|
|||||||||||||||||||||||
 |
>Kenbak Repro |
|
|
|||||||||||||||||||||||||||
|
1 INTRODUCTION: 1.1 Why was the Kenbak-1 Reproduced? 
The Kenbak-1 was reproduced as a
tribute to the ingenious machine of John V. Blankenbaker. It
is hoped
that by spreading the knowledge of this device to a broader audience,
more people will understand just what an incredible machine it was.
While many web sites mention this computer, details are so
sparse about how it was used, and what it could do, that many people
fail to believe it can be considered a true "Personal Computer."
I have found that even collectors who own an original
Kenbak-1 may be hesitant to turn it on, as the switches were fragile
(held
on by brittle glue) and they definitely don't want to damage their
valuable relic. But with a lower-cost copy, I hope that many
people can
power it up, try to write programs, and get the true machine-language
"blinking lights" experience that was the 1960's and 1970's computer
revolution. I hope that even museums which have an original
Kenbak-1 could let visitors toggle in a simple program, and run it,
without worry of damage to the original, and a much wider knowledge of
the Kenbak-1 computer will be realized. This project was for
fun, not money, and I'll never recoop the money I spent on it.1.2 How was this Kenbak-1 Reproduced?  This is the result of over 300 hours of part-time work. While the schematics of the Kenbak are freely available from many sources, including the inventor, it was a real challenge to construct a working PC board. I got the idea of making one after viewing Erik Klein's Kenbak-1 computer at www.vintage-computer.com. His web site seemed to show a reasonably simple circuit board, so I figured it shouldn't be hard to make one. I then saw that schematics and logic diagrams were available from Herbert Eisengruber of The Computer Museum of Nova Scotia (www.computermuseum.20m.com) for a small fee, so I got a copy. While this would have been enough information to make a computer, I really wanted the circuit board to be similar to the original, with the same general location of components and traces, so I asked Erik if he had any higher-resolution pictures of his Kenbak-1. To my amazement, he sent me such high-resolution pictures, that I could read part numbers on diodes, color codes on resistors, and could follow traces almost perfectly. With out Erik's pictures, I doubt I could have gotten this project working. Laying out the PC board wasn't easy. From the photos I settled on a 16 x 9.25 inch circuit board, and found two free programs, TinyCAD (tinycad.sourceforge.net) and FreePCB (www.freepcb.com) which allowed me to go from schematics to printed circuit board in only about 100 hours of time. I tried to make the boards look a lot like the originals, keeping in the spirit of the original machines, but all traces were laid out free-hand. When I was happy with the board layout, I had several fabricated by APCircuits, the least expensive web-friendly maker I could find. Just a week after sending them my data, I received the boards. 
The only error I found on the printed
circuit board was an inadvertent connection between VCC and GND which
took a lot of work to find. This wasn't caught by the FreePCB
program due to a bug which was later fixed. Other than that,
the only proposed changes on the circuit board for a second run would
be cosmetic. The computer worked perfectly after fixing a few
early failing integrated circuits, and a short between a signal line
and VCC, caused by some hastily repaired power-plane cuts when I had
to find the previous VCC/GND connection. The very last error,
however, eluded me for over a year. My project sat in my
basement, with a logic analyzer hooked up to it, untouched, for almost
a year. Then, in November 8, 2006, a full copy of
the Kenbak-1 "Theory of Operation" showed up on www.bitsavers.org
on
November 8, 2006. I think this was directly from John
Blankenbaker, based on the copy he placed on his web site (he doesn't
seem to rely on any other sources for his web site, so I anticipate he
is the original source. This showed state diagram, with
