EP1495464B1 - Device and method for encoding a time-discrete audio signal and device and method for decoding coded audio data - Google Patents

Abstract

According to the invention, a time-discrete audio signal is processed (52) in order to provide a quantization block with quantized spectral values (52). In addition, a whole-number spectral representation is generated from a time-discrete audio signal, using a whole-number transformation algorithm (56). The quantization block, which has been generated using a psychoacoustic model (54), is inverse quantized and rounded (58) to form a differential between the whole-number spectral values and the inverse quantized rounded spectral values. The quantization block alone produces a psychoacoustic encoded/decoded audio signal affected by loss after the decoding process, whereas the quantization block together with the combination block provides a loss-free, or practically loss-free encoded and decoded audio signal during said decoding process. The generation of the differential signal in the frequency range allows a simpler encoder/decoder structure to be produced.

Description

The present invention relates to audio coding / audio decoding and more particularly to scalable encoding / decoding algorithms with a psychoacoustic first Scaling layer and a second scaling layer, the additional audio data for lossless decoding includes.

Modern audio coding methods, such. Eg MPEG Layer3 (MP3) or MPEG AAC use transforms such as the so-called modified discrete cosine transformation (MDCT), to a block-wise frequency representation of an audio signal to obtain. Such an audio encoder usually receives a stream of time discrete audio samples. The stream of audio samples is windowed to one windowed block of, for example, 1024 or 2048 windowed To obtain audio samples. To the fenestration different window functions are used, such as B. a sine window, etc.

The windowed discrete-time audio samples are then by means of a filter bank in a spectral representation implemented. In principle, a Fourier transformation, or for special reasons a variety of Fourier transform, such as. B. an FFT or, as it has been executed, an MDCT be used. Of the Block of audio spectral values at the output of the filter bank can then be further processed as needed. Both The above-mentioned audio coders are followed by quantization the audio spectral values, the quantization levels typically be chosen so that by quantizing introduced quantization noise below the psychoacoustic Masking threshold is, d. H. "Masked away" becomes. The quantization is a lossy one Encoding. To get a further data volume reduction, the quantized spectral values are subsequently added entropy-coded for example by means of a Huffman coding. By adding page information, such as B. scale factors, etc. is from the entropy-coded quantized spectral values by means of a bitstream multiplexer a bitstream is formed, which is stored or can be transferred.

In the audio decoder, the bitstream is by means of a bit stream demultiplexer in coded quantized spectral values and split page information. The entropy-coded quantized spectral values are first entropy-decoded, to obtain the quantized spectral values. The quantized spectral values are then inversely quantized, to obtain decoded spectral values, the quantization noise but below the psychoacoustic level Masking threshold is and therefore inaudible will be. These spectral values are then determined by means of a Synthesis filter bank converted into a temporal representation, to obtain time discrete decoded audio samples. In the synthesis filter bank must be a transformation algorithm used inverse transformation algorithm become. In addition, after the frequency-time inverse transformation the windows are undone.

To achieve good frequency selectivity, use Modern audio coders typically have a block overlap. Such a case is shown in Fig. 4a. First, for example, 2048 become discrete-time audio samples taken and windowed by means 402. The window that embodies device 402, has a window length of 2N samples and provides output a block of 2N windowed samples. To achieve a window overlap is by means of a Device 404, which for reasons of clarity shown separated from the device 402 in Fig. 4a , a second block of 2N windowed samples is formed. The 2048 samples fed to the device 404 however, they are not those immediately adjacent to the first window subsequent discrete-time audio samples, but involve the second half of the through the institution 402 windowed samples and additionally include only 1024 "new" samples. The overlap is through a device 406 in Fig. 4a shown symbolically, the causes a degree of overlap of 50%. Both the through means 402 output 2N windowed samples and the 2N output by means 404 windowed samples are then sent by means 408 and 410 subjected to the MDCT algorithm. The device 408 provides N according to the known MDCT algorithm Spectral values for the first window while the device 410 also provides N spectral values, but for the second window, wherein between the first window and the second window has an overlap of 50%.

In the decoder, the N spectral values of the first window, as shown in Figure 4b, means 412, which is an inverse modified discrete cosine transformation performs, fed. The same applies to the N spectral values of the second window. These become a facility 414, which is also an inverse modified discrete Cosine transformation. Both the decor 412 and device 414 each provide 2N Samples for the first window or 2N samples for the second window.

In a device 416, which is referred to in Fig. 4b with TDAC (Time Domain Aliasing Cancellation), the fact is taken into account that the two windows are overlapping. In particular, a sample value y _{1 of} the second half of the first window, that is to say with an index N + k, is summed with a sample value y ₂ from the first half of the second window, ie with an index k, so that on the output side, ie in the decoder, N decoded temporal samples result.

It should be noted that by the function of the device 416, which is also referred to as an add function, in the coder schematically represented by FIG. 4a performed fencing taken into account, so to speak automatically is so that in the illustrated by Fig. 4b Decoder no explicit "inverse windowing" take place Has.

When implemented by means 402 or 404 Window function with w (k) is called, where the index k represents the time index, the condition must be satisfied that the window weight w (k) squared adds to the window weight w (N + k) squared together 1 where k is from 0 to N-1. If a sine window is used, whose window weightings of the first Half wave follow the sine function, so is this condition always fulfilled, since the square of the sine and the square of the Cosines for each angle together give the value 1.

A disadvantage of the window method described in FIG. 4a with subsequent MDCT function is the fact that the fenestration by multiplying a discrete-time Sample when thinking of a sine window, is achieved with a floating-point number, since the sine of a Angle between 0 and 180 degrees apart from the angle 90 Degree does not yield an integer. Even if integer discrete time Samples are windowed Windows so floating point numbers.

Therefore, even if no psychoacoustic encoder is used is, d. H. when a lossless coding is achieved is to be at the output of the devices 408 and 410 a Quantization needed to be reasonably manageable Entropy coding to perform.

So if known transformations, as based on 4a have been operated for lossless audio coding must be used, either a very fine quantization can be used to the resulting Neglect errors due to the rounding of floating point numbers to be able to, or the error signal must additionally for example, be coded in the time domain.

Concepts of the former kind, that is to say quantization adjusted so fine that the resulting error due the rounding of floating point numbers is negligible For example, in German Patent DE 197 42 201 C1 discloses. Here is an audio signal in its spectral Presentation converted and quantized to quantized To obtain spectral values. The quantized spectral values are again inversely quantized, transferred to the time domain and compared to the original audio signal. Is the error, so the error between the original Audio signal and the quantized / inverse quantized Audio signal, above an error threshold, then the Quantizer feedback set finer, and the Comparison is performed again. The iteration is over, if the error threshold is fallen below. That then still possibly existing residual signal is with a Time domain encoder encoded and written to a bitstream, the next to the time domain encoded residual signal also includes coded spectral values corresponding to the quantizer settings have been quantified at the time the termination of the iteration were present. It was noted that the quantizer used is not must be controlled by a psychoacoustic model, so that the coded spectral values are typically more accurate are quantized as this is due to the psychoacoustic Model would have to be.

In the technical publication "A Design of Lossy and Lossless Scalable Audio Coding ", T. Moriya et al., Proc. ICASSP, 2000, a scalable encoder is described first lossy data compression module z. As an MPEG encoder comprising a block-wise digital waveform as input and generates the compressed bitstream. In an existing local decoder the coding is undone again, and it will generates a coded / decoded signal. This signal is compared with the original input signal by the coded / decoded signal from the original one Input signal is subtracted. The error signal then becomes fed into a second module where a lossless bit conversion is used. This conversion has two Steps. The first step is a conversion of a two's complement format in a sign magnitude format. The second step is the conversion of a vertical magnitude sequence into a horizontal bit sequence in a processing block. The lossless data conversion is executed to maximize the number of zeros or the number of consecutive zeros in a sequence to maximize compression as much as possible the temporal error signal due to digital Numbers to reach. This principle is based on a bit-slice arithmetic coding (BSAC) scheme, in the technical publication "Multi-Layer Bit Sliced Bit Rate Scalable Audio Coder ", 103rd AES Convention, Preprint No. 4520, 1997.

