over the last year developer Mike Simone (MikeS) has been working hard implementing high-quality composite1 and s-video output capability in the MiSTer FPGA project.
he's finally done cooking. result? largely fantastic. look at that shot of NES Gun.Smoke in composite. it's exquisitely analog, perfectly imperfect (the display is my PVM-2950Q). i love it and am excited about likely adopting composite video, and sometimes s-video, as my daily driver for various consoles.
but, getting ahead of myself. let's talk about how this is working, why i like it, and that time two of Sonic's creators went on record shit-talking Sega's penny-pinching on the Mega Drive.
chasing imperfection
for retro gaming, authentic or even just clean-looking composite video output is a surprisingly tough nut to crack. while RGB-to-composite converters are a dime a dozen, generic off-the-shelf solutions that are not tuned to precisely mimic the output timing characteristics of a given console's hardware introduce extremely annoying artifacts like rainbowing and the dreaded "dot crawl." various consoles produce those artifacts already2 and generic converters make them worse. such artifacts are a key reason a lot of gamers developed a distaste for composite video. but it can look pretty good, actually3.
MikeS attacks this challenge in two key ways: creating custom MiSTer FPGA cores that natively simulate the timing characteristics of each console's luma / chroma signals4, and by devising an open-source hardware-based "luma trap" that cleans up the composite signal further.
a little while ago MiSTer hardware vendors like Ultimatemister began selling the non-powered, "passive" version of MikeS' luma trap, which gave the best-yet composite signal. last week Misteraddons became the first to produce the superior powered, "active" version, which, having been watching MikeS' progress for nearly a year, i instantly ordered.
the Misteraddons active board's design requires power from either USB-C or the MiSTer i/o board itself, which can be jumpered to output 5V on VGA pin 9 (iirc?). the latter is very convenient as no extra power cable is required. it is a bare board with no case and my retro setup uses metal racks, so i wrapped it in its own packing foam to protect it (cute 🥰). it outputs s-video and composite simultaneously, making switching between the two signals as easy as changing display inputs.
one drawback of Misteraddons' implementation is that, unlike MikeS' open-source hardware spec, it only supports NTSC video. my CRTs are PAL-capable and there are various PAL-only games across various systems i'm interested in, so that's a bummer. my understanding is that Ultimatemister's imminent active luma trap product will include an NTSC / PAL switch, so keep that in mind if you care about PAL.
testing some Y/C cores
i don't want to focus on picking apart screenshots because i don't have satisfactory ones handy. that said, here is the MikeS Y/C MiSTer NES core playing Super Mario Bros. in s-video (PVM-2950Q, Sony CXA palette):
as far as i can tell so far s-video is identically sharp and colorful across every core, or close enough. however, everyone always says composite -> s-video is a much larger leap than s-video -> component/RGB, and they're right. comparing s-video to RGB, the pixels are 80% as clear and the colors just somewhat more drab (not apparent in this simple SMB scene). it's sharp enough that dithering, either in pixel art or, say, PSX 3D, still fails to blend, looking pixelated instead. so, the picture is appreciably more "organic," but we can do considerably better. err, worse. betterworse.
enter composite. again, not the most telling photo here, as it came out a bit blurry and it's not the most illuminating scene. but some of the actual differences should still be noticeable if you compare to the s-video shot:
the image is further softened; edges become less defined. colors take a more serious hit (again, less apparent than it could be in this particular shot). in motion, moving objects exhibit a modest pixel-crawling effect, but dot crawl on still objects is close to non-existent thanks to MikeS' painstakingly careful software/hardware implementation.
let me just summarize composite findings so far: NES and SNES cores look cozy af, a perfect level of softness and just 1990s TVness that grants the glowing phosphors additional life. PC Engine (TG16) composite is extremely clean, maybe even moreso than a real Duo's--great. PSX looks nice, and i have much more to say below. Neo Geo MVS5 is terrible, with diagonal artifacts in many colors; will stick with s-video or RGB. and MD looks like total vom, which means MikeS is doing a faithful job of reproducing that machine's notoriously awful composite signal.
