Method and unit for substracting quantization noise from a PCM signal

1. A method for substracting quantization noise from a pulse code modulated PCM signal being segmented into frames, comprising the steps of:

calculating for each frame of said PCM signal a constant quantization noise level Bq according to the following equation:

B

q

=

∑

n

=

0

W

-

1

⁢

{

(

s

min

*

⁡

[

n

]

-

s

max

*

⁡

[

n

]

)

·

w

⁡

[

n

]

}

2

12

wherein

n: indicates a specific sample of the PCM signal; S* min [n] : represents the minimum quantization noise level for a specific sample value s*[n] of said PCM signal; S* max [n] : represents the maximum quantization noise level for the specific sample value s*[n] of the PCM signal, wherein S* min [n] −S* max [n] has a different value for at least two specific samples n respectively;

w[n]: represents a window-function; and

W: represents the number of samples per window; and

substracting the quantization noise as represented by said quantization noise level Bq from said PCM signal.

2. The method according to claim 1 , characterized in that the minimum quantization level S* min as well as the maximum quantization level S* max are known.

3. The method according to claim 1 , characterized in that the minimum quantization level S* min and the maximum quantization level S* max are predicted according to the following equations:

S* min =i[n ]−( i[n]−i min [n] )/2

S* max =i[n ]+( i max [n]−i[n ])−/2

wherein

i: represents one out of a plurality of possible representation levels predefined due to the specific PCM quantization method applied to an original signal;

i[n]: represents that predefined representation level which corresponds to the sample value s*[n] for a specific n;

i min [n]: represents that representation level which is—started from i[n]—the next smaller non-zero representation level for which u[n]=1;

i max [n]: represents that representation level which is—started from i[n]—the next bigger non-zero representation level for which u[n]=1;

with the usage array u[i] being defined to:

u

⁡

(

i

)

=

min

(

1

,

∑

n

=

0

L

-

1

⁢

{

0

,

s

*

⁡

[

n

]

≠

i

1

,

otherwise

)

,

-

2

N

-

1

≤

i

<

2

N

-

1

wherein

L: represents the number samples of the whole PCM-signal; and

N: represents the number of bits used for quantizing an original sample value by using PCM to generate the PCM sample values s*[n].

4. The method according to claim 1 , characterized in that the substracting of the quantization noise represented by said quantization noise level Bq from the PCM-signal is carried out in the frequency domain according to the following steps:

computing the spectrum S*[k] of the PCM signal s*[n] and forming the magnitude |S*[k]| thereof;

computing a signal-to-noise ratio SNR[k] of said spectrum S*[k] according to:

SNR[k]=|S*[k]|/Bq;

calculating from said signal-to-noise ratio SNR[k] a filter magnitude F[k] according to a predefined filter algorithm based on at least one filter update parameter;

calculating an output spectrum S b [k] at least substantially free of said quantization noise by multiplying both the real part R{S*[k]} and the imaginary part I{S*[k]} of the spectrum S*[k] with said filter magnitude F{k}; and

transforming the output spectrum S b [k] back into a signal s b [n] in the time domain.

5. The method according to claim 4 , characterized in that the filter update parameter and thus the filter magnitude F[k] are adjusted such that the quantization noise in the remaining output spectrum S b [k] is as low as possible.

6. The method according to claim 4 , characterized in that it further comprises the steps of:

weighting the frames of the input PCM signal with a first window w 1 [n] and calculating the spectrum S*[k] from said weighted signal;

generating a weighted output signal s b w [n] by weighting the signal s b [n] received after the re-transformation with a second window W 2 [n]; and

calculating a final output signal S w b [n] for a current frame of the PCM-signal from said weighted output signal s b w [n] such that the transition between two successive output frames and is smoothed.

7. The method according to claim 4 , characterized in that the computation of the spectrum S*[k] of the PCM signal is done by using a Fast Fourier Transformation FFT; and

the re-transforming the output spectrum S b [k] back into a time domain signal s b [n] is done by using an inverse FFT.