descriptions, which was the key to debugging the last problem.1.3 How hard is it to Make a Kenbak-1? With a circuit board, and a little soldering experience, knowledge of basic electronics, the Kenbak-1 can be assembled in about 4-8 hours, depending on how elaborate or authentic the front-panel needs to be. But a bit of debugging is very likely to be needed. Even though I tested all of the 130+ TTL chips prior to assembly, 3 seemed to be non-functional during debugging. Debugging the computer isn't easy. A logic probe "may" be enough for some people, but I required a logic analyzer to get it running correctly. Since the machine has a serial architecture (all data rotates around in circles, and all math/adding/comparisons occur one bit at a time) it really can be harder to debug than a similar parallel byte architecture. At this point, I've actually developed some testing hardware and attachments, so I may be able to help debug other people's machines, but can't make any promises. I tend to be very busy, and do have a day job and busy family life. 2.0 PARTS AND CONSTRUCTION DETAILS 2.1 The Parts List: In addition to the printed circuit board, there are about 131 7400 series TTL, a handful of 1N914 diodes, several capacitors, many resistors, and a few bipolar transistors. All of these are readily available, and can be acquired from internet dealers, such as unicorn electronics (www.unicornelectronics.com) for under $40.00. But there are two other integrated circuits which are a bit hard to find. The memory is implemented as Intel 1404A dynamic shift registers. I have purchased several of these on Ebay for $25 or less, but they are becoming scarce. I found several vintage IC dealers on the network who promised to get me large quantities at $30 each, plus some expensive shipping. But luckily, I found a source of AMD second-source components out of some old medical equipment. These shift registers may be difficult to find. If you have more money than time, I may supply a few, but may request around $100 each to discourage depleting my supply, depending on how many I have and if I find a new source. Here's the full list of all parts. First the quantity, then the part number or description
2.2 Layout of Parts on Circuit Board: Here's the component layout mask, click on this image for larger image. If any errors are found, let me know. 
Here's a detailed list of each part number, as found on the circuit board markings, and the component placement diagrams. This is not a perfect list, and if errors are suspected, compare with the true part-number list from the TinyCAD partslist which can be found <HERE>. Integrated Circuits
The following are schemataics/logic diagrams of the circuit board, redrawn on TinyCAD and then downloaded into the PC board layout software. They complement the schematics in the Kenbak Theory of Operation Manual, in that these diagrams correctly tell "which" of several same-name signal lines. For example, there may be two of the "CP" clock lines, and this schematic properly tells which one is "CP-1" and "CP-2".             2.4 How Authentic can the Reproduction be?: The reproduction can be as authentic as desired, but there are tradeoffs in cost and reliability.
3.1 Populating the Printed Circuit Board:
There is a single jumper wire on the back side of the board. The location of this is on the layout diagram, but is also pictured on the left to assist in figuring out exactly which pads are involved. There are two capacitors which are mounted on the back of the board. Location of these C16 and C17 are pictured above on the left. Soldering from the solder-side of the board only will be needed for the one that is under an IC.
3.2 Making the Front Panel:
Note that the labels below the number 2 data light show the "4" shifted
a bit to the left. This was the layout of the production Kenbak-1,
but was corrected in the CTI-5050 computer. I chose to copy this
mistake, in tribute to the true spirit of the machine. Clicking on
the link below will open a template for the front panel, at 300 DPI.
This is too large for an 8.5x11 inch page, so you would need a large
format printer. If you print at 300 DPI, it will fit on an 17x11
sheet (usual double-sided sheet.) I made my front panel by
spraying the steel sheet with primer, then contact adhesive, then
placing the printed sheet over the top of the metal, and pressing it
down. Brushing the panel with sand-paper, then applying a clear
plastic transparency sticker would probably be even more authentic,
keeping with the "brushed aluminum" look of the originals. The rear view of the front panel shows how the wires are all attached. A nicer job with heat-shrink tubing could be done, but surely isn't necessary. Notice that connectors are installed to aid in connecting and disassembling the front panel. These will plug into headers, aiding in debugging and assembly. The LED's here are standard T1 3/4 (5 mm) size, with cheap plastic clips available at any supply place (digikey/jameco) and these switches are C&K 8126 snap-action momentary switches. The Key caps were pulled off a vintage keyboard, and epoxy attached to the ends. This didn't work so well. For a second version, I drilled out the inside of the caps with a milling machine, then epoxy glued in tiny key-caps made for the C&K 8126 switches. I made a simple jig to get the caps centered inside the keyboard caps. This worked well, allowed easy removal and made the switches much better lined up. Epoxy appliance paint is a good way to get black and white buttons. Notice that I'm not using the electrolytic capacitors as shown in the original kenbak schematics on the buttons. With modern buttons, they are not needed, but could be added if you want.. 3.3 Wiring up the Power Supply and Panel:
There are many power supplies that are available that fit well. An industry standard 30-40 watt switching size is 3 inches by 5 inches, by 1.2 inches tall, that usually supplies 5 volts at around 3-5 amps, and -12 volts at 500 ma, (as well as not-needed +12 volts at 1-2 amps.) I got several of the power supplies on the right, which I'm quite happy with. They do have a minimum current, so trying to measure voltage without connection to a load, will not work. Of course, it's possible to build a linear supply as shown in the original Kenbak schematics, but that wouldn't be easy, and may be more harsh on the fragile computer circuitry. 3.4 A Simple Case: I made a case out of 22 gauge weldable steel. Needed tools are a nice metal bending break, and a sheer to cut it. This is one part that may be out of the reach of many hobbyists, but I figure that if ready-cut sheets could be obtained by a metal supply house (who would definitely have a nice sheer) the bending could maybe be done without too much work, or a cheap metal bender from "Harbor Freight" or on Ebay. The measurements of layout for the top and bottom sheets are here: (photos pending) First step, lay out markings on the sheets, a fine-point permanent ink pen is easier to see than a scribe. Next, a few little pieces must be removed with tin-snips. Then, the front and back flanges must be bent, but don't completely flatten them. a small L-bracket will be placed later. Next, I bent up the handle-tabs... And next, the hexagon-angles are bent. For this, scrap pieces of 22 gauge steel keep the flanges from being flattened too much, preventing the L-bracket from sliding in. Now, the l-brackets are slid under the flanges. While a spot welder or soldering would be traditional ways to attach these, I cheated, and used a bit of epoxy glue, then firmly flattened the flanges. This requires de-greasing (spraying with 409, and thoroughly rinsing and drying works fine) so that the glue will firmly stick. Last, after checking all angles, and seeing that they fit, with the front and back panels, sand with some fine sand-paper, thoroughly clean again, and then spray with a good quality primer paint. Add 2 or more coats of color, and finish it off with 2 or more light coats of a clear sealer. For the rear panel, I used a dremel and abrasive wheel to cut out the polygon holes, smoothed out the edges with a flat file, then I used epoxy to glue-in a screen-like insert. Painting with the case-color works well. A bracket for the power-supply was made of just a strip of scrap metal. This doesn't need to be painted. Lastly, the side-handles were cut out of wood. A table saw can be used to quickly cut up many of these. I used oak, because it's very strong, but the deep grain in modern oak makes this a real challenge to remove the grain pattern by sanding and painting alone. I got a good metalic effect by applying shinny aluminum duct tape, and carefully trimming it. A matte finish clear top-coat will take away excess gloss. If you have a milling machine, real aluminum handles could be made, but this was more work than I thought I could devote.
4.0
TESTING AND DEBUGGING: When the item is first powered up, don't expect it to run.
Anticipate the real work may have just started. Here's a logical
number of steps to try to get it checked out.
It should be noted that the slew rate of this clock signal is pretty low. That causes some issues if you feed this into both inputs of a 2-input NAND gate. The resulting output of the NAND gate may not be a nice-smooth square wave, it may have a little irregular glitch, just after it switches. This glitch interferes with the later flip-flop divider it feeds into. Solutions are to slow down the clock to make the signal more square (I don't like that solution.) Or you can tie one input of the NAND gate to +5V (this could be done by just bending the pin out of the socket, and jumping a little wire from it to pin 14) or alternatively placing a 40-100 pF capacitor between the output and VCC or Ground (I did experiment with this, and it removed the glitch) or lastly, substituting a 74LS series NAND gate for this will be more tolerant of the slower slew rates, and likely remove the glitch. I've experimented with several of these methods. I'd like to hear if other people had other thoughts. Next, make sure that the signal Phi-1 and Phi-2, the 2-phase clocks after the level shifters, which go to the shift registers, are like this...
