• Please review our updated Terms and Rules here

Mini-Omnibus backplane for debug and minimal system

gwiley

Experienced Member
Joined
Nov 12, 2021
Messages
249
Location
San Diego, CA, USA
6 months ago I built a small Omnibus backplane. For a long time I didn’t have a PDP-8/E/F/M chassis with backplane and power supply and wanted to get a system running using my Omnibus board collection.

I’ve had a number of requests to release the design files of this mini backplane to use it for debugging and to build a system without a real chassis+backplane+power supply, so I was thinking to make the design files available on my github. There were some issues with the first mini backplane version which have been fixed recently but I’ve not yet built the updated version (v2). Before calling the v2 design “done” it would be good to get feedback from experts here to see if there’s anything that should be added or modified.

So, the following is a short description so people can comment. I’ll post a more detailed description with design files and component sourcing on github.

The schematic is in the attached pdf. Circuit-wise, there’s not much here, just connectors for nine Omnibus slots, a Power OK circuit (copy from the H740 power supply), and power input terminals.

The board design is shown in the following two images (top & bottom). The layout considerations were primarily to minimize voltage drop in the +5V net. The -15V was also fairly robust but I realized later that not much core memory would be installed, so priority was given to have a solid +5V rail.
Mini Omnibus Backplane layout top v2 r01a.jpg Mini Omnibus Backplane layout bottom v2 r01a.jpg
The first version was designed for 2 oz copper. But to reduce the cost, v2 can be fabricated using 1 oz copper and two AWG 14 jumpers are included to distribute +5V across the long dimension of the board. There’s a voltage drop analysis that I’ll upload to github, or if anyone is curious I can share it here too.

The connectors with 0.125" (3.175 mm) contact spacing available from AliExpress aren't the nice deep slot connectors as on a real DEC backplane. They're more shallow and allow a wider "tounge" (the part with the gold fingers). I made a pair of acrylic strips with a kerf every ½ inch that holds the boards a little more steady. To make the plug-in boards align well it's necessary to use either dual 40 or dual 43 pin connectors, cut of both sides, remove some pins and install shims between the A and B, and between C and D. I learned this trick from the Douglas Electronics Omnibus module extender. The shim is a very simple piece designed using Microsoft 3D builder. I had some shims 3D printed and they fit nicely. I can share the STL file. There are two sources for for the same connector (the 2x40 or 2x43, again, some modifications are required):
 

Attachments

  • Mini Omnibus Backplane schematic v2 r01.pdf
    1.7 MB · Views: 24
The reason for 9 slots is the following minimal system:
1. Front panel, KC8E or KC8F
2. M8330 Timing Board
3. M8310 Major Registers Control
4. M8300 Major Registers
5. M837 Extended Memory Control
6. M8650 Asynchronous Data Control
7, M8357 RX01/RX02 interface or 2nd M8650 for serial disk
8. 32KW memory board
9. M8320 Bus Loads
 
Very promising! Thanks for sharing your design; I'd like to build one of these for benchtop purposes. Are you using KiCAD? Where's your GitHub located?

IMO the 1oz with supplemental heavy wiring ought to work OK. I'm interested in seeing your voltage drop analysis.

Keep up the good work :-}.

paul
 
Nice work! Would be fun to make a DW08(B) on a single Omnibus card for that backplane to expand a Posibus (Negibus) machine. So many possibilities!
 
This mini-Omnibus backplane has amazing potential for many use cases.

A possible small improvement for ultimate flexibility would be to add another slot to make it a 10 slot backplane.
This would allow one to trouble-shoot a 3 board core memory set on the work-bench outside of the full system.

As an alternative to the shim used by Douglas Electronics it should be possible to 3D print an overlay to increase the height of the backplane for more stability and correct board alignment.

Thank you for creating this backplane board!

In the past two years I have seen only one real Omnibus backplane on Ebay and the seller wouldn't ship to Australia.

Tom
 
Nice work!

