JPS6085630A - Signal processing device - Google Patents

Abstract

PURPOSE:To prevent a rounding error in applying D/A conversion to an input signal by using data of converting output from a memory written with a data conversion table as data having a higher bit number than the bit number of an input digital signal in applying nonlinear correction to the input digital signal obtained through A/D conversion. CONSTITUTION:A digital color video signal read out of a frame memory 4 is fed to a digital AM modulation circuit 7. On the other hand, a carrier signal of, e.g., 1,700Hz is fed to the modulation circuit 7 through a terminal 8. Digital modulation is applied at the modulation circuit 7, from which a DSB signal is obtained. Inverse gamma correction is applied at the same time to the color video signal at the digital AM modulation circuit 7 and also the signal is extracted into a data having a higher bit, e.g., >=10 bits per sample than 8 bits per sample from a frame memory as the output digital signal, thereby attaining countermeasure against rounding errors.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application This invention is a digital signal that converts a human information signal into a digital signal, performs various necessary digital processing, and then converts it into a digital signal.
The present invention relates to a signal processing device that converts signals back to analog signals.

BACKGROUND TECHNOLOGY AND PROBLEMS When processing analog information signals such as video signals and audio signals, once they are converted into digital signals and subjected to various signal processing in the digital state, it is difficult to adjust them with respect to unopened rectification and temperature characteristics. Recently, digital processing as described above has been attracting attention for egg weight, as it has various benefits such as stabilization.

In this case, the input information signal will always be subjected to D/D conversion and D/A conversion, but if the input-to-output characteristics are linear, the signal will be output with the same number of bits as the input bit number. This allows the input data to be reproduced well.

However, when applying nonlinear correction to input data to obtain output data, outputting the same number of bits as the number of input bits will cause so-called rounding errors and dither noise. There are drawbacks to this.

That is, for example, lightning strikes, television signals or V 1'
As shown in Figure 1, the reproduced video signal from R is subjected to gamma correction according to the characteristics of the cathode ray tube so that the density (equivalent to reflectance) appears linear on the cathode ray tube screen. . Therefore, when outputting this television signal to something other than a cathode ray tube, such as a printer, to obtain a predetermined gradation image, the density of the gradation image cannot be determined manually as shown in Figure 2. However, since it has non-linearity, it is not possible to obtain a density linear image.

Therefore, when outputting to a printer in this way and trying to obtain a good gradation image, it is necessary to apply inverse gamma correction, which has nonlinear characteristics opposite to gamma correction, to the human input signal as shown in Figure 3. ,
As shown in FIG. 4, the density of the gradation image is made to be linear with respect to human power.

This inverse gamma correction is performed before D/A conversion because it is more stable and accurate when performed at the digital signal stage. Therefore, conversion errors during D/A conversion become a problem.

Is the human power-output characteristic linear as shown by the solid line a in Figure 5?
In such a case, even if the number of output bits is the same for the large number of bits, as is clear from the figure, the human bit level corresponds one-to-one to the output bit level, and when the D/A changes sufficiently, the human power will not be reproduced. I know it will happen. However, when the human power-output characteristic is a non-linear characteristic as shown by the solid line in Figure 5, if the number of bits of human power and output are equal, in the example shown in the figure, the two human power bit levels are 1. Since the output bit level is set to two, the human input bits are damaged and a so-called rounding error occurs, and when D/A conversion is performed, the appearance of the human input signal deteriorates due to this error.

Purpose of the Invention In view of the point 1, the present invention is to solve the problem that even when non-linear correction such as inverse gamma correction is performed on a digital signal, a so-called rounding error occurs in D/A conversion. This makes it possible to reproduce input data with good accuracy.

Summary of the Invention In this invention, when nonlinearly correcting a digital signal of a human input signal obtained by A/D conversion, a memory in which a data conversion table is written is used, and the data of the conversion output from this memory is used. This is to prevent rounding errors from occurring when the human input signal is D/A converted by using data with a quotient bit number rather than the bit count of the human input digital signal.

EXAMPLE An example of the present invention will be described below with reference to the drawings. In this example, a color video signal obtained by a still video camera is transmitted through a telephone line, and the receiving side receives a predetermined gradation. This is an example in which the invention of "C" is applied to a transmission system for obtaining hard copies of still images.

In this case, the color video signal is transmitted with AM conversion.

