WEAK SIGNAL COMMUNICATIONS

Try the headline as a phrase in Yahoo. You would get about two thousand hits and all of it is about amateur radio. If one excludes JT65, WSJT and K1JT, the number of hits goes down by 25%.

It seems to me that there is a lot of confusion about what weak signal communication really is. My personal interest in amateur radio has always been in this area and I have only used Morse coded CW for the purpose. In recent years, JT65 has appeared as an alternative and that has caused a heated and destructive discussion between different groups. One thing is that CW operators in the EME community do not want accept that JT65 contacts are considered equal to CW for diplomas, contests and records. That is understandable since the signal level required for JT65 is quite a bit lower than what is needed for CW. It is no excuse however for using false arguments to invalidate JT65.

In EME, SSB is much more difficult than CW while CW is much more difficult than JT65. This inequality between modes should be taken care of by the contest and diploma rules. It is the organizer who sets the rules and has the responsibility to act as a "police" and make sure that rules are followed. There are no general rules and no authority with the power to set up general rules. IARU can only give recommendations of what should be required for a minimal QSO. The reason I write now is an article in QTC 11/2007 by Peter, SM2CEW which reflects opinions that I find harmful to our hobby and counter-productive for the long-time survival of CW, my own favourite mode. The article can be found on the Internet at
http://www.sm2cew.com/Analys%20av%20WSJT%20Deep%20Search.pdf (Swedish)
http://www.sm2cew.com/Digital%20communications%20using%20minimal%20transfer.pdf (English)

What is a QSO?

The answer to this question is actually the issue that all the discussion circles around. It is at the same time both trivial and complicated. All amateurs know the answer in everyday terms - but if one tries an exact definition it becomes extremely difficult. The following definition that Peter presents in his article was the recommendation for a minimal QSO for a long time:

The definition of a minimum valid QSO is that both stations have copied all of the following:

1. Both callsigns from the other station.
2. Signal reports from the other station.
3. R from the other station, to acknowledge
   complete copy of 1 and 2.

This definition has no meaning until the word "copied" has an exact definition. It has not been much discussed before the deep search decoder of JT65 appeared, but the problem has always been there. Weak signal communications and the technical sides of it in particular has been my main interest in radio for nearly 50 years. Here is the rule I follow myself since long before the above one was established:
1. A QSO is valid if both operators independently (without telephone, 
   Internet etc) consider it valid on their own side. 

2. I consider a QSO valid when I have convinced myself that the QSO
   partner has written my call correctly into his logbook with the
   associated report and that I know his correct call sign. 

3. The information has to be transferred in a generally accepted 
   format that allows anyone to follow the QSO by listening to it
   provided he has a good enough system to get the required S/N.
It is as simple as this. Note that the seemingly objective "copied" now is replaced by "have convinced myself" which is explicitly subjective. I have never had any reason to worry about limitations in space and time other than that a QSO has to be continuous.

In a weak signals context the word "copy" is misleading. When S/N is low there are always errors. Even if we use a machine (mechanical RTTY printer or WSJT in a computer) that actually produces a call sign for us. The printed call can be incorrect. Also when our brain leads our hand to write characters on paper the word copy is misleading even if we assume the operator to have a perfect brain that never makes an error. The signal that reaches the ear is the sum of what the operator at the other end has transmitted and random noise. There is a finite probability for the random noise to form exactly the waveform required to make the summed signal that actually reaches the ear to become a different call sign from what the other station transmitted.

What we have is the operators judgment or more provocatively "guesses". Errors that occur due to random noise do not repeat synchronously with the real signal. It is obvious to everyone who worked with weak signal communications that call signs and reports are repeated to allow the other station to listen several times. An experienced operator will listen long enough to practically never make incorrect "guesses". Yet the risk for an error is not zero.

