JPH0771009B2 - Digital filter circuit and band division transmission system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital filter circuit used for voice transmission or image transmission and a band division transmission system using the digital filter circuit.

[0002]

2. Description of the Related Art In recent years, as a high-efficiency coding method for image signals or audio signals, band-division coding, in which an input signal is separated and sub-sampled for each frequency component by a plurality of digital filters, and then coded and transmitted, is noted. Has been done.

An example of a voice transmission system using this band division coding is shown in FIG. In the configuration of FIG. 1, a digital filter (h ₁ (n)) having a low-pass characteristic is provided on the transmission side.
2. Digital filter with high-pass characteristics (h ₂ (n))
The input signal x (n) is band-divided by 3, and each band signal is sub-sampled by the sub-sampling circuits 10 and 11 at half the Nyquist speed with respect to the input, and its output is quantized and encoded by an encoding circuit (not shown). Send to the receiver. On the receiving side, first, the transmission information is decoded by a decoding circuit (not shown), and 0 values are interpolated by the interpolation circuits 16 and 17, and each output is a low-pass digital filter (g ₁ (n)) 6 and a high-pass characteristic. After passing through the digital filter (g ₂ (n)) 7 of the above, the synthesized signal is added by the adder circuit 30 to obtain the decoded signal x (n) (in the drawing, the decoded signal x is a decoded signal above the symbol). Symbol, but omitted in the description).

This band-division coding method of a voice signal has an advantage that the total voice quality can be improved by allocating a larger number of quantization bits to a band where voice energy is concentrated. Further, there is an advantage that the quantization noise can be prevented from affecting other bands.

Further, when this band division coding method is applied to an image signal, in addition to the advantage of the above-mentioned voice, there is an advantage that the problem of block distortion is solved.

It should be noted that this band division coding system is not only used for two band division as shown in FIG.
As shown in FIG. 2, multi-stage connection may be used for arbitrary band division. Especially for a signal source such as an image signal having a high correlation between pixels, only low-frequency information is often band-divided by multistage connection.

In general, it is impossible to realize an ideal frequency characteristic with a constant filter with a finite number of taps, and three kinds of distortion, that is, aliasing distortion, amplitude distortion, and phase distortion occur in a decoded signal. But in the z region between the filters,

[Equation 2] If the relationship of is satisfied, the decoded signal x (n) is converted to the original signal x
It is possible to completely match (n).

Therefore, on the basis of the equation (1), a filter capable of completely reconstructing a decoded signal is being studied.

As an example of the structure of the perfect reconstruction filter, first, Q proposed by Crochiere is proposed.
MF (Quadrature Mirror Filter, hereinafter referred to as QMF) may be used.

Literature Crotier, "Subband Coding", Bell Systems Technical Journal No. 60 (September 1981), pp. 1633-1653 (RECrochier
e, "Subband Coding," BSTJ, 60, pp.1633-1653 (Sep.198
1))

In this method, the filter {h ₁ (n)} is a coefficient-symmetric linear phase FIR filter, and the other filters {h ₂ (n), g ₁ (n), g ₂ (n)} are respectively used. _{h 2 (n) = (-} 1) n h 1 (n) g 1 (n) = 2 · h 1 (n) g 2 (n) = - 2 · (-1) n h 1 (n) ... ( By defining in (2), the aliasing distortion and phase distortion in the restored signal are completely removed. However, the removal of the amplitude distortion is limited to the case where the number of taps of the filter is infinity, which is 2, and in an environment of less than that, there is no choice but to rely on approximation. Therefore 16 to apply to the image
It was inevitable that more taps than blocks were required, which was difficult to achieve.

[0012] In addition, Smith et al.
ate Quadrature Filter Hereinafter referred to as CQF. ) Is proposed.

