EP2385521A1 - Steganography in digital signal encoding - Google Patents

Abstract

The method involves providing data information, particularly voice information, selecting steganographic information, generating a code word from a code book provided by the signal encoder on the basis of the code elements. The code word generated within the scope of a transmission standard associated with the code book. The data information is encoded into signal information containing the code word. The code word generated having an additional feature that is calculated on the basis of the code elements forming the code word, and represents the steganographic information.

Description

The invention is directed to a method for embedding a steganographic information in a signal information of a signal encoder.

In addition to analogue audio, video and video transmissions, digital transmission is becoming increasingly important. Among other things, this is due to the fact that digital signal information is processed more easily, i. can be copied or compressed. In particular, the compression of digital signal information means that information can be transmitted by means of signal transmission channels with limited data transmission rates with high information density.

In addition to the compression of signal information as a type of processing, the embedding of "invisible" - steganographic - information into signal information has recently gained acceptance. Such an embedding of additional information allows, for example, a marking of copyrights - if the signal information is, for example, a piece of music - or, generally speaking, a general indication of origin, ie a "digital watermark".

Even if such an embedding of steganographic information in music and / or video signals has already largely become established, the embedding of steganographic information in coded signal information, in particular if it is to be transmitted in "real time", continues to be difficult. This is due to the fact that certain encodings do not provide redundancy and thus no room for steganographic information or that the steganographic information is lost in decoding the encoded signal information.

Such an initial situation in which signal information, in this case primarily voice information, is transmitted and received over a channel and encoded and decoded in real time, and in which unrestricted transmission resources are available, can be found, for example, in mobile telephony. At best, the GSM network allows a maximum transmission rate of 13.0 kbit / s. An uncoded speech information, i. here a non-compressed voice information, would be difficult to understand because of the very low transmission rate on the receiver side. In order nevertheless to transmit intelligible speech information, for example from one mobile radio device to another, the so-called voice codecs have emerged as an effective means for compressed voice signal transmission. If additional information, i. Steganographic information to be embedded in such a signal information, so the peculiar to the coding peculiarities are taken into account.

In the field of mobile telecommunications, for example, in GSM (Global System for Mobile Communications) mobile radio networks or UMTS (Universal Mobile Telecommunications Standard) mobile radio networks, the voice information to be transmitted by means of the well-known CELP (Code-book Excited Linear Predictive Coding) or ACELP (Algebraic Code Excited Linear Prediction) or in the future the AMR (Adaptive Multi-Rate) coding coded. These speech codings are all based on a model of speech production in which the formation of the speech signal in an excitation stage and a filtering stage is generated in a first nutrition. A signal coder, such as a CELP coder, an ACELP coder or an AMR coder, generates a codebook entry, typically a vector of a so-called codebook, with the codebook entry code elements - ie usually the vector components - information regarding the (filter) excitation included. Filter coefficients, gain factors, etc. are encoded as time information by means of dedicated codebooks.

A codebook for excitation coding usually consists of a set of vectors, for example each with 10 components in the ACELP encoding according to the Enhanced Full Rate (EFR) standard, which encode the speech information to be transmitted for a certain length, for example 5 milliseconds. From the fixed codebook, which comprises a total of a large variety of vectors, wherein the vectors are constructed according to known criteria, usually a subset of the codebook, a sub-codebook is used, which is often sufficient to the ordinary speech information with to transfer a good quality.

In order to distinguish between the complete codebook specified in the coding, the sub-codebook used in practice is called a "practical codebook".

To quickly find a suitable codebook entry, the practical codebook is only searched heuristically, i. a complete search for a suitable codebook entry is not performed.

A method which considers the decomposition of a fixed codebook is described in the article " Watermarking Combined with CELP Speech Coding for Authentication "by Zhe-Ming Lu et al (in IEICE TRANS.INF. & SYST., Vol. E88-D, No. 2 February 2005 ) disclosed. In this case, first a codebook is divided into three sub-codebooks, from which in turn two codebooks are generated, which have different properties. Depending on which steganographic information is to be transmitted, is now a codebook entry from the for this particular sub-codebook selected and used to encode the language information to be transmitted. This speech information can be decoded on the receiver side, wherein the actual decoder can also recognize at the same time from which decomposition of the codebook the codebook entry originates. In order to provide a sufficiently good coding from one of the sub-codebooks, the publication also describes the known analysis-by-synthesis method. In this method, the selected codeword is evaluated, ie the quality of the coding is checked. This is essentially achieved by decoding, ie synthesizing, the coding after speech information has been coded, and comparing the result of the decoding, which in turn represents speech information, with the original speech information. Thus, a synthesis is performed in advance on the transmitter side - encoder side - which, after any transmission, is also performed on the receiver side decoder side. With such an analysis-by-synthesis loop, it is possible to find a codeword, that is to say generally a vector from a codebook, which firstly has the desired property, ie originating from the correspondingly decomposed subcodebook, and thereby the speech information coded with sufficient quality.

