation: A class of provably good
methods for digital watermarking and information embedding," IEEE TRANSACTION
ON INFORMATION THEORY, May 2001, vol. VOL. 47) .
n EP non pre-published 10 54964,0-1224 differential encoding has been introduced in
combination with BPSK- (binary phase shift keying) signaling to obtain a system which is
robust with respect to movement of the decoding device (for example if the signal is
picked up by a microphone), potential frequency mismatch between the local oscillators in
the transmit (Tx) and receive (Rx) sides and potential phase rotations introduced by a
frequency selective channel, such as the propagation in the reverberant environment.
The robustness comes from the fact that the information is coded in the phase difference
between two adjacent symbols, so that the system is virtually unaffected by a slowly
drifting phase rotation of the modulation constellation.
Although the method described in EP 10154694.0-1224 uses information about the host
signal a by scaling the watermark signal wm in order to make it imperceptible, the host
signal a is still an additional source of unknown noise from the communication system's
perspective, n other words, the watermark signal wm (prior to the perceptually motivated
scaling) is generated regardless of any knowledge of the host signal a .
Several watermarking systems use some kind of informed embedding method but only a
few belong to the group of host-interference rejecting methods. Examples of these are lowbit
modulation (LBM) (Mitchell D. Swanson; Bin Zhu; Ahmed H. Tewf k " Data hiding
for video-in-video," IEEE International Conference on image Processing, 1997, vol 2,
pp. 676 - 679; Brian Chen and Gregory W. WornelL Quantization index modulation
methods for digital watermarking and information embedding of multimedia," Journal of
VLSI Signal Processing, vol, 27, pp. 7 - 33, 20 ) and quantization index modulation
(QIM) that was introduced in (Chen and Wornell, " Quantization index modulation: A
class of provabiy good methods for digital watermarking and information embedding,"
IEEE TRANSACTION ON INFORMATION THEORY, May 20 , vol. VOL 47, and
Brian Chen and Gregory Wornell, " System, method, and product for information
embedding using an ensemble of non-intersecting embedding generators," 1999,
WO99/60514A).
In QIM, it is first necessary to choose one or more parameters of a signal representation,
e.g.. the complex coefficients of a time-frequency representation. The parameters chosen
are then quantized according to information to be embedded. In fact, each informationcarrying
symbol is linked with a certain quantizer; alternatively a whole message is linked
with a sequence of quantizers. Depending on the information to b transmitted, the signal
is quantized with the quantizer or sequence of quantizers associated with the information.
For instance, if the parameter to be quantized was a positive real number, the quantizer to
be used to embed a 0 could be defined by the quantization steps 0, 2, 4, 6, . . . whereas the
quantizer for a 1 could be 1, 3, 5, . , .. If the current value of the host signal was 4.6 the
embedder would change the value to 4 in case of a bit 0 and to 5 in case of a . At the
receiver, the distance between the received signal representation and all possible quantized
representations is calculated. The decision is made according to the minimum distance. In
other words, the receiver attempts to identify which of the available quantizers has been
used, By doing so, host-interference rejection can be achieved.
Of course, quantizing certain signal parameters may introduce perceivable distortion to he
host signal in order to prevent this th quantization error may be partly added back to the
signal which is referred to as distortion-compensated QIM (DC-QIM) (see also Antonius
Kalker, "Quantization index modulation (QIM) digital watermarking of multimedia
signals," 2001, WO03/053064). This is an additional source of distortion at the receiver.
Although it has been shown that DC-QIM is optimal for the AWON (additive white
Gaussian noise) channel and regular QIM is near-optimal (see also Chen and Wornell,
" Quantization index modulation: A class of provabiy good methods for digital
watermarking and information embedding," IEEE TRANSACTION ON
INFORMATION THEORY, May 2001, vol. VOL, 47) the methods have certain
drawbacks. They allow for high bit rates bu are especially sensitive to amplitude scaling
attacks (see also Fabricio Ourique; Vinicius Licks; Ramiro Jordan; Fernando Perez-
Gonzalez, "Angle qim: A novel watermark embedding scheme robust against amplitude
scaling distortions," IEEE International Conference on Acoustics, Speech, and Signal
Processing (ICASSP), March 2005).
Another method (derived from QIM) is named Angle Q M (AQIM) and was proposed in
the article of Fabricio Ourique; Vinicius Licks; Ramiro Jordan; Fernando Perez-Gonzalez,
Angle qim: A novel watermark embedding scheme robust against amplitude scaling
distortions," IEEE international Conference on Acoustics, Speech, and Signal Processing
(ICASSP), March 2005. There the information is embedded via the quantized angular
coordinates. By doing so, robustness against amplitude scaling can be achieved. This
method does not provide differential modulation and is therefore not robust against phase
drift.
Other watermarking systems exist where the information is embedded into the phase of the
audio signal. The methods presented in the article of W. Bender, D, Gruhl, N. Moriraoto,
and Aiguo Lu, " Techniques for data hiding, " IBM Sys J., vol. 35, no. 3-4, pp. 313—
336, 1996 and S. K o J.D, Johnston,W. Turin, and S,R. Quackenbush, " Covert audio
watermarking using perceptually timed signal independent multiband phase modulation,"
IEEE International Conference on Acoustics, Speech, and Signal Processing, (ICASSP),
2002, vol 2, pp. 1753 - 1756 are non-blind methods and therefore limited to only a sma
number of applications. In the article of Michael Arnold, Peter G. Baum, and Walter
Voesing, " A phase modulation audio watermarking technique," pp. 2- 16, 2009, a
blind phase modulation audio watermarking technique is proposed which is called
Adaptive Spread Phase Modulation (ASPM). Additionally, these phase modulation
methods do not have the host-interference rejection property and do not take the
differential coding nto account.
Many more watermarking methods exist, including spread spectrum or echo-hiding
methods. But as already stated in EP 54964.0-1 224 these methods may not be
applicable to certain tasks of interest, e.g. transmitting a watermark over an acoustic path
in a reverberant environment.
t is an object of the present invention to create a watermarking concept, which allows for
an improved robustness of a watermark signal which is embedded into a host signal and
transmitted in the host signal over a communication channel.
u u of the n n
This object is achieved by a watermark generator according to claim 1, a watermark
decoder according to claim 1 , a method for providing a watermark signal based on
discrete valued data according to claim 14, a method for providing discrete valued data in
dependence on a watermarked signal according to claim 5 and a computer program
according to claim 16.
An embodiment according to the present invention creates a watermark generator for
providing a watermark signal as a sequence of subsequent watermark coefficients based on
discrete valued data. The watermark generator optionally comprises an information
processor configured to provide, in dependence on information units of the discrete valued
data, a stream of subsequent stream values, such that the stream represents the discrete
valued data. The watermark generator further comprises a differential encoder configured
to provide the watermark signal The differential encoder is further configured to apply a
phase rotation to a current stream value of the stream values representing the discrete
valued data or to a curren watermark symbol, the current watermark symbol
corresponding to a current stream value of the stream values representing the discrete
valued data, to obtain a current watermark coefficient of the watermark signal. The
differential encoder is further configured to derive a phase of a previous spectral
coefficient of a watermarked signal which is a combination of a host signal and the
watermark signal. The differential encoder is further configured to provide the watermark
signal such that a phase angle of the phase rotation applied to the current stream value or
the current watermark symbol is dependent on the phase of the previous spectra!
coefficient of the watermarked signal.
It is an idea of the present invention that a watermark signal is more robust, especially with
respect to a degradation, for example, by Doppler effect, if a differential encoding of a
watermark coefficients is performed such that a phase of a current watermark coefficient is
based on a phase of a previous spectral coefficient of a watermarked signal which is to be
embedded in a host signal. Embodiments of the present invention combine host
interference rejection with differential encoding. This concept of deriving the phase of the
current watermark coefficient based on the phase of a previous spectral coefficient of the
watermarked signal reduces the distortion induced by the host signal and thus improves a
decoding process, for example at a decoder being configured to extract the watermark
signal out of the watermarked signal.