A disadvantage of the concepts described above the fact that the data for the lossless enhancement layer, d. H. the additional data that is needed to achieve a lossless decoding of the audio signal, must be won in the time domain. This means, that is a complete decoding including a frequency / time conversion required to encode / decode Receive signal in the time domain, so by means of a sample-wise difference between the original audio input signal and the encoded / decoded one Audio signal due to the psychoacoustic Coding is lossy, the error signal is calculated becomes. This concept is particularly relevant disadvantageous that in the encoder which generates the audio data stream, both a complete time-frequency converter, such as B. a filter bank or z. B. requires an MDCT algorithm for the forward transformation is, and at the same time, only to the error signal generate, a complete inverse filter bank or a complete Synthesis algorithm is needed. The encoder must therefore in addition to its inherent encoder functionality also contain the complete decoder functionality. If the encoder is implemented by software, so For this purpose, both storage capacities and processor capacities are used needed to an encoder implementation leads with increased effort.

The object of the present invention is a less elaborate concept by which an audio stream can be generated, which is at least almost lossless is decodable.

This object is achieved by a device for coding a Discrete-time audio signal according to claim 1, by a method of encoding a discrete-time audio signal according to claim 21, by a device for decoding of coded audio data according to claim 22, by a method for decoding encoded audio data Claim 31 or by a computer program according to claim 32 or 33 solved.

The present invention is based on the knowledge that the additional audio data, a lossless decoding allow the audio signal to be obtained thereby that a block of quantized spectral values as usual is provided, and then inversely quantized, to have inverse quantized spectral values due to quantization by means of a psychoacoustic model are lossy. These inverse quantized spectral values are then rounded to a rounding block of rounded to obtain inverse quantized spectral values. As a reference for difference formation according to the invention a Integer transformation algorithm uses which a block of integer discrete-time samples Integer block of spectral values, which is only integer Spectral values generated. According to the invention now becomes the combination of the spectral values in the rounding block and spectral valuewise in the integer block, ie in the frequency domain performed so that in the encoder itself no Synthesis algorithm, ie an inverse filter bank or a inverse MDCT algorithm etc. is needed. The combination block, which comprises the difference spectral values due to the integer transform algorithm and of the rounded quantization values are only integer ones Values that entropy-encode in some known manner can be. It should be noted that the Entropy coding of the combination block of any entropy coder can be used, such. Huffman coder or arithmetic coders etc.

For coding the quantized spectral values of the quantization block can also be used any encoder be such. B. the well-known for modern audio encoder usual tools.

It should be noted that the inventive coding / decoding concept compatible with modern coding tools, such as Window switching, TNS or center / page encoding for multi-channel audio signals.

In a preferred embodiment of the present invention Invention is used to provide a quantization block of quantized using a psychoacoustic model Spectral values used an MDCT. Furthermore it is preferred as an integer transform algorithm to use a so-called IntMDCT.

In an alternative embodiment of the present invention Invention can be dispensed with the usual MDCT, and it can use the IntMDCT as an approach to the MDCT to the effect that the integer spectrum, that is obtained by the integer transform algorithm is fed to a psychoacoustic quantizer, to obtain quantized IntMDCT spectral values, then again inversely quantized and rounded to coincide with the original integer spectral values. In this case, only a single Transformation is needed, namely the IntMDCT, which is integer discrete-time samples integer spectral values generated.

Typically, processors work with integers, or Each floating-point number can be represented as an integer. When using integer arithmetic in a processor is, so can on the rounds of the inverse quantized Spectral values are waived because of arithmetic of the processor anyway rounded values, namely within the accuracy of the LSB, d. H. the least significant bit, available. In this case, a completely lossless Processing achieved, d. H. a processing within the accuracy of the processor system used. alternative However, a rounding to a coarser accuracy be performed, in that the difference signal in the combination block on the by a rounding function specified accuracy is rounded. Introducing a Rounding over the inherent rounding of a processor system in addition, flexibility allows the "degree" to influence the losslessness of the coding in the Meaning of a data compression, a nearly lossless encoder to accomplish.

The decoder according to the invention is characterized that from the audio data both the psychoacoustically coded Audio data as well as the additional audio data are extracted, subject to possibly existing entropy decoding and then processed as follows. First the quantization block in the decoder becomes inverse quantized and using the same rounding function, which has also been used in the encoder, rounded, then added to the entropy-decoded auxiliary audio data to become. The decoder then has both a psychoacoustically compressed spectral representation of the audio signal as well as a lossless representation of the audio signal before, the psychoacoustically compressed spectral Representation of the audio signal in the time domain implement is a lossy coded / decoded To receive audio signal while the lossless presentation using an integer transform algorithm inverse integer transformation algorithm implemented in the time domain is going to be a lossless or, as it has been stated, to obtain almost lossless coded / decoded audio signal.

Fig. 1

ein Blockschaltbild einer bevorzugten Einrichtung zum Verarbeiten von zeitdiskreten Audio-Abtastwerten, um ganzzahlige Werte zu erhalten, aus denen ganzzahlige Spektralwerte ermittelbar sind;

Fig. 2

eine schematische Darstellung der Zerlegung einer MDCT und einer inversen MDCT in Givens-Rotationen und zwei DCT-IV-Operationen;

Fig. 3

eine Darstellung zur Veranschaulichung der Zerlegung der MDCT mit 50-Prozent-Überlappung in Rotationen und DCT-IV-Operationen;

Fig. 4a

ein schematisches Blockschaltbild eines bekannten Codierers mit MDCT und 50 Prozent Überlappung;

Fig. 4b

ein Blockschaltbild eines bekannten Decodierers zum Decodieren der durch Fig. 4a erzeugten Werte;

Fig. 5

ein Prinzipblockschaltbild eines bevorzugten erfindungsgemäßen Codierers;

Fig. 6

ein Prinzipblockschaltbild eines alternativen erfindungsgemäß bevorzugten Codierers;

Fig. 7

ein Prinzipblockschaltbild eines erfindungsgemäß bevorzugten Decodierers;

Fig. 8a

eine schematische Darstellung eines Bitstroms mit einer ersten Skalierungsschicht und einer zweiten Skalierungsschicht;

Fig. 8b

eine schematische Darstellung eines Bitstroms mit einer ersten Skalierungsschicht und mehreren weiteren Skalierungsschichten; und

Fig. 9

eine schematische Darstellung von binär codierten Differenz-Spektralwerten zur Verdeutlichung möglicher Skalierungen hinsichtlich der Genauigkeit (Bits) der Differenz-Spektralwerte und/oder hinsichtlich der Frequenz (Abtastrate) der Differenz-Spektralwerte.

Preferred embodiments of the present invention will be explained below in detail with reference to the accompanying drawings. Show it:

Fig. 1

a block diagram of a preferred means for processing discrete-time audio samples to obtain integer values from which integer spectral values can be determined;

Fig. 2

a schematic representation of the decomposition of an MDCT and an inverse MDCT in Givens rotations and two DCT-IV operations;

Fig. 3

a representation to illustrate the decomposition of the MDCT with 50 percent overlap in rotations and DCT-IV operations;

Fig. 4a

a schematic block diagram of a known coder with MDCT and 50 percent overlap;

Fig. 4b

a block diagram of a known decoder for decoding the values generated by Fig. 4a;

Fig. 5

a schematic block diagram of a preferred encoder according to the invention;

Fig. 6

a schematic block diagram of an alternative preferred encoder according to the invention;

Fig. 7

a schematic block diagram of a preferred according to the invention decoder;

Fig. 8a

a schematic representation of a bit stream having a first scaling layer and a second scaling layer;

Fig. 8b

a schematic representation of a bit stream with a first scaling layer and a plurality of other scaling layers; and

Fig. 9

a schematic representation of binary coded difference spectral values to illustrate possible scaling in terms of accuracy (bits) of the difference spectral values and / or in terms of the frequency (sampling rate) of the difference spectral values.