It turns out, however, that the fixed distribution of a practical codebook - which is already a subset of a higher-level codebook - reduces the number of usable codewords per sub-codebook in several sub-codebooks such that an audible reduction of the speech quality is not ruled out ,

The invention is therefore based on the object to provide a solution which makes it possible steganographic information in a signal information of a signal encoder such to embed that a reduction of voice quality is largely avoided.

In a method for embedding a steganographic information in a signal information of a signal coder, the object is achieved by providing data information, in particular speech information, selecting a steganographic information from a set of steganographic information, generating a codeword from a provided codebook by means of a signal Coder on the basis of codeword-forming code elements such that, using the generated codeword in the context of a codebook-associatable transmission standard, the data information is coded into a codeword-containing and / or codeword referring signal information; and that the generated codeword has an additional property computable on the basis of the codewords forming the codeword, the additional property representing the steganographic information.

Such a method for embedding steganographic information in signal information of a signal coder in which a codeword is generated from a provided codebook by means of the signal coder on the basis of codeword-forming code elements makes it possible to provide a codeword having a calculable property on the one hand, ie represents a steganographic information, and on the other hand at the same time provides signal information which for a data information, in particular a voice information, encoded. By not performing a decomposition of the codebook from the outset, but instead creating a codebook entry based on the code elements forming the codeword, codewords not in the practical codebook and / or the decompositions of the practical codebook may be considered were present. This significantly expands the number of code elements that can be accessed so that either decomposition in more sub-codebooks compared to the prior art or with an equal number of sub-codebooks provides improved speech quality over the prior art can.

Preferably, in one embodiment of the invention, an evaluation of the generated codeword is carried out in the context of a transmission standard associable with the provided codebook by decoding the codeword and then comparing the decoded data information with the original data information.

This has the advantage that the quality of the generated codeword with respect to the coding-decoding fidelity, i. the loss of (speech) quality due to encoding and decoding, can be evaluated.

In another embodiment of the method according to the invention, the code word is generated from the provided codebook by means of the signal coder on the basis of the code elements forming the codeword, taking into account the evaluation.

The consideration of the evaluation in the generation of the codeword from the provided codebook makes it possible to generate codewords which have a high speech quality with regard to the information to be coded.

In one embodiment of the invention, the use of a based on the GSM (Global System for Mobile Communications) - and / or UMTS (Universal Mobile Telecommunications Standard) transmission standards based coder and codebook is provided.

This has the advantage that the method for embedding a steganographic information can also be used in mobile radio networks.

For further embodiment of the invention, the generation of a codeword based on the CELP, the ACELP (Algebraic Code Exited Linear Prediction) - and / or the AMR encoding is provided.

The high prevalence of CELP or ACELP coding allows the use of the method according to the invention in many fields of technology, in particular mobile telecommunications. This also applies to the future by analogy for the AMR coding.

Furthermore, in one embodiment, the calculation of the property of the code word is performed as a result of applying at least one operation to at least one of the code elements forming the code word.

It is thus possible, on the basis of the code elements forming the codeword, to determine a property of the codeword which represents the steganographic information using at least one, preferably mathematical, operation.

Furthermore, in an embodiment of the method according to the invention, the provision of the code word is carried out such that the code word implicitly satisfies the property.

This has the advantage that a codeword can be generated in such a way that it already fulfills the property which represents the steganographic information during its generation.

Furthermore, in one embodiment of the method according to the invention, the selection of the steganographic information is carried out in such a way that the steganographic information is used for signal improvement, in particular for voice transmission, such as artificial bandwidth expansion and / or noise reduction.

It has been found that the additional transferable steganographic information can be used, for example, to describe a property of the data information actually to be transmitted, so that the steganographic information can be used for signal enhancement. This means that - if the code word coding for the data information is generated by the method according to the invention - the marginal loss in the transmission quality can not only be compensated by the steganographic additional information, but can even be overcompensated.

Furthermore, in one embodiment, the selection of steganographic information is performed such that the steganographic information is used as a digital watermark.

When using the steganographic information as a digital watermark not only the originality and the origin of data information can be marked, but also copyright to data information can be introduced using the steganographic information in the form of a digital watermark. The data information is qualitatively hardly impaired in the context of the method according to the invention.

In a further embodiment of the invention, a transmission of the codeword-containing or referring to the codeword signal information is performed to a receiver.

Advantageously, by transmitting to a receiver, it is possible to transmit the data information and the steganographic information in the form of signal information over a spatial distance.

Furthermore, in one refinement of the method according to the invention, provision of data information by means of decoding of the codeword is carried out in the context of a transmission standard associable with the provided codebook.

By decoding the codeword, the data information included in the signal information as well as the steganographic information can be recovered and used.

Furthermore, in an embodiment of the method according to the present invention, provision of the steganographic information on the receiving side is carried out by means of calculating the additional property of the code word on the basis of the code elements forming the code word.

Depending on the configuration of the receiver, it is possible to calculate the steganographic information contained in the signal information. This possibility can optionally be used, ie systems which are not able to recalculate the steganographic information will only extract the data information from the signal information without getting to the steganographic information.