FIG. 6 shows an example of this still image transmission system. (1) is a video signal reproducing device that reproduces still color video signals, and the still color video signals are transmitted in one field or one frame on a disk-shaped magnetic sheet. The unit is “
ζ It is recorded as concentric tracks, and the playback device (
At step 11, a reproduced still color video signal is obtained by repeatedly reproducing each track.

Further, this still color video signal is recorded on a magnetic sheet by, for example, a video still camera.

Zunawaji, the magnetic sheet is stored in a cassette, and this cassette is loaded into a video still camera.In this camera, the subject light is converted into an electrical signal by an image sensor such as a COD, and the recording process is performed as described above. Still color video signals are recorded on single concentric tracks on a magnetic sheet. In this case, the still color video signal is, for example, 6
It has a band of Mllz.

The still color video signal from this video signal reproducing device (11) is supplied to the A/D converter (3) via the clamp circuit vFI (21), and the still color video signal from the terminal (5), e.g.
8 per °ζl sample by Hz clock signal WC
This is converted into a bit digital color video signal, which is written to the frame memo (January 4) by the clock signal WC.

Frame mechanism 'f-IJ' (41 is for time axis expansion of the color video signal, and the read clock signal Ji=l-Rc whose frequency is 1/6400 of the corrupted clock signal WC is sent through the terminal (6). Frame memo January 4th
).

Therefore, from this frame memory (4), the band of the color video signal is, for example, 8001 as shown in FIG. 7B.
A compressed image of 1z is obtained, which enables this transmission through the telephone line and image reproduction on the receiving side.

The digital color video signal read from this frame memory (4) is supplied to a digital AM modulation circuit (7). On the other hand, through the terminal (8), for example, 1700112
A carrier signal of is supplied to this modulation circuit (7).

Digital AM modulation is then performed in this modulation circuit (7) as will be described later, and a DSB signal is obtained from the treble. Here, in this digital AM modulation circuit (7), as will be described later, inverse gamma correction of 1 is performed on the color video signal at the same time, and the output digital signal from this is 1 from the frame memory. The 8-bit sample is converted into data of quotient focus, for example, one sample IO bit or more, and taken out, and measures are taken to prevent rounding errors.

The DSB signal from this digital AM modulation circuit (7) is sent to a D/A converter (with the same number of bits, that is, 10 bits or more) (
9), the signal is converted into an analog AM/MTS signal, which is then transmitted through a telephone line via a low-pass filter, an amplifier (11), and a line transformer (12). This signal has a carrier frequency of 1700 as shown in Figure 7C.
The frequency spectrum is distributed in a band centered on Hz and extending to 80011 z at the top.

Next, the digital AM modulation circuit (7) of this example and its operation will be explained.

That is, in this digital AM modulation circuit, the human input data is converted into a signal S1 corresponding to the level of the human input data by one analog signal in A in FIG. 8, and a signal S1 in FIG. By converting signals 81 and S2 into signals 81 and 81 and selecting them alternately with carrier frequency pulses Sc, an AM change i not shown in FIG.
An IA output signal can be generated.

In this case, the conversion of this data is performed using a data conversion table, and here the content of the data is determined by the signal S.
Two data conversion tables corresponding to s and S2 are prepared.

In addition, non-linear correction is performed at the stage of digital AM modulation.

That is, since the output of the video signal portrait device (1) is intended to be reproduced by a color cathode ray tube, it has been subjected to gamma correction. Therefore, when transmitting over a telephone line and using a printer on the receiving side to obtain a predetermined interfloor image, it is necessary to apply inverse gamma correction to set the gamma of the entire system to 1. This inverse gamma correction is performed during this digital AM modulation, and in this case,
When converting human data, this is done by using inverse gamma correction as the converted data.

Figure 9 is an analog explanation of the conversion from input data to output data. When the non-linear complementary IF is not applied, the signal S1 has a neo-like conversion level as shown by the broken line (13) in the figure. , the signal S2 has a normal conversion level as indicated by the broken line (I4) in the figure, but when non-linear correction is applied as in the present invention, the signal S1 changes to the solid line (15) in the figure. The signal S2 has a conversion level as shown by the solid line (16) in the figure, and the signal S2 has a conversion level as shown by the solid line (16) in the figure.

Furthermore, as mentioned above, rounding for non-linear correction is required! In order to reduce the j error, the converted output data is the west bit data compared to the human input data.

In other words, ζ is the inverse gamma correction curve and ζ is the foul rate (electric ratio E
) The following formula can be derived from the relational expression % formula %) vs. concentration, and this is used.

