JPH0423586A - Pseudo picture coding signal generator - Google Patents

Abstract

PURPOSE:To obtain a signal well approximated to an audocorrelation function with high accuracy over a long time even when the degree of the filter is not increased by generating a coefficient deciding the transfer function of the filter variable with a sampling time of the filter. CONSTITUTION:The circuit consists of a change component signal generator 1, an average value adder circuit 2, a maximum value control circuit 3 and a minimum value control circuit 4. In order to simulate a change in a variable rate picture coding signal, a synthesis filter 12 receiving white gauss noise is employed and a smoothing filter 132 receiving other white gauss noise is provided and an output of the smoothing filter 132 is used as an output of the synthesis filter 12 to vary the coefficient of the filter 12. Thus, a signal well approximated to an audocorrelation function of a variable rate picture coding signal with high accuracy is obtained over a long time even when the degree of the filter is not increased.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a pseudo image encoded signal generator for evaluating multiplexing characteristics in an asynchronous transfer mode.

(Prior art) Asynchronous transfer mode
nsferMode (ATM) is a method of transmitting information in short fixed-length packets called cells.

In this method, various media with different properties can be efficiently transmitted by using statistical multiplexing effects. Therefore, it is attracting attention as a promising transmission method for wideband ISDN. Among various media, moving image signals in particular can be transmitted at television broadcasting quality even with intensely moving images by performing variable rate encoding. Therefore, it is expected to be a new medium for broadband I5DN. However, in addition to the above-mentioned advantages, ATM has quality deterioration factors such as cell discard and fluctuations and increases in delay time, so evaluation of statistical multiplexing of various media is an important issue for realizing broadband I5DN. ing.

FIG. 3 is an explanatory diagram of an outline of ATM multiplexing.

The video signal from the television camera is sent to an image encoder 3.
Coded by 0. The output signal of the image encoder 30 is packetized by cell assemblers 31 and 32 and input to the multiplexer 33. In the multiplexer 33, cells arriving independently from N human-powered channels are stored in an output buffer 34 and sequentially sent out to the output channels. At this time, the cells that did not fit into the output buffer 34 are
Will be discarded. The multiplexing characteristics are evaluated by observing the state of the cells in the output buffer 34 of the multiplexer 33 and calculating the cell waiting time and discard rate. In order to obtain output signals from N independent image encoders 30, it is disadvantageous in terms of cost and technique to prepare N actual image encoders.

For this reason, a single pseudo-image encoding system that generates a pseudo-signal having the same statistical properties (average value, variance value, autocorrelation function, etc.) as the output signal output from each of the N image encoders is used. using a signal generator (e.g. J, Zedps
ki et al, “Packet Transpo
rtof VBRInterframe DCT Co
mpressed DigitalVideo on
a C3MA/CD LAN, , IEEE GL
(See OBECOM'89pp, 886-892).

FIG. 2 is a block diagram of a conventional pseudo image encoded signal generator.

The illustrated generator includes a change signal generator 21, an average value addition circuit 22, a maximum value limit circuit 23, and a maximum value limit circuit 2.
It consists of 4.

The variation signal generator 21 includes a normal random number generator 51 and a synthesis filter 52.

First, the variation signal generator 21 generates white Gaussian noise e with an average of 0 and appropriate variances a and 2 using a normal random number generator 51, and inputs it to the synthesis filter 52. The output signal Xk of the synthesis filter 52 has a certain appropriate correlation and is a signal representing a changing component of the variable rate image encoded signal. Therefore, an average value of 0 is added to this xk by an average value adding circuit 22. Further, the maximum value limiting circuit 23 and the minimum value limiting circuit 24 limit the value so that it does not actually occur, and generate the pseudo image encoded signal 1. The signal Ik thus obtained is input to the cell assembler 31, 32 of FIG.

FIG. 4 is a block diagram of a conventional synthesis filter.

As a synthesis filter, a filter having an all-pole transfer function expressed by the following equation (1) is generally used.

H(Z) = 1/ (1+Σ+=+ Mal z-1
) ”' (1) However, al is the coefficient of the filter,
M is the order of the filter.

The input signal ek and output signal Xk of the filter are expressed by the following relationship.

To χ=−Σ+=+ Mat X*−+ +e++ −
(2) When a white signal is input, a signal with a certain correlation is obtained. Therefore, the filter coefficient 81 is determined so that the autocorrelation function of the output signal x1 has the same form as the autocorrelation function of the variable rate encoded signal of the actual moving image. in this case,
Since the transfer function is assumed to be the all-pole function of equation (1), the autocorrelation function of the actually measured variable rate image encoded signal data is calculated using the widely known coefficient estimation method of the autoregressive model. It can be found from

Furthermore, as shown in FIG. 2, the first order M of the filter is the simplest and is therefore widely used.

In this case, the autocorrelation function of the filter output xk is a function that decreases exponentially.