Finally, in one refinement of the method, the method is also carried out in a mobile radio device.

The method is particularly suitable for signal information transmission, i. Voice or other information, by means of mobile devices, which are operable in a mobile network.

The above-mentioned and claimed components to be used described in the embodiments are subject in their size, shape, design, choice of materials and technical concepts no special conditions of exception, so that the well-known in the field of application selection criteria can apply without restriction.

Further details, features and advantages of the subject matter of the invention will become apparent from the subclaims and from the following description of the accompanying drawings, in which - by way of example - a preferred embodiment of the invention is shown.

Figur 1

zeigt einen Codierer 100 gemäß einem Ausführungsbeispiel der Erfindung.

Figur 2

zeigt ein Codier/Decodiersystem 200 gemäß einem Ausführungsbeispiel der Erfindung.

Figur 3

zeigt einen Codierer 300 gemäß einem Ausführungsbeispiel der Erfindung.

Figur 4

zeigt einen Codierer 400 gemäß einem Ausführungsbeispiel der Erfindung.

Figur 5

zeigt ein Codebuch 500 gemäß einem Ausführungsbeispiel der Erfindung.

For the described coding methods according to embodiments of the invention, corresponding coders are provided as well as corresponding methods for decoding and corresponding decoders.

FIG. 1

shows an encoder 100 according to an embodiment of the invention.

FIG. 2

FIG. 10 shows an encoding / decoding system 200 according to an embodiment of the invention.

FIG. 3

shows an encoder 300 according to an embodiment of the invention.

FIG. 4

shows an encoder 400 according to an embodiment of the invention.

FIG. 5

shows a codebook 500 according to an embodiment of the invention.

Fig.1 shows an encoder 100 according to an embodiment of the invention.

The coder 100 is supplied with a signal 101 to be coded, for example a speech signal 100. In addition, data 100 to be embedded is supplied to the encoder 100. The coder generates a coded signal 103 from the signal 101 to be coded, into which the data 102 to be embedded is embedded, that is, from which a corresponding decoder can determine the data 102 to be embedded.

The coded signal 103 is transmitted, for example, to a receiver, for example by means of a computer network or by means of a radio network.

In the embodiment described below, it is assumed that the encoder 100 is used in a GSM (Global System for Mobile Communications) mobile radio network. In other embodiments of the invention, the encoder can also be used within the framework of a mobile radio network in accordance with UMTS (Universal Mobile Telecommunications Standard), CDMA2000 (CDMA: Code Division Multiple Access) or according to FOMA (Freedom of Mobile Access).

In the embodiment described below, it is assumed that the signal 101 to be encoded is a speech signal outputted from the encoder 100 according to an ACELP (Algebraic Code Excited Linear Prediction) speech compression method. for example, according to an "Enhanced Full Rate" ACCELP voice compression method as used in a GSM mobile radio network.

The encoder 100 uses a fixed (in other words stochastic) codebook, which is decomposed into N sub-codebooks, for information embedding, that is, for embedding the data 102 to be embedded into the coded signal 103 in an embodiment for coding the so-called residual signal. For the actual residual signal coding, the corresponding sub-codebook is used depending on the information to be embedded in accordance with a binning scheme (binning scheme).

Since the codebook is not searched exhaustively (but heuristically) in a CELP speech coder, the sub codebooks may have a similar scope to the searched part of the fixed codebook, and the quality of the CELP coding suffers little from information embedding. Furthermore, the information embedding can be performed with low algorithmic complexity.

In the embodiment described below, the encoder 100 uses a codebook defined as follows:

mit

• c₀, c₅ ∈ {0, 5, 10, 15, 20, 25, 30, 35}, c₁, c₆ ∈ {1, 6, 11, 16, 21, 26, 31, 36}
• c₂, c₇ ∈ (2, 7, 12, 17, 22, 27, 32, 37) , c₃, c₈ ∈ {3, 8, 13, 18, 23, 28, 33, 38}
• und
• c₄, c₉ ∈ {4, 9, 14, 19, 24, 29, 34, 39}

The codebook C used in this embodiment is the codebook of the GSM EFR (Enhanced Full Rate) codec and is given by the vectors c of the ACELP pulse positions (unsigned in this embodiment) for each 5ms subframe: $$\mathrm{c}\mathrm{=}\left\{\underset{\mathrm{}}{\mathrm{c}}\mathrm{=}\left({\mathrm{c}}_{\mathrm{0}}\mathrm{...}{\mathrm{c}}_{\mathrm{9}}\right)\right\}$$

With

• c ₀ , c ₅ ∈ {0, 5, 10, 15, 20, 25, 30, 35}, c ₁ , c ₆ ∈ {1, 6, 11, 16, 21, 26, 31, 36}
• c ₂ , c ₇ ∈ (2, 7, 12, 17, 22, 27, 32, 37), c ₃ , c ₈ ∈ {3, 8, 13, 18, 23, 28, 33, 38}
• and
• c ₄ , c ₉ ∈ {4, 9, 14, 19, 24, 29, 34, 39}

A codeword c from the codebook (ie, the set of all possible codewords) C is thus a ten component vector, with each component describing a position of a pulse within a subframe. The codebook C in this embodiment has a circumference of 2 ^ (10 * log_2 (8)) = 2 ^ 30 codewords.

