United States Patent [19J
`Wiese
`
`I lllll llllllll Ill lllll lllll lllll lllll lllll 111111111111111111111111111111111
`US005991715A
`[11] Patent Number:
`[45] Date of Patent:
`
`5,991,715
`Nov. 23, 1999
`
`[54] PERCEPTUAL AUDIO SIGNAL SUBBAND
`CODING USING VALUE CLASSES FOR
`SUCCESSIVE SCALE FACTOR
`DIFFERENCES
`
`[75]
`
`Inventor: Detlef Wiese, Neufahrn, Germany
`
`[73] Assignee: Institut Fiir Rundfunktechnik GmbH,
`Miinchen, Germany
`
`[21] Appl. No.: 08/521,817
`
`[22] Filed:
`
`Aug. 31, 1995
`
`Related U.S. Application Data
`
`[63] Continuation of application No. 08/094,028, filed as appli(cid:173)
`cation No. PCT/EP91/01211, Jun. 27, 1991, abandoned.
`Foreign Application Priority Data
`
`[30]
`
`Jan. 26, 1991
`
`[DE]
`
`Germany ............................. 41 02 324
`
`Int. Cl.6
`........................................................ GlOL 7/04
`[51]
`[52] U.S. Cl. .......................... 704/204; 704/205; 704/227;
`704/503
`[58] Field of Search ..................................... 704/204, 205,
`704/227, 503
`
`[56]
`
`References Cited
`
`U.S. PATENT DOCUMENTS
`
`8/1984 Beraud et al. .......................... 704/212
`4,464,783
`4,972,484 11/1990 Theile et al. ............................ 704/227
`
`FOREIGN PATENT DOCUMENTS
`
`064119A3 11/1982 European Pat. Off ..
`3328344Cl 12/1984 Germany .
`
`3639753C2
`
`6/1988 Germany .
`
`OTHER PUBLICATIONS
`
`Plenge et al., "Combined Channel Coding and Conceal(cid:173)
`ment," IEE Colloquim on Terrestrial DAB - Where is it
`Going?, (Digest No. 042) p. 3/1-8, Feb. 17, 1993.
`
`Primary Examiner-David R. Hudspeth
`Assistant Examiner-Talivaldis Ivars Smits
`Attorney, Agent, or Firm-Venable; George H. Spencer;
`Robert Kinberg
`
`[57]
`
`ABSTRACT
`
`A method of transmitting digitized block coded audio sig(cid:173)
`nals includes forming scale factors of the digitized audio
`signals. The n(k-1) differences are formed from k succes(cid:173)
`sively in-time scale factors for each frequency sub-band or
`for a group of spectral values of the audio signal. The n(k-1)
`differences are grouped into at least two value classes. New
`scale factors are selected for each of the n sub-bands or
`spectral value groups based on a sequence of n(k-1) value
`classes. Identifying information, including the control infor(cid:173)
`mation indicating at which locations in the sequence of
`n(k-1) value classes the selected new scale factors are
`disposed, is associated with each sequence of n(k-1) value
`classes. The associated selected new scale factors are
`assigned to each sequence of the sampled signal values and
`to the identifying information associated with each sequence
`of sampled signal values. A transmission pattern of new
`scale factors is determined separately for each of the n
`sub-bands or spectral value groups. Lastly, audio signals are
`regenerated from the sampled signal values and from the
`assigned selected new scale factors.
`
`2 Claims, 1 Drawing Sheet
`
`FORMING SCALE FACTORS
`
`51
`
`52
`
`FORMING n(k-1) DIFFERENCES
`
`53
`GROUPING THE DIFFERENCES INTO VALUE
`CLASSES
`
`54
`SELECTING SCALE FACTORS BASED ON A
`SEQUENCE OF VALUE CLASSES
`
`FORMING CONTROL INFORMATION
`
`S5
`
`56
`ASSOCIATING CONTROL INFORMATION
`WITH IDENTIFYING INFORMATION
`
`57
`ASSIGNING SELECTED SCALE FACTORS TO
`EACH SEQUENCE OF SIGNAL VALUES
`
`SS
`DETERMINING A TRANSMISSION PATTERN
`
`REGENERATING AUDIO SIGNALS
`
`59
`
`IPR2016-01710
`UNIFIED EX1021
`
`

`
`U.S. Patent
`
`Nov. 23, 1999
`
`5,991,715
`
`51
`
`52
`
`FORMING SCALE FACTORS
`
`1 r
`
`FORMING n(k-1) DIFFERENCES
`
`' r
`
`53
`GROUPING THE DIFFERENCES INTO VALUE
`CLASSES
`
`, ..
