EP2119073A1 - Method for generating mutually orthogonal signals having a controlled spectrum - Google Patents

Abstract

The invention relates to a method for generating mutually orthogonal signals having a controlled spectrum, comprising the generation of a plurality of mututally orthogonal, controlled-power discrete spectra s(i) having dimension Q, i designating spectrum number. The aforementioned spectra represent time signals in the spectral range and have a modulus μ that is constant in a spectral line designation set G and zero everywhere else. The inventive method comprises the following steps: determining (20) at least part of a complex Hadamard matrix H of order dR = w in the case of spectra of real signals and dR = 2.w in the case of spectra of complex signals; determining (22) the extension P of matrix H from G and dimension Q; and obtaining (22) controlled-power spectra s(i) = μ.P(H[.][i]), wherein H[.][i] designates the i-th column of matrix H. In addition, a plurality of mutually orthogonal time signals s(i) is also generated from said discrete complex signals.

Description

 METHOD FOR GENERATING MUTUALLY ORTHOGONAL SIGNALS WHERE THE SPECTRUM IS CONTROLLED

The present invention relates to a method of generating mutually orthogonal signals whose spectrum is controlled.

It has many and varied fields of application, including the tattooing of signal files such as audio signals.

Spectral spread audio tattooing uses time signals with a broad spectrum - the extended spectrum. The tattoo uses either one or more signals stored in a dictionary for symbol modulation. When several signals are used, it is preferred to use signals whose inter-correlation product is zero, as this facilitates correlation detection. Cross-correlation of signals is a particular form of the dot product. It can thus be said that to look for signals that are not correlated with one another is to choose a family of mutually orthogonal signals.

In tattooing, we need signals whose spectrum is under control, that is to say that correspond to a particular template. For example, an AAC (Advanced Audio Coding) coded signal at 24 kbit / s per channel occupies a band in the order of

7 kHz, hence the interest of controlling the spectrum of the mark by limiting its bandwidth to 7 kHz. In addition, the tattoo must be as discreet as possible, it will be modulated and formatted taking into account the properties of psychoacoustics. To guarantee a precise shaping, below the masking curve guaranteeing the inaudibility of the mark, the signal to be modulated must have a perfectly white spectrum.

The generation of real time signals of mutually orthogonal given length is generally done, either by an orthogonalization technique of a family of time signals, generally random, or by use of the rows or columns of the actual Hadamard matrices. Each of these techniques of the prior art has the disadvantage of producing signals whose power spectrum is difficult to control. By way of example, for the orthogonalization technique, variations of dynamics of 60 dB and more are observed on the power spectrum, the spectrum being chopped and not very regular. In addition, the technique of actual Hadamard matrices can produce only signals of length 2 or multiples of 4.

In transmission systems with code division multiple access (Code Division Multiple Access) mobiles, for example in CDMA systems such as IS95 (International Standard 95), the signals used to separate users are orthogonal and real Hadamard sequences whose spectrum depends on the number of the sequence. This way of proceeding is described in Pierre LESCUYER's work entitled "UMTS, origins, architecture, the norm", Dunod, 2nd edition, 2002. As an example, the first line of the Hadamard matrix contains only 1. It follows that this step must be followed by a step spreading spectrum itself, or scrambling, before the modulation (pages 116 to 119 of the aforementioned work), which increases the complexity such a system.

WO-A-0077962 discloses a method for generating orthogonal complex spectra in low numbers, typically seven, by discretizing the phase of each complex sample. This method has the disadvantage of providing only a limited number of sequences.

To improve the situation, it is sought to generate real signals and mutually orthogonal complex spectrums of arbitrary length and whose power spectra can be controllable in the measurement of the length of the signals generated and the number of orthogonal signals desired.

Thus, the present invention provides a method of generating a plurality of discrete spectra s (i) of Q dimension, mutually orthogonal and with controlled power, where i denotes the spectrum number, these spectra representing time signals in the spectral domain and being of constant μ module in a set of spectral line designation G and zero everywhere else, the method being remarkable in that it consists of:

determining at least a portion of a complex Hadamard matrix H of order dR = w in the case of real signal spectra and dR = 2.w in the case of complex signal spectra;

determining the extension P of the matrix H from G and the dimension Q; and

to obtain the power-controlled spectra s (i) = μ.P (H [.] [i]), where H [.] [i] designates the ith column of the matrix H. Thus, the invention allows the generation mutually orthogonal discrete spectra of any length, in desired number and at controlled power. In addition, it is not necessary to construct the entire complex Hadamard matrix.

In a particular embodiment, the step of determining at least a portion of the complex Hadamard matrix consists in obtaining a column of a rotation matrix calculated from predetermined rotation and permutation keys applied to a matrix. of reference Hadamard.

This system of use of keys makes it possible to generate families of varied spectra and little correlated. In a particular embodiment, the method further comprises a step of decomposition of the order d _R of the reference Hadamard matrix into a product of factors and a sub-step of calculating the least common multiple of the set. of these factors for the determination of the reference Hadamard matrix. Thus, choosing the least common multiple means that the phases of the complex spectrum generated will be as far apart as possible. This increases the robustness to noise.

In a particular embodiment, the method further comprises a step of determining a signal

s (i) [k] =, or : • i denotes the number of the signal,

• Q denotes a set of coordinates,

• X is the number of spatial dimensions,

• T denotes the number of temporal dimensions and • F _(qJ denotes the Fourier matrix of order q _α , so as to generate a plurality of mutually orthogonal time signals s (i).

The present invention also proposes a method for generating a family of time signals, which is remarkable in that it consists in combining mutually orthogonal time signal families generated by a method as briefly described above and whose spectral supports are disjoint.

The use of families of complex spectra with disjoint power spectra makes it possible to quantify the signal in the transformed domain by using progressive signal frequency modeling by orthogonal complex spectra. This way of proceeding gives directly to each step optimal scale factors for the complex spectra, without requiring reoptimization of the preceding factors, contrary to the prior art, which uses a Gram-Schmidt method for this reoptimization. The present invention also provides the use of mutually orthogonal complex time signals or spectra generated by a method as briefly described above for spectral spreading in spread spectrum transmission systems.

