Who else is building a Beeb from scratch?

[!FOZ!]

Small modification to tidy up the wiring required for the BooBip.com Flash and SRAM modules. I fitted pin headers under the board for the fly leads that would normally require connections on the top of the board with grippers (shown in last pic). It still also means the modules are easily removed as needed and the remaining pin headers are fairly unobtrusive. Quite pleased with the result both top and bottom - and I do plan to have some standoffs when I fit into the eventual case. That said, does anyone see any problems here?

[BeebMaster]

That's quite a nice thing to do. On one Beeb I saw recently, somebody had soldered an 8-pin DIP socket on top of IC76, presumably to facilitate easier connection to pins 11 & 12 for ROM addressing.

[1024MAK]

If you are going to solder, it's better IMHO to do it on the underside of the board.

Mark

[!FOZ!]

Thanks both, glad that mod gets a thumbs up! I thought about a few approaches but soldering another socket onto a chip was not one of them…

Meanwhile my UHF output is working having fitted C45 (10nf)!

[!FOZ!]

How is everyone doing with your builds?

[thedark]

Not that good. Someone take a look at it. Got a beeeeb of death.... so hope that its a small issue.

[!FOZ!]

Not that good. Someone take a look at it. Got a beeeeb of death.... so hope that its a small issue.

Sorry to hear you are still having issues.

Try this minimal configuration (you will not need a keyboard) and see if it boots after removing these ICs

viewtopic.php?p=220996&sid=1871ffd5d1e6 ... be#p220996

[thedark]

Roland has checkt it
Its seeing everyting dubble. He got taken it at home to see if it can be fixed. I do not have the equipement for it.

[!FOZ!]

Roland has checkt it
Its seeing everyting dubble. He got taken it at home to see if it can be fixed. I do not have the equipement for it.

I hope he can get it going for you.

[maniacminer]

I've got three dead boards and I've got three new Issue 7 remake boards for Xmas (lucky me) I'm wondering why the "A" capacitors are 33nF rather than 100nF given the frequencies used in the Beeb?

[eelco108]

How is everyone doing with your builds?

Work got in the way ... hopefully the winter break helps to get more soldering time in?
I have all the components, including a keyboard ... so just need time. A precious commodity, ism't it ...

[eelco108]

I've got three dead boards and I've got three new Issue 7 remake boards for Xmas (lucky me) I'm wondering why the "A" capacitors are 33nF rather than 100nF given the frequencies used in the Beeb?

Three dead boards? Be more gentle to your beebs!

The main reason (I think) these are not 100nF is that the poor power supply will charge them all when switched on, and because there are so many of them this would cause quite a spike in demand on the power supply when you flick the switch. Probably too much, so the designers picked the lowest that did the job (the 100nF is only a recommendation). Also, they are all in parallel, so they help each other out, if you like, to get rid of these nasty ripples that all these switching circuits are causing ...

[wiggy]

Probably cost was a driver for the smaller value BITD. The "A" designation is a clue that the value isn't actually that important.

Contemporary recommendations proved harder than I expect to track down, but Philips' HCT family User Guide from 1997 suggests:

A local voltage regulator on a printed circuit board can be decoupled using an electrolytic capacitor of 10 to 50uF. Localized decoupling of devices can be provided by 22nF per every two to five packages and a 1uF tantalum capacitor for every ten packages. The Vcc line of bus driver circuits and level-sensitive devices can be decoupled from instantaneous loads by a 22 nF ceramic capacitor connected as close to the package as possible."

My memory of other similar recommendations was that they tended to call-up one 22nF or 33nF for every couple of ICs. So this value is correct for the timescale.

[maniacminer]

Probably cost was a driver for the smaller value BITD. The "A" designation is a clue that the value isn't actually that important.

