CW Operating Practice and Network Protocols: The Same Problems, Solved by Ear

Posted on 2026-11-06
TL;DR
A CW contact needs framing, turn-taking, acknowledgement and channel access, just like a network protocol. Annotating three of my own QSOs against ITU, ARRL and networking sources shows how old these solutions are and where the comparison stops working.

When I learned CW, I learned characters. What I did not expect to learn was how much procedure sits around them. Every contact follows a script: who calls, who answers, who may talk when, how to ask for a repeat, how to say goodbye. In Germany this is a subject of its own in the amateur radio exam, called Betriebstechnik, roughly operating practice1.

Working as a software engineer, I kept recognizing things. Framing, turn-taking, acknowledgements, retries, addressing: these are the problems a network protocol has to solve. I have not come across a comparison of CW operating practice with network protocols, though my search was limited. So here is my attempt, built on three contacts from my notes and on what I could verify in the original sources.

This is not a claim that CW is a protocol stack, and certainly not that it is an ancestor of TCP/IP. The claim is smaller: the same problems keep coming back, whatever the medium, and operators solved them with conventions long before computers did.

A pileup is ALOHA by ear

When a rare station calls CQ, dozens of operators answer at the same time. They cannot hear each other, because each of them is transmitting while the others are. The station at the center hears a mess, picks out a few letters of one callsign and sends them back with a question mark. Everyone else stays quiet and tries again.

That is close to how ALOHAnet worked at the University of Hawaii in the early 1970s: transmit without listening, the hub acknowledges what it received, and a sender without acknowledgement waits a random time and tries again2. The IARU operating manual explains why listening is not enough before you call CQ: you may not be able to hear the station already using the frequency, because of propagation. So you ask, by sending QRL? at least twice, a few seconds apart, and silence counts as permission3. Networking calls the underlying problem the hidden terminal problem. Operators call the remedy “QRL?”.

Two short contacts

On 6 August 2026 I worked LZ1XV in Bulgaria on 14.0217 MHz from the club station in Hamburg. Reports were 569 sent and 599 received, at 20:04 UTC. This is what I copied down of the other side:

OP VAL QTH in Botevgrad botevgrad hw? dl2xu de lz1xv
OK dl2xu de lz1xv ok ben fb sigs from hamburg = QSR 73 best dx
es hpe cuagn gb l dl2xu de lz1xv

These are his two turns, and my own are missing from my notes. The first gives name and place, repeats the place, and hands over with hw? and the two callsigns. The second acknowledges, says goodbye and signs. Both turns end with the two callsigns, so the address travels with every turn. My note says QSR before the goodbye, which is a different Q code with a different meaning, so I read it as QSL, “I acknowledge receipt”.

On 23 August I worked I5ABV, who signed IA5/I5ABV, on 14.0345 MHz at 17:50 UTC. I sent 589 and received 449:

lbr is, vr rst 449 dm0reqse more qsb hw? bk
fb dr ben in hamburg fb tnx fer rst es qso vy 73 es all best ciao
dl2xu de ia5/i5abv 73

Here the signal is fading (qsb), and the operator says so, reports it and hands the turn back with bk. Part of my note is unclear to me today, and I will not guess at it.

One contact, line by line

The third contact is the one I annotate in full. Fifteen minutes after LZ1XV, at 20:19 UTC, I worked Manuel, CT1CBP, in Portugal on 14.040 MHz. I sent 539 and received 559. He ran 500 watts into an end-fed antenna. The table lists what I wrote down of his side. My own turns are missing from my notes, and I will not reconstruct them. That fits the medium: only one station transmits at a time, so my notes hold exactly one side.

