Hellduino: Sending Hellschreiber from an Arduino

Update: Welcome Hack-a-day readers! If you are looking for the schematics for this “transmitter” (really just a simple oscillator, send some love to radio guru Steve Weber over at his website. You could really use any oscillator you like, even a canned oscillator (although the square waves would generate lots of harmonics).

Yesterday’s project coupled a simple Colpitt’s oscillator (snipped from Steve Weber, KD1JV) with an Arduino. Steve used it to send temperature telemetry in Morse code back to his shack from an outdoor thermometer. But I thought that something else could be done: sending telemetry in Hellschreiber.

Hellschreiber is an early type of facsimile teleprinter system developed in the 1920s, which has enjoyed a certain amount of popularity in the amateur radio community. It sends characters using a conventional on-off keyed transmitter, just like Morse, but instead of sending dots and dashes of different lengths, it scans characters left to right, and top to bottom, and keys the transmitter on where each letter is “on”. Thus, a Morse transmitter can be modified pretty simply to send Hellschreiber.

So I did.

Here are some details. Hellschreiber is normally defined as sending characters defined on a 7×7 matrix, at 122.5 dots (2.5 characters) per second. But the actual font is actually defined on a 7×14 matrix. To keep the bandwidth of the signal down, the font doesn’t ever define a character that requires turning single dots on or off: the minimum signal changes are two dots long. These “half dots” are sent at 245 baud, or about 4.08ms per dot. Because I needed to account for the time spent looking up the character, I tuned that down to about 4.045ms. I was concerned that because I was keying the oscillator on and off, the startup time (which I estimated at about 2ms) could be a problem, but I suspect the shutdown time is about 2ms as well, so the overall system works better than you might imagine. The startup and shutdown keying waveforms are a bit erratic though, and the bandwidth of the signal is probably too wide. I think a better way to do this would be to build an oscillator that runs continuously, and then key a buffer amp with a filtered pulse to keep the bandwidth low. But for a 500 microwatt transmitter (estimated, and represents power going into the antenna, not radiated) it probably works just fine.

Here’s the source code:

int radioPin = 13 ;

typedef struct glyph {
    char ch ;
    word col[7] ;
} Glyph ;

Glyph glyphtab[] PROGMEM = {
{' ', {0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000}},
{'A', {0x07fc, 0x0e60, 0x0c60, 0x0e60, 0x07fc, 0x0000, 0x0000}},
{'B', {0x0c0c, 0x0ffc, 0x0ccc, 0x0ccc, 0x0738, 0x0000, 0x0000}},
{'C', {0x0ffc, 0x0c0c, 0x0c0c, 0x0c0c, 0x0c0c, 0x0000, 0x0000}},
{'D', {0x0c0c, 0x0ffc, 0x0c0c, 0x0c0c, 0x07f8, 0x0000, 0x0000}},
{'E', {0x0ffc, 0x0ccc, 0x0ccc, 0x0c0c, 0x0c0c, 0x0000, 0x0000}},
{'F', {0x0ffc, 0x0cc0, 0x0cc0, 0x0c00, 0x0c00, 0x0000, 0x0000}},
{'G', {0x0ffc, 0x0c0c, 0x0c0c, 0x0ccc, 0x0cfc, 0x0000, 0x0000}},
{'H', {0x0ffc, 0x00c0, 0x00c0, 0x00c0, 0x0ffc, 0x0000, 0x0000}},
{'I', {0x0ffc, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000}},
{'J', {0x003c, 0x000c, 0x000c, 0x000c, 0x0ffc, 0x0000, 0x0000}},
{'K', {0x0ffc, 0x00c0, 0x00e0, 0x0330, 0x0e1c, 0x0000, 0x0000}},
{'L', {0x0ffc, 0x000c, 0x000c, 0x000c, 0x000c, 0x0000, 0x0000}},
{'M', {0x0ffc, 0x0600, 0x0300, 0x0600, 0x0ffc, 0x0000, 0x0000}},
{'N', {0x0ffc, 0x0700, 0x01c0, 0x0070, 0x0ffc, 0x0000, 0x0000}},
{'O', {0x0ffc, 0x0c0c, 0x0c0c, 0x0c0c, 0x0ffc, 0x0000, 0x0000}},
{'P', {0x0c0c, 0x0ffc, 0x0ccc, 0x0cc0, 0x0780, 0x0000, 0x0000}},
{'Q', {0x0ffc, 0x0c0c, 0x0c3c, 0x0ffc, 0x000f, 0x0000, 0x0000}},
{'R', {0x0ffc, 0x0cc0, 0x0cc0, 0x0cf0, 0x079c, 0x0000, 0x0000}},
{'S', {0x078c, 0x0ccc, 0x0ccc, 0x0ccc, 0x0c78, 0x0000, 0x0000}},
{'T', {0x0c00, 0x0c00, 0x0ffc, 0x0c00, 0x0c00, 0x0000, 0x0000}},
{'U', {0x0ff8, 0x000c, 0x000c, 0x000c, 0x0ff8, 0x0000, 0x0000}},
{'V', {0x0ffc, 0x0038, 0x00e0, 0x0380, 0x0e00, 0x0000, 0x0000}},
{'W', {0x0ff8, 0x000c, 0x00f8, 0x000c, 0x0ff8, 0x0000, 0x0000}},
{'X', {0x0e1c, 0x0330, 0x01e0, 0x0330, 0x0e1c, 0x0000, 0x0000}},
{'Y', {0x0e00, 0x0380, 0x00fc, 0x0380, 0x0e00, 0x0000, 0x0000}},
{'Z', {0x0c1c, 0x0c7c, 0x0ccc, 0x0f8c, 0x0e0c, 0x0000, 0x0000}},
{'0', {0x07f8, 0x0c0c, 0x0c0c, 0x0c0c, 0x07f8, 0x0000, 0x0000}},
{'1', {0x0300, 0x0600, 0x0ffc, 0x0000, 0x0000, 0x0000, 0x0000}},
{'2', {0x061c, 0x0c3c, 0x0ccc, 0x078c, 0x000c, 0x0000, 0x0000}},
{'3', {0x0006, 0x1806, 0x198c, 0x1f98, 0x00f0, 0x0000, 0x0000}},
{'4', {0x1fe0, 0x0060, 0x0060, 0x0ffc, 0x0060, 0x0000, 0x0000}},
{'5', {0x000c, 0x000c, 0x1f8c, 0x1998, 0x18f0, 0x0000, 0x0000}},
{'6', {0x07fc, 0x0c66, 0x18c6, 0x00c6, 0x007c, 0x0000, 0x0000}},
{'7', {0x181c, 0x1870, 0x19c0, 0x1f00, 0x1c00, 0x0000, 0x0000}},
{'8', {0x0f3c, 0x19e6, 0x18c6, 0x19e6, 0x0f3c, 0x0000, 0x0000}},
{'9', {0x0f80, 0x18c6, 0x18cc, 0x1818, 0x0ff0, 0x0000, 0x0000}},
{'*', {0x018c, 0x0198, 0x0ff0, 0x0198, 0x018c, 0x0000, 0x0000}},
{'.', {0x001c, 0x001c, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000}},
{'?', {0x1800, 0x1800, 0x19ce, 0x1f00, 0x0000, 0x0000, 0x0000}},
{'!', {0x1f9c, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000}},
{'(', {0x01e0, 0x0738, 0x1c0e, 0x0000, 0x0000, 0x0000, 0x0000}},
{')', {0x1c0e, 0x0738, 0x01e0, 0x0000, 0x0000, 0x0000, 0x0000}},
{'#', {0x0330, 0x0ffc, 0x0330, 0x0ffc, 0x0330, 0x0000, 0x0000}},
{'$', {0x078c, 0x0ccc, 0x1ffe, 0x0ccc, 0x0c78, 0x0000, 0x0000}},
{'/', {0x001c, 0x0070, 0x01c0, 0x0700, 0x1c00, 0x0000, 0x0000}},
} ;