Indeed, by their construction, these signals or spectra are spread spectrum in the transmitted band and power of value 1, unlike the Hadamard sequences used in the current American system IS95 code division multiple access. In audio coders, they can be used as quantization dictionaries in predictive coders. In this case, the quantization is performed by two fast algorithms: on the one hand, by fast Fourier transform for the passage in the frequency domain and, on the other hand, for the scalar product involved in the quantization. The present invention also provides the use of mutually orthogonal complex time signals or complex spectra generated by a method as briefly described above for audio tattooing and its detection. In audio tattooing, these spectra can be directly used for tattooing in the frequency domain, their ideally white power spectrum in the transmitted band (s) allowing their precise shaping by psychoacoustic weighting, which is an important property to ensure the character inaudible tattoo. The spectra of the prior art do not allow this precise formatting, since their power spectrum has a significant variation in dynamics. The complex spectra of the invention also have the advantage of being easily modulated as such, in defined frequency bands, in code access multiplex transmission systems. As for the time domain signals, in audio tattooing, they can be directly used for spread spectrum tattooing, where a symbol of K _b bits is represented by K _b orthogonal signals. Moreover, the same advantages as those mentioned above are found for complex spectra. The present invention also provides the use of mutually orthogonal complex spectrums with controlled power spectrum generated by a method as briefly described above for encoding or representing audio signals, the audio signals being quantized by means of a dictionary or family of dictionaries with real or complex values.

Indeed, in the audio coders, these spectra can be used as quantization dictionaries for the signals coming from a discrete Fourier transform. In this case, the quantization is for example performed by a fast algorithm whose structure is derived from the way in which the dictionary is constructed, in a particular embodiment, by Kronecker products of small-size basic matrices. The present invention also provides the use of mutually orthogonal complex time signals or spectrums with controlled power spectrum generated by a method as briefly described above for the optimization of metrology excitation data. For the calculation of the impulse response of acoustic rooms in metrology, the fact that the generated sequences have an ideally flat spectrum in the band is likely to improve the accuracy of the detection compared to the current use of the pseudo-noise sequences generated by shift registers, whose correlation function has a parasitic term with respect to its ideal value, which is a Dirac.

Correlatively, the invention proposes a device for generating a plurality of discrete spectra S (i) of dimension Q that are mutually orthogonal and with controlled power, i denoting the number of the spectrum, these spectra representing time signals in the spectral domain and being of constant μ module in a set of spectral line designation G and zero anywhere else, the device being remarkable in that it comprises:

a module for determining at least a portion of a complex Hadamard matrix H of order dR = w in the case of real signal spectra and dR = 2.w in the case of complex signal spectra; a module for determining the extension P of the matrix H from G and the dimension Q; and

- a module to obtain the power-controlled spectra s (i) = μ.P (H [.] [i]), where H [.] [i] denotes the ith column of the matrix H.

The present invention further provides a computer program product that can be loaded into a programmable apparatus, characterized in that it includes instruction sequences for implementing a method as briefly described above, when this program is loaded and executed by the programmable device.

The particular features and advantages of the method of generating a family of time signals, the various uses of time signals or complex spectra, the device for generating a plurality of discrete spectra and the computer program product being similar to those of the method of generating a plurality of discrete spectra, they are not repeated here.

Other aspects and advantages of the invention will appear on reading the following detailed description of particular embodiments, given by way of non-limiting examples. The description refers to the accompanying drawings, in which:

FIG. 1 is a graph representing the power spectrum of the signals generated, in a particular embodiment of the invention;

FIG. 2 is a flowchart illustrating the main steps of a time signal generation method according to the present invention, in a particular embodiment;

FIGS. 3 to 6 are flowcharts detailing various operations performed to obtain a column of a matrix of rotations of a complex Hadamard matrix according to the present invention, in a particular embodiment;

FIG. 7 is a flowchart illustrating the process according to the present invention in all its generality;

FIG. 8 is a flowchart illustrating in more detail the so-called "band calculation" method implemented in the flowchart of FIG. 7; FIG. 9 is a flowchart illustrating in more detail the so-called "band preparation" step implemented in the flowchart of FIG. 8;

FIG. 10 is a flowchart illustrating in greater detail the spectral calculation step implemented in the flowchart of FIG. 8;

FIG. 11 is a flowchart illustrating in more detail the so-called "exponential" step implemented in the flowchart of FIG. 10;

FIG. 12 is a flowchart illustrating in greater detail the step of calculating the extension of FIG. 9 in the case of complex spectra;

Fig. 13 is a flowchart illustrating in more detail the step of extending Fig. 10 in the case of complex spectra; Fig. 14 is a flowchart illustrating in more detail the step of preparing the extension in the case of complex spectra; Fig. 15 is a flowchart illustrating in more detail the step of extending Fig. 10 in the case of real signals;

FIGS. 16 to 19 illustrate examples of application of the present invention, in particular embodiments; and - Figure 20 schematically shows a device adapted to implement the present invention, in a particular embodiment.

The invention applies to the case of real or complex discrete signals. Below are some notations and definitions that will be used in the rest of the description.

By convention, the index variables start at zero. Thus, a vector v of dimension 3 will have three components noted: v [θ], v [i], v [2]. Similarly, for the matrices, we note m [l] [c] the component of the line I and the column c (starting at zero) of the matrix m.

The number of signals generated is noted Ns. The signals to be generated will be noted s (i), where 0≤i <Ns, where i is the variable that designates the signal number.

Each signal s (i) can be seen indifferently, either as a signal of length Ls, or as an application of [0; Ls-1] c N (N being the set of natural numbers), or as a vector of dimension ls. This last interpretation, that is to say the vectorial interpretation, is preferred here.

From the vectorial point of view, each signal s (i) is a vector of E ^Ls (the Cartesian product of the vector space E at the power Ls) where E is either the body of the reals 9? (in the case of a real-valued signal), that is the body of the complexes C (in the case of a signal with complex values). The values of the signal s (i) are denoted s (i) [n], where 0≤n <Ls, where n is the variable which designates the time sample.

In the following, we will use Fourier matrices of any order (or dimension) which will be denoted F _(n ). The Fourier matrix of order n is the square matrix defined by F _(n) <≡ C ^nxn and F _(n) [l] [c] = e "(1.1)

This matrix has the following properties: F _(n) [l] [c] | = 1 and t -

F _(π) -F _(π) = n1 _(n) (1.2) where F _(n) denotes the conjugate matrix of F _(n) , F _(n) denotes the transposed matrix of the matrix F _(n) and _(n) denotes the identity matrix of order n. We

deduce F ₍ - _n ¹ , = - ÏÇ ₎ -

F and F are respectively the Discrete Fourier Transform (DFT) matrix and the inverse Discrete Fourier transform matrix. In the present invention, the matrix F will be used in two different contexts: to transform a complex spectrum with a controlled power spectrum into a time signal, and

as a real or basic complex Hadamard matrix to generate any order Hadamard matrices by Kronecker products. The problem we are trying to solve is to find a family

S of Ns signals such that the scalar product of two signals s (i) and s (j), denoted (s (i) | s (j)), is zero if i is different from j, which means that the signals are mutually orthogonal according to the scalar product under consideration.

In other words, using the mathematical notations introduced previously, it is necessary to find S = {s (i)} such that (s (i) | s (j)) = 0 <=> i ≠ j.

