What graphic chips could be used with a W65C816

You won't be able to get a v.92 link with both sides of the connection being analog. I'm a little rusty, but I was the technical lead for a large (10k+ subscriber) dialup ISP in the late 90s - early 00s. I think the maximum speed you'll be able to achieve with both ends being analog is 33.6k. Wikipedia seems to agree with me on this: http://en.wikipedia.org/wiki/V.92

You could achieve this fairly easily if you were willing to grab an old Cisco AS5248, AS5300 or MaxTNT and a channel bank, possibly putting a simple software PBX system in the middle of it for doing the actual call routing. This will still cost you at least $1500 in equipment if you've got good access to surplus telco gear (eBay) and probably the better part of a week of time. Even then, the AS5248 might only support v.90. If you can live with 33.6k then just grab any old modem, throw a 9V battery in the loop and configure one to answer (ATA) and the other to dial without waiting for dialtone (ATX0D if memory serves).

Unless you are willing to invest a LOT of time and effort into it, you won't be able to build a v.92 softmodem in less time than you'd be able to acquire some used equipment and lab up a simple digital test bench. It's doubtful you'd find anything inexpensive or open-source for this, as the patents on v.92 are still in effect.

(edit to address theoretical question in comments)

In theory, you should be able to hit 56k easily. You don't have a PSTN that is stealing the LSB every 6 frames (1ms) and you also don't have a hard requirement to keep the bandwidth of the 20 or so feet of copper within some telecom spec. Without the PSTN you could just connect the devices through 20 feet of CAT3 and probably get a good megabit out of it with RS-485 without much effort at all, but that's not what the question was asking.

I do know that the consumer-end modems (the ones you buy and hook up to a computer) are not designed to negotiate a v.90/v.92 connection with themselves. If you take two v.92 modems and hook them up as I describe above, the fastest rate you'll see is a symmetric 33.6k. They were never designed to respond as if they were the digital end, and it's not possible to "adjust" them to do so. You'd have to completely re-engineer their firmware which, if you're going to go that route, you may as well build your own modem instead of trying to reverse-engineer someone else's.

First, I'm going to assume this is what you are looking for:

1. An Atmel xmega micro-controller
2. Has IO pins configurable as input (with optional pull-up resistors), or as output
3. Has both slave and master SPI functionality (preferably hardware support)

Comment below if this isn't exactly what you're looking for.

Pretty much all xmega (maybe even all) devices feature IO pins with select-able internal pullup/pulldown resistors.

This is the IO block diagram for an xmega-AU device:

Notice that the top-right section of the general IO pin block diagram has select-able internal pull-up and pull-down functionality, as well as a pull keep or totem pole input options. This is common across all IO pins configurable as general IO pins.

That just leaves hardware SPI functionality. Luckily, the same xmega-AU device supports hardware SPI in both master and slave modes. See chapter 22 for full details on what is supported.

This is not exclusive to the xmega-AU series, this just happened to be the first one I clicked on. IIRC, pretty much all xmega chips have the same or very similar general IO block diagram, and they also support hardware SPI in master and slave modes.