Many error-correcting data modes are well suited tofile transfers, yet - PDF document

Many error-correcting data modes are well suited tofile transfers, yet most hams still prefer error-proneBaudot for everyday chats. PSK31 should fix that.It requires very little spectrum and borrows somecharacteristics from Morse code. Equipment? Freesoftware, an HF transceiver and a PC withWindows and a sound card will get you on the air. By Peter Martinez, G3PLX PSK31: A NewRadio-Teletype ModeodeBritain for permission to reprint thisarticle. It originally appeared in the1960s, and was instrumental in intro-the end of the Õ70s. This improved thepaved the way to further develop-but further away from the rest ofThere is now a gap opening betweencontact fans who are still using thealthough of course using keyboard andscreen rather than teleprinter. Thereniques now available to bring RTTYneeds of Òlive QSOÓ operatingÑas op-live contacts. PSK31 is now becominging process used in modern datacontacts. I have identified several fac-tors; the first revolves around the factor PACTOR, there is a fixed transmis-sion cycle of 450 ms or 1.25 s or more.correction systems, there is also anmation is spread over time. In a livetwo-way contact, the delay is doubledhanded over. I believe that these de-lem as a human one. Another factor inas the quality of the radio link varies.In an analogue transmission systemrelationship between the two. Theand take account of it subconsciously:conversation topic to suit the condi- 4 QEXthe screen is not so smooth. The mod-ern error-correcting digital modes areparticularly bad at this, with copyabove a certain level and stoppingthis level. The effect is of no conse-quence in an automatic mailbox-for- ASCII*Varicode0 (NUL)10101010111 (SOH)10110110112 (STX)10111011013 (ETX)11011101114 (EOT)10111010115 (ENQ)11010111116 (ACK)10111011117 (BEL)10111111018 (BS)10111111119 (HT)1110111110 (LF)1110111 (VT)110110111112 (FF)101101110113 (CR)1111114 (SO)110111010115 (SI)111010101116 (DLE)101111011117 (DC1)101111010118 (DC2)111010110119 (DC3)111010111120 (DC4)110101101121 (NAK)110110101122 (SYN)110110110123 (ETB)110101011124 (CAN)110111101125 (EM)110111110126 (SUB)111011011127 (ESC)110101010128 (FS)110101110129 (GS)111011101130 (RS)101111101131 (US)110111111132 (SP)1!111111111#111110101$111011011%1011010101&1010111011(11111011)11110111*101101111+111011111,1110101-110101.1010111/110101111010110111110111101211101101311111111410111011151010110116101101011711010110181101010119110110111:11110101;110111101111101101=1010101&#x-439;
.60;111010111?1010101111@1010111101A1111101B11101011C10101101D10110101E1110111F11011011G11111101H101010101I1111111J111111101K101111101L11010111M10111011N11011101O10101011P11010101Q111011101R10101111S1101111T1101101U101010111V110110101W101011101X101110101Y101111011Z1010101101[111110111\111101111]111111011^1010111111_101101101a1011b1011111c101111d101101e11f111101g1011011h101011i1101j111101011k10111111l11011m111011n1111o111p111111q110111111r10101s10111t101u110111v1111011w1101011x11011111y1011101z111010101{1010110111|110111011}1010110101~10110101111271110110101 ASCII*VaricodeASCII*Varicodewarding link, but can badly inhibit theerror-correcting modes, you only getother station. The copy is decidedlyThese factors lead me to suggest based on the use of cific application is that of live con-tacts. The continued popularity of tra-system is proof of this hypothesis:There is minimal delay (150 mS), theflow of conversation is continuous, theerror-rate is tolerable, and it is easy toHow, then, do we go about using mod-ern techniques that were not availableRTTY? First, since we are talking aboutlive contacts, there is no need to discussfaster than can be typed by hand. Sec-ond, modern transceivers are far moreWe should be able to use much nar-could use better coding. The drift-toler-on the start-stop code, without intro-correcting or synchronization pro-cesses, is based firmly on anothertradition, namely that of Morse code.duration of a character. In addition, ifwe think of it in terms that we nor-code is self-synchronizing: We donÕtcode doesnÕt suffer from the Òerror-cascadeÓ problem that results in thestep if a start or stop-bit is corrupt.This is because the pattern used tocode a gap between two charactersThe code I have devised is thereforelength. The letter-gap can also beshortened to two bits. If we representkey-up by 0 and key-down by 1, thenself. The next is 11, then 101 and 111,then 1011, 1101, 1111, but not 1001since we must not have two or moreconsecutive zeros inside a code. A fewminutes with pencil and paper willgenerate more. We can do the 128-of the ASCII characters, then allo-, and I call it the alphabet. With English text, Varicodehas an average code lengthÑinclud-character. By simulating random bitcorrupted characters, I find thatcode, thus verifying that its self-syn-The shortest code in Morse is themost-common letter: ÒeÓ, but in compares the coding ofthe same word in ASCII, RTTY, MorseTo transmit Varicode at a reason-per minute needs a bit-rate of about 32it can be easily