DE69532932T2 - Method for non-linear quantization of an information signal - Google Patents

Description

Territory of invention

The The invention relates to signal compression in which an information signal using a reduced number of bits as well as a signal stretching in which the number of bits, which represent the information signal is restored.

background the invention

If a digital information signal for storage and / or transmission is compressed at a reduced bit rate, the samples become the information signal often requantized using a reduced word length. The reduction the word length, by which the samples are respectively shown reduced the number of bits required to represent the information are at the expense of an increase in quantization errors. The requantization usually takes place linear through even distribution of quantization values corresponding to the available quantization levels over the Range of values of the information signal and by requantization each sample of the information signal closest to nearest quantization value. The same technique or procedure can be used for requantization of transformation coefficients are used, which is the Represent information signal.

at some applications, such as the so-called ISO / MPEG Layer 3 audio coding standard, are or are the quantization values according to the available quantization levels distributed unevenly. The uneven distribution allows the quantization values, that certain values of the information signal are more accurate than other values be quantized. This gives better results than a linear one Quantization, for example, the values that quantizes more accurately become, more often occur as the values that are quantized with less accuracy or there are errors in the values that are quantized more accurately become, stronger noticeable. Both of these possibilities can occur during low-bit-rate audio compression.

Lots Audio compression systems take an orthogonal quantization before Transformation of the audio input signal into the frequency domain in front. When this happens, one of the most common input signals will occur low amplitude spectral coefficients more frequently as high amplitude spectral components. Because the psychoacoustic phenomenon concurrent masking between the members of a group of consecutive spectral coefficients is beyond that noise due to quantization errors in the presence from a big one Amplitude having lower likelihood coefficients to listen as in the presence of low amplitude coefficients. Out For this reason, a nonlinear quantization system, which more quantization level at low amplitudes concentrated, too lead to a significant improvement in sound quality.

A Non-linear quantization is usually realized by that a linear quantization to a non-linearized version of the data that is the subject of quantization. The data to be quantized is usually a block or a set of data, such as a set of spectral coefficients, using the same number of quantization levels, or the same word length to be quantized. In the case of nonlinear quantization the data that is the subject of quantization by a nonlinear Function preprocessed, and the resulting preprocessed Data then becomes uniformly spaced apart quantization values provided. If Q [x] is a linear Quantization of x and f (x) indicates a nonlinear function of x, then the non-linear quantization can be represented as Q [f (x)]. This method is exemplified by the ISO / MPEG Layer 3 audio coding standard specified.

1 illustrates a known linear quantization system Q [x] with five quantization levels. 1 shows the relationship between the data value x of the data subject to quantization and the resulting quantization levels. The five possible quantization levels are indicated by circles in integers above the horizontal axis. Each quantization level has a corresponding quantization value. For example, in 1 the quantization level 2 has a quantization value of 1.0, and the quantization level -1 has a quantization value of -0.5. The data value x of the data subject to quantization and the quantization level of the quantization values are in 1 indicated below the horizontal axis by the values -1.0, -0.5, 0.5 and 1.0.

At the in 1 In the linear quantization system shown, the data x is quantized to the quantization level with a quantization value closest to x. In 1 the decision values between adjacent quantization values are indicated by dashed lines. For example, the dashed Li never D ₀₁ the decision value between the quantization values 0 and 0.5. The data values between adjacent decision values are all quantized to the same quantization level and thus to the same quantization value.

The data values of the data undergoing quantization may be limited to the range between -1.0 and 1.0. Alternatively, the data may have data values outside this range; in this case, data values greater than the highest decision value D _{12 are} quantized to the highest quantization level, and data values lower than the lowest decision _value D _-1-2 are quantized to the lowest quantization level. At the in 1 As the result of the quantization system shown, the decision values fall halfway between the quantization values as a result of the data values being quantized to the next quantization value.

As mentioned above, may be a nonlinear quantization system according to the prior art by a first preprocessing or transformation of the data object quantization, using a nonlinear function and then linearly quantizing the resulting preprocessed data can be realized. If f () is the nonlinear Is function and the linear quantization of x is expressed as Q [x], then the nonlinear quantization of x can be expressed as Q [f (x)].

In 2 Fig. 12 is a block diagram of a known non-linear quantizer. According to 2 For example, the data that is subject to quantization will be passed through the input port 10 the preprocessor 21 in this example, the non-linear function f () from the non-linear function block 23 get here. The preprocessor pre-processes the respective data that is subject to quantization according to the non-linear function that it receives from the non-linear function block. The preprocessed data is from the preprocessor 21 the linear quantizer 22 fed. The linear quantization device 22 uses a set of equally spaced quantization values according to those from the input terminal 11 received data indicating wel the number of quantization levels. The linear quantization device 22 quantizes the preprocessed data to the closest equally spaced quantization value and outputs the resulting nonlinear quantized data to the output port 14 from.

It should be noted that linear quantization is a special case of the nonlinear quantization just described. If the nonlinear function f () is defined as f (x) = x, then the data is linearly quantized as in 3 is illustrated. According to 3 First, the data a is preprocessed by the linear function f (x) = x, and the resulting preprocessed data a 'on the vertical axis is then linearly quantized by Q []. Since the function f (x) is linear with a slope of 1.0, the result of Q [f (x)] is the same as Q [x]. At the in 3 For example, the pre-processed data value is greater than the decision value D ₁₂ between the quantization levels 1 and 2, and therefore the data a is quantized to a quantization level of 2.

If f (x) is a nonlinear function, the quantization system Q [f (x)] becomes a nonlinear quantization system according to the prior art. 4 illustrates an example of this form of nonlinear quantization. As in the example of the above with reference to 3 As described above, the data is preprocessed by the function f (x) and the preprocessed data is then stored in the non-linearized region as shown on the vertical axis in FIG 4 is illustrated, linearly quantized. In this case, however, the function f (x) is nonlinear.

A preprocessing using the nonlinear function f (x) has the effect of changing the quantization value in the input area (the horizontal axis in FIG 4 ) each of the quantization levels. 5 illustrates the result of this effect. In 5 is the same quantization system as in 4 shown, with the exception that the respective quantization value q from the transformed area (the vertical axis in 5 ) has been inversely transformed back into the input area (that is, the horizontal axis) by the inverse magnitude g (q) of the nonlinear function f (x). Thus shows 5 the level in the input range of the quantization value of each quantization level in the transformed range. It can be seen that the quantization values are uniformly spaced in the transformed region, but are not uniformly spaced in the input region. In this example, since the slope of the nonlinear function f (x) is greater for data values that are close to zero than for data values near 1.0 and -1.0, the illustrated non-linear quantization system is more accurate for small data values.

6 illustrates the quantization values of 5 in the same format as 1 , using only the entrance area. 6 also illustrates the corre sponding lowing decision values. The decision values are calculated in the same way as the quantization values in 5 that is, by inverse transforming the decision values in the transformed area (the vertical axis of 5 ) back to the entrance area (this is the horizontal axis of 5 ). It can be seen that the decision values are no longer halfway between the quantization values in the input range. With this in mind, certain data values can not be quantized with a minimum quantization error.

The by the above-mentioned ISO / MPEG Layer 3 audio coding standard prescribed nonlinear quantization device has additional Disadvantages. First, the quantization effects vary predetermined nonlinear function f () according to the number the quantization level used to quantize the data values, after having preprocessed the nonlinear function f () accordingly have been. By the standard or the concerned Norm prescribed quantizer changes the number of Quantization level while a fixed nonlinear function f () is used. In a quantization device, in which the number of quantization levels changes dynamically, such as In most high-efficient compression systems, one performs Preprocessing using a fixed nonlinear function f () does not produce optimal results for all values of the number of Quantization levels.

To the Second, the calculation of f (x) with respect to the respective data for the non-linear function f () too much time and / or signal processing power require. To save time or processing power, the Data values using a simpler, but less optimal nonlinear function f () must be preprocessed.

Summary the invention

The Invention provides a method for quantizing data which represent an information signal in order to derive the information signal representing corresponding quantized data using to produce fewer bits. The respective data will be in the case that they are quantized, represented by a quantization level, which is selected from a number of quantization levels. In the method, the data and word length information, which indicates the number of quantization levels received. One Set of quantization tables is provided. The set of quantization tables contains a quantization table for every possible one Number of quantization levels. Each quantization table contains one Table entry for each quantization level in the respective number of quantization levels. The quantization table for the number of quantization levels indicated by the word length information are selected from the set of quantization tables. Finally will the quantization level whose table entry matches the data value of the corresponding data from the quantization table as appropriate Data of the quantized data selected.

The table entries in the quantization tables provide a non-linear quantization characteristic, and you can be derived from a single nonlinear function that the nonlinear quantization characteristic.