After having realized that a QSO actually is based on a guess on who is the other station it becomes totally uninteresting, at least to me, to use skeds. Contrary to nearly every other EME operator I have not accepted any sked proposals because I do not want to make guesses about something I already know. The report is always "O" so there is nothing I could guess incorrectly. It is a long time since I lost my antenna in a storm, but during many years I was very active with CW on EME. There were many stations who only worked on skeds. I could hear them easily, but never with a CQ and they would never answer a CQ from me. There was a list on the Internet, about three pages with all skeds scheduled for the week-end. I tried to listen for stations I wanted to work according to the list. Now and then the sked partner was absent and I tried to take over the sked. I did it about 10 times with the following result:

.....
nothing
DL1ABC DE W5XYZ DL1ABC DE W5XYZ...
nothing
DL1ABC DE W5XYZ DL1ABC DE W5XYZ...
W5XYZ DE SM5BSZ SM5BSZ SM5BSZ W5XYZ
DL1ABC DE W5XYZ OOO OOO OOO DL1ABC DE ....
W5XYZ DE SM5BSZ SM5BSZ SM5BSZ W5XYZ
DL1ABC DE W5XYZ OOO OOO OOO DL1ABC DE ....
SM5BSZ SM5BSZ SM5BSZ SM5BSZ SM5BSZ ....
DL1ABC DE W5XYZ OOO OOO OOO DL1ABC DE ....
nothing
DL1ABC DE W5XYZ OOO OOO OOO DL1ABC DE ....
nothing
Only one single time I was successful in taking over a sked. The probability for an incorrect guess in a sked is extremely large in real life. I am convinced that operators normally do not listen at all but base their guess about hearing the expected station on the assumption that nobody else is supposed to use that particular frequency at that particular time. In this connection I must say that I used equipment that made me absolutely sure that the sked partner was absent.

The above text is in no way intended to declare sked-QSOs made by others non-valid. The assumption that it has to be the sked partner if something is heard is practically always correct in EME. I am also convinced that sked operators listen carefully for the "O" report and the final string of "RRRR". When the transmission of these starts or if they start at all is truly unknown. The operator has to hear enough to be sure enough that the other station will consider the contact valid. There is no certainty however, not when both S/N and time is limited.

One year ago, at the Vienna meeting, the IARU Region 1 VHF/UHF/Microwave committee decided on recommendations for minimum requirements for a valid QSO on VHF and higher bands:

A valid contact is one where both operators during the contact have
(1) mutually identified each other
(2) received a report
(3) received a confirmation of the successful identification
    and the reception of the report.
It is emphasized that the responsibility always lies with the operator for the integrity of the contact.

The new recommendation is clearly better than the older one. It avoids becoming meaningless by not using the undefined word "copy". I do however believe my personal rule is better and also useful on all bands.

Detect limits.

In CW the detect limit depends on the bandwidth. Slower keying gives less bandwidth and better S/N, but there are limitations. On EME one has fading due to multi-path propagation. The moon has a rough surface and it twists so that reflections from different parts sometimes add and sometimes cancel (libration fading). For this reason the keying speed should not be too low because then the received fragments become too short and do not contain enough information. A single letter occasionally on QSB peaks will never give a QSO. One has to receive groups of at least four characters, preferably more, to be able to construct an unknown call sign. On 144 MHz the optimum keying speed is around 12 WPM. The optimum bandwidth is then about 13 Hz. I can copy CW down to -16 dB (in 2500 Hz bandwidth) by use of a narrow filter. I can imagine good operators can copy such levels using an ordinary CW filters even though I personally can not. By use of coherent detection using the carrier phase (available for example in Linrad and predecessors) I can receive 12 WPM down to -18 dB well enough to work random EME. Download SimJT, a program by K1JT. Generate your own audio files with weak signals and try yourself. Note that the sensitivities given in this article are key-down power vs RMS noise in 2.5 kHz bandwidth to conform with the noisy CW generated by SimJT.

In real EME the QSB peaks are something like 5 dB stronger than the average power level so in real life a signal level of about -23 dB is enough for me to work random contacts. Such QSOs take long time because there are only a few QSB peaks each minute that give any information at all and it takes me a long time to convince myself that I have received the correct call sign and that the other station has my own call correctly. My random contacts at the minimum level typically lasted half an hour and the other station used very low power so he would typically copy my transmissions reasonably well and hear how I change from CQ to QRZ after having seen a response on my waterfall graph. Random contacts at the minimum S/N in both directions are highly unlikely because small stations do not call CQ using CW.