In another embodiment, the components of the vectors c , as provided in accordance with EFR, are signed. The use of signed components allows for improved information embedding. However, in one embodiment, EFR dispenses with the use of signed components for complexity reasons.

The codebook C is in one embodiment decomposed into two sub-codebooks C (1) and C (2) such that one codeword of the coded signal 102 can embed one bit of the data to be embedded 101 per codeword and accordingly one bit of the data to be embedded 101 per Subframe is transmitted, which corresponds to a sub-frame duration of 5ms a data rate of 200bit / s.

(d.h. Summe der Komponenten ist gerade) und alle Codeworte aus C(2) die Bedingung

(d.h. Summe der Komponenten ist ungerade) wobei trunc die Abrundungsoperation, das heißt das Abrunden auf die nächstkleinere ganze Zahl, bezeichnet.The codewords of the sub-codebooks differ in that the sum of the components c _{i of} a codeword is even from the one sub-codebook and odd from the other sub-codebook. For example, all code words from C (1) satisfy the condition

(ie sum of components is even) and all code words from C (2) the condition

(ie sum of components is odd) where trunc denotes the rounding operation, that is, rounding down to the next smallest integer.

If a first message (in this example consisting of one bit, for example the bit value 0) is to be transmitted, then a codeword from C (1) is used for coding (the current signal values of the signal to be coded 101), and a second message ( in this example consisting of one bit, for example bit value 1), a codeword from C (1) is used for coding. A receiver or decoder may determine whether the first message or the second message has been embedded based on the membership of a received codeword to C (1) or C (2).

in Binärdarstellung eine gerade Anzahl von Einsen aufweist und zu C(2) andernfalls.In another embodiment, the division of C is performed in even and odd parity of the sum of the components of the codewords. For example, a codeword belongs to C (1) if

has an even number of ones in binary representation and to C (2) otherwise.

anschaulich wird hierbei die Unterscheidung basierend auf Geradzahligkeit bzw. Ungeradzahligkeit der Summe der Komponenten mit geradem bzw. ungeraden Index durchgeführt.In one embodiment, four bits are embedded per subframe, thereby achieving a data rate of 400bit / s. This is done by dividing the codebook C into four sub-codebooks C (1) to C (4), wherein the codewords of the sub-codebooks satisfy, for example, the following conditions:

In this case, the distinction is made on the basis of the even-numbered or odd-numberedness of the sum of the components with even or odd index.

bzw.

Analogous to the above alternative, even when dividing the codebook C into four sub-codebooks, it may be performed based on the parity of a binary representation of the sum of even-numbered components, that is, based on the parity of

respectively.

In the above conditions for the codebooks C (1) and C (2) or C (1) to C (4), alternatively, instead of a component c _i itself, the expression trunc (c _i / 5) may also be used Pulse position clearly indicated within a so-called track. Alternatively, the respective graycoded version GRAY (ci) or GRAY (trunc (c _i / 5)) may be used as intended for channel coding in EFR. Taking into account the actual transmission of the code words over a GSM mobile radio channel, this possibility for codebook decomposition proves to be particularly advantageous. This ensures that particularly few channel bits have an influence on the embedded data, which reduces the bit error rate of the transmitted embedded data in case of disturbed transmission.

erfüllen, wobei A_i eine Indexmenge, b_j die Komponenten des jeweiligen Codeworts bezeichnen. Die Summation wird dabei modulo 2 durchgeführt, so dass gefordert wird, dass die Summe modulo 2 mehrerer Codewortbits b_j gleich dem einzubettenden Nachrichtenbit m_i ist.Generally, in one embodiment, the codebook C may be decomposed such that the codewords of the sub-codebook used for coding when a message bit mi is to be transmitted are the condition ${\mathrm{m}}_{\mathrm{i}}\mathrm{=}\underset{\mathrm{j}\mathrm{\in }{\mathrm{A}}_{\mathrm{i}}}{\mathrm{\oplus }}{\mathrm{b}}_{\mathrm{j}}$

where A _i denotes an index set, b _j the components of the respective codeword. The summation is carried out modulo 2, so that it is required that the sum modulo 2 of several codeword bits b _{j is} equal to the message bit m _i to be embedded.

To reconstruct an embedded message in a codeword to be received or decoded, it is only necessary for the decoder to determine to which sub-codebook the codeword belongs. If the codewords are transmitted undisturbed to the decoder, the embedded information can also be reconstructed without error.

The above-described procedure for embedding information can also be used in other coders, for example in all CELP speech coders but also in other signal coders such as video coders, image coders, etc.