In the following, with reference to FIGS. 5 to 7 according to the invention Encoder circuits (FIGS. 5 and 6) or an inventively preferred decoder circuit (Fig. 7). The inventive encoder shown in Fig. 5 includes an input 50 into which a discrete-time Audio signal can be fed, and an output 52, off the encoded audio data can be output. This at the entrance 50 fed discrete-time audio signal is in a device 52 is fed to provide a quantization block, the output side a quantization block of provides discrete-time audio signal using of a psychoacoustic model 54 quantized spectral values the discrete-time audio signal 50 has. The inventive Encoder also includes means for Generating an integer block using an integer transform algorithm 56, where the integer algorithm is effective to discrete integer discrete Samples to generate integer spectral values.

The encoder according to the invention further comprises a device 58 for inverse quantization of the quantization block, issued by the device 52, and, if a different accuracy than the processor accuracy is required, a rounding function. If to the accuracy of the processor system as it has been executed is to be gone, so the rounding function is already inherent in the inverse quantization of the quantization block included as a processor that uses integer arithmetic has, anyway, is unable, non-integer Deliver values. The device 58 provides thus a so-called rounding block, which inverse quantized Includes spectral values that are integer, d. H. inherent or have been rounded explicitly. Both the rounding block as well as the integer block become a combination device fed using a Difference formation a difference block with difference spectral values returns, where the term "difference block" should indicate that the difference spectral values Values are the differences between the Include integer block and the rounding block.

Both the quantization block resulting from the device 52 is output, as well as the difference block, from the Difference generator 58 is output, one Processing device 60 is supplied, the z. Legs usual processing of the quantization block, and the further z. B. an entropy coding of the difference block causes. The device 60 for processing indicates the output 52 coded audio data containing both information about the quantization block as well as information over the difference block.

In a first preferred embodiment, such as it is shown in Fig. 6, the discrete-time audio signal by means of an MDCT converted into its spectral representation and then quantized. The device 52 for delivering the Quantization block thus consists of the MDCT device 52a and a quantizer 52b.

In addition, it is preferred to use the integer block an IntMDCT 56 as an integer transformation algorithm to create.

FIG. 6 also shows the processing device shown in FIG 60 as bit stream encoder 60a for bit stream encoding of the quantization block that passes through the Means 52b, and by an entropy coder 60b for entropy coding the difference block shown. The bit stream encoder 60a outputs the psychoacoustically coded audio data while the entropy coder 60b outputs an entropy-coded difference block. The two output data of blocks 60a and 60b can be suitably combined into a bit stream, as the first scaling layer the psychoacoustically coded audio data, and the second scaling layer the additional audio data for lossless decoding Has. The scaled bitstream then equals the in Fig. 5 shown coded audio data at the output 52 of the encoder.

In an alternative preferred embodiment the MDCT block 52a of FIG. 6 is omitted, as is indicated by a dashed arrow 62 in Fig. 5 is. In this case, the integer spectrum that passes through the integer transformation means 56 is supplied, both fed to the difference generator 58 and in the quantizer 52b of FIG. 6. The spectral values, which are generated by the integer transformation, become here as an approximation for a usual MDCT spectrum used. This embodiment has the advantage that only the IntMDCT algorithm in the encoder is present, and that not both the IntMDCT algorithm as well as the MDCT algorithm present in the encoder have to be.

Referring again to Fig. 6, it should be noted that the solid blocks and lines a conventional audio encoders to represent one of the MPEG standards while the dashed blocks and lines the extension of a represent such conventional MPEG encoder. So it's too see that no fundamental change of the usual MPEG encoder are required, but that the inventive Obtaining the additional audio data for a lossless Encoding using an integer transformation without change be added to the encoder / decoder framework can.

Fig. 7 shows a schematic block diagram of an inventive Decoder for decoding at the output 52 of Fig. 5 output coded audio data. These will be first in psychoacoustically coded audio data on the one hand and the supplemental audio data, on the other hand, decomposed. The psychoacoustic encoded audio data becomes a conventional bit stream decoder 70 while the additional audio data, if they have been entropy-coded in the coder, entropy-decoded by means of an entropy decoder 72 become. At the output of the bit stream decoder 70 of FIG. 7 are quantized spectral values that are inverse Quantizers 74 are supplied, which are identical in principle to the inverse quantizer in the device of FIG. 6 can be constructed. If an accuracy is sought, the does not match the processor accuracy, so is in the decoder a rounding device 76 is also provided, which the same rounding algorithm or the same rounding function to map a real number to an integer, as also implemented in the device 58 of FIG can be. In a decoder-side combiner 78 become the rounded inverse quantized spectral values with the entropy-coded additional audio data spectral valuewise preferably combined in an additive manner, so that in the decoder on the one hand inversely quantized spectral values at the output of the Device 74 are present and secondly integer spectral values present at the output of the combiner 78.

The output spectral values of the device 74 may then by means 80 for performing a inverse modified discrete cosine transformation in the time range are implemented to be a lossy one psychoacoustically coded and decoded audio signal to obtain. By means of a means 82 for performing an inverse integer MDCT (IntMDCT) also becomes the output signal of the combiner 78 in its temporal representation converted to a lossless coded / decoded audio signal or a, if a corresponding coarser rounding has been used, a nearly lossless coded and to generate again decoded audio signal.

In the following, a particular preferred embodiment of the entropy coder 60b of FIG. Having multiple in a conventional modern MPEG encoder Code tables that depend on an average Statistics of quantized spectral values selected are present, it is preferred to use the same code tables or codebooks also for the entropy coding of the difference block at the output of the combiner 58 to use. After the amount of the difference block, ie the residual IntMDCT spectrum, on the accuracy of quantization may be a codebook selection for the entropy coder 60b performed without additional page information become.

In an MPEG-2 AAC encoder, the spectral coefficients are ie the quantized spectral values in the quantization block grouped into scale factor bands, where the Spectral values are weighted by a gain factor, that of a corresponding scale factor, which is a scale factor band is assigned, is derived. Because in this known coder concept an uneven quantizer is used to quantize the weighted spectral values, depends on the size of the residual values, ie the spectral values at the output of combiner 58, not just from the Scale factors, but also from the quantized values However, after both the scale factors and also the quantized spectral values in the bit stream, the is generated by the device 60a of Fig. 6, ie in contained in the psychoacoustically coded audio data, it is preferred to depend on a codebook selection in the coder on the size of the difference spectral values and further in the decoder the codebook used in the coder based on both the bitstream transmitted To determine the scale factors as well as the quantized values. After the entropy coding of the difference spectral values no page information at the output of the combiner 58 must be transmitted, performs the entropy coding only to a data rate compression without any signaling bits in the data stream as page information for the entropy coder 60b would have to be.

In an audio encoder according to the MPEG-2 AAC standard used a window switch to transpose Vorechos into transients Avoid audio signal areas. This technique is based on the possibility of individual window forms in each Half of the MDCT window, and allows you to to vary the block size in successive blocks. Similarly, the integer transform algorithm in the form of IntMDCT, to the reference 1 to 3, with reference to FIGS. to likewise different window forms at Windows and the time-domain aliasing section of the MDCT decomposition to use. It is therefore preferred for both the integer transformation algorithm as well as for the Transformation algorithm for generating the quantization block to use the same window decisions.