If one runs JT65B and uses the KV decoder at -23 dB, the chance to get the correct message on screen is about 80 % in a single receive period. There is of course also a risk that an incorrect message would appear, but that risk is extremely small and can be neglected. The ultimate limit for JT65B is the limit for sync detection. It is somewhere near -28 dB. Provided that the QSO partner repeats the same message over and over again, it takes about 20 minutes to collect the about eight successful sync detects required for the KV decoder to produce a result from the averaged waveform and with nearly 100% probability show the correct calls on the screen. When that has happened the operator can "guess" in good faith that it really was what the QSO partner transmitted. It is generally accepted that the KV decoder can be trusted.

Deep search in JT65

I have made some experiments with files generated with SimJT at -28 dB and it turns out that deep search with aggressive decoding gives about 60% correct results and no error in the 10 passes normally required to get an output from the KV decoder. I have not found it worthwhile to make many enough tests to have any opinion about the number of incorrect results. The big controversy is about whether the operator can use the result from the deep search decoder to make his guess about what the other station has transmitted. It is quite obvious that at -28 dB the information actually received from the antenna was not enough to judge who out of all radio amateurs in the whole world has called who. On the other hand, it is highly unlikely that the result would be wrong. To be on the safe side one could wait for another deep search to print the same result. It is obviously not necessary to wait until eight sync detects have been successful. Then one would have about six identical independent and correct results from the deep search decoder on screen plus the same output from the KV decoder.

I made my tests with WSJT version 5.9.7 and I can say nothing about how earlier versions may have behaved. I can also say nothing about how spurs and other interferences that are not white noise might affect the result.

What I find important and think is missing in Peters article is the role of the operator. The WSJT program takes no decisions. It is the operator who makes decisions based on his knowledge and experience. The operator has to make his own judgement about what becomes visible on the computer screen. Can I be sure enough that this really represents a message that was received? What is the confidence level? Is it high enough for the current interference situation? Do I have to wait for another "decode" to be sure enough that it is true? He takes the decision whether he is sure enough that the text on the screen is in agreement with what the other station transmitted. Peter writes that there have been shortcomings on this point manifested as false QSL cards. Those (if they exist in a significant amount) are however not caused by JT65 but by bad operators who did not properly convince themselves that they could trust the output from the DS decoder to be true on a single detect regardless of QRM and confidence levels.

A good operator would base his decisions on his own personal experience. Particularly when WSJT was new, every sensible operator should expect reliability problems. Who has ever heard of a software package that was completely free from bugs and mistakes of any kind from start? Surely theory tells us what to expect - but that is under the assumption of no mistakes of any kind. The operator is fully responsible for making sure he knows enough to make very few incorrect QSOs. Regardless of mode!

JT65 vs CW.

Peter writes in his article that a QSO using the KV decoder (in a single receive period) requires approximately the same signal level as a CW QSO. This is correct as shown above, but the amount of time required is vastly different.

Comparing what one mode can do in a couple of minutes to what another mode can do in half an hour is irrelevant. Both modes have a threshold where the time increases dramatically for small reductions of S/N. The limit S/N at times in the order of half an hour does not change much if the QSO time is allowed to double both in CW and in JT65..

At the JT65 limit, I tried it a couple of times at -28.5 dB and then typically 27 receive periods, i.e. nearly one hour QSO time, was required to collect the about 9 successful sync detects needed for the KV decoder to make a detect on basis of the averaged waveform. Out of these about 9 sync detects, typically five gave the correct result using the deep search decoder (aggressive) while the others gave no output from the deep search. I have not seen a single false decode, but my tests were in a background of pure white noise so I just do not know what to expect in a real on-the-band situation. An operator that decides that two independent decodes with the same result from deep search is enough is likely to have decided upon the call sign within 15 minutes based on my tests at -28.5dB. Only in case the other station is listed with the correct locator in call3.txt. Otherwise one has to wait for about an hour for the KV-decoder.

