XMODEM, CRC XMODEM, WXMODEM File Transfer Protocols

    Please circulate this document anyway that you see
    fit without alteration except on the page at the
    end titled: "Notes and Comments".  It is requested
    that anyone using these protocols within a commer-
    cial product not charge for them as an option or
    surcharge, but include XMODEM and its derivations
    as part of the basic product.


                                     Peter Boswell
                                     June 20, 1986
                                     People/Link email: TOPPER

 This document was converted to Wiki syntax by Rob Swindell on July-13-2011.

In the years that have past since Xmodem was first developed as a file transfer protocol, many thousands of people have been involved in finding reasonable ways to move data via asynchronous telephone communications. I appreciate the opportunity that I have had to meet and learn from many of these people. There is nothing in this document that did not actually come from someone else. Indeed, whether it is WXMODEM, X.PC, Synchronous dial-up X.25, SNA, ZMODEM, Blast, Kermit or any other protocol that becomes the dominant dial-up file transfer protocol for personal and home computers is just not important. What is important is that the public domain have a high speed file transfer protocol that is reasonably popular and commonly available for many types of personal computers, for bulletin boards and for services such as People/Link, Delphi, CompuServe, GEnie and The Source.

Here are a few people that all of us should thank and I would especially like to recognize:

• Ward Christensen

Ward, a true pioneer in the microcomputer communications area, is the author of the original Checksum Xmodem protocol. Thanks for reminding me to “keep it simple stupid”.

• Chuck Forsberg

Chuck has edited perhaps the best work on Xmodem and has provided both YMODEM (1K Xmodem) and ZMODEM (Windowed YMODEM) to the public domain. Thanks for showing me a protocol which would deal with the X-On/X-Off problem and reminding me that there is such a thing as a DLE character.

• Richard (Scott) McGinnis

Scott is the architect, the moving force, for the People/Link software system. His ideas, comments and encouragement have been wonderful. Wait until you see his visual conference program for the IBM PC! Thanks for showing me how to use a DLE.

• Gene Plantz

Gene operates a major IBM PC bulletin board in the Chicago area and has been active in the National SYSOP Association. Thanks for pushing me to do something about performance.

In a historical perspective, there seems to be a common pattern in all computer systems development that can shed some light on where we stand and how we got here. The pattern is function first, then integrity and finally performance.

Any kind of software must first do something worthwhile. There is no point in being error free, or inexpensive to operate if we do not want the function. Back in 1977, Ward Christensen had a need to move data between microcomputers. Within a year it became obvious that the function Xmodem provided met a real need to many microcomputer users.

Once we have a new function and we accept it, there is a normal desire for the function to be correct. No one can't count the times that new software users have pointed out ... “that new function is super, but the results are wrong”. The effort changes from providing new function to providing integrity. The development of CRC Xmodem is a clear response to the integrity phase of a service as it reduced undetected transmission errors by many orders of magnitude.

After the integrity has been accepted, people begin to look toward cost and performance. XMODEM entered this phase in 1984-1985. Chuck Forsberg's YMODEM is a major step in this effort providing larger block sizes, batch mode and more. His ZMODEM is a major step toward making XMODEM derivative protocols work effectively with Public Data Networks and most importantly, provides for restart of a file transfer at the point of failure. WXMODEM, presented here, is an alternate solution to ZMODEM which is, hopefully, an easier solution to the most important performance problems.

No one really knows where XMODEM and the file transfer function will go in the coming years. Perhaps X.PC from Tymnet, MNP from Microcom or Synchronous X.25 will slowly push XMODEM, et. al, into history. I think this will happen, but not for maybe 5 to 10 years. Perhaps when 50% of the households outgrow the Commodore 64, or when modem manufac- turers can provide a $50 synchronous modem we will see the beginning of the end for XMODEM, but not today. 

XMODEM and its derivatives have become the primary method for file transfer for personal computers. Hopefully this document will help people to understand these protocols and to implement them on their own. In particular, this document presents an additional XMODEM derivation to the public domain: WXMODEM.

After working with bulletin boards, Public Data Networks such as Tymnet and Telenet, and commercial host systems such as People/Link, Delphi, CompuServe and others, a number of people came to believe that hobbyist, home and business users would benefit significantly from a new, conceptually simple file transfer protocol which would provide improved performance and fully support the public data networks such as Tymnet, Telenet and Datapac.

But before WXMODEM can be presented, XMODEM and CRC XMODEM must be described in detail.

I've elected to use two special terms: transmitter and receiver. The transmitter is the computer/software which is transmitting data packets and receiving acknowledgement characters. The receiver is the computer/software receiving the data packets and transmitting acknowledgement characters.