The transmission of page information (embedded information) by means of steganography can also improve the signal A receiver without knowledge of the embedded information can use the (speech) signal in which the information was embedded as usual, that is, as in the case of no embedding of information with only a small amount of information Decode losses. If the recipient knows the embedded information, the page information can be used to improve the signal. A corresponding embodiment will be described below with reference to Fig.2 described.

Fig.2 shows an encoder / decoder system 200 according to another embodiment of the invention.

The encoding / decoding system 200 includes an encoder 201 as described with reference to FIG Fig.1 described on. Accordingly, a signal 202 to be encoded and data 203 to be embedded are supplied to the encoder 201. The data to be embedded are used for signal enhancement and are accordingly generated in a signal-enhancement-encoding signal 202 by a signal analyzer 204 to which the signal 202 to be coded is supplied.

Analogous to Fig.1 The encoder 201 outputs a coded signal 205 into which the data 203 to be embedded is embedded. The coded signal 205 can now be transmitted to a receiver, for example, as described above, by means of a mobile radio communication network.

If the receiver has a "conventional" decoder 206, that is, a decoder that can not detect the embedded data from the encoded signal 205, the decoder 206 decodes only the encoded signal 205 to a decoded signal 207 which (except for transmission errors and coding / decoding losses) corresponds to the signal 202 to be coded.

If the receiver has an "extended" decoder 208, that is, a decoder capable of detecting the embedded data from the encoded signal 205, then the embedded data is extracted and the extracted data 209 is used by a signal enhancement unit 210 which decodes a signal and (compared to the decoded signal 207) improved signal 211 generated.

As a signal enhancement, e.g. artificial bandwidth expansion, or noise reduction can be used. The coefficients of a postfilter determined by the transmitter can also be transmitted by steganography.

In particular, the application of artificial bandwidth expansion is advantageous because the telephone network is historically limited to an acoustic bandwidth of 3.1kHz (300Hz-3.4kHz), a transmission of broadband (50Hz-7kHz) but only with enormous effort on the part of the network operators and the terminal manufacturers would be accomplished. Changes in the (mobile) transmission network, however, are not required to implement the embodiments described above. Corresponding (powerful) bandwidth expansion algorithms are disclosed, for example, in the publication of Peter Jax, Bernd Geiser, Stefan Schandl, Hervé Taddei, and Peter Vary, "An Embedded Scalable Wideband Codec Based on the GSM EFR Codec", in Proceedings of ICASSP, Toulouse, May 2006 described. Furthermore, in the publication of Peter Jax and Peter Vary, "Bandwidth Extension of Speech Signals: A Catalyst for the Introduction of Wideband Speech Coding?", IEEE Communications Magazine, vol. 44, no. 5, May 2006 the introduction of broadband voice over the Detour of the bandwidth extension (possibly with the support of digital watermarks) mentioned.

In the following, a further possibility for a decomposition of the above-defined codebook C (EFR ACELP codebook) will be described.

1. 1) Zunächst wird die erste Pulsposition i₀ ∈ {0,...,39} heuristisch ermittelt und bleibt während der gesamten Suche fest. Der zu i₀ gehörige "Track" ist beispielsweise x = 4 oder x = 9. Für die x-te Komponente c_x des entsprechenden Codeworts gilt c_x = i₀.
2. 2) Die Position des zweiten Pulses i₁ ∈ {0, ... , 39} wird ebenfalls heuristisch ermittelt, wobei für jede der vier Iterationen des unten stehenden Algorithmus (Schritt 3) ein anderer Wert angenommen wird. Beispielsweise gehöre für die erste Iteration zur gewählten Position i₁ der Track y=3 oder y=8, d.h. C_y = i₁.
3. 3) Für die verbleibenden acht Tracks werden für jede von vier Iterationen die Pulse sukzessive in Paaren zu je zwei Tracks durch erschöpfende Suche optimiert. In jeder der vier Iterationen werden die Track-Paare durch Permutation neu zusammengestellt, wobei c_x und c_y nicht wiederverwendet werden

First, for a better understanding, the search strategy of the EFR codec is explained briefly:

1. 1) First, the first pulse position i ₀ ∈ {0, ..., 39} is determined heuristically and remains fixed during the entire search. The "track" belonging to i ₀ is, for example, x = 4 or x = 9. For the x-th component c _{x of} the corresponding code word, c _x = i ₀ .
2. 2) The position of the second pulse i ₁ ∈ {0, ..., 39} is also determined heuristically, whereby a different value is assumed for each of the four iterations of the algorithm below (step 3). For example, for the first iteration to the selected position i _1, the track y = 3 or y = 8, ie C _y = i ₁ .
3. 3) For the remaining eight tracks, for each of four iterations, the pulses are successively optimized in pairs of two tracks by exhaustive search. In each of the four iterations, the track pairs are reassembled by permutation, with c _x and c _y not reused

• for(i = 0, 5, 10,..., 35) // Iteration über alle zulässigen Elemente für c₀
• for(j = 1, 6, 11,..., 36) // Iteration über alle zulässigen Elemente für c₆
• Teste Puls-Paar (c₀=i, c₆=j) entsprechend CELP-Kriterium auf Optimalität

For example, for the track pair c ₀ / c _6, the optimization is performed according to the following pseudocode:

• for (i = 0, 5, 10, ..., 35) // iteration over all valid elements for c ₀
• for (j = 1, 6, 11, ..., 36) // iteration over all valid elements for c ₆
• Test pulse pair (c ₀ = i, c ₆ = j) according to CELP criterion for optimality

In this search strategy according to EFR, a total of 1024 combinations (4 iterations * 4 track pairs * 8 pulse positions * 8 pulse positions = 1024 combinations) are examined and from this the optimal pulse pairs are selected.