In a preferred embodiment of the present invention the information processor may be
configured to provide the stream representing the discrete valued data n a time frequency
domain, such that each stream value of the stream is associated to a frequency subchannel
with a center frequency and a time slot. The differential encoder may be configured to
obtain the current watermark coefficient in the time frequency domain, such that a
frequency subchannel associated to the current watermark coefficient is identical to a
frequency subchannel associated to the current stream value and such tha a time slot
associated to the current watermark coefficient is identical to a time slot associated to the
current stream value In other words, the current stream value and the current watermark
coefficient which corresponds to the current stream value may be associated to the same
frequency subchannel and time slot
The information processor may for example perform a time spreading and a frequency
spreading of the information units of the discrete valued data, such that every information
unit of the discrete valued data is represented by at least two different stream values of the
stream representing the discrete valued data, wherein different stream values representing
the same information unit differ in their associated frequency subchannels and/or time
slots.
Furthermore, the differential encoder may be configured to derive the spectral coefficients
of the watermarked signal in a time frequency domain such that each spectral coefficient of
the watermarked signal corresponds to a frequency subchannel and a time slot. The
differential encoder may further be configured to determine the phase rotation, such that a
time slot associated to the previous spectral coefficient of the watermarked signal in
dependence on which the phase angle of the phase rotation applied to the current stream
value or the current watermark symbol is chosen, and the time slot associated to the current
stream value are adjacent in time. The watermarked signal may for example be a sequence
of subsequent spectral coefficients and a current spectral coefficient of the watermarked
signal may be adjacent in time to (or may follow) the previous spectra! coefficieni of the
watermarked signal. Each spectral coefi cient of the watermarked signal may be a
combination of a spectral coefficient of the host signal associated to the same frequency
subchannel and time slot like the spectral coefficient of the watermarked signal and of a
watermark coefficient associated to the same frequency subchannel and time slot like the
spectral coefficient of the watermarked signal, The current spectral coefficient of the
watermarked signal may therefore be based on a current coefficient of the host signal and
the current watermark coefficient, wherein the phase of the current watermark coefficient
is based on (or is in some cases even identical to) the phase of the previous spectral
coefficient of the watermarked signal. The frequency subchannel associated to the previous
spectral coefficient of the watermarked signal may be identical to the frequency
subchannel of the current spectral coefficient of the watermarked signal and therefore also
to the frequency subchannels of the current coefficient of the host signal of the current
stream value and of the current watermark coefficient.
5
Brief Description of the Figures
Embodiments according to the present invention will be described taking reference to the
enclosed figures, in winch:
10
Fig 1 shows a block schematic diagram of a watermark generator, according to an
embodiment of the present invention;
Fig. 2 shows a bock schematic diagram of a differential encoder, for use in an
5 embodiment of the present invention;
Fig. 3a to 3c show diagrams of an example for phase rotation and scaling applied in the
differential encoder of Fig. 2;
20 Fig, 4 shows block schematic diagram of a differential encoder for use in an
embodiment of the present invention;
Fig. 5 shows a diagram of an example of mapping stream values to watermark
symbols;
-
Fig. 6a shows diagrams of possible outputs for different stream values in
dependence on a maximum number of watermark symbols associated to the
same stream value;
30 Fig 6b shows a diagram of an example on how a watermark coefficient is derived
in an M-point constellation using the differential encoder from Fig 4;
Fig. 7 shows a block schematic diagrain of a watermark generator according to a
further embodiment of the present invention;
35
Fig, 8a shows as a comparison example to Fig. 7, a block schematic diagram of a
watermark generator as described in EP 0154964;
Fig 8b shows a diagram as an example for the embedding principle of the
watermark generator from Fig, 8a;
Fig. 9 shows block schematic diagram of a watermarked audio signal provider
with a watermark generator according to an embodiment of the present
invention;
Fig. 10 shows a flow diagram of a method according to an embodiment of the
present invention:
Fig 11 shows a block schematic diagram of a watermark decoder according to an
embodiment of the present invention; and
Fig. 12 shows a diagram of an example of the mapping of different: phase angle
ranges to discrete values of discrete valued data like it is performed by the
watermark decoder from Fig. 11
Detailed Description of odi e of the Present .
Before embodiments of the present invention are explained in greater detail, taking
reference to the accompanying figures, it is to be pointed out that the sa e or functionally
equal elements are provided with the same reference numbers and that a repeated
description of these elements shall be omitted. Descriptions of elements provided with the
same reference numerals are therefore mutually interchangeable.
Watermark generator according to Fig. 1
n the following a watermark generator 00 will be described taking reference to Fig. ,
which shows the block schematic diagram of such a watermark generator. The watermark
generator 00 is configured to provide a watermark signal 102, also designated as "wm",
as a sequence of subsequent watermark coefficients. The watermark generator comprises
an optional information processor 106 and a differential encoder 108. The information
processor 06 is configured to provide, in dependence on information units (for example
bits) of the discrete valued data 104 (for example binary data), a first stream 10 of
subsequent stream values, such that the stream 1 0 represents the discrete valued data 104.
The differential encoder 108 is configured to provide the watermark signal 02 and to
apply a phase rotation 2 to a current stream value (for example, a stream value b(i, j)) of
the stream values representing the discrete valued data 4 or to a current watermark
symbol (for example, a watermark symbol Xk(i, j)) corresponding to a current stream value
(for example, to the stream value b(i, j)) of the stream values representing the discrete
valued data 104, to obtain a current watermark coefficient (for example a watermark
coefficient wm(i, j)) of the watermark signal 2.
The differential encoder 108 may therefore perform an optional stream value to watermark
symbol mapping 4.
The differential encoder 108 is further configured to derive a phase 6 of a spectral
coefficient (for example a spectral coefficient awm(i, j 1)) of a watermarked signal. The
watermarked signal is a combination of a host signal 8 and the watermark signal 102.
The watermarked signal may also be designated as "awm" and the host signal may also be
designated as "a".
The differential encoder 8 is configured to provide the watermark signal 102 such that a
phase angle of the phase rotation 2 applied to the current stream value or the current
watermark symbol is dependent on the phase 6 of the previous spectral coefficient of the
watermarked signal. In a preferred embodiment of the present invention, the phase angle of
the phase rotation 2 applied to the cun-ent stream value or the current watermark symbol
is equal to the phase angle of the previous spectral coefficient of the watermarked signal
To derive the phase 16 of the previous spectral coefficient of the watermarked signal the
differential encoder 108 may perform a phase derivation 120 on the previous spectral
coefficient of the watermarked signal. The previous spectral coefficient may, for example,
be provided from a stage which is external to the watermark generator 100, or the
differential encoder 108 may be configured to determine spectral coefficients of the
watermarked signal by combining watermark coefficients and spectral coefficients of the
host signal 18. For example, the differential encoder 108 may be configured to derive the
previous spectral coefficient of the watermarked signal based on a combination of a
previous watermark coefficient (for example a watermark coefficient wm(i, j - 1)) and a
previous spectral coefficient, for example a spectral coefficient a(i, j - 1)) of the host signal
, n other words, the differential encoder may not only derive watermark coefficients
b ut also spectral coefficients of the watermarked signal,
The information processor 106 may be configured to provide the first stream 0
representing the discrete valued data 104 in a time frequency domain, such that each
stream value of the stream 0 is associated to a frequency subchannel and a time slot.
The index "i" used above ay indicate the frequency subchannel and the index "j" may
indicate the "symbol number" or, i other words, the time slot of the corresponding
coefficient or symbol n other words "i" denotes a frequency subchannel (at center
frequency ;) and "j" denotes the temporal index or the time slot of the value corresponding
to it
Therefore each stream value of the stream 110 is associated to a frequency subchannel i
and a time slot j . Furthermore, the differential encoder 08 may be configured to obtain the
current watermark coefficient wm(i, j ) in th time frequency domain, such that a frequency
subchannel i associated to the current watermark coefficient wm(i, j ) is identical to a
frequency subchannel i associated to the current stream value b(i, j ) and such that a time
s o j associated to the current watermark coefficient m(i, j) is identical to a time slot j
associated to the current stream value b(i, j).