Mega Drive? more like mega dirty
so, Mega Drive. Genesis. it's the console that both famously benefits from the blurring afforded by composite--many, many games relied on strategic dithering to simulate smoother gradients--and one that had an absolutely horrid composite signal.
straight from the lips of Sonic's creators6:
and here's an enthusiast weighing in on various MD video artifacts:
The video circuitry of early Mega Drives is one big mess, not only do you get the rainbow banding through composite (due to the CXA1145 just not being up to the job), you also get the jailbarring via RGB on the blue channel due to the blue trace being right next to the composite subcarrier on the PCB (which has become infinitely more noticeable when connecting to an LCD, but is still there on a decent CRT TV or monitor). Different VA revisions can suffer at different levels of this, but AFAIK, there isn't a single board from VA0-VA6.8 that doesn't have some kind of video issue on composite or RGB.
(heh, just the other week i posted that 2016 photo of my early-model Genesis + Sega CD exhibiting blue bars in the black background when hooked up via RGB. so that's why!)
anyway, i've really, really been looking forward to being able to play MiSTer's MD core in composite. i'd grown dissatisfied with how RGB / component (and also now, s-video) more or less sabotage the extensive dithering often used in MD graphics. the ur-example everyone always cites comes in the first Sonic the Hedgehog's waterfalls, which look like sharply defined vertical lines in RGB or s-video but turn into beautiful transparencies (accompanied by less-beautiful rainbow artifacting) when displayed in composite.
here's a great video dissecting what's happening there:
for more examples, Chris Covell has many well-done MD RGB / composite comparison images on his website. however, imo the composite colors i'm seeing on my CRT setup don't feel quite as muted as in many of his examples. the Gun.Smoke, SMB, and Sonic photos i took are much more representative of what i'm seeing.
so, first thing i did when i got the MikeS cores + hardware all set up was run to the Sonic waterfalls. sure enough, they were nice and transparent again. but i was mildly disappointed that even with the tuned Y/C core and the active luma trap rainbowing was still apparent (and is more so when you scroll the screen). however, i'm pretty sure you'd see the same mild rainbowing on a real MD--MikeS is mostly going for accuracy, not improving on reality.
what bothered me more was some jarring line artifacts on certain details, like the dark undersides of palm tree leaves (screenshot is from MiSTer forums). i found that if i increased the display's sharpness even a little (i usually play at 0 sharpness) these artifacts rapidly became visible on other objects too, like Sonic's golden rings.
now, it's been decades since i've used a Genesis in composite so i don't fully recall how bad it got, and i don't remember this vertical line stuff. (i will test real Genesis hardware soon to check.) but again, MikeS is going for accuracy and this is likely just how it was, assuming he got all the timings dialed in properly. the MiSTer MD Y/C core's composite certainly lives up to the real thing's blighted reputation, and to be honest, the PQ is a bit too ugly for my tastes. but it's this, or ruin all the dithering again.
in response to MiSTer forums users remarking on the MD and arcade crud, MikeS posted screenshots of a further tweaked MD core whose timings allow it to surpass a real MD's composite PQ (aside from the improved tree, note the reduction of waterfall rainbowing). it's unclear if he will actually release it (please?), but if nothing else it proves that a cleaner MD composite signal is absolutely possible.
i don't know which signal i'll favor for Mega Drive in the short term. if its composite is just too dirty for me, maybe i'll forego good dithering and such and fall back to the comparative beauty of s-video, which has few artifacts, great colors, and is at least a little fuzzier than RGB / component.