In a coder according to MPEG-2 AAC moreover exist several other coding tools, of which only TNS (TNS = Temporal Noise Shaping) and center / side (MS) stereo encoding be mentioned. In a TNS encoding will be the same as with MS coding, a modification of the spectral values performed before quantization. Consequently, take the difference between the IntMDCT values, ie the integer block, and the quantized MDCT values. According to the invention is the integer transformation algorithm designed around both TNS coding and center / page coding also allow integer spectral values. The TNS technique is based on adaptive forward prediction the MDCT values over the frequency. The same thing Prediction filter that is signal adaptive from a standard TNS module is calculated, is preferably also to used to predict the integer spectral values, where if this results in non-integer values, one Downstream rounding can be used to get back to generate integer values. This rounding is preferably done after each prediction step. In the decoder the original spectrum can be reconstructed again, by the inverse filter and the same rounding function be used. In a similar way, the MS encoding also applied to IntMDCT spectral values be done by making rounded Givens rotations at an angle of π / 4, based on the lifting scheme. This allows the original IntMDCT values in the Decoders are reconstructed again.

It should be noted that the inventive concept in its preferred form with the IntMDCT as integer Transformation algorithm on all MDCT-based audio-coded encoder. Only By way of example, such coders are coders MPEG-4 AAC Scalable, MPEG-4 AAC Low Delay, MPEG-4 BSAC, MPEG-4 Twin VQ, Dolby AC-3 etc.

It should be particularly noted that the inventive Concept is backwards compatible. The hearse fitted Encoder or decoder is not changed, only extended. Additional information for the lossless components can be backward compatible in listening coded bitstream transmitted, for example, MPEG-2 AAC in the field "Ancillary Data". The addition to the previous one Hearing-matched decoder, which is shown in dashed lines in Fig. 7 is, these additional data can evaluate and together with the quantized MDCT spectrum from the listener-matched decoder reconstruct the IntMDCT spectrum without loss.

The inventive concept of psychoacoustic coding, supplemented by a lossless or almost lossless Coding is particularly suitable for generation, transmission and decoding scalable data streams. It is It is known that scalable data streams have different scaling layers include, at least the lowest Scaling layer independent of the higher scaling layers can be transmitted and decoded. Further Scaling layers or enhancement layers are used in a Scalable processing of data from the first scaling layer or base layer added. A fully equipped Encoder can produce a scaled data stream which has a first scaling layer and which in principle Any number of other scaling layers has. An advantage of the scaling concept is that in In the case where a broadband transmission channel for Available is the scaled data stream generated by the encoder complete, including all scaling layers be transmitted over the broadband transmission channel can. On the other hand, it is only a narrowband transmission channel present, so the coded signal can nevertheless transmitted over the transmission channel, but only in the form of the first scaling layer or a specific one Number of further scaling layers, where the certain number smaller than the whole of the encoder generated number of scaling layers. Of course can the encoder already adapted to the channel, to which he is connected, the basic scaling layer or first scaling layer and one dependent on the channel Create number of further scaling layers.

On the decoder side, the scalable concept also has the Advantage that it is backwards compatible. This means, a decoder that is only capable of the first scaling layer to process, just the second and more Scaling layers in the data stream are ignored and on can produce useful output signal. Is the decoder on the other hand, a typically more modern decoder that has multiple Scaling layers from the scaled data stream can process this encoder with the same data stream how a basic decoder is addressed.

In the present invention, the basic Scalability in that the quantization block, so the Output of the bitstream encoder 60a into a first scaling layer 81 of FIG. 8 which, when FIG. 6, psychoacoustically encoded data e.g. For includes a frame. The preferably entropy-coded Difference spectral values obtained by the combination device 58 are created with easy scalability written in the second scaling layer shown in FIG. 8a is designated 82 and thus the additional audio data for includes a frame.

Is the transmission channel from the encoder to the decoder on broadband transmission channel, so both scaling layers 81 and 82 are transmitted to the decoder. However, if the transmission channel is a narrow-band transmission channel, in the only the first scaling layer "fits", so just imagine the second scaling layer the transmission are removed from the data stream, so that a Decoder addressed only with the first scaling layer becomes.

On the decoder side can be a "basic decoder", which only can process the psychoacoustically encoded data that simply omit the second scaling layer 82 if it does receive them over a broadband transmission channel Has. However, if the decoder is a full-featured decoder, Both a psychoacoustic decoding algorithm and an integer decoding algorithm, so this fully featured decoder can handle both the first scaling layer as well as the second scaling layer to decode to a lossless coded and to generate decoded output again.

In a preferred embodiment of the present invention Invention, as shown schematically in Fig. 8a is again psychoacoustically in a first scaling layer encoded data for one frame. The second scaling layer of Fig. 8a is now scaled finer, so that from this second scaling layer in FIG. 8a, several scaling layers arise, such as e.g. a (smaller) second scaling layer, a third scaling layer, a fourth scaling layer, etc.

The difference spectral values output by the adder 58 are particularly good for another Sub-scaling as illustrated with reference to FIG becomes. Fig. 9 schematically illustrates binary coded spectral values Each line 90 in FIG. 9 represents a binary coded one Difference spectral value. In Fig. 9 are the difference spectral values sorted by the frequency as it passes through an arrow 91 is indicated. So has a difference spectral value 92 is a higher frequency than the difference spectral value 90. The first column of the panel in FIG. 9 represents the most significant bit of a difference spectral value in front. The second digit represents the bit with the significance MSB-1. The third column represents a bit of significance MSB-2. The third to last column is one bit valence LSB + 2. The penultimate column is set Finally, the bit with the significance LSB + 1 represents last column one bit with the significance LSB, ie the least significant one Bit of a difference spectral value.

In a preferred embodiment of the present invention, precision scaling is performed by taking, for example, the 16 most significant bits of a difference spectral value as the second scaling layer to then be entropy coded by the entropy coder 60b, if desired. A decoder using the second scaling layer receives on the output side difference spectral values with an accuracy of 16 bits, so that the second scaling layer together with the first scaling layer delivers a losslessly decoded CD-quality audio signal. It is known that CD-quality audio samples have a width of 16 bits.
On the other hand, if a studio quality audio signal is input to the encoder, that is, an audio sample having samples, each sample comprising 24 bits, the encoder may further generate a third scale layer comprising the last eight bits of a difference spectral value and also entropy as needed is encoded (means 60 of Fig. 6).

A full-featured decoder that uses the data stream the first scaling layer, the second scaling layer (16 most significant bits of the difference spectral values) and the third scaling layer (8 least significant bits a difference spectral value), so the decoder using all three scale layers lossless coded / decoded audio signal in studio quality, So with one present at the output of the decoder Word width of a sample of 24 bits.

It should be noted that higher in the studio area Word lengths of the samples are common than in the consumer sector. In the consumer sector, the word width is 16 bits in an audio CD, while in the studio area 24 bit or 20 Bit be used.

Based on the concept of scaling in the IntMDCT range Thus, as has been shown, all three accuracies (16-bit, 20-bit or 24-bit) or any at minimum 1-bit scaled accuracies are scalable scalable.

This will be the audio signal presented with 24-bit accuracy with the help of the inverse IntMDCT in integer Spectral range shown and fitted with an ear MDCT-based audio encoder output scalable combined.

The integer difference values, the lossless representation are present, are now not completely in a scaling layer coded, but first with lesser Accuracy. Only in another scaling layer become the residual values necessary for the exact representation transfer. Alternatively, however, could also be in another Scaling layer a difference spectral value total, So with 24 bits, for example, be represented, so that for decoding this further scaling layer underlying scaling layer is not needed. However, this scenario leads to a higher overall Bitstream size, but may be when the bandwidth of the Transmission channel is unproblematic, to a simplification in the decoder, because in the decoder then not more scaling layers must be combined, but always only one scaling layer is sufficient for decoding.

If, for example, the lower eight LSBs, as shown in FIG. 9 is shown, not initially transmitted, so achieved one scalability between 24 bits and 16 bits.

