EP2380169B1 - Method and means for the scalable improvement of the quality of a signal encoding method - Google Patents

Description

The invention relates to a method for the scalable improvement of the quality of a signal coding method. In digital communication systems, to reduce the required data rates, the audio signals to be transmitted are compressed by means of coding methods and decompressed after the transmission.

Such a coding method, which is intended for transmission of a speech signal in a frequency range of 300 to 3400 Hz with a data rate of 8 kbit / s, is known, for example, from ITU-T Recommendation G.729.

For transmission of a higher quality, the transmission of an extended frequency range of 50 Hz to 7000 Hz is known. A designated, so-called broadband voice codec is described, for example, in ITU-T Recommendation G.722.EV.

The method uses the so-called Subband-Adaptive Differential Pulse Code Modulation (SB-ADPCM) for encoding audio signals, which in the ITU-T Recommendation G.722, "7 kHz Audio Coding within 64 kbit / s"; G.722 (11/88), International Telecommunication Union, Geneva, CH, November 25, 1988 , XP017400870.

To further increase the quality of the transmitted audio signal, there is a demand for a scalable coding method.

On the one hand, scalability allows backward compatibility with conventional decoding methods and on the other hand, it offers a simple possibility, in the case of limited data transmission capacities in the transmission channel, to carry out a transmitter and receiver-side adaptation of the data rate and the size of transmitted data frames.

The invention has for its object to provide a method for scalable improvement of the quality of a coding method according to the Subband Adaptive Differential Pulse Code principle.

aufgebaut, wobei off = [0 .. M-1], den Abstand des ersten Impulses von dem Beginn des Vergleichsabschnittes angibt, α _p ∈{α₀,α₁,···,α _L-1} den Amplitudenwert , M den Abstand zwischen zwei einzelnen Pulsen, N die Anzahl der Pulse und L die Anzahl der unterschiedlichen Pegel ά ;
Die Information über das Referenzsignal mit dem minimalen Fehlersignal wird übertragen.The object is achieved by claim 1 with a method for scalable improvement of the quality of a coding method according to IT-U Recommendation G.722 with the following method steps,
In an iterative process with a number of repetition steps depending on the extent of the extension, a digital error signal determined from an input signal to be coded and a prognosis signal is compared in sections with a number of M * L ^N different reference signals, and from this the reference signal with a minimum error signal with respect to a predetermined one Error criterion determined,
the reference signals c (n) are each made of equidistant Dirac pulses δ ( n ) according to $$c\left(n\right)={\displaystyle \underset{p=0}{\overset{N-1}{\Sigma }}}{\alpha }_p\cdot \delta \left(n-\mathit{off}-M\cdot p\right)$$

where off = [0 .. M-1] , indicates the distance of the first pulse from the beginning of the comparison section, α _p ∈ {α ₀ , α ₁ , ···, α _{L -1} } the amplitude value, M den Distance between two individual Pulses, N the number of pulses and L the number of different levels ά;
The information about the reference signal with the minimum error signal is transmitted.

ermittelt wird und wenn über den Zeitraum des Vergleichabschnittes gemäß $$E_n={\displaystyle \sum _{n=0}^{\mathit{Ma}}}{\mathrm{e}}_{\mathrm{H}\mathrm{1}}{\left(\mathrm{n}\right)}^2$$

ein Fehlerbetrag ermittelt und zur Bestimmung des minimalen Fehlersignales herangezogen wird.It is advantageous if an extended error signal e _H1 (n) according to $${\mathrm{e}}_{\mathrm{H}\mathrm{1}}\left(\mathrm{n}\right)={\mathrm{e}}_{\mathrm{H}}-\mathrm{c}\left(\mathrm{n}\right)$$

determined and if over the period of the comparison section according to $$e_n={\displaystyle \underset{n=0}{\overset{\mathit{Ma}}{\Sigma }}}{\mathrm{e}}_{\mathrm{H}\mathrm{1}}{\left(\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="imgf0001.png"\; state="\{\{state\}\}">n</figure-callout>}\right)}^2$$

an error amount is determined and used to determine the minimum error signal.

Also advantageous is an arrangement for carrying out the inventive method according to claim 3, in which in addition to a conventional encoder (ADPCM) according to the subband Adaptive Differential Pulse Code principle according to IT-U Recommendation G.722 means for generating reference signals are provided for each Stage of the extension in each case a signal generator EHDS1, ... EHDSS for generating the reference signals c (n) and a control unit CB 1, ... CB S have.

In the following, an embodiment of the method according to the invention will be explained with reference to the figures.

• Fig.1 den Aufbau eines erfindungsgemäßen Referenzsignales
• Fig. 2 die Struktur eines erfindungsgemäßen Codecs und
• Fig. 3 die Struktur eines erfindungsgemäßen Decoders.

They show by way of example:

• Fig.1 the structure of a reference signal according to the invention
• Fig. 2 the structure of a codec according to the invention and
• Fig. 3 the structure of a decoder according to the invention.

The reference signal according to Fig. 1 comprises a number of N Dirac pulses δ ( n ). The distances between the individual pulses are each M sampling periods, the distance of the first pulse δ (1) from the beginning of the comparison section is off = [0 .. M-1] sampling periods. The Dirac pulses may have a predetermined number of amplitude values L.

The mathematical definition of a reference signal is as follows: $$c\left(n\right)={\displaystyle \underset{p=0}{\overset{N-1}{\Sigma }}}{\alpha }_p\cdot \delta \left(n-\mathit{off}-M\cdot p\right)$$

By varying the parameters amplitude value α with different values of L and offset off = [0 .. M-1] will be generated by different reference signals a group having a number M · L ^N.