For embedding information, in the following example of a single (watermark) bit b, into a codeword c = (c ₀ ,..., C ₉ ), according to an embodiment of the invention, the above algorithm is modified as described below.

• c6_offset = 5 * ((c1 + b + 1) mod 2)
• for(i = 0, 5, 10,..., 35) // Iteration über alle zulässigen Elemente für c₀
• for(j = 1,11,21,31) // Iteration über alle zulässigen Elemente für c₆ (gegenüber oben halbe Anzahl)
• Teste Puls-Paar (c₀=i, c₆=j+c6_offset) entsprechend CELP-Kriterium auf Optimalität

If a pulse for one track, for example, c _1, was found in a search, may, at the choice of the (identically configured) track c ₆ by a watermark bit b in the pulse position pair _c1 and c are embedded ₆ by c ₆ is selected depending on the bit b. For example, the pairwise search is modified as follows:

• c6_offset = 5 * ((c1 + b + 1) mod 2)
• for (i = 0, 5, 10, ..., 35) // iteration over all valid elements for c ₀
• for (j = 1,11,21,31) // iteration over all permissible elements for c ₆ (opposite half number above)
• Test pulse pair (c ₀ = i, c ₆ = j + c6_offset) according to CELP criterion on optimality

ermittelt werden.The embedded bit b may be in the receiver or decoder through the operation $$\mathrm{b}\mathrm{=}\left({\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="imgf0005.png"\; state="\{\{state\}\}">c</figure-callout>}}_{\mathrm{1}}\mathrm{+}{\mathrm{c}}_{\mathrm{6}}\right)\mathrm{<figure-callout\; id="2"\; label="mod"\; filenames="imgf0005.png"\; state="\{\{state\}\}">mod</figure-callout>}\mathrm{2}$$

be determined.

Instead of c6_offset = 5 * ((c1 + b + 1) mod 2), other combinations of already determined pulse positions and embedded bis can also be used. In the above example, the search space for the pulse position c _{6 was divided} into two equal parts (odd / even values). Further divisions (for example, first / second half of the value) are also possible, with the equation for c6_offset being adapted accordingly.

gesunken.The number of pulse position combinations examined is due to the bit embedded in this way $$\mathrm{4}\mathrm{iterations}\mathrm{*}\left(\mathrm{3}\mathrm{pairs}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{+}\mathrm{1}\mathrm{Pair}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{4}\mathrm{Pos}\mathrm{,}\right)\mathrm{=}\mathrm{896}\mathrm{combinations}$$

declined.

To embed several bits in a codeword c , the search space for c ₆ can be halved once again, or an identical method can be used for a second pair of pulses. It is favorable to couple precisely such pulses as part of the information embedding by c6_offset (or the bit b) which lie in a track. Otherwise, due to the sign encoding of the EFR, no definite assignment of the pulses can be made in the receiver. By appropriate additional expenditure in the transmitter this restriction can be canceled. In the receiver, c1 and c6 are not different to distinguish. The data extraction via b = (c1 + c6) mod 2 does not cause any problems, but it is difficult to compute b = (c1 + c5) mod 2, for example (because of the sign) "accidentally" b = (c6 + c5 ) mod 2 could be calculated. The transmission-side "additional effort" consists in considering the sign encoding in the optimization loops and, as it were, anticipating it per optimization step.

• for(i2 = 0,5,10,...,35) // Iteration über die zulässigen Werte von c0
• for(i3 = 1,6,11,...,36) // Iteration über die zulässigen Werte von c6
• for(i4 = 2,7,12,...,37) // Iteration über die zulässigen Werte von c7
• Suche optimales Tripel (c0=i2, c6=i3, c7=i4)

Due to the reduced number of combinations examined (896 instead of 1024 in the above example), a lower coding quality results through watermark embedding. This can be compensated by an extended search, ie an extension of the search space. For this purpose, the tracks are no longer searched in pairs but in groups of 3 or 4 (or even more) tracks together. The common search (without watermark embedding) for 3 tracks (eg c ₀ , c ₆ and c ₇ ) is realized for example as follows:

