Network Working Group A. Katz
Request for Comments: 798 ISI
September 1981 DECODING FACSIMILE DATA FROM THE RAPICOM 450
This note describes the implementation of a program to decode
facsimile data from the Rapicom 450 facsimile (fax) machine into an
ordinary bitmap. This bitmap can then be displayed on other devices
or edited and then encoded back into the Rapicom 450 format. In
order to do this, it was necessary to understand the how the
encoding/decoding process works within the fax machine and to
duplicate that process in a program. This algorithm is descibed in
an article by Weber  as well as in a memo by Mills , however,
more information than is presented in these papers is necessary to
successfully decode the data.
The program was written in L10 as a subsystem of NLS running on
TOPS20. The fax machine is interfaced to TOPS20 as a terminal
through a microprocessor-based interface called FAXIE.
Grateful acknowledgment is made to Steve Treadwell of University
College, London and Jon Postel of Information Sciences Institute for
II. Interface to TOPS20
The fax machine is connected to a microprocessor-based unit called
FAXIE, designed and built by Steve Casner and Bob Parker. More
detailed information can be found in reference . FAXIE is
connected to TOPS20 over a terminal line, and a program was written
to read data over this line and store it in a file. The decoding
program reads the fax data from this file.
The data comes from the fax machine serially. FAXIE reads this data
into an 8-bit shift register and sends the 8-bit byte (octet) over
the terminal line. Since the fax machine assigns MARK to logical 0's
and SPACE to logical 1's (which is backward from RS232), FAXIE
complements each bit in the octet. The data is sent to TOPS20 in
octets, the most significant bit first. If you read each octet from
most significant bit to least significant bit in the order FAXIE
sends the data to TOPS20, you would be reading the data in the same
order in comes into FAXIE from the fax machine.
The standard for storing Rapicom 450 Facsimile Data is described in
RFC 769 . According to this standard, each octet coming from
FAXIE must be complemented and inverted (i.e. invert the order of the
bits in the octet). Thus, the receiving program did this before
storing the data in a file. When the decoding program reads this
file, it must invert and complement each octet before reading the
Each data block from the fax machine is 585 bits long. The end of
this data is padded with 7 0's to make 592 bits or 74 octets.
According to RFC 769, this data is stored in a file preceded by a
length octet and a command octet. The possible commands are:
56 (70 octal)--This is a Set-Up block (the first block of the
file, contains information about the fax image)
57 (71 octal)--This is a data block (the rest of the blocks in the
file except for the last one)
58 (72 octal)--End command (the last block of the file)
The length field tells how many octets in this block and is always 76
(114 octal) except for the END command which can be 2 (no data). The
length and command octets are NOT inverted and complemented.
Below is a diagram of each block in the file:
| length | command| data | data | ... | |
III. The Rapicom 450 Encoding Algorithm
An ordinary 8 1/2" by 11" document is made up of about 2100 scan
lines, each line has 1726 pels (picture elements) in it. Each pel
can be either black (1) or white (0).
The Rapicom 450 has three picture quality modes. In fine detail
mode, all of the document is encoded. In quality mode only every
other scan line is encoded and it is intended that these missing
lines are filled in on playback by replicating the previous line.
There is also express mode, where only every third line is encoded.
Data is encoded two lines at a time, using a special two dimensional
run-length encoding scheme. There are 1726 pels on top and 1726 pels
on the bottom. Each pair (top-bottom) of pels is called a column.
For each of the 1726 columns you can have any one of four
configurations (called states):
(top-bottom) pels state
------------ ---- ------
W-W 0,0 0
W-B 0,1 1
B-W 1,0 2
B-B 1,1 3
The encoding algorithm can be described in terms of a
non-deterministic finite-state automaton shown in Fig. 1 (after Mills
). You start out in a state (0-3) and transform to another state
by emitting the appropriate bits marked along the arcs of the
diagram. For example, suppose you are in state 1 (WB). To go to
state 2 (BW), you would output the bits 101 (binary); to go to state
0 (WW) you would output the bits 1000. Note that the number of bits
on each transition is variable.