`54
`SELECTING SCALE FACTORS BASED ON A
`SEQUENCE OF VALUE CLASSES
`
`1 r
`
`55
`
`FORMING CONTROL INFORMATION
`
`1 •
`
`56
`ASSOCIATING CONTROL INFORMATION
`WITH IDENTIFYING INFORMATION
`
`' .
`
`57
`ASSIGNING SELECTED SCALE FACTORS TO
`EACH SEQUENCE OF SIGNAL VALUES
`
`58
`DETERMINING A TRANSMISSION PATTERN
`
`1'
`
`1 '
`
`59
`
`REGENERATING AUDIO SIGNALS
`
`FIGURE 1
`
`

`
`5,991,715
`
`1
`PERCEPTUAL AUDIO SIGNAL SUBBAND
`CODING USING VALUE CLASSES FOR
`SUCCESSIVE SCALE FACTOR
`DIFFERENCES
`
`This application is a Continuation of application Ser. No.
`08/094,028, filed Jul. 26, 1993 (now abandoned), which is a
`371 of PCT/EP91/01211 Jun. 6, 1991.
`
`BACKGROUND OF THE INVENTION
`
`2
`disposed, assigning the associated selected new scale factors
`to each sequence of the sampled signal values and to the
`identifying information associated with each sequence of
`sampled signal values, determining a transmission pattern of
`5 new scale factors separately for each of the n sub-bands or
`spectral value groups based on psychoacoustic aspects with
`respect to pre-masking and post-masking effects of a human
`auditory system with a distinction being made between
`10 psycho-acoustically relevant changes in the scale factors,
`and regenerating audio signals from the sampled signal
`values and from the assigned selected new scale factors.
`
`1. Field of the Invention
`The invention relates to a method of transmitting digitized
`block coded audio signals using scale factors formed during
`block coding of the digitized audio signals based on a peak
`value of a sequence of signal values of the digitized audio 15
`signals, especially during irrelevance and redundance reduc(cid:173)
`ing methods. Such a method is disclosed in German Patent
`3,328,344.
`2. Description of the Related Art
`It is known (German Patent 3,328,344) to transmit digi(cid:173)
`tized audio signals by forming a scale factor during block
`coding from the amount of the peak value of a sequence
`(block) of signal values, with this scale factor indicating in
`which one of several magnitude ranges the amplitude of the 25
`peak value lies. In addition, as disclosed in German Patent
`3,639,753, the sampled signal values may be represented in
`a plurality of spectral sub-band signals and the sampled
`values of the individual sub-bands may be changed as
`determined by the respective masking threshold of the 30
`human auditory system in the sense of a reduction of
`irrelevance and redundance. Instead of dividing the digitized
`audio signal into spectral sub-bands, it is also possible to
`subject the digitized audio signal to a Fourier analysis and to
`determine the scale factors for groups of spectral values as 35
`well as to perform a reduction of irrelevance and redun(cid:173)
`dance.
`
`BRIEF DESCRIPTION OF THE DRAWING
`
`The invention will be described in greater detail below
`with reference to the sole FIGURE showing a flow chart of
`a method according to the invention for transmitting digi-
`20 tized block coded audio signals.
`
`DETAILED DESCRIPTION OF THE
`PREFERRED EMBODIMENT
`
`The invention will now be described in greater detail with
`reference to an embodiment thereof.
`
`The bit saving scale factor transmission is based on the
`detection and coding of patterns, thus eliminating redun(cid:173)
`dance as well as irrelevance.
`
`The input matrix is composed of k columns and n rows,
`with k representing the number of scale factors that are
`successive in time and n representing the number of sub(cid:173)
`bands (step Sl of the FIGURE). The description hereinafter
`generally is given in the left-hand column while an example
`is given in the right-hand column. The example (right)
`relates to three successive scale factors comprising a total of
`32 sub-bands (hereinafter called a frame). The scale factors
`are quantized with 6 bits and are thus able to take on 26=64
`possible values. A block for which a scale factor is formed
`has a length of 8 ms.