In other words, a frequency subchannel and a t e slot or symbol number of a watermark
coefficient of the watermark signal wm, which are associated to a stream value of the
stream value of the stream 0, may be identical to the frequency subchannel and time slot
or symbol number of the corresponding stream value
Furthermore, the differential encoder 1 8 ay be configured to derive spectral coefficients
of the watermarked signal in a time frequency domain too, such that each spectral
coefficient of the watermarked signal is associated to a frequency subchannel and a time
slot. The differential encoder 108 may therefore be configured to determine the phase
rotation 1 2 such that a time slot j - 1 which is associated to the previous spectral
coefficient awm(i, j - ) of the watermarked signal, in dependence on which the phase
angle of the phase rotation 1 2 applied to the current stream value b( j ) or the current
watermark symbol X j (i, j) is chosen, and the time slot j corresponding to the current stream
value b(i, j ) are adjacent in time. n other words, a current spectral coefficient awm(L j),
which is a combination of the current watermark coefficient wm( j ) and the current
spectral coefficient a( j ) of the host signal 8. may directly follow the previous spectral
coefficient awm(i, j - ) in ti e when the watermarked signal is viewed as a sequence of
subsequent spectral coefficients awm(i, j ) (i, j e N). Furthermore, the differential encoder
8 may be configured such that the frequency subchannels i of the current stream value
b(i, j), the current watermark coefficient wm(i, j ) and the spectral coefficient awm(i, j) are
identical. This means, the differential encoder 108 may perform the watermark signal
derivation process for every frequency subchannel onto which the information units of
discrete valued data 104 are mapped. This is advantageous because in a reverberant
environment different phase rotations to the transmitted signal may be applied to different
frequency subchannels. The phase of the current watermark coefficient wm(i, j ) may
therefore only be based on the previous spectral coefficient awm(i, j 1) of the
watermarked signal being associated to the same frequency subchannel i like the current
watermark coefficient wm(i, j).
n the following, a differential encoder 208 for use in a watermark generator according to
an embodiment of the present invention shall be explained taking reference to the Figs. 2,
3a to 3c. n the following, the host signal 8 will be an audio signal and may also be
designated as host audio signal in which the watermark signal is to be embedded.
Nevertheless, embodiments of the present invention may also be used for embedding
watermark signals in other signals than audio signal, for example in video signals. The
functionality of the differential encoder 208 may be equivalent to the functionality of the
differential encoder 108 and the differential encoder 208 may comprise the further
functionalities shown in Fig. 2.
The differential encoder 208 is configured to receive the host signal as a host audio
signal a(t) in the time domain. The differential encoder 208 may therefore comprise, as
shown in Fig. 2, an analysis filter bank 202 configured to obtain the spectral coefficients
(for example the current spectral coefficient a( j)) of th host signal 1 8 in the time
frequency domain. This analysis filter bank 202 may also be used in a corresponding
watennark decoder. In other words the audio coefficients (the spectral coefficients of the
host signal ) in the encoder may be obtained by applying the same analysis filter bank
202 that is used in a decoder. Obtaining the spectral coefficient of the host signal 1 may
be pa l of the phase derivation 120 which is performed by the differential encoder 208, To
obtain the current spectral coefficient awm(i, j ) of the watermarked signal th differential
encoder 208 may perform, during the phase derivation 120, a combination 204 of the
current watennark coefficient wm(i, j ) and the current spectral coefficient a(i, j ) of the host
signal 8 for example according to the following equation:
awm(i, j ) = a(i, j ) + wm(i, j). (1)
The obtained current spectral coefficient a m i j ) of the watermarked signal may be
stored in the differential encoder 208 (symbolized by a delay element 206 of the
differential encoder 208). The stored current spectral coefficient awm(i, j ) of the
watermarked signal may then be used to determine a following watermark coefficient
wm(i, j + 1).
To find the current watermark coefficient wm(i, j ) for the. frequency subchannel i, the
phase 6 of the previous spectral coefficient awm(i, j - I) which is also designated as
previous watermarked audio coefficient of the watermarked signal, is calculated, for
example, in a phase calculating process 1 of the phase derivation process 120 using the
following equation:
In this application " " designates the imaginary unit (square root of -1), i should not be
mixed up the index j for the time slots.
The phase 16 (represented by the complex value e i, ,, - ) or the phase angle p(i , j 1) of
the previous spectral coefficient awm(i, j - 1) is used by the differential encoder 208 in the
phase rotation 12, which is applied to the current stream value b(i, j ) of the stream 0. If
no phase change is to be transmitted, for example if the current stream value b(i, j ) is equal
to a phase of the current watermark coefficient wm(i, j ) is identical to the phase 6 of
th previous spectral coefficient a mi j 1) of the watermarked signal, or in other words
the current water ark coefficient wm(i, j ) points in the same direction as the previous
spectral coefficient awm(i, j - 1) of the watermarked signal If a phase change by %( 80°)
is to be transmitted (for example if the current stream value b(i, j ) is -1), the current
watermark coefficient wm(i, j ) may point in the opposite direction when compared to the
previous spectral coefficient awm(i, j - 1) of the watermarked signal The stream values
may, for example, be binary data, for example, the stream values may be:
b(i, i) (3)
At the beginning of the stream 10. that is for j = 0 or if the phase (e ' is undefined,
the phase 16 or ' may be set equal .
The differential encoder 208 may further perform a scaling of the current stream value b(i,
j ) by a current scaling factor y(i, j ) (i, j s N, g I ) or by a current factor which is smaller
than the current sealing factor ( j), The scaling 2 0 may be applied to the current stream
value b(i, j ) before applying the phase rotation 2 or after applying the phase rotation 2
(as it is shown in Fig, 2). The current scaling factor y(i, j ) is provided by a
psychoacoustical processing module (not shown in Fig. 2) in dependence on the host signal
118, into which the watermark signal 102 is to be embedded. The scaling factor i, j )
describes a masking characteristic of the host signal . The current scaling factor y(i, j )
may determine a maximum amplitude of the current watermark symbol wm(i, j ) such that
the current watermark coefficient w (i j ) stays inaudible in the watermarked signal.
In a . preferred embodiment of the present invention the maximum ampliiiide of the current
watermark coefficient wm(i, j ) allowed by the psychoacoustical model is always used.
The differential encoder 208 may therefore determine the current watermark coefficient
wm(i, j ) as:
wni(i, j ) b(i, j ) · i j ) - e . (4)
By using equation 4, the encoding strategy may get optimal, this means that a signal-tonoise
ratio at a decoder after differential decoding can be maximized. From equation 4 it
can be seen that a differential encoding is carried out implicitly, so that a signal b^i, j )
b(i, j ) i, j-1), which has to be computed in EP 10 154964, does not need to be
computed in embodiments of the present invention.
Furt hermore it is to be pointed out that in embodiments of the present invention the phase
rotation 2 during differential encoding is introduced, to archive a host interference
rejection with an implicit differential encoding, which is a significant advantage, especially
when compared to the embedding method stated in EP 1 154964
In Figs. 3 to 3c an example for the embedding process given.
Fig. 3a shows in a diagram the host signal plus watermark for the previous temporal slot,
namely the previous spectral coefficient awm(i, of the watermarked signal, as a
vector 3 0 in the complex plane. Furthermore, the current spectral coefficient a(i, j ) of the
host signal 18 is shown as another vector 3 2 in the complex plane. The current audio
signal afi, j ) or the current spectral coefficient a(i, j ) of the host signal 8 represents the
center of a circle on which the phase rotation 2 can be applied to the current stream
value b(i, j). Furthermore, the radius of the circle after scaling the current stream value b(i,
j ) may delimit some masking region of the host signal .
In other words, the radius of the circle may be scaled based on the current scaling factor
g (ί , j). As can be seen from Fig, 3a, the current stream value b(i, j ) may comprise a phase
of 0 or a phase of , depending on its value. As shown in equation 3, the current stream
value b( j ) may either take the value - or + , this rule may apply if the discrete valued
data 4 is binary data, the stream 0 therefore may comprise only binary stream values.
Th vector b ,( i, j ) may therefore correspond with a first value (for example -1) of the
current stream value b( j ) and the vector b ( i, j ) may correspond with a complementary
value (for example +1) of the current stream value b(i, j).