Contemporary recommendations proved harder than I expect to track down, but Philips' HCT family User Guide from 1997 suggests:

A local voltage regulator on a printed circuit board can be decoupled using an electrolytic capacitor of 10 to 50uF. Localized decoupling of devices can be provided by 22nF per every two to five packages and a 1uF tantalum capacitor for every ten packages. The Vcc line of bus driver circuits and level-sensitive devices can be decoupled from instantaneous loads by a 22 nF ceramic capacitor connected as close to the package as possible."

My memory of other similar recommendations was that they tended to call-up one 22nF or 33nF for every couple of ICs. So this value is correct for the timescale.

Ah, ok, so 100nF should be fine, the LS TTL uses 5x the power of HC/HCT logic. It's just I've got an Acorn style problem, a big sack of 100nF 5mm pitch polystyrene capacitor so it'll save me a bit not having to buy 200+ 22/33/47nF (depending where I look)

Just a mathematical thought 70x 100nF is 7μF - I think that'll be fine given the output capacitors on the 5V rail are 2,200μF I don't think the PSU will be too bothered.

[1024MAK]

I don't think the amount of capacitance is a problem for the PSU. The PSU power limits anyway. All that would happen, is that the supply voltage would rise at a very, very slightly slower speed. Not a problem as the reset pulse is more than long enough for the 1980s technology chips.

The important thing with decoupling capacitors is more about using a type that is not too inductive and providing them close to each IC. Now, it's considered the default value is 100nF, but that was not the norm back in the early 1980s.

In practice, even with 22nF capacitors, it's likely that the system would run without a substantial number (say ¼) missing. However, to try to prevent edge cases caused by signal level problems (due to a chip that is trying to drive a logic line having a temporarily reduced supply voltage, hence risking the drive level being outside the defined TTL levels), its always better to go on the cautious side.

Hence in practice, any ceramic (or better type) capacitor of any value used for power rail decoupling can be replaced with a higher capacity capacitor of the same or better type, within reason. So absolutely no problem using 100nF for all the capacitors marked as 'A' on the Beeb board.

Somewhere on this board, there is a discussion on decoupling.

Mark

[wiggy]

In practice, even with 22nF capacitors, it's likely that the system would run without a substantial number (say ¼) missing.
...
So absolutely no problem using 100nF for all the capacitors marked as 'A' on the Beeb board.

Absolutely!

Although that said, I wouldn't like to try my luck reducing the number that much around the DRAM - DRAM is mildly famous for large current spikes (due to the way it has to refresh a whole row at a time on every read) and there are stories around of other systems that could get enough current through the ground track resistance/inductance to lift the whole memory up in voltage to the point that the data output levels were corrupted...

The Beeb notably gives the DRAM area it's own connection from the PSU, so it's clear Acorn had at least heard of this.

[maniacminer]

The Beeb notably gives the DRAM area it's own connection from the PSU, so it's clear Acorn had at least heard of this.

The Issue 1-3 boards have isolated power zones, issue 4+ has a kind of ring main linking the three Vcc zones together, I'm sure thickening up of the tracks also helped where there wasn't sufficient voltage hysteresis to allow the DRAM, buffers and clock circuitry to work properly, especially as Acorn treated some of the digital signals in the analogue domain, thus are affected by noise and ground bounce.

[1024MAK]

The thinking for the three separate +5/VCC feeds, is the BBC wanting a conventional PSU rather than a SMPSU. One of the designs for these conventional PSUs uses three 7805 voltage regulators. As it is unwise to parallel these, the circuitry had to be divided into three sections...

Similar to the Atom having two 7805 voltage regulators and having its circuitry divided into two sections...

By the time of the issue 4 boards, the ASTEC SMPSU had become standard. Hence now it was okay to parallel up the mainboard +5/VCC rails, as they all came from the same point in the SMPSU anyway.

Mark

[maniacminer]

Ah, yes, the original PSU had three 7805 regulators (I have given several of them away, they ran very hot) the parallelization of 7805s causes oscillation and random shutdowns.

Version 2 of that linear PSU used a power transistor and that design had poor voltage regulation, but ran cooler.

I noticed a 0.5V voltage drop using the PiTube and RGB2HDMI and ended up soldering some extra wire between the power lugs on the underside of the board to the RGB connector and the Tube connector.