My immediate thought was that 9 slots is not quite enough but your card list and minimal machine reasoning makes sense. How much additional cost would there be to adding 4 more slots? Would that exceed the 5 volt design constraints or does it make the card size exceed some physical dimension for reasonable cost? The slot contents I am thinking of are:
  1. KC8E front panel.
  2. M8330 System Timing
  3. M8340 EAE (connects to M8330 and M8341)
  4. M8341 EAE (connects to M8340 and M8310)
  5. M8310 Major Register Control.
  6. M8300 Major Registers
  7. M837 Extended Memory and Time Share Control
  8. M8650 or M8655 Console serial port (and Serial Disk).
  9. Peripheral controllers
  10. Peripheral controllers
  11. Peripheral controllers
  12. M847 combined with a 32k memory board
  13. M8320 Bus Loads
I am thinking of the RK05 three board controller set or a three board 4k or 8k core set in 9-11.

Another way to do this is to do what DEC did and allow some method for chaining backplanes together. Minimally this could be as simple as 144 jumpers between boards. This would not work in a production environment but for hobbyist purposes it would be fine. This would steal two of the 18 slots giving a generous 16 slot system. This is only generous if you replace six slots used by core memory with one slot of 82k.

Thanks for posting.
 
254434938_6641475425870184_8214646048616124487_n.jpg


I think Paul Lennous has a version that uses the ECS-type of DEC connectors. You'll probably find the ECS 2418-13-30 connectors are easier to work with and a better match for DEC modules than the Chinese ones.
There's also some other items in my Available Omnibus options.. post that may be of interest if you're working on 8A/8E systems.
 
Very promising! Thanks for sharing your design; I'd like to build one of these for benchtop purposes. Are you using KiCAD? Where's your GitHub located?
Thanks! I'm using EasyEDA which runs in a browser. Not my favorite CAD package but there are some nice features such as huge component library linked to a component supplier in China. I can export a BOM and have them do SMT (but no SMT on this board).
My GitHub doesn't have the files uploaded yet, but I plan to do that soon. I'll post a link then.
IMO the 1oz with supplemental heavy wiring ought to work OK. I'm interested in seeing your voltage drop analysis.
Two images are attached that show the analysis. I started using Excel to compute the drop but the mesh was too complex to get an accurate result, so Excel computes the resistance of all important trace segments and those segments are in a mesh simulated in LTSpice. The load is difficult to model accurately because it depends on how the Omnibus boards are wired to the +5V pins. I assumed a total load of 12 amps is distributed equally over 27 contacts (9 slots and three +5V pins on each slot). 5V / 12A = 0.4167 ohms, so 27 * 0.4167 ohms = 11.25 ohms on each pin.

My present system with the 9 boards mentioned in the second comment draws about 8 amps from the +5V.
backplane v2 voltage drop Excel.jpg backplane v2 voltage drop LTSpice.jpg
 
This mini-Omnibus backplane has amazing potential for many use cases.
thanks!
A possible small improvement for ultimate flexibility would be to add another slot to make it a 10 slot backplane.
This is tempting. I was concerned about keeping the cost low, and it's not so fun to solder so many connectors. Are you thinking something like this?
1. Front panel, KC8E or KC8F
2. M8330 Timing Board,
3. M8310 Major Registers Control
4. M8300 Major Registers
5. M837 Extended Memory Control
6. M8650 Asynchronous Data Control
7, core memory XY
8. Core memory stack
9. core memory sense inhibit
10. M8320 Bus Loads
This would allow one to trouble-shoot a 3 board core memory set on the work-bench outside of the full system.
If there were an FPGA Omnibus board that implemented M8330+M8310+M8300 and some of M837, that might free up enough slots to test various things?
As an alternative to the shim used by Douglas Electronics it should be possible to 3D print an overlay to increase the height of the backplane for more stability and correct board alignment.
Or maybe keep the shim and in addition print a sort of card guide for the right side that bolts onto mounting holes H5 and H10. For the KC8E programmer's console the guide needs to be offset a little more than the other slots. My acrylic version has this.

acrylic kerf card guide.jpg

but instead of just floating this could extend lower toward the backplane and bolt onto two of the backplane mounting holes.

Regarding the shim, I was thinking about a better version of the shim that sticks up into the keying slot that's on Omnibus boards between AB, BC and CD, to prevent boards from being inserted backwards.
 