The comparison according to the invention of the thus obtained reference signals c (n) is based on the Figures 2 and 3 explained in more detail. Fig. 2 shows the structural structure of an encoder according to the invention, which in addition to a conventional encoder ADPCM according to the subband Adaptive Differential Pulse Code principle according to IT-U Recommendation G.722 means for generating reference signals comprising for each stage of the extension in each case a signal generator EHDS1, .. EHDSS for generating the reference signals c (n) and a control unit CB 1, ... CB S has.

According to the invention, the reference signals c (n) are compared over a given time interval, a so-called frame, with a digital error signal e _H , which in a conventional coding method according to IT-U Recommendation G.722 was determined from an input signal to be coded and a prognosis signal.

ein Fehlerbetrag ermittelt wird.In accordance _therewith , according to e _H1 (n) = e _H -c (n), an extended error signal e _H1 (n) results at that over the period of the comparison section according to FIG $$e_n={\displaystyle \underset{n=0}{\overset{\mathit{Ma}}{\Sigma }}}{\mathrm{e}}_{\mathrm{H}\mathrm{1}}{\left(\mathrm{<figure-callout\; id="2"\; label="n"\; filenames="imgf0001.png"\; state="\{\{state\}\}">n</figure-callout>}\right)}^2$$

an error amount is determined.

By means of the control unit CB 1, ... CB S, the reference signal c (n) with the lowest error amount E _n is now determined and the information about this signal is transmitted as additional information I _H1min ,... I _HSmin and used in the receiver for decoding the useful signal ,

The following parameters have proven useful in practice for the construction of the reference signal c (n).

The starting point is a sampling rate of 8 KHz and thus a duration of a sampling interval of 125 μsec. The duration of a comparison section is 5 msec, the number of possible amplitude values L for the Dirac pulses is 2. The number of Dirac pulses itself in a comparison section is N = 5. The distance between every 2 Dirac pulses is M = 8 sampling intervals ,

The above-described comparison process of the reference signals c (n) with the digital error signal e _H is iteratively repeated several times in dependence on the selected scaling, which is described in the Fig.2 for the repeating process by a function block with signal generator EHDSS, control unit CB S and additional information _signal I _{HSmin is} shown.

For the first repetition step, this means that the reference signals c (n) are compared with the extended first error signal e _H1 (n) and an extended second error signal E _H2 (n) is generated therefrom. This process is typically repeated four times.

Fig. 3 shows the structure of a decoder according to the invention in which the audio signal is obtained from the received signal I _H , I _H1 , I _H2 ... I _HS . The received signal comprises, in addition to the output signal I _{H of} the conventional coder ADPCM, the additional information I _H1min ,... I _HSmin obtained with the invention as a function of the number of expansion _{stages selected} in the transmitter.

In this case, a significant advantage is also that not all information that is present in the received signal, actually need to be evaluated. It is thus possible that a receiver with only one conventional decoder core decoder receives a signal which also contains the additional information I _H1min ,... I _HSmin , but does not use it for the extraction of the audio signal

This possibility is called backward compatibility.

In the case of a receiver which includes the inventive expansion stages EDS1, EDS2,... EDSS for decoding the additional information I _H1min ,... I _HSmin , the signal is decoded in full quality, unless a restriction is required for other reasons.

Claims (3)

1. A method for scalable improvement of the quality of an encoding method, in which the Subband Adaptive Differential Pulse Code Modulation (SB-ADPCM) for encoding audio signals pursuant to IT-U Recommendation G.722 is used, comprising the following method steps,

   - comparing a digital error signal (e_H), derived from an input signal to be encoded and a prognosis signal, in sections to a number of M*L^N different reference signals c(n) in an iterative process having a number of repeated steps depending on a scope of an expansion, and deriving from said comparison a reference signal having a minimum error signal with respect to a prescribed error criterion,

   - each of the reference signals c(n) is made up of equidistant Dirac impulses δ(n) according to the formula $$c\left(n\right)={\displaystyle \sum _{p=0}^{N-1}}{\alpha }_p\cdot \delta \left(n-\mathit{off}-M\cdot p\right)$$

   

   wherein off = [0.. M-1] indicates the distance of the first impulse from the beginning of the comparison segment,
   α _P ∈{α₀,α₁,···,α _L-1} indicates the amplitude value, M the distance between two individual pulses, N the number of pulses, L the number of different levels α; transmitting the information (I_Himin, ... I_HSmin) about the reference signal with the minimum error signal.
2. The method according to Claim 1, wherein an expanded error signal e_H1(n) is determined as an error criterion according to e_H1(n) = e_H- c(n), and an error amount is determined over the period of the comparison segment according to $$E_n={\displaystyle \sum _{n=0}^{\mathit{Ma}}}{\mathrm{e}}_{\mathrm{H}\mathrm{1}}{\left(\mathrm{n}\right)}^2$$

   

   and is used to determine the minimum error signal.
3. An arrangement designed for carrying out the method according to Claims 1 or 2, wherein in addition to a conventional encoder (ADPCM) operating according to the Subband Adaptive Differential Pulse Code principle pursuant to IT-U Recommendation G.722, means are provided for generating reference signals which, for each step of the expansion, have a signal generator (EHDS 1 ,... EHDSS) for generating the reference signals c(n), and a control unit (CB 1, ... CB S).

Images