dither king PlayStation
though it doesn't get talked about as much, the PlayStation also has an intimate relationship with dithering, though in a different way. while the system can display 16.7 million colors, during gameplay it's always running in a much lower color depth due to limited memory, etc. obviously this has drawbacks. too few colors in a gradient, for example, will lead to ugly / obvious banding. to address this and similar issues with low color counts Sony implemented a very distinctive dithering algorithm which games can turn on or off at will.
as you might expect, viewing PSX through higher-fidelity signals makes this dithering stand out a lot more, giving everything touched by the obvious dither a mottled appearance that feels more like an artifact than an intended visual benefit. on the other hand the inherent blurring of a composite signal better combines the dithering with the objects it covers, resulting in a smoother image and arguably richer / more "complete" / more evocative / however-you-might-want-to-describe-it visual presentation. (some may disagree.)
here's a great video (same channel as the one above!) that describes it all very succinctly.
to see how composite and s-video fared on the PSX Y/C core i booted up Silent Hill, one of the system's most obviously dithered games. it loaded into a low-light scene and, sure enough, the dithering was both extensive and extremely obvious in s-video, just patterns of dots overlaid everywhere. switching to composite improved the dither + original image blending by an order of magnitude. the "dottiness" was still quite apparent, but less in your face than with s-video.
here's an attempt to photograph the difference in a very dark scene. note that you have to click through to see the full-size gif or the reduced width of the Cohost column will induce tons of moire artifacts:
as such, for heavily dithered PSX games my choice may well be composite. however, other 3D games look just fine to me in RGB / component (and thus s-video), so i imagine it'll be a case-by-case basis. also--to make a mild historical argument--while s-video wasn't a thing on NES, MD, or PCE, it was pretty available on PlayStation, so it's definitely a period-appropriate video signal for enjoyin' 32-bit.
tldr; MiSTer Y/C is great
i'm really into it. MiSTer's capabilities have expanded yet again, making what's imo the best retro platform in the world even better. i used to think RGB or component was the be-all end-all for all classic consoles but these days i prefer a more nuanced approach that takes into account each platform's hardware, history, and signature characteristics.
in many cases this has led me to value a more lo-fi aesthetic--albeit still on really nice CRT displays. i'm chasing a certain balance between (the loaded word) "authenticity" and prudent, targeted improvements made possible through modern tech. as we've seen, the MD Y/C core's composite might be a little too authentic for me, lol.
one question that remains is if Sorg, the MiSTer FPGA project's famously cautious / grumpy7 maintainer, will agree to accept MikeS' Y/C work into MiSTer proper. if he did there would no longer need to be separate Y/C cores, and Y/C output would just be a normal feature of most MiSTer cores. that would be great, but what MikeS has already created is plenty good as it is.
the last thing i'll add is that i don't mean to be too prescriptivist here, as most of the above is just my particular opinion and there's no right or wrong way to enjoy games.
update: @copySave shared some info on how to set this up
- you know, the single yellow plug? the type of connection cable that most consoles (i'm givin' you the side-eye, Jaguar) started including as standard around 1991. meanwhile, s-video is another single-cable video standard that fancier U.S. TVs starting including in the early '90s. it never got as much adoption, but most console makers sold s-video cables separately. MiSTer, btw, typically outputs HDMI, 15khz analog RGB, and YPbPr aka component.
- '90s kids know~
- another major factor in how nice composite looks is the comb filter of the display screen.
- Luma and Chroma are the brightness and color signals from which composite and s-video are derived, commonly abbreviated "Y" and "C". hence MikeS' custom composite / s-video-capable MiSTer cores are typically referred to as the Y/C cores.
- this is fun: MikeS has implemented his Y/C in many dozens of MiSTer arcade cores (arcade games normally only output RGB). interestingly, i read that Neo Geo core set to AES mode looks better, perhaps because it's mimicking the real AES consoles' timings vs. injecting composite timings into essentially an arcade platform? idk though.
- interview was translated by this person, and this image version is from Reddit.
- in a mostly lovable way. mostly.