For retransforming the lower-precision transmitted values into the time domain, the transmitted values are preferably scaled back to the original range, for example 24 bits, by multiplying them by 2 ^8, for example. An inverse IntMDCT is then applied to the corresponding scaled-back values.

In the inventive accuracy scaling in the frequency domain it is also preferred, the redundancy in the LSBs. Has an audio signal, for example in the upper frequency range very little energy, so expresses this also applies to the IntMDCT spectrum in very small values, for example, much smaller than the z. B. at 8 bits are possible values (-128, ..., 127). This expresses in compressibility of the LSB values of the IntMDCT spectrum. It should also be noted that at very small difference spectral values typically a number of bits from MSB to MSB-n are equal to zero, and then that first with a bit with a weight MSB-n-1 the first, leading 1 in a binary coded difference spectral value occurs. In such a case, if a difference spectral value only zeros in the second scaling layer includes an entropy coding is suitable for further Data compression especially good.

According to another embodiment of the present invention Invention is for the second scaling layer 82 of Fig. 8a prefers sample rate scalability. A sample rate scalability is achieved by that in the second scaling layer, as shown in Fig. 9 right is, the difference spectral values up to a first Limit frequency are included, while in a third scaling layer the difference spectral values with one frequency between the first cutoff frequency and the maximum Frequency are included. Of course, another can also be Scaling be done so that out of the total Frequency range multiple scaling layers are made.

In a preferred embodiment of the present invention The invention includes the second scaling layer in FIG. 9 Differential spectral values up to a frequency of 24 kHz, which corresponds to a sampling rate of 48 kHz. The third scaling layer then contains the difference spectral values from 24 kHz to 48 kHz, which corresponds to a sampling rate of 96 kHz.

It should also be noted that in the second scaling layer and the third scale layer not necessarily all bits of a difference spectral value must be coded. So could in one another form of combined scalability the second Scaling layer, the bits MSB to MSB-x of the difference spectral values up to a certain cutoff frequency. A third scaling layer could then be the bits MSB to MSB-x of the difference spectral values from the first one Limit frequency up to the maximum frequency include. A fourth scaling layer could then be the remaining bits for the difference spectral values up to the cutoff frequency. The last scaling layer could then be the rest Bits of the difference spectral values for the upper frequencies include. This concept becomes a division of the panel in Fig. 9 in four quadrants, each Quadrant represents a scaling layer.

In the frequency scalability is at a preferred Embodiment of the present invention a scalability between 48 kHz and 96 kHz sampling rate described. The 96 kHz sample signal is in the IntMDCT range in the lossless extension layer first only up to half coded and transmitted. If the upper Part is not transmitted in addition, it is in the decoder assumed to be zero. In the case of the inverse IntMDCT (same Length as in the encoder) then creates a 96 kHz signal, the in the upper frequency range contains no energy and therefore be scanned to 48 kHz without loss of quality can.

The above scaling of the difference spectral values in Quadrant of Fig. 9 with fixed limits is in terms the size of the scaling layers low, because in one Scaling layer actually only z. B. 16 bits or 8 bits or the spectral values up to the cutoff frequency or must be included over the cutoff frequency.

An alternative scaling is the quadrant boundaries in Fig. 9 in a sense "soften". Exemplary frequency scalability would mean that, not one so-called "brickwall low-pass filter" to apply, that the difference spectral values before a cutoff frequency unchanged are equal to zero after the cutoff frequency. Instead, the difference spectral values could also be used with be filtered by any low-pass filter that the Spectral values below the cutoff frequency already somewhat impaired, but above the cutoff frequency leads to that energy is still present here as well, although the Differential spectral values are decreasing in terms of energy. In a then generated scaling layer are also contain spectral values above the cutoff frequency. There however, these spectral values are relatively small, they are efficiently encoded by entropy coding. The highest scaling layer would have the difference in this case between the complete difference spectral values and the spectral values contained in the second scaling layer.

The accuracy scaling can be similar in a sense be softened. So can the first scaling layer also spectral values with z. B. have more than 16 bits, where next scaling layer then has the difference. Generally speaking, the second scaling layer thus the difference spectral values with lower accuracy, while in the next scaling layer the rest, so the difference between the complete spectral values and the spectral values contained in the second scaling layer is transmitted. This is a variable accuracy reduction reached.

The inventive method for encoding or decoding is preferably on a digital storage medium such. As a floppy disk, with electronically readable control signals stored, the control signals so with a programmable computer system can work together, that the coding and / or decoding method are carried out can / can. In other words, that is a computer program product with on a machine readable Carrier stored program code for performing the Coding method and / or the decoding method, if the program product runs on a computer. The invention So procedures can be in a computer program with a program code for carrying out the invention Procedure if the program is on a computer expires, be realized.

The following is an example of an integer Transformation algorithm to the IntMDCT transformation algorithm entered in "Audio Coding Based on Integer Transforms' 111th AES Assembly, New York, 2001, is described. The IntMDCT is special cheap because it has the attractive features of MDCT, such as B. a good spectral representation of the audio signal, has a critical scan and a block overlap. The good approximation of MDCT by an IntMDCT further allows, in the encoder shown in Fig. 5 only to use a transformation algorithm as determined by an arrow 62 in Fig. 5 is shown. With reference to FIG. 1 to 4 are the essential characteristics of this particular Form of an integer transformation algorithm explained.

Fig. 1 shows an overview diagram for the invention preferred device for processing discrete-time Samples that represent an audio signal to integer ones Obtaining values based on the int MDCT integer transform algorithm is working. The discrete-time ones Samples are passed through the device shown in FIG fenestrated and optionally in a spectral representation implemented. The discrete-time samples that are connected to a Input 10 are fed into the device with a window w of length 2N discrete time Samples corresponds windowed to an output 12th to achieve integer windowed samples which are suitable to, by means of a transformation and in particular, the device 14 for executing an integer DCT converted into a spectral representation too become. The integer DCT is designed to consist of N input values N output values to produce what, in contrast to the MDCT function 408 of Fig. 4a, which is windowed from 2N Samples due to the MDCT equation only N spectral values generated.

For windowing the discrete-time samples are first in a device 16, two discrete-time samples are selected, together form a vector of time-discrete samples represent. A time-discrete sample, the is selected by means 16, lies in the first Quarter of the window. The other discrete-time sample lies in the second quarter of the window as it is based of Fig. 3 is executed in more detail. The through the device 16 generated vector is now with a Rotary matrix of dimension 2 x 2 applied, this Operation is not performed immediately, but by means of several so-called lifting matrices.

A lifting matrix has the property of being only one Element, which depends on the window w and unequal Is "1" or "0".

Factorization of wavelet transforms in lifting steps is presented in the technical publication "Factoring Wavlet Transforms Into Lifting Steps", Ingrid Daubechies and Wim Sweldens, Preprint, Bell Laboratories, Lucent Technologies, 1996. Generally, a lifting scheme is a simple relationship between perfectly reconstructive filter pairs having the same low pass or high pass filter. Each pair of complementary filters can be factored into lifting steps. This applies in particular to Givens Rotations. Consider the case where the polyphase matrix is a Givens rotation. It then applies:

Each of the three lifting matrices to the right of the equals sign has the value "1" as main diagonal elements. Furthermore, in each lifting matrix is a secondary diagonal element equal to 0, and a minor diagonal element from the rotational angle α dependent.

The vector is now filled with the third lifting matrix, i. H. the lifting matrix on the far right in the above equation, multiplied, to get a first result vector. This is illustrated in FIG. 1 by a device 18. It is now the first result vector with any Rounding function, which is the set of real numbers in the set of integer numbers, rounded, as it is in Fig. 1 is shown by a device 20. At the exit The device 20 becomes a rounded first result vector receive. The rounded first result vector becomes now in a device 22 for multiplying the same with the middle, d. H. second, fed to the lifting matrix, to obtain a second result vector which is in a device 24 is in turn rounded to a rounded to obtain the second result vector. The rounded one second result vector is now in a device 26th fed, to multiply the same with the listed on the left in the above equation, d. H. first, Lifting matrix to obtain a third result vector finally rounded by means 28 is finally fenced at the output 12 integer Get samples, which now, if one spectral representation of the same is desired by the Device 14 needs to be processed to work on one Spectral output 30 to obtain integer spectral values.

