56k Single Chip Modems

Question

I'm currently working on a project that's intended for getting old electronics that access networks through the use of a phone line to access the internet. To do this, I'm emulating a simple phone system, but I will need a modem (single chip modem if possible) that will answer at the end of the line.

Hopefully, any modems I'd like to use would have these features:

• Up to 56k speeds (V.92 performance - degrades gracefully down to about 1200 baud)
• Single chip (not required, but it'd be best if that were possible)

I'd like to interface the modem with an ARM through a serial port, although it'd be fine if the modem just had an 8/16 bit I/O that can be used with older microprocessors. Wouldn't be too bad if I designed this around one of those CPUs, anyways..

I've done some looking on the internet, and really the only modem I could find was the CMX869B. I found no modems so far that have the V.92 performance I would like.

Thanks if anyone has any responses.

Answer 1

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.

Answer 2

Here's a V92 modem module we use in one of our products. It takes RS232 style inputs (at 3.3V). While not a single chip, it is a single module. My only complaint is that it's not surface-mount.

Answer 3

In addition to what @goblinbox commented, another possibility to consider, would be to build a soft-modem yourself. A modern ARM processor should have enough chops to handle most of the signal processing stuff in software.

According to the ARM guide, the only extra hardware needed is a 14-16bit linear CODEC at 8kHz and the PSTN interface circuit. The guide is only for V.22 but it's a start.

There seems to be other sources for soft-modems as well.