To base a decision on a single result from deep search might be hazardous because the results are only certain (to within less than about 0.1%) in a background of pure white noise and that is based on theory under the assumption of no software errors. DS results are independent and uncorrelated so two identical results on two different receive periods gives an extremely high confidence level and "guesses" based on that will virtually always be correct. Generally, what I find is well in line with the performance figures published by K1JT.

Just as QSB improves the detect limit for CW it will also improve the limit for JT65, but to a much smaller extent. With normal EME fading the JT65 threshold is probably -29.5 dB and I think it will be correct to state that the JT65 advantage over CW is about 6.5 dB (29.5-23) on typical 144MHz EME signals for operators that use the best digital technology for CW or have extremely well trained ear-brain systems.

Philosophical questions.

In Dubus 3/2007 Klaus, DJ5HG writes that deep search can not transfer call signs. This is of course quite correct. Deep search is just a tool that may help the operator to make a guess faster. What to think about that is controversial.

When Klaus uses the word "mode" he refers to "communication mode", the whole chain from the transmitting operator to the guess made at the receive side. I prefer the provocative "guess" over the conventional (more neutral) wording "decision". The computer program computes that one particular message out of a given set of messages has the best fit to the received signal and writes the best fit on the screen. The operator may guess that the letters on the screen actually represent the message that was transmitted by the QSO partner.

As a parallel, consider the transfer of books: I sit in my library and have a couple of hundred books around me. It is enough for me to have access to a "communication mode" capable of transfering "ARRL HANDBOOK 1994" in order to transfer over 1000 pages of printed information without any error to someone who has access to a similar library. Surely I do not have to transfer the book letter by letter to bring all the information in the book to someone else. An operator who uses the deep search decoder as a basis for his own guesses will not be likely to make mistakes. Note the dramatic difference to my experiences with trying to take over skeds in CW where the mere presence of a signal obviously was enough to provoke the guess that the sked partner is present.

In the context of discussing what should constitute a minimum QSO the amount of information is not a trivial problem. As I see it, the "transmit mode" has to be capable of encoding contacts between all possible (normal?) amateur call signs in the world. As a consequence the number of transmitted information bits can not be much smaller than the number of bits in the JT65 protocol. Peter tries to invalidate JT65 with a reference to the papers by DJ5HG, but Klaus uses the concept "communication mode" from information theory. It is relevant for describing a communications link where for example a random text on a piece of paper should be transfered to another place as printed text on paper. In amateur radio it is different. The total information content is very low and the real desicions are taken by a human. It is like testing in medicine. The only valid test is the double blind test. As an example, (interesting to radio amateurs) read this article: "Blood-brain barrier permeability in rats exposed to electromagnetic fields used in wireless communication." (Wireless Networks 3, 1977, 455-461) Particularly table 1. Rat brains are damaged by microwave radiation after absorbing only 1.5 Joules per kilogram. Increasing the exposure to 48000 J/kg makes no difference! (No more damage.) The study used 1260 rats and it is statistically over-significant. Personally I am convinced that the study shows that the human operator that had to classify a rat as damaged or not had prior knowledge of whether the rat was exposed or not. I do not intend to say that someone was cheating - I just want to say that all experiments where human judgement is involved have to be made double-blind to hold scientifically.

The important factor in an amateur contact is the operator. He makes various things that are hard to model scientifically. Introducing a rule that declares contacts made by use of the DS decoder non-valid would be meaningless. Too many operators would not accept it and use DS secretly anyway. There is no way for others to know. Declaring JT65 non-valid because there is a DS associated with it will not work. If someone designs a different mode without a DS decoder it will always be possible for someone else to add a DS decoder. It is possible to add a DS decoder to any decoder. The reason there is not a DS decoder for CW is that there is no good CW decoder (yet?) in software. As I see it, we have to accept that the method used to learn the content of a signal (i.e., to "decode" it) is nobody's business but the receiving operator. I do not say I like it - but I see no alternative.