Preferably, the device 14 is an integer DCT or Integer DCT executed.

The Discrete Cosine Transform of Type 4 (DCT-IV) of length N is given by the following equation:

The coefficients of the DCT-IV form an orthonormal N x N Matrix. Each orthogonal N x N matrix can be transformed into N (N-1) / 2 Givens rotations be decomposed, as in the technical publication P. P. Vaidyanathan, "Multirate Systems And Filter Banks ", Prentice Hall, Englewood Cliffs, 1993 is. It should be noted that also further decompositions exist.

Regarding the classifications of the different DCT algorithms to H. S. Malvar, "Signal Processing With Lapped Transforms, "Artech House, 1992. General the DCT algorithms differ by type their basic functions. While the DCT-IV, which prefers here is comprised of non-symmetric basis functions, i. H. a cosine quarter-wave, a cosine 3/4 wave, a cosine 5/4 wave, a cosine 7/4 wave, etc., has the discrete one Cosine transformation z. B. type II (DCT-II), axisymmetric and point-symmetric basis functions. The 0th basic function has a DC component, the first basic function is a half cosine wave, the second basis function is a whole cosine wave, etc. Due to the fact that the DCT-II takes the DC component into account, it is used in video coding, but not in audio coding, as opposed to audio coding for video coding the DC component is not relevant is.

In the following will be discussed how the angle of rotation α of the Givens rotation depends on the window function.

An MDCT with a window length of 2N can be reduced into a discrete cosine transformation of type IV with a length N. This is accomplished by explicitly performing the TDAC operation in the time domain and then applying the DCT-IV. For a 50% overlap, the left half of the window for a block t overlaps the right half of the previous block, ie, the block t-1. The overlapping part of two consecutive blocks t-1 and t is preprocessed in the time domain, ie before the transformation, as follows, ie processed between the input 10 and the output 12 of FIG. 1:

The values denoted by the tilde are the values at the output 12 of FIG. 1, while those without tilde in the above Equation x values denote the values at input 10 or behind the device 16 for selection. The running index k runs from 0 to N / 2-1, while w represents the window function.

From the TDAC condition for the window function w, the following relationship applies:

For certain angle α _k, k = 0, ..., N / 2-1, this preprocessing may be written in the time domain as a Givens rotation, as it has been executed.

The angle α of the Givens rotation depends on the window function w as follows: □ = arctan [w (N / 2-1-k) / w (N / 2 + k)]

It should be noted that any window functions w can be used as long as they have this TDAC condition fulfill.

In the following, with reference to FIG. 2, a cascaded Encoder and decoder described. The discrete-time samples x (0) to x (2N-1), which share a window "windowed", are initially so through the device 16 of FIG. 1, that the sample x (0) and the sample x (N-1), d. H. a sample from the first quarter of the window and a sample from the second quarter of the window, be selected to the Vector at the output of the device 16 to form. Which Crossing arrows represent schematically the lifting multiplications and subsequent rounding of the facilities 18, 20 and 22, 24 and 26, 28, respectively, at the entrance the DCT-IV blocks the integer windowed samples to obtain.

If the first vector is processed as described above, Furthermore, a second vector of the samples x (N / 2-1) and x (N / 2), d. H. again a sample from the first one Quarter of the window and one sample from the second Quarter of the window, selected and in turn by the in Fig. 1 described algorithm processed. Similarly all other pairs of samples from the first and second quarter of the window worked. The same processing will be for the third and fourth quarters of the first Window performed. Now there are 2N windowed windows at the exit 12 integer samples that now look like it is shown in Fig. 2, in a DCT-IV transformation be fed. In particular, the integer ones windowed samples of the second and third quarters fed into a DCT. The windowed integer samples the first quarter of the window will be in one precedent DCT-IV along with the windowed integer ones Samples of the fourth quarter of the previous one Window processed. Similarly, the fourth quarter the windowed integer samples in Fig. 2 with the first quarter of the next window together into a DCT-IV transformation fed. The middle one shown in FIG integer DCT-IV transformation 32 now provides N integer spectral values y (0) to y (N-1). This integer Spectral values can now be simple, for example Entropy-coded without any intervening Quantization is required because the fenestration and Transformation provides integer output values.

In the right half of Fig. 2, a decoder is shown. The decoder consisting of inverse transformation and "inverse windowing" works inverse to the encoder. It is it is known that an inverse of the reverse transformation of a DCT-IV DCT-IV can be used as shown in FIG is. The output values of the decoder DCT-IV 34 become now, as shown in Fig. 2, with the corresponding Values of the previous transformation or the subsequent transformation is inversely processed to the integer windowed samples at the output of the Device 34 or the preceding and following Transformation again discrete-time audio samples x (0) to produce x (2N-1).

The output side operation is done by inverse Givens rotation, ie, such that the blocks 26, 28, 22, 24 and 18, 20, respectively, are traversed in the opposite direction. This is illustrated in more detail with reference to the second lifting matrix of Equation 1. If (in the encoder) the second result vector is formed by multiplying the rounded first result vector by the second lifting matrix (means 22), the following expression results: ( x . y ) ↦ ( x . y + x sin α )

The values x, y on the right side of Equation 6 are integers. However, this does not apply to the value x sin α. Here, the rounding function r must be introduced, as in the following equation ( x . y ) ↦ ( x . y + r ( x sin .alpha)) is shown. This operation executes the device 24.

The inverse mapping (in the decoder) is defined as follows: ( x ' y ') ↦ ( x ' y '- r ( x 'Sin .alpha))

Due to the minus sign before the rounding operation becomes It can be seen that the integer approximation of the lifting step can be reversed without an error is introduced. The application of this approximation to each The three lifting steps leads to an integer Approximation of the Givens rotation. The rounded rotation (in the encoder) can be reversed (in the decoder), without that an error is introduced by the inverse ones rounded lifting steps in reverse order to go through, d. H. when decoding the algorithm of Fig. 1 is carried out from bottom to top.

If the rounding function r is point symmetric, then the inverse rounded rotation is identical to the rounded rotation with the angle -α and reads as follows:

The lifting matrices for the decoder, d. H. for the inverse Givens rotation, arises in this case immediately from equation (1), merely using the expression "sin" is replaced by the term "-sin α".

In the following, with reference to FIG. 3 again Disassembly of a standard MDCT with overlapping windows 40 to 46. The windows 40 to 46 each overlap to 50%. For each window, Givens rotations are created within the first and second quarters of a window or within the third and fourth quarters of a window executed, as shown schematically by the arrows 48 is. Then the rotated values, i. H. the windowed ones integer samples such as in an N-to-N DCT fed that always the second and third quarters a window or the fourth and first quarters of a subsequent Window together using a DCT-IV algorithm is converted into a spectral representation.

It therefore becomes the usual Givens rotation in lifting matrices decomposed, which are executed sequentially, where after each lifting matrix multiplication a rounding step is introduced, such that the floating point numbers be rounded immediately after their formation, that before each multiplication of a result vector with a Lifting matrix the result vector only integers Has.

The output values therefore always remain integer, whereby it it is preferred to also use integer input values. This is not a limitation, as any example PCM samples as stored on a CD are, are integer numerical values whose value range varies depending on the bit width, d. H. depending on whether the time-discrete digital input values 16-bit values or 24-bit values are. Still, that's how it's been done is, the entire process is invertible by the inverse Rotations are performed in reverse order. It Thus there exists an integer approximation of the MDCT perfect reconstruction, ie a lossless transformation.