In states 0 (WW) and 3 (BB), a special run length encoding scheme is
used. There are two state variables associated with each of these
states. One variable is a run-length counter and the other is the
field length (in bits) of this counter. Upon entry to either of
these two states, the counter is initialized to zero and is
incremented for every additional column of the same state. At the
end of the run, this counter is transmitted, extending with high
order zeros if necessary. If the count fills up the field, it is
transmitted, the field length is incremented by one, and the count
starts again. This count is called the run length word and it is
between 2 and 7 bits long.
For example, suppose we are in state 0 (WW) and the run length for
this state (refered to as the white run length) is 3. Suppose there
are three 0's in a row. The first 0 was encoded when we came to this
state, there are two more 0's that must be encoded. Thus we would
send a 010 (binary). Similarly, if there are seven 0's in a row, we
would send a 110, but eight 0's would be sent by 111 followed by 0000
and the white run length becomes 4. (Ten 0's would be encoded as 111
followed by 0010 and the white run length would be 4).
Run length word lengths must be between 2 and 7. The field length is
decremented if the run is encoded in one word and:
1. If the run length is 3 and the highest order bit is 0.
2. Or, if the run length is 4, 5, 6, or 7 and the highest order 2
bits are 0.
In addition to all this, there is a special rule to follow if the run
occupies at least two run words (and can involve incrementing the run
word size) and the run ends exactly at the end of a scan line. In
this case, the last word of the run is tested for decrement as if the
previous words in the run did not exist.
To confirm the reader's understanding of the encoding procedure,
suppose we had the following portion of a document (1=black,
top row: 0 1 1 1 1 1 0 0 0 0 0 1 1 0 0 0 ...
bottom row: 1 1 1 1 1 0 0 0 0 0 0 0 0 1 0 0 ...
state: 1 3 3 3 3 2 0 0 0 0 0 2 2 1 0 0 ...
Suppose also that the black run field length is 2, the white run
length is 3, and the state is 1. (This example comes from
This portion would be encoded as:
1 1011 11 000 1 0100 100 1 0 010 1000 ...
NOTE: It turns out that the Rapicom 450 sends the bits of a field
in reverse order. This will be discussed in the section V.
However, since each run length field is sent reversed, the above
encoded bit pattern would actually be sent as:
1 1011 11 000 1 0100 001 1 0 010 1000 ...
|-this is actually 100 reversed
This example illustrates the rule for decrementing the run length
top row: 0 1 1 0 0 1 1 1 1 1 0 0 ...
bottom row: 1 1 1 1 1 0 1 1 1 1 1 0 ...
state: 1 3 3 1 1 2 3 3 3 3 1 0 ...
Here, let us suppose that the black run field length is now 4, the
white is still 3, and the state is 1.
This portion would be encoded as:
1 1011 0001 1 1 101 0111 011 1 1000 ...
|-goes to 3 |-blk cnt goes to 2
When we reverse the order of the run fields, the bit pattern that
is actually sent is:
1 1011 1000 1 1 101 0111 110 1 1000 ...
|-this is actually 0001 reversed, etc.
IV. The Setup Block and the Data Header
Each data block from the fax machine is 585 bits long. The number of
blocks in a picture is variable and depends on the size and
characteristics of the picture. It should be emphasized that a block
can end in the middle of a scan line of the document. There can in
fact be many scan lines in a block.
The 585 bit data block is composed of a 24 bit sync code which is
used to recognize the beginning of a block, a 37 bit header, 512 bits
of actual data, and a 12 bit CRC checksum:
| 24-bit | 37-bit | 512-bit | 12-bit |
|sync code | header | data | checksum |
The number of useful data bits is variable and can be between 0 and
512 (although there are always 512 bits there, some of them are to be
ignored). The number of data bits to be used is given in the header.
The 37 bits of header is composed of:
| 2-bit |5-bit| 10-bit | 12-bit | 3-bit | 3-bit |2-bit|
|seq num|flags|data count| x position|black size|white size|state|
An explanation of these fields follows:
IMPORTANT NOTE: Most (but not all) of these fields are sent by
the fax machine in REVERSE ORDER. The order of each n-bit field
must be inverted.