`
`Input matrix:
`
`1scf11 scf12 ... scf1kl I 1scf11 scf12 ... scf13 13 * 6 bits= 18 bits
`
`I scf21
`
`I 1 scf21
`
`3 * 6 bits = 18 bits
`
`lscfn1 ....... scfnkj I lscf321 ...... scf32J 3 * 6 bits= 18 bits
`
`total frame 32 * 18 bits = 576 bits
`
`Now the differences d of successive values of each row
`and/or column are calculated (step S2). They are stored in an
`intermediate matrix. The magnitude of the differences is a
`function of the absolute block length (row) or of the width
`of the sub-bands (column). Hereinafter only timely succes(cid:173)
`sive scale factors are considered, for example:
`
`d1 •21 =sc/12-sc/11 lditto
`
`The intermediate matrix has k-1 columns (and when
`considering the columns, also n-1 rows).
`
`SUMMARY OF THE INVENTION
`
`In contrast thereto it is the object of the invention to 40
`perform an information reduction also for the scale factors
`by means of which the data rate which is necessary for a
`digital audio transmission is further reduced.
`This is accomplished according to the invention by a
`method of transmitting digitized block coded audio signals 45
`including the steps of forming scale factors of the digitized
`audio signals based on a peak value of a sequence of
`sampled signal values of the digitized audio signals, forming
`n(k-1) differences from k successively in-time scale factors
`for each frequency sub-band or for a group of spectral values 50
`of the audio signal with n being greater than or equal to 1,
`grouping the n(k-1) differences into at least two value
`classes with each value class including a quantity of at least
`one possible difference of scale factors, selecting new scale
`factors for each of the n sub-bands or spectral value groups 55
`based on a sequence ofn(k-1) value classes with the number
`of successive different selected scale factors within each
`sequence of n(k-1) value classes being less than or equal to
`a number of successive different scale factors of each
`sub-band or spectral value group, associating identifying 60
`information with each sequence of n(k-1) value classes, the
`identifying information identifying an association of each of
`the selected new scale factors with at least one of the k
`sequences of the sampled signal values for each respective
`sub-band or spectral value group and including control 65
`information indicating at which locations in the sequence of
`n(k-1) value classes the selected new scale factors are
`
`

`
`3
`
`Intermediate matrix:
`
`5,991,715
`
`4
`one, two or three scale factors must be transmitted and a
`control information of 2 bits.
`
`5 Class
`Sequence
`
`Cambi-
`nation
`
`Transmission
`Pattern
`
`Number of
`SCF
`
`Control
`Inform.
`
`c,
`C2
`C3
`C4
`
`Co
`C7
`
`scfl scf2 scf3
`scfl scf2
`scfl scf2
`scfl scf3
`
`3 (18 bit)
`2 (12 bit)
`
`11 (2 bit)
`01 (2 bit)
`
`scfl scf3
`scfl
`
`2 (6 bit)
`1 (6 bit)
`
`10 (2 bit)
`00 (2 bit)
`
`L 1 L 1
`L 1 L 2
`L 1 L 3
`10 L, L4
`L1 Ls
`L 2 L 1
`L2 L2
`L2 L3
`L2 L4
`L2 Ls
`15 L3 L,
`
`Ls Ls
`-,
`
`20
`
`With the aid of a table, the possible differences D 1 are
`grouped into L classes. One class 1 is a quantity of one or
`more differences D. For the example on the righthand side,
`the differences D are assigned to five classes L (step S3),
`with one or several possible differences belonging to one
`class:
`
`L1 = {01 , 0 2 . . . 0 0 }
`Lo = {07' 0 8 . . . 0 12}
`L, = {013, 0 14, ... 0 17}
`L 4 = {0 18}
`
`if d < -2 and d > -64
`if d > -3 and d < 0
`if d = 0
`if d < 3 and d > 0
`if d > 2 and d < 64
`
`D,}
`
`All possible sequences of classes L result m (k-lf
`possibilities for combination c (step S4):
`
`CJ
`
`=L1L1 ... L1
`
`=L1 Li ... L1
`
`c(k-1) L = Lk-1 Lk-1 ... Lk-1
`
`25
`
`30
`
`35
`
`Four possible transm1ss10n patterns and thus a control
`information of 2 bits result for this example:
`
`1. three different scale factors
`
`sl s2 s3
`
`00
`
`data flow for a frame (3 * 6 bit+2 bit) * 32=640 bits;
`2. the first scale factor for the first position, the second
`scale factor for the second and third positions
`
`sl s3
`sl s2
`
`01
`
`data flow for a frame (2 * 6 bit+2 bit) * 32=448 bits;
`3. the first scale factor for the first and second positions,
`the third scale factor for the third position
`
`s2 s3
`sl s3
`
`10
`
`data flow for a frame (2 * 6 bit+2 bit) * 32=448 bits;
`4. one scale factor for all three positions
`
`sl
`s2
`s3
`
`11
`
`Data reduction is realized in that the combinations c are 40
`each assigned a scale factor transmission pattern. This
`pattern is composed of a control information and a sequence
`of scale factors (steps S6 and S7), with the quantity of
`different scale factors within a sequence being smaller than
`or equal to the quantity of the different scale factors from the 45
`input matrix. The transmission patterns evolving from the
`classified differences are determined (step SS) on the basis
`of statistical knowledge of the signal and according to
`psycho-acoustic aspects with reference to the pre-masking
`and post-masking effects of the human auditory system. If, 50
`for example, the scale factors of a row in the input matrix do
`not change, it is not necessary to transmit all scale factors
`since this is redundant information (redundance ). Ascending
`scale factors must be transmitted more accurately than
`descending scale factors since it is known from psycho- 55
`acoustics that the human auditory system exhibits distinct
`post-masking in a range up to 200 ms, but pre-masking only
`in a range up to 20 ms (irrelevance). The assigned selected
`scale factors are then used to regenerate the audio signal
`(step S9).