(Problems to be Solved by the Invention) However, the above-described conventional technology has the following problems.

That is, since - boat fishing uses a first-order all-pole function synthesis filter, only an autocorrelation function that decreases exponentially can be obtained. Therefore, there is a drawback that the autocorrelation function of an actual variable rate image encoded signal cannot be accurately approximated over a wide range. Therefore, there is a problem in that the multiplexing characteristics of image signals in an ATM communication network cannot be accurately simulated.

Furthermore, when attempting to accurately approximate the autocorrelation function of an actual variable rate image encoded signal over a wide range, the fourth
As shown in the figure, there is a problem in that the order of the synthesis filter needs to be extremely large, making it difficult to implement.

In this case, the approximation accuracy does not improve as the order increases; for example, even with 5th order, the autocorrelation up to about 5 samples is well approximated, but the autocorrelation beyond that deviates greatly from the actual one. As a result, the error becomes larger overall.

The present invention has been made with attention to the above points, and the first-order synthesis filter can only obtain an autocorrelation function that decreases exponentially, which accurately approximates the autocorrelation function of a variable rate image encoded signal. However, if you try to accurately approximate the autocorrelation function of a variable rate image encoded signal over a wide range, the order of the synthesis filter will become very large, making it difficult to achieve. It is an object of the present invention to provide a pseudo image encoded signal generator that can remove the autocorrelation function of a variable rate image encoded signal and accurately approximate the autocorrelation function of a variable rate image encoded signal with a small number of orders.

(Means for Solving the Problems) The pseudo image encoding signal generator of the present invention includes a normal random number generator that generates random numbers according to a normal distribution, and a filter that receives the random numbers generated by the normal random number generator as input. , and a coefficient generator that generates coefficients that change every sampling time of the filter, and the output of the filter is used as a changing component of a pseudo image encoded signal.

(For A) In the pseudo image coding signal generator of the present invention, the coefficients that determine the transfer function of the filter are generated by the coefficient generator while changing at each sampling time of the filter. Therefore, the autocorrelation function can be accurately approximated over a wide range without increasing the order of the filter.

(Embodiment) FIG. 1 is a block diagram showing an embodiment of a pseudo image encoding signal generator of the present invention.

The illustrated generator includes a variation signal generator 1, an average value adding circuit 2, a maximum value limiting circuit 3, and a minimum value limiting circuit 4.

The variation signal generator 1 includes a normal random number generator 11, a synthesis filter 12, and a coefficient generator 13.

The normal random number generator 11 generates random numbers that follow a normal distribution.

The synthesis filter 12 combines the input signal ek generated by the normal random number generator 11 and the coefficient a1(k
) and outputs an output signal Xk. This synthesis filter 1
The details of the configurations of the coefficient generator 2 and the coefficient generator 13 will be described in FIGS. 5 and 6, respectively, which will be described later.

The average value addition circuit 2 adds the average value ■o to the output signal x2 to calculate a pseudo image encoded signal ■□.

The maximum value limiting circuit 3 stores the maximum value ■□a of the pseudo image encoded signal ■1, and compares this maximum value I max with the pseudo image encoded signal Ik to determine the pseudo image encoded signal When Ik is small, a pseudo image encoded signal signal is output. On the other hand, when the pseudo image encoded signal Ik is large, the maximum value ■□8X is output.

The minimum value limiting circuit 4 receives the pseudo image encoded signal ■.

The minimum value of ■,. This minimum value ■, is stored. When the pseudo image encoded signal 1 is larger than the pseudo image encoded signal 1, the pseudo image encoded signal Ik is output. On the other hand, when the pseudo image encoded signal 5 is small, the minimum value Imin is output.

FIG. 5 is a block diagram of an embodiment of a synthesis filter according to the present invention.

The illustrated synthesis filter includes a plurality of coefficient generators 61, 62, etc.

Coefficient generators 61 and 62 generate coefficients a+(k) and a, respectively.
2(k). The transfer function of the synthesis filter shown in this figure and the relationship between the human input signal e and the output signal xk are expressed by the following equations (3) and (4), respectively.

H(Z) = 1/ (1+Σr=+ 1at(k)z
-') +・(3)Xk=-Σl=+ Mat(k)
Xb-+ +ek・++ (4) These equations express the filter coefficient a in the above equation (1) or (2) as a function of time, and the synthesis filter 12
The coefficient a+ that changes at the same time interval as the sampling time of
(k) is given by each coefficient generator. Note that there is only one coefficient generator.

FIG. 6 is a block diagram of a coefficient generator according to the present invention.

The illustrated coefficient generator includes a normal random number generator 131, a smoothing filter 132, an average value adding circuit 133, a maximum value limiting circuit 134, and a minimum value limiting circuit 135.

The normal random number generator 131 generates random numbers that follow a normal distribution.

The smoothing filter 132 inputs the random number n+(k) generated by the normal random number generation circuit 131 and calculates the change component Δa + (k
) is output.