Here is a table of special ASCII characters that are used throughout this paper:

      Name      Decimal        Hexadecimal    Description

      SOH          01           H001          Start Of Header
      EOT          04           H004          End Of Transmission
      ACK          06           H006          Acknowledge (positive)
      DLE          16           H010          Data Link Escape
      X-On (DC1)   17           H011          Transmit On
      X-Off(DC3)   19           H013          Transmit Off
      NAK          21           H015          Negative Acknowledge
      SYN          22           H016          Synchronous idle
      CAN          24           H018          Cancel

Xmodem is a popular error recovery type protocol for transferring files between computers via serial, asynchronous communications. Before learning more about Xmodem, it is important to hear what its author has to say:

    "It was a quick hack I threw together, very unplanned (like
    everything I do), to satisfy a personal need to communicate
    with some other people.  ONLY the fact that it was done in
    8/77, and that I put it in the public domain immediately,
    made it become the standard that it is"....."People who
    suggest I make SIGNIFICANT changes to the protocol, such as
    'full duplex', 'multiple outstanding blocks', 'multiple
    destinations', etc etc don't understand that the incredible
    simplicity of the protocol is one of the reasons it survived
    to this day in as many machines and programs as it may be
    found in!"
      

The protocol is Asynchronous, 8 data bits, no parity bit, one stop bit. Modems which are commonly used are AT&T 103 (300 baud), AT&T 212A (1200 baud) and CCITT V.22 (2400 baud).

Typically, the data in a file is transmitted without change (if a 7 bit machine, the left most, high order, bit is always zero) except that CP/M and MS/DOS operating systems want a ^Z (decimal 26) to represent end-of-file.

Prior to entering the protocol, both the transmitting and receiving computer must know where to get the data (what file is to be transmitted) and where to put the data (file to store the data or buffer area). In Xmodem one side of the file transmission is always in charge (local computer), asking the other side (remote computer) to either transmit a file or to accept a file. Through a dialog outside of Xmodem the local computer (your PC) first sends commands to the remote computer to select a file name to prepare to transmit or receive a file via XMODEM. Once this is completed the remote computer enters the XMODEM protocol. Now the local computer must be told what file to transmit or receive and it enters the XMODEM protocol, and hopefully data starts moving.

Upon entering the Xmodem protocol, the transmitting computer waits between 10 seconds and a minute to receive an NAK character from the receiving computer. The receiving computer is said to drive the protocol. The transmitter may retry any number of times. If any character other than a NAK or CAN is read by the transmitter, it is ignored. The CAN character implies cancellation of the Xmodem file transfer and that the transmitter should leave the Xmodem protocol. Once the receiver has sent a NAK, it will wait 10 seconds for data to begin to arrive. If none arrives in 10 seconds, the receiver will send another NAK and continue to repeat 10 times at which point the receiver will leave the Xmodem protocol (typically with a super cryptic error message such as “aborted”, “NAK retry maximum exceeded”).

      Transmitter                        Receiver

      [wait for one minute]         <    [NAK]

      [begin block transmission]    >

The transmitter takes the data, divides it into 128, 8 bit byte pieces and places it in an Xmodem Packet.

The Xmodem Packet looks like this:

       [SOH] [seq] [cmpl [seq] [128 data bytes] [csum]

       SOH       Start of header character (decimal 1).

       seq       one byte sequence number which starts at 1, and
                 increments by one until it reaches 255 and then
                 wraps around to zero.

       cmpl seq  one byte 1's complement of seq.  This can be
                 calculated as cmpl = 255 - (255 and seq) or using
                 xor as cmpl = (255 and seq) xor 255.

       data      128, 8 bit bytes of data.  Note than when sending
                 CP/M and MS/DOS files a ^Z (decimal 26) must be
                 added to then end of the file.  If the last block
                 of data is less than 128 bytes, the Xmodem packet
                 must be padded with characters, usually ^Z's.

       csum      one byte sum of all of the data bytes where any
                 overflow or carry is discarded immediately.  For
                 example, if the first 3 bytes are 255, 5 and 6 the
                 checksum after the first 3 bytes will be 10.

 Once Xmodem Initiation has completed, the transmitter sends the first Xmodem packet and then waits. After the receiver has the full packet, it will compare its own checksum calculation with the checksum that was sent by the transmitter. If the checksums match, the receiver will send an ACK. If the checksums are different, the receiver will send a NAK.

After receiving an ACK the transmitter will send the next Xmodem packet. If a NAK is received, the transmitter will resend the same XMODEM packet again.

Once the transmitter has sent the last Xmodem packet and has received an ACK, the transmitter will send an EOT and then wait for a final ACK from the receiver before leaving the Xmodem protocol. When the receiver sees an EOT instead of an SOH (the first character the next packet), the receiver transmits an ACK character, closes its files and leaves the Xmodem protocol.

Let's look at a three block file transfer:

       Transmitter                                  Receiver

                                     <<<<<          [NAK]
       [SOH][001][255][...][csum]    >>>>>
                                     <<<<<          [ACK]
       [SOH][002][254][...][csum]    >>>>>
                                     <<<<<          [ACK]
       [SOH][003][253][...][csum]    >>>>>
                                     <<<<<          [ACK]
       [EOT]                         >>>>>
                                     <<<<<          [ACK]

Seems easy, right? And it is, until something goes wrong.