Personally I have no problems at all with it. In random contacts incorrect guesses will not produce false contacts. The situation is the same as when one mistakenly transmits a wrong call in CW to someone who answered a CQ. One would not receive reports, but repeated call signs from the other station or alternatively, the other station would disappear.

In skeds it is different. It is impossible to guess something one already knows so the operator is actually doing something else. I have solid experience in what that means in CW!! JT65 should be better in this respect. Personally I feel that the reasons that made skeds un-interesting in CW are much stronger in JT65. To have agreed about what the computer is expected to print on its screen in advance and also feeding precisely that information into the computer would not make me feel comfortable - but that is just my personal feeling.

Klaus advocates QSO rules for a valid QSO that acknowledge that we are doing guesses although he calls them decisions. He suggests that the operators decision has to be based on information that would give the same probability for all possible call signs in the world if the decision were completely at random. In JT65 the total number of possible results of the deep search decoder is only 14466 with the data base distributed with the latest version of WSJT. With the DJ5HG criterion, a guess based on it would not constitute a valid contact.

If one on the other hand awaits a detect on the average waveform with the KV decoder, then the rule that Klaus proposed will be fulfilled with a very large margin.

When running traditional modes, the criteria for a valid QSO are less strict. We all know a great deal about likely call signs and guessing occurs wildly all the time. I have personally had a good antenna with adaptive polarization, so I have been able to listen to the traffic and I know with certainty that this is happening. Sometimes one can hear station A in QSO with B who is in QSO with C. Reports and final RRRR are happily sent without the faintest idea that all of it is totally wrong. If one uses the computer with JT65, the number of incorrect contacts would decrease - at least with modern program versions.

I also want to point out that in case old programs had a deep search decoder that was likely to produce false messages, the decision that a single such message is a true decode would be an operator error. The operator alone is responsible for all decisions. The tool that allows a guess prior to the KV decoder output from averages must be used only in situations where the operator knows that it is reliable enough. He has to know how QRM, spurs and aggressive decode affects the reliability and he must know how to interpret the confidence level.

Lower signal levels, less time and less effort is required to make a JT65 QSO as compared to a CW QSO. Peter obvoiusly thinks that ARRL should have made retroactive changes to the DXCC rules in order to make JT65 contacts non-valid. The rule for "mixed" is: "Contacts may be made using any mode since November 15, 1945." Retroactive changes would cause many problems to ARRL in the future and are needless. One can have the DXCC diploma for CW and then all contacts have to be made in CW. Possibly Peter means that ARRL should not have recognized JT65 as a mode. That would have been even more problematic for the future as can be understood by the text above.

More on deep search.

As I see it, the deep search is just a tool that helps the operator to make his guess faster under the assumption that the other station is present in the data base. If it is missing, one has to be patient and wait for the KV decoder to show the message based on averages. It may look like DJ5HG and K1JT have different opinions about the way JT65 works. That is not at all the case as far I can understand from what they write. They do however have incompatible opinions how to value the mode. For that reason I now make a detailed description about the workings of the deep search decoder in a way that I believe both Joe and Klaus will find correct and honest. Hopefully this will help the reader to make his own decision how to value JT65.

WSJT computes a correlation (similarity) between what is received and all possible AND interesting messages that can be constructed from the data base. With 4822 call signs, this amounts to 14466 examined messages. Each message consists of 63 tones, each of one can have 64 different values. The total number of patterns that could in principle be received, 64 to power 63 (about 6 x 10^113) is a vastly larger number. Only 2^72 (about 5 x 10^21) of these patterns represent valid WSJT messages so there are about 10^92 tone patterns that can not be a JT65 message for each one of the 2^72 possible messages. It is not difficult to accept the claim that the encoded forms of valid messages are all very different from each other and that the pattern of every valid message differs from that for every other valid message in at least 52 of the 63 data tones. Clever mathematicians have figured out how to achieve this. The number 2^72 is very large; it can represent all pairwise combinations of all normal amateur call signs in the world, and a great many more messages as well. Nevertheless it is tiny, it relates to the total number of possible patterns as one to 10^92 a number that needs 92 digits. The process of rejecting everything that is impossible is the origin of coding gain, the qualitative difference between JT65 and CW.

