Classifications

• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/63—Joint error correction and other techniques
• • G—PHYSICS
  • G10—MUSICAL INSTRUMENTS; ACOUSTICS
  • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
  • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
  • G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
  • G10L19/08—Determination or coding of the excitation function; Determination or coding of the long-term prediction parameters
  • G10L19/10—Determination or coding of the excitation function; Determination or coding of the long-term prediction parameters the excitation function being a multipulse excitation
  • G10L19/113—Regular pulse excitation
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/25—Error detection or forward error correction by signal space coding, i.e. adding redundancy in the signal constellation, e.g. Trellis Coded Modulation [TCM]
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/65—Purpose and implementation aspects
  • H03M13/6577—Representation or format of variables, register sizes or word-lengths and quantization
  • H03M13/6594—Non-linear quantization
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L27/00—Modulated-carrier systems
  • H04L27/18—Phase-modulated carrier systems, i.e. using phase-shift keying
  • H04L27/183—Multiresolution systems
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L27/00—Modulated-carrier systems
  • H04L27/32—Carrier systems characterised by combinations of two or more of the types covered by groups H04L27/02, H04L27/10, H04L27/18 or H04L27/26
  • H04L27/34—Amplitude- and phase-modulated carrier systems, e.g. quadrature-amplitude modulated carrier systems
  • H04L27/3488—Multiresolution systems