It has become a defacto standard that the receiver may cancel the file transfer by sending a CAN character and then leaving the protocol. If the transmitter receives a CAN character when expecting either a NAK or ACK, the transmitter is to terminate and leave the protocol. Likewise, if the receiver sees a CAN when expecting an SOH (start of packet) it should terminate the file transfer. Many implementations now require two CAN characters before recognizing a cancel condition.

Error detection and recovery are the primary purposes of the Xmodem protocol. The transmitter and receiver should continue to retry until 10 errors in a row have occurred. Some of the common error conditions are listed below:

If the sequence number does not match the complement sequence number, the packet must be discarded and a NAK sent to the transmitter.

If the sequence number is the same as the sequence number of the last packet received, the packet should be discarded and an ACK sent to the transmitter.

If the sequence number matches the complement sequence number and it is neither the expected sequence number nor the last sequence number, the receiver should send two CAN characters and leave the Xmodem protocol (e. g. abort the file transfer).

When expecting data, if 10 seconds ever pass without receipt of a character, the receiver should send another NAK. This should be repeated 10 times. Some implementations will timeout after 10 seconds waiting for the first character of a packet, SOH, and then reduce the timeout for characters in a packet. The timeout should not go below 5 seconds if the protocol is to be used with public data networks.

In the original protocol, the transmitter would wait 10 seconds for an ACK, NAK or CAN and then retransmit the last Xmodem packet as if a NAK had been received. Most implementations either have the transmitter wait for a very long time (30 seconds to a minute) and then terminate the file transfer if an ACK, NAK or CAN has not been receive or wait about 30 seconds and retransmit the last packet.

Since extraneous characters are frequently generated when using asynchronous communications, the receiver should not count on receiving exactly 132 characters for each Xmodem packet. One algorithm for re-synchronization goes as follows:

• Assume that the checksum algorithm will cause re-transmission of Xmodem packets which contain extraneous characters.

• If the character received when expecting the start of a packet is not a SOH then ignore the character and continue to search for a SOH.

• Once a SOH is found, assume that the next two characters will be a valid sequence number and complement. If they are complements then assume that the packet has begun. If they are not complements, continue to search for a SOH.

• Send a NAK if a timeout occurs while attempting to re-synchronize (e.g. continue to process timeouts as described above).
• If no re-synchronization occurs within 135 characters then send a NAK character and retry receiving the packet.

When the receiver sees an EOT (which was not sent by the transmitter, but generated out of a communications error) instead of a SOH character, the receiver assumes incorrectly that the complete file has been transmitted. This is typically an unrecoverable error and it does occur especially when the transmitting and receiving UARTs are clocked slightly differently. An algorithm to detect false EOT might return a NAK for the first EOT received and only assume true end of transmission after receiving two EOT's.

        Transmitter                   Receiver

        [last block .. ]    >>>>>
                            <<<<<     [ACK]
        [EOT]               >>>>>
                            <<<<<     [NAK]
        [EOT]               >>>>>
                            <<<<<     [ACK]

Just in case the transmitter was not prepared to resend the EOT, it might be wise to set the timeout to about 3 seconds and retransmit the NAK up to 3 times and then issue a warning message but assume end of transmission.

Some Xmodem implementations will terminate on a single CAN character. Occasionally a CAN character will be generated by a communications error and if this occurs and is seen by the receiver between packets or is ever seen by the transmitter, the file transfer will be incorrectly canceled. Many implementations now require two CAN characters in a row before assuming that the file transfer is to be aborted.

CRC Xmodem is very similar to Checksum Xmodem. The protocol initiation has changed and the 8 bit checksum has been replaced by a 16 bit CRC. Only theses changes are presented.

One of the earliest and most persistent problems with Xmodem were transmission errors which were not caught by the checksum algorithm. Assuming that there is no bias in asynchronous communications errors, we would expect that 1 out of every 256 erroneous complete or oversized Xmodem packets to have a valid checksum. With the same assumption, if the checksum were 16 bits, we would expect 1 out of every 65,536 erroneous complete or oversized packets would have a valid checksum.

Considerable theoretical research has shown that a 16 bit cyclical redundancy check character (CRC/16) will detect a much higher percent of errors such that it would only allow 1 undetected bit in error for every 10^14 bits transmitted. That's 1 undetected error per 30 years of constant transmission at 1 megabit per second. However, my personal experience indicates that something around 10^9 to 10^10 is more realistic. Why such a vast improvement over the checksum algorithm? It is caused by the unique properties that prime numbers have when being divided into integers. Simply stated, if an integer is divided by a prime number, the remainder is unique. The CRC/16 algorithm treats all 1024 data bits in an Xmodem packet as an integer, multiples that integer by 2^16 and then divides that 1040 bit number by a 17 bit prime number. The low order 16 bits of the remainder becomes the 16 bit CRC.