Fig 3b shows the circle from Fig. 3a after the phase rotation 2 has been applied to the
current stream value b(i, j). It. be seen from Fig, 3b, that the phase angle f( , j - 1) by
which the circle from Fig. 3a is rotated is identical to the phase angle f(ί j - 1) of the
previous spectral coefficient awm(i, j - 1) of the watermarked signal. In other words, the
current stream value b( j ) is phase shifted by the phase 116 of the previous spectral
coefficient awm(i, j - 1) of the watermarked signal
Fig, 3c shows the circle from Fig 3b after scaling 210 of the phase rotated current stream
value b( j). The circle therefore delimits the masking region. According to embodiments
of the present invention the watermark wm(i, j ) or the current watermark coefficient wm(i
j ) is constructed either as wm (i, j ) or as wm (i, j ) depending on the current stream value
b(i, j). As a comparison example, the possibilities for the traditional method stated in EP
10154964 namely wm( (i,j)or wm (i,j) , are also shown can be seen that for the
traditional method the current watermark coefficients m (i, j)or wm (i,j) are
constructed independent on the phase of a previous spectral coefficient of a watermarked
signal n other words, comparing the two strategies it can be observed that embodiments
of the present invention rotate the traditional solution by an angle dependent on the
previous watermarked signal awm(i, j - 1) or the previous spectral coefficient awm(i, j - 1)
of the watermarked signal.
To summarize, the current spectral coefficient awm(i, j ) of the watermarked signal can be
obtained by a summation of the current spectral coefficient a(i, j ) of the host signal and
a rotated (and, optimally, scaled) version of the current stream value b(L j ) such that the
current spectral coefficient awm(i, j ) lies on or within the circle, the center of which is
defined by the current spectral coefficient a(i , j ) of the host signal 118 a radius of which is
determined by a magnitude of the current stream value b(i, j ) and a range of values of the
current scaling factor g( j).
This embedding strategy according to embodiments of the present invention reduces the
distortion induced by the host signal 8 and thus improves the decoding process a a
decoder.
From Fig. 3c it can be seen that the current spectral coefficient a m(i, j ) of the
watermarked signal can be calculated depending on the current stream valu b(i, j ) as:
awm (i j ) - a(i, j ) ÷ wm ( , j ) (5)
or
aw r (i j ) = a(i, j ) + wm (i, j), (6)
depending on the value of the current stream value b(i, j).
The current scaling factor g (1, j ) and the phase 16 e w - i ) of the previous spectral
coefficient awm(i, j-l) of the watermarked signal are already included in wm '(i, j ) and
, j ) (see equation 4).
The circle of radius g ( , j ) delimits the area of the complex plane in which the current
watermark coefficient m(i, j), which is expressed as a vector centered in the current
spectral coefficient a(i, j ) of the host signal 1 , can be defined. The current scaling factor
g ( , j ) is provided by the psychoacoustical model and ensures that the watermark will be
inaudible. To achieve the highest SIR, namely the highest signal to interference (i.e. the
host signal ) ratio, it can be optimal to place the current watermark coefficient wm(i, j )
on the circle rather than within it. In other words, it may be optimal to use the maximum
allowed power of the watermark coefficients of the watermark signal 2. In other words it
may be optimal to use the current scaling factor (i, j ) provided by the psychoacoustical
model for the scaling 210 rather than another (also allowed) scaling factor which is smaller
than the current scaling factor y(i, j).
It has been found that with such a masking circle shown in the Fig. 3a to 3c not only
m(i ) = b (i, j ) · g (ί , j ) with b (i, j ) e { - 1, 1} is allowed, but that instead also
wm(i, j ) = i, j ) · e · b(i, j ) with Q [0, 2p) can also be used. Embodiments of the present
invention make use of this finding by applying the phase rotation 2 to the current stream
values of the stream 0 based on phases of previous spectral coefficients of the
watermarked signal Other masking regions might also be possible. For example, if
investigations showed that phase changes of an audio coefficient (of a spectral coefficient
of the host signal ) are less critical to the ear than amplitude changes, the masking
region (which is a circle in Figs. 3a to 3c) may have a kidney shape.
For simplicity reasons a circular masking region is assumed in the embodiments described
in this application.
n the embodiment described above the invention was presented in a more specific way,
this means in a two-point solution, wherein no stream value to watermark symbol mapping
is performed. This two-point solution may be of great importance for current practical
applications. However, more general multi-point solutions may be of interest in future
applications. Therefore, in the following, another embodiment of the invention will be
described which extends the specific two-point solution to a more general higher order
solution.
Following the same principles used n the previous sections, the invention is now
generalized to an M-point constellation. To do so, we allow a different symbols to map the
same information as shown in Fig. 5. In this plot, two symbols are allowed for each bit
state. For example, for the bit state "1" a first complex symbol 510 with a phase of 0 and a
second complex symbol 5 2 with a phase of p are allowed. For a second bit state "0" a
third complex symbol 520 with a phase of /2 and a fourth complex symbol 522 with a
phase of 3p/2 are allowed. The choice between the different symbols can be carried out,
once more following the informed embedding idea, this means by considering the current
host audio signal (i.e. the current spectral coefficient a(i, j ) of the host signal 8) as well
as (for the differential encoding) the previous watermarked signal (i.e. the previous spectral
coefficient awm(i, j -1) of the watermarked signal).
Fig. 4 shows a block schematic diagram of a differential encoder 408, for use in a
watermark generator according to an embodiment of the present invention. The differential
encoder 408 differs from the differential encoder 208 in that it is configured to perform the
stream value to watermark symbol mapping 1 4, which was also mentioned as an optional
feature of the watermark generator 100 in Fig. 1, To perform this stream value to
watermark symbol mapping 4, the differential encoder 408 comprises a subconsteilation
selector 402 and a decider 404. The decider 404 may also be called multi-point decider
404. The subconsteilation selector 402 is configured to selectively provide a plurality of
current watermark symbols x (i, j ) (i, j , k e N), which constitute a subconsteilation, in
dependence on the current stream value b(i, j). The index "k" is associated to a symbol
number of the current watermark symbol. If one stream value is mapped to a subset of M
watermark symbols belonging to two different subconstellations, for each subconsteilation
k may range from k=l to k M/2, A first subconsteilation may be associated with a first
stream value (e.g. +1) a d a second subeonsteilation may be associated with a second
stream value (e.g. -1).
Furthermore, the differential encoder 408 may be configured to apply a phase rotation 2
to each of th current watermark symbols x (i, j ) of the subeonsteilation corresponding to
th current stream value b(i„ j), to obtain a plurality of current candidate watermark
coefficients wrrs (i, j). As can be seen from Fig. 4. the differential encoder 408 further ay
be configured to perform a scaling 210 of each of the current watermark symbols X|.: i, j )
based on the current scaling factor g ( . j ) provided by the mentioned psychoacoustica!
module. The current scaling factor g(ϊ , j ) is equal for each of the current watermarksymbols
( j ) of the subeonsteilation corresponding to the current stream value b( j).
The differential encoder 408 may therefore also be configured to apply the phase rotation
2 to a scaled version of the current watermark symbols X (.i, j).
According to further embodiments of the present invention the differential encoder 408
may firstly perform the phase rotation 2 and then perform the scaling 210.
The decider 404 may be configured to choose one out of the current candidate watermark
coefficients , j ) as the current watermark coefficient w (i, j).
As an example, the differential encoder 408, or, more precise the subeonsteilation selector
402 may always code bit (one stream value of the stream 10) always with M symbols.
In the folio wing it is assumed that the stream 0 only contains binary values or in other
words the stream values of the stream 0 can only take binary values. For example, the
stream values may be either - 1 or +1. The current stream value b(i, j ) may therefore he
{-1, 1 . The current stream value b( , j ) or the bit b(i, j ) enters the subeonsteilation
selection block or the subeonsteilation selector 402 An output of the subeonsteilation
selector 402 comprises, as stated in Fig. 6, M/2 complex watermark symbols xi(i, j), . . .
, j). The M/2 points constitute the subeonsteilation, which corresponds to one b it state
or in other words the subeonsteilation corresponds to the current stream value b(i, j). Put in
mathematical terms, the k-th point of the subeonsteilation can be computed as:
xk(i, j ) exp ( (2D · (k - 1) ÷ .j))) (7)
where
(8)
and
h(ί , ) ·· 0 for b(i, j ) = 1
, ) fo (i ) ] .
As can be see in Fig, 6a for M = 2 there is only one symbol in each subconstellation and
we have simply x (i, j ) b(i, j), which was the case with the differential encoder 208 of
Fig. 2.
From Fig. 6 it can be seen tha for each state of the bit b( j) or for each value of the
current value b(i, j ) different subconstellations are provided by the subconsieliation
selector 402. Furt hermore it can be seen that the subconsieliation selector 402 may
provide the plurality of current watermark symbols ( j ) as complex values, such that
different current watermark symbols only differ in phase and such that phase differences of
different adjacent current watermark symbols associated with the same current stream
value b(i, j ) are equal. For example, for M 8, a phase difference between watermark
symbols of the same subconsieliation is always /2.