The table entries in the quantization tables a different non-linear quantization characteristic in different quantization tables. In this Trap can the table entries in the quantization tables for more quantization level a more nonlinear one Provide quantization characteristics as the table entries in the Quantization tables for less quantization level.

The table entries can a quantization value or an input range for each quantization level contain. If the quantization value is included, the quantization level becomes of the table entry where the quantization value is the data value closest to one data is, selected. If the input range is included, the quantization level becomes of the table entry selected, where the input range is the data value of the one concerned Includes data.

The Procedure may additionally perform a time-frequency analysis on the information signal to to generate the data.

The Procedure can also additionally the exercise block gliding on the data to block through a Blockglei to generate processed data; in this case the quantization level, whose table entry is the block-sliced data value corresponds to the respective data processed by block gliding, selected as appropriately quantized data.

The invention also provides a method of nonlinear quantization of data representing an information signal to produce correspondingly quantized data representing the information signal using fewer bits. The respective data will be, if not are linearly quantized, represented by a quantization level selected from a number of quantization levels. The method provides a set of non-linear functions. The nonlinear functions each define a non-linear quantization characteristic. The data and the word length information indicating the number of quantization levels are received. One of the non-linear functions is selected from the set of non-linear functions, preferably the word-length input signal. The data is preprocessed to generate preprocessed data by multiplying the data by the selected nonlinear function to produce corresponding preprocessed data. Finally, the preprocessed data is linearly quantized in response to the word length information to produce the quantized data.

For every number the quantization level in the permissible Range of the number of quantization levels can be a different be provided non-linear function.

The nonlinear functions can in nonlinearity vary; in this case, the one selected function indicates the nonlinear Functions a stronger one nonlinearity on, if the word length information indicates more quantization levels than the one selected function of the nonlinear function when the word length information has less quantization level indicates.

A piecemeal linear approximation of a nonlinear function can be considered as respective Function of the nonlinear functions in the nonlinear function set be provided. One piece at a time Approximation of a nonlinear function would be for any number of quantization levels within the permissible range the number of quantization levels available. Each piecewise contains linear approximation a plurality of nodes involving the piecewise nonlinear approximation coincident with the nonlinear function. The knots are in the Number equal to the corresponding number of quantization levels. Each piecewise In addition, linear approximation points linear segments connecting successive nodes. Replacing the nonlinear function by the piecewise linear approximations of the nonlinear functions themselves diminished the quantization error generated by this method.

The through the piecewise linear approximations of approximated nonlinear functions can be found in of nonlinearity vary; in this case, the nonlinear functions are the through the piecewise linear approximations are approximated, for larger numbers of quantization levels, preferably stronger nonlinear as the nonlinear functions by piecewise linear approximations for smaller numbers of quantization levels are approximated. alternative can the piecewise linear Approximations all approximate the same nonlinear function.

The Procedure may additionally execution include a time-frequency analysis on the information signal, to generate the data.

The Procedure can also additionally the exercise block glide on the data to be processed by block sliding Generate data; in this case, the preprocessed data generated by the fact that each processed by block sliding Data with the selected nonlinear function multiplied.

Finally, poses the invention a method for dequantization of the straight described method nonlinear quantized data available. at The method becomes a set of inverse nonlinear functions provided. The inverse nonlinear functions correspond each one of the plurality used to quantize the data of nonlinear functions. The nonlinear quantized data and the word length information, which indicates the number of quantization levels with which the nonlinear Quantized data are quantized are received. The nonlinear quantized data becomes linear in response to the word length information dequantized to produce non-linear data. From the sentence of Nonlinear functions become one of the inverse nonlinear functions selected. After all the non-linear data undergoes post-processing, to generate the dequantized data by selecting the selected inverse nonlinear function applied to the respective nonlinear data to generate corresponding dequantized data.

The one of the inverse nonlinear functions may be on the word length information selected become.

The inverse nonlinear functions may vary in nonlinearity; in this case, the selected one of the inverse nonlinear functions preferably has a stronger nonlinearity, if the word length information indicates more quantization levels than in the case that the word length information indicates less quantization level.

On the dequantized data can be additionally postprocessed resolution block gliding.

Short description the drawings

1 Figure 11 illustrates in an input domain diagram a linear quantization according to the prior art.

2 Fig. 13 is a block diagram showing a prior art nonlinear quantization device in which data is subjected to nonlinear transformation prior to linear quantization.

3 Figure 11 illustrates in an input / output diagram a linear quantization based on a nonlinear quantization model.

4 Fig. 11 illustrates in an input / output diagram a nonlinear quantization according to the prior art.

5 Fig. 11 illustrates in an input / output diagram the quantization value of the respective quantization level generated by non-linear quantization according to the prior art.

6 FIG. 4 illustrates, in an input range diagram, the quantization values and the decision values represented by the in. FIG 5 illustrates nonlinear quantization methods according to the prior art.

7 shows a block diagram of a non-linear quantization device.

8th illustrates in a flow chart the calculation of the quantization values for the respective quantization level.

9 Fig. 3 is a graph illustrating an exemplary nonlinear function that could be used by the first non-linear quantization method, the graph also showing a linear function for comparison.

10A Fig. 9 is a graph showing the absolute value of the quantization error resulting from the non-linear quantization with three quantization levels.

10B FIG. 4 is a graph showing the absolute value of the quantization error resulting from nonlinear quantization with seven quantization levels. FIG.

11 Fig. 11 is a block diagram of an audio broadcasting or recording system in which the audio signal is compressed using the non-linear quantization techniques prior to transmission or recording.

12 illustrates in a block diagram the audio compression device of the in 11 audio signal transmission system shown, wherein the audio compression device concerned comprises a non-linear quantization device embodying the non-linear quantization method.

13 FIG. 16 is a block diagram illustrating a time-frequency analyzing means in the audio compression apparatus of FIG 11 represented audio signal transmission system.

14 Fig. 11 is a block diagram illustrating a nonlinear quantization device embodying a second non-linear quantization method.

15 FIG. 12 shows an example of the quantization table for five quantization levels used in the second non-linear quantization method.

16A Figure 4 is a graph illustrating a piecewise linear approximation of a nonlinear function for five quantization levels for use in the non-linear quantization method.

16B Figure 4 is a graph illustrating a piecewise linear approximation of a non-linear function for seven quantization levels for use in the non-linear quantization method.

17 illustrates in a block diagram a non-linear quantization device.

18 illustrates quantization values and decision values in an input domain diagram.

19 Figure 4 is a graph illustrating an exemplary set of non-linear functions for use in the fourth non-linear quantization method according to the invention.

20 Fig. 13 is a block diagram showing a nonlinear quantization device embodying the non-linear quantization method according to the invention.

21 Fig. 12 is a block diagram illustrating a nonlinear embodying the non-linear dequantization method according to the invention re dequantization device.

22 Fig. 11 is a block diagram showing an audio-stretching device of the present invention 11 represented audio signal transmission or recording system.

detailed Description of the invention

In 7 Fig. 10 is a block diagram of a non-linear quantizer embodying a non-linear quantization method 30 illustrated. The nonlinear quantization device 30 and the other non-linear quantizers and non-linear dequantizers described herein are shown as block diagrams. The functional elements of these block diagrams may be implemented in practice using appropriate LSI circuits (that is, high-density circuits) or by using lower-integrated integrated circuits or by discrete components. Alternatively, the functional elements of the non-linear quantizers and dequantizers may be implemented by programming a suitable microcomputer or digital signal processor and using such additional circuitry as may be required.

In the non-linear quantizer 30 takes the quantizer 32 the data that is the subject of the quantization, via the input terminal 10 on, the respective data subject to quantization quantizes to the quantization level corresponding to the next quantization level in the set of non-uniformly spaced quantization values from the quantization value calculator 31 are supplied, and outputs the resulting quantized data to the output terminal 14 from. The data that is the subject of quantization and that is the input terminal 10 are usually a data block or data set, such as a set of spectral coefficients to be quantized using the same number of quantization levels or word length.

The quantization value calculator 31 receives the non-linear function f () from the non-linear function block 33 , and it receives from or via the input port 11 an input signal indicating the number of quantization levels at which the data to be quantized is to be quantized. The quantization value calculator calculates a quantization value for the respective quantization level, and outputs the resulting quantization values to the quantizer 32 from. The number of calculated quantization values is equal to the number of quantization levels indicated by the input signal indicating the number of quantization levels. The quantization value calculator 31 can via the input port 11 an input signal indicating the word length with which the data subject to quantization is quantized, in place of the input signal indicating the number of quantization levels.

8th illustrates in a flowchart in more detail the calculation of the quantization value of the respective quantization level by the quantization value calculator 31 , At the in 8th 1, the corresponding linear quantization value x in the transformed region is first calculated for the respective quantization level i. Then, the corresponding quantization value q [i] in the input range is calculated as q [i] = g (x), where g () is the inverse function of the nonlinear function f (), that is, g (f (x)) = x.