The 17 bit prime number in CRC Xmodem is 2^16 + 2^12 + 2^5 + 1 or 65536 + 4096 + 32 + 1 = 69665. So calculating the CRC is simple, just multiply the 128 byte data number by 65536, divide by 69665 and the low order 16 bits of the remainder are the CRC. The only problem is, I've never seen a computer which has instructions to support 130 byte integer arithmetic! Fortunately for us, Seephan Satchell, Satchell Evaluations, published a specification a very efficient algorithm to calculate the CRC without either 130 byte arithmetic or bit manipulation. Appendix A contains the source code, in IBM/PC BASIC, for the calculation of a CRC.

Other methods of calculating CRC's for Xmodem involve bit level logic.¹⁾

The initiation of CRC Xmodem was designed to provide for automatic fall back to Checksum Xmodem if the transmitter does not support the CRC version.

The receiver requests CRC Xmodem by sending the letter C (decimal 67) instead of a NAK. If the transmitter supports CRC Xmodem, it will begin transmission of the first Xmodem packet upon receipt of the C. If the transmitter does not support CRC Xmodem, it will ignore the C. The receiver should timeout after 3 seconds and repeat sending the C. After 3 timeouts, the receiver should fall back to the checksum Xmodem protocol and send a NAK.

This section assumes that the reader is already familiar with Xmodem and CRC Xmodem presented above.

First, Xmodem provided the basic file transfer function, then CRC Xmodem improved the data integrity, now we come to WXmodem which provides better cost/performance.

A few people began discussing improvements to Xmodem with me in late 1985, over time we developed the following criteria:

1. The protocol must be as similar as possible to the XMODEM originally developed by Ward Christensen. The popularity of XMODEM, I believe, is based on its conceptual simplicity. More software writers are familiar with this protocol than any other. More files are transferred everyday by this protocol than any other asynchronous protocol. Simplicity here implies a limited number of rules for timing, error recovery and initiation.
2. The protocol must overcome the propagation delay that is characteristic of the public data networks. While the cost of long distance communication is 50 to 90% less via the public data networks than via the public voice networks, the propagation delays inherent in public data networks both reduces the cost savings and increases the aggravation that occurs while watching a computer slowly perform a file transfer.
3. The protocol must overcome the flow control problems which are characteristic in many public data network situations. Basically, in most situations, the X-On and X-Off characters must always be used for flow control and only for flow control when using public data networks.
4. The protocol should improve error recovery by simplifying the manner in which a programmer can determine the beginning of an XMODEM block. Since the Start of Header character (SOH) can appear in the data or CRC, the programmer must use a relatively sophisticated method to determine if a SOH actually represents the beginning of a XMODEM block.

This protocol provides special public data network support for non-X.25 hosts and PC-Pursuit access to bulletin boards. In order to accomplish this, the transmitter is not permitted to transmit the X-On and X-Off characters in the Xmodem packets. The reason for this restriction is simple:

• By the very nature of X.25 public data networks, without flow control, buffer overruns and lost data are inevitable from time to time at any baud rate.

• To avoid data loss public data networks must always assume that any X-Off and X-On character is a flow control character when supporting PC-Pursuit for bulletin boards and when supporting non-X.25 host systems.

Since many non-X.25 hosts, bulletin boards and communications programs use X-On and X-Off as flow control characters, public data networks must support those X-Off and X-On requests at the point of connection where the X-Off is received by the public data network. Otherwise, as many as several hundred characters backed up in the network would be transmitted by the public data network before the X-Off used for flow control reached the transmitter. The public data network has no way to know whether an X-On/X-Off protocol or Xmodem is operational at any point in time. Therefore a Xmodem packet which contains an X-Off character and no succeeding X-On character will cause the public data network to stop forwarding the ACK or NAK.

In addition, error recovery requires sophisticated programming for the receiver to determine the start of an XMODEM packet. This protocol simplifies this task by dedicating a special character as an indicator that an XMODEM packet is about to begin. The SYN (synch, decimal 22) character is used for this purpose. The presence of one or more SYN characters in a data stream always indicates that the next non SYN character is the beginning of an XMODEM packet (e.g. SOH).

The method used here to handle these situations is through the insertion of the DLE character (H010, decimal 16, data link escape character) as an indicator that the character following the DLE is in fact a modified DLE, SYN, X-On, or X-Off character.