Definitions

• the invention relates generally to coding with data compression.
• data compression is motivated by more efficient management of the channel resource during data transmission or by a reduction in the size of the memory in a data logger.
• Data compression is based on an extraction of the redundancy of the symbols to be transmitted according to a predetermined law.
• the redundancy is added to the reception, during the reconstruction of the data.
• the compression is said to be without distortion.
• some distortion is accepted in the reconstruction process.
• image coding for example, transform coding achieves fairly low coding rates.
• speech coding analysis and synthesis coding achieves low coding rates.
• Transform compression provides very low coding rates with reduced distortions. Compression is then achieved through a representation of the information, and discrimination is made through an unequal distribution of the frequency components. In general, the goal of any source coding is precisely the discovery of this structure in the information to be coded.
• the non-repeat information becomes more sensitive to the effects of noise and consequently the performances compared to the model without noise deteriorate quickly.
• the invention finds applications in very numerous fields, in particular in the coding of vo x and the coding of images.
• digitized speech is transmitted at low bit rates expressed in kbit / s, with correct reconstruction quality.
• the digitized image is transmitted at low bit rates, expressed in tens, hundreds or even thousands of kbit / s, and this with a correct reconstruction quality.
• the field of images always requires considerable amounts of information to manage.
• the invention relates more particularly to block coding according to which the source and channel coding operations are integrated into a single entity.
• This coding technique is called joint source-channel coding.
• the implementation of this technique is based on a redefinition of the quantification of the coefficients of the source linked to the property of bi] action between the no ore ⁇ e levels of the quantifier and the number of states of the modulations used.
• compression we always refer to two algorithms: the first generates a representation of the source in the form of a limited number of bits; - the second reconstructs the source from the representation bits, possibly tainted with errors.
• the compression schemes are classified into lossless schemes and lossy schemes.
• the lossless or distortion-free diagrams are diagrams where the reconstruction of the data on decoding gives a source identical to the source to be transmitted. This is typical of compressing written text or computer generated information.
• the perfect reconstruction constraint cannot be verified. Distortion between the original and the reconstructed is tolerated to obtain a higher compression ratio.
• the reconstructed voice signal samples are not necessarily the same as those of the transmitted signal.
• a certain degree of distortion is tolerated without compromising the understanding of the message sent.
• the higher the compression ratio the greater the loss introduced by the coding process.
• Group 1 Joint Source-Channel Codings according to which the source and channel coding operations are integrated into a single entity.
• Group 2 Concatenated Source-Channel Encodings whereby a given source encoder is concatenated to a given channel encoder and the source and channel encoding bit rates are determined, so as to maximize the performance of the set.
• Group 3 Joint Codings with Constraints according to which the coder and / or the decoder are modified so as to take into account channel errors; for example, a source encoder optimized for a noise-free channel is re-optimized to take into account the statistics of the channel.
• the main objective of the invention is to provide a joint source-channel coding of group 1 in particular for the coefficients at the output of a vocoder, by avoiding adding redundancy by means of a code for correcting errors, as in a group 2 coder, and to be very sensitive to the slightest variation in channel characteristics, as with a group 3 coder.
• a joint source-channel coding comprising a quantization and a phase modulation with a constellation of NC points
• the constellation of the pnase modulation is a rotated constellation deduced from a predetermined rotation of a uniform phase constellation at NC state points so that the NC points of the constellation are projected onto any one of the axes of the constellation in NC all different projections
• the quantization presents a non-uniform quantization law of NS levels which are NS projections among NC projections different from the constellation rotated.
• the numbers NC and NS are integers, with NS ⁇ NC.
• the constellation of NC points is a phase modulation constellation, or an amplitude modulation constellation
• the invention also relates to a speech or image coder whose quantization means, also called coders, producing coefficients and parameters to be multiplexed in frames are joint source-channel coders implementing conforming joint source-channel codings to the invention.
• the joint source-channel coders for long-term analysis filter gain coefficients, excitation grid position parameters, and short predictor filter coefficients high rank terms include QAM-type phase constellation modulators, and joint source-channel coders for short-term short-term predictor filter coefficients, long-term analysis filter delay coefficients, and amplitude block parameters include phase constellation modulators