Heard What it does Network counterpart
Manuel viana His name and, I assume, his place A fixed header field, filled in
my pwr is 500 5TT ant is end fed Power and antenna Optional header fields. 5TT is 500 as a cut number, with T for zero
= Separates two parts of the transmission (BT) Delimiter. ITU-R M.1172: “Signal to mark the separation between different parts of the same transmission”
wx is clr es temp 23C Weather Free-form payload
dl2xu de ct1cbp Called station, “from”, sending station Destination and source address in every single turn. ITU-R M.1172: DE is “From …”
ok ben Acknowledges my last turn, addressing me by name An ACK that says nothing about whether the content was understood
QSL ok via buro Confirms the contact, card follows via the QSL bureau A delayed confirmation through a second channel
tnx fer fb qso gb gl es 73 73 73 hpe cuagn Thanks, goodbye, good luck, best regards, hope to meet again Politeness, but also an announcement that the end is near
+ End of transmission (AR) End of a frame. ITU-R M.1172: “End of transmission”
dl2xu de ct1cbp sk tu End of contact, thank you End of the session. SK is amateur usage; the ITU texts have VA (“End of work”) instead

Two fragments of this contact, sp sg ee and rb call, I can no longer read, so they are not in the table.

The rules were written down early

None of this is improvised. The Allied military radiotelegraph procedure manual ACP 124 (in the 1983 edition I read) reads like a protocol specification: “Every transmission shall end with either the prosign K (invitation to transmit) or the prosign AR (end of transmission).”4 The amateur version is older than the ITU-R texts I could read. In 1926 the first edition of the ARRL’s Radio Amateur’s Handbook said that “the proper use of AR, K, and SK signals is required for uniformity in understanding what the other operator intends to do.”5 Use AR between messages, K at the end of each turn, SK only when signing off.

That is the difference between the end of a frame and the end of a session, and my notes show both: the + after Manuel’s goodbyes, then sk after the callsigns. In the ITU texts the end of the session is not SK but VA (“End of work”), sent by both stations.6 SK is amateur usage, and the 1946 ARRL manual allows either: “SK (or VA) shall be used by each station when signing off”.7

Listening before transmitting is older than amateur radio, too. The 1912 London convention told a station, before calling, to “make sure that no other correspondence is being carried on within its radius of action; if it finds otherwise, it shall wait for the first pause”.8 The 1939 ARRL manual added retry limits that read like a timeout policy: a call should not repeat the other station’s callsign more than three times, a CQ not more than five times without identification, and afterwards “the dial should be covered thoroughly for two or three minutes looking for replies”.9

The gaps are the protocol

Underneath all of this, Morse code is a binary signal with durations: a dash lasts three dots, the gap inside a letter one dot, the gap between letters three dots, and the gap between words seven dots. The letter gap is not decoration. E is a single dot and I is two dots, so without the gap a receiver cannot tell E E from I. Silence is the framing layer.

Even this was settled by a standards body. The international telegraph regulations of Cairo 1938 define the space between two words as “five dots”, and the Paris regulations of 1949 as “seven dots”.10 The current ITU-R recommendation repeats seven.11 Both texts are regulations for wired telegraphy, so I cannot say how radio operators handled the change. What I can say is that no sender ever holds the gaps exactly, and the receiver adapts to the sender’s rhythm by ear. A network engineer would call that clock recovery.

A prosign breaks the rule on purpose. The military manual defines the notation: “An overscore (a line over two or more letters) indicates that the letters overscored are to be transmitted as a single character; that is, without pause between letters.”12 Letters that are sent without the usual gap become one control symbol. Line codes in networking do something similar when they use a pattern that cannot occur in ordinary data.

A check borrowed from the telegraph

The most surprising thing I found is about messages, not about contacts. Amateurs relay formal messages called radiograms through traffic nets, and every radiogram carries a check: the number of words in the text. It is not an error-correcting code. It is a count, and it exists so that the receiver can tell whether a word was lost along the way.