We will consider the usual scalar products:

- in the real case (E = 9): (a | b) = 'a.b

- in the complex case (E = C): (a | b) = 'ïï.b

It will also be necessary to consider the discrete complex spectra s (i) <≡ C ^Ls defined from the time signals s (i) by: s (i) - F _Ls ^{1 •} s (i) (1.3) In other words, the complex spectrum is obtained by TFD of the real signal. The spectra are said to be discrete since they come from the DFT of a discrete signal, that is to say, sampled in time at a given frequency F _e .

The power spectrum of the time signal is given as a function of the complex spectrum by:

P. (i) M - | s (i) [k] | ² = (s (i) [k] | s (i) [k]) = l (ïjjkj ^• s (i) [k]

In the following, we will say either indifferently that the time signal and the complex spectrum are controlled power.

Since the matrices F _(Ls) and F _(Ls) are orthogonal (equation (1.2)), the time signal is given as a function of the complex spectrum by: s (i) = F _(LB) ^• s (i) (1.4) c that is, inverse discrete Fourier transform of the complex spectrum.

For the signal s (i) to be real (E = 9i), we show that a necessary and sufficient condition is that the following equalities are satisfied: s (i) [0] = s (i) [O] and Vk e [i-Ls-iJs (i) [k] = s (i) [Ls-k] (1.5)

In other words, for the inverse Fourier transform of a complex spectrum to be real, the complex spectrum s (i) must verify the previous symmetry property.

The complex Hadamard matrices are the square matrices H of any order denoted d (that is to say H e C ^dxd ) having the following properties: ^ι HH = dl _(d) (orthogonality property) and | H [l] [c] | = 1. Note that Fourier matrices are complex Hadamard matrices. The complex Hadamard matrices also have the following properties: - if H is a complex Hadamard matrix, if P, Q are permutation matrices, if C, D are diagonal matrices whose non-null elements of the diagonal have a module of 1, then PCHD Q is also a complex Hadamard matrix; - If G and H are two complex Hadamard matrices of respective orders g and h, then G ® H is a complex gham Hadamard matrix. The operation <8> is the product of Kronecker or tensor product.

Remember that: (G H) [I] [C] = (G [ _g ] [c _g ]) (H [ _{h h} ] [c _h ]) (1.6) where I = 1 _g. H + 1 _h c c = _g · h + _h c.

The basic principle of the solution proposed by the present invention is now described.

For the basic discrete solution, it is desired that the family S of the real desired signals satisfies the following properties:

(s (i) | s (i)) = 1

(s (i) | s (j)) = 0 «i ≠ j _{(1 7)} ke [S _max -S _mιn] u [Ls - -LS S _max - S _mιn]" | s (i) [k] | Μ = k * [S _max -S _mιn] u [Ls - -LS S _max - S _mιn] "s (i) [k] = 0

These properties are illustrated by the graph of FIG. 1 which represents the value of the power spectrum of the signals generated.

These properties express the fact that the time signals have a unitary power over a period, that the complex spectrum of these signals is of constant modulus, μ, between a low frequency index S _m i _n and a high frequency index S _max and no one everywhere else.

The low frequency F _min is linked to the index S _min and to the sampling frequency F _e by: s ¹ ^ mIn - F min - p

where Ls is the number of samples of the DFT defined by equation (1.3).

Phases are the only degrees of freedom of the system. In addition, the signals form a free family in the space of the signals of length Ls, they are orthogonal to each other, or, in other words, they are not correlated with each other. It may be noted that the particular case S _m i _n = 0 and S _max = Ls / 2, where there is no zero sample, is taken into account in equation (1.7) because of the periodicity of the complex spectrum s (i) (period Ls). The spectrum of signals Frequencies are then perfectly white over the entire frequency band and the correlation function of the orthogonal temporal signals, derived by inverse Fourier transform, is equal to a Dirac distribution, an interesting property in many applications. The time signals will be real when the property of symmetry with complex conjugate (equation (1.5)) will be verified.

By noting that time signals are given by inverse Fourier transform of complex spectra (equation (1.4)), we show that:

(s (i) | s (j)> = (F _(Ls) .s (i) | F _(Ls) .s (j)) ^t I (r) ^t F ^ ;. F _(Ls) . s (j) = Ls. ('((ï) .s (j)) = Ls. <s (i) | s (j)>.

The problem of the orthogonality of the signals in the domain of complex spectra s (i) is then reduced.

If the spectral lines outside the interval set [S _min- S _max ] u [Ls-S _max -Ls-S _min ] are zero, it is sufficient to ensure the orthogonality for the signal values with discrete spectrum s (i) for ke [S _min- S _max ] u [Ls-S _max -Ls-S _min ] Let w = 1 + S _max - S _{min be} the number corresponding to half the number of μ module lines.

This operation, which restricts the study to nonzero values, is called restriction. We write R: C ^Ls → C ^2w . Its inverse operation is called extension and is denoted P: C ^{2 w} → C ^Ls .

In the real case, it suffices to restrict for ke [S _mn --- S _max ]. The remaining values, indexed ke [Ls-S _max --- Ls-S _mn ], are deduced from the previous ones by conjugation following the property s (i) [k] = s (i) [Ls-k] of the signals real, recalled above. We then have: R: C ^Ls → C ^w and P: C ^w → C ^Ls . Let r (i) be the restriction of s (i). _R d include the size of the restriction r (i), that is to say r (i) ^dR e C. In the case of real signals, we have dR = w and in the case of complex signals, we have dR = 2.w. We have af (i) = R (s (i)) and we search for a family R = {f (i)} of Ns vectors r (i), such that s (i) = F _(Ls) .P (r ( i)),

that is, S = F _(Ls) .P (R). The satisfaction of the properties required for s (i) leads to writing the following properties for r (i):

J | r (i) [p] | - μ Vi Vp [(f (i) | f (j)) = O o i ≠ j

These properties or constraints are close to those of the vectors of complex Hadamard matrices. The solution to this problem is thus obtained by choosing for the family R a free family resulting from a complex Hadamard matrix H of order dR multiplied by μ.

In the simple case where r (i) corresponds to the ith column of the matrix H, we write r (i) = μ.H [.] [I], we have: s (i) = F _{(Ls )} .P (μ.H [.] [I]) = μ.F ₍ Lβ) .P (H [.] [I])

(s (i) | s (j)> = Ls. ['ê (i) .s (j)] = Ls. [' p (P (i)). P (P (j

(s (i) | S (J)) = ^ ('HTm ™.) = - ((' HΉW

R R

So :

<s _(i> is _{(i) M,} ^w ,, _(Wiω) <*: tr ²

The dimensionality analysis shows that the following relation must be respected: Ns≤dR, and, in the real case, d _R ≤ | Ls and in the complex case, dR≤Ls. These are the constraints of the system.