• the modulators with phase constellation of MAQ and MAQ types are respectively preceded by quantifiers having non-uniform quantization laws the levels of which are respectively subsets of projections relating to points of a constellation of phase predetermined turned to at least four dimensions.
• the predetermined tour phase constellation is that resulting from a rotation of the constellation of the
• the invention quantifies the coefficients at the output of the vocoder to transform them into samples taking a finite number of possible values. These samples were chosen to take into account the disturbances of the channel. In addition, this coding does not show sensitive bits as is the case with conventional tandem techniques. Thus, the resistance of the coding of the invention to channel noise is greater.
• the vocoder of the invention provides a compressed speech signal resistant to disturbances of the transmission channel, allowing the production of transmitters and receivers with complexity and limited cost, in particular as regards coding and decoding.
• the bandwidth of the transmission channel is reduced by several units compared to the prior art.
• FIG. 1A and 1B are diagrams of phase constellations of an MDP4 modulation not turned and turned respectively;
• FIG. 2 is a diagram showing the transformation between a constellation ⁇ e channel corresponding to the phase modulation MDP4 turned and a source constellation according to the invention;
• - Figure 3 is a schematic block diagram of a joint source-channel encoder and a joint channel-source ⁇ ecoder according to the invention connected by a faded transmission channel;
• FIGS. 4A and 4B are diagrams of phase constellations of an MAQ-16 modulation not turned and turned respectively;
• FIG. 5 is a block diagram of a speech coder according to the prior art for cellular GSM mobile radio system
• FIG. 6 is a block diagram of a joint source-channel coder in blocks according to the invention for a cellular GSM mobile radio system.
• coefficients at the output of an encoder are quantified according to a joint code in blocks.
• This quantization scheme labels the coefficients and output of the vocoder, called subsequently sample, not by bits, but by symbols which are transmitted directly on the channel.
• h (t) ' denotes the pulse response ⁇ of the transmission filter and T is the inverse of the modulation speed expressed in bauds
• the symbols a ⁇ are directly transmitted on the channel and represent all or part of a quantified sample from the vocoder.
• the relation that there is between the quantized samples of the vocoder and the symbols transmitted has, identifies the coding in blocks.
• the code is the set of all the possible symbol sequences a k that can be transmitted on the channel.
• the encoder transforms samples from the source not yet quantified, or more finely quantified, into code words. For this, the coder searches for the code word closest to the sequence of samples sent by the vocoder according to a certain predetermined distance criterion. Very often, this distance criterion is the Euclidean distance between the sample leaving the voco ⁇ eur and the symbol of the source. These source symbols are then transformed into channel symbols a.
• the channel ⁇ istor ⁇ these co ⁇ e words by adding to the noise, or by introducing interference between symbols, for example.
• the decoder then reconstructs the vocoder samples from the noisy observations. For this, he chooses the code word located - at the shortest distance according to a certain criterion of the word received. Then, it associates the quantified sample which corresponds to it. Finally, thanks to the coefficients of the reconstructed vocoder, the voice decoder reconstructs the spoken word.
• the invention implements the following joint source-channel coding in the particular mode of joint source-channel block codes.
• the coding In joint coding, during digital transmission through fading channels, such as radio channels for mobile radiotelephones, the coding consists of passing from a source code, called source dictionary, composed of points of a space a DS dimensions to a channel code, called a channel dictionary, composed of the same number of points, ma s in a space with DC dimensions. Assuming that DC> DS, the points of the source dictionary in a number equal to the points of the channel dictionary must all be different, but in a space of smaller dimension than the channel space. For transmission in a fading channel, the dimensions where the fading takes place are precisely similar to the missing dimensions of the source dictionary. The resolution of the problem of transmission in a Rayleigh channel calls upon the notion of diversity which consists in distributing the information in the greatest possible number of components of the code word, so as to recover it in the components which are not not fainting.
• the approach followed by the invention is the reverse of the classic spouse approach. Diversity is used when going back from the channel to the source.
• the invention introduces the notion of turned constellations. Modulation techniques using point arrays have become very powerful tools for the design of digital transmission systems with high spectral efficiency both for a Gaussian channel and for a fading channel, and in particular for a Rayleigh channel.
• a diversity order L of a multidimensional constellation is the minimum number of different components between any two points of the constellation.
• diversity is designated by L.
• To introduce diversity into a multidimensional constellation it is rotated so that all pairs of points have a maximum number of different components.
