Jonathan Mordosky posts a loader in cosmacelf
---------------------------------------------

I built my Membership Card a few weeks back and have been having a blast
 programming it.  The 1802 has such an awesome architecture.  I built
 it with 64K of ram with the idea that I could try different software 
without having to program and swap EPROMs.  

Accordingly, I wrote a BIN loader that I could toggle in to bring
 in base software.  I figured I would post it here in case anyone 
else had a use for it.  It's written for eight data bits, and no
 Parity at 300 Baud (what's your hurry, this is a hobby after all)
 but could probably do 1200 with no modifications.  It's still
 a little long at 46 bytes and I'd welcome ideas to shorten it.
 
 I'm currently working on a small Monitor (< 1k) with XModem 
capability to go with it.  Thank You Lee Hart for a great product!

Jonathan later added: "A main goal for me in the design 
of the program, was to keep it as short as possible so that 
it is easier to toggle in, thus cutting out all the bells 
and whistles.  Hopefully, the code is understandable enough
 that it will be possible to add any desired extra features
 on the fly as needed."


Jonathan, Oct 24 2015

Lee Hart modifies the program
-------------------------

Jonathan Mordosky published a great little serial
loader for 1802 Elf/Membership card computers. I tried it, and it works!

However, the timing is a little off, which limits it to low baud rates.
I spent some some this evening to see if I could improve it. The
attached file has Jonathan's program, and my enhancements. I haven't
fully tested it, but thought [some of you] might like to see it.

Analysis by Josh Bensadon
-----------------------------

Very well done.  I followed it through for the most part and I do believe
it will work. I am going to try it on my VELF (I just need to change
to EF2 and flip the polarity).

Very nice and efficient use of the registers, and especially nice touch 
using long jump if DF for that extra 1/2 cycle needed.

At first, I was concerned about the autobaud for 9600, since your resolution
isn't very fine (increments of 2 cycles=3D 17.9 uSec)
But, the loop will count to 2 in just 102.8uSec... Ahh, but wait... the 
testing of EF2 will happen within a 1 instruction cycle time, that's a 9uSec
window! [From a timing drawing I made] there is a 1 in 8 chance that the
EF signal will happen late enough to cause the counter to reach 3.

I believe if you replace the NOP with just 2 more SEX 2's, you will push
this error in favor of getting 2 every time for 9600 baud. [If not, says Lee,
just assume 9600 and force it to 2 if you wish.]

PS.  For the purpose of mapping this out, I assumed the EF line is sampled
on the 7th clock in the execute machine cycle, but it doesn't matter where
it gets done, since it will just result in a sliding window anyway.

[Lee added: It actually tests the EFn lines in the middle of the second clock pulse 
of the Execute cycle.]

Now, I have entered the autobaud portion into my VELF and just output the
Delay value then loop back. The VELF uses a 3.579+ MHz colour burst crystal 
[divided by 2 to run at about 1.79Mhz]. Here's the results of many times operation:

Hitting <SPACE> at 9600, 100 times caused a delay of 2, 76 times and delay 3, 24 times.

I then changed the NOP to another pair of SEX instructions and repeated the test.
I  got a delay of 2, 100 times out of 100.

The display was so steady, I wondered if it was working, so I choose to test other bauds and got this:
4800:   delay of 8   (steady)
2400:   delay of 0x13 or 0x14  (about 50/50 but I'm not counting)
1200:   delay of 0x2B (steady)
300:   delay of 0xB6 (with the occasional 0xB7 about 10% of the time)

To account for the high percentage of 3's (it was predicted 1/8th = 12.5%, 
not 24%).  I will guess there must be a little lag in the RS-232, I mean 
to say some capacitance might be extending the pulse duration beyond 
104.1666 uSec. Or the exact high/low voltage levels 
on the EF line may have an influence.

Bottom line, I think Lee's program works great,  it just needed 
a little more SEX.

Lee Hart explains results
--------------------------

[Josh is] right on needing 10 (not 9.5) instructions/loop. Theory is good,
 but practice makes perfect. I stared at it some more, and now see why.

We don't want the *same* delay for baud rate detection and bit timing.
 Baud rate detection needs a "window" on each side of the desired 
number. At 1.8 MHz and 9600 baud, a bit is 11.65 instructions 
long. 11.5 instructions/bit is the closest the 1802 can get. The 
auto-baud routine therefore needs to check early, and late... at 
something like 11 and 12.

