Philips NMS 8255/8250 memory upgrade (again!)

I finally got the mapper working, after being busy with other duties

Long story short: Original instructions way of wiring up the 139 for RAM selection makes no sense. A working example below.

At first I built it the way I described - almost identical to the instructions, but with STROBE (CAS2/E) and 670's Q2 swapped. MSX booted, but MSXDOS2 was not usable (would not boot / be recognized), and memory test hang before it displayed the mapper size. But Maarten Veerheijdt's testmap 4.2 came in handy, and there was address errors reported above 0x8000 (32kb). 512kb was found with a basic testing program, however. So, I concluded that maybe just the first 32kb is usable, altough the mapper itself is working.

Also, as a sidenote: if one owns a memory mapper and inserts it in a lower slot - for example, I have PlaySoniq - the mapper on the lowest slot will be used. This came in handy while testing the built-in mapper, since I could always have a working mapper / memory, and test the faulty one!

So I did some more testing - there's really three alternatives that can fail that I can think of: the mapper circuit (which seemed to be working), the RAM chips, or the RAM chip selection (which can not be seen by mapper test programs). So I went back to the drawing board, and looked at the circuit once more with thought. I was at first assuming that there is only some small error in the instructions, and working on that assumption. But I started to suspect otherwise and started to think how the mapper should really work (with help from RetroTechis post), and I concluded that only one CAS should be active from the 139 (I'm calling the Y outputs CASX from now on). I can not see how the original instructions could achieve that, even with the modification I had already made.

If using 139, the first thing is that one (and only one!) of the 139's G (enable) inputs should be active if CAS2/E is active (low). The original mapper on the MB achives this by splitting CAS2/E into CAS1 and 2. A lot of other mapper circuits use only the original 670 (it has one unused register bit, which can be used to expand up to 256kb), and use CAS2/E as "STROBE" and generate the selections signals by other means - or, add their own register bits and bypass the original 670 altogether in this regard. Seeems like someone just draw up the instructions based on some other instructions (perhaps 256kb mapper instructions), with selection bits from the new 670 and the old 670's bit randomly connected into the 139 (note: I didn't actually built it that way, but I still can not see how it could work).

So I modified the circuit, so that I take original CAS1 and CAS2 signals already split on the MB to G1 and G2 (the order shouldn't matter) and the bits added by the new 670 to both A selectors and both B selectors. This way the mapper works (at least according to maptest). Now I need to find a way to stress test the memory to actually test if the chips are good or not

Let this post be as a reminder and a lesson here for everyone else considering doing something simiilar - or different - modification. There really can be faulty instructions out there, so understanding them is quite important.

Cheers!

Hi RetroTechie,

I was also wondering why not usea a 138 instead! But I already had the 139 othwerwise wired up. I actually have already bought a spare 138, in case I need it (for this project or later). I believe you'd need to pull up G2A&B (with a resistor to Vcc?) and connect CAS2/E to G1, is that correct?

I also believe I have the D lines correctly connected, and there is only errror in the picture introduced by me . At least testmap didn't report any errors. I will fix the picture later, and check my connections (though I'm not sure if flickr will allow swapping pictures, so this thread might look weird if someone is reading it later on). In any case, it's good that someone is proofreading this!

As for why using 4464 - well I already bought them before I knew what I was doing . I'd not do it in hindsight, and would not recommend using them to anyone. But if someone reading this still wants to use 4464's: make sure your source is good. I got used ones - with slightly bent pins (and one was clearly faulty - luckily I ordered some extra). As a reasult of the bent pins, the 4-storey houses of RAM look like they've been in an earthquake, but left standing. That may introduce some short circuits, since there is only 2mm (or so) space between the RAM chips on the PCB. If one chip is 0,3mm off, and the block of RAM is off in the opposite direction... there is suddenly 1,5mm clearance... hope the next floor is off in the other direction!

You can try to straighten the pins for as long as you like before soldering, if you like this kind of tinkering - but you probably get the picture.

I believe I've done just this (I did test in BASIC, too, expecially with the broken 139 connections - and it worked OK). The error is just in the picture. I draw it late the other night and made an error =).

I'm not sure how I've been looking at the datasheet : . Of course it is like you said.

One thing that is still unclear to me: when to use pull-up / -down resistors - and is it needed here? This is probably a very novice question, which Google will tell me, soon...

Thanks!

P.s. the picture has been replaced and can be found here. Hopefully flickr will delete / swap the original picture in the static URL reference in the previous post some day (I can't edit the post anymore).

Whoops... fixed

Chances are it's the exact same file as this.