Eo = C(akI!' -1) ...121 Dither noise due to quantization rounding error becomes a problem in the part where the inverse gamma correction curve has a gentle slope, but if the number of conversion output bits is increased, , this rounding error becomes smaller and the problem of dither noise is solved. Figure 5 above is an enlarged view of the part where the inverse gamma correction curve has a gentle slope. If it is converted into 10-bit data, the human character image quality will not be impaired and the error will be small, as shown by the broken line C in the figure. In other words, 256 (11i+) data in 8-bit representation are each mapped to data in 10-bit representation based on the above-mentioned equation (2), and a data conversion table is created.

In order to reduce the rounding error and prevent the occurrence of dither noise, a sufficient effect can be obtained by converting the output data to a bit number that is one or more bits higher than the output bit number.

However, in this example, we go further, and for any inverse gamma correction curve, the quotient bits are set such that the receiving printer can achieve the required gradation expression capability.

Now, assume that the inverse gamma correction curve is y=f(x), the required gradation expression capability is a bit, that is, 21 bits, and the required number of output bits is n bits.7 bits. Human power is more than a bit.”
It is expressed as 2 °C, and the output y is 1:1 for the human power
If so, the gradation after inverse gamma correction will naturally be expressed as 2&gradation or higher. So, the minimum increase amount ΔV nun of output y
The required number of output bits n can be calculated by subtracting the number of bits that can be used in the table IMIiJ.

Since the minimum amount that can be expressed in ribi and 7 points is ] / 2 '', it becomes Δymu + - 1/ 211. Inverse gamma correction curve y = convex and increasing in R key
In the case of f(x), ΔYwn=f(1/2″), so f(1/2&)=1/2n, °, n=Iog2 (1/f(1/2&)). If the number of bits is n or more, it has the ability to express 28 gradations.

Next, let's explain the above using a specific example.

As the inverse gamma correction curve, y = x 2.2, which is the inverse correction curve of the gamma correction currently implemented in cameras, is used.

Generally, gradation expression becomes severe when the amount of increase in output y is small, as mentioned above, but curve y =
For x 2.2, the d-cycle expression becomes severe near the value of 0 for y. An enlarged view of this curve near y=0 is shown in Figure 1O.

In FIG. 10, the solid line d represents the 10-bit output, the dashed line e represents the 12-bit output, and the broken line f represents the 14-bit output for the input cuff bits.

As is clear from Fig. 10, when the output is expressed as 10-bit data, in the worst case, four human forces correspond to one output value (the output is 1-0).
). Therefore, the gradation in this case is 2''/4=32nd floor IM,°, log232=5 focal points.

Similarly, when the output is 12-bit data, three inputs correspond at the worst point (output is at rOJ), so the gradation in this case is 2'/3 = 4
2.7 floors, 42.7 = 5.4 bits.

Similarly, when the output is 14 bit data, the two inputs correspond at the worst point (output is 10"), so the gradation in this case is 2'/2. =64 gradations; Iog264-6 bisoto.

Here, since the required gradation II & 1 representation capability is 64 gradations, it will be satisfied if 14 bits are used as the conversion output data. Therefore, for each data of input cuff bits, ′ζ is 14 according to the non-linearity inverse correction curve.
If you allocate bit data and use this as a conversion table, the first floor has the ability to express up to 64 gradations! ＋ll! il
will fall.

The above is a method for calculating from the inverse correction curve of FIG. 10, but C can also be calculated by division as follows.

First, since the required gradation expression ability is 64 gradations,
It is only necessary to obtain an output y corresponding to 64 gradations for an input X of 64th floor steel. In other words, it is sufficient if the output y can be distinguished for different human forces X in places where the gradation is severe (where the amount of increase in y is small).

When the amount of increase in y is small, input x = Q and X = 1/6
Between 4 and + (D out;) J y (7) increase W ha A'
i w= (1/64)'・2t,? (7) If the number of output bits that can represent the minimum increase N A y W is n bits, the minimum unit that can be represented by n bits is l / 2 ''. Therefore, in this minimum unit, ΔY
If man is expressed, the output after inverse complementation +E will be on the 64th floor side.

To summarize the above, 1/2 n - Δ )'mn' = (1/64)
2・ノ,',2+t−□(1/64) ”Yotsu゛<, n=13.2.