We will write synthetically that S is a free family of μ. F. P (H), where F is the Fourier matrix, P is the extension deduced from S _max and S _m i _n and H is the complex Hadamard matrix of order d _R deduced from S _max and S _m i _n . It has been explained above how to construct a family S of time signals s (i) having the properties of equation (1.7):

It is enough to find a matrix of complex Hadamard H then to take:

s (i) = μ.F _(Ls) .P (H [.] [i]) with μ = ¹ where P is the deduced extension of S _max and S _mιn and nature of E (actual case or complex cases), and where F (Ls) is the Fourier matrix of order Ls, which, in other words , corresponds to the so-called "inverse discrete Fourier transform in the complex domain", which is often denoted TFD _(L ¹ _S) . We often have one of the following two definitions:

TFD ₍ ^ (X) = - ^ _{.F (Ls)} .x or TFD _(L ¹ _s) (x) = F _(Ls) .x.

This discrete Fourier transformation will advantageously be achieved by a Fast Fourier Transform, an operation well known to those skilled in the art.

This method of constructing s (i), the basic object of the present invention, is shown diagrammatically in FIG.

The module 20 calculates the ith column of a matrix of rotations of a complex Hadamard matrix H _cle whose generation depends on a key.

Rotations are expressed as whole numbers, that is:

H _cle [.] [I] _e Z ^dR .

The value θ allows, by the formula z _θ (n) = 0 ^2π ë ¹ , to go from the entire rotation to the corresponding complex number of module 1. A key is used to obtain distinct signal families. In an application of the present invention tattooing signal files, this allows for separate tattoos that can be superimposed without interacting.

The key is an integer which makes it possible to generate two permutations of Ns elements and twice Ns integer rotations of order θ, that is to say a number from 1 to [NS! .Θ ^NS ] ² . In practice, since these numbers are generally very large, the keys are used to initialize a generator making it possible to extract the necessary information. The introduction of the key allows to introduce a secret in the process of generating the sequences and thus to restrict the possibility of their use to the owners of the key only.

The module 22 transforms a vector of rotations of Z ^dR into a complex vector with a discrete spectrum of C ^Ls .

The transformation depends on the nature of the signal but does not depend on i.

In the case of real signals:

S _mn <k <S _max

Ls - S _max <k <Ls - S _mιn otherwise

In the case of complex signals:

S _mn <k <S _max = * s (i) [k] = μ.z _θ (H _cle [k - S _mn ] [i])

Ls - S _max ≤ k ≤ Ls - S _mn => s (i) [k] = μ.z _θ (H _dβ [k + 2.S _max -M-Ls-SJi]) otherwise = ^> s (i ) [k] = 0

In the case where one limits oneself to the generation of mutually orthogonal complex spectra, the procedure is finished at this stage.

The complex spectrum s (i) can be used as such, which has been shown in Figure 2 by an arrow output algorithm. Otherwise, the generation of mutually orthogonal time signals is performed by the module 24 which transforms a complex vector s (i) in C ^Ls and controlled spectrum into a controlled discrete spectrum time vector of E ^Ls .

The module 24 is an application module of a reverse discrete Fourier transform of order Ls. This operation is well known to those skilled in the art. It does not depend on i.

The operations carried out by the calculation module of H _cle [.] [I] are described below.

It has been seen above that this module calculates a column of the matrix H _cle . This has the advantage of avoiding the construction of the complete matrix and thus simplifying and accelerating calculations. Formally, the matrix of the rotations is expressed thus: ~ θ

CIE ^ ^~ / ^ - ^'n ~ 0' _of), CIE = ^0R ^ CNIE hG PCIe hG H _ref. PcIe col -RcIe col, WHERE:

2πv- 1

- Pcie hg and Pcie coi are the permutation matrices corresponding to the permutations σ _c ι _e hg θt σ _c ι _e cor; - R _c ia hg and CNIE coi are diagonal matrices of rotations (rows and columns) corresponding to the rotations R _c ia hg [x] [x] = z _θ (p _c ia hg (x)) and RciecoiMM = z _θ (Pcie coi (x)) - In practice, any rotations can be used and not only the rotations of order θ, that is to say that the matrices R _c ι _g and Rcie coi can also be defined as the diagonal matrices R _{cle hg} [x] [x] = _e ^{2π / lp} ^ ^(x) and R _cle [x] [x] = _e ^{2π / lp} - ^(x) . In practice, the application of the rotations is optional;

- H _r ef is the reference complex Hadamard matrix canonically constructed by the process as a function of d _R ;

- θ is the smallest order of the root of the unit which makes it possible to express the rotations by integers.

The calculation becomes:

The expression of the auxiliary functions σ _c hg, σ _c iecoi, Pcie hg and p _c iecoi is trivial. The permutations can be performed for example by scrambling algorithms well known in cryptography and the rotations can be initialized by a random function for introducing a secret. In the context of the description of the process, these calculated and known auxiliary functions are assumed.

The number dR, which represents the order of H _re f, the reference complex Hadamard matrix, is written as a product of numbers. We denote DF (dR)

= {f,} a decomposition of dR into a product of factors f, (thus, d _R =) arbitrarily ordered according to an index i varying from 1 to Nf, where Nf denotes the number of factors of the decomposition of dR, that is to say Nf = | DF (d _R ) | . In practice, one can take for DF the decomposition into prime factors.

For example, d _R - 100 => DF (d _R ) - {2; 2; 5; 5} and Nf - 4.

The complex Hadamard matrix H _ref of reference constructed by the method is defined by:

H ^ret . = RV _n il RV RV _f i, I --- --- RV RV _f m, I = i <ι < _{N N} R _n ((1.9)

The basic matrices F _(f) , i = 1, ..., Nf can be matrices of

Fourier as defined above. More generally, we can take complex Hadamard matrices of size fi. The generation of H _re f by a Kronecker product from base matrices can be exploited in the case where the scalar product of all complex spectra by a vector x is required, ie (s (i) | x ^ i = 0, ..., Ns-1, to carry out the computations with minimal complexity, resulting in a dRlog2dR complexity algorithm having a structure close to that of the Hadamard transform According to the definition of H _re f by Equation (1.9), the order of the root of the unit is the smallest common multiple of the set of factors of d _R , hence θ = ppcm (DF (d _R )) = ppcm ({ For example, θ (100) = 10.

If the prime factor decomposition of dR is taken as DF, this procedure has the advantage that θ is the smallest possible, which means that the phases of the complex spectrum are as far apart as possible. Incidentally, this increases the robustness to noise.