derived from the 8-kHzsample-rate used in many DSP sys-tems. In theory, we only need a band-nary data, and the frequency stabilitythe amateur bands, to my knowledge,by SP9VRC. Instead of frequency-shifting the carrier, which is wastefulof spectrum, or turning the carrier onand off, which is wasteful of transmit-ter power capability, the ÒdotsÓ of thethe wires to your antenna feeder. Thisuses the transmitted signal more effi-comparing the signal present in thedot to no-signal in the gap. But if we31.25 baud, it would generate terriblecode, and low-pass filter it to the theo-at the dot rate. The spectrum is a cen-down on either side. A signal that issending continuous reversals, filteredto the minimum bandwidth, is equiva-carrier emission, that is, to two toneson-off keying is thus equivalent to thetextbook improvement in changingwith full carrier to double-sidebandwith suppressed carrier. I have called 6 QEXthis technique Òpolarity-reversal key-ingÓ so far, but everybody else calls it shows the envelope of BPSKTo generate BPSK in its simplestform, we could convert our data1 V, for example,modulator is the desired carrier fre-practice we use a standard SSB trans-could signal logic zero by continuousBPSK, but they all start with a band-pass filter. For the speed chosen forPSK31, this filter can be as narrow as31.25 Hz in theory. A brick-wall filterhowever, not only in monetary termsbut also in the delay time throughwidth (62.5 Hz) at the 60-dB-downpoints with a delay-time of two bitsWe could extract the informationfrom the demodulated signal by mea-code. It helps to pick the data out ofthe noise, however, if we know whento expect signal changes. We caneasily transmit the data at an accu-rately timed rate, so it should be pos-though the term ÒcoherentÓ is some- shows block diagrams of a typicalneed to make sure that there are nomodulation, so we could never predictwhen the next reversal was due. For-need, provided we choose the logic lev-sal and one to a steady carrier. The idlesignal of continuous zeros thus gener-ates continuous reversals, giving us awith continuous keying, there will al-tween characters. The average num-ber of reversals will therefore be morethan two in every 6.5 bits, and therewill never be more than 12 bits withno reversal at all. If we make sure thatthe transmission always starts withan idle period, then the timing willpull into sync quickly. By making theof reversals to squelch the decoder.Hence, the screen doesnÕt fill withtheory, but how do you get on the airwith this mode? In the first experi-ments on this mode in early 1996, thetain one of several DSP starter kits.These are printed-circuit cards, usu-ally with a serial interface to a PC,turers at low cost to help engineersDSP programming. Some radio ama-for these, not just for RTTY but alsofor SSTV, packet, satellite and digital-input and output and some general-purpose digital input/output. The con-struction work needed is limited towiring up cables, building a powerscreened box. The DSP software isfreely available, as is the software thatboard and screen, and can be obtainedmost easily via the Internet. It wouldPSK31 modem in hardware, althoughHowever, it became clear late in 1998software running in the PC. At Christ-soundcard. The availability of this pro- So far, weÕve looked at requirementscommunication system, and proposedthe narrow-band PSK31 mode as atraditional RTTY. This mode has nowbeen in use on the HF bands by a smallabout two years. Now, letÕs look at twoBPSK modulation and the Varicodealphabet, several people urged me toadd error correction to it in the beliefthat it would improve it still further. Itransmission, the discontinuous traf-for live contacts. There is another rea-son. All error-correcting systems workby adding redundant data bits. Sup-mitted bits. If I wanted to maintaintraffic throughput, I would need toresults for bursts of noise, which can-not be simulated on the bench, so Ison tests. The automatic repeat (ARQ)out. Forward error correction (FEC)seemed to deserve a second look, pro- phase-shifted BPSK carrier atdata, but as a single-channel that can phase-shift values. By the way, the clock-just as well for QPSK, because the en-velope still has a modulation compo-data that are organized in blocks ofby transmitting longer blocks. I knowof variable-length blocks like Vari-code. There are ways of reducing er-rors in continuous streams of datawith no block structure. (This seems anatural choice for a radio link, sincetional codes. One of the simplest formseach data bit to be sent, but from asequence of them. This means thateach bit is effectively spread out intime, intertwined with earlier andwe spread it out, the better will be thenoise, but we must not go too far or we 8 QEX dants of the encoders that made theÒancestral genes.Ó If we record the fam-survivor, we can trace it back to findthe transmitted bit-stream. We mustods), however, before all survivorsing system based on the running totalreceived pattern is corrupted. Althoughwe may need to wait a bit longer thanbecome clear. In other words, theThe longer we wait, the more accu-to the other (800 ms), giving a round-seconds. I think this is about the limitbefore it becomes a nuisance. In anycase, the decoder could change to tradevery good, but it is sometimes disap-pointing. In bench tests with whitenoise, it is actually Fig 4ÑThe spectrum of a BPSK signal, idling and sending data, compared with anunmodulated carrier at the same signal level. The carrier is the center pip; thesmaller pips are the PSK31 reversals, and the large, ragged hump is noise shapedby the filter. Fig 5ÑComparison of the PSK31 spectrum with 100-baud, 200-Hz-shift FSK(AMTOR/PACTOR). The taller, three-hump signal at center is PSK31. The smaller,double-peak (
