ST536 STE EDITION

[dml]

It involves improving the 5v / ground by running wires on the rear of the motherboard, adding some polymer caps around the glue chip, and adding a capacitor on the blitter to stop blitter related crashes.

I was going to mention this (decoupling problems - not the pdf, which I hadn't seen) because some Atari boards seem to have decoupling caps in strange places, not always near the chips or distributed properly around the ground pins on each side of the bigger chips.

Also... MLCC ceramic capacitors (which were not used then but are now) are really, really bad at decoupling high frequency stuff - too high impedance. Normal ceramic caps have better frequency characteristics. There are other types which are even better (film types used for RF stuff) but bulky to put near ICs.

I was debugging a <1GHz homebrew radio hat for someone recently and the radio was unstable because the decoupling caps were not keeping up. 8MHz is much less of a problem but anything MHz upwards is still randomly sketchy without good decoupling near the ICs and anywhere track resistance has been adding up along the way.

[coonsgm]

Well the new STE just arrived yesterday. Listed as untested for parts, I was able to pick it up for a steal....on the ebay no less as the only bidder on it. I guess not too many folks were interested in a project if it was a project.

As you can see from the photos, once "tested" it works fine....even the floppy drive worked right away and would boot from the Ultrasatan too.

IMG_2950.jpg (377.71 KiB) Viewed 372 times

IMG_2949.jpg (236.79 KiB) Viewed 372 times

On the plus side, I have one more STE to test the STE536 with.

It is a native US version and also uses the external blitter.

[exxos]

Good stuff @coonsgm. Currently the external blitter boards are sulky it seems. So worth making note which ones you test the 536 on

[exxos]

I was debugging a <1GHz homebrew radio hat for someone recently and the radio was unstable because the decoupling caps were not keeping up. 8MHz is much less of a problem but anything MHz upwards is still randomly sketchy without good decoupling near the ICs and anywhere track resistance has been adding up along the way.

I was reading some applications notes last week. The general rule seems to be 10-100nf for up to 100mhz. But 0402 to 0603 sized caps. Though I started using the COG type as they better for high frequency. But they only go to about 10nf.

Mostly its the track length to the caps where the main "delay" is. But not simple to physically place. Its one reason I went for SMT FPU socket on the DFB1X as slightly smaller lead lengths.

Mix in larger sized caps are slowler due to physical dimensions. Then going smaller helps, but to small and you don't get the value capacitance you think you do. Dave Jones did a interesting video on it all a while back.

I'm using 100nf on every power pin. Then a 47uf under every chip. On the next rev I've added more 47uf caps and a high speed regulator.

Problem is with boosters, the ground length ends up rather long. Even the smaller V1/STE boosters is enough to break even just with a passthrough adapter. You end up with DC offsets and a bit of noise mixed in never ends well.

I actually have a GND strip on the right side on the STE536 as its close to the motherboard outer ground. It makes easy connection to beaf up the gnd. But that alone may not be enough. It's why I added a power header so thicker cables can power the board direct from the PSU. Less current though the STE power traces can only be a good thing.

[exxos]

I've mentioned this in passing before, but the guys who created the TwiSTer booster have a whole section describing improvements to be made on an STe motherboard to make the booster stable. It would not surprise me if these modifications were beneficial for the STe536. It involves improving the 5v / ground by running wires on the rear of the motherboard, adding some polymer caps around the glue chip, and adding a capacitor on the blitter to stop blitter related crashes.

I wonder if they put a cap on BR to snub the noise? But I did try 100pF and didn't help. It's why I went with a SR type latch instead.

Add more wires to beaf up the power rails makes sense to a point. But with RF frequencies you need surface area. The impedence of wires is likely higher than the flat copper traces. May help for raw DC power, but RF its probably not helping all that much. Would need a lot of analysis to figure it all out.

[exxos]

Did some "quick'n'dirty" GND wire tests.

IMG_3320.JPG (269.38 KiB) Viewed 337 times

IMG_3322.JPG (278.24 KiB) Viewed 337 times

GND from booster to outer MB GND.

IMG_3323.JPG (271.84 KiB) Viewed 337 times

IMG_3324.JPG (269.52 KiB) Viewed 337 times

Wire from bulk cap to booster.

IMG_3326.JPG (345.32 KiB) Viewed 337 times

IMG_3325.JPG (252.24 KiB) Viewed 337 times

Another wire from bulk cap to booster.

IMG_3328.JPG (314.56 KiB) Viewed 337 times

IMG_3327.JPG (279.96 KiB) Viewed 337 times

Another wire from bulk cap to booster.

IMG_3330.JPG (348.63 KiB) Viewed 337 times

IMG_3329.JPG (263.87 KiB) Viewed 337 times

That's what I mean with GND wires. Needs 3 thick multi strand cables to even make a dent. Really needs some good Litz wire. I do have some, but I don't see any need to continue. It didn't help with the resetting issues.

I've done some more tweaks to the busgrant logic and its been sat at the desktop while I been doing this post. I just hope it's not a case of fix it on the external blitter and break it on the internal blitter type of "fix".. Will have to try it out next..

Currently its still running with the delayed clock.. So need to look into that as well ...

[ijor]

Btw, and hope this is not too off topic. How many layers have the ST/STE motherboards? And it is known which kind of stack-up they use?

[exxos]

Btw, and hope this is not too off topic. How many layers have the ST/STE motherboards? And it is known which kind of stack-up they use?

@sporniket May be able to answer that.

[sporniket]

The STe is mostly 2 layers, and there is an inner layer in the area of the digital audio circuitry. On the original ST, it's a 2 layer board. I guess that's the same for STf.

[exxos]

8MHz benchmarks. @dml it must be those 32bit reads slowing it all down.

IMG_3332.JPG (103.99 KiB) Viewed 321 times

I'm finding all sorts of bugs now... Things a lot easier at 8MHz