[maniacminer]

I just started one of the three. I screwed up already and bought a box of the wrong capacitors for "A", I misread the line about 2.54mm and bought 5mm instead, d'oh! It's not a total loss, I can use them in other things. I ordered the correct part and that should arrive next week

[1024MAK]

Not building using 'order of height/size' then?

Almost always (except when modification, 'select on test' or experimentation is involved) I build up boards by fitting components in order of height, so typically in this order: SMD components, signal diodes, resistors, axial ceramic or similar capacitors, rectifier diodes, other low profile axial components, small capacitors that are lower in height than sockets, chips that will not be socketed, IC sockets, connectors of a similar height, board mounted 5mm LEDs, radial electrolytic capacitors, resistors and diodes that are mounted vertically, larger components or connectors etc.

Mark

[maniacminer]

Not on this build as I'm testing it as I go. Reset and clock first, then address bus, data bus, OS ROM read and so on. Given that I already have three dead boards that I've invested hours of probing, IC removal, socketing, etc. I'm not willing to get into another dead board scenario, I'd rather throw it up in the air and give it both barrels Funnily enough I do have access to a wave soldering bath (it hasn't been used in two decades) if I were to make tens of boards I'd be sorely tempted to get it running, although, if I were to make BBC boards commercially, I'd move to SMD and probably use the SCL design of a 32kx8 SRAM + buffers instead of the horrible 16x2k DRAMs... (wouldn't need to use 81LS95s either as I wouldn't be driving 16 ICs with hard, fast edges)

[!FOZ!]

Not on this build as I'm testing it as I go. Reset and clock first, then address bus, data bus, OS ROM read and so on. Given that I already have three dead boards that I've invested hours of probing, IC removal, socketing, etc. I'm not willing to get into another dead board scenario, I'd rather throw it up in the air and give it both barrels Funnily enough I do have access to a wave soldering bath (it hasn't been used in two decades) if I were to make tens of boards I'd be sorely tempted to get it running, although, if I were to make BBC boards commercially, I'd move to SMD and probably use the SCL design of a 32kx8 SRAM + buffers instead of the horrible 16x2k DRAMs... (wouldn't need to use 81LS95s either as I wouldn't be driving 16 ICs with hard, fast edges)

Good luck with the build! I did some incremental testing as well but mostly when it was built, minus most connectors, tested the frequencies then and using the minimal “Language ?” setup - after that added the rest. It would be great to capture your component list as you go through testing here perhaps.

Also I see your using standoffs - its a great idea - i had them on both sides it does eliminate some of the need to start small - but not all - some components would be very tricky to fit after fitting sockets (at least for me).

[maniacminer]

It lives! Well the reset and clock work. 44mA at 5V so no horrors yet

[maniacminer]

Stuffing A into two more boards

[!FOZ!]

It lives! Well the reset and clock work. 44mA at 5V so no horrors yet

Congrats! I love the look of these boards.

[maniacminer]

It lives! Well the reset and clock work. 44mA at 5V so no horrors yet

Congrats! I love the look of these boards.

Thanks. I'm trying a different approach with the next two by fitting resistors in value order. Then it hit me that there's no guide to where the resistors, capacitors etc goes. I'm spending more time looking for spots on the board than soldering and cropping... Perhaps I should make one. I've also found a couple of errors on the BoM with the wrong pitch part, I left a comment accordingly on the Google Sheets document. I also replaced the 10R resistors to the LM324 with 100uH inductors, a lot more expensive, but should be more effective at reducing noise.

EDIT: I have worked out where all the resistors and capacitors are and created an index for both part number and value. (Updated:01/05/2024)

[smokey]

we need some obolete chips reproducing

[!FOZ!]

we need some obolete chips reproducing

Which ones are you struggling with?

[maniacminer]

I've got no serial ULAs left, I was hoping for a CPLD replacement, I guess sacrificing the cassette function and replacing the ULA with a few wire links would allow the RS423 to work.