8. The method according to claim 6 , characterized in that the first and the second window w 1 and w 2 are identical.

9. A quantization noise substracting unit for substracting quantization noise from a pulse code modulated PCM signal being segmented into frames, comprising:

a quantization noise level calculating unit ( 100 ) for calculating for each frame of said PCM signal a constant quantization noise level Bq according to the following equation:

B

q

=

∑

n

=

0

W

-

1

⁢

{

(

s

min

*

⁡

[

n

]

-

s

(

3

)

⁢

max

*

⁡

[

n

]

)

·

w

⁡

[

n

]

}

2

12

wherein

n: indicates a specific sample of the PCM signal;

S* min [n]: represents the minimum quantization noise level for a specific sample value s*[n] of said PCM signal;

S* max [n]: represents the maximum quantization noise level for the specific sample value s*[n] of the PCM signal, wherein S* min [n] −S* max [n] has a different value for at least two specific samples n respectively;

w[n]: represents a window-function; and

W: represents the number of samples per window; and

a background noise substracting unit for substracting the quantization noise as represented by said quantization noise level Bq from said PCM signal.

10. The noise substracting unit according to claim 9 , characterized in that it is located at a decoder's side.

11. The noise subtracting unit according to claim 9 , characterized in that the minimum quantization level S* min and the maximum quantization level S* max are predicted according to the following equations:

S* min =i[n ]−( i[n]−i min [n ])/2

S* max =i[n ]+( i max [n]−i[n ])−/2

wherein

i: represents one out of a plurality of possible representation levels predefined due to the specific PCM quantization method applied to an original signal;

i[n]: represents that predefined representation level which corresponds to the sample value s*[n]for a specific n;

i min [n]: represents that representation level which is—started from i[n]—the next smaller non-zero representation level for which u[n]=1;

i max [n]: represents that representation level which is—started from i[n]—the next bigger non-zero representation level for which u[n]=1;

with the usage array u[i]being defined to:

u

⁡

(

i

)

=

min

(

1

,

∑

n

=

0

L

-

1

⁢

⁢

{

0

,

s

*

⁡

[

n

]

≠

i

1

,

otherwise

)

,

-

2

N

-

1

≤

i

<

2

N

-

1

wherein

L: represents the number of samples of the whole PCM-signal; and

N: represents the number of bits used for quantizing an original sample value by using PCM to generate the PCM sample values s*[n].

12. The noise subtracting unit according to claim 9 , characterized in that the background subtracting unit subtracts the quantization noise represented by said quantization noise level Bq from the PCM-signal in the frequency domain according to the following steps:

computing the spectrum S*[k] of the PCM signal s*[n] and forming the magnitude |S*[k]| thereof

computing a signal-to-noise ratio SNR[k] of said spectrum S*[k] according to:

SNR[k]=|S*[k]|/Bq;

calculating from said signal-to-noise ratio SNR[k] a filter magnitude F[k] according to a predefined filter algorithm based on at least one filter update parameter;

calculating an output spectrum S b [k] at least substantially free of said quantization noise by multiplying both the real part R{S*[k]} and the imaginary part I{S*[k]} of the spectrum S*[k] with said filter magnitude F{k}; and

transforming the output spectrum S b [k] back into a signal s b [n] in the time domain.

13. The noise subtracting unit according to claim 12 , characterized in that the filter update parameter and thus the filter magnitude F[k] are adjusted such that the quantization noise in the remaining output spectrum S b [k] is as low as possible.

14. The noise subtracting unit according to claim 12 , characterized in that the background subtracting unit subtracting the quantization noise represented by said quantization noise level Bq from the PCM-signal in the frequency domain further comprises the steps of:

weighting the frames of the input PCM signal with a first window w 1 [n] and calculating the spectrum S*[k] from said weighted signal;

generating a weighted output signal s b w [n] by weighting the signal s b [n] received after the re-transformation with a second window W 2 [n]; and

calculating a final output signal S w b [n] for a current frame of the PCM-signal from said weighted output signal s b w [n] such that the transition between two successive output frames is smoothed.

15. The noise subtracting unit according to claim 12 , characterized in that the computation of the spectrum S*[k] of the PCM signal is done by using a Fast Fourier Transformation FFT; and

the transforming of the output spectrum S b [k] back into the time domain signal s b [n] is done by using an inverse FFT.

16. The noise subtracting unit according to claim 14 , characterized in that the first and the second window w 1 and w 2 are identical.