#define NGLYPHS         (sizeof(glyphtab)/sizeof(glyphtab[0]))

encodechar(int ch)
    int i, x, y, fch ;
    word fbits ;

    /* It looks sloppy to continue searching even after you've
     * found the letter you are looking for, but it makes the 
     * timing more deterministic, which will make tuning the 
     * exact timing a bit simpler.
    for (i=0; i<NGLYPHS; i++) {
        fch = pgm_read_byte(&glyphtab[i].ch) ;
        if (fch == ch) {
            for (x=0; x<7; x++) {
                fbits = pgm_read_word(&(glyphtab[i].col[x])) ;
                for (y=0; y<14; y++) {
                    if (fbits & (1<<y))
                        digitalWrite(radioPin, HIGH) ;
                        digitalWrite(radioPin, LOW) ;
                    delayMicroseconds(4045L) ;

encode(char *ch)
    while (*ch != '\0') 
        encodechar(*ch++) ;

  Serial.begin(9600) ;
  pinMode(radioPin, OUTPUT) ;

    encode("K6HX QTH CM87UX TMP 72F PWR 500 MICROWATTS ") ;

Let me know if you use this project for anything!

Late night pondering about the micro-power Morse beacon…

Before toddling off to bed last night, I did a bit more tinkering, and a bit of thinking, and then a bit of research.

The YouTube video I made showed that the spurious radiation from just attaching a clip lead from the oscillator to my oscilloscope gave enough signal to inject itself into my RFSpace SDRIQ software defined radio, even without any antenna attached. I hypothesized the signal should be stronger with my regular 40M dipole antenna, so I tested that, and sure enough, my FT-817 attached to the 40m dipole easily is S7, even without doing any antenna tuning. This suggests that for any application on my small property, I could reduce the power significantly (and use little/no antenna) and still have very legible signals.

I used a 3.68Mhz crystal because I have something like 200 of them (I bought them for something like $3 for a big bag) but that’s kind of a ham unfriendly frequency. While it’s not very likely for such a weak signal to propagate significantly at the power levels we are talking about, but it’s perhaps possible that someone within a radius of a mile or so might be able to hear it, and be annoyed. Steve Weber used 4Mhz crystals, which are at the edge of the 80m band, and probably less well trafficked. A good, friendly idea.

Last, I was trying to understand what the regulations actually say about legal Part 15 operation. The regulations allow for experimenters to build five transmitters of this type for experimentation. On the 80m, you are suppose to keep the average field strength at just 15 microvolts per meter at a distance of 30m. Working through the approximations presented in Understanding the FCC Regulations for Low-Power, Non-Licensed Transmitters, this suggests that the product of power and antenna gain should be less than 6.6E-8 or so. While I know the input power, I don’t know how to measure (or estimate) either the output power or the antenna gain. Can anyone point me at a reference which shows how I might calculate the properties of a small 6″ long stub antenna at 4Mhz or so? I know it’s mostly an academic exercise, but it’d be nice to know the rough limits are, and the FCC regulations for Part 15 do say that we are supposed to “employ good engineering practices in order to ensure compliance with Part 15 standards”.

One final thing: as you can see from the oscilloscope, the output is far from a clean sine wave. It’s not horrible, but I was able to detect it at 2x and 3x the carrier frequency. The obvious thing would be to design a filter to clean it up, but most filter designs assume a 50 ohm (or thereabouts) output load, which this clearly doesn’t have. Is that a problem?