Digital PAL/NTSC Video encoder

Digital PAL/NTSC Video encoder

Details

Category: Video Controller

Created: February 16, 2007

Updated: January 27, 2020

Language: VHDL

Other project properties

Development Status: Stable

Additional info: Design done

WishBone compliant: No

WishBone version: n/a

License: LGPL

Description

Connecting to the world outside

This part was completely redesigned due to variant output inpedances of different CPLD/FPGA and to reduce the moving pattern from older versions.
The two transistors are used for impedance transformation so the output inpedance has not more much effect
to the result. In the upper part the luminance signal is generated and in the lower part the chrominance
signal. The two 1,2KOhm resistors on fbh and fbl pin sets the chrominance output to
a defined level. The chrominance signal is mixed with the luminace signal by an simple capacitor, this is
also used in cheap s-video to FBAS adapters. So this works, it should be easy to create s-video by
splitting this connection, but this feature is untested.

Summary

The design written in VHDL fits in a cheap XC9536 CPLD if only PAL or NTSC signals are generated.
Walking pattern visible on some TV at older versions of the design are minimized. The design is more easy
expandable but by the cost of an increased amount of external hardware. For some cases ther are ready-to-use
jedec files in the jedec folder.

For a german version have a look at my homepage http://www.jcwolfram.de/projekte/vhdlmain.php

Changelog

7.2.2007 the first public version (0.21)

* PAL and version NTSC for 16, 20, 32 and 40 MHz clock frequency

9.3.2007 version (0.31)

* give up the way with a long ROM table in favor of an mixed signal generation
* a more clearly design and improved image quality
 

Luminance signal generation

The luminance signal generation is the same as generate B/W only signals. It uses a little table to calculate
the different levels. This component is located in file luma_gen.vhd. In some cases it is also possible to generate the luminance signal directly by the rgb source without using this component. For example, if your MCU is already generating an BAS signal you can add the chrominance signal in the simple way of a capacitor.

A simple example

To get a chip, who's working a little bit of additional logic (glue) is necessary.
The rgb signals are stored in regsters to get a more synchronous design and to avoid glitches by different
input delays. Also 2 threestate outputs for the chrominance-signals (burst and colour) and if needed, a
clock divider for the CPU/MCU clock. The luminance signal generation is clocked only by the CPU clock because faster clocking has no advantage and provides undesirable noise.
The cgsel signal can used as an fourth bit to select one of 8 grey levels or with an external switch. '0' selects
colour mode and '1' selects greylevel mode. With the cg\_pnsel signal it is possible to switch between PAL ('0')
and NTSC ('1') without any hardware changes. If only PAL or NTSC is needed, feeding this signal with statically '0' or '1' results in decreasing the amount of logic cells.

A little bit of theory

There are many usefull addresses on the Internet and so I only want to tell briefly the most important to know
about the PAL system. Horizontal timing and vertical timing are identical to the black and white BAS signal
as much as possible. The main difference is the colour carrier signal which is used to encode the additional
information.
To avoid moires on black'n white pattern the colour carrier has a crooked frequency.
Namely the horizontal frequency of 15,625 hertz * 283.75 plus the half vertical frequency (25 hertz).
With it one comes on 4,433,618.75 hertz. To generate this we can use a quartz oscillator or find a way to use
the system clock from teh rgb source (e.g. microcontroller). This can be done by using DDS (Direct Digitally Synthesis)
and a error of just abovementioned 25 hertz must be accepted, otherwise we need very wide counters to the signal production.
With a clock frequency of 16 MHz we need such a counter (accumulator register) with 12-bit width.
Every clock the counter adds 1135 (283.75 * 4) and so we get the (approximate) value of the colour carrier.
With 20 MHz clock speed this would be theoretical 908, a 10-bit-wide counter should be also enough with addition of 227.

To identify a colour signal and to synchronise the quartz oscillator in the TV, the normally quadrature-modulated
colour carrier will be sent unmodulated shortly after the synchronous impulse for approx. 10 periods. This is named
as the burst. If burst is sent the phase changes from line to line between -135 and +135 degrees.
However, after my experiences 8 periods are also sufficient. The amplitude of the Bursts amounts to 0,15V exactly 50%
of the maximum amplitude with maximum colour saturation. For the basic colours a saturation of about 50% is
completely sufficient but its also possible to create a table with another saturation. The FBAS (CVBS) signal consists
of three signals:

* the luminance signal Y = 0,299*R + 0,587*G + 0,114*B
* the U colour signal U = 0,493 * (B-Y)
* the V colour signal U = 0,877 * (R-Y)

The Y signal will transfer directly, the U and V signals modulates the colour carrier by quadrature-modulation.
In brief the U signal is multiplied by the cosinus wave of the colour carrier and the V signal with about
0 or 180 degrees of the sinus wave of the colour carrier. Besides, the tone is encoded in the phase and the colour
saturation in the amplitude of the modulated colour carrier. In the last stage the three signals will be added and
during the synchronous impulses it must be set to 0 V.

