Date: Saturday, 14 December 1985
From: Keith Petersen, W8SDZ
Re:   MODEM/XMODEM robustness improvement on EOT tests successful

One of the most serious shortcomings of the Christensen protocol is
that a single EOT (04h) received during a protocol transfer is taken
by the receiver as end-of-file.  In the presense of noise, this can -
and does occur.  The result is an incomplete file on the receiving end
and an abort after ten retries on the sending end.

Frank Wancho and I have been experimenting with the idea of the
receiving end sending a NAK when it gets the first EOT (or what it
thinks is an EOT).  The sender will send another EOT in response to
this if it's really EOT.  Otherwise, it will send SOH and the next
record, thus allowing the receiver to recover from a false EOT.

I have tried this with an experimental version of XMODEM (no source
for IMP or MEX, so can't try them).  It works perfectly.

The code looks something like this:

[Already have character here, expecting SOH for new record]

TSTSOH	if SOH then Jump to GOTSOH which gets rest of header, etc.
	if not EOT then Jump to SENDNAK

	else wait for one second of no received chars, send NAK

	then do:

	quick purge of input port to clear garbage
	wait up to 10 seconds for another character
	if timeout Jump to timeout routine
	if character is EOT return to caller with flag set
	Character was not EOT, Jump to TSTSOH to test for others

--END OF ROUTINE--

SENDNAK is same as always, wait for one second of no characters, send
NAK, increment error count, Jump to receive loop.

The only visable evidence that this improved code is in place is that
the sender shows one error message: ** NAK received, not ACK **
after it sends the real EOT.  It just sends another, so no problem.

Earlier versions of various Christensen MODEM programs would send EOTs
forever (it was a closed loop).  I have a max ten counter in the
SENDEOT routine.  The latest versions of IMP and KMD have only a three
counter there, which I consider too low.  The program should attempt
to send EOT ten times, same as it does for any other received NAK
during the file transfer.

After we're through testing it, Frank will be releasing the improved
Tops-20 MODEM program.  I am still working on XMODEM for CP/M-80.  It
has so many conditional assembly switches and code dependant on other
programs, such as BYE and ZCPR, that I am making a new program without
all this rampent featuritus, which I will be releasing soon as XMD.

--Keith