Furthermore, a first subconsieliation may con espond with a first value of the current
stream value b(i, j), for example, b(i, j ) = 1, and a second subconstellation may correspond
with a second value of the current stream value b(i, j), for example, b(i, j ) = - 1,
Furthermore, a phase difference between two adjacent watermark symbols of different
subconstellations is always equal a d is half of the phase difference between two adjacent
watermark symbols of the same subconstellation
As can be seen from Fig 4, each symbol in the subconstellation, in other words each
watermark symbol c ·( , j ) of the subconstellation is scaled according to the current scaling
factor g ( , j ) given by the psychoacoustic model and is then rotated according to the phase
6 of the previous spectral coefficient awm(i, j - 1) of the watermarked signal. Each
scaled and rotated symbol in tire subconstellation entering the decider 404, for example as
the current watermark candidates wn¾(i, j), a candidate for the current watermark
coefficient wm(i, j).
The decider 404 chooses which of the candidates, denoted by wn (i j),.,, w (i, j )
should be used as watermark (as the current watermark coefficient wm(i, j ))
One possibility is to choose the candidate out of the current candidate watermark
coefficients wmk(i, j ) (k . ..M/2) which maximizes a signal~io~noise ratio of the
watermarked signal with respect to channel noise. In this case the decider 404 may he
configured to add each candidate w , j ) (k = 1. ..M/2) to the current spectral coefficient
a(i, j ) of the host signal 1 8 to obtain watermarked signal candidates awn (i j), ...
awni 2 j), which are also denoted as current candidate watermarked spectral coefficients
aw i, j ) (k= M/2) and choose the one with the highest power n mathematical terms:
so thai the watermark signal or the current watermark coefficient is
(«, j ) — W (i j ) ,
In other words the decider 404 may be configured to derive the plurality of current
candidate watermarked spectral coefficients awn¾(i, j ) (k=l..,M/2) based on
combinations of the current spectral coefficient a(i, j ) of the host signal 8 with the
plurality of candidate watermark coefficients ( , j ) (k=l...M/2), to determine the
current candidate watermark spectral coefficient with the highest power out of the plurality
of current candidate watermarked spectral coefficients awms (i, j) (k=l . . .M/2) to choose
the current candidate watermark coefficient corresponding to the current candidate
watermarked spectral coefficient having the highest power as the current watermark
coefficientit
should be pointed out once more that embodiments of the present invention implement a
differential encoding implicitly.
For using a higher order M the signal-to-noise ratio of the watermarked signal can be
improved, meaning that the watermark may survive more easily the distortions introduced,
for example by microphone noise On the other hand, the symbols in the subconstellation
are closer for larger M (this means the phase difference gets smaller). This implies that the
bit error rate BER will increase. Given this tradeoff, d e choice of M depends on the
desired application.
Fig, 6b shows in an example how the decider 404 decides which current candidate
watermark coefficient out of the plurality of current candidate watermark coefficients
wm (i, j ) (k= ,, .M 2) to use as the current watermark coefficient wm(i, j). n the example
it is assumed that the subconstellation selector 402 codes one bit with one of a total out of
eight symbols. This means M = 8. Furthermore, it is assumed that the value of the current
stream value b(i, j ) = 1. From Fig. 6a the subconstellation for this ease can be found in the
first column, third row of the table show in Fig. 6a, The subconstellation therefore
comprises four current candidate watermark symbols X (i, j ) to x4(i, j), which are spaced
apart from each other with a phase of p/2. In Fig. 6b the current watermark symbols X i , j )
to X , j ) have already been scaled and phase rotated to obtain the current candidate
watermark coefficients wmi(i, j ) to wm (i j). The multipoint decider 404 derives, based on
the combination of the current spectral coefficient a( j ) of the host signal 8 and the
current candidate watermark coefficients wmi(i, j ) to w (i, j), the current candidate
watermarked spectral coefficients awm^i, j ) to awm (i, j). The decider 404 then chooses
the current candidate watermark coefficient which corresponds to the current candidate
watermarked spectral coefficient with the highest power. In the example shown in Fig. 6b
the decider 404 would choose the current candidate watermark coefficient (i, j ) as the
current watermark coefficient wm(i, j), because its corresponding current candidate
watermarked spectral coefficient awn , j ) has the highest power out of the current
candidate watermarked spectral coefficients awm (i, j ) to awn (i, j).
In other words, the watermark coefficients wm(i, j ) are chosen such that awm(i, j ) lies
within the masking region and the signal-to-noise ratio at the decoder after differential
decoding is maximized, e.g. when a decoding rule like in equation 3 is used,
It should be noted that to obtain the special case shown in the embodiment of Fig. 2,
namely M = 2, the scheme in Fig. 4 simplifies greatly. As already mentioned, the
subconstellation selector 402 can be superfluous as we have only one candidate which is
xi(i j ) = b(i, j). Therefore, the decider 404 may also be removed, as no choice is required.
After describing the two differential encoders 208, 408 n the following another watermark
generator, which may use the differential encoders 208, 408 will be explained taking
reference to Fig. 7.
The Watermark Generator According to Fig. 7
Fig. 7 shows a block schematic diagram of a watermark generator 700 according to an
embodiment of the present invention. A iunctionality of the watermark generator 700 may
be similar to functionality of the watermark generator 100. The watermark generator 700
may comprise the optional features shown in Fig. 7 , The watermark generator 700
comprises an information processor 706, a differential encoder 708 and a modulator . A
functionality of the information processor 706 may be similar to the functionality of the
information processor 06 and the information processor 706 may comprise the additional
features shown in Fig. 7 The differential encoder 708 may be the differential encoder 08,
the differential encoder 208, the differential encoder 408 or another differential encoder
according to an embodiment of the present invention.
For the description of the watermark generator 700 a stream 04 of binary data expressed
as 1} is assumed A signaling block 712 of the information processor 706 organizes
the data in packets of equal length and appends overhead bits, A packet of payioad bits
together with the overhead is denoted as a message.
A channel encoder 714 of the information processor 706 adds redundancy to the message
for forward error correction purposes.
Afterwards, the data is spread in frequency, for example binary data for the different
subchannels i is generated by a frequency spreader 716 of the information processor 706
In order to facilitate decoding, a synchronization signal is inserted by multiplying the
matrix of binary information by a concatenation of synchronization sequences. This
synchronization scheme insertion can be performed by a synchronization scheme inserter
718 of the information processor 706,
A time spreader 720 of the information processor 706 carries out a spreading in time
domain, this means adds further redundancy in order to gain more robustness against noise.
An output of the information processor 706 is binary data (for example the strea 0 of
subsequent stream values, also denoted as b(i, j)), where i indicates a subchannel and j
indicates the time slot or symbol number.
The differential encoder 708 carries out the host interference rejection and differential
encoding process on the stream 0 of subsequent stream values which is provided by the
time spreading block 720. The differential encoder 708 may also be denoted as host
interference rejection and differential encoding block 708. The differential encoder 708
may for example be similar or equal to the differential encoder 08 from Fig. 1, the
differential encoder 208 from Fig. 2 or the differential encoder 408 from Fig. 4. The
differential encoders 108, 208, 408 have been described before, therefore a repeated
description of the differential encoder 708 is omitted,
Th differential encoder 708 provides the watermark signal 02 in the time frequency
domain as a sequence of subsequent watermark coefficients wm(i, j ) (i, j N) for a
plurality of frequency bands. In other words the output of the differential encoder 708
consists of the watermark coefficients wm(i, j ) (i, j N). The modulator 710 is configured
to derive the watermark signal in a time domain based on the subsequent watermark
coefficients of the watermark signal 02 in the lime frequency domain The modulator 10
therefore provides the watermark signal 722 as a time domain watermark signal 722, which
is also denoted as wni(t). In other words the job of the remaining modulator 710 s to
convert the watermark coefficients wm(i, j), (i, j N) into the time signal wm(t).
The resulting time domain watermark signal 722 ( (t)) is the watermark which can be
added to the (audio) host signal a(t).
In the following, the watermark generator which is described in EP 10154964 shall be
explained in short as a comparison example to the watermark generator 700 according to
an embodiment of the present invention shown in Fig. 7.