`If an input matrix has, for example, three columns, and it
`is considered row by row, two differences result that must be
`classified and 25 possible combinations c. Each combination
`c is assigned a scale factor transmission pattern. This results
`in a certain number of scale factors to be transmitted and in 65
`a control information at which position the scale factors are
`located or change (step SS), respectively. In this example,
`
`data flow for a frame (1 * 6 bit+2 bit) * 32=256 bits.
`If one considers only three timely successive scale factors
`(in the example the frame corresponds to exactly 24 ms), the
`advantage of a low decoder delay and a small access unit
`60 (smallest unit to be decoded) results. If these advantages are
`not important, a greater data reduction can be realized as
`follows if larger time sections are considered:
`by means of an input matrix that has a larger number of
`columns;
`by transmitting the information "do not transmit scf' for
`a frame.
`
`

`
`5,991,715
`
`5
`
`Numerical example:
`
`successive scale factors:
`calculated differences:
`assigned class:
`transmission pattern:
`transmitted scale factor(s):
`control information:
`decoded scale factors:
`
`10
`
`40
`
`38
`-2
`30
`L2
`Ls
`scfl scf2 scf2
`10,
`40
`01
`40
`
`10
`
`36
`
`37
`
`38
`-1
`+2
`L4
`L2
`max( scfl, scf2, scf3)
`38
`
`00
`38
`
`38
`
`38
`
`5
`
`10
`
`6
`(k-1) value classes being less than or equal to a
`quantity of successive different scale factors of each
`sub-band or spectral value group;
`assigning identifying information to each sequence of
`(k-1) value classes, the identifying information iden(cid:173)
`tifying an association of each of the selected new scale
`factors with at least one of the k sequences of the
`sampled signal values for each respective sub-band or
`spectral value group and including control information
`indicating at which locations in the sequence of (k-1)
`value classes the selected new scale factors are dis(cid:173)
`posed;
`determining a transmission pattern of new scale factors
`separately for each of the n sub-bands or spectral value
`groups based the selected new scale factors and the
`assigned identifying information and based on psy(cid:173)
`choacoustic aspects with respect to pre-masking and
`post-masking effects of a human auditory system with
`a distinction being made between psycho-acoustically
`relevant changes in the scale factors;
`transmitting the determined transmission pattern of the
`new scale factors;
`receiving the transmitted new scale factors; and
`regenerating audio signals from the sampled signal values
`and from the assigned selected new scale factors.
`2. A method according to claim 1, wherein the control
`information indicates when no new scale factor is being
`transmitted, that a preceding new scale factor is applicable
`30 for all relevant k sequences of sampled signal values.
`
`I claim:
`1. A method of transmitting digitized block coded audio
`signals comprising the steps of:
`forming nK scale factors of the digitized audio signals 15
`based on a peak value of a sequence of sampled signal
`values of the digitized audio signals, where n is the
`number of frequency sub-bands or groups of spectral
`values of the digitized audio signals and n is greater
`than or equal to 1, and K is the number of successive 20
`in-time scale factors for each frequency sub-band or for
`a group of spectral values of the audio signals;
`forming n(k-1) differences from the k successively
`in-time scale factors for each frequency sub-band or for
`a group of spectral values of the audio signals;
`grouping the n(k-1) differences into at least two value
`classes with each value class including at least one
`possible difference of scale factors;
`selecting new scale factors for each of the n sub-bands or
`spectral value groups based on a sequence of (k-1)
`value classes, a quantity of successive, different
`selected new scale factors within each sequence of
`
`25
`
`* * * * *