History and features

After some projects of rgb colour video output with AVR microcontrollers there was the question
about FBAS (cvbs) signal generation on a simple and easy way. For example, to be able to feed a
modulated HF signal into an aerial arrangement.
Because a microcontroller is rather inexpedient for it on account of the necessary processing speed,
the decision fell in favour of a CPLD. Of course there is for such things also special-IC's,
but I did not want to choose this way. As result there is a simple design which also fits in a XC9536 from Xilinx.
The encoder components can be also integrated into other designs, provided, the regulations of the LGPL are kept.

* the colour carrier frequency is generated from the system clock of 16 or 20 MHz
* inputs: hsync, vsync and rgb
* 8 basic colours / 8 grey steps eligible
* the same hardware can be used for PAL and NTSC

the following image shows a "screenshot" of a test program. Because of the spectral sensitivity of my digicam the colours which contains blue are brighter as viewed by eyes.

Timing

Due to the complex logic, the timing in chroma-path is a little bit critical and limits clock speed.
For correct results for 32MHz a 7ns CPLD is needed, for 40MHz a 5ns one. So fast spikes are filtered
by lowpassing at the output, a 15ns Device works well in all cases. For better timing more macrocells
(38-40) are necessary to flatten the design. Another way is to store stages of the chrominance signal
generation in stages (pipelining). This causes in delay between chrominance and luminance signals which
should not be visible on TV. Or, the luminance signal must also beeing delayed. For PAL and a little bit
less of image quality the 16- and 20MHz versions are an good alternative.

Realisation

And now te big question, how can we fit these functions in a small CPLD.
The colour carrier and the Burst signal do the least problems, the modulation and mixture with the Y signal
is less trivial. After some experiences with long ROM tables including the first official version 0.21 I
decided this was not a good way. Poor image quality and the need for an external script were the reasons
for searching another way. With the first attempts appeared that it is possible theoretically to generate
a FBAS (CVBS) signal, while rebuild the signal path digitally. But, the logic expenditure is rather high.
The new way I've found is to generate chrominance and luminance signals digitally and add them on an
analog way after lowpassing. So it's possible to limit the bandwidth of the chrominance and the luminance
signals separately.

Chrominance signal generation

The DDS synthesis and burst generation are like the older versions, but the signal generation is completely
redesigned. The component which does this is located in the file chroma\_gen.vhd.
The necessary phase shift is calculated of the 4 MSB of the carrier signal, rgb and burst. The resolution
is limited to 4 bits resulting in phase steps of 22,5 degrees.

-burst ha a phase of +135 and -135 degrees, equivalent shift vallues are 6 and 10 (16-6)
-because of V=0 on blue colour the phase shift in this case is 0 deg n odd and even lines.
-the phase shifts of the other colours can be determined by the colour circle, the value is changing between n and 16-n between the lines.
-in the case of NTSC generation the phase shift of burst and colour does not alter between the lines

The necessarey phase shift is been calculated by a little table and was added to the colour carrier
value to get the phase of the modulated carrier. For simplification only the MSB of the addition was
used to generate a rectangular signal. For black, white and grey levels the chrominance signal is switched
off, for the burst signal a output to generate a reduced amplitude is activated.
This concept is easy expandable to generate more colours in higher phase resolution and/or multiple
carrier levels.
In the result of problems with XST and configurations (ignored by 7.1) in file main.vhd the component
declaration is followed by the selected architecture. For different clock speeds there are exists 4 architectures, clock16, clock20, clock32 and clock40.

The testing of the image quality is mostly subjective by using a 5\" LCD-TV (the only i can use to thest NTSC). At PAL modes the quality is better than at NTSC modes, but this is more problem of the TV standards. The 16- and 20MHz verions are highly unuseable for NTSC generation because of \"blowing\" from coloured to non coloured areas in the image. Better lowpassing of the chrominace signal should help but is not tested.