1. Whenever an X-On, X-Off, SYN or DLE character is about to be transmitted as any part of an actual XMODEM packet including the CRC, the transmitter will transmit instead a DLE character followed by the original character which has been modified by exclusive or'ing it with 64 to its value. 1
2. The inserted DLE characters are not counted in the 128 byte data length of the data portion of the XMODEM packet. Indeed, it would be possible to have a packet which is physically 264 bytes in length because the Xmodem block sequence number (or its complement), all of the 128 data characters and two CRC characters are all either X-On, X-Off, SYN or DLE characters.
3. Neither the DLE nor the adjusted characters are used in the CRC calculation, rather the original character is always used in the CRC calculation.
4. When the receiver sees a DLE character, it does not count it in the XMODEM block length calculation, nor compute it in the CRC calculation but discards it and then remembers to exclusive or the next character with 64 and to verify that the result character is either a DLE, SYN, X-On or X-Off (the receiver will reject the packet unconditionally, if not one of those four characters) and then include the character as part of the packet.
5. Prior to transmission of a XMODEM packet, the transmitter will send one or more SYN characters (recommend two) as a positive indicator to the receiver of the beginning of a Xmodem packet.2
6. Except for the character received after a DLE, the receiver will test each incoming character to see if it is a SYN character. If it is, it will discard the character and assume that the next character will be another SYN or SOH. If a SYN character is received in the middle of a packet, the receiver will NAK that packet. The purpose of the SYN character is to simplify recognition of the beginning of a XMODEM packet by the receiver. Once an out of synch condition occurs on incoming data, the receiver can just ignore every incoming character until it sees a SYN. Existing XMODEM code which already properly deals with this situation could just always discard any SYN character at time of receipt with no further action.
7. The transmitter must support flow control characters (X-On, and X-Off) during transmission of packets. Upon receipt of an X-Off it will wait 10 seconds for an X-On and will start transmission again after 10 seconds or an X-On is received, whichever occurs first. Any extraneous X-On characters received by the transmitter will be ignored and discarded. (Note that this does NOT apply to X.25 host computers which use X.25 L2 and L3 windows for flow control.)

An initial handshake is provided to permit the receiver to indicate to the transmitter whether it can support checksum Xmodem, CRC Xmodem, or Windowed Xmodem:

1. WXMODEM - The receiver will send a character W (decimal 87) and wait 3 seconds for the beginning of a Xmodem packet. This will be repeated 3 times and then the receiver will drop down to CRC Xmodem.
2. CRC XMODEM - The receiver will send a character C (decimal 67) and wait 3 seconds for the beginning of a Xmodem packet. This will be repeated 3 times and then the receiver will drop down to Checksum Xmodem.
3. Checksum XMODEM - The receivers will send a NAK and wait up to 3 seconds for the beginning of a Xmodem packet. This will be repeated 4 times and if no valid SOH is received, the receiver will abort the file transfer request.

In order to overcome the propagation delays inherent with public data networks such as Tymnet, Telenet, Datapac, IPSS, Transpac and dozens more, the protocol must permit the transmitter to send more than one packet before receiving an acknowledgement from the receiver. The number of packets that the transmitter will send before stopping transmission if an acknowledgement has not been received is called the “window”. WXmodem uses a window of 4 packets for several reasons. Most importantly, it uses a single set of timing rules which would deal reasonably well with a wide range of baud rates (that implied keeping the window fairly small). Secondly, the window sequence number is directly related to the Xmodem packet sequence number which, hopefully, will simplify implementation of windowing. 

1. The window is always 4 Xmodem packets. That is, the transmitter will send 4 unacknowledged packets. Transmission will not cease and the time out interval will not begin until 4 unacknowledged packets have been transmitted. Note that the window may be less than 4 Xmodem packets for short files or at end-of-file.

2. The receiver will transmit acknowledgements in the form:

 ACK[sequence]

The [sequence] field is an 8 bit number where the high order or most significant 6 bits are always zero and the low order or least significant 2 bits are always the same as the low order 2 bits of the XMODEM block sequence number of the XMODEM packet being acknowledged (value in decimal may range from 0 to 3).

3. The receiver does not have to acknowledge every packet, but must acknowledge at minimum every fourth packet. The transmitter will accept one ACK[sequence] for multiple XMODEM packets. For example, after an unknown number of packets:

         Transmitter                             Receiver

         ....
         ....
         ....
         [Block Sequence Number H0FE]
         [Block Sequence Number H0FF]            ACK[H002]
         [Block Sequence Number H000]            ACK[H003]
         [Block Sequence Number H001]
         [Block Sequence Number H002]            ACK[H001]
         .....

Since some transmitters must close the window and cease all communications before doing disk I/O to read more data, it is suggested that acknowledgements be sent for every packet (except when the receiver can easily determine that another packet is already being received at the point in time that the ACK[sequence] is about to be sent).3

4. The receiver will reject a packet (request retransmission) by sending:

  NAK[sequence]

Where [sequence] is then next window sequence number (between H000 and H003) after the [sequence] of the last good block. The receiver will discard up to 3 Xmodem packets received after the NAK is transmitted until it receives the packet with the sequence number that had previously been nak'ed by the receiver. The receiver will not send a second NAK until another packet with the same sequence number is received which is also invalid or a timeout has occurred.