The computation of the matrices of integers H _ref [l] [c] uses the fact that a number n such that 0≤n <dR decomposes uniquely in the numerical system deduced from DF (d _R ), namely:

n = Σ MV] IM = Σ ⁿ '- ^{b | = n} i ^{+ f} i- ( ⁿ 2 ^{+ f} 2 - (n ₃ + ^{• • •} ))

"l≤i≤Nf ^ 1 <j <ι J" l≤i≤Nf

This is in fact a generalization of equation (1.6) to the case of the decomposition of d _R into a product of factors. Thus, having defined, i.e., decomposed the line number I = Σl, .b, and the column number c = Σc ,. b, we obtain, after the calculations

"l≤i≤Nf" l≤i≤Nf following:

H _1Bf Pl [C] θ I, .c

^H - ^{[I] [} C ^] = ii≤≤Iii≤≤NNff- - -ciππ.v - I ψ ¹ I ² * ^ the value sought: θ ~

The factor - being integer, H _ref is an integer matrix, which we

was actually looking to get.

Figure 3 shows the organization of the process of calculating the ith column of the rotation matrix H _de [.] [I].

Step 300 for the preparation comprises: a) calculating _R d of the decomposition product of factors fi, denoted DF (d _R) = {fi}; b) calculate the least common multiple of the set DF (dR), denoted θ = ppcm (DF (d _R )); c) calculating with the key the auxiliary functions σ _c ié, i ig, σ _c ié, i, P P c and

Step 302 consists in performing the permutation σ _c ie, cor (i) and then to calculate directly _ref H [.] [Σ _{collar key} (i)]. To this last matrix is added, at step 304, the contribution of the rotations according to equation (1.8).

A basic example of workflow a) and b) is illustrated in the flowchart of Figure 4. During an initialization step 400, a variable n is initialized to the value dR, a variable D to the value 2, a variable i to the value 0 and a variable θ to the value 1.

Then we perform a test 402 consisting in checking if n is congruent to zero modulo D, that is, if n is a multiple of D.

If the test 402 is negative, a test 404 consisting in verifying if n is 1 is carried out.

If the test 404 is negative, the value of the variable D (step 406) is incremented by one unit and the test 402 is returned. If the test 404 is positive, the variable Nf, which designates the number of factors, is assigned. from the decomposition of dR, the value of i (step 408) and the algorithm ends.

If the test 402 is positive, a test 410 is performed consisting in checking whether θ is congruent to zero modulo D.

If the test 410 is negative, the variable θ is assigned the value of the variable D.θ (step 412) and step 414 is passed. If the test 410 is positive, it goes directly to step 414.

Step 414 consists in assigning the variable n the value of the variable n / D, incrementing the value of the variable i by one, and assigning the variable fi the value of the variable D. Then, we return to test 402. Returning to FIG. 3, the operation 302 for calculating the column

Θ ^Λ

H _ref [] [σ _{key col} (i)] consists in calculating the H _ref [l] [c] = Σ hr I c where c = σ _c ié, coi (i). i≤i≤Nf l TJ ₁ J

A basic example of implementation of this step is illustrated by the flowchart of Figure 5.

I _j and q are calculated progressively. Thus, during an initialization step 500, a variable I is initialized to the value 0 and a variable c to the value of the auxiliary function σ _c ié, _∞ ι (i) -

Then one checks during a test 502 if I = d _R. If the 502 test is positive, the algorithm ends. Otherwise, a variable j is initialized to the value 1 and a variable a to the value 0 (step 504). Then, one tests during a test 506 if j> Nf, Nf denoting the number of prime factors of dR. If the 506 test is positive, the value of the variable a to the variable H _ref [1] [c] (step 508), the value of the variable I (step 510) is incremented by one unit and returns to the test 502.

If the test 506 is negative, the value of (the sign [J

designating the integer part) to the variable x and calculate I _j = I - xf, (step 512).

Then, we assign the value of the variable y and calculate η = c - yf _j (step 514).

The next step 516 is to assign the value of a + - - c, - I, to

the variable a.

Then, during step 518, the value of variable j is incremented by one and the value of variable x is assigned to variable I and the value of variable y to variable c.

We then return to test 506.

Turning back to FIG. 3, operation 304 consists of calculating the column H _of [.] [I] by calculating the

H _cle [l] [i] - h _ref [σ _{cle lιg} (l)] [σ _cle (i)] + p _{cle lιg} (I) + p _cle (i).

One embodiment of this operation is illustrated by the flowchart of FIG.

As shown in FIG. 6, an initialization step 600 firstly consists in initializing a variable I at the value 0, a variable c at the value of the auxiliary function σ _c ié, coi (i), a variable h the value of H _ref [c] and a variable r to the value of the auxiliary function p _c ie, cor (i) [.] -

Then during a test 602, we check if I = d _R. If this test is positive, the algorithm ends. Otherwise, the variable H _of [l] [i] is assigned the value of the expression r + h [σ _c ié, hg (l)] + Pcie, hg (l) (step 604). Then increment the value of variable I (step

606) then return to test 602. It is now explained how the basic method of generating mutually orthogonal time signals described above is applied in all its generality, according to the present invention.

It has been seen above that the basic method generates mutually orthogonal time signals or mutually orthogonal complex spectrums according to a power spectrum mask as shown in FIG. 1. This approach was didactic and made it possible to expose the operation of base: the generation of H _of [l] [i].

In all its generality, the present invention makes it possible to generate families S = {s (i)} of mutually orthogonal discrete signals of the system evolution space A:

A _ t ^γ i ^xγ 2 ^{x x Y} x) ^{Zi XZ2><XZτ} = E ^Y 1 ^XY 2 ^XX Yx ^X Z ₁ ^X Z ₂ ^X Z _τ having X actual spatial dimensions (E = 9i) or complex (E = C) and T temporal dimensions and to generate the family S = (s (i)} of their mutually orthogonal complex spectra in:

ZpY ₁ XY ₂ X xY _x xZ _τ _ pY _{i X} γ ₂ χ XY _x XZ ₁ XZ ₂ XZ ₁ - according to spectral constraints which will appear during the description of the process in its greater generality.

Let [n] denote the set [n] = {0; 1; 2; ...; n-2; n-1} of positive or zero integers less than n.

Let Q = [Yi] ^χ ... ^χ [Y _x ] ^χ [Zi] ^χ ... ^χ [Zτ] = [qo] ^x - - ^x [qx + τ-i] be the set Q of the coordinates of system, that is, the elements of Q are coordinates. For practical reasons, we define c \ _\ in correspondence with the Y and Z dimensions of the system _1. Q defines the dimensions of the system: | Q | , its number of elements or coordinates is the number of freedoms of system A.

The set Q makes it possible to express the values of a signal f of A as a function of Q in E, ie f: Q ^ E; x ι-> f (x). In other words, a signal is comparable to f, an element of the set of functions of Q in E, denoted by F (Q ₁ E), ie f <≡ F (Q ₁ E). The set F (Q ₁ E) is known as a vector space, and thus, as before, we will equate the signals f with vectors of the vector space F (Q ₁ E).

