How to approach software based composite video in P1+
Tubular
Posts: 4,702
I've been thinking this morning about how to approach a composite video driver for the P1+. Chip has mentioned this will need to be done in software, but since we'll have greatly improved MIPS compared to P1, this shouldn't be too much of an issue.
The P3- DE0 emulation we have now can do 80 MIPS, the upcoming P1+ probably 100 MIPs, so anything started now on the existing P3- platforms shouldn't be hard to port across.
I've been reading up on the RGB to YUV space conversion, and its requirement for scaling (multiplications) and blending in the carrier.
Because I'm new / not expert in this area I'm thinking it might be best to use 3 cogs, at least initially,
1 to do the 'U' portion of the signal and carrier and colorburst on the back porch
1 to do the 'V' portion of the signal and carrier
1 to do the Luminance Y signal, add the U and V signal (with carrier pre-calculated), timing, and send the final value to the pins
However I suspect there are a lot of tricks already employed over the years, and I'm not yet aware of these. The thing I like about the above approach is there would be plenty of room for lookup tables, and the U and V calculations can be slower and still get some kind of result even if not updated on every pixel.
Then try and work out how to interleave it into a single cog, if that looks possible.
Has anyone else had experience in this kind of area? Any advice would be appreciated
The P3- DE0 emulation we have now can do 80 MIPS, the upcoming P1+ probably 100 MIPs, so anything started now on the existing P3- platforms shouldn't be hard to port across.
I've been reading up on the RGB to YUV space conversion, and its requirement for scaling (multiplications) and blending in the carrier.
Because I'm new / not expert in this area I'm thinking it might be best to use 3 cogs, at least initially,
1 to do the 'U' portion of the signal and carrier and colorburst on the back porch
1 to do the 'V' portion of the signal and carrier
1 to do the Luminance Y signal, add the U and V signal (with carrier pre-calculated), timing, and send the final value to the pins
However I suspect there are a lot of tricks already employed over the years, and I'm not yet aware of these. The thing I like about the above approach is there would be plenty of room for lookup tables, and the U and V calculations can be slower and still get some kind of result even if not updated on every pixel.
Then try and work out how to interleave it into a single cog, if that looks possible.
Has anyone else had experience in this kind of area? Any advice would be appreciated
Comments
One thing we might think about is getting a basic signal done. That will be my first step. Monochrome, interlaced and not. Then PAL. From there, add the colorburst for NTSC and PAL.
At that point, it's an empty container. How color gets done and modulated can vary.
P2 (it's P2, not P1+) will have the ability to do component out (which a lot of tv's have as in input). That can be upconverted to HDMI or whatever you want externally.
Why have an RGB frame-buffer? Wouldn't it be much faster to have a YUV frame-buffer and do any color conversions when drawing?
Marty
-Phil
-Phil
If we've got killer VGA, which Chip says we will, then doing composite, component, S-video NTSC color isn't going to be a problem. PAL might, and that one comes down to jitter, same as it did on P1. PAL requires a very precise color reference and phase shift every scan line where NTSC can tolerate very significant variations without serious artifacts. We shall see when we get the FPGA.
IMHO, a smaller 4 conductor connector does component, and that works all the way up to 1080i, if desired. Nearly all TV's that have higher resolutions than SD have component inputs on them. Interestingly, they don't all have VGA. I see fewer VGA ports on TV's than I do component here in the States. I've not yet seen one that doesn't accept composite. I have seen a few that do not accept S-video.
Anyway, shouldn't be an issue given P1 can do an all software NTSC color display --a pretty nice one actually, with just that 3 pin DAC. Seems to me, that same technique Eric Ball did for us should work very nicely with a real DAC. Or some variation on that... Should be one COG, assuming there isn't advanced color management going on.
@Lawson: Totally. A native color buffer will be fast, and we can do a lookup table on 8 bits and less per pixel too.
Analog is nice, because it does handle the full HD resolution, and it's free of the DRM / licensing hassles inherent in HDMI. I like the idea of having that packaged up and managed by anybody but us, just feed it clean analog signals and it's going to be tough to tell the difference. A clean component signal appears pixel perfect on the sets I've tried it on.
That sounds like a good place to start. But do these need a PLL, or are they bitbanged? Do you have links?
Yes good idea. For a proof of concept on P1 I even just did a single line repeated over and over. The monitors seemed happy enough with that
Thats an interesting idea Marty. Have them all pre-scaled and everything, and just add the values and apply color carrier as the final step.
@all, I need to put out composite video, its non-negotiable unfortunately. But as Phil points out its a simple two wire output, and is second to none for ease of capture, recording etc.
David see Chip's first post his thread, towards the end
http://forums.parallax.com/showthread.php/155132-The-New-16-Cog-512KB-64-analog-I-O-Propeller-Chip
There are a few techniques. No, they aren't bit banged in the sense of manually stuffing values into the DACS, or toggling pins. Waitvid is used. Yes, they are bit banged, in that waitvid is used with "pixels" and "colors" that result in signals, including the colorburst, that the silicon would normally do. A couple of these were just enhancements on P1 color / signal characteristics. Others were to best exploit NTSC artifact color, seen in that Nyan Cat thing I did a while back.