References Smith, Barnwell, “Exact Reconstruction Techniques for Tree Structured Subband Coders”, IE
EE Transactions of Acoustic Speech and Signal Processing Volume 3 ASSP-34 Volume 3
Issue (June 1986), pp.434-441 (M. Smith and TP
Barnwell III, "Exact Reconstruction Techniques for
Tree-Structured Subband Corders ", IEEE Trans.on ASSP
P pp.434-441 (June.1986))

In this method, a method called spectrum decomposition is used to solve the problem of the amplitude distortion, which is a drawback of the above QMF, and realizes perfect reconstruction. However, the problem of application to images remains where the filter characteristics are not linear.

On the other hand, Le Gall et al. Consider SSKF (Symmetric) from the viewpoint that a long tap such as QMF or a non-linear phase filter such as CQF is not desirable when considering application to an image. Short Kernel F
ilter) is proposed.

Reference Lugol, Tabata Bai, "Subband Coding of Digital Images, Using Symmetric Short Kernel Filters and Alice Metictic Coding Technics", IEEE Icasb 88 (1988) 6
Mon), pp. 761-764 (DLGall and A.Tabataba
i, "Subband Cording of Digital Images Using Symmetr
ic Short Kernel Filters and Arithmetic Cording Tech
niques, "Proc. of IEEE ICASSP'88, pp.761-764 (June.19
88))

In this method, the constraint between filters is QMF.
G ₁ (z) = H ₂ (−z) G ₂ (z) = H ₁ (−z) (3) From F (z) −F (−z) = 2 · z ⁻ A polynomial F (z) of z satisfying ^k ... (4) is defined as H ₁ (z) and H ₂ (−
A perfect reconstruction filter is realized by using z) as a factorization problem. At this time, F (z) is a polynomial in which the odd-numbered terms have only one non-zero term, but Lugol et al.
A concrete solution is sought by imposing a coefficient symmetry condition centered on an odd-order term of and performing factorization based on the constraint condition.

Here, as a comparison of the data compression characteristics of the above three filter configurations, after band division into two bands,
Alternatively, when characteristics are compared when quantization and coding are performed after band division into a plurality of bands by multistage connection, SSKF has the best characteristic as long as the number of band divisions is small.

[0019]

When using QMF in band division coding, as described above, in order to realize sufficient removal of amplitude distortion, the length of the filter (specified by the number of taps) is used. Is a problem. Also, CQ
When F is used, perfect reconstruction is realized, but the filter characteristic is nonlinear phase, and for applications such as hierarchical coding in which low resolution images are displayed in stages, phase distortion The problem of occurrence of. Furthermore, the data compression characteristics of these methods are inferior to SSKF as long as the number of band divisions is small.

However, when SSKF is used, the problem of the length of the filter and the problem of the phase distortion have been solved, but the configurations proposed as specific examples are limited to four types, and the configurations thereof are limited. Is not generalized. That is, there are an infinite number of filters that realize perfect reconstruction,
Of these, the best configuration from the viewpoint of data compression has not been proposed, and no configuration of a perfect reconstruction filter with high data compression efficiency has been found.

The present invention provides a digital filter circuit having a high data compression efficiency under such an environment that the filter length is a constraint condition among the infinite perfect reconstruction filters.

[0022]

According to the present invention, a plurality of first digital filters (h ₁ (n) to h ₁ (n) to h ₁ (n) to h) having different pass bands and passing an input digital signal in the respective bands are provided.
_M (n)), a plurality of sub-sampling circuits for sub-sampling the output of the first digital filter, a plurality of interpolating circuits for interpolating the sub-sampled signals with 0-value, and the 0-value interpolation A plurality of second digital filters (g) having a signal as an input and having a pass band corresponding to the first digital filter
₁ (n) to g _M (n)) and an adder circuit that takes the sum of the outputs of the second digital filter, in which the following coding gain G _SBC is set using the filter length as a parameter.