My immediate thought was that 9 slots is not quite enough but your card list and minimal machine reasoning makes sense.
I agree, 9 slots is kind of on the threshold of being barely enough, but cost and complexity were key factors.
How much additional cost would there be to adding 4 more slots? Would that exceed the 5 volt design constraints or does it make the card size exceed some physical dimension for reasonable cost?
The board is a bit larger, so it will take up more space on the desk or bench top for those who don't need the extra slots. Cost of more board area plus extra connectors is a factor. Also assembly is significant unless you have access to wave soldering equipment.
Another way to do this is to do what DEC did and allow some method for chaining backplanes together. Minimally this could be as simple as 144 jumpers between boards. This would not work in a production environment but for hobbyist purposes it would be fine. This would steal two of the 18 slots giving a generous 16 slot system. This is only generous if you replace six slots used by core memory with one slot of 82k.
So, jumpering two backplanes is definitely a do-able thing. I did this with the first Mini Omnibus prototype because it had only 8 slots which was insufficient to add a floppy controller.
Here's a simple solution, it's a homebrew version of an M935 Omnibus bridge.
Just need 8 of these:
Omnibus Extender.jpg
and four 34-pin flat cables to connect two backplanes:
backplane jumper in backplane cropped.jpg
 
I think Paul Lennous has a version that uses the ECS-type of DEC connectors. You'll probably find the ECS 2418-13-30 connectors are easier to work with and a better match for DEC modules than the Chinese ones.
I searched some but wasn't able to find a distributor for the ECS connectors. Can you recommend a source? I'm definitely interested.
The connectors from China take forever to ship but they sure are inexpensive. Cost adds up fast when stuffing a backplane with so many connectors. I think there are solutions to solve the horizontal alignment and wobble issues but would like to investigate the ECS connectors. The ECS and Chinese versions look like the footprints are similar, but I haven't checked all of the dimensions yet.

There's also some other items in my Available Omnibus options.. post that may be of interest if you're working on 8A/8E systems.
Thanks for the link. I've actually built or bought some of these. Currently have:
Rolands 32KW memory board
M8357R (RX8E)
RX/01/RX02 emulator (2 -layer version)
Douglas Electronics board extender/riser
Had printed some of the PDP-8 switch handles o
 
This is tempting. I was concerned about keeping the cost low, and it's not so fun to solder so many connectors. Are you thinking something like this?
1. Front panel, KC8E or KC8F
2. M8330 Timing Board,
3. M8310 Major Registers Control
4. M8300 Major Registers
5. M837 Extended Memory Control
6. M8650 Asynchronous Data Control
7, core memory XY
8. Core memory stack
9. core memory sense inhibit
10. M8320 Bus Loads

Yes - that is exactly what I had in mind to use as a trouble-shoot platform with enough slots for the 3 board DEC core memory set.
Any trouble shooting on my LAB-8/e means working hunched over the rack mounted system with poor access.
Your backplane board would allow me to trouble shoot in a much more comfortable and accessible environment on top of my work bench.

Thanks
Tom
 
Another way to do this is to do what DEC did and allow some method for chaining backplanes together. Minimally this could be as simple as 144 jumpers between boards. This would not work in a production environment but for hobbyist purposes it would be fine. This would steal two of the 18 slots giving a generous 16 slot system. This is only generous if you replace six slots used by core memory with one slot of 82k.
I lean against growing this board into "be-all-and-end-all" and instead favor supporting extensibility so that two backplanes could be chained together, preferably without sacrificing slots and using up expensive backplane connectors (as DEC does with the M935). That said, one could compromise by adding IDC connectors to just the _rear_ of the backplane so that the existing 9 slots are preserved for module-use. Those IDC connectors could then either serve as built-in pick-up points for connecting a logic analyzer or use with a modified homebrew version of the M935 Omnibus bridge to extend to a second board to add 8 more slots (and the logic analyzer could then be connected at the rear of the second backplane). Agree with gwiley about the various rationales for keeping the backplane at a reasonable size. To me the bench-top-appropriate sizing is an important consideration. Adding one more slot to total 10 would be OK (and then be prepared to sacrifice it for bridging to a second backplane, if desired).

Anyway, just a perspective; I'll work with whatever Greg comes up with. I already have a "monster" 20 slot 8/A backplane that I could put up on a bench, but of course it has the full paraphenalia on the back for connecting two PS and I'd still need to do something about those -- which double the overall thickness; and of course it's 6-slots wide overall. With those massive DEC SU blocks it's a real doorstop, and then some. I'd prefer something more light-weight and good-enough for standard OMNIBUS-based servicing use on a bench. 10 slots would be OK; IDC connectors (regardless of slot-count) would be nice, but certainly isn't a showstopper, as gwiley illustrates with his M935 surrogate.
 