FIG. 11 shows a specific example of the digital AM modulation circuit (7). At the opening of the park, (21) indicates ROM, and the R, O
M (21) has two data conversion tables (
22A ) and (22B > are incorporated into ζ.

One conversion table (22^) is a conversion table on the positive side, for example as shown by the curve (15) in FIG.
It consists of, for example, 14-bit conversion data for each of the human input data according to the correction curve. The other conversion table (22B) is a negative side conversion table shown by the curve (16) in FIG. 9, and similarly consists of 14-bit conversion data.

Data for 256 samples of 14 bits is written in the ROM (21), and 8-bit address data is supplied to this ROM (21) and outputted.

Then, as the address data of ROM (21), one sample of human data is 8 byy l (bo = by)
Seven bits excluding the least significant bits b and o are supplied, and a pulse signal Sc of a carrier frequency (1700 Hz) is supplied from a terminal (23) as the tenth bit of the address data. The table (22A) is selected during the "1" period of the pulse signal Sc, and the table (22B) is selected during the "1-0" period of the pulse signal Sc.

For example, when the input data (7 bits) is 0, the ROM
8192 data are read from (21).

When the human power data is 127, data of 16383 or 0 is read from the ROM (21). As mentioned above, the data read from this ROM (21) is supplied to the D/A converter (9) with the same focus number of 14 bits, and an AM modulated output as shown in Fig. 8C is obtained. This is converted into a 1 sinusoidal wave by a low-pass filter a.

Also, as usual in Fig. 12, the conversion table (22A>(
In addition to 22B), for example, 7 types of conversion tables can be stored in ROM (
21) may be stored in ζ.

One set of tables uses 256 bytes, so by using a ROM with a capacity of 2 bytes, it is possible to prepare 8 types of tables. In this case, the 8 types of tables have different data conversion characteristics. The top 3 of the 11-bit dresses) a+o
+ a9. What's wrong with AS?

You can choose which table to use. And ROM
A pulse signal SC is supplied from the terminal (23) as corresponding to bit a7 of the address for (21),
As the remaining 7 bits, human data (bx to bt) which does not include the most-1st bit is supplied.

By using a plurality of tables in this manner, it is possible to provide versatility, such as selecting an appropriate NL as the inverse correction for a plurality of gamma corrections with different gamma coefficients.

Note that in the above example, the most significant bit of the 8-bit data was removed and it was set to ``ζ7 bits.''
This is because the ROM of the hood is easy to handle, so 8-bit data may be used as it is, and a 9-bit word ROM may be used in combination with the pulse signal of the carrier frequency.

Alternatively, instead of the ROM, a RAM may be used and the table may be written in the RAM in advance.

A balanced modulation circuit may be constructed by making the data to be mapped complementary rather than positive or negative.

Further, the present invention can be applied not only to a still image transmission system as shown in FIG. 1, but also to an AM broadcast modulation circuit.

Furthermore, the invention is not limited to the case where it is applied to the AM modulation circuit as described above, but can also be applied to any case in which non-linear correction is applied to a human input signal at the digital signal stage.

Effects of the Invention According to this invention, when it is necessary to apply ``ζ non-linear correction to a human-powered digital signal, the human-powered data is converted into corrected output data of quotient bits, so the rounding error is reduced. In addition, according to the present invention, nonlinear correction and rounding error countermeasures can be performed simultaneously using a data conversion table.
Another advantage is that the configuration is not complicated.

[Brief explanation of the drawing]

1 to 4 are diagrams for explaining the case of applying inverse gamma correction as an example of nonlinearity correction, FIG. 5 is a diagram for explaining the present invention in detail, and FIG. A block diagram of a transmitting side of a still image transmission system to which the invention is applicable;
7 to 9 are diagrams for explaining the invention, FIG. 10 is a diagram for explaining the invention, FIG. 11 is a block diagram of an example of the main part of the invention, and FIG. 12 is another example. FIG. (3) is an A/D converter, (21) is a ROM as a memory in which a data conversion table is written, +91 is a D
/A converter. 1st factor Figure 3 Input 21 @4 factor λ force

Claims (1)

[Claims] an A/D converter that converts a human input information signal into a digital signal, a memory in which a conversion table is written that converts the digital signal into data obtained by applying non-linear correction to the input information signal, and a D/A converter. The converted data has a bit number of bits greater than the number of bits of the digital signal of the human input information signal, and the D/A converter also has the same number of bits as this converted data. A signal processing device in which the conversion data is converted into an analog signal by the D/A converter.