[Equation 1] Here, K is the number of bands to be band-divided, A _k is a constant determined from the filter coefficient h _k (n) and the correlation coefficient of the input signal, and B _k is the filter coefficient g _k (n) and the quantization error. The constant α _k determined from the correlation coefficient is characterized in that the filter coefficients of the first and second digital filters are determined so as to maximize the constant determined from the sub-sampling rate of the k-th band.

Further, the band division transmission system of the present invention has a plurality of first digital filters (h) having different pass bands and passing an input digital signal in each band.
₁ (n) to h _M (n)), a plurality of subsampling circuits for subsampling the output of the first digital filter, and transmission for transmitting the subsampled signals to the respective subband transmission lines. Means and
Receiving means for decoding the signal received on the subband transmission line, a plurality of interpolation circuits for interpolating each of the decoded signals with a zero value, and the zero-value interpolated signal as an input, and the first digital filter described above. A plurality of second digital filters (g ₁ (n) to g having a pass band corresponding to
_M (n)) and an adder circuit that takes the sum of the outputs of the second digital filters, in the band division transmission system, the filter length is used as a parameter and the next coding gain G _SBC
To

[Equation 1] Here, K is the number of bands to be band-divided, A _k is a constant determined from the filter coefficient h _k (n) and the correlation coefficient of the input signal, and B _k is the filter coefficient g _k (n) and the quantization error. The constant α α _k determined from the correlation coefficient and α _k are characterized in that the filter coefficients of the first and second digital filters are determined so as to be a constant maximum determined from the sub-sampling rate of the k-th band.

[0024]

The present invention maximizes the coding gain G _SBC , which is an evaluation function, from the viewpoint of data compression characteristics when determining filter coefficients for band-division coding, which have been designed with emphasis on filter characteristics. It is decided to do.

An example of a one-dimensional two-band linear phase band division system (when k = 1) as shown in FIG. 1 will be described. α ₁ = α ₂ = 0.5 and g ₁ (n) = (− 1) ⁿ h ₂ (n) g ₂ (n) = − (− 1) ⁿ h ₁ (n).

The length of the filter h ₁ (n) is 1,
When {h ₁ (n)} = {a}, the length of the filter h ₂ (n) is 3, and {h ₂ (n)} = {− q, 1, −q}, the perfect reconstruction condition is a = 1. When ρ is the correlation coefficient of the input signal, A ₁ , A ₂ , B ₁ , and B ₂ are respectively A ₁ = 1 A ₂ = (1 + 2q ² ) -4q · ρ + 2q ² · ρ ² B ₁ = 1/2 + q ² Given by B ₂ = 1/2.

Next, the length of the filter h ₁ (n) is 5,
{H ₁ (n)} = {c, b, a, b, c}, filter h
The length of ₂ (n) is 3, {h ₂ (n)} = {-q, 1,-
In the case of q}, the perfect reconstruction condition is given by a + 2 · q · b = 1 q · b + 1 = 0, and A ₁ , A ₂ , B ₁ and B ₂ are respectively A ₁ = (a ² + 2b ² + 2c ² ). +4 (ab + bc) ρ +2 (b ² + 2ac) ρ ² + 4bc · ρ ³ + 2c ² · ρ ⁴ A ₂ = (1 + 2q ² ) -4q · ρ + 2q ² · ρ ² B ₁ = 1/2 + q ² B ₂ = (1 / 2) given by a ² + b ² + c ² .

If a filter coefficient that maximizes the coding gain G _SBC , which is an evaluation function, is determined based on these conditions, a filter for optimum band division coding for data compression is obtained using the filter length as a parameter. The coefficient can be determined.

[0029]

Embodiments of the present invention will be described below with reference to the drawings.

The digital filter circuit of this embodiment is
In the circuit shown in FIG. 1, the digital filter (h
₁ (n)), (h ₂ (n)), (g ₁ (n)), (g ₂
In (n)), G _SBC as the evaluation function is

[Equation 1] The feature is that it is decided to maximize.
Here, K is the number of bands divided into bands, A _k is a constant determined from the filter coefficient h _k (n) and the correlation coefficient of the input signal, and B _k is the phase of the filter coefficient g _k (n) and the quantization error. Α _k is a constant determined from the subsampling rate of the k-th band.