100 Hz) signal is FSK.Fig 6ÑA screenshot of the PSK31 Windows program control panel, receiving aslightly noisy QPSK signal (notice the scope display at left). Fine-tuning controlsfor receive and transmit audio tones are near the bottom-center of the panel. tioned earlier, but in conditions ofbeen recorded. This performancefrom the transmission delay, whichcan be a bit annoying, QPSK is twiceas critical in tuning as BPSK. A QPSKsignal will start to decode wrong whenthe phase shift is greater than 45error is only 3.9 Hz. This could be aproblem with some older radios. Whattends to happen is that contacts startworth doing and there is no drift.be kept in mindÑit is important forIn English-speaking countries, vir-communications are present in the Some armchair lawyers have questioned the legality of PSK31 since its Varicode is not specifically mentioned as reads in part: This Order amends Section 97.304(a) of the Commissionto clarify that amateur stations mayuse any digital code that has its technical characteristics publicly documented. This action was initiated by a letter fromThe Order goes on to note that The technical characteristics of CLOVER, G-TOR and PACTOR have been docu-mented publicly for use by amateur operators, and commercial products are readily available that facilitate the transmis-sion and reception of communications incorporating these codes. Including CLOVER, G-TOR and PACTOR in the ruleswill not conflict with our objective of preventing the use of codes or ciphers intended to obscure the meaning of the com-munication. We agree, therefore, that it would be helpful to the amateur service community for the rules to specificallyGiven that PSK31 is in the public domain for amateur use, that software is readily and freely available and that itsemission characteristics clearly meet the standards of Section 97.307 for RTTY/data, there is little doubt that its use byHowever, just to complete the documentation, in a letter to the FCC dated January 27, 1999, ARRL General CounselChristopher D. Imlay, W3KD, documented the technical characteristics of PSK31 in a manner similar to how CLOVER,G-TOR and PACTOR were previously documented. There is no need for PSK31 to be mentioned specifically in the rules,The left-most numbers in each column contain the 32 combinations of a runof five Varicode bits, transmitted left bit first. The right-most number is the meaning no shift, meaning advance by 90 meaning polarity reversal and meaning. A signal that is advancing in phase continuously is higher in ra-00000201000010000111000300001101001310001211001000010301010110010011010200011001011210011311011100100301100110100011100200101001101210103111011001102011100101101111103001111011113101112111110As an example, the symbol, a single 1 preceded and followed by00010, 00100, 01000, 10000, 00000, which results in the transmitter sendingNote that a continuous sequence of zeros (the Varicode idle sequence) gives text generated on computers. Theseextra symbols are now standardizedworldwide in the ANSI alphabet, thetical to ASCII, and the second 128 con-tain all the special symbols. Since the operating system usesANSI, and most PC programs are nowwritten for It is very easy to add extra charac-backwards-compatibility problems. Inthe early PSK31 decoders, if there wasnored as a corruption. In the extendedalphabet, I let the transmitter legallysend codes longer than 10 bits. The oldas extra characters. To get anothercodes. There seemed little reason to beclever with shorter common charac-merical order, with code number 128255 being 101101011011. The vastmajority of these will never be used, sothe characteristics of modern HFdata-transmission modes that havecontributed to the decline in live QSOoperation on these modes, while tradi-concentrating on the special nature oflive-QSO operation, a new RTTY mode(I donÕt call it a ÒdataÓ mode) has beendevised, which uses modern DSPother telegraphy mode. shows compares this to the bandwidthAt the time of writing (February1999) PSK31 is available for the Texasten by G0TJZ, the Analog DevicesADSP21061 ÒSHARCÓ kit with soft-the SoundBlaster card, DL9RDZ haswritten a -based program forthe PC. Some commercially availablehave already been upgraded to includePSK31 and more will follow. However,the most popular implementation ofsoundcard program, which I havegorithms for PSK31 are being made