#multi-user basic

A few months ago I built an I/O expansion board for my homebrew 68030 project with a 4-port serial card to go with it, and got BASIC running for four simultaneous users. It worked, but not as well as I had hoped. I wanted to be able to run two of those serial cards to support 8 total users, but it had proven unstable enough that with just the one card I had to slow down the whole system to 8MHz.

So I designed a new serial card.

I had previously been running this computer without any issues at 32MHz with a mezzanine card with FPU & IDE as well as a video card. The main board by itself can clear 56MHz. Having to go all the way down to 8MHz just didn't sit well with me. I want this machine to run as fast as possible for its 8 users.

I put extra time into reviewing worst-case timing for all components and graphing out how signals would propagate. The 16C554 quad UARTs I'm designing around are modern parts that can handle pretty fast bus speeds themselves — easily up to 50MHz with no wait states on the 68030 bus — assuming all the glue logic can get out of the way fast enough.

Signal propagation delays add up quickly.

My first draft schematic used discrete 74-series logic for chip selection, signal decoding, timing, etc. At slower bus speeds this wouldn't have been a problem. But I want this thing to run as fast as possible. By the time critical signals had made it through all those logic gates, I was looking at already being well into one wait state by the time the UART would see a 50MHz bus cycle begin.

I needed something faster. I was also running low on space on the board for all the components I needed. The obvious answer was programmable logic. I settled on the ATF22V10 as a good compromise of speed, size, availability, and programmability. It's available in DIP with gate delays down to 7ns. Where discrete gates were necessary, I selected the fastest parts I could. The final design I came up with showed a worst case timing that would only need one wait state at 50MHz and none for anything slower.

It ended up being a tight fit, but I was able to make it work on a 4-layer board within the same footprint of my main board, putting some components on the back side. (It may look like a bunch of empty space, but there's actually a lot going on running full RS232 with handshaking for 8 ports).

New problem. I had blown my budget for the project. As much as I love those stacked DE9 connectors, they're expensive. And there's no getting around the $10 pricetag for each of those quad UARTs. Even using parts on-hand where possible, I was looking at a hefty Mouser order.

[jbevren] suggested using ganged RJ45 connectors with the Cisco pinout instead of stacked DE9, to save space & cut costs. [Chartreuse] suggested buffering the TTL serial TX/RX signals to drive the LEDs that are frequently included on PCB-mount RJ45 connectors. Both great ideas. I was able to cut 20% off my parts order and add some nice diagnostic lights to the design.

Two weeks later, I received five new PCBs straight from China. I of course wasted no time setting into starting to assemble one.

I really set myself up for a challenge on this one. I learned to solder some 25 years ago and have done countless projects in that time. But I think this might be the most compact, most heavily populated, most surface mount board I've ever assembled myself. (There are 56 size 0805 (that's 2x1.2mm) capacitors alone!)

After a few hours soldering, I had enough assembled to test the first serial port. If the first port worked then the other three on that chip should work too, and there's a great chance the other chip would work as well.

And it did work! After some poking around with the oscilloscope to make sure nothing was amiss, I started up the computer and it ran just fine at 8MHz.

And at 16MHz.

And at 25MHz.

And at 32MHz.

And at 40MHz.

And almost at 50MHz!

Remember what I said about my timing graphs showing one wait state for 50MHz? The computer actually booted up and ran just fine at 50MHz. The problem was when I tried typing in a BASIC program certain letters were getting switched around, and try as I might, BASIC just refused to 'RQN' my program. It was pretty consistently losing bit 3, likely from that signal having to travel just a tiny bit farther than the others. A problem that will probably be resolved with an extra wait state.

Good enough for a first test! A few hours more and I finished assembling the card.

I did have some problems with cleaning up flux off the board, and I had to touch up a few weak solder joints, but so far everything seems to be working. I've updated my little multi-user kernel to run all 8 users from this new card and it's running stable at 40MHz.

I need to update my logic on the 22V10 to fix a bug in the wait state generator. I would love to see this thing actually running at 50MHz — a 25% overclock for the 40MHz CPU I am currently running. I also want to expand my little kernel program to add some new features like the ability to configure the console serial ports and maybe even load programs from disk.