But the loop that detects D5 has a 0 to 1 instruction delay, so
 the width of D5 appears to be 10.65-11.65 instruction times.
 Checking at 10 and 12 neatly "windows" the possible variations
 in apparent width of D5 so it always selects Delay=2.

I modified my assembly listing accordingly, and attached a new
 version.

Two other changes: I made the 1/2-bit delay 1 instruction shorter,
 by moving the STR 2 and GHI 3 to 0015-0016. When Delay=2, this
 makes it 3+Delay instructions from Start detect at 0017 to Bit
 test at 0020. I.e nominally 5 instructions at 9600 baud (where 
the actual time is something like 4.8-5.8 instruction times at
 9600 baud). I could make it 5.5, but it would add bytes to 
this minimal loader and it may not be necessary.

And Herb, I added an optional "test" patch at 0011-0013 to
 simply display the value of Delay. You can type <spaces> 
repetitively and observe the values for any given baud rate.

NOtes on application to the 1802 M/S card - Lee Hart
----------------------------------------

Early ones (rev.A-E) had an RC clock, adjusted with a trimpot.
 They'll work at lower baud rates, where clock speed variations
 with temperature and supply voltage won't matter. {Or, the
 owner can choose to add a ceramic resonator and resistor to
 run at full speed. - Herb]

Later ones (rev.F-present) have a 1.8 MHz ceramic resonator
 (which actually tends to run at 1.79 MHz).

- When the pot is set to high resistance (5k to 1meg), it runs at a low
    frequency set by the pot and 40pf capacitance of the resonator.
- When the pot is set to zero, the clock runs as fast as the 4093 can
    go (about 2 MHz with a 5v supply).
- When the pot is set for an RC frequency near 1.8 MHz (about 1-3k
  roughly 1/2 turn from the end),
    the resonator makes it "lock in" at 1.8 MHz. The frequency is
    very stable despite variations in temperature or supply voltage.

Speculation on higher baud selection by Lee Hart
-------------------------------------------

Actually, you might be able to send/receive at 19.2k, 28.8k, or even 
38.4k baud! At 3.58MHz/2 = 1.79MHz, each 2-cycle instruction is 8.924uS. 
The number of instructions/bit needed for each of the standard higher 
baud rates is:

baud	instructions	bus
rate	per bit		cycles	error
----	------------	------	-----
9600	11.5		23	1.3%
19200	6		12	3%
28800	4		8	3%
38400	3		6	3%

A 1.3% is obviously acceptable, and works. Using a 3-cycle instruction 
(an instruction and a half) gets us pretty close. A 3% error *might* 
still be close enough. If so...

Leave out the Delay test (i.e a fixed baud rate routine). Here's the 
minimum loop I can think of, which is just 4 instructions:

; Enter with Start bit detected, and in the middle of Start bit,
; D=FF, and R(X) pointing to RAM where the bytes are to be stored

NextBit	BN3 Zero	;1  test next bit
	SKP		;2   if bit=1 (EF3 pin low), leave DF=1
Zero	SHR		;2   if bit=0 (EF3 pin high), set DF=0
	SHRC		;3  shift bit into byte (DF into D7, D0 into DF)
	BDF NextBit	;4  if DF=1, then Start bit hasn't shifted all
			;   the way thru byte. Loop until it does.
;This code should end in:
	STR 2
	INC 2		; (if X=2)
or
	STR 2
	OUT 4		; to display the byte received on LEDs

; and then loop either way with
	STXD		;   store byte in RAM, increment pointer,
			;   and loop back to find next byte

19200 needs a 6-instruction loop. So just add two 2-cycle do-nothing 
instructions to the loop.

28800 needs a 4-instruction loop. The above does that just as shown.

38400 needs 3 instructions per loop. We can't do that; but we *can* 
unwind the loop. Leave off the final BDF, and simply repeat the other 4 
instructions nine times, once for Start and each of 8 data bit.

I wonder if it would really work? :-) - Lee


Notes on achieving a high baud rate - Herb Johnson
--------------------------------------

But in the end...it's "I tried it and got to X baud rate". 
There's nothing wrong with a M/S card kit that only gets to 
4800 baud, or 2400 baud. The point of the kit is to learn, have
 fun, get something working, to tinker around. The documentation
 on my site (searchable from Google search, Bing search, etc.) 
reflects years of versions of kits, each of which 
have their own particular "features". Learning about these features
 is informative; a kit that works perfectly without effort or
 puzzles provides no reasons to dig so deep. 

This document last updated Nov 2 2015 HRJ