For collision I'd look at line, circle intersection - there's an example in the code snippets section.
Anyway, I'd have lines representing the edge of collisions, so a loop could be a circle. Because you'd use segments to make the circle you could have sections active and not active depending on what other lines the player was standing on. For instance if Sonic is scooting along and goes onto a loop, then the loop has to start when Sonic hits the bottom right, then let him follow it round and go over the loop section at the start again. So the part where Sonic would exit the loop has to be disabled for collision when Sonic is entering the loop, like let him in then close up again. This could be based on checking what lines are being stood on, and deciding if a line is collidable based on that. Would get pretty complex but it's an option, and would let you have practically any shape of level.
The main reason that I mention line collision though, is that there are several benefits that may not be apparent...
* Lines can easily provide extra info, like what angle they are pointing in. So responding to collision can be very neat, as that vector can be got.
* Lines could be moved, rotated, even linked to an object or sprite. For instance a big Pinball paddle. You would have to let the paddle take presedence over the player, like if Sonic is touching a paddles line when it flips, it would affect the player movement and speed accordingly.
* Line collision is as smooth and accurate as you could get in 2D - instead of 2D pixel collision, you have a smooth line, no worries with getting stuck on individual pixels - it's just a much better way to get curves and loops.
* Each line could have it's own properties - for instance a line could be bouncy!, or sticky, or deadly, or whatever you feel like coding.
* Line and Circle collision are like peas and carrots, they go great together - Sonic is practically a ball, the level can be represented in collision lines - add in a good collision response system and you'd have a very convincing and smooth platform engine.
What I would do is get some line-circle collision code and experiment with adding lots of lines and checking collision on them with a small sphere, work it over a few times till you have something your happy with. Then you'll need both a tile editor, and a collision editor - so you can design levels then draw the collision lines. These lines could be applied to individual blocks - like a loop image, with a file containing all the line information for it, then just place these things in an editor to design the level. Takes a bit of work to get started in this, but really it's the only way I'd tackle Sonic.
I would make a sonic *inspired* game though - you could spend years trying to perfect Sonic, but why bother when you can take it a different and ultimately more interesting route. If people want to play Sonic, they will go and play sonic on their DS, 360, or PC through emulation, or one of the standalone handhelds, or one of those TV plug-in things, or online in a website emulator, or on a Megadrive itself. The list of ways to play Sonic goes on and on and on - so I say people should re-imagine games rather than remake them. Most of the games I've made are re-imaginings, taking the origial game basics, what made it fun, why do I want to redo it, what would I change... With Sonic the gameplay is solid, and you could take that and make it your own - invent your own character and scenario and try and get all the best bits of Sonic games in there. Your not cloning something, your re-imagining a game you enjoyed as a kid, and have the chance to take out some of the stuff that frustrates you as well as opening the game up to people who might not be interested in playing an old game like Sonic.
Right now I'm finishing off a remake of Bruce Lee from the 8-bit era, a platforming-kung-fu-collect-em-up - and besides taking a lot longer than I thought it would, it's been a blast - I haven't felt once like I should remake Bruce Lee, it's too much fun to disregard all that convention and concentrate on making it enjoyable instead.
A good test engine is key, as I said, go nab the Line-circle collision stuff and get a circle rolling around. With that as a start, everything you add onto it makes it more and more fun, and interesting to develop. Consider what cool ideas you could bring to the table.
Hope that doesn't sound too much like a rant!, just wanted to post up my thoughts seeing as retro tends to dominate my projects. Personally I can't imagine myself ever doing a clinical remake, I'm far too arrogant for that ;D.