This is used to synchronize on each block. The value of this
24 bit field is 30474730 octal (not reversed).
This number cycles through 0, 1, 2, 3 for the data blocks. It
is 0 for the Set-Up block (not reversed).
Each of these flags are 1 bit wide:
Purpose unknown, it always seems to be 1.
Purpose unknown, it always seems to be 0.
1 for Set-Up blocks (which are repeated when coming from
the fax machine though only one of them is transfered by
FAXIE to TOPS20 and stored in the file) and 0 for data
Purpose unknown, doesn't matter.
1 if this is a Set-Up block.
Number of useful bits to use out of the 512 data bits. NOT ALL
of the 512 data bits are used, only this number of them. This
number can be 0 (usually in one of the first data blocks) which
means to throw away this block. (This field is reversed!)
Current position on the scan line, a value between 0 and 1725.
If this number is greater than where the previous block left
off, the intervening space should be filled with white (0's).
If this number is less than where the previous block left off,
set the X position to this value and replace the overlapped
data with the new data from this block. If this number is
greater than 1726, ignore this field and continue from where
the previous block left off. (This field is reversed!)
The size of the black run length field, must be between 2 and
7. This is the correct value for the black size. It may
differ from what was found at the end of the previous block.
(This field is reversed!)
The size of the white run length field, must be between 2 and
7. It may differ from what was found at the end of the
previous block. (This field is reversed!)
The current state. This is the correct state. It may differ
from the state at the end of the previous block. (This field is
512 bits of the actual encoding of the document. NOT ALL of
this data is used in general, only the amount specified by the
data count. If this is a set up block, the data contains
information about the type of document (see below).
CRC checksum on the entire block. Uses polynomial
In a setup block, the data portion of the data block consists of:
| 6-bit | 5-bit | 1-bit | 20-bits | 480-bits
| flags | spare |multi page| of zeros | 1's and 0's
Specifically these are:
6 flags (each are 1 bit):
Is 1 if express mode.
Is 1 if detail mode. (NOTE: If the Detail and Speed flags
are both 0, then data is in Quality mode).
14 inch paper
is 1 if 14 inch paper length.
5.5 inch paper
is 1 if 5.5 inch paper length. (NOTE: If the 14 inch and 5
inch flags are both 0, then paper length is 11 inch).
is 1 if paper is present at scanner (should be always 1).
These 5 bits can be any value.
1 if multi page mode
Rest of data of set-up block:
The above fields are followed by twenty 0 bits and the rest of
the 512 bits of the block are alternating 0's and 1's.
There are a number of important points to be remembered in regard to
the header of a data block. First of all, the data count, the
x-position, and the black and white run sizes must be read IN REVERSE
ORDER. The reason for this is that the fax machine sends these bits
in reverse order. However, the sequence number and the state fields
ARE NOT REVERSED. In addition to this, each run field in the data IS
REVERSED. This reversing of the bits in each n-bit field is
completely separate from the reversing and complementing of each
octet mentioned earlier.
Second, only the first n bits, where n is the value of the data count
field (remember its reversed!), of the data is valid, the rest is to
be ignored. If n is zero, the whole block is to be ignored.
Third, if the x position is beyond where the last block ended, fill
the space between where the last block ended and the current x
postion with white (0's). If the x postition is less then where the
last block ended, replace the overlapped data with the data in the
new block. If the x postition is greater than 1726, ignore it and
continue from where the previous block left off.
Fourth, the black size, white size (reversed), and state (not
reversed!) given in the header are the correct values even if they
disagree with the end of the previous block.
Finally, the sequence number (not reversed) should count through
0,1,2,3. If it does not, a block is missing.
V. The Decoding Algorithm
Upon first glance at the finite state diagram in Figure 1, it may
seem that it would be difficult to create a decoding procedure. For
example, if you are in the WW state, and the next bit is a 1, how do
you know whether to do a transition to WB or BW? The answer to this
is to recognize that every arc out of the BW state begins with 0 and
every arc out of WB begins with 1. Thus, if you are in the WW state,
and the next bit is 1, followed by a 0, you know to go to the BW
state. If the next bit is 1, followed by a 1, you know to go to the
In writing the decoding program it was necessary to have two ways of
reading the next bit in the data stream. The first way reads the bit
and "consumes" it, i.e. increments the bit pointer to point at the
next bit. The other way does not "consume" it. Below are four
statements which show how to decode fax data. The numbers in
parentheses are not to be consumed, that is to say they will be read
again in making the next transition.