Conventional Watermark Ge erato s Comparison E amp 1e
Fig. 8a shows a block schematic diagram of a watermark generator 800 as it is described in
EP 10154964. The watermark generator 800, like the watermark generator 700, may also
perform the signaling, channel encoding, frequency spreading, synchronization scheme
inserting and time spreading to the binary data at its input, to obtain a stream 804 of binary
data as a sequence of subsequent binary values b(i, j) (i, j e N) for a plurality of frequency
bands. The watermark generator 800 comprises a differential encoder 802 which then
performs a differential encoding on the stream 804 of binary data. An output of the
differential encoder 802 is:
d ,j ¾ f (i,j--l)- b(i ( 1 1)
The output of the differential encoder 802 is a stream 8 8 of subsequent differentially
encoded binary coefficients , j) (i, j N). A modulator 806 of the watermark
generator 800 transforms the resulting binary data b ( j ) (i, j e N) into a time signal and
performs amplitude scaling at the same time according to scaling factors (for example
y(i,j)) given by a psychoacoustica! model. One can regard the differentially encoded
binary coefficients bdiff (i,j (i. j N) as coefficients and the modulator 806 as a synthesis
filter bank that first scales the coefficients and then transforms them into time domain. A
resulting time signal wm(t) is the watermark which can be added to an audio host signal
a(t).
Fig. 8b shows the embedding principle of the system proposed i EP 0154964. Please
note th a the scaling factor ( ,j) for the coefficient amplitudes are already included in
wm(i ) . n fact wm(ij) bdiff (i,j). y(i,j).
The watermark m(i j) is chosen between wm -'(i,j) or wm (i,j) depending on
bdjff . In other words either (i, j ) or wm ,'' (i, ) s chosen as the watermark wni(i, j )
depending on (i, j .
It can be observed that both m (i,j) and m (i,j) have been constructed regardless of
the host audio signal (with the exception of the factor provided by the
psychoacoustical model, which has (obviously) analyzed the host signal and adjusted the
factor j ) in accordance with frequency masking effects and temporal masking effects).
In the following a short example for an application, in which a watermark generator
according to an embodiment of the present invention can be used, sha l be shown.
Example of an Application using the Watermark Generator 700
Fig. 9 shows a block schematic diagram of the watermark generator 700 from Fig. 7 in
conjunction with a psychoacoustical processing module 902 to provide a time domain
watermarked audio signal awm(t). In this example the watermark generator 700 is only
used as an example. The watermark generator therefore may be substituted with any other
watemmrk generator according to embodiments of the present invention. As can be seen
from Fig. 9 the watermark generator 700 receives as inputs discrete valued data 04 (in the
present example binary data 04), a host signal 1 (in the present example a time domain
audio host signal 1 8) and a current scaling factor g ( j). The scaling factor g ( , j ) is
provided by the psychoacoustical processing module 902 based on the time domain audio
host signal . The psychoacoustical processing module 902 provides scaling factors y(i,
j ) (i, j s ) for each stream value b(i, j ) (i, j N) which are internally generated in the
watermark generator 700. In other words, the psychoacoustical processing module 802
provides the current scaling factor y(i, j ) in each subchannel i (at center frequency f ) and
for each time slot j .
As mentioned before, the resulting signal of the watermark generator 700 is the time
domain watermark s gnal wn (l) This resulting time signal wm(t) is the watermark which
is added to the time domain audio host signal a(t). The watermarked host signal:
awm(t) —a(t) + wm(t) (12)
can be transmitted over a communication channel and constitutes a received signal y(t) at
the receiver.
In the following a method for generating a watermark signal will be explained.
Method for Generating a Watermark Signal According to Fig. 10
Fig. 10 shows a flow diagram of a method 1000 for providing a watermark signal as a
sequence of subsequent watermark coefficients based on discrete valued data.
The method 1000 comprises a step 002 of providing in dependence on information units
of the discrete valued data a first stream of subsequent values, such that the first stream
represents the discrete valued data.
The method 000 further comprises a step 1004 of applying a phase rotation to a current
stream value or to a current watermark symbol, the current watermark symbol
corresponding to a current stream value of the stream values representing the discrete
valued data, to obtain a current watermark coefficient of the watermark signal
The method 000 further comprises a step 1006 of deriving a phase of a spectral
coefficient of a watermarked signal which is a combination of a host signal and the
watermark signal
The method 1000 further comprises a step 008 of providing the watermarked signal such
that a phase angle of the phase rotation applied to the current stream value or to the current
watermark symbol is dependent on the phase of the previous spectral coefficient of the
watermarked s nal
In the following, a decoder for decoding a watermark signal generated by a watermark
generator according to an embodiment of the present invention will be described.
Watermark Decoder According to Fig. .
In a receiver, which comprises a watermark decoder, typically the inverse of the mentioned
operations for generating the watermark signal is carried out in reverse order to decode the
watermark. For the case that the differential encoder 208 from Fig. 2 is used n the
watermark generator to generate the watermark signal the differential decoding can be
performed by:
bi ) R b ) - ( - ) = (13)
fl ·f H I sfe (j) - )) (14)
where f ) = Zb" (j ) . ™ are normalized complex coefficients given by an analysis
filter bank in subchannel i (at center frequency ) representing a received signal y(t) and
the variable j is the temporal index. The resulting real valued soft bits bi (j)are the
estimates of b(i, j). The differential decoding works in that if the phase difference is 0 the
cosine is 1, whereas for a phase difference equal to p the cosine becomes -1. y(t)
represents the watermarked signal awm(t) which has been transmitted over a
communication channel. This differential decoding principle works fine for watermark
signals generated by the differential encoder 208 in which the phase rotation 112 is directly
applied to the current stream values b(i, j ) (i j e N) of the stream 10. n other words this
decoding principle works for differential encoders in which no stream value to watermark
symbol mapping 4 is applied. Therefore, a decoder, which is configured to decode the
watermark signal 102 generated by the differential encoder 208 can be similar to a
decoder, which is configured to decode a watermark signal generated by the differential
encoder 802 of the watermark generator stated in EP 10154964.
In contrast to this, the use of a M-point constellation with M > 2 implies a use of a different
decoder. Such an M-point constellation has been shown with the differential encoder 408
according to Fig, 4, which applies a stream value to watermark symbol mapping 4 to
each of the stream values b(i, j ) (i, j N) of the stream 0.
Fig. shows a block schematic diagram of a decoder 1 00 according to an embodiment
of the present invention, which is configured to decode such M-point constellation
watermark signals. The watermark decoder 100 for providing discrete valued data 02
comprises an information processor 1 04 and a diflereniial decoder 06, the information
processor 04 is configured to provide a stream 08 of complex valued spectral
coefficients b fi (j) (i, j N), the stream 1108 representing the watermarked signal 0 1.
The information processor 1104 may be configured to provide a stream 1108 for each
frequency subchannel i
The differential decoder 106 is configured to determine a phase angle difference 1 0
(also denoted as d,ff (j)) between a previous complex valued spectral coefficient 2
(also denoted as b "™(j -1)) and a current complex valued spectral coefficient 114 (also
denoted as h™"" (j)). The differential decoder 06 is further configured to map phase
angle differences within at least two different phase angle ranges to a first discrete value
6 of the discrete valued data 1102 and to map phase angle differences within at least
another two different phase angle ranges to a second discrete value 1 of the discrete
valued data 1102. The discrete valued data 1 02 may for example be binary data and the
first discrete value 6 may for example correspond to a logical 1 and the second discrete
value 1 may for example correspond to a logical — or 0.
n other words, the differential decoder 06 may be configured to choose in response to
the determined phase difference 0 falling into the phase angle ranges mapped to the
first discrete value 116, the first discrete value as a value for a current element of the
discrete valued data 02 and o choose in response to he determined phase difference
0 falling into the phase angle ranges mapped to the second discrete value 18, the
second discrete value as a value for the current element of the discrete valued data 1102.
The information processor 104 may be configured to provide he stream 0S of complex
spectral coefficients in a frequency time domain, such that each spectral coefficient
corresponds to one frequency subchannel i and one time slot j . The differential decoder
06 may be configured such that the previous complex spectral coefficient 11 2 and the
current complex spectral coefficient 4 correspond to adjacent time slots j , j - 1 and to
the same frequency subchannel i.