The transformation shown provides integer output values instead of floating-point values. It provides a perfect reconstruction, so no mistake is introduced if one Forward and then a reverse transformation executed become. The transformation is according to a preferred embodiment the present invention is a substitute for the modified discrete cosine transformation. Others too However, transformation techniques can be performed in integers be as long as a decomposition into rotations and one Disassembly of the rotations in lifting steps possible is.

The integer MDCT has the most favorable features the MDCT. It has an overlapping structure, which means a better frequency selectivity than non-overlapping ones Block transformations is obtained. by virtue of the TDAC function already in the window before the transformation is taken into account, becomes a critical scan maintained so that the total number of spectral values, which represent an audio signal, equal to the total number of input samples.

Compared to a normal MDCT, the floating-point samples shows, shows in the described preferred integer transformation that only in the Spectral range in which there is little signal level, the noise compared to normal MDCT is increased while up this Rauscherhöhung not at significant signal levels makes noticeable. This is the integer processing for an efficient hardware implementation, since only multiplication steps are used without more in move-add steps (shift / add steps) can be disassembled, which is easy and fast can be implemented in hardware. Of course is also a software implementation possible.

The integer transformation provides a good spectral Presentation of the audio signal and still remains in the range the whole numbers. When referring to tonal parts of an audio signal is applied, this results in a good E-nergiekonzentrierung. This can be an efficient lossless Coding scheme can be constructed by simply placing the in Fig. 1 shown fenestration / transformation with a Entropy coder is cascaded. In particular, a stacked Coding (Stacked Coding) Using Escape Values, as it is used in MPEG AAC, is favorable. It is preferable to set all values by a certain power of Shorten two until they are in a desired code table fit, and then the omitted low-order ones Code bits in addition. Compared to the alternative the use of larger code tables is the one described Alternative in terms of memory consumption to save the code tables cheaper. An almost lossless Encoder could also be obtained by: simply omit certain of the least significant bits become.

In particular for tonal signals allows entropy coding the integer spectral values have a high coding gain. For transient parts of the signal is the Coergewinn low, due to the flat spectrum transient signals, d. H. due to a small number spectral values that are equal or close to zero. Like it in J. Herre, J.D. Johnston: "Enhancing the Performance of Perceptual Audio Coders by Using Temporal Noise Shaping (TNS) "101st AES Convention, Los Angeles, 1996, Preprint 4384, however, this flatness can be used be, by a linear prediction in the frequency domain is used. An alternative is a prediction with open loop. Another alternative is the predictor with closed loop. The first alternative, d. H. the open-loop predictor is called TNS. The quantization after the prediction leads to an adaptation of the resulting quantization noise to the temporal structure of the audio signal and therefore prevents Vorechos in psychoacoustic audio encoders. For a lossless Audio coding is the second alternative, d. H. with a closed loop predictor, more appropriate, because closed-loop prediction is accurate Reconstruction of the input signal allowed. If those Technique applied to a generated spectrum must be Rounding step after each step of the prediction filter be performed to stay in the range of integers. By using the inverse filter and the same Rounding function can accurately restore the original spectrum getting produced.

To the redundancy between two channels for data reduction can also exploit a mid-page coding lossless be used when a rounded rotation with an angle □ / 4 is used. Compared to the alternative of calculating the sum and difference of the left and right channel of a stereo signal has the rounded rotation the advantage of energy conservation. The use of so-called Joint-stereo coding techniques can be used for each band be turned on or off, as it is in the standard MPEG AAC is performed. Other angles of rotation can also be taken into account for a redundancy between two channels more flexible to reduce.

Claims (33)

1. Apparatus for coding a time-discrete audio signal to obtain coded audio data, comprising:

   means (52) for providing a quantization block of spectral values of the time-discrete audio signal quantized using a psychoacoustic model (54);

   means (58) for inversely quantizing the quantization block and for rounding the inversely quantized spectral values to obtain a rounding block of rounded inversely quantized spectral values;

   means (56) for generating an integer block of integer spectral values using an integer transform algorithm formed to generate the integer block of spectral values from a block of integer time-discrete samples;

   combination means (58) for forming a difference block depending on a spectral value-wise difference between the rounding block and the integer block, to obtain a difference block with difference spectral values; and

   means (60) for processing the quantization block and the difference block to generate coded audio data including information on the quantization block and information on the difference block.
2. Apparatus of claim 1, wherein means (52) for providing is formed to
   generate a MDCT block of MDCT spectral values from a time block of temporal audio signal values by means of an MDCT, and
   quantize the MDCT block using a psychoacoustic model to generate the quantization block comprising quantized MDCT spectral values.
3. Apparatus of claim 2,
   wherein the means (56) for generating the integer block is formed to execute an IntMDCT on the time block to generate the integer block comprising IntMDCT spectral values.
4. Apparatus of one of the preceding claims,
   wherein the means (52) for providing is formed to calculate the quantization block using a floating-point transform algorithm.
5. Apparatus of one of claims 1 to 3,
   wherein the means (52) for providing is formed to calculate the quantization block using the integer block generated by means (56) for generating.
6. Apparatus of one of the preceding claims,
   wherein the means (60) for processing is formed to subject (60a) the quantization block to entropy coding, to obtain an entropy-coded quantization block,
   to subject (60b) the rounding block to entropy coding, to obtain an entropy-coded rounding block, and
   to convert the entropy-coded quantization block to a first scaling layer of a scaled data stream representing the coded audio data, and to convert the entropy-coded rounding block to a second scaling layer of the scaled data stream.
7. Apparatus of claim 6,
   wherein the means (60) for processing is further formed to use one of the plurality of code tables depending on the quantized spectral values for the entropy coding of the quantization block, and
   wherein the means (60) for processing is further formed to select one of a plurality of code tables depending on a property of a quantizer usable in a quantization for generating the quantization block for the entropy coding of the difference block.
8. Apparatus of one of the preceding claims,
   wherein the means (52) for providing is formed to use one of a plurality of windows for windowing a temporal block of audio signal values depending on a property of the audio signal, and
   wherein the means (56) for generating is formed to make the same window selection for the integer transform algorithm.
9. Apparatus of one of claims 1 to 8,
   wherein the means for generating is formed to use an integer transform algorithm, comprising:

   windowing the time-discrete samples with a window (w) with a length corresponding to 2N time-discrete samples, to provide windowed time-discrete samples for a conversion of the time-discrete samples to a spectral representation by means of a transform capable of generating N output values from N input values, wherein the windowing comprises the following substeps:

   selecting (16) a time-discrete sample from a quarter of the window and a time-discrete sample from another quarter of the window to obtain a vector of time-discrete samples;

   applying a square rotation matrix the dimension of which matches the dimension of the vector to the vector, wherein the rotation matrix is representable by a plurality of lifting matrices, wherein a lifting matrix only comprises one element dependent on the window (w) and being unequal to 1 or 0, wherein the substep of applying comprises the following substeps:

   multiplying (18) the vector by a lifting matrix to obtain a first result vector;

   rounding (20) a component of the first result vector with a rounding function (r) mapping a real number to an integer to obtain a rounded first result vector; and

   sequentially performing the steps of multiplying (22) and rounding (24) with another lifting matrix, until all lifting matrices are processed, to obtain a rotated vector comprising an integer windowed sample from the quarter of the window and an integer windowed sample from the other quarter of the window, and

   performing the step of windowing for all time-discrete samples of the remaining quarters of the window to obtain 2N filtered integer values; and