By definition of complex spectra in the Fourier domain, for every signal F of F (Q ₁ E): and

Let QP be the set of parts of Q (QP = ^ (Q)), that is, the set that contains all the subsets of Q. Let G = {cg _a = (g _a , key _a) / g _a <≡ QP and ae [n]} a family of parts n g _a Q mutually disjoint, that is to say such that if a ≠ b g _has then ng _b = 0. g _a are called "bands". A key named key _has been associated to them, possibly unique, which allows the generation of a variety of weakly correlated signals families. G is the definition of the constraints of the system: each constraint cg _has

= (g _a , key _a ) expresses what are the spatio-temporal frequencies composing the band g _a ; the frequencies outside g _a have a zero amplitude.

The process in all its generality allows to create n families S _a =

{s _a (i)} of signals and S _a = {s _a (i)} their versions in the Fourier domain, having spectra respecting the constraints determined by G and formalized by the following properties:

1. the spectrum of a family S _a is controlled by its band g _a , ie keg _a "| s (i) [k] | = μ _a

Va e [n], Vi e S _a , Vk and Q (1.10) k * g _a "Is (J) [K] I = 0

2. the families are decorrelated between them, ie V (a, b) e [n ^ a ≠ b ^ (vf e S _a , Vg e S _b , (f | g) = O and (f | g) = θ )

3. the signals of a family are decorrelated between them, either Va e [n], V (U) e | S _a | ² , (s _a (i) | s _a (j)) = 0 "(s _a (i) | s _a (j)) = 0" i ≠ j

4. The power of a signal is controlled, either

Va e [n], Vi e S _a , (s _a (i) | s _a (i)) = 1

The method allows, from families S _a = {s _a (i)} of signals and S _a = (s _a (i)} of their complex spectra, to generate a family S = {s (i)} of mutually orthogonal signals (decorrelated) and the family S = (s (i)} of their complex spectra which are themselves orthogonal, and whose power spectrum in each of the bands g _a is independent and arbitrary.

The method allows, from families S ₃ = {s _a (i)} of signals and S _a = (s _a (i)} of their complex spectra, to generate a family S = {s (i)} of signals according to two different methods whose choice depends on the use to which the family S is intended:

- Dependence of the bands between them, which means that the orthogonality in one band implies the orthogonality in the other bands, and thus that the signals of the family S are orthogonal to each other;

- independence of the strips together, which means that the families _S signals can be combined without restriction and implies that S family signals are not always orthogonal pairs.

The flowchart of FIG. 7 illustrates the process according to the present invention in all its generality.

The constraint G is segmented into its constraints by ego band, cgi, cg ₂ , etc. Each band g _a gives rise to the generation (via the modules 70, 71, 72, etc.) of a signal s _a (i) and of its variant s _a (i) in the Fourier domain, having a controlled spectrum . The signals thus created are multiplied (via the modules 700, 701, 702, etc.) by a factor c _a which controls the power of the signal in this band. The signals are then added in a module

7000 to give the complex spectrum s (i, c) and / or the signals s (i, c) sought (s).

The factors c _a can be modified without changing the orthogonality (non correlation) of the spectra and signals, they modify only the balancing of the power of the signals s (i, c) and s (i, c). The method in all its generality illustrated in FIG. 7 uses the so-called "band calculation" method (in the modules 70, 71, 72, etc.).

The band calculation is illustrated in more detail on the flowchart of FIG. 8. It includes a step 80 called "band preparation" which makes it possible to determine once and for all the generation parameters that will be used by the so-called "calculation" method. spectral "illustrated by block 82.

Spectral calculation method allows the generation of the complex spectrum _S (i) of the ith band signal g _a. The signal s (i) is obtained at the output of the module 84, by a discrete inverse Fourier transform of its complex spectrum s _a (i), by a simple implementation of the formula

s (i) [k], possibly optimized according to classical techniques of fast Fourier transformation.

The band preparation step is illustrated in greater detail in the flowchart of FIG. 9. This mechanism is composed of a so-called "extension calculation" module 90 which determines, as a function of the constraint cg _a and the dimensions of the system, the dimension _a, R Hadamard matrices (and thus the maximum number of the family of signals) and PRLG extension _data to be used for the extension. The preparation module 300 is the one described above in conjunction with FIGS. 3 and 4. The flowchart of FIG. 10 shows in greater detail the organization of the spectral calculation step implemented by the block 82 of FIG. 8. This step includes:

step 302 of calculating H _ref [.] [σ _{a of c0 |} (i)] illustrated in FIG. 3 and explained in FIG. 5; step 304 of calculating H _{a of} [.] [i] illustrated in FIG. 3 and explained in FIG. 6;

a so-called "exponential" step 106 which consists of calculating the complex numbers corresponding to the calculated rotations and normalizing them by a scale factor making it possible to obtain the power property controlled. The formula is: H _{a key} [l] [i] = . This step is illustrated in more detail in Figure 11;

a step 108 of extension which makes it possible to constitute the complex spectrum s _a (i) with controlled power sought. As shown in FIG. 11, an initialization step 110 firstly consists of initializing a variable I at the value 0. Then a test 112 consists in checking whether the value of the variable I has reached the maximum number d _a , R of family signals. If this is the case, the procedure ends. Otherwise, we calculate

^^ H ₃ of [I] [I]

H _{has key} [l] [i] = μ _a .e ^θ during a step 114, then increments the variable I by one unit (step 116) and returns to the test 112.

The extension operations 108 (FIG. 10) and the calculation of the extension 90 (FIG. 9) differ according to the nature of the signal sought, that is to say according to whether a real (E = 9) or complex ( E = C).

The flow charts of FIGS. 12 and 13 show the calculation steps in the case of complex signals. This case is the simplest because there is no symmetrization constraint.

FIG. 12 illustrates in greater detail the step 90 of calculating the extension of FIG. 9 in the case of complex signals. In a step 120, the value of the band g _a is assigned to the variable prlg _a . Then during a step 122, the value of | g _a | to the variable d _a , R. We calculate

then μ _a = (step 124), which ends the procedure. Respect for property of controlled power, that is to say (s _a (i) | s _a (i) ^ = 1, leads indeed to

ask μ a

FIG. 13 illustrates in greater detail step 108 of extension of FIG. 10 in the case of complex spectra. During a first step 130, the variable s (i) [.] Is initialized to the value 0 and during a step

132, one initializes a variable I to the value 0. Then a test 134 consists of checking if the variable I is equal to d _{a, R.} If this is the case, the procedure ends. Otherwise, it calculates, in a step 136, s (i) [pr IgJk]] = _{H key} [l] [i]. The value of I is then incremented by one unit (step 138) and then returns to test 134.

In the case of real signals, it is necessary to ensure equality s _a (i) = s _a (i). This equality is true when Vk e Q, s _a (i) [k] = s _a (i) [k]. This definition uses the conjugate definition of the k coordinate. This bijection of Q in Q is defined as follows: k = (K ₀ , K ₁ , ..., K _{x +} J _-1 ) -w- k = (k ₀ , K ₁ , ..., K _{x +} J _-1 ) f k. = 0 <=> k. = 0 where <_. We see that k, = k, and q, even => q, / 2 = q, 12.