Fig. 2 shows how the differential decoder 06 may perform the mentioned phase angle
ranges mapping. Fig. 12 shows the special case for M = 4. This means two different phase
angle ranges are mapped to the first discrete value 16 of the discrete valued data 02
and another two different phase angle ranges are mapped to the second discrete value 1
of the discrete value data 1102. in Fig. 12 a phase angle is drawn in an anti clockwise
direction starting from a point 12 0 with the a phase angle of 0.
A first angle range 1202 which ranges from - /4 (or 7p/4) to /4 and a second phase angle
range 1204 which ranges from 3p /4 to 5p /4 are mapped to the first discrete value 1116 by
the differential decoder 06. A third phase angle range 206 which ranges from p/4 to 3p
/4 and a fourth phase angle range 1208 which ranges from 5p /4 to n /4 are mapped to the
second discrete value 8 of the discrete value data 1202 by the differential decoder
06. Comparing this diagram shown in Fig. 2 with the diagrams shown i Fig. 6a for the
ease in which M = 4 it can be seen that the mapping performed n the decoder matches the
mapping performed in the encoder. Widths of the phase angle ranges 1202, 1204, 1206,
1208 are equal for a l of the phase angle ranges 1202, 1204, 1206, 1208 and are 2p/M (in
the special cas which is shown in Fig. 12, where M=4 the width is /2). As can be seen
from of the combination of Fig. 6a and Fig. 12 a phase drift through the communication
channel smaller than p/4 may not lead to a bit error,
As can be seen from Fig. 12 the differential decoder 06 may be configured to map the
phase angle ranges to the discrete values such tha adjacent phase angle ranges are mapped
to two different discrete values of the discrete value data 102 .
From the explanations made above it becomes clear that the use of an M-point
constellation with M > 2 implies the use of a different decoder. One significant difference
to conventional (traditional) decoders is the mapping of the bits, in that a traditional system
typically codes l g ) bits with M symbols, whereas at least some of the proposed
systems always code 1 bit with M symbols.
n the following some aspects of the present invention will be summarized n short.
A watermark scheme employed in embodiments of the present invention comprises a
m ti-cham e differential BPSK method for embedding digital information in an audio
signal. Each of several subchannels i is related to a frequency (fj) of a time frequency
representation of the audio signal a(t). The information to be transmitted in one subchannel
i is contained in the phase difference of consecutive coefficients b(i, j ) (i, j e N) of a time
frequency representation.
Embodiments of the present invention have been presented in a more specific way using
the differential encoder 208 in Fig. 2 and have been presented in a more generalized way
using the differential encoder 408 in Fig, 4. The two-point solution shown in Fig, 2 may be
of greater importance for the current practical application. However, the more general
multipoint solution presented in rig. 4 might be of interest in future applications.
At least some embodiments of the present invention relate to digital audio watermarking,
i.e. some modification of an audio signal in order to hide digital data and the corresponding
decoder capable of receiving th s information while the perceived quality of the modified
audio signal remains indistinguishable (inaudible) to the one of the original.
Embodiments of the present invention implement differential encoding implicitly by
providing a current watermark coefficient based on a previous spectral coefficient of the
watermarked signal
Embodiments of the present invention create a method for generating an inaudible
watermark featuring differential encoding in the time frequency domain The watermark is
shaped optimally, or at least approximately optimally or signal-adapted, considering the
host audio signal to maximize the decoder performance. Moreover, the choice of the order
of the symbol constellation allows to tradeoff robustness against external noise sources (i.e.
better signai-to-noise ratio of the watermarked signal) versus better bit error rates
Embodiments of the present invention create a (partly) host interference rejecting
watermark embedding method which implicitly contains a differential encoding scheme,
mp ie ie rsta o n Alternatives
Although some aspects have been described in the context of an apparatus, it is clear that
these aspects also represent a description of the corresponding method, where a block or
device corresponds to a method step or a feature of a method step. Analogously, aspects
described in the context of a method step also represent a description of a corresponding
block or item or feature of a corresponding apparatus. Some or all of the method steps may
be executed by (or using) a hardware apparatus, like for example, a microprocessor, a
programmable computer or an electronic circuit. n some embodiments, some one or more
of the most important method steps may be executed by such an apparatus.
Depending on certain implementation requirements, embodiments of the invention can be
implemented in hardware or in software. The implementation can be performed using a
digital storage medium for example a floppy disk, a DVD, a Blue-Ray, a CD, a ROM a
PROM, an EPROM, an EEPROM or a FLASH memory, having electronically readable
control signals stored thereon, which cooperate (or are capable of cooperating) with a
programmable computer system such tha the respective method is performed. Therefore,
the digital storage medium may be computer readable.
Some embodiments according to the invention comprise a data carrier having
electronically readable control signals, which are capable of cooperating with a
programmable computer system, such that one of the methods described herein is
performed.
Generally, embodiments of the present invention can be implemented as a computer
program product with a program code, the program code being operative for performing
one of the methods when the computer program product runs on a computer. The program
code may for example be stored on a machine readable carrier,
Other embodiments comprise the computer program for performing one of th methods
described herein, stored on a machine readable carrier.
n other words, an embodiment of the inventive method is, therefore, a computer program
having a program code for performing one of the methods described herein, when the
computer program runs on a computer.
A further embodiment of the inventive methods , therefore, a data carrier (or a digital
storage medium or a computer-readable medium) comprising, recorded thereon, the
computer program for performing one of the methods described herein. The data carrier,
the digital storage medium or the recorded medium are typically tangible and/or nontransiiionary.
A further embodiment of the inventive method is, therefore, a data stream or a sequence of
signals representing the computer program for performing one of the methods described
herein. The data stream or the sequence of signals may for example be configured to be
transferred via a data commimication comiection, for example via the internet.
A further embodiment comprises a processing means, for example a computer, or a
programmable logic device, configured to or adapted to perform one of the methods
described herein.
A further embodiment comprises a computer having installed thereon the computer
program for performing one of the methods described herein.
A further embodiment according to the invention comprises an apparatus or a system
configured to transfer (for example, electronically or optically) a computer program for
performing one of the methods described herein to a receiver. The receiver may. for
example, be a computer, a mobile device, a memory device or the like. The apparatus or
system may, for example, comprise a file server for transferring the computer program to
the receiver .
n some embodiments, a programmable logic device (for example a field programmable
gate array) may be used to perform some or all of the functionalities of the methods
described herein. In some embodiments, a field programmable gate array may cooperate
with a microprocessor in order to perform one of the methods described herein. Generally,
the methods ar preferably performed by any hardware apparatus.
The above described embodiments are merely illustrative for the principles of the present
invention, t is understood that moditications and variations of the arrangements and the
details described herein will be apparent to others skilled in the art. It is the intent,
therefore, to be limited only by the scope of the impending patent claims and not by the
specific details presented by way of description and explanation of the embodiments
herein.
Naturally, the concept described herein may also be used for waiemiarking of video signals
or image signals.