The processing starts at step S0. In step S1, the non-linear function block is used 33 receive an input signal indicating the nonlinear function f (). In step S2, the inverse function g () of the function f () is calculated from the nonlinear function f (). At step S3, via the input terminal 11 an input signal indicating the number of quantization levels lev. At step S4, the quantization level i is initialized to the lowest quantization level using equation (1): i = - (lev - 1) / 2 (1)

Of the Quantization value q [i] in the input range for the respective quantization level i is then calculated by the loop resulting from the steps S5, S6, S7 and S8 until it is determined at step S7 that the quantization values for all Quantization levels have been calculated.

At step S5, the linear quantization value x in the transformed range for the quantization level i is calculated by using equation (2). x = (2 × i) / (lev-1) (2)

At step S6, the quantization value q [i] in the input region is calculated from the quantization value in the transformed region and the inverse function g () using equation (3). q [i] = g (x) (3)

At step S7, a test is performed to determine whether the quantization values for all quantization levels have been calculated by determining whether the value of i equals the highest quantization level through equation (4): i = (lev - 1) / 2 (4)

If at step S7, it is determined that the value of i does not correspond to the highest quantization level, the execution goes proceed to step S8, where the value of i increments by one and then proceeds to step S5 where the loop is repeated.

If it is determined in step S7 that the value of i corresponds to the highest quantization level, this indicates that quantization values have been determined for all the quantization levels, and execution proceeds to step S8. In step S8, the values determined by the quantization value calculator 31 calculated quantization values of the quantization device 32 for use in quantizing the data undergoing quantization. The processing then proceeds to step S9, at which control returns to the main routine.

The quantization device 32 quantizes the respective data subjected to the quantization by connecting the output terminal 14 a quantizing level selected from the number of quantizing levels designated by the input signal indicative of the number of quantizing levels (or the number of quantizing levels corresponding to the input word length) supplied through the input terminal 11 to be obtained. The quantization device 32 also obtains a quantization value for each quantization level from the quantization value calculator 31 for use in quantizing the data subjected to the quantization to the next quantization value. The quantization device 32 For example, it may determine which of the quantization values obtained by the quantization value calculator 31 is closest to the respective data subjected to the quantization, and becomes the quantization level corresponding to the quantization value concerned at the output terminal 14 when they submit a quantized data. Each the output terminal 14 The quantization level supplied is the number of bits indicated by the word length input signal (or has the number of bits corresponding to the input signal indicative of the number of quantization levels) which is output by the quantization device 32 from the input terminal 11 is recorded. The quantization device 32 may determine which of the quantization values is closest in value to the respective data undergoing quantization, for example, by determining differences between the one data and the respective quantization values, and that the output terminal 14 the quantization level corresponding to the quantization value with respect to which the difference is lowest is supplied.

The quantization levels by the quantization value calculator 31 executed processing results in quantization values that coincide with quantization values as shown in FIG 5 are illustrated by a known nonlinear quantization device using the same nonlinear function f (). However, the nonlinear quantization performed by the quantization method differs from that in 5 represented processing, since the quantization device 32 operates in the input domain and quantizes the respective data undergoing quantization to the quantization level corresponding to the next unevenly spaced quantization value. This results in decision data lying on the midpoints between adjacent quantization values, as distinct from the decision values in the above with reference to FIGS 5 and 6 described known non-linear quantization device. As a result, the first nonlinear quantization method produces quantization errors that are less than or equal to the quantization error (s) generated by the known nonlinear quantization.

The fully extended line in 9 shows an exemplary nonlinear function; the dashed line indicates a linear function for comparison. The 10A and 10B The quantization errors generated by the nonlinear quantization method (full line), as compared to the quantization errors produced by the known nonlinear quantization (dashed line), show both quantizations with the in 9 working through the full line exemplary nonlinear functions. 10A Figure 12 shows the quantization errors that result when the data values in the range -1.0 to +1.0 are quantized using three quantization levels. It can be seen that the non-linear quantization method produces significantly smaller quantization errors for data values near -0.4 and +0.4. 10B Figure 12 shows the quantization errors that result when data values are in the same range using seven quantization levels be quantized. In both illustrated cases, the quantization errors produced by the non-linear quantization method are less than or equal to the quantization errors (n) produced by the known nonlinear quantization.

As an alternative to the output of the quantization value for the respective quantization level to the quantization device 32 can the quantization value calculator 31 calculate an input range for each quantization level in the number of quantization levels and the quantizer 32 respectively. This allows the quantizer 32 to simplify. The quantization value calculator 31 can determine a decision value in the middle between adjacent quantization values from the quantization values calculated as described above. The determination of the decision value in the middle between adjacent quantization values minimizes the quantization errors. The quantization value calculator may then derive an input range for each quantization level from the calculated decision values of adjacent quantization levels. The quantization value calculator may apply the calculated input range for the respective quantization level to the quantization device 32 in place of the respective quantization value. The quantization device 32 Then, the data subjected to the quantization can be simply quantized by determining which of the input ranges obtained from the quantization value calculator comprises the data value of the respective data subjected to the quantization, and can supply the respective quantization level to the output terminal 14 submit.

Now, referring to the 11 to 13 an example of the application of a non-linear quantization device embodying the non-linear quantization method in an audio signal transmission or recording system. The non-linear quantization method, however, is not limited to this application. The non-linear quantization method will provide the advantages described above when used in other non-linear quantization applications. For example, the non-linear quantization method can be used for nonlinear quantization of the coefficients resulting from orthogonal transformation of a square or quadrilateral block from a motion picture signal or from a square or quadrilateral block of errors resulting from motion compensation with respect to a motion picture.

11 illustrates the general structure of the transmission or recording system 110 for an audio signal. In the illustrated system, the audio input signal is compressed prior to recording or transmission to reduce the storage capacity requirements for the recording medium and / or the bit rate requirements of the transmission medium. The audio input signal is through the audio compressor 111 compressed to provide a compressed signal to the transmission or recording medium 113 is supplied. The compressed signal reproduced from the transmission or recording medium is then passed through the audio-stretching device 112 stretched to provide an audio output. The system 110 usually works according to the psychoacoustic characteristic of the human sense of hearing to minimize the audibility of errors resulting from the process of compressing and stretching the audio input signal.

12 shows a block diagram of the audio compressor 111 which works according to the psychoacoustic characteristic of the human sense of hearing. The time-frequency analyzer 121 receives the audio input signal and divides the audio input signal in time and in frequency into sets of spectral components. For example, the time-frequency analyzer may divide the audio input signal into blocks in time and then orthogonally transform the samples of the audio input signal in the respective block from the time domain to the frequency domain to produce the sets of spectral components. Alternatively, the time-frequency analyzing means may divide the audio input signal into a plurality of frequency bands using, for example, a plurality of filters, and may divide the signal component in each of the frequency bands into sets of spectral components over time. The time-frequency analyzer 121 gives the sets of spectral components to the bit allocation device 122 and the spectral quantizer 123 from.

At the in 12 In the audio compressor shown in FIG. 1, the accuracy with which the spectral components in each set of spectral components are represented using a certain number of bits is increased by applying block-slip processing to each set of spectral components before their quantization. The bit allocation device 122 determines a scaling factor for each set, and the spectral quantizer 123 divides the respective spectral component in the sentence by the scaling factor. The resulting mantissas are then quantized. In practice, like this will be described in more detail below, dividing the spectral components in each set of spectral components by the frequency into frequency bands, and the bit allocation means 122 determines a scaling factor for the respective frequency band. The spectral quantizer 123 applies block sliding processing to the spectral components in the respective frequency band using the respective scaling factor.

The bit allocation device 122 It adaptively determines the word length and scaling factor to quantize the spectral coefficients in the respective set of spectral components and gives the scaling factor and word length to the spectral quantizer 123 from. The spectral quantizer 123 uses the scaling factor and word length for non-linear quantization of the spectral components in the respective set of spectral components, for example, by using block-slip processing and non-linear quantization to generate a set of nonlinear quantized spectral components. Finally, the bit allocation means 122 the bit allocation parameters, and the spectral quantizer 123 gives the respective set of nonlinear quantized spectral components to the multiplexer 125 from. The multiplexer multiplexes the bit allocation parameters and the nonlinear quantized spectral components to obtain the compressed signal for delivery to the recording or transmission medium 113 to build.

To ensure that the quantization noise resulting from the quantization by the spectral quantization 123 The result is a nonlinear quantization that is subjectively inaudible to the listener, or is minimally audible, through the bit allocation facility 122 executed bit allocation processing is performed on the basis of a psychoacoustic model, which is schematically represented by the psychoacoustic model 124 is shown. The psychoacoustic model used by the bit allocation processing could typically include such data as threshold detection curves, equal volume curves, simultaneous masking characteristics, and temporal masking characteristics.

