Comments on “Document B08”
(as presented to the IARU Region 1 VHF/UHF/Microwave Committee)

Document B08, prepared for the Interim Meeting in Vienna on February 24–25 2007, asserts that the digital protocol known as JT65 fails to meet accepted minimum standards for valid QSOs. The document proposes that:

"The IARU Region 1 should decide on minimum requirements for what is considered to be a valid digital QSO. This decision should also include the report system, i.e., what is normally called “unknown QSO information.”
The decoding process should at all times allow any randomly formatted text message to be decoded. This would be true when the information represent calls and reports, formatted as existing and well established radio amateur QSO procedures dictate. Injecting known information to the decoding process should not be allowed for a valid digital QSO. Anyone following the contact should be able to decode the same information, without having the information present on their computer.

The proposal seeks to impose standards for “digital mode” QSOs very different from those used for traditional modes like CW and SSB. It offers no justification for such a distinction; moreover, its supporting arguments are flawed and readily disproved.

Background

Every method of communication suffers degraded performance when the signal-to-noise ratio (SNR) is low. This is true for all forms of amateur radio voice communication, for “ear-and-brain” CW, and for computer-assisted “digital modes.” Whatever the form of coding and modulation, if the SNR is high enough then copy can be perfect, or nearly so. When SNR is low the rate of information transfer decreases, approaching zero at the threshold of signal detectability.
Operators using traditional weak-signal modes such as CW and SSB quickly learn that at low SNR it is helpful to format transmissions in standard ways and to include redundant information in every one. The simplest form of redundancy is repetition, so it is not surprising that when signals are marginal, wise operators include many repetitions of all essential information.

The JT65 protocol

 Modern communication technologies use structured redundancies that are much more powerful than simple repetition. The JT65 protocol uses a Reed-Solomon code optimized for the needs of amateur radio weak signal communication, together with a form of modulation (multi-tone frequency shift keying) known to be much more efficient than simple On-Off keying. JT65 has been implemented in a computer program called WSJT, of which I am the principal author. Standard JT65 transmissions convey exactly 72 bits of arbitrary “user information.” As a consequence, any one of 272 4.7 × 1021 distinct messages can be conveyed in a single transmission. The 72 user bits are augmented with an additional 306 bits of mathematically encoded redundancy; the redundant symbols are created in such a way that the exact transmitted message can be decoded, with extremely small probability of error, even if many symbols are corrupted or lost in the noise during transmission. Rather than being transmitted character-by-character, as in Morse code, message information is mathematically spread throughout a whole transmission. Signal dropouts do not cause the loss of isolated portions of a message; JT65 messages are copied in their entirety, or not at all. If enough channel symbols are received with adequate SNR, copy is complete and error-free, with very high confidence. If not, the decoder produces no result and a repeat transmission is required.

Prior information and the JT65 decoder

Experienced weak-signal operators know that with marginal signals it is much easier to recognize and copy one’s own callsign (or a familiar one) than unknown calls or random characters. Exactly analogous distinctions apply for the JT65 decoder implemented in WSJT. A fully general algorithm reliably decodes any JT65 message down to an SNR limit (1) of about −24 dB. In addition, WSJT offers a secondary decoder that yields reliable copy of some signals about 4 dB weaker. This “deep search” decoder is not sensitive to the full range of 2⁷² possible messages; instead, it is programmed to determine specifically whether one of a large number of hypothetical messages was the exact message transmitted. Hypothetical messages are generated with the help of a callsign database maintained by the user: calls found there are combined with “CQ”, with the receiving station’s own callsign, and with optional numerical signal reports. With the default database of more than 4800 callsigns known to have been active in VHF weak-signal communication, this procedure yields more than 14,400 hypothetical messages. If one of the hypothetical messages matches the transmitted one in every detail, that message can be decoded with high confidence down to about −28 dB SNR. The slightest difference between hypothetical and received messages — for example, a single-character having been changed or omitted — will cause the deep-search decoder to reject the hypothesis and produce no result. Again, it should be emphasized that the fully general decoder will decode any JT65 message whenever the SNR is adequate. Slightly weaker signals can be decoded if the computer is given some information about the most plausible and interesting message contents. The situation is really no different than with human decoding of traditional-mode signals.

(1 Note: WSJT’s reference bandwidth for SNR measurements is 2500 Hz)

False assertion.

Proposal B08 asserts that the JT65 decoding process:

“. . . is comparing fragments of information, matching this with known calls and locators from a database, reconstructing and then printing the full information on the screen as if it had been received via the airwaves.”

This statement is false, as can be easily confirmed in a number of ways. Perhaps most fundamentally, the source code for WSJT is openly available.2 Anyone can examine the code, compile it for him- or herself, and test it — as many interested amateurs have done.The deep search algorithm is wholly contained in 155 lines of straightforward, easy-to-read code. It contains no “comparing” or “matching” of “fragments of information.” Instead,
hypothetical messages are encoded in their entirety, just as they would be for transmission. The entire received signal is then correlated against every one of the hypothetical encodings. The structured redundancy built into the 272 possible Reed-Solomon codewords ensures that any difference between a transmitted message and a hypothetical test message, no matter how small, ensures that at least 52 of the 63 six-bit channel symbols will be different and the correlation will be small. As a consequence, the procedure can confidently determine whether a hypothetical test message is (or is not) identical to the message transmitted — even at very low SNR.
The deep-search algorithm is well understood mathematically and is highly reliable. Its rate of “false positives” is extremely low — certainly lower than that achievable by a skilled CW operator under marginal conditions, even if the CW signals are many dB stronger than the JT65 signals. This fact is the result of the improved efficiencies made possible by using a code with strong “forward error correction,” together with the well known SNR advantage of multi-tone frequency-shift-keying over On-Off keying.