There is no question that the transmitted message can represent all combinations of normal call signs in the world. (More below on unusual ones.) All the controversy is about the receive side.

The ideal - and uncontroversial procedure at the receive side would be to compare all 2^72 possible patterns to the received pattern and if the best fit exceeds some threshold criterion, guess that it is the message that really was transmitted. Assuming a white noise background and no QSB we can then say that the signal level required to receive 50% correct messages, 49.9% no decode and 0.1% false decodes would have to be about -26 dB in 2.5 kHz bandwidth. As far as I can understand both K1JT and DJ5HG are in full agreement this far.

Now, if the comparison were made only to the 14466 messages of interest that are consistent with the data base while using the same threshold, one would still have 50% correct detects at -26 dB, but the number of false detects would be 10^17 times smaller. What Joe did in the WSJT program and what Klaus finds unacceptable is to make the threshold a function of the database size. This way the probability for a false decode is kept at a level that might approach 0.1% regardless of the size of the data base. The advantage is that the lower threshold improves the sensitivity - provided that the call sign really is present in the data base. Surely most of the information is fetched from the library, the hard disk, but the amount of transferred information is large enough to make it 99.9% sure that the "correct book" is fetched from the library. If the received "title" is not in the library we will find the shelf empty - a "no decode".

In the terminology, with mode equal to "communication mode" that Klaus uses in his articles, JT65DS is a large number of different modes. Each time the contents of the data base is changed, a new mode is created. In contrast, the "transmit mode" JT65 is well defined.

The ideal decoder, the one which is uncontroversial and correlates to all the 2^72 possible tone patterns is not technically possible. The nearest one can reach today is the KV decoder and it provides a detect limit of -24 dB, 2 dB worse than the theoretically ideal limit.

If one compares to only the 14466 patterns that correspond to the interesting messages that can be constructed from the data base, the detect limit becomes -28 dB if one wants the risk of making an incorrect guess to become not far below 0.1%. If one extends the data base for the number of tested patterns to become two million, the detection limit becomes -27 dB. That is only one dB better than what one would get by including all 2^72 legal patterns. Peter writes that a larger data base gives a large probability for incorrect call signs. This is completely wrong. The mistake is caused by a misunderstanding of what Klaus has written which is only about the "communication mode" JT65DS with 4822 call signs in the data base. In Klaus terminology, JT65DS with 2 million calls is another mode!

Read about this in Dubus 4/2007 where both Joe and Klaus have written: Joe argues that it is enough that the operator has made sure that the risk for a faulty QSO is negligible. Klaus does not accept guesses when the number of possible guesses is so small as three times the number of calls in the data base. It has to be at least two millions in order for the actually transferred amount of information to correspond to what is needed to identify an unknown station.

It is obvious that the data base in JT65 in principle could be extended to include all radio amateurs in the world. The performance of the deep search decoder would not be degraded by more than about 1 dB. Then Klaus would accept that the deep search decoder is OK and it would still provide a 3 dB better sensitivity than the KV decoder.

Personally I have a simple-minded attitude to the discussion. The operator has a responsibility to make sure that what he writes in his log book is correct. There has to be unknown components in the QSO. The call sign of the QSO partner or the report must not be known in advance and a confirmation that the QSO is OK must have been received within the QSO without any form of information exchange over other channels. When these criteria are fulfilled I think a QSO is OK. A QSO must also be open to third parties. Anyone interested must be able to follow the progress of a QSO just by listening to the radio waves. Peter writes in his article that this is not true with JT65 but that is clearly false. The third party station can not use the DS decoder (not until he knows the call sign of one of the qso participants) so he may need a better receive system that permits decoding with the KV decoder.

I apply exactly the same criteria to all modes on all bands. It is highly questionable whether the strict rules proposed by Klaus could be applied to traditional modes.

DX expeditions in JT65.