5. When the transmitter receives a NAK[sequence], it will complete transmission of any XMODEM block currently being transmitted and then begin re- transmission starting with the block which was nak'ed.

6. The receiver will discard duplicate packets but count them in the window for purposes of deter- mining the maximum receive window without an ACK in response. For example, if the receiver gets packet sequence number 127 four times in a row, it must send an ACK H003 even if the receiver has previously acked that block.

7. The timeout intervals at various points in processing are:

• Waiting for a character on receive, start of packet not yet recognized: 15 seconds
• Waiting for a character on receive, start of packet has been recognized: 15 seconds
• Waiting for an Ack or Nak on transmit side after the window has closed: 15 seconds4
• Waiting for an X-On after receipt of an X-Off by the transmitter: 10 seconds

8. When the transmitter times out waiting for an ACK or NAK when the window is closed (e.g. four blocks have been transmitted), the transmitter will retransmit the last block transmitted and wait again. Only after 10 consecutive timeouts have occurred will the transmitter cancel the transmission.

9. Where possible, it is recommended that the receiver return an ACK[sequence] for every packet or at least 50% of the Xmodem packets. When the receiver must wait for the window to close (e.g. receive 4 Xmodem packets without an acknowledgement), some performance benefit will be lost.

If the receiver cannot overlap disk I/O and communications I/O, the receiver can temporarily stop transmission by either:

1. “Closing the window” (e.g. receiving 4 blocks without sending an ACK[sequence]) performing the disk I/O and then sending an ACK[sequence].
2. Transmitting an X-Off followed by an X-On when the receiver is ready to resume accepting data. Note that the receiver should be prepared to accept data for about a 1/4 of a second after the X-Off is sent to cover situations where satellite propagation delay may occur. One possible implementation would let the computer user set the “X-Off delay time” so that in the normal case the X-Off delay could be set to 25 milleseconds. A sophisticated implementation might set the initial X-Off delay at 250 milleseconds and then reduce it based on experience during the file transfer.
3. Each approach has its advantages, but the X-Off approach will provide the best performance in most cases especially when using a public data network. Note, however, that some computers, notably the Commodore 64 and the IBM PC Jr cannot receive communications data while writing to disk.

Host computer systems which utilize the X.25 protocol (examples: People/Link, Delphi, CompuServe, The Source) to interface with the various public data networks may send special control packets which change the manner in which the network will communicate with the remote personal computer, bulletin board or terminal. For the purposes of this paper, it is assumed that the X.25 host can already support CRC and/or Checksum Xmodem and present only the changes for WXMODEM.

1. When an X.25 Host is the transmitter, it must be sure to set the distant X.3 PAD parameters to assure that the receiver can use X-Off/X-On for flow control. This is accomplished by sending a Q-Bit command packet to set X.3 parameter 12 to a 1 prior to the initial handshake. Note that if the receiver cannot support WXMODEM, the X.25 Host must send the appropriate Q-Bit packet to reset parameter 12 to a 0 before transmitting the first CRC or Checksum Xmodem packet.
2. When an X.25 Host is the receiver and in WXMODEM mode, it must be sure to set the distant X.3 PAD parameters to assure that the network will use X-Off/X-On for flow control between the network and the transmitter to prevent its buffers from overflowing. This is accomplished by sending a Q-Bit command packet to set X.3 parameter 5 to a 1 prior to the initial handshake.

The purpose of this appendix is to give non-technical and non mathematical software writers a cook book approach to calculating the CRC-16 used in Xmodem. We have half accomplished that goal. The BASIC code in the examples below has been tested on an IBM PC and found to work effectively even at 9600 with compiled Basic. Some BASIC languages do not offer an XOR function and others do not have MKI$ and CVI functions which simplified the movement of data between data types. Someday we hope to provide a Commodore C-64/C-128 implementation which simulates XOR, but not today!

My thanks go to Chuck Forsberg, Joe Noonan, John Byrns and Stephen Satchell. Without their help and public domain documents, this would have never been possible.

The Intel 8080 and upward has a feature, convenient only to some electrical engineer somewhere, which places 2 byte (16) bit integers in BYTE REVERSE order in memory. That is, the least significant byte is placed in memory before the most significant byte for integer operations. If A$ is one byte containing the number 52 and it is assigned to I% using the ASC function, the binary value (52) ends up in the first byte of I% and the second byte is zero.

                          Result

      I%=0                [x'0000']
      I%=1                [x'0100']
      A$="A"              [x'41']
      I%=ASC(A$)          [x'4100']
      B$=MKI$(I%)         [x'4100']  letter "A" then binary zero
      I%=CVI(CHR$(0)+A$)  [x'0041']
      A$=CHR$(65)         [x'41']

Once this is understood, many problems with these algorithms goes away.

The bit shift method here was converted from the “C” logic presented in Chuck Forsberg's “Xmodem/Ymodem” protocol reference and from an old IBM two page reference guide that Joe Noonan carries with him in his appointment calendar!