Demonstrated proof of integrity

A live demonstration of the JT65 decoder was provided at the August 2006 EME Conference in W¨urzburg, Germany. JT65 signals could be transmitted at any chosen SNR, with any desired message content, and sent to a receiving computer running the standard WSJT program. Conference participants were invited to explore the operation of the decoder at different SNR levels, perhaps by selecting callsigns included (or not included) in the decoding computer’s database. Many tests were also made with completely arbitrary messages, including random cipher groups. Participants were especially invited to try to “trick” the receiving computer into decoding a message that had not been sent: for example, by transmitting at very low SNR a message different in only one character from one that would surely be tested by the deep-search decoder. The JT65 decoders made zero mistakes during all of these tests: they either produced the correct result, or no result at all. It was also plainly demonstrated that even with “stranger” callsigns or random cipher groups, correct decoding was always achievable (down to about −29 dB SNR) by using WSJT’s ability to average several successive transmissions.  Everyone who bserved and participated in the demonstration was persuaded of the full integrity of the WSJT decoding process.

(2 Note: See URL http://developer.berlios.de/projects/wsjt/ )

 

Minimal valid QSOs

For many years it has been accepted by weak-signal amateur VHF/UHF operators worldwide that a minimum valid QSO requires each station to copy both callsigns, a signal report or some other piece of previously unknown information, and an acknowledgment of complete copy. This easy-to-understand guideline wisely leaves other details concerning the validity of a contact up to the integrity of individual operators. For example, what does it mean to have copied both callsigns during a scheduled QSO attempt, when all of the necessary information is known in advance to both operators? Personal integrity requires that even if the information is already known, it must still be copied over the air, with confidence. Why is it commonly understood that completing a scheduled QSO has a 3 or 4 dB advantage over a contact with an unknown station answering one’s CQ? The answer, of course, is that it is “several dB easier” to be sure that you copied something correctly, if you know in advance what to expect. The sensitivity advantage of the WSJT deep-search decoder is the computer equivalent of this well-known fact for human operators. The principal difference is that the computer can be quantitative about how many dB the advantage amounts to — and what it means to be “sure” that information has been copied correctly. Pages 14–16 of Document B08 reveal a fundamental lack of understanding of how JT65 works in practice, and about modern coding techniques in general. The document fails to recognize that state-of-the-art design of coding and modulation methods can be as important to a reliable
communication system as other engineering choices involving state-of-the-art antennas, pre-amplifiers, and power amplifiers. The document makes the nonsensical inference that decoding a Morse-code “O”, “RO” or “R” by ear conveys legitimate QSO information, but decoding JT65’s “OOO”, “RO”, and “RRR” messages (whether “by eye” or by a computer) does not.
Contrary to assertions made in Document B08, the JT65 protocol is anything but “private.” Full details of its motivation, design philosophy, and implementation were published some 18 months ago in the ARRL technical journal QEX,3 and the same information was made publicly available six months before that. The basic ideas, message structure, encoding techniques, QSO procedures, and other details of the protocol and its predecessor, JT44, have been presented and discussed in a number of open forums and widely attended amateur radio conferences in North America, Europe, and Australia. The most recent example is the August 2006 EME Conference in W¨urzburg, the published Proceedings of which contain a paper describing the history and capabilities of WSJT with special emphasis on JT65.4 Additional publications and a complete copy of my presentation at W¨urzburg can be found on the WSJT web site.(5).

I do not consider it worthwhile to devote space here to detailed corrections of a number of additional misleading or false assertions in the text of Document B08. Instead, I will simply call the Committee’s attention to the need (if they should consider proposal worthy of any further consideration) to solicit input from individuals who actually use the techniques that the document attempts to criticize.

(3 Note:J. Taylor, K1JT, “The JT65 Communications Protocol”, QEX, September-October 2005, pp. 3–12)

Summary

 I believe it is self-evident to nearly everyone that the basic guidelines for minimal valid QSOs should be independent of operating mode. The fact is that our long-established guidelines are sound, and JT65 QSOs meet them every bit as well as QSOs using traditional modes. Indeed, under the most marginal SNR circumstances the reliability, accuracy, and information content of JT65 QSOs far exceeds that of many CW QSOs. In writing that statement I am not questioning the integrity of any CW operator or the validity of any CW QSO; I am simply emphasizing the fact that forward error correction makes the decoded content of JT65 messages much more reliable than the best achievable under marginal CW conditions.
Yes, EME QSOs using the technology advances in JT65 are significantly easier to make than CW or SSB QSOs using otherwise identical equipment. Similar quantum leaps in weaksignal capabilities have occurred in the past several decades when other new technologies (for example, low-noise GaAsFET transistors and computer-optimized antennas) became available. We should be proud that radio amateurs have quickly learned and adopted these state-of-the-art additions to the technology of radio communication — and that our hobby has welcomed the advances.
 

Joe Taylor, K1JT
January 10, 2007
 

(4 Note:“Open Source WSJT: Status, Capabilities, and Future Evolution.” J. Taylor, K1JT, in Proceedings of the 12th International EME Conference, W¨urzburg, August 25–27, 2006.)

(5 Note: See URL http://pulsar.princeton.edu/~joe/K1JT/Documentation.htm )