Two images are attached that show the analysis. I started using Excel to compute the drop but the mesh was too complex to get an accurate result, so Excel computes the resistance of all important trace segments and those segments are in a mesh simulated in LTSpice. The load is difficult to model accurately because it depends on how the Omnibus boards are wired to the +5V pins. I assumed a total load of 12 amps is distributed equally over 27 contacts (9 slots and three +5V pins on each slot). 5V / 12A = 0.4167 ohms, so 27 * 0.4167 ohms = 11.25 ohms on each pin.
Thank you, but apparently I'm not the smartest crayon in this box. Please elaborate a bit, particularly in regards to:

1. Where are points H2R, H3R, H4R, and H5R? H3L, H4L, and H5L?

2. It seems to me that the model needs to also add (at least) 2x 14AWG on the ground return plane when considering the effect of adding the same on the +5v distribution plane. There's not much point in fortifying just half of the +5v power circuit. Considering that the ground return plane is shared with two other power distributions IMO it's the one most subject to "droop" -- which would be in practice "raise" -- at the edge-connectors.

3. Why not treat (and install) the added wires as a distributed drop in resistance, rather than a single point-attachment? I would strip some 14 or 12 AWG solid wire, form it to the proper "C" shape, and then solder it directly to the copper-plane, ideally continuously but certainly at multiple locations -- the idea being to closely approximate the effect of (distributed) 2oz copper. I've seen this done in various fashions in various other designs, including just being liberal in applying solder to traces to build up some thickness (not as good as pure copper, but it helps). At least I'd add wires dedicated to each of the third and fourth-ranked slot-sets rather than a single (set) added somewhere between them. Pragmatically I'd focus on building up the "crossbars" that are the terminal distribution legs to the connectors through fully-soldered pieces of heavy copper wire. Particularly the one down-the-middle. Eyeballing (spitballing?) the copper distribution that's where the pinch-point looks to fall given that we're assuming an equally-distributed +5v load across the three connectors of a given module.

Thanks for the further enlightenment :-}.
 
2. It seems to me that the model needs to also add (at least) 2x 14AWG on the ground return plane when considering the effect of adding the same on the +5v distribution plane. There's not much point in fortifying just half of the +5v power circuit. Considering that the ground return plane is shared with two other power distributions IMO it's the one most subject to "droop" -- which would be in practice "raise" -- at the edge-connectors.
Addendum: While I can imagine that your intent here was to model a lumped solution and that in practice you really meant one-wire-on-top-and-another-on-bottom given all of the other specificity in the calculations (! :->) this is an area in which the model is IMO not realistic WRT the calculation of actual voltages (or maybe voltage-differentials) at module-pins ... as opposed to some lumped resistance as seen by the power supply.

But as I noted earlier, I may very well not understand your intent(s) and result(s) here!
 
I have coded the M837 substitute and had planned to add it as an option in a future 32K card. (The CPLD is also set up so that it can also be reprogrammed to be a serial line instead.)

Vince
Could your design be adopted & implemented by others in their own prototyping/development environment? Agree that adding it to an improved 32K design is great, but in my case I have sufficient memory but lack any M837 at all :-{. So I'd be comfortable implementing it on a multi-purpose module where I might also implement a separate CPLD-based serial line, or whatever.
 
Could your design be adopted & implemented by others in their own prototyping/development environment? Agree that adding it to an improved 32K design is great, but in my case I have sufficient memory but lack any M837 at all :-{. So I'd be comfortable implementing it on a multi-purpose module where I might also implement a separate CPLD-based serial line, or whatever.
Sure, in theory. In practice, I suspect the easiest thing is to order the memory board kit (assuming I ever get them into production), then just not populate the boot loader and/or memory sections.
Or order two and put the serial device CPLD and connectors in the other one :).

Heck, we might even work a deal where you can help me debug the prototypes. At the moment, the technical hitch is that I want to verify correct bootloader operation in the 8/A before I commit to ordering prototype boards with the CPLD site on them. Otherwise, I potentially have to order an extra board spin. (It's not that expensive, but it can be confusing to have lots of versions lying around. There are also supply chain issues that prevent kit production anyway, and so little reason to rush.)

Vince
 
  • Like
Reactions: smp
Back
Top