In the audio compressor 111 can the in 12 illustrated spectral quantization device 123 apply the non-linear quantization method as described above with reference to 7 and 8th has been described. Other embodiments of the non-linear quantization method described below may also be used for the first non-linear quantization method in the spectral quantizer 123 be used. The spectral quantizer 123 will now be with reference to 12 described.

In the spectral quantizer 123 the block sliding processor receives 132 the sets of spectral components from the time / frequency analyzer 121 and it receives the respective scaling factor for the particular set from the bit allocator 122 , The block sliding processor 132 applies block-slip processing to the respective set of spectral components by dividing the spectral components in the set by the respective scale factor. The resulting sets of the block-glide processed spectral components become the non-linear quantizer 133 via the input connection 10 fed. The nonlinear quantization device 133 also obtains the word length for the particular set of spectral components from the bit allocator 122 via the input connection 11 and performs a nonlinear quantization of the respective set of the spectral components processed by block gliding for the word length. The non-linear quantizer gives the resulting set of quantized spectral components over the output port 14 to the multiplexer 125 from.

If the above with reference to 7 described nonlinear quantization device embodying the first non-linear quantization method as a non-linear quantization device 133 is applied quantizes the quantization device 32 each of the block gliding processed spectral components in the set of block gliding spectral components processed by the block glide processor 132 via the input connection 10 are included. The quantization device 32 quantizes the spectral components processed by block gliding to the next quantization value in the number of quantization values generated by the quantization value calculator 31 are calculated. The quantization value calculator 31 calculates the number of quantization values using the nonlinear function of the nonlinear function block 33 and from the respective word length indicative input signal via the Eingangsan circuit 11 has been recorded. The quantization device 32 gives the resulting set of quantized spectral components over the output port 14 to the multiplexer 125 from.

In the audio compressor 111 can the time-frequency analyzer 121 Although based on a hybrid form of transformation and subband coding, it is not limited to such a structure. An example of the time-fre frequency analyzer 121 incorporating the ATRAC hybrid coding technique, which is used, for example, to compress a PCM audio signal to produce the compressed signal for recording on a MiniDisc, is disclosed in US Pat 13 illustrated. The input connection 151 a PCM audio input signal in the frequency range of, for example, 0 Hz to 20 kHz is supplied. The audio input signal is passed through the frequency domain split filter 153 divided into a frequency domain signal in the frequency range of 0 Hz to 10 kHz and in a signal of a high frequency range of 10 to 20 kHz. The frequency domain signal in the frequency range of 0 Hz to 10 kHz is provided by the frequency domain split filter 155 further divided into a signal of a lower frequency range of 0 Hz to 5 kHz and into a signal of a middle frequency range of 5 to 10 kHz. As a frequency domain subdivision filter 153 and 155 Preferably, quadrature mirror filters (QMFs) are used.

alternative in the case that the PCM audio input signal comprises a frequency range of 0 Hz to 22 kHz, the signal of the high frequency range has a Frequency range from 11 kHz to 22 kHz, the signal of the middle Frequency range covers a frequency range from 5.5 kHz to 11 kHz, and the lower frequency range signal includes a frequency range from 0 Hz to 5.5 kHz.

The frequency domain signal in the high frequency range is from the frequency domain split filter 153 to the orthogonal transformation circuit 157 delivered for the high frequency; the frequency domain signal in the middle frequency range is from the frequency domain split filter 155 to the orthogonal transformation circuit 159 for the middle frequency, and the frequency domain signal in the lower frequency range is output from the frequency domain dividing filter 159 to the orthogonal transformation circuit 161 delivered for the lower frequency.

The orthogonal transformation circuits 157 . 159 and 161 , which are preferably Modified Discrete Cosine Transform (MDCT) circuits, take an orthogonal transform of frames of the respective frequency domain signals to collectively generate corresponding sets of spectral components for output to the bit allocator 122 and the spectral quantizer 123 in front. Those through the orthogonal transformation circuits 157 . 159 . 161 collectively generated sets of spectral components become the output port 165 supplied by the bit allocation device 122 and the spectral quantizer 123 be supplied.

That from the input terminal 151 The received audio input signal is assumed to be divided into frames of a certain number of samples. The orthogonal transformation circuits 157 . 159 and 161 Orthogonal transform the frames of the three frequency domain signals derived from the respective frame of the audio input signal to collectively produce a set of spectral components. To provide further flexibility and eliminate audible time domain artifacts, such as noise modulation, the orthogonal transform circuits take 157 . 159 and 161 an orthogonal transformation of the respective frame of the frequency domain signals into blocks with a variable block length. The block length to which the respective frequency domain signal is orthogonally transformed is equal to the frame length divided by 2 ⁿ , where n is a positive integer equal to or greater than zero. The block lengths into which the respective frame of the audio input signal is orthogonally transformed into the three frequency ranges are determined by the block length decision block 163 based on the dynamic spectral characteristics of the digital input signal. In general, the signal of the high frequency range is orthogonally transformed into blocks having a shorter block length than the signals of the lower and middle frequency ranges. The block length decision block 163 Also reduces the block lengths in frames of the digital input signal in which the level of the digital input signal changes in comparison to frames of the digital input signal in which the level remains relatively constant.

In another possible configuration of the time-frequency analyzer 121 in the audio compressor 112 For example, a time-frequency transform similar to that specified by the MPEG Layer 3 audio coding standard is used. In this case, the audio input signal is divided by a set of filters into 32 subbands of equal width, followed by an MDCT transform.

In the audio compressor described above 111 For example, the spectral components in the sets of spectral components may be quantized by the spectral quantizer 123 be divided into frequency bands. In this case, the bit allocation device becomes 122 assign a scale factor and a word length to quantize the spectral components in the respective frequency band, and the block sliding processor 132 will apply block-slip processing to the spectral components in the respective frequency band; the nonlinear quantization device will not line up the spectral components in the respective frequency band to the respective word length for the band quantize ar. When the non-linear quantizer 133 The first non-linear quantization method according to the invention embodies the quantization value calculator 31 therein a quantization value for each level from the number of quantization levels indicated by the respective word length in the respective frequency band, and the quantization means 32 in it, the spectral components subjected to block-slip processing in the respective frequency band quantizes to the next quantization value, that for the band by the quantization value calculator 31 is determined.

In the audio transmission or recording system 110 takes the stretching device 112 a precise elongation of that of the recording or transmission medium 113 as long as the non-linear dequantizing device in the expander device utilizes the inverse function of the nonlinear function provided by the non-linear quantizer 133 in the compressor 111 is used. As far as the expander is concerned, it is irrelevant whether the quantized spectral components in the compressed signal were quantized using nonlinear quantization or the prior art, since the quantization values of the quantization levels are the same in both cases. However, when the quantized spectral components in the compressed signal have been nonlinearly quantized by the non-linear quantization method, however, the quantization errors that result when the expander becomes 112 the compressed signal stretches smaller than when the nonlinear quantization has been performed according to the prior art. The reason for this is that the non-linear quantization uses decision values different from those used in the known non-linear quantization using the same nonlinear function.

The non-linear quantization device which uses the above with reference to FIGS 7 and 8th describes nonlinear quantization methods, uses a fixed non-linear function f (), and recalculates the quantization values each time the word length or the number of quantization levels is changed. This can consume considerable computing power. Moreover, the quantization effects of a given nonlinear function f () may change according to the number of quantization levels.

Accordingly used a non-linear one embodying a second non-linear quantization method Quantization means a look-up table, which has an input range x with a quantized output value T [x] for each Quantization level defines or defines. It does not need preprocessing using the nonlinear function f () during the Quantization processing to be executed. Instead becomes the nonlinearity quantization by the distribution of those specified in T [] Input ranges determined. These input areas can be arbitrary be fixed without the complexity of the quantization process to change. A separate lookup table T [] must be used for each possible number of quantization levels be specified. When the input ranges and the quantization values in the respective lookup table T [] on a single nonlinear Function f () based, then you can the input ranges and the quantization values in such Chosen way be that one the second non-linear quantization method embodying non-linear quantization device lower quantization errors generated as a known nonlinear quantization device based on the same nonlinear function.

In 14 Fig. 10 is a block diagram of a non-linear quantizer embodying the second non-linear quantization method 40 illustrated. In this second non-linear quantization method, the data subjected to quantization is non-linearly quantized by using a look-up table selected from a set of look-up tables to generate the nonlinear quantized data.

In 14 receives the table quantizer 42 in the non-linear quantizer 40 the data that is the subject of the quantization, via the input terminal 10 , quantizes the respective data that is the subject of quantization using a look-up table provided by the table selector 42 has been obtained to generate the respective nonlinear quantized data, and outputs the resulting quantized data to the output terminal 14 from. The data subjected to the quantization and that via the input terminal 10 normally represent a block or set of data, such as a set of spectral coefficients to be quantized using the same number of quantization levels or word length.