   converting (14) N windowed integer samples to a spectral representation by an integer DCT for values with the filtered integer samples of the second quarter and the third quarter of the window, to obtain N integer spectral values.
10. Apparatus of one of the preceding claims,
    wherein the means (52) for providing the quantization block is formed to perform a prediction of spectral values over the frequency using a prediction filter prior to a quantization step (52b), to obtain prediction residual spectral values representing the quantization block after a quantization;
    wherein also a prediction means is provided, which is formed to perform a prediction over the frequency of the integer spectral values of the integer block, wherein also rounding means is provided to round prediction residual spectral values due to the integer spectral values representing the rounding block.
11. Apparatus of one of the preceding claims,
    wherein the time-discrete audio signal comprises at least two channels,
    wherein the means (52) for providing is formed to perform center/side coding with spectral values of the time-discrete audio signal to obtain the quantization block after quantization of center/side spectral values, and
    wherein the means (56) for generating the integer block is formed to also perform center/side coding corresponding to the center/side coding of the means (52) for providing.
12. Apparatus of one of the preceding claims,
    wherein the means (60) for processing is formed to generate a MPEG-2 AAC data stream, wherein in a field Ancillary Data ancillary information for the integer transform algorithm is introduced.
13. Apparatus of one of the preceding claims,
    wherein the means for processing (60) is formed to output the coded audio data as data stream with a plurality of scaling layers.
14. Apparatus of claim 13,
    wherein the means for processing (60) is formed to insert information on the quantization block into a first scaling layer (81), and to insert information on the difference block into a second scaling layer (82).
15. Apparatus of claim 13,
    wherein the means for processing (60) is formed to insert information on the quantization block into a first scaling layer, and to insert the information on the difference block into at least a second and a third scaling layer.
16. Apparatus of claim 15,
    wherein in the second scaling layer difference spectral values with reduced accuracy are contained, and in one or more higher scaling layers a residual part of the difference spectral values is contained.
17. Apparatus of claim 15 or 16,
    wherein the information on the difference block includes binarily coded difference spectral values,
    wherein the second scaling layer for difference spectral values includes a number of bits from a most significant bit (MSB) to a less significant bit (MSB-x) for a difference spectral value, and
    wherein the third scaling layer includes a number of bits starting from a less significant bit (MSB-x-1) to a least significant bit (LSB).
18. Apparatus of claim 17,
    wherein the time-discrete audio signal is present in form of samples with a width of 24 bits, and
    wherein the means for processing (60) is formed to insert more significant 16 bits of difference spectral values into the second scaling layer, and to insert residual 8 bits of a difference spectral value into the third scaling layer, so that a decoder reaches CD quality using the second scaling layer, wherein a decoder reaches studio quality using also the third scaling layer.
19. Apparatus of claim 15,
    wherein the means (60) for processing is formed to insert at least part of difference spectral values for representation of a low-pass filtered signal into a second scaling layer, and to insert a difference between the difference spectral values in the second scaling layer and original difference spectral values into at least one further scaling layer.
20. Apparatus of claim 15 or 19,
    wherein the means (60) for processing is formed to insert at least part of different spectral values up to a certain cut-off frequency into a second scaling layer, and to insert at least part of difference spectral values from the certain cut-off frequency to a higher frequency into a third scaling layer.
21. Method of coding a time-discrete audio signal to obtain coded audio data, comprising:

    providing (52) a quantization block of spectral values of a time-discrete audio signal quantized using a psychoacoustic model (54);

    inversely quantizing (58) the quantization block and rounding the inversely quantized spectral values to obtain a rounding block of rounded inversely quantized spectral values;

    generating (56) an integer block of integer spectral values using an integer transform algorithm formed to generate the integer block of spectral values from a block of integer time-discrete samples;

    forming (58) a difference block depending on a spectral value-wise difference between the rounding block and the integer block, to obtain a difference block with difference spectral values; and

    processing (60) the quantization block and the difference block to generate coded audio data including information on the quantization block and information on the difference block.
22. Apparatus for decoding coded audio data having been generated from a time-discrete audio signal by providing (52) a quantization block of spectral values of the time-discrete audio signal quantized using a psychoacoustic model (54), by inversely quantizing (58) the quantization block and rounding the inversely quantized spectral values to obtain a rounding block of rounded inversely quantized spectral values, by generating (56) of an integer block of integer spectral values using an integer transform algorithm formed to generate the integer block of spectral values from a block of integer time-discrete samples, and by forming (58) a difference block depending on a spectral value-wise difference between the rounding block and the integer block, to obtain a difference block with difference spectral values, comprising:

    means (70) for processing the coded audio data to obtain a quantization block and a difference block;

    means (74) for inversely quantizing and rounding the quantization block to obtain an integer inversely quantized quantization block;

    means (78) for spectral value-wise combining the integer quantization block and the difference block to obtain a combination block; and

    means (82) for generating a temporal representation of the time-discrete audio signal using the combination block and using an integer transform algorithm inverse to the integer transform algorithm.
23. Apparatus for decoding of claim 22,
    wherein the coded audio data is scaled and includes a plurality of scaling layers,
    wherein the means (70) for processing the coded audio data is formed to ascertain the quantization block from the coded audio data as first scaling layer, and to ascertain the difference block from the coded audio data as second scaling layer.
24. Apparatus of claim 22,
    wherein the information on the difference block includes binarily coded difference spectral values,
    wherein the coded audio data is scaled and includes a plurality of scaling layers,
    wherein the means (70) for processing the coded audio data is formed to ascertain the quantization block from the coded audio data as first scaling layer, and to extract a representation of the difference spectral values with reduced accuracy as second scaling layer.
25. Apparatus of claim 24,
    wherein the means (70) for processing the coded audio data is formed to extract a number of bits starting from a most significant bit to a less significant bit, which is more significant than a least significant bit of a difference spectral value, as second scaling layer, and
    wherein the means (82) for generating a temporal representation of the time-discrete audio signal is formed to synthetically generate missing bits for a difference spectral value before using the integer transform algorithm.
26. Apparatus of claim 25,
    wherein the means (82) is formed to perform an upscaling of the second scaling layer for the synthetical generation, wherein in the upscaling a scale factor is used, which equals 2ⁿ, wherein n is the number of less significant bits not contained in the second scaling layer, or to employ a dithering algorithm for the synthetical generation.
27. Apparatus of claim 22,
    wherein the coded audio data is scaled and includes a plurality of scaling layers, and
    wherein the means (70) for processing coded audio data is formed to ascertain the quantization block from the coded audio data as first scaling layer, and to ascertain low-pass filtered difference spectral values as second scaling layer.
28. Apparatus of claim 22 or 27,
    wherein the coded audio data is scaled and includes a plurality of scaling layers, and
    wherein the means (70) for processing the coded audio data is formed to ascertain the quantization block of the coded data as first scaling layer, and to ascertain difference spectral values up to a first cut-off frequency as second scaling layer, wherein the first cut-off frequency is smaller than the maximum frequency of a difference spectral value, which may be generated in a coder.
29. Apparatus of claim 28,
    wherein the means (82) for generating a temporal representation is formed to set input values in an integer transform algorithm of full length, which are above the cut-off frequency of the second scaling layer, to a predetermined value, and to downsample the temporal representation of the time-discrete audio signal after using the inverse integer transform algorithm by a factor chosen corresponding to a ratio of a maximum frequency of a difference spectral value, which may be generated by a coder, and the cut-off frequency.
30. Apparatus of claim 29,
    wherein the predetermined value for all input values above the cut-off frequency is zero.
31. Method of decoding coded audio data having been generated from a time-discrete audio signal by providing, inversely quantizing, generating, forming, and processing, comprising:

    processing (70) the coded audio data to obtain a quantization block and a difference block;

    inversely quantizing (74) the quantization block and rounding to obtain an integer inversely quantized quantization block;

    spectral value-wise combining (78) the integer quantization block and the difference block to obtain a combination block; and

    generating (82) a temporal representation of the time-discrete audio signal using a combination block and using an integer transform algorithm inverse to the integer transformation algorithm.
32. Computer program with a program code for performing the method of coding of claim 21, when the program is executed on a computer.
33. Computer program with a program code for performing the method of decoding of claim 31, when the program is executed on a computer.

Images