     /*
      * This function calculates the CRC used by the XMODEM/CRC Protocol
      * The first argument is a pointer to the message block.
      * The second argument is the number of bytes in the message block.
      * The function returns an integer which contains the CRC.
      * The low order 16 bits are the coefficients of the CRC.
      */
     int calcrc(ptr, count)
     char *ptr;
     int count;
     {
         int crc, i;

         crc = 0;
         while (--count >= 0) {
          crc = crc ^ (int)*ptr++ << 8;
          for (i = 0; i < 8; ++i)
              if (crc & 0x8000)
               crc = crc << 1 ^ 0x1021;
              else
               crc = crc << 1;
          }
         return (crc & 0xFFFF);
     }

But in IBM PC BASIC, our implementation looks like:

     100 DEFINT A-Z 'DEFAULT IS TWO BYTE INTEGERS
     2000 REM * V$ CONTAINS 133 CHARACTER COMPLETE XMODEM PACKET
     2010 REM * CRC$ IS TWO BYTE CRC WITH MOST SIGNIFICANT BYTE FIRST
     2020 CRC$=CHR$(0)+CHR$(0)                      'START AT ZERO
     2030 FOR I2=4 TO 131
     2040   A$=MID$(V$,I2,1)
     2050   GOSUB 4000
     2060 NEXT I2
     2070 REM * CRC$ CONTAINS CALCULATED CRC!

     3000 IF CRC$=MID$(V$,132,2) THEN ....    'IT'S GOOD!!!

     4000 REM * CRC BITWISE CALCULATION (WHAT A JOKE!)
     4010 CRCH1=ASC(LEFT$(CRC$,1)) XOR ASC(A$)
     4020 CRCL1=ASC(RIGHT$(CRC$,1))
     4030 FOR I3 = 0 TO 7
     4040   CARRY=0 : IF CRCH1 > 127 THEN CARRY=-1  'IS HIGH BIT ON IN CRC?
     4050   CRCH1=(CRCH1*2) AND 255                 'CRCH << 1 AND 255
     4060   IF CRCL1>127 THEN CRCH1=CRCH1+1 'IF CRCL CARRIES THEN INCR CRCH
     4070   CRCL1=(CRCL1*2) AND 255                 'CRCL << 1 AND 255
     4080   IF CARRY=0 THEN GOTO 4105               'IF HIGH BIT WAS ON,
     4090   CRCH1=CRCH1 XOR 16                      'XOR WITH &H1021
     4100   CRCL1=CRCL1 XOR 33
     4110 NEXT I3
     4130 CRC$=CHR$(CRCH1)+CHR$(CRCL1)
     4140 RETURN 'WHEW

That routine will execute 128 * 7 + 128 * 9 * 8 BASIC statements for each Xmodem packet or 10112 statements per Xmodem packet! It will work for low baud rates in compiled BASIC, but just is too much for interpretive BASIC.

This method is based on routine M4 in Steven Satchell's paper, “Test of CRC Routines for CRC-CCITT”, but has some very signifi- cant differences. A table of 256 CRC's, originally calculated with the bit shift method is used to avoid performing the bit shift during communications. The table contains the CRC's for each byte value from 0 to 255 when the original CRC is zero. The result of this calculation is included in the DATA statements in the code.

The comments are intended to show what is logically happening rather than physically. Because of the “byte reverse” nature of integers in the 8088, a logical shift of 8 bits to the left is a physical shift of eight bits to the right!

     200 DEFINT A-Z  'ALL INTEGERS
     210 DIM CRCTB(256)

     300 GOSUB 9000 'INITIALIZE CRC TABLES

     6200 REM * CRC CALCULATION USING TABLE METHOD, V$=XMODEM PACKET
     6210 CRC$=CHR$(0)+CHR$(0)                 'INITIALIZE TO ZERO
     6220 FOR Q=4 TO 131
     6230   CRCH1=ASC(LEFT$(CRC$,1))           'CRC >> 8 AND 255
     6240   CRCL2=CVI(CHR$(0)+RIGHT$(CRC$,1))  'CRC << 8 AND 255
     6250   CRC1$=MKI$(CRCTB(CRCH1 XOR ASC(MID$(V$,Q,1))) XOR CRCL2)
     6260   CRC$=RIGHT$(CRC1$,1)+LEFT$(CRC1$,1) 'SET IT BACK!
     6270 NEXT Q
     6280 IF CRC$ <> MID$(V$,N,2) THEN ....... 'GOTO ERROR ROUTINE
     6290 REM * END OF CRC CALC