[k, ≠ O o k, = q, - k, This constraint is taken into account during the step of

"preparation of the extension" illustrated in detail by the flowchart of Figure 14.

We begin with steps of initialization of d _a , R at the value 0 (step 140), extension data prg _a to the empty set (step 142) and variable I to the value 0 (step 144). .

We then proceed to a test 146 to check if I = | g _a | . If this is the

1 case, one calculates μ _a =, = (step 148) and the procedure ends. If not,

^J a, R is assigned to the variable k the value of g _a [l] (step 150). Note that the value of μ _a is different depending on whether the signals are real or complex. Then a test 152 consists in determining if k = k. If this is the case, the value of the variable I is incremented by one (step 154) and then returns to the test 146. Otherwise, it is checked whether ke prlg _a (test 156). If this is the case, proceed to step 154. Otherwise, it is determined in a test 158 whether ke prlg _a . If this is the case, go to step 154. Otherwise, add k to the extension data (step 160), increment by one unit the value of d _a , R (step 162) and then go to step 154. This extension of preparation removes the elements of g which _is undesirable. Thus, the dimension d _a , R can have a value different from

The extension in the case of real signals is illustrated by the flowchart of FIG. 15. The procedure is identical to that of the complex case illustrated in FIG. 13 with regard to the steps 130 to 136 described above. In the case of real signals, step 136 is further followed by a step 164 of calculating s _a (i) [prlg _a [k]] = H _{a cle} [1] [i]. The variable I is then incremented by one unit (step 166) and then returns to test 134.

This procedure ensures the property necessary for the real ones:

Vk <≡ Q, s _a (i) [k] = s _a (i) [k]. In this case, the calculation shows that the factor μ _has to be worth μ _a =

The maximum number of signals of the family S generated according to G is given according to the use cases:

- band dependence between them, which means that orthogonality in one band implies orthogonality in the other bands,

= min S 'a ₃ = min d ae [n] ae [n] a R'

- independence of the bands between them, | S | =] ^~ [| S _a | =] ^~ [d _{a R.} [n] ae [n] The invention is applicable in many fields. A first example concerns the tattooing of audio files.

Orthogonal signals are also very useful in transmission, where they are suitable for orthogonal, bi-orthogonal and CDMA modulations. FIG. 16 shows, by way of nonlimiting example, that the method according to the invention makes it possible to provide time signals or their Fourier variants with discrete spectra controlled at the level of the modulation of the messages. At the demodulation, the data received from the transmission channel are processed in order to reconstitute the messages sent. The process of Figure 16 can be used to power a read-only memory (ROM), which avoids boarding the process.

In FIG. 16, the binary message is modulated (within a block 1600) by the signal allocated to the user #i for whom the binary data is intended. The signal is then processed to be transmitted on the mobile radio channel 1602. In reception, the user #i performs the correlation of the received signal (block 1604). Since the signals are orthogonal, the user # i will only detect, in the stream he receives, the data that is intended for him.

These signals can also be used in metrology, where they can make it possible to optimize the excitation data to be supplied to the studied system, as shown in FIG. 17. This makes it possible to increase the relevance of the result of the measurement performed on the system. system studied. The method of Figure 17 can be used to power a ROM, which avoids boarding the process. A typical case of use in metrology is the measurement of the impulse response of acoustic rooms.

To carry out this measurement, a long periodic signal sequence is emitted by a loudspeaker, each period of which is flat spectrum or controlled according to the case. The periodic signal will be filtered by the impulse response of the room. The signal is retrieved from a microphone for processing. The percussion response of the room is obtained by making the cross correlation between the signal received by the microphone and the transmitted sequence.

To obtain a precise room response, it is necessary that the transmitted sequence has a correlation function equal to a Dirac distribution. This is precisely the case of the time signals which are the subject of the present invention.

Correlation is performed by taking a sequence of twice the size of the desired impulse response. An efficient way of performing the correlation is to perform the discrete Fourier transform of the received signal, to multiply the frequency signal obtained by the conjugate complex of the Fourier transform of the orthogonal sequence and to perform a fast inverse Fourier transform of the received signal. resulting signal. Like the signals are generated in the spectral domain, it is sufficient to store in ROM the frequency version of the sequence.

These signals can also be used as a basis for encoding or representing signals. For example, these signals may be used in digital audio coding as illustrated by way of example in FIG. 18. The use of orthogonal complex spectrums in the frequency domain makes it possible to code the speech or audio signal directly in this domain.

Since the complex spectra used have a frequency-controlled extent, the quantization noise shaping can be performed so that the noise faithfully follows the masking curve over the specified frequency bands.

The coding device illustrated in FIG. 18 comprises a dictionary in which the complex spectra generated according to the method of the invention are stored. This type of dictionary is used in the coding or decoding of audio signals to implement the quantization and inverse quantization step.

In a particular example, the signals are generated by Kronecker products of base matrices, the scalar product of the signal to be quantified by all the waveforms of the dictionary can then be efficiently obtained by a fast algorithm involving butterflies such as of the Fast Fourier Transform (TFR) or that of the actual Hadamard transform.

For a matrix of dimension Ns generated by Kronecker products according to equation (1.9), the flow per sample will be given by:

Flow = log ₂ (Ns) / Ns

For usual values of Ns of a dozen, this flow will be relatively low. To increase the rate, the dictionary is expanded by taking different matrices as generating matrices. For example, for an order 2, there are 64 possible basic matrices, the elements of each matrix being orthogonal and chosen in (1, -1, i, -i). More specifically, Figure 18 illustrates the use of mutually orthogonal complex spectra in a time-domain predictive coder.

The contribution of the filtering of an emitted signal s _e (n) with zero excitation (block 180) is first removed from the signal (subtractor 182) to give the target t. The target t in the frequency domain is obtained by Fast Fourier Transform of t (block 184). The complex samples of the signal are quantized by a quantizer 186 defined by a dictionary containing the Ns orthogonal complex vectors s (0), ..., s (Ns - 1).

Reverse operations of those just described are carried out after passing through a transmission channel 188, finally resulting in a received signal s _re c (n).

Quantification amounts to minimizing the following criterion:

E, = | t - ghs (i) | ² i = 0, ..., Ns - 1 where h is a diagonal perceptual weighting matrix in the frequency domain and g is the normalization gain. By minimizing Ej, we find: t 'h' s (i), ^ j. . . - -,. -. g = - - (or denotes ICI the transposed matrix) s (i) h 'hs (i) and the index i _op t which minimizes the error criterion is given by: Ns - 1 The numerator is equal to the scalar product of t ^τ h ^τ by all the waveforms of the dictionary generated by a Kronecker product of elementary matrices for which an effective algorithm of the "Hadamard transform" type is realized. The resulting structure is based on a butterfly structure similar to that of the TFR. In the particular case where the normalization gain g is known, the computation of the optimal index amounts to computation of the index i _op t which maximizes [t ^τ s (i)] ² . A particular example of implementation is obtained by segmenting the spectrum into frequency bands, contiguous or not, of variable length, possibly with areas of zero amplitude at high frequencies.