Peter does not like the way temporary DXCC prefixes are used in JT65. I agree with him on this point and would prefer a fixed code that always has the same meaning. I suggest that every JT65 user simply puts for example UNKN into the temporary prefix box. Nothing would change in the transmitted message and the operator who receives UNKN/PA3CNX and works that station would have to make sure he knows the correct prefix through the QSO itself. During CQ, PJ4/PA3CNX can use his correct call as plain text. If UNKN/PA3CNX answers my CQ I would have to type "PSE TX PREFIX" or equivalent as a plain text message and wait for a response.

Final remarks.

We may communicate with very little information. If enough information is agreed upon in advance, a single bit is enough. For an example look here: http://www.sm5bsz.com/kk7ka/kk7ka.htm Such communication does not constitute a QSO however. It was just an interesting experiment.

Peter fears a development towards usage of code tables where every interested operator has a code that is valid within a limited time frame and uses that code rather than his call sign for communication. Peter writes that "single letter calls" have been proposed as a method to minimize information transfer. This is obviously unacceptable for QSOs - but could be used for experiments like mine with KK7KA.

For formal reasons I want to interpret mode as "transmit mode." Frequency allocation is by mode and it is related to the bandwidth of the transmitted signal. A mode has to be generally accepted and have a definition that is stable with time. It must also allow encoding of all possible pair-wise combinations of amateur call signs although it could be done in separate transmit periods provided that the mode makes it possible to make sure that the different transmissions come from the same station. A mode must also have a set of other messages to allow reports and confirmation. I am convinced these constraints on modes are implicitly understood. "single letter calls" is not the future of digital communication.

Peters article has a final remark: "If anyone can make a software that is capable to decode two completely unknown call signs and a report in reasonable real-time at -27 dB I would very much like to test it." I suggest JT65A. I have seen it decode two calls and a report with the KV decoder using the average of anything between two and four receive periods at -27 dB. The time from start until an output from the KV decoder was anything from just below 3 minutes up to just below 9 minutes. Reasonable real-time compared to CW at low signal levels. The message can contain a report or a locator and no information whatsoever is needed in advance. The deep search decoder is not involved. With the KV decoder 72 bits of information is transferred with extremely high integrity in typically 5 minutes at -27 dB. This is easy to verify for anyone who is willing to spend half an hour with SimJT and WSJT. If you want to verify that the S/N levels created by SimJT and reported by WSJT are correct you can use Linrad.

From a technical point of view, Peters article is incorrect in many respects. JT65 is not based on the deep search decoder at all. Showing that text appended to call signs in the database appear in the deep search results is pointless - everyone who uses JT65 must know that a suffix can not be more than three characters in JT65 and that any surplus characters are not involved in the coding or decoding.

Surely there are many things in JT65 that can be criticized. The message format is limited and some special event call signs may not fit together with another 6-character call sign in a 13-character free text message. The temporary DXCC arrangement can also be questioned. I think future trends in digital communication will go in the opposite direction to what Peter fears. I would rather think that there would be an interest in somewhat longer free text messages.

Parallel communication via the Internet chat sites, DX clusters, telephone or other amateur bands is in no way a problem to digital modes only. I do not want to try to impose my personal attitude towards skeds on others but there is obviously a grey zone. I find it deplorable that the ARRL EME contest has a category "Assisted" for operators who use real time Internet to solicit contacts or make skeds. This category was introduced by ARRL in 2005. It was a time when most JT65 operators were newcomers because the mode was new. It was a bad idea then and it is getting worse as years go by. The fact that the Internet has been mis-used much more by JT65 operators than by CW operators does not mean that JT65 is better helped by the Internet than CW. Actually it is quite the opposite. What we have seen is the difference between a mature mode with experienced operators and a new mode with many unexperienced operators. The digital operators need help to understand that amateur radio is done with radio waves and not by Internet. They also need help to understand that should require ARRL to remove the assisted cathegory.

New amateurs do not have to learn CW. I think it is very important in the EME community to have at least one mixed contest every year. A contest in which the operator who wants to win will have to use both CW and JT65. Separating the modes in time will not encourage newcomers to try CW. In a long time scale there is a risk that CW will disappear completely - those of us who use CW should try to make newcomers try it and discover the charm of this fast and extremely flexible mode!

73 de SM5BSZ, Leif