     9000 FOR I%=0 TO 255 ' INITIALIZE CRC TABLE
     9010   READ CRCTB(I%)
     9020 NEXT I%
     9025 RETURN
     9030 DATA 0, 4129, 8258, 12387, 16516, 20645, 24774, 28903
     9040 DATA -32504,-28375,-24246,-20117,-15988,-11859,-7730,-3601
     9050 DATA 4657, 528, 12915, 8786, 21173, 17044, 29431, 25302
     9060 DATA -27847,-31976,-19589,-23718,-11331,-15460,-3073,-7202
     9070 DATA 9314, 13379, 1056, 5121, 25830, 29895, 17572, 21637
     9080 DATA -23190,-19125,-31448,-27383,-6674,-2609,-14932,-10867
     9090 DATA 13907, 9842, 5649, 1584, 30423, 26358, 22165, 18100
     9100 DATA -18597,-22662,-26855,-30920,-2081,-6146,-10339,-14404
     9110 DATA 18628, 22757, 26758, 30887, 2112, 6241, 10242, 14371
     9120 DATA -13876,-9747,-5746,-1617,-30392,-26263,-22262,-18133
     9130 DATA 23285, 19156, 31415, 27286, 6769, 2640, 14899, 10770
     9140 DATA -9219,-13348,-1089,-5218,-25735,-29864,-17605,-21734
     9150 DATA 27814, 31879, 19684, 23749, 11298, 15363, 3168, 7233
     9160 DATA -4690,-625,-12820,-8755,-21206,-17141,-29336,-25271
     9170 DATA 32407, 28342, 24277, 20212, 15891, 11826, 7761, 3696
     9180 DATA -97,-4162,-8227,-12292,-16613,-20678,-24743,-28808
     9190 DATA -28280,-32343,-20022,-24085,-12020,-16083,-3762,-7825
     9200 DATA 4224, 161, 12482, 8419, 20484, 16421, 28742, 24679
     9210 DATA -31815,-27752,-23557,-19494,-15555,-11492,-7297,-3234
     9300 DATA 689, 4752, 8947, 13010, 16949, 21012, 25207, 29270
     9310 DATA -18966,-23093,-27224,-31351,-2706,-6833,-10964,-15091
     9320 DATA 13538, 9411, 5280, 1153, 29798, 25671, 21540, 17413
     9330 DATA -22565,-18438,-30823,-26696,-6305,-2178,-14563,-10436
     9340 DATA 9939, 14066, 1681, 5808, 26199, 30326, 17941, 22068
     9350 DATA -9908,-13971,-1778,-5841,-26168,-30231,-18038,-22101
     9360 DATA 22596, 18533, 30726, 26663, 6336, 2273, 14466, 10403
     9370 DATA -13443,-9380,-5313,-1250,-29703,-25640,-21573,-17510
     9380 DATA 19061, 23124, 27191, 31254, 2801, 6864, 10931, 14994
     9390 DATA -722,-4849,-8852,-12979,-16982,-21109,-25112,-29239
     9400 DATA 31782, 27655, 23652, 19525, 15522, 11395, 7392, 3265
     9410 DATA -4321,-194,-12451,-8324,-20581,-16454,-28711,-24584
     9420 DATA 28183, 32310, 20053, 24180, 11923, 16050, 3793, 7920

This method uses 128 * 6 BASIC statements per Xmodem packet or a miserly 768 BASIC statements per packet. And, if you want, the code can be tightened still more. Unfortunately, any further tightening that we could see would eliminate most of the already limited readability of the code. 

Please add your notes and comments here or send them to me and I'll get them added to the current copy on People/Link.

1. This was originally set up to ADD 32 to the character on transmit and SUBTRACT 32 on receive. By using exclusive or with 64, the logic is the same on transmit and receive.
2. The use of the SYN character was added at the request of several people who have coded Xmodem routines and have struggled valiantly to improve their error recovery routines. Peter Boswell 6/10/86
3. The suggestion that ACK[sequence] be sent for every block received was added. Peter Boswell 6/10/86
4. The original value for the ACK/NAK timeout was 10 seconds. This was changed to 15 seconds the situation where the receiver is operating at 300 baud and using X-Off to stop receipt of characters during disk I/O. Peter Boswell, 6/10/86

 XMODEM and its derivatives have become the primary method for file transfer for personal computers and is a popular error recovery type protocol. Before learning more about Xmodem, it is important to hear what its author has to say:

    "It was a quick hack I threw together, very unplanned (like
    everything I do), to satisfy a personal need to communicate
    with some other people.  ONLY the fact that it was done in
    8/77, and that I put it in the public domain immediately,
    made it become the standard that it is"....."People who
    suggest I make SIGNIFICANT changes to the protocol, such as
    'full duplex', 'multiple outstanding blocks', 'multiple
    destinations', etc etc don't understand that the incredible
    simplicity of the protocol is one of the reasons it survived
    to this day in as many machines and programs as it may be
    found in!"

Ward Christensen, quoted from a message posted on CompuServe in 1985. Edited by Chuck Forsberg, “X/Ymodem Protocol Reference”, unpublished, 10/20/1985.

The protocol is Asynchronous, 8 data bits, no parity bit, one stop bit.

• Reference Library