The table selector 41 stores a set of lookup tables for use by the table quantizer 42 , The table selector 41 receives via the input port 11 an input signal indicating the number of quantization levels which are to be used to quantify the data to be subjected to quantization. In response to the input signal indicating the number of quantization levels, the table selector selects 41 one of the lookup tables from the set of lookup tables stored therein, and passes the selected lookup table to the table quantizer 42 from. The table selector 41 For example, instead of the input signal indicating the number of quantization levels, an input signal indicating the word length to be used for quantizing the data that is the subject of the quantization may alternatively be included.

The lookup tables in the set of the in the table selector 41 stored lookup tables are usually derived from a nonlinear function f () written in 14 schematically by the non-linear function block 43 is specified. The set of the in the table selector 41 stored lookup tables for a given nonlinear function consists of a unique lookup table for each possible value of the number of quantization levels (or word length). Each look-up table consists of a table entry for the respective quantization level in the number of quantization levels that pass through the input terminal 11 recorded quantization level data are specified. Each table entry consists of the input range and the output quantized quantized level.

The Combining the input areas for all table entries in each Lookup table must span the range of data values of the data which are the subject of quantization, making a table entry for each potential Data value in the range of data values exists. Furthermore must not overlap the input ranges of the table entries in the respective lookup table be present, so no more than one table entry for each potential Data value in the range of data values exists. In other words, this means that in every lookup table one and only one table entry may or must be present, which has an entrance area, which each possible Data value in the range of data values corresponds.

In 15 an example of the lookup table for five quantization levels is illustrated. To simplify the table, the input data ranges are represented by decimal numbers with a resolution of 0.01 instead of binary numbers with a resolution of the least significant bit. In practice, the input ranges would be represented by binary numbers having a number of bits equal to or greater than the number of bits in the respective data that are the subject of the quantization. To ensure that the input ranges cover the entire range of data values without overlapping, the maximum of the respective data range would differ from the minimum of the next higher input range by the least significant bit.

Coming back to 14 It should now be noted that the table selector 41 selects the candidate look-up table from the set of look-up tables stored therein for the input signal that passes through the input port 11 which indicates the number of quantization levels for the quantization of the data subject to the quantization, wherein the selection means concerned is the table look-up table of the quantization means 42 supplies. Then, the quantizing means for the respective data subject to the quantization determines the one range of the input ranges in the selected lookup table into which the data value falls, and sets the corresponding output quantized value from the lookup table as the respective one nonlinear quantized data nonlinear quantized data available.

In the 14 illustrated non-linear quantization device 40 provides a great deal of flexibility, since it can perform quantization using some nonlinear function f () and because its complexity does not increase with increasing complexity of the nonlinear function f ().

When creating the in the table selector 41 stored look-up tables, the boundaries of the input areas may be determined in any desired manner, for example by conversion from a nonlinear function f () or by direct calculation.

When the input ranges are calculated directly, the non-linear quantizer generates 40 the lower quantization errors provided by the first non-linear quantization method operating with the same nonlinear function. These quantization errors are lower than those in 2 The known nonlinear quantization device which operates with the same nonlinear function. As in 6 illustrates the decision values between the quantization levels at the in 2 The known nonlinear quantization device shown does not appear at or at the midpoints between the quantization values. As a result, some data values are not quantized to the next quantization value. This represents an artifact of the known nonlinear quantization devices. The second quanti On the other hand, the non-linear quantization device embodying the invention allows the input ranges in the look-up tables to be selected such that the decision values are made to fall exactly at the midpoints between the quantization values. When this happens, the quantization errors produced by the second non-linear quantization method are as in 10A and 10B is illustrated by the solid lines, for a 3-level and 7-level quantization. The quantization errors produced by the known non-linear quantization are shown by dashed lines in these drawing figures for comparison.

As mentioned above, the nonlinear quantization device embodying the first non-linear quantization method 30 by the nonlinear quantization device embodying the second non-linear quantization method 40 as a non-linear quantizer 133 in the compressor 111 of in 11 to 13 replaced audio signal transmission or recording system. The second non-linear quantization method brings the advantages described above when used in other non-linear quantization applications.

When a nonlinear quantization device embodying the second non-linear quantization method is a nonlinear quantization device 133 is used receives the table selector 41 the word length for each set of spectral components from the bit allocation device 122 via the input connection 11 , and the table quantizer 42 receives the sets of the spectral components subjected to block-slip processing from the block floating-point processor 132 via the input connection 10 , The table quantizer then quantizes the block-sliced spectral components in each sentence using the table selector 41 selected look-up table to the word length for that of the bit allocation device 122 received sentence. The table quantizer gives the resulting set of quantized spectral components over the output port 14 to the multiplexer 125 from.

If the sets of the spectral components are divided into frequency bands before quantization, the bit allocation means becomes 122 assign a scale factor and a word length to quantize the spectral components in the respective frequency band, and the block sliding processor 132 will perform block-slip processing on the spectral components in the respective frequency band. In this case, the table selector becomes 41 a look-up table for each frequency band for delivery to the table quantizer 42 and the table quantizer 42 is used to quantize the spectral components subjected to the block sliding processing in the respective band by using the band look-up table obtained from the table selector. Alternatively, the table selector 41 It may select a look-up table corresponding to each of the different word lengths with which to quantize the set of spectral components, and may pass those look-up tables to the table quantizer 42 along with correlation data indicating which look-up table to use to quantize each of the frequency bands. The table quantizer would then quantize the spectral components in each of the frequency bands using the look-up table designated by the correlation data.

The quantization value for the respective quantization level can be found in the tables in the table selector 41 stored instead of the input range for the respective quantization value. In this case, the table quantizer will quantize the respective data that is the subject of the quantization to the next quantization value.

Now, referring to the 16A . 16B and 17 a nonlinear quantizer embodying a third non-linear quantization method 50 described. The nonlinear quantization device 50 based on the in 2 illustrated nonlinear quantization device 10 and the data subject to quantization are pre-processed according to a nonlinear function before being quantized using a linear quantizer. In the non-linear quantizer 50 are quantization errors with respect to those caused by the in 2 illustrated non-linear quantization device 10 is reduced by replacing the non-linear function used for preprocessing the data that is the subject of the quantization with a selected piecewise linear approximation of the non-linear function. The selected piecewise linear approximation depends on the number of quantization levels (or word length) at which the data subject to quantization is quantized to pre-processing. The preprocessed data resulting from preprocessing the data that is the subject of quantization with the selected piecewise linear approximation is then linearly quantized with the appropriate number of quantization levels, the equally spaced quantization values to provide the nonlinear quantized data. The third non-linear quantization method provides lower quantization errors than the known non-linear quantization using the same non-linear function prototype.

In accordance with quantization using different numbers of quantization levels (or different word lengths), a set of piecewise linear approximations of the nonlinear prototype functions is provided. Each piecewise linear approximation of the nonlinear function in the set represents a different piecewise linear approximation of the nonlinear prototype function corresponding to the respective number of quantization levels (or word length). Each piecewise linear approximation of the nonlinear function is equal to the nonlinear prototype function each of the quantization values, and is linear between successive quantization values. The 16A and 16B illustrate an exemplary nonlinear prototype function indicated by the dashed line, and the solid line indicates the corresponding piecewise linear approximation for five quantization levels and seven quantization levels, respectively.

17 FIG. 10 is a block diagram of the non-linear quantizer embodying the third non-linear quantization method. FIG 50 , In the non-linear quantizer 50 The data that is the subject of the quantization is passed through the input port 10 the preprocessor 21 fed. The data that is subject to quantization is usually a block or set of data, such as a set of spectral coefficients to be quantized using the same number of quantization levels or word length. The preprocessor 21 preprocesses the respective data subject to quantization according to the piecewise linear approximation of the nonlinear function; this approximation is derived from the approximator provided for the piecewise linear approximation 51 and the subject preprocessor outputs the resulting preprocessed data to the linear quantizer 22 from. The linear quantization device 22 uses the evenly spaced quantization values, which are fixed according to an input signal, which passes through the input terminal 11 is recorded and indicates the number of quantization levels. The linear quantization device 22 quantizes the preprocessed data to the quantization level corresponding to the next evenly spaced quantization value, and outputs the resulting nonlinear quantized data to the output terminal 14 from.

The approximate linear approximation providing approximator 51 gets the nonlinear prototype function from the nonlinear function block 53 supplied, and also receives it via the input terminal 11 an input signal indicative of the number of quantization levels to be used for the quantization of the data subject to quantization. The approximator generates the input signal indicating the number of quantization levels and the non-linear prototype function 51 the piecewise linear approximation of the nonlinear function corresponding to the number of quantization levels (or word length) and gives this approximation to the preprocessor 21 from. Instead of the input signal indicating the number of quantization levels, the approximator providing a piecewise linear approximation 51 and the linear quantizer 22 alternatively receive an input signal indicative of the word length with which the data subject to quantization is quantized.