Where does it come from? In August 1916, the ARRL’s QST printed a proposal for a standard message form by Lindley Winser: “It is my intention to take the methods used in commercial work and adapt them to the needs of amateur work.”13 His example reads hr nr 1 ck 16 losangeles 820 pm 3rd via 6zw to mr john jones, and for the case where operators disagree on the check, he proposes “for the sending operator to transmit the first letter of every word counted and let the receiving operator compare.” That is not his invention. The international telegraph regulations of 1885 contain the same recovery procedure: the receiving clerk reports a difference in the word count, and if the sender cannot explain it as a slip in the announced number, “he repeats the first letter of each word until he comes to the passage omitted, which is then inserted.”14

In protocol terms, this is a selective repeat built on top of a very weak error detector. The count alone tells you that something is missing, and the first letters tell you where. It cannot detect a wrong letter, and it was never meant to.

The header that carried the check was revised like a protocol header. In January 1917 QST showed a message that began Hr msg No. 1 ck 9 radio Los Angeles filed 10:15 p. m. 23rd. To ..., so the check came right after the number and the destination came last.15 A reader complained in June of the same year about stations that “become unreadable, either from fading or QRM, just as they get to the ‘To:’” and asked for the Western Union preamble instead.16 In the 1926 Radio Amateur’s Handbook the same kind of header ends with the check (NR 88 217p MAY 8 CK 51), and by 1939 the ARRL manual gave the order as “NR-STN-CK-PLACE-TIME-DATE”.17 In the same manual the League chose the simpler land-line count over the cable count used in marine work, “this simplification being the reason for its adoption”.18 The ARRL’s own wording for a simpler header is the kind of sentence you find in a changelog.

Where radio hams met TCP

Amateurs did not only borrow from the telegraph, they also fed back into networking. Phil Karn, KA9Q, and Craig Partridge described in 1987 how to estimate round-trip times for TCP when a segment has been retransmitted, which is now called Karn’s algorithm. In the paper they write that “a TCP implementation using Karn’s algorithm and Mills’ nonlinear filter has been in heavy use on perhaps the worst medium ever used to pass IP datagrams: amateur packet radio”, with packet loss rates “often exceeding 50%”.19 Three years later Karn presented MACA at an ARRL computer networking conference, a scheme for channel access that tackles the hidden terminal problem20. I did not verify here how directly it led to the RTS/CTS mechanism in Wi-Fi.

The digital mode FT8 turns the CW contact into a state machine. Its designers describe a standard QSO as a sequence of six messages, starting with CQ K1JT FN20 and ending with 73, and give it a formal description with six states.21 The paper says most QSOs follow this sequence or a close variant, and that “in normal usage, failure to decode a response from a QSO partner implies repetition of the previous transmission.” That is stop-and-wait. The paper also records a design decision I like: WSJT-X requires that every QSO be initiated by a human operator.

Where the comparison breaks

I do not want to oversell the parallel.

  • A human fills the gaps. Context, expectation and knowing the other operator recover most errors, and no specification captures that. I could not read two fragments in my own notes of Manuel’s contact, but the contact worked, because both of us understood what was going on.
  • There is almost no integrity checking. The radiogram check counts words. It does not protect letters.
  • The Q codes are conventions, not enforced rules. QRL? asks whether a frequency is in use, and the other station may simply not answer.
  • There is no shared state machine. Two operators can disagree about whether the contact is over without noticing.
  • The layers are not separable. A prosign mixes framing, session control and application meaning in one or two characters.
  • The vocabulary moved over time. “Are you busy?” was QRW in the 1909 and 1912 lists and became QRL only in 1927, when QRL’s old meaning, “are you receiving badly?”, was dropped.22

I think the more interesting reading is not that CW is a primitive protocol, but that a community of human operators converged on the same solutions as protocol designers when they faced the same constraints: a shared channel, no clock, errors, and no way to ask for help.