• each point of the uniform phase constellation has one of its coordinates equal to another point of the constellation, and in FIG. 1B, each point of the constellation turned to each of its two coordinates , XI, X2 different from the corresponding coordinates of the other points of the constellation being rotated.
• the turned constellation offers an additional degree of protection.
• the rotated constellation gives an gain of 8 dB compared to the unpurned MDP4 constellation.
• DC is an integer such that DC> 2
• the principle is the same.
• a block coded modulation seen as a finite subset of a network of points is "rotated" by a rotation at dimensional DC so as to have the greatest possible diversity up to DC. In this case, the expected gains are even greater than those for a 2-dimensional constellation.
• the invention establishes an application between the levels resulting from a source coding and the points of the channel constellation in order to obtain the desired robustness in terms of distortion of the source reconstructed on reception.
• the joint source-channel block coding of the invention exploits the diversity of the constellations turned to find this application.
• Source coding is derived from channel coding.
• the invention applies the property that, if a code comes from a constellation turned to maximum diversity, which is equal to the dimension of the constellation, then the projections on any one of the axes of the constellation constitute a set with a cardinal equal to that of the code.
• This property is the basis of joint block coding according to the invention, and is deduced from the very definition of the diversity of a constellation.
• the source dictionary is constituted by the points or levels indicated by a cross on the axis 0-X1, while the channel dictionary consists of the points [XI, X2] of the constellation MDP4 tour.
• the method of the invention is advantageous in that it adapts the scalar quantizer to the distribution of the source by simple modification to the angle of rotation.
• FIG. 2 an example of a quantification scale for a source sample is shown in thick lines.
• Figure 3 shows an encoder 1 in a transmitting device and a decoder 2 in a receiving device according to the invention, connected through a fading channel 3.
• the coder 1 comprises a source 11, a joint source-channel coding circuit in blocks 12 and an interleaver 13.
• the source 11 is for example a vocoder.
• the coding circuit 12 includes a scalar quantizer 121 followed by a modulator
• the scalar quantizer is associated with a programmable circuit 123 which establishes a non-uniform quantization lo according to the source dictionary as shown at the bottom of FIG. 2.
• the quantization lo is chosen as a function of the initial constellation (FIG. 1A) of the modulator 122 after a rotation thereof with a predetermined angle ⁇ in order to maximize the cardinality of the points projected on a predetermined axis O-Xl of the constellation CC channel.
• Each sample provided by the vocoder 11 and representing for example a parameter of the vocoding model implemented in the vocoder 11 is quantified in the quantizer 121 according to the quantization law called by rounding, by corresponding to the level or point of the source constellation CS , closest to the sample.
• the modulator 122 converts by "reverse" projection each point of the constellation CS transmitted by the quantifier into the corresponding symbol of the channel constellation CC which, according to FIG. IB or 2, is defined by Cartesian coordinates XI (t) and X2 (t), or polar, amplitude p (t) and phase ⁇ (t) that are different from the coordinates of all the other symbols of the constellation CC.
• the interleaver 13 interleaves the symbols produced by the modulator 122 to introduce time diversity therein in a known manner.
• the transmission channel 3 is shown diagrammatically in FIG. 3 by a generator 31 generating multiplicative noise due to fading and a generator
• a deinterleaver 21 deinterleaves the interleaved symbols according to the initial order, and an estimation circuit 22 estimates the amplitude and the instantaneous phase of the channel.
• a demodulator 23 demodulates each deinterleaving symbol in points of the source constellation CS, which had been emitted, by carrying out the transformation of the constellations CC to CS of FIG. 2 by projection on the predetermined axis O-Xl.
• each set of projections resulting from the projection of the points ⁇ e the constellation MAQ-16 on one of the axes 01, OQ before application of the rotation matrix has a cardinal of 4, i.e. there is no has only 4 different component values relative to each axis 01, OQ for the 16 points.
• the rotation and projection operations for designing a scalar quantifier with a bijection between the points of the channel constellation and the points or levels of the linear source constellation is a characteristic of the invention.
• the invention is applied to the full-rate speech coder included in a base station or in a mobile station of the cellular mobile radio network according to the GSM standard.
• the base station is connected to the switched telephone network via ISDN 64 kbit / s channels
• the speech signal initially sampled at 8 kHz on 8 bits is compressed into a speech signal at 13 kbit / s to be transmitted in a radio channel.
• the coder COP essentially comprises a segmentation circuit SEG for segmenting a speech signal SP at 64 kbit / s, an analysis filter LPC (Lmear Predictive Coding) with linear predictor PL, a long-term prediction analysis_ filter LTP (Long Term prediction) and an RPE excitation signal calculation circuit
• the LPC short-term predictor filter models variations in the short-term vocal tract.