• for (i2 = 0,5,10, ..., 35) // iteration over the permissible values of c0
• for (i3 = 1,6,11, ..., 36) // iteration over the permissible values of c6
• for (i4 = 2,7,12, ..., 37) // iteration over the allowable values of c7
• Search optimal triple (c0 = i2, c6 = i3, c7 = i4)

untersucht werden, was gegenüber den 1024 Kombinationen gemäß EFR einen erheblichen Mehraufwand bedeutet.For each triple, this means that 8 * 8 * 8 = 512 combinations are searched. If the entire search for the 8 variable pulse positions (according to steps 1 and 2, two pulse positions are fixed) so that 2 triples and 1 pair are optimized together, it follows that $$\mathrm{4}\mathrm{iterations}\mathrm{*}\left(2\mathrm{triples}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}*\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{+}\mathrm{1}\mathrm{Pair}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\right)\mathrm{=}4352\mathrm{combinations}$$

which, compared to the 1024 combinations according to EFR, means a considerable additional effort.

d.h. exakt die Anzahl der untersuchten Kombinationen im Standard-EFR-Codec. Die Wasserzeichendatenrate liegt in diesem Fall bei (2+3)bit/5ms = 1kbit/s.However, if, for example, in the course of the optimization of the first triple 3 watermark bits, embedded in the course of the optimization of the second triple 2 watermark bits as described above, while for the pulse position pair no embedding is made, then the number of combinations to be examined $$\mathrm{4}\mathrm{iterations}\mathrm{*}\left(\mathrm{1}\mathrm{triples}\mathrm{*}\mathrm{4}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{4}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{4}\mathrm{Pos}\mathrm{,}\mathrm{+}\mathrm{1}\mathrm{triples}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{4}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{4}\mathrm{Pos}\mathrm{+}\mathrm{1}\mathrm{Pair}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\mathrm{*}\mathrm{8th}\mathrm{Pos}\mathrm{,}\right)\mathrm{=}\mathrm{1024}\mathrm{combinations}\mathrm{.}$$

ie exactly the number of tested combinations in the standard EFR codec. The watermark data rate in this case is (2 + 3) bit / 5ms = 1kbit / s.

Finally, another way to expand the search space is to increase the iteration number, i. the investigation of further track permutations.

The idea underlying an embodiment can be seen in that the information embedding is known to the signal encoder, which is achieved by a common data embedding and signal coding, that is, for example, the watermark embedding is integrated into the encoder. This can be done within an analysis-by-synthesis loop ("closed-loop") as described in Figure 3 is shown.

Figure 3 shows an encoder 300 according to another embodiment of the invention.

The encoder is supplied with a signal to be encoded and data 302 to be embedded.

An encoded signal 303 is generated from the signal 301 to be encoded by means of a loop comprising a codebook 304, a synthesizer 305 and a comparator 306. In this case, a possible coding of the signal 301 to be coded is generated by the codebook 304 and it is checked by means of the synthesizer 305 and the comparator 306 how well it reflects the signal 301 to be coded and, if appropriate, modified based on the output of the comparator 306.

The data to be embedded 302 are embedded in the encoded signal 303 during the encoding process, for example according to one of the procedures described above.

For example, based on the data 302 to be embedded, a sub-codebook of the codebook 304 is selected as shown in FIG Figure 4 is shown.

Figure 4 shows an encoder 400 according to another embodiment of the invention.

Analogous to the in Figure 3 The encoder 300 to be embedded is supplied with data 402 to be embedded in the encoder and a signal 401 to be encoded, and generates a coded signal 403 in which the data 402 to be embedded is embedded. In addition to a synthesis device 405 and a comparator 406 analogous to the encoder 300 in FIG Figure 3 Thus, the encoder 400 has a plurality of sub-codebooks 404, a codebook divided into a plurality of sub-codebooks 404. Based on the data to be embedded, the sub-codebooks are selected in coding the signal 401 to be coded. For example, a data word of the signal to be coded 401 is input Codeword assigned from a first sub-codebook, when a first page information from the data to be embedded 402, for example a bit with the value 0, to be embedded and it is assigned a codeword from a second sub-codebook, if a second page information from the einzubettenden Data 402, for example a bit with the value 1, should be embedded.

The division of a codebook into several sub-codebooks is in Figure 5 illustrated.

Figure 5 shows a codebook 500 according to another embodiment of the invention.

The codebook 500 is labeled C. For efficiency reasons, if the codeword extent of the codebook 500 is very large, the codebook 500 is only partially searched during coding, i. Codewords for coding are selected only from a codebook subset 501 denoted C '(practical codebook).

In the above embodiments, the codebook 500 for data embedding is decomposed into codebooks as explained above, for example, four subcodebooks 502 designated C (1) to C (4).

Since a sub-codebook 502 has a smaller codeword extent than the codebook subset 501, and thus the quality of the signal coding (depending on the number of subcodebooks 502) would decrease from using the entire codebook subset 501, in one embodiment Code extent of the sub-codebooks 502 extended, so that a total of an extended codebook subset 503 is used for coding. The algorithmic complexity increases only slightly, the quality of the coding does not decrease and in special cases even an increase in quality can be achieved.