FIG. 4 shows an example of a band division image transmission system using this digital filter circuit. In this embodiment, image information is divided into two bands and transmitted.

The input digital signal x (n) is divided into two bands by the digital filters 2 and 3 according to the present invention. The signals that have passed through the two digital filters 2 and 3 are subsampled by 1/2 of the Nyquist frequency by the subsampling circuits 10 and 11, respectively, and coded by the coding circuits 21 and 22. The band-divided and coded two-band signals are multiplexed by the multiplexer 24 and time-division multiplexed transmitted to the receiving side by the multiplex transmission line 25.

On the receiving side, the demultiplexer 26 demultiplexes and demultiplexes two band signals. After both signals are decoded by the decoding circuits 27 and 28, the interpolation circuits 16 and 17 further perform zero value interpolation by the Nyquist frequency. The signals which are zero-value interpolated by the interpolation circuits 16 and 17 are input to the digital filters 6 and 7 on the receiving side of the present invention, respectively. The signals that have passed through the digital filters 6 and 7 are combined by the adder circuit 30, and the digital signal x
The restoration of (n) is performed.

Here, in this band division image transmission system, the feature of this embodiment is that the filter coefficients of the digital filters 2, 3, 6 and 7 are different.

[Equation 1] Is determined to be the maximum. Here, K is the number of bands to be band-divided, A _k is a constant determined from the filter coefficient h _k (n) and the correlation coefficient of the input signal, and B _k is the filter coefficient g _k (n) and the quantization error. Α _k is a constant determined from the correlation coefficient and α _k is a constant determined from the sub-sampling rate of the k-th band.

Next, an example in which an algorithm for determining the filter coefficients of the digital filters 2, 3, 6 to 7 is actually applied will be described.

FIG. 5 shows a digital filter (h) which is a low-pass filter on the transmission side when q is a parameter.
₁ (n) 6 has 5 taps, and the digital filter (h ₂ (n)) that is a high-pass filter has 3 taps. That is, the gain evaluation formula (2) is maximized. Has been decided. As a comparative example, the case of SSKF with 5 × 3 taps is also shown. As shown in FIG. 5, by optimizing the parameters, almost SS
Coding gain equal to or higher than KF can be obtained.

Further, in FIG. 6, the number of taps of the high-pass filter is fixed at 3, and the number of taps of the low-pass filter is 1, 5,
9 is a plot of theoretical optimum values in the case of 9, and the case of SSKF is also shown as in FIG. In this case Rukoto the Ki de be obtained similarly SSKF equal to or more coding gain and FIG 5 is also shown.

On the other hand, the number of taps of the low-pass filter is 5,
When the number of taps of the high-pass filter is 3, the simulation result when band-division-encoding the monochrome image GIRL when using the filter coefficient obtained from the above algorithm is compared with the simulation result using SSKF.

First, when multi-stage connection is not performed, that is, in the case of four bands because it is two-dimensional, in the method of the present invention, high efficiency encoding of 50 dB can be realized with 4.503 bits per pixel. On the other hand, SSKF required 4.507 bits per pixel for 50 dB encoding.

Further, in the case of multi-stage connection, in the case of 3 layers based on subdivision of low frequency components only (in case of 2 dimensions and 10 bands), in the method of the present invention, 50 dB at 3.515 bits per pixel. It was possible to achieve higher efficiency. On the other hand, in SSKF, per pixel is required for 50 dB encoding.
It took 3.517 bits.

As described above, according to the method of the present invention, it is possible to obtain better characteristics than the conventional method and to obtain a digital filter circuit having a high coding gain. Further, based on the present invention, it becomes possible to determine the optimum filter coefficient according to the size of the filter.