We will just need to see what Chip does with WAITVID in the next FPGA, then the work can start.
I have found it best to work out the whole signal with a basic horizontal test pattern, such as grey bars (simplest), checkerboards, etc... That can be evaluated on a scope and tuned nicely. Once that is all done, adding things one at a time works well. Add a color burst. Then phase shift it, etc...
http://youtu.be/-gyO2lRXLyg BTW, dig that killer P1 sound, Retronitus by Ahle (miss you man, hope all is well)
True, but I have seen some Chinese Monitors use Mini/Micro USB connectors for VGA, and they have USB-VGA passive cables.
I'm not sure if this is has a documented standard pin mapping anywhere, but the existence of cable-sets suggests a standard ?
I understand that the P1 has some special hw that creates the color signals for both ntsc and pal. My understanding this cannot be done by sw.
I suggested thatmaybe the NTSC hw couldbe added to one cog or to the hub. Composite monitors are plentiful and cheap and imho are not going to disappear anytimesooon.
You can do separate chrominance/luminance signals with internal pin dacs to a s-video connector and then use a s-video to rca conversion cable (as I know they are simply resistive/capacitive adders of the two signals) or add them on the pcb before going to an rca connector.
This approach should easier the signals generation, unify the composite/s-video drivers and allows for a wider device compatibility.
@Cluso, it can and has been done in software on P1. Really, it's just going to be a question of what method makes sense.
-Phil
I think in the P1+ the video shifter will go direct to the DACs, which are not clocked, or perhaps clocked by the PLL. We anyway need an analog output for composite, so "DAC only" is not a problem here (in opposite to to a TFT driver).
About NTSC/PAL generation:
I think the simplest methode to generate an NTSC or PAL signal with colors is to make a waitvid for every pixel with for example 8 samples. For every color you want to use you have a 8-sample table that contains the luminance level with overlayed color burst. Phase and amplitude of the color burst define the color. For PAL we need two tables per color, one shifted by 180°.
Bitmap drivers should be quite simple if this works.
Andy
-Phil
Yes, the any-integer will allow base burst Crystals, and also the x4 burst crystals.
( I guess that PLL N can be > present 16, maybe 6-8 bits ? )
But to generate the chroma-sine in software if we use those crystals (very common) and a Prop2 16xPLL for sysclock = 229.09 mhz
A sysclk that should be the upper limit of P2, you should probably get all the colors NTSC can show.
27.000 Mhz crystal (used in PAL/NTSC DVD players, exact multiple of the PAL and NTSC line frequencies) and with a 8xPLL =216MHz sysclk could also be good.
Interesting idea Andy. But wouldn't you need to also have the system clock a multiple of the colorburst, so you don't introduce a phase shift depending on which horizontal pixel you're outputting?
I'm assuming for now we don't have separate video PLLs. If we do, it'll all be easier.
One very nice advantage of that is precise alignment of the pixel clock to the color burst. This allows for artifact colors as well as ones generated normally.
Do you have a link?
BTW I have never bothered to understand the color generation for either NTSC or PAL.
In the P1, the video PLL is 16x the color burst frequency, which allows for 16 distinct hues. Color modulation is done by wiggling the video output by one LSB. Although this results in fairly low color saturation, you can achieve "supersaturation" by wiggling it between max brightness and overflow to zero (sync level). Because this results in inverted modulation, hues are shifted by 8 places (180 degrees).
-Phil
Step1: create a burst of color outside left viewable area, this will be the reference.
Step2: You keep a sine wave going for color and the height of this sine determines the saturation you want.
Step3: You offset this sine wave compared to the reference color burst to send the color (hue) you want.
Step4: Mix this in (analog sum) with the black&white/sync information= composite video.
P.S with Step 3 it's not nice to jump to opposite side of the color wheel as that phase shift will really mess up the sine wave.
So graduated color changes at max 45deg per pixel is better looking.
If we want to use external analog ICs (but we prefer not), we could do something like the C64
The VIC-II used the 14.31818 MHz master clock input (4 times the NTSC color burst frequency of 3.579545 MHz) to produce quadrature square-wave clocks. These clock signals were then integrated into triangle waves using analog integrators. The triangle waves were then integrated again into sine waves (actually rounded triangle waves, but good enough for this application). This produced a 3.579545 MHz sine wave, inverse sine wave, cosine wave and inverse cosine wave.
An analog summer was used to create the phase-shifts in the Chroma signal by adding together the appropiate two waveforms at the appropiate amplitudes. The Color Palette data went to a look-up table that specified the amplitude of the waves by selecting different resistors in the gain path of the summer. The end result was that we could create any hue we wanted by looking at the NTSC color wheel to determine the phase-shift and then picking the appropiate resistor values to produce that phase-shift.
Color Saturation was controlled by scaling the gain of the summer. When we picked the resistor values to determine the output phase-shift, we also scaled them to produce the desired output amplitude. Luminance was controlled using a simple voltage divider which switched different pull-down resistors into the open-drain output. We could create any Luminance we wanted by choosing the desired resistor value.