In the first case, we can for example encode the whole spectrum in areas of increasing frequency or increasing perceptual importance (deduced in this case from scale factors), by means of families of complex orthogonal spectra as represented in FIG. FIG. 18, each zone of the spectrum being quantified by a matrix generated by

Kronecker. Scalar product calculations involved in quantization are performed by fast transformations using the matrix structure. It is thus possible, progressively, to obtain a hierarchical binary train, the desired property of compression of the signals.

Fig. 19 shows an example of using mutually orthogonal signals in the case of audio tattooing. The complex signals or spectra are generated according to the description given above, the method illustrated being either on-board or off-line, the signals being stored once and for all in a ROM.

If we want to transmit a binary message of index i, we will filter and modulate the index signal i (block 190) and add the result of this operation to the audio signal (adder 192).

After transmission (block 194), on reception, the resynchronization and the correlation (block 196) are first carried out between the received signal and the set of signals of the reception dictionary. The detected signal is the one that gives the maximum correlation. The fact that the signals are spectrum controlled makes it possible not to disturb the signal unnecessarily by inserting the potentially audible mark in areas of the spectrum which, in any case, will not be transmitted by the codecs.

It could furthermore be envisaged to use the signals obtained according to the invention in the field of digital filtering.

A device 200 implementing the methods in accordance with the invention is illustrated in FIG. This device may be for example a microcomputer 200 connected to different peripherals, at least some of which may provide information to be processed according to the invention.

The device 200 may comprise a communication interface 2000 connected to a network (not shown). The device 200 furthermore comprises storage means 2002, such as a hard disk. It may also include a reader of 2004 information carriers such as floppy disks, CD-ROMs or memory cards 2006. The information carrier 2006 and the storage means 2002 may contain software implementation data of the invention. as well as the code of the invention which, once read by the device 200, will be stored on the storage means 2002. In a variant, the program enabling the device to implement the invention may be stored in read-only memory (by example a ROM) 2008. In the second variant, the program can be received via the network, to be stored in the same manner as described above.

The device 200 is connected to a microphone 2010 via an input / output card 2012. The data to be processed according to the invention will in this case be an audio signal.

This same device includes a 2014 screen to visualize the information to be processed or to interface with the user, who can set some modes of treatment, using a keyboard 2016, a mouse or any other way.

A CPU 2018 executes the instructions for implementing the invention, instructions stored in the ROM 2008 or in the other storage elements. When powering on, the programs and processing methods stored in one of the (non-volatile) memories, for example the ROM 2008, are transferred to a random access memory (for example, a RAM) 2020, which then contains the executable code of the invention as well as the variables necessary for the implementation of the invention. A communication bus 2022 allows communication between the different sub-elements of the microcomputer 200 or linked to it. The representation of the bus 2022 is not limiting and in particular, the central unit 2018 is capable of communicating instructions to any sub-element of the microcomputer 200 directly or via another sub-element of the microcomputer 200.

The device described here is likely to contain all or part of the treatment described in the invention.

Claims

A method for generating a plurality of discrete spectra s (i) of dimension Q, mutually orthogonal and with controlled power, i denoting the number of the spectrum, said spectra representing time signals in the spectral domain and being of constant μ module in a set G of spectral line designation and zero everywhere else, said method being characterized in that it consists of:

determining (20) at least a portion of a complex Hadamard matrix H of order dR = w in the case of real signal spectra and dR = 2.w in the case of complex signal spectra;

determining (22) the extension P of the matrix H from G and the dimension Q; and

obtaining (22) said power-controlled spectra s (i) = μ P (H [.] [i]), where H [.] [i] denotes the ith column of the matrix H.

2. Method according to claim 1, characterized in that the step (20) of determining at least a portion of the complex Hadamard matrix consists in obtaining a column (H _cle [.] [I]) of a rotational matrix calculated from predetermined keys of rotation and permutation applied to a reference Hadamard matrix (H _ref ).

3. Method according to claim 2, characterized in that it further comprises a step of decomposition of the order dR of said reference Hadamard matrix into a product of factors (f,) and a sub-step of calculating the least common multiple (θ) of all of said factors for the determination of the reference Hadamard matrix.

4. Method according to any one of the preceding claims, characterized in that it further comprises a step (24) of determining a signal

s (i) [k] = [ ^k N] [x [α]] or : • i denotes the number of the signal,

• Q denotes a set of coordinates,

• X is the number of spatial dimensions,

• T denotes the number of temporal dimensions and • F _(qJ denotes the Fourier matrix of order q _α , so as to generate a plurality of mutually orthogonal time signals s (i).

5. A method of generating a family of time signals t (i), characterized in that it consists in combining mutually orthogonal time signal families generated by a method according to claim 4 and whose spectral supports are disjoint.

6. Use of mutually orthogonal complex time signals or complex spectra generated by a method according to any one of the preceding claims for spectrum spreading in spread spectrum transmission systems.

7. Use of mutually orthogonal complex time signals or complex spectra generated by a method according to any one of claims 1 to 5 for audio tattooing and detection thereof.

8. Use of mutually orthogonal complex spectrums with controlled power spectrum generated by a method according to any one of claims 1 to 4 for coding or representing audio signals, the audio signals being quantified using a dictionary or a family of dictionaries with real or complex values.

9. Use of time signals or mutually orthogonal complex spectrums with controlled power spectrum generated by a method according to any one of claims 1 to 5 for the optimization of metrology excitation data.

10. Device for generating a plurality of discrete spectra s (i) of dimension Q, mutually orthogonal and with controlled power, i denoting the number of the spectrum, said spectra representing time signals in the spectral domain and being of constant μ module in one spectral line designation set G and zero everywhere else, said device being characterized in that it comprises:

means for determining at least a portion of a complex Hadamard matrix H of order dR = w in the case of real signal spectra and d _R = 2.w in the case of complex signal spectra;

means for determining the extension P of the matrix H from G and the dimension Q; and

means for obtaining said power-controlled spectra s (i) = μ.P (H [.] [i]), where H [.] [i] denotes the ith column of the matrix H.

11. Device for encoding audio signals, characterized in that it comprises a dictionary in which are stored complex mutually orthogonal spectra generated by a device according to claim 10.

12. A computer program product that can be loaded into a programmable device, characterized in that it comprises sequences of instructions for implementing a method according to any one of claims 1 to 5, when the program is loaded and executed by the programmable device.