[0042]

As described above, according to the present invention, it is possible to obtain a digital filter circuit having good characteristics that the data compression efficiency is higher than that of the conventional band division coding transmission system.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a 2-band band division transmission system.

FIG. 2 is a diagram showing a configuration of a multi-band band division transmission system based on multi-stage connection.

FIG. 3 is a diagram showing a configuration of a parallel multiple band band division transmission system.

FIG. 4 is a diagram showing a configuration of a 2-band band image transmission system according to an embodiment of the present invention.

FIG. 5 is a diagram showing the coding gain of the perfect reconstruction filter when the size of the transmission-side low-pass filter is 5, the size of the high-pass filter is 3, and the linear phase is satisfied.

FIG. 6 The size of the high-pass filter on the transmission side is fixed at 3, and the sizes of the low-pass filters are 1, 5, and 9
FIG. 6 is a diagram showing the optimum value of the coding gain of the perfect reconstruction filter in the case of satisfying the linear phase.

[Explanation of symbols]

 2, 3, 4, 6, 7, 8 Digital filter 10, 11, 12 Sub-sampling circuit 16, 17, 18 Interpolation circuit 21, 22 Encoding circuit 24 Multiplexer 25 Multiplex transmission line 26 Demultiplexer 27, 28 Decoding circuit 30 Adder circuit

Continuation of the front page (56) Reference JP-A 64-42938 (JP, A) JP-A 63-201700 (JP, A) JP-A 2-98232 (JP, A) JP-B 2-29239 (JP , B2) Japanese Patent Publication 6-36157 (JP, B2) "Performance Evaluation of Subbing Coding and Optimizer on of Its Filter Co efficient", SPIE Vol. 1605, PP. 95-106 Nov. 1991

Claims (2)

[Claims] 1. A plurality of first digital filters (h ₁ (n) to h _M (n)) having different pass bands and passing an input digital signal in each band, and an output of the first digital filter. A plurality of sub-sampling circuits for respectively sub-sampling, and a plurality of interpolating circuits for zero-value interpolating the sub-sampled signals,
This zero-value interpolated signal is input, and a plurality of second digital filters (g ₁ (n) to g _M (n)) having a pass band corresponding to the first digital filter and this second digital filter In a digital filter circuit provided with an adder circuit that sums the outputs of digital filters, the following coding gain G _SBC is expressed as Here, K is the number of bands to be band-divided, A _k is a constant determined from the filter coefficient h _k (n) and the correlation coefficient of the input signal, and B _k is the filter coefficient g _k (n) and the quantization error. A constant determined from the correlation coefficient and α _k is a constant determined from the sub-sampling rate of the k-th band, and the filter coefficients of the first and second digital filters are determined so as to be maximum. Digital filter circuit. 2. A plurality of first digital filters (h ₁ (n) to h _M (n)) having different pass bands and passing input digital signals in the respective bands, and outputs of the first digital filters. A plurality of sub-sampling circuits for respectively sub-sampling, a transmitting means for transmitting the sub-sampled signals to the respective sub-band transmission paths, a receiving means for decoding the signals received on the sub-band transmission paths, and a decoding means. A plurality of interpolating circuits for interpolating each of the generated signals into 0-values, and a plurality of second digital filters (g having a pass band corresponding to the above-mentioned first digital filter, to which the 0-value-interpolated signals are input.
₁ (n) to g _M (n)) and an adder circuit that sums the outputs of the second digital filters, in the band division transmission system, the following coding gain G _SBC is set with the filter length as a parameter. [Equation 1] Here, K is the number of bands to be band-divided, A _k is a constant determined from the filter coefficient h _k (n) and the correlation coefficient of the input signal, and B _k is the filter coefficient g _k (n) and the quantization error. A constant determined from the correlation coefficient and α _k is a constant determined from the sub-sampling rate of the k-th band, and the filter coefficients of the first and second digital filters are determined so as to be maximized. Band division transmission method.