O 2012/038344 PCT/EP2011/066118
Cl
1. A watermark generator for providing a watermark signal (102, w ) as a sequence
of subsequent watermark coefficients (wm(i, j ) (i, j N)) based on a stream ( 10)
of subsequent stream values (b(i, j)) representing discrete valued data (104), the
watermark generator comprising:
a differential encoder (108, 208, 408, 708) configured to provide the watermark
signal (102, wm), wherein the differential encoder (108, 208, 408, 708) is
configured to apply a phase rotation ( 12) to a current stream value (b(i, j)) of the
stream values (b(i, j)) representing the discrete valued data (104) or to a current
watermark symbol (c , j)), the current watermark symbol (x (i j)) corresponding
to a current stream value (b(i, j)) of the stream values (b(i, j)) representing the
discrete valued data (104) to obtain a current watermark coefficient (wm(i, j)) of
the watermark signal ( 2);
wherein the differential encoder ( 08, 208, 408, 708) is configured to derive a
phase ( , (i, of a previous spectral coefficient (awm(i, j - 1) of a
watermarked signal (awm) which is a combination of a host signal ( 18, a) and the
watermark signal (102, wm); and
wherein the differential encoder (108, 208, 408, 708) is configured to provide the
watermark signa (102) such that a phase angle (f ( , j)) of the phase rotation ( 1 2)
applied to the current stream value (b(i, j)) or the current watermark symbol ( k( ,
j)) is dependent on the phase ( q>(i, of the previous spectral coefficient
(awm(i, j - 1)) of the watermarked signal (awm),
The watermark generator according to claim 1,
wherein the watermark generator comprises an information processor ( 06);
wherein the information processor ( 6) is configured to provide the stream ( 10)
representing the discrete valued data (104) in a time frequency domain, such that
each stream value of the stream ( 0) is associated to a frequency subchannel (i)
and a time slot (j); and
wherein the differential encoder (108, 208, 408) is configured to obtain the current
watermark coefficient (wm(i j)) in the time frequency domain, such that a
frequency subchannel (i) associated to the current watermark coefficient (wm(i, j))
is identical to a frequency subchannel (i) associated to the current stream value
(b(i, j)) and such that a time slot (j) associated to the current watermark coefficient
(wm(i, j)) is identical to a time slot (j) associated to the current stream value
W )
3. The watermark generator according to claim 2,
wherein the differential encoder (208, 408) is configured to derive spectral
coefficients (awm(i, j)) of the watermarked signal (awrri) in a time frequency
domain, such that each spectral coefficient of the watermarked signal is associated
to a frequency subchannel (i) and a time slot (j), and
wherein the differential encoder (208, 408) is configured to determine the phase
rotation ( 112), such that a time slot (j~l) associated to the previous spectral
coefficient (awm(i, j - 1)) of the watermarked signal (awm), in dependence on
which the phase angle ( ( , j)) of the phase rotation ( 2) applied to the current
stream value (b( j)) or the current watermark symbol (xk(i, j)) is chosen, and the
time slot (j) associated to the current stream value (b(i, j)) are adj cent in time.
4. The watermark generator according to claim 3,
wherein the differential encoder (208, 408) is configured such that a frequency
subchannel (i) associated to the previous spectral coefficient (awm(i. j - 1)) of the
watermarked signal (awm) and the frequency subchannel (i) associated to the
current stream value (b(i, j)) are identical.
5. The watermark generator according to one of the claims 1 to 4,
wherein the differential encoder (208, 408) is configured to additionally scale (210)
the current stream value (b(i, j)) or the current watermark symbol (xk(i, j)) by a
current scaling factor (g( , j)) or by a cun-ent factor which is smaller than th current
scaling factor (v(i, j)); and
wherein the current scaling factor (g (ί , j)) is provided by a psychoacoustical
processing module (902) in dependence on a host signal ( 8, a) into which the
watermark signal (102, wm) is to be embedded, and such that the current scaling
factor (v(i, j)) describes a masking characteristic of the host signal ( , a).
O 2012/038344 PCT/EP2011/066118
6. The watermark generator according to c aim 5,
wherein the differential encoder (208, 408) is configured to scale the current stream
value (b(i, j)) or the current watermark symbol (c ( , j)) by the current scaling factor
(g ( , j)) to adjust an amplitude of the current watermark coefficient (wm(i, j)) such
that a watermark is inaudible in a watermarked signal (awm(t)) determined by a
combination of the host signal ( 8, a) and the watermark signal (102, wm).
7, The watermark generator according to one of the claims 1 to 6,
wherein the differential encoder (408) comprises a subconstellation selector (402)
configured to selectively provide a plurality of current watermark symbols (x (i, j )
to h j)) which constitute a subconstellation in dependence on the current
stream value (b(i, j)).
wherein the differential encoder (408) is configured to apply the phase rotation
( 1 12) to each of the current watermark symbols (xi(i, j ) to j)) of the
subconstellation corresponding to the current stream value (b(i, j)) or to a scaled
version thereof, to obtain a plurality of current candidate watermark coefficients
(wrn (i, j ) to . 2( j)); and
wherein the differential encoder (408) comprises a decider (404) configured to
choose one out of the current candidate watermark coefficients (wm ( j ) to
t as the current watermark coef fi cient (wrr (i j)).
8 The watermark generator according to claim 7,
wherein the decider (404) is configured to derive a plurality of current candidate
watermarked spectral coefficients based on combinations of a current spectral
coefficient (a(i, j)) of the host signal ( 18, a) with the plurality of candidate
watermark coefficients (wmi(i, j ) to w (h j)) to determine the current candidate
watermarked spectral coefficient with the highest power out of the plurality of
current candidate watermarked spectral coeffi cients to choose the current candidate
watermark coefficient corresponding to the current candidate watermarked spectral
coefficient having the highest power as the current watermark coefficient
(wm(i, j)).
O 2012/038344 PCT/EP2011/066118
9 The watermark generator according to one of the claims 7 to S,
wherein the suhconst lat on selector (402) is configured to provide the plurality of
current watermark symbols (xi(i, j ) to x )} as complex values such that
different current watermark symbols only differ in phase and such that phase
differences of different adjacent current watermark symbols associated with the
same current stream value are equal.
10. The watermark generator according o one of the claims 1 to 9,
further comprising a modulator ( 0) configured to derive the watermark signal in a
time domain based on the subsequent watermark coefficients.
A watermark decoder for providing discrete valued data ( 02) in dependence on a
watermarked signal ( 101), the watermark decoder comprising:
an information processor ( 1 04) to provide a stream ( 108) of complex valued
spectral coefficients, the stream ( 1108) representing the watermarked signal ( 101);
and
a differential decoder ( 106) configured to determine a phase angle difference
p s (j)) between a previous complex valued spectral coefficient ( 12, b " (j -1))
and a current complex valued spectral coefficient ( 4, b"" "(j)),
configured to map phase angle differences within at least two different phase angle
ranges (1202, 1204) to a first discrete value ( 1 16) of the discrete-valued data
( 102) and to map phase angle differences within at least another two different
phase angle ranges (1206, 1208) to a second discrete value ( 18) of the discrete
valued data ( 2).
12. The watermark decoder according to claim ,
wherein the information processor ( 1 04) is configured to provide the stream
( 1108) of complex spectral coefficients in a time-frequency-domain such that each
complex spectral coefficient is associated to one frequency subchannel (i) and one
time slot (j); and
O 2012/038344 PCT/EP2011/066118
wherein the differential decoder ( 106) is configured such that the previous
complex spectral coefficient ( 1 12, b ™(j ~l)) and the current complex spectral
coefficient ( 1 14 b " m (j)) are associated to adjacent time slots (j - 1, j ) and the
same frequency subchannel (i).
3. The watermark decoder according to one of the claims 1 or 2,
wherein the differential decoder ( 06) is configured to distinguish a least between
four different phase angle ranges (1202, 1204, 206, 1208), and
wherein the differential decoder ( 6) is configured to map adjacent phase angle
ranges to different discrete values ( 1 6, 1 18) of the discrete-valued data ( 1102).
14. A method for providing a watermark signal as a sequence of subsequent watermark
coefficients based on discrete-valued data, the method comprising:
Providing ( 1102), in dependence on information units of the discrete-valued data, a
stream of subsequent stream values such that the stream represents the discretevalued
data;
Applying a phase rotation (1004) to a current stream value of the stream values
representing the discrete valued data or to a current watermark symbol, the current
watennark symbol corresponding to a current stream value of the stream values
representing the discrete valued data, to obtain a current watermark coefficient of
the watennark signal;
Deriving ( 06) a phase of a previous spectral coefficient of a watermarked signal
which is a combination of a host signal and the watermark signal; and
Providing (1008) the watermark signal such that a phase angle of the phase rotation
applied to the current stream value or to the current watermark symbol is dependent
on the phase of the previous spectral coefficient of the watermarked signal.
. A method for providing discrete-valued data in dependence on a watermarked
signal the method comprising:
Providing a stream of complex -valued spectral coefficients, the stream representing
the watermarked signal;
O 2012/038344 PCT/EP2011/066118
Determining a phase angle difference between a previous complex valued spectral
coefficient and a current complex-valued spectral coefficient;
Mapping phase angle differences within at least two different phase angle ranges to
a first discrete value of the discrete-valued data and mapping phase angle
differences within at least another two different phase angle ranges to a second
discrete value of the discrete valued data.
16. A computer program for performing the method according to claim 4 or 5, when
the computer program runs on a computer.
1 . Method according to claim 14, wherein the host signal is an audio signal, an image
signal or a video signal, and the watermarked signal is an audio signal, an image
signal or a video signal.
. Method according to claim 15, wherein the host signal is an audio signal, an image
signal or a video signal, and the watermarked signal is an audio signal, an image
signal or a video signal .