The reduction of the quantization errors caused by the third non-linear quantization method compared to the known nonlinear quantization using the same prototype function can be seen from a comparison of the quantization values and the decision values between successive quantization values in the input domain. If the in 4 By using the nonlinear function represented as a nonlinear prototype function, the quantization values would by definition be identical to those values given in 5 and 6 are shown. In other words, this means that the in 6 decision values indicated by dashed lines would not be at the midpoints between successive quantization values. Meanwhile, since the piecewise linear approximations of the non-linear function used by the third non-linear quantization method are linear between successive quantization values, the decision values fall on the centers between successive quantization values as shown in FIG 18 is illustrated. This reduces the quantization errors without changing the nonlinearity of the quantized data points.

The quantized data generated by the third non-linear quantization method can be dequantized to the non-linear prototype function using a known non-linear quantizer based on the inverse function. Such dequantization would result in reduced quantization errors. For example, if the in 9 by the solid line shown nonlinear function as a nonlinear prototype function be would be used, the quantization errors would be as in 10A and 10B for three and seven quantization levels, respectively, by the solid line when the data resulting from the quantization by the third non-linear quantization method according to the invention is dequantized using a known dequantizer in which the inverse function is applied to the non-linear prototype function , For comparison, the quantization errors resulting from the quantization of the data using the known nonlinear quantization are in 10A and 10B indicated by ge dashed lines. In all cases, the quantization errors produced by the quantization method are less than or equal to the quantization errors (n) caused by the known nonlinear quantization.

As mentioned above, the non-linear quantization device 30 , which represents the first non-linear quantization method, by a non-linear quantizer embodying the third non-linear quantization method as a non-linear quantizer 133 in the compressor 111 of the audio signal transmission or recording system, as shown in FIG 11 to 13 is shown. However, the third non-linear quantization method is not limited to this application. The third non-linear quantization method brings the advantages described above when used in other non-linear quantization applications.

When a nonlinear quantization device embodying the third non-linear quantization method is a nonlinear quantization device 133 is used to receive the piecewise linear approximation approximator 52 and the linear quantizer 22 the word length for each set of spectral components from the bit allocation device 122 via the input connection 11 , and the preprocessor 21 receives the sets of the block glide processed spectral components from the block sliding processor 132 via the input connection 10 , The preprocessor 21 performs preprocessing of the block glide processed spectral components in the respective set using the piecewise linear approximation of the nonlinear function provided by the piecewise linear approximation approximator 52 to the word length for that of the bit allocation device 122 obtained sentence is generated. The linear quantization device 22 performs linear quantization of the resulting preprocessed spectral components in the set to produce a set of non-linear quantized spectral components. The linear quantizer provides the set of nonlinear quantized spectral components across the output port 14 to the multiplexer 125 from.

If the sets of the spectral components are frequency-divided into frequency bands before quantization as mentioned above, the bit allocation means becomes 122 assign a scale factor and a word length to quantize the spectral components in the respective frequency band, and the block sliding processor 132 will perform block-slip processing on the spectral components in the respective frequency band. In this case, the approximator providing a piecewise linear approximation 52 a piecewise linear approximation of the nonlinear function for each frequency band for delivery to the preprocessor 21 generate, and the preprocessor 21 the block spectral processed spectral components in the respective frequency band are pre-processed using the piecewise linear approximation for the band obtained from the respective approximator.

The described above nonlinear quantization and the non-linear quantization devices according to the prior art The technique uses the same nonlinear function f () independently of the number of quantization levels (or word length). This limits the performance of the quantizer since the quantization effects a predetermined function corresponding to the number of quantization levels (or the word length) that change be used to quantize the data. One is a nonlinear one Quantization method according to the invention embodying nonlinear ones Quantizer turns off the nonlinear function f () a stimulus of nonlinear functions according to the number of quantization levels (or word length) selected for quantization the data is or will be allocated. The set of nonlinear ones Functions may consist of a set of piecewise linear approximations of the respective non-linear prototype functions that exist corresponding to the number of quantization levels allocated for quantizing the data differ.

19 FIG. 12 illustrates an exemplary set of non-linear functions for use in the non-linear quantization method according to the invention. The row in question contains seven terms and is therefore suitable for a quantization system in which the number of quantization levels has seven permissible values. Alternatively, this series of nonlinear functions may also be used in a non-linear quantization system in which the number of quantization levels has more than seven allowable values if two or more of the values are off the number of quantization levels use the same non-linear function.

In the in 19 The set of non-linear functions represented in FIG. 1 would preferably use the function closest to a linear function with the least number of quantization levels, that is, the shortest quantization word length. Increasingly larger values in the number of quantization levels or larger quantization word lengths would select non-linear functions that are increasingly more nonlinear. This greatly facilitates the performance of nonlinear quantization using large quantization word lengths while avoiding the high distortion that would result if a strongly: nonlinear function with small quantization word lengths were used.

For example, a nonlinear function such as y = x ^0.55 having a word length of about 2 or 3 bits could be used, and a nonlinear function such as y = x ^0.4 could be applied with a word length of 4 bits.

20 shows a block diagram of a non-linear quantization device 60 which embodies the non-linear quantization method according to the invention. In the quantization device 60 For example, the data that is the subject of quantization is the preprocessor 21 preprocessed according to a nonlinear function that is selected by the non-linear function selecting means 62 is obtained. The resulting preprocessed data is then received by the preprocessor 21 to the linear quantizer 22 which linearly quantizes the data in question to produce the nonlinear quantized data that it sends to the output port 14 emits.

The nonlinear function block 63 stores a number of different nonlinear functions. Alternatively, the nonlinear function block may store a piecewise linear approximation of the respective nonlinear function instead of the nonlinear function itself. The application of the piecewise linear approximation of the nonlinear function rather than the nonlinear function itself reduces quantization errors by changing the decision values to occur at the midpoint between adjacent quantization values. For simplicity, unless otherwise noted, references below to the "nonlinear function" mean a nonlinear function and a piecewise linear approximation of the nonlinear function.

The nonlinear function selector 62 applies a particular non-linear function selection method to select one of the series in the nonlinear function block 63 stored non-linear functions to select one of the non-linear functions for use for non-linear quantization of the data that are the subject of quantization. The nonlinear function selector 62 gives the selected nonlinear function to the preprocessor 21 from. The preprocessor 21 receives the data that is the subject of the quantization via the input terminal 10 and preprocessing the respective data subject to quantization according to the selected nonlinear function, and supplying the resulting preprocessed data to the linear quantizer 22 from. The linear quantization device 22 performs linear quantization of the preprocessed data to provide the non-linear quantized data that the device in question sends to the output port 14 emits.

In the most general implementation of the embodiment of the non-linear quantization device 60 contains the series of in the nonlinear function block 61 stored nonlinear functions have a non-linear function for each allowed number of quantization levels (or any permissible word length). On via the input connection 11 The input signal received indicates the number of quantization levels to be used for the quantization of the data subject to quantization and becomes the linear quantizer 22 and additionally the non-linear function selector 62 fed. The non-linear function selector selects the input signal indicative of the number of quantization levels (or word length) to be used for the quantization of the data subject to quantization 62 one of the in the nonlinear function block 63 stored nonlinear functions for delivery to the preprocessor 21 out.

In another implementation, the non-linear function selector may 62 be configured so that two or more possible values from the number of quantization levels select the same non-linear function. If a piecewise linear approximation of the nonlinear function is used instead of the nonlinear function itself, however, a different piecewise approximation must be applied for each value of the number of quantization levels, although two or more possible values from the number of quantization levels select the same nonlinear prototype function ,

As mentioned above, the non-linear quantizer embodying the first non-linear quantization method may be replaced by a non-linear quantizer a quantization device embodying the non-linear quantization method according to the invention as a non-linear quantization device 133 in the compressor 111 of in 11 to 13 replaced audio signal transmission or recording system. However, the non-linear quantization method according to the invention is not limited to this application. The non-linear quantization method according to the invention provides the advantages described above when used in other non-linear quantization applications.

When the nonlinear quantization device embodying the non-linear quantization method according to the invention is a non-linear quantization device 133 is used receives the non-linear function selector 62 the word length for the respective sentence or the respective series of spectral components from the bit allocation device 122 via the input connection 11 , and the preprocessor 21 receives the sets of the block glide processed spectral components from the block sliding processor 132 via the input connection 10 , The preprocessor 21 performs preprocessing of the block glide processed spectral components in the respective sentence using the nonlinear function provided by the nonlinear function selector 62 to the word length for the sentence, that of the bit assignment device 122 has been obtained is selected. The linear quantization device 22 performs linear quantization of the resulting preprocessed spectral components in the set to produce a set of nonlinear quantized spectral components that the device in question transmits via the output port 14 to the multiplexer 125 emits.