• the LPC predictor generates eight LAR (Log Area) coefficients
• LAR (2) and LAR (3) 5-bit quantization
• LAR (4) and LAR (5) 4-bit quantization
• LAR (6) and LAR (7) quantization on 3 Dits, i.e. 6 bits.
• the LTP long-term analysis filter models rapid variations in the vocal tract. It provides an LTP-Lag delay coefficient and an LTP-Gam gain coefficient, per sub-frame of 5 milliseconds, ie a speed of 1.8 kbit / s; the long-term coefficients on a 20 ms frame are: LTP-Lag: quantization on 7 bits, ie 4 x 7
• the RPE calculation circuit produces an excitation signal consisting of a block of pulses which are regularly distributed over time and coded at 9.4 kbit / s in position and amplitude by means of a grid.
• the excitation signal is composed of gate position parameters Mi at the output of a low-pass filter FBP, ⁇ e parameters ⁇ e amplitude block X axi, and values relating to _13 pulses x ⁇ (0) to x ⁇ (12 ) by subframe î of 5 ms.
• Excitation pulses x ⁇ (0) to x ⁇ (12): 3-bit quantization of the 13 pulses, i.e. 4 x 39 156 bits.
• the above coefficients and parameters are multiplexed in an MX multiplexer at the output of the encoder to produce frames at 260 bits every 20 ms, ie at the coding rate of 13 kbit / s. Then in a COC channel coder followed by an interleaver, these bits are separated into three different relative importance classes. The bits of the first two classes are protected by different error correcting codes, and the bits of the last class, the least important, are concatenated without protection.
• the channel encoder delivers a 456-bit code unit of data every 20 ms, a rate of 22.8 kbit / s. This selective protection by class is known as an Unequal Protection technique against errors.
• a joint source-channel speech coder in blocks COPa in accordance with the invention for GSM network comprises the functional circuits SEG, PLa, LPCa, LTPa, FPB and RPEa of the coder COP, but with the following changes.
• the channel coder COC is deleted, and the sets of quantizers, also called coders, QUI, QU2_ and QU3 at the output of the PL, LTP and RPE circuits are replaced by joint source-channel coding circuits in blocks. of the type of that 12 described with reference to FIG. 3 in the circuits PLa, LTPa and RPEa, with the exception of the pulse output of the RPEa calculation circuit.
• the decoders (not shown) in the LPC, LTP and RPE circuits respectively performing the inverse functions of the coders (quantifiers) in the PL, LTP and RPE circuits are replaced by joint channel-source decoding circuits of the type of that 23-24 shown in Figure 3.
• the modulation is carried out on 16 or 32 levels for example.
• An MAQ -64 constellation is provided for each of the low rank coefficients LAR (0) and LAR (l)
• the MAQ -256 constellation is the Cartesian product of two MAQ-l ⁇ constellations.
• the process of the invention can be generalized. Instead of considering a projection of a CC channel constellation of dimension 2 or 4 towards a source constellation CS of dimension 1, we can consider in the same way, the projection of a channel constellation of dimension DC towards a source constellation of dimension DS, with DODS.
• the source dictionary and the channel dictionary are chosen to be identical.
• the task of the channel code is to provide the necessary protection to prevent loss in the quality, during transmission.
• the bits which result from the quantization of the parameters are conventionally coded by analog-digital conversion into binary code words and then coded in the channel coder to be transmitted in the channel. The separation has become almost universal between source coding and channel coding.
• the method of the invention performs all of the source coding and the channel coding differently.
• the parameters to be transmitted are quantified in a non-uniform scalar quantizer which results from the projection of a rotated channel constellation.
• the DC dimension depends on the degree of importance of the coefficients to be transmitted. The higher the coefficient, the greater the DC.
• the coefficients of less importance correspond to scalar channel constellations, while the others will correspond to channel constellations with 2, 4 or even 8 dimensions for the most important coefficient.
• the performance of the coding of the invention is illustrated in terms of reduction in bandwidth for an auoio quality identical to that existing at present.
• Another approach is to present joint source-channel block coding as a means of building a communications system . which, for a fixed bandwidth, is capable of transmitting with a lower average power.
• the coding of the invention is then advantageous, for example in communication systems by satellites where the smallest decibel gained is very significant on the cost of the payload.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Signal Processing (AREA)
• Probability & Statistics with Applications (AREA)
• Computer Networks & Wireless Communication (AREA)
• Nonlinear Science (AREA)
• Computational Linguistics (AREA)
• Health & Medical Sciences (AREA)
• Audiology, Speech & Language Pathology (AREA)
• Human Computer Interaction (AREA)
• Acoustics & Sound (AREA)
• Multimedia (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)

Applications Claiming Priority (1)

Publications (1)

Family

ID=9542424

Family Applications (1)

Country Status (2)

Families Citing this family (1)

Family Cites Families (4)

• 1999
  • 1999-09-21 WO PCT/FR1999/002241 patent/WO2001022676A1/fr not_active Application Discontinuation
  • 1999-09-21 EP EP99943017A patent/EP1131928A1/de not_active Withdrawn

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events