Footnotes


  1. The German amateur radio exam has a part called Betriebstechnik und Vorschriften, operating practice and regulations: Amateurfunkprüfung on Wikipedia. The topic has its own literature in German.↩︎

  2. ALOHAnet began at the University of Hawaii in 1968, and the first packet broadcast unit ran in June 1971. A sender without an acknowledgement waited a random time and retransmitted: ALOHAnet on Wikipedia. I am describing a similarity of problems here, not a family tree.↩︎

  3. IARU, Ethik und Betriebshinweise für den Funkamateur, German edition 2010, hosted by the DARC, section II.9.18 and the CW section before it: PDF. The manual gives the reason in parentheses: one may not be able to hear the transmitting station “aufgrund der Ausbreitungsbedingungen”. I paraphrased the German text.↩︎

  4. ACP 124(D), Communication Instructions: Radiotelegraph Procedure, October 1983, section 222, archive.org copy.↩︎

  5. The Radio Amateur’s Handbook, 1st edition, ARRL 1926: archive.org copy.↩︎

  6. ITU-R Recommendation M.1172, Miscellaneous abbreviations and signals to be used in radiotelegraphy communications, 1995, which lists VA as “End of work”, and ITU-R M.1170 on Morse procedures. Neither text contains SK: M.1172, M.1170.↩︎

  7. Operating an Amateur Radio Station, ARRL, 25th edition 1946: archive.org copy.↩︎

  8. Service Regulations of the International Radiotelegraph Convention, London 1912, Article XXIV, section 3, reprinted in the American Journal of International Law 7 (1913): archive.org copy.↩︎

  9. Operating an Amateur Radio Station, ARRL, September 1939: archive.org copy.↩︎

  10. International Telegraph Regulations, Cairo 1938, paragraph 241, and Paris 1949, paragraph 253, from the ITU history portal: Cairo 1938, Paris 1949.↩︎

  11. ITU-R Recommendation M.1677-1, International Morse code, 2009: ITU-R M.1677-1.↩︎

  12. ACP 124(D), section 111, see the first reference to it above.↩︎

  13. Lindley Winser, “Suggestions on Message Form and Transmission”, QST, August 1916: scan at worldradiohistory.com. I read the scan’s text layer, so some letters in the original OCR are rough.↩︎

  14. International Telegraph Regulations, Berlin 1885, Regulation XXXIX, from the ITU history portal: 1885 regulations. The OCR prints “cornes” for “comes”.↩︎

  15. QST, January 1917, at worldradiohistory.com, in the same series as the August 1916 scan.↩︎

  16. QST, June 1917, in a reader letter seconding a request by 8AEZ. I did not find 8AEZ’s original letter.↩︎

  17. The 1926 handbook and the 1939 manual cited above. The 1926 example is NR 88-217p MAY 8 CK 51.↩︎

  18. Operating an Amateur Radio Station, ARRL, September 1939, section on the land line check. The 1946 edition repeats the rule.↩︎

  19. Phil Karn and Craig Partridge, “Improving Round-Trip Time Estimates in Reliable Transport Protocols”, ACM SIGCOMM 1987, available from Karn’s site at ka9q.net/papers/. I quote the text version of the paper.↩︎

  20. Phil Karn, “MACA - A New Channel Access Method for Packet Radio”, 9th ARRL Computer Networking Conference, London, Ontario, 1990.↩︎

  21. Steve Franke (K9AN), Bill Somerville (G4WJS) and Joe Taylor (K1JT), “Motivation and design of the digital modes FT4 and FT8”, QEX, July/August 2020: PDF. The formal description has six states, from Calling to Signoff.↩︎

  22. Q-code lists of the British Post Office handbook (October 1909, via a web archive copy of telegraph-office.com), the London convention of 1912 (see above) and the Washington convention of 1927 (archive.org, item mracInternationalRadiotelegraphConventionProceedingsWashingtonDC1927-11-25). The 1927 list is a scan where the codes and questions are in separate columns, so I paired them by order.↩︎

Portrait of the blog's author. Dude with full beard and short hair.
Author: Benjamin Brunzel I'm a software engineer based in Weilheim i.Ob., Germany. If you want to get in touch contact me in the fediverse