If the sets of spectral components are frequency-divided into frequency bands prior to quantizing, as mentioned above, the bit allocation means 122 a scaling factor and a word length for quantizing the spectral components in the respective frequency band, and the block sliding processor 132 applies block sliding processing to the spectral components in the respective frequency band. In this case, the non-linear function selector selects 62 for the respective frequency band, a nonlinear function from the set of nonlinear functions included in the nonlinear function set 61 are stored. The non-linear function selector supplies the nonlinear function selected for the particular frequency band to the preprocessor 21 off, and the preprocessor 21 performs preprocessing of the block glide processed spectral components in the respective frequency band using the non-linear function for the frequency band received from the non-linear function selector.

The first and second non-linear quantization methods may also apply different non-linear functions depending on the word length or the number of quantization levels. In the nonlinear quantization device embodying the first non-linear quantization method, disclosed in US Pat 7 This can be done via the input port 11 recorded input signal also the non-linear function block 33 be fed as indicated by the dashed line 34 is indicated, and the non-linear function block 33 can to the quantization value calculator 31 give a non-linear function, which depends on the number of quantization levels or the word length.

In the nonlinear quantization device embodying the second non-linear quantization method, as shown in FIG 14 The data entries in the tables in the set of tables in the table selector can be illustrated 41 stored tables using different nonlinear functions that depend on the number of quantization levels. The selection of the lookup table for the input signal indicating the number of quantization levels (or word length) and which is via the input terminal 11 would then automatically select a non-linear function that differs according to the number of quantization levels.

The application of different non-linear functions in the third non-linear quantization method according to the invention is described above with reference to FIG 20 described.

If the quantized data is quantized using a non-linear function selected according to the number of quantization levels at which the data subject to quantization is quantized, then the quantized data can not be dequantized using a known dequantizer. The reason for this is that the known dequantizer is based on only a single non-linear function. 21 illustrates in a block diagram a non-linear dequantizer 70 , which embodies the non-linear dequantization method according to the invention. In the non-linear dequantizer 70 receives the linear dequantizer 71 the nonlinear quantized data over the input terminal 15 as well as via the input connection 16 a corresponding input signal indicating the number of quantization levels (or word length) indicates with which the nonlinear quantized data is quantized. The input signal indicating the number of quantization levels also becomes the inverse non-linear function selector 74 fed. The linear dequantizer 71 performs dequantization of the nonlinear quantized data according to the input signal indicating the number of quantization levels, and outputs the resulting nonlinear data to the post processor 72 which performs post-processing according to a selected inverse nonlinear function and the resulting dequantized data to the output terminal 17 emits.

As the dequantizer 70 To quantize quantized data that has been quantized using different non-linear functions that have been selected according to the number of quantization levels contains the dequantization device in question 70 the inverse nonlinear function block 73 in which a set of a plurality of inverse nonlinear functions is stored. Each inverse nonlinear function in the set corresponds to one of the nonlinear functions originally in the compressor 111 ( 11 ) have been used to quantize the data that is the subject of quantization. For example, each of the inverse nonlinear functions in the set may correspond to one of the nonlinear functions contained in the nonlinear function block 63 in the quantization device 60 ( 20 ) are stored.

The inverse nonlinear quantization device 70 also contains the inverse non-linear function selector 74 , which is an inverse nonlinear function of the set of inverse nonlinear functions contained in the inverse nonlinear function block 73 are selected using the same selection method as that used to select the nonlinear function in the quantizer. The selected inverse nonlinear function becomes the post or post processor 72 which postprocesses to the non-linear data from the linear dequantizer 71 using the selected inverse nonlinear function. Thus, in the illustrated example, the post or post processor applies 72 post processing on the nonlinear data using the inverse function to the nonlinear function present in the quantizer 60 was used to pre-process the data that is the subject of quantization. The post-processor 72 gives the resulting dequantized data to the output terminal 17 from.

In the most general implementation of the non-linear dequantization device 70 contains the set of inverse nonlinear functions contained in the inverse nonlinear function block 73 are stored, an inverse nonlinear function for each allowable value of the number of quantization levels (or word length). The input signal, which is the linear dequantizer 71 via the input connection 11 and indicating the number of quantization levels (or word length) used to quantize the nonlinear quantized data, additionally becomes the inverse nonlinear function selector 74 fed. In response to the input signal indicating the number of quantization levels used to quantize the non-linear quantized data, the inverse non-linear function selector selects one of the inverse non-linear function blocks 73 stored inverse non-linear functions for delivery to the post-processing block 72 from.

In embodiments where there are fewer inverse nonlinear functions in the set of nonlinear functions than the number of allowable values of the number of quantization levels (or word length), the inverse nonlinear function selector requires 74 nevertheless, that an input signal from the terminal 11 indicates the number of quantization levels (or word length).

22 Fig. 12 is a block diagram showing an example of the application of a dequantizing means embodying the dequantizing method according to the invention in the audio-stretching means 112 of in 11 illustrated audio transmission or recording system 110 , However, the non-linear dequantization method according to the invention is not limited to this application. The method provides the advantages described above when used in other non-linear dequantization applications.

In the stretching device 112 the compressed signal is transmitted from the transmission or recording medium by means of the demultiplexer 225 receive. The demultiplexer 225 It extracts the sets of quantized spectral components and the respective bit allocation parameters from the compressed signal, and also extracts the word length and scaling factor for the respective set of quantized spectral components from the bit allocation parameters. The demultiplexer gives the sets of spectral components to the input port 15 the non-linear dequantizer 233 and delivers the respective word lengths to the input terminal 16 the non-linear quantizer 233 , The demultiplexer also supplies the scaling factors to the block-slip enable processor 232 from.

The non-linear dequantizer 233 takes a non-linear dequantization of the respective set of nonlinear quantized spectral components from the multiplexer 225 using the respective word length before the linear dequantization of the quantized spectral components by the linear dequantizer 71 ( 21 ), and she chooses those by the post-processor 72 used inverse nonlinear function to perform post-processing on the nonlinear spectral components resulting from linear dequantization by the linear dequantizer 71 result.

The sets of the block glide processed spectral components produced by the non-linear quantizer 233 are generated, the Blockgleit Freigabeprozessor 232 which reverses the block sliding processing applied to the respective set of spectral components in the compressor. The block-slip enable processor multiplies the block-sliced spectral components in each set by the scaling factor for the set obtained by the demultiplexer 225 has been obtained to produce a set of reconstructed spectral components.

The block-slip enable processor 232 returns the set of reconstructed spectral components to the time-frequency synthesizer 221 from. The time-frequency synthesizer 221 transforms the respective set of reconstructed spectral components back into the time domain to reconstruct a block of the audio output signal.

If the sets of spectral components before quantization in the compressor 111 in frequency are divided into frequency bands, the demultiplexer extracts 225 each band of the quantized spectral components and their respective scaling factor and the word length from the compressed signal. The non-linear dequantizer 233 performs non-linear dequantization of the quantized spectral components in the respective frequency band using the word length for the band, and the block-slip enable processor 232 resolves the block slip exerted on the respective resulting spectrally-processed spectral components to produce a band of the reconstructed spectral components. The time-frequency synthesizer 221 Then, an orthogonal transform of the reconstructed spectral components in all the frequency bands back into the time domain and synthesizes the resulting blocks of the frequency domain signals to reconstruct a block of the audio output signal.