If I am in state BW (2) and the next bits are:
0 (0): go to BW
0111: go to BB
010 (1): go to WB
0100: go to WW
If I am in state WB (1) and the next bits are:
1 (1): go to WB
1000: go to WW
101 (0): go to BW
1011: go to BB
If I am in state WW (0), then first go through the run length
algorithm, then if the next bits are:
0: go to BB
1 (0): go to BW
1 (1): go to WB
If I am in state BB (3), then first go through the run length
algorithm, then if the next bits are:
0: go to WW
1 (0): go to BW
1 (1): go to WB
For the run length algorithm, remember, look at the next n bits,
where n is the length of either the black or white run length
word, REVERSE the bits, and output that many BB's or WW's
(depending on whether black or white run). If the field is full,
increment the size of the word, and get that many bits more, i.e.
get the next n+1 bits, etc. Also, the run length word length can
be decremented according to the rules given in section III.
You always go through the run length even if there is only one WW
or BB, in this case, the run field will be 0.
Let us look at the first example given in section III. Suppose we
want to decode the bits: 110111100010100100100101000... (we have
already reversed the run lengths to make things easier).
We are in state 1 (WB) and the black run length word length is 2
and the white length is 3. We get these initial values either
from the block header, or by remembering them from the previous
transitions if this is not the start of the block. According to
our rules, we would parse this string as follows:
1(1) 1011 11 000 1(0) 0100 100 1(0) 0(0) 010(1) 1000...
The numbers in parentheses are numbers that were read but not
"consumed", thus the next number in the sequence is the same as
the one in parentheses. First, we see a 1 and that the next bit
is a 1, this means that we go to WB. Then we have a 1011 which
means to go to BB. Then we do a run, we have a 11 followed by a
000 which means the black run length gets incremented by 1 (it is
now 3) and we get 3 MORE BB's. Now we have a 1 followed by 0
which means go to BW. Next a 0100 which is WW. Then we have a
run, 100, which means four more WW's. We keep going like this and
we get the original bit pattern given in the first example of
It is important to always start fresh when dealing with each
block. There are many reasons for this. The first is that
sometimes blocks are dropped, and you can recover from this if you
resynchronize at the start of each block. Also, if at the end of
the previous block, there is about to be a transition, instead of
making it at the beginning of the next block, the fax machine
gives the new state in the header of the next block and goes from
there. Thus it is important to always start at whatever state is
given in the header, and to align yourself at the current X
position given there also.
Sometimes, while decoding a block, a bit pattern will occur which
does not correspond to any transition. If this happens, the rest
of the block may be bad and should be discarded.
The decoding program decodes the fax data block by block until it
comes to an END command in the data, or runs out of data.
VI. Program Performance
The L10 NLS program takes about two CPU minutes to run on TOPS20 on a
DEC KL10 to decode the average document in fine detail mode. In this
mode, the picture is about 1726 by 2100 pels, and takes about 204
disk pages to store.
We have a program which displays bit maps on an HP graphics terminal
and have been able to display portions of documents. (not all of an
8.5" by 11" document will fit in the display). We can use the
terminal's zoom capability to look at very small portions of the
 Weber, D. R., "An Adaptive Run Length Encoding Algorithm",
International Conference on Communications, ICC-75, IEEE, San
Francisco, California, June 1975.
 Mills, D. L., "Rapicom 450 Facsimile Data Decoding",
WP2097/MD33E, COMSAT Laboratories, Washington D.C., undated.
 Casner, S. L., "Faxie", ISI Internal Memo, USC/Information
Sciences Institute, February 1980.
 Postel, Jon, "Rapicom 450 Facsimile File Format", RFC 769,
USC/Information Sciences Institute, September 1980.