The average value addition circuit 133 calculates the coefficient at(k) by adding the coefficient average value alo to the change component Δa+(k).

The maximum value limiting circuit 134 sets the maximum value al of the coefficient a+(k)
□8 is stored, and this maximum value a 1may and coefficient a
+(k), and if the coefficient a r (k) is small, the coefficient a+(k) is output. On the other hand, the coefficient at(k)
When is large, the maximum value a1□8 is output.

The minimum value limiting circuit 135 sets the minimum value a1 of the coefficient a+(k)
□. is stored, and this minimum value a 1mIn and coefficient a
+(k), and if the coefficient at(k) is larger, the coefficient a+(k) is output. On the other hand, when the coefficient at(k) is small, the minimum value a lmIn is output.

FIG. 7 is a block diagram of an embodiment of a smoothing filter according to the present invention.

The illustrated smoothing filter has an L-order all-pole function with an IIR
It's a filter. Here, the coefficient α, α12,...
αIn is a predetermined constant. The transfer function in this case and the relationship between the input signal n+(k) and the output signal Δa+(k) are expressed by the following equations (5) and (6), respectively.

Hs I(Z) = 1/ (1+Σ, , LαIJZ
-') ... (5) Δat(k)=-ΣJ=l
' aiJΔat (k-j) "n+ (k) -
(6) Variance On1′ of white Gaussian noise, coefficient a 1(k
) average value alO1 maximum value limit a l, n, , minimum value limit a 1 mIn and smoothing filter coefficient α1
．． is obtained as follows.

First, a coefficient a r (k) is estimated based on equation (4) from actually measured variable rate image encoded signal data using a parameter estimation method of a time-varying system such as a Kalman filter. Then, the average value, maximum value, and minimum value of the estimated coefficient a+(k) are respectively a ios a 1m
1 lX, a 1 mIn.

Next, the smoothing filter coefficient 01. For, Equation (6)
has the same form as the L-order autoregressive model, so Le
It can be determined from the autocorrelation function of the estimated coefficient a+(k) using the Vinson-Durbin method (for example, Nakamizo: "Signal Analysis and System Identification", Corona Publishing, 1988). At this time, the variance of the residual signal obtained at the same time is the variance σ of white Gaussian noise. 12, will be given.

In fact, the results of analyzing coded signal data of a certain moving image show that using the first-order synthesis filter 12 and the second-order smoothing filter 132 has approximately the same approximation accuracy as a conventional 30th-order synthesis filter. was gotten.

Note that the specific method of configuring the synthesis filter and smoothing filter in the present invention is not limited to those shown in the embodiments, but may also include a method of configuring a filter that realizes the transfer functions of equations (3) and (4) above. There are various other things. Any of these construction methods can be used to implement the present invention.

(Effects of the Invention) As described above, according to the present invention, in a pseudo image encoded signal generator using a synthesis filter that receives white Gaussian noise as input in order to simulate changes in a variable rate image encoded signal, , we provided a smoothing filter that manually uses white Gaussian noise, and changed the coefficients of the synthesis filter by using the output of the smoothing filter as the output of the synthesis filter. The autocorrelation function of a variable rate image encoded signal can be accurately approximated over a long period of time with a small number of filter orders.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the pseudo image encoded signal generator of the present invention, FIG. 2 is a block diagram of a conventional pseudo image encoded signal generator, and FIG. 3 is an overview of multiplexing by ATM. An explanatory diagram, FIG. 4 is a block diagram of a conventional synthesis filter,
FIG. 5 is a block diagram of an embodiment of a synthesis filter according to the present invention, FIG. 6 is a block diagram of an embodiment of a coefficient generation circuit according to the present invention, and FIG. 7 is a block diagram of an embodiment of a smoothing filter according to the present invention. It is a diagram. 1... Change signal generator, 2... Average value addition circuit,
11... Normal random number generator, 12... Synthesis filter,
13... Coefficient generator, 131... Normal random number generator,
132... Smoothing filter, 133... Average value addition circuit, 134... Maximum value limiting circuit, 135... Minimum value limiting circuit.

Claims (1)

[Claims] 1. A normal random number generator that generates random numbers according to a normal distribution, a filter that receives the random numbers generated by the normal random number generator as input, and generates coefficients that change at each sampling time of the filter. A pseudo-image encoded signal generator, comprising: a coefficient generator for generating a pseudo-image encoded signal; 2. The coefficient generator includes: a normal random number generator that generates random numbers according to a normal distribution; a filter that inputs the random numbers generated by the normal random number generator; and a predetermined signal that is added to the output signal of the filter. It is characterized by comprising an adding circuit, a maximum value limiting circuit that limits the output signal of the adding circuit to a predetermined maximum value, and a minimum value limiting circuit that limits the output signal of the adding circuit to a predetermined minimum value. A pseudo image encoded signal generator according to claim 1.