Claims (28)

A method of nonlinear quantization of data representing an information signal to therefrom generate corresponding quantized data representing the information signal using fewer bits, the respective data being represented by a quantization level (i) in the event that they are non-linearly quantized selected from a number of quantization levels (lev), comprising the steps of: providing a set of nonlinear functions ( 63 ), each defining a non-linear quantization characteristic, receiving ( 10 ) of the data, each having a data value, receiving ( 11 ) word length information indicating the number of quantization levels (lev), selecting ( 62 ) one of the nonlinear functions from the set of nonlinear functions to the word length information, preprocessing ( 21 ) of the data for the generation of preprocessed data by applying the at step ( 62 ) selected nonlinear function on the respective data for generating corresponding preprocessed data and linear quantization ( 22 ) of the preprocessed data on the word length information to produce the quantized data. The method of claim 1, further comprising a step of establishing an allowable range for the number of quantization levels, and wherein at the step of providing ( 63 ) of a set of nonlinear functions, providing a different nonlinear function for each quantization level from the number of quantization levels (lev) in the allowable range for the number of quantization levels (lev). The method of claim 1 or 2, wherein at the step of providing ( 63 ) of a set of nonlinear functions which vary nonlinearity nonlinear functions, and wherein at the selection step ( 62 ) the selected function of the nonlinear functions has greater nonlinearity when the word length information indicates more quantization levels (lev) than the one selected function of the nonlinear functions when the word length information indicates less quantization levels. The method of claim 1 or 2, wherein it additionally comprises a step for determining a zulässi range for the number of quantization levels (lev), wherein at the step of providing ( 63 ) of a nonlinear function set as each of the nonlinear functions in the nonlinear function set is a piecewise linear approximation ( 51 ) is provided to a nonlinear function, providing a piecewise approximation of a nonlinear function for each number of quantization levels (lev) in the allowable range for the number of quantization levels (lev), each piecewise linear approximation ( 51 ) includes a plurality of nodes at which the piecewise nonlinear approximation coincides with the nonlinear function and which are equal in number to the respective number of quantization levels (lev), and comprises linear segments connecting successive nodes of the respective nodes. The method of claim 4, wherein at the step of providing ( 63 ) of a set of nonlinear functions, the nonlinear functions approximated by the piecewise linear approximations ( 51 ) in nonlinearity, where the nonlinear functions approximated by piecewise linear approximations are more nonlinear for larger numbers of quantization levels (lev) than the nonlinear functions represented by piecewise linear approximations for smaller (n) numbers of Quantization levels (lev) are approximated. A method according to claim 4 or 5, wherein in the step of providing a set of non-linear functions ( 63 ) all piecewise linear approximations approximate the same nonlinear function. Method according to claim 1, comprising the method steps: providing ( 43 . 41 ) of a set of quantization tables derived from said set of nonlinear functions, the set of quantization tables comprising a quantization table for each possible number of quantization levels (lev), each quantization table comprising a table entry for each quantization level (lev) in the corresponding number contains the quantization level (lev), select ( 41 ) of the quantization table from the set of quantization tables for the number of quantization levels indicated by the word length information and selecting ( 42 ) of the quantization level from the quantization table selected at the selection step, the table entry of which corresponds to the data value of the respective data as one of the quantized data. The method of claim 7, wherein at the step of providing ( 43 . 41 ) of a set of quantization tables, the table entries in the quantization tables provide a different non-linear quantization characteristic in different tables of the quantization tables. The method of claim 8, wherein at the step of providing ( 43 . 41 ) of a set of quantization tables, the table entries in the quantization tables for more quantization levels (lev) provide a more non-linear quantization characteristic than the table entries in the quantization less quantization tables (i). The method of claim 7, wherein at the step of Providing a set of quantization tables of each quantization level (lev) has a corresponding quantization value, in which each table entry in the quantization tables has an input area for the includes respective quantization level (lev), where the input area Values between decision values in the middle between the quantization value the quantization level and the quantization values of adjacent ones Includes quantization levels, in which the selection step sets the quantization level of the table entry selects and wherein the input area is the data value of the respective one data includes. The method of claim 7, wherein in the step of providing a set of quantized tables, each quantization level (lev) has a corresponding quantization value and each table entry in the quantization tables has the quantization value for the quantization level (lev) concerned, and wherein the step of selecting ( 42 ) selects the quantization level in the table entry, where the quantization value in value is closest to the data value of the respective one data. Method according to one of claims 1 to 11, wherein it additionally comprises a step for carrying out a time-frequency analysis ( 111 ) with respect to the information signal for generating the data. The method of claim 12, wherein the information signal comprises a stream of a plurality of samples, wherein in the step of performing the time-frequency analysis ( 111 ) the time-frequency analysis is performed with respect to a block of the samples of the information signal ( 132 ) to generate data, and wherein the method additionally includes a step of deriving ( 122 ) of the word length information obtained in the step of receiving the word length information received from the block of samples of the information signal. Method according to claim 12 or 13, wherein the step of execution ( 111 ) the time-frequency analysis comprises the steps of: deriving ( 132 ) of a set of spectral components from the block of samples of the information signal, deriving ( 122 ) of a scaling factor from the set of spectral components, applying a block glide ( 122 ) to the set of spectral components using the scale factor to generate the data, and wherein the step of deriving ( 122 ) the word length information comprises a step of deriving the word length information from the set of spectral components. The method of any one of claims 12 to 14, wherein the information signal comprises one of a plurality of samples, the step of performing the time-frequency analysis (10). 111 ) comprises a step of orthogonally transforming a block of the samples of the information signal to generate the data. The method of claim 15, wherein the step of orthogonally transforming a bucket of the samples to generate the data comprises the steps of: orthogonal transforming ( 153 . 155 ) of a block of the samples for generating spectral components divided into frequency bands, deriving a scaling factor for the spectral components in one of the frequency bands, applying block gliding to the spectral components in the one frequency band of the frequency bands to the scaling factor, and providing the resulting, by block slipping processing spectral components as data, and wherein the method additionally comprises a step of deriving the word length information received in the step of receiving the word length information from the spectral components of the one frequency band of the frequency bands. The method of any one of claims 12 to 14, wherein the step of performing the time-frequency analysis ( 111 ) a step for filtering ( 153 . 155 ) of the information signal for generating a sequence of time-domain spectral components in a sub-band of a plurality of sub-bands as data. The method of claim 17, wherein the step of filtering ( 153 . 155 ) comprises the steps of deriving a scaling factor for a sequence of time domain spectral components in the one subband of the plurality of subbands, applying block slippage to the sequence of time domain spectral components in the one subband of the plurality of subbands to the scaling factor, and providing the resultant block-processed spectral components as data, the method further comprising a step of deriving the word length information received in the step of receiving the word length information from the sequence of the time-domain spectral components in the one frequency band of the frequency bands. The method of claim 1, wherein the method additionally comprises a block clutter application step ( 123 ) in the data for generating data processed by block sliding, and wherein at the preprocessing step ( 21 ) the preprocessed data are generated by the fact that in the selection step ( 62 ) selected non-linear function is applied to all of the data processed by block sliding. Method according to claim 7, wherein it additionally comprises a Step to Apply Block Gliding to Generation Data the data processed by block gliding comprises the One block glide processed data one by block gliding have processed data value, and wherein in the selecting step the quantization level, the table entry of which by block slip processed data value of the respective processed by block sliding Data equals when appropriately quantized data is selected. A method for dequantizing nonlinear quantized data for generating dequantized data therefrom, wherein the nonlinear quantized data is derived by nonlinear quantization of original data representing an information signal corresponding to a selected one of a plurality of nonlinear functions, wherein the nonlinear quantized data is the Representing information signals using fewer bits than the original data, the respective nonlinear quantized data having a quantization level selected from a number of quantization levels, comprising the steps of: providing ( 73 ) of a set of inverse nonlinear functions, each corresponding to one of the plurality of nonlinear functions, receiving ( 15 ) of nonlinear quantized data, receiving ( 16 ) word length information indicating the number of quantization levels, linear dequantization ( 71 ) of the nonlinear quantized data to the word length information Generation of nonlinear data, Select ( 74 ) one of the inverse nonlinear functions from the set of nonlinear functions and postprocessing ( 72 ) of the nonlinear data for generating the dequantized data by applying the inverse nonlinear function selected at the selecting step to the respective nonlinear data to generate corresponding dequantized data. The method of claim 21, wherein in the selecting step ( 74 ) one of the inverse nonlinear functions is selected in response to the word length information. The method of claim 21 or 22, further comprising a step of establishing an allowable range for the number of quantization levels, and wherein at the step of providing ( 73 ) of a set of inverse nonlinear functions, a different inverse nonlinear function is provided for each number of quantization levels in the allowable range for the number of quantization levels. The method of any one of claims 21 to 23, wherein at the step of providing ( 73 ) of a set of inverse nonlinear functions, the inverse nonlinear functions in the nonlinearity vary, and wherein at the selection step ( 74 ) a selected one of the inverse nonlinear functions has greater nonlinearity if the word length information indicates more quantization levels than the one of the selected functions of the inverse nonlinear functions if the word length information indicates less quantization levels. Method according to one of claims 21 to 24, wherein it additionally comprises a step for carrying out a time-frequency synthesis ( 221 ) with respect to the dequantized data to recover the information signal. The method of claim 25, further comprising the steps of: receiving a scaling factor and before executing the time-frequency synthesis step ( 121 ) Enabling the block sliding on the scaling factor applied to the dequantized data to generate block-released data, and in the time-frequency synthesis step (FIG. 121 ) the time-frequency synthesis is applied to the data released by block sliding. The method of claim 25 or 26, wherein the information signal comprises a stream of a plurality of samples, and wherein in the time-frequency synthesis step (16) 121 ) the application of the time-frequency synthesis to the block-slipped data recovers one block of the samples of the information signal. Method according to one of Claims 21 to 27, additionally comprising a step for applying a block-slip release ( 132 ) to the dequantized data on execution of the post-processing step.