Next, make sure that signals T0-T7 from chips U81 and U82 correctly sequence through the 8 bit-times. Only one signal should be active at a time. Insert the shift registers only when you verify that the voltage and
signals on each pin are reasonable. 5 volts, -5 volts, and
the clock signals should be acceptable. 4.2 Set and Read Memory: There's a lot that must be working to set and read memory contents, so this isn't really a trivial operation. With the machine in "STOP" mode (state QC) press the CLEAR button, and all the data lights should turn out. The INPUT light should turn on. Then one at a time, press some of the data buttons, and the corresponding light should turn on. Press clear again, to turn off all the data lights, and press "SET ADDRESS" to set the address to zero. Then press "READ MEMORY" 3 or 4 times. Each time you press this button, you should get different data on the data displayed representing those memory locations. You may write down a note of this data, to verify it stays the same. Note this isn't actually random data. The shift registers memories tend to be alternating ones and zeros, much of the time. After reading out 3 or 4 memory locations, press "DISPLAY ADDRESS" to see what the current address is. This should give the number 3 or 4, depending on how many times you pressed the "READ" button. This verifies that the "W" register is correctly incrementing the address after each data read. Next, press "CLEAR" again and "SET ADDRESS" to reset back to the first memory location, and then press "STORE MEMORY" a few times to store zero in the first few memory locations. Again, press "DISPLAY ADDRESS" to see that the address counter incremented after each of these. Then try to read those locations back, to see if you're setting and reading memory correctly. 4.3 Try to Run a Program: Once you verify you can read and write memory, let's take the plunge and try to run a program. This, again, is a big jump, so don't be surprised if it doesn't work right away. Here's simple program. It simply starts with the value zero in the A register, and keeps adding one to it. After each add, it stores the result in location 200, which is also the output register, so the result is displayed on the front panel. It then loops back to the top of the program, repeating it.
Addr Dat Label Mnemonics CommentTo enter this program, press clear, and "ADDRESS STORE", then for each of the program codes (000, 000, 000, 004, 003, 001, 034, 200, 344, 004) set the input register to them (press clear then press on the correct bits on) and then press "MEMORY STORE". After you've double checked that the correct information is in all of those locations, press the "START" key. If you're lucky, the run light will remain lit, and the front panel lights will continuously count up ward from zero to 377-octal, and back to zero. If it didn't work, start with something more basic, play around. See if you can just execute some "no-operation" no-op instructions. See if the P register changes at all when hitting the "RUN" button. Sometimes an instruction is executed, but the P register (program counter) doesn't update. Play around, and see what works, and what doesn't. 4.4 Debugging the STATE Machine: After the above steps, you'll be well on your way to find out what your machine can and can't do. With luck, it will run the above program, in which case you can test out all of the instruction set one item at a time. If a particular instruction or operation isn't working, I've found a deep-memory logic analyzer (consider the PalmLogic, sometimes available on ebay for about $160) with at least 10-16 inputs, and at least 10,000 samples, really helps. I set up 5 of the inputs to record the state outputs (the outputs of U130, U131, and U132, which are the latched version of JS0-JS4) and then I can see where the states are mis-functioning, and debug why.
Above are copies of the Kenbak-1 State Machine state diagrams. In red, however, are the representatives for S4, S3, S2, S1, and S0, which will aid in debugging with a logic analyzer. My debugging technique is to find an instruction that does not correctly operate, write a single-instruction program with that instruction, then hook the logic analyzer up to the state outputs. I can then trigger on a state the program is expected to hit (falling edge of S0, which will occur as soon as the "RUN" key is released, is often a good place.) Then I run the program, and see that the correct states are sequenced. If not, it's a simple review of associated logic values will determine why.
Here is a view of a PalmLogic logic analyzer display. The state lines are on the bottom, an octal representation of the state is just above that. The octal nomenclature was done so that it corresponds with the red state designations in the above state diagrams. The upper level signals were some intermediate signals 4.5 A List Test Suite to find Non-working Instructions: The huge variety of instructions on the Kenbak-1 makes debugging challenging. A program which will test out all instructions, and then more limited programs to hone-in on particular functions are needed. A set of debugging programs are in the top-level Kenbak-1 web pages.
|
|
|
|||||||||||||||||||||||||