In one embodiment, an algebraic codebook is used. In contrast to an ordinary codebook in tabular form, an algebraic codebook exists only in the sense of an algebraic design rule. This means that the individual codebook entries (codewords) are generated during the signal coding by a codeword generator. The "binning scheme" for information embedding, that is, the decomposition of the codebook in sub-codebooks and selection of the sub-codebook used for coding depending on the information to be embedded, in an encoder with algebraic codebook now no longer consists only in the division of the codebook in a number of sub-codebooks, but moreover in the modification of the codeword generator in that only codewords associated with the sub-codebook C (i) selected by the currently-to-be-embedded message i are output.

Claims (21)

Method for embedding steganographic information in signal information of a signal coder (100), characterized by Providing data information, in particular voice information, as the signal to be coded (101), Selecting steganographic information as data to be embedded (102), the steganographic information being selected from a set of steganographic information, Generating a codeword from a provided codebook (500) by means of the signal coder (100) on the basis of codewords forming codewords such that Using the generated codeword in the context of a transmission standard which can be associated with the codebook, the data information is coded into a codeword-containing and / or codeword referring signal information as coded signal (103); and that The generated codeword has an additional property which can be calculated on the basis of the code elements forming the codeword, the additional property representing the steganographic information. A method according to claim 1, characterized by generating the codeword such that the codebook (500) is divided into a number of sub-codebooks and that only one sub-selected by the steganographic information currently to be embedded by the signal coder (100) Codebook associated codeword is issued. A method according to claim 1 or 2, characterized by evaluating the generated codeword in a transmission standard associable with the provided codebook by decoding the codeword and then comparing the decoded data information with the original data information and further characterized by optionally generating the codeword from the provided codebook by means of the signal coder (100) based on the codeword code elements taking into account the evaluation. Method according to one of the preceding claims, characterized by the use of a coder (100) and codebook based on the GSM and / or the UMTS transmission standard. Method according to one of the preceding claims, characterized by generating the codeword from a provided algebraic codebook (500) in the sense of an algebraic design rule and / or on the basis of the CELP, the ACELP or the AMR coding. Method according to one of the preceding claims, characterized by the calculation of the property of the code word as a result of applying at least one operation to at least one of the code elements forming the code word. Method according to one of the preceding claims, characterized by the provision of the codeword such that the codeword implicitly satisfies the property. A method according to claim 7, characterized by generating the code word such that it already fulfills the property representing the steganographic information during its generation. Method according to one of the preceding claims, characterized by the selection of the steganographic information such that the steganographic information is used on the receiving side for improving the signal, in particular in voice transmission, such as an artificial bandwidth extension and / or noise reduction. Method according to one of the preceding claims, characterized by the selection of the steganographic information such that the steganographic information is used as a digital watermark. Method according to one of the preceding claims, characterized by transmitting the signal information containing the code word or referring to the code word to a receiver. Method according to Claim 11, characterized by providing data on the receiving side by means of decoding of the codeword in the context of a transmission standard associable with the provided codebook. A method according to claim 11 or 12, characterized by providing the steganographic information on the receiving side by means of calculating the additional property of the codeword on the basis of the code elements forming the codeword. Method according to one of the preceding claims, characterized by carrying out the method in a mobile radio device. Method according to one of the preceding claims, wherein the information to be embedded is supplied to the signal encoder (100). Method according to one of the preceding claims, wherein the signal encoder (100) determines a first code element and a second code element of the code word, wherein the second code element is determined in dependence on the first code element and the information to be embedded. A signal encoder (100) for embedding a steganographic information in a signal information of the signal encoder (100), characterized by a device for receiving data information, in particular voice information, as the signal to be coded (101), means for receiving steganographic information as data to be embedded (102), the steganographic information being selected from a set of steganographic information, a device for generating a codeword from a provided codebook (500) by means of the signal coder (100) based on the code word forming code elements such that Using the generated codeword in the context of a transmission standard which can be associated with the codebook, the data information is coded into a codeword-containing and / or codeword referring signal information as coded signal (103); and that The generated codeword has an additional property which can be calculated on the basis of the code elements forming the codeword, the additional property representing the steganographic information. A method of receiving steganographic information embedded by a signal encoder (100) in signal information by generating a codeword from a provided codebook (500) by means of the signal coder (100) on the basis of codeword code elements that Using the generated codeword in the context of a transmission standard which can be associated with the codebook, data data as coded signal (101) is encoded as coded signal (103) in a signal information containing the codeword and / or referring to the codeword; and that The generated codeword has an additional property which can be calculated on the basis of the code elements forming the codeword, the additional property representing the steganographic information; wherein the reception-side provision of the steganographic information comprises calculating the additional property of the code word based on the code elements forming the code word. The method of claim 18, wherein the receiving side, a first code element and a second code element of the codeword are determined, wherein the steganographic information is determined in dependence on the first code element and the second code element. The method of claim 18 or 19, further comprising determining the embedded data from a coded signal (205) and using the thus determined data (209) for signal enhancement, in particular for artificial bandwidth expansion and / or noise reduction, or for determining the coefficients of a postfilter determined on the transmitter side , A decoder (208) adapted to perform the method of any one of claims 18 to 20.

Images