MIL-STD-188-110C appendix D, 3 kHz

hfmodem.bateleur.wbhf sends and receives the 3 kHz waveforms of MIL-STD-188-110C appendix D, the wideband HF family: waveforms 0 to 13, 75 to 16000 bit/s. It was written from MIL-STD-188-110C with Change 1, appendix D. No other implementation was read, and nothing has been tested against another sender.

The appendix covers channels of 3 to 24 kHz. Only 3 kHz is here: the rest need wider channels than an amateur SSB transceiver passes. At 3 kHz it is 2400 symbols a second on 1800 Hz, the same as 188-110's serial tone and STANAG 4539, so it goes through the same rig unmodified.

The waveform

The test suite checks the appendix's worked examples:

• the trinomial's first 32 values;
• the dibit 01 scrambled with them;
• the 24-symbol mini-probe and its shifted form (table D-XXII);
• the 1/3 code as 2/3 applied after repeating;
• the waveform identifier and count codings.

Where the text left a choice

• Preamble scrambling. Each section's table is 256 long, which is one symbol at 24 kHz. A 3 kHz Walsh symbol is 32 chips, and the appendix does not say which 32 of the 256 it takes. Here every symbol takes the first 32, as the level-control blocks do.
• The checksum bit. The appendix asks for d2, the identifier's checksum bit, and also says w1's least significant bit, which is d2, shall be 0. The checksum is sent here, and either is accepted.
• The filter. The appendix recommends a 35% root-raised cosine. Its requirement is only that the signal stay inside 3 kHz. bateleur sends its own pulse, the one all its serial waveforms share, which fits the station's filter.

Modes

Modes are named by rate and interleaver: US, S, M or L, for example 9600L, 150US or 75S. Waveform 0 has no ultra short interleaver. Each rate belongs to one waveform:

The receiver reads the waveform and interleaver from the preamble, so only the sender sets the mode: MODE WBHF <mode> on the host port, or offair render WBHF --mode. The modem always listens for it, as it does for 188-110 and 4539. A STANAG 5066 node set to it changes rate to any of its rates.

The receiver

Acquisition looks for the preamble's 288 fixed symbols. Waveform 0 is to work at -6 dB, where the fixed symbols score about 0.2 against the 0.35 the other serial waveforms use. This one takes 0.13. Noise alone, searched over minutes at three offsets, never went above 0.075. Nor did a 188-110, 4285 or 4539 transmission. A preamble is taken only once twice its fixed section has arrived. The fixed section has sidelobes a Walsh symbol either side, and these clear 0.13. Taken any sooner, the search had once settled on one of them, and the header read was wrong.

The header is read from a symbol-spaced channel fitted to the fixed section. Each count and identifier symbol's four values are scored against what arrived, one symbol at a time, the chosen one then counted as known for the next. The count and identifier taken are the valid codewords those scores favour most. A longer preamble's count says how many superframes are still to come.

Equalization is the 4539 receiver's: the block path equalizer, its gains from the mini-probes, beside the RLS one. It takes the mini-probes' fifth roots of unity (the 48-symbol probe of waveforms 1 and 2) as known points. Waveform 0 is decided a Walsh symbol at a time by the path model alone, as 188-110's 75 bit/s is.

Losing the signal. These waveforms work below 0 dB, where the known symbols' error never drops below the other receivers' "signal gone" line: at -6 dB it is about 4 with the signal there. So reception ends when the error is both past that line and six times what it was in the preamble, sustained over a second or two. With the signal gone the error is past 30.

Streaming (stream.ReceiverWbhf) releases each interleaver block's bytes as soon as it is decoded, less the few that might begin the end of message.

Not here

• The 6-24 kHz bandwidths.
• The K=9 code. It is optional, and the receiver does not accept it.
• The single-symbol fixed section. A sender may shorten the fixed section to one Walsh symbol when it sends one superframe. Only the nine-symbol form is sent or looked for.
• Late entry at a shifted mini-probe. It needs the waveform agreed beforehand, which is not how the modem is set up.
• The optional end-of-transmission marker. It is not sent, and not looked for.

Measured performance

Table D-LII asks for a coded BER of 1e-5 with the long interleaver and a 20-superframe preamble, on white noise and on the 2 ms / 1 Hz poor channel. Each figure below is from eight transmissions of about 25 s each, through bateleur.sim with a 23 Hz offset, at the SNR the table gives (python -m hfmodem.bateleur.measure wbhf <mode> <channel> <snr>). That is 15 kbit to 3.2 Mbit a point, so the rates are shown to the bits counted, not to 1e-5 with any confidence.

Three points miss, each by about a decibel:

• 600 bit/s on the poor channel. The one transmission that failed lost a single interleaver block in a fade at 7 dB, and decoded whole at 8 dB. The failure at 4800 bit/s was in the same fading run.
• 12000 bit/s on white noise. It had no errors in eight transmissions at 25 dB. The equalizer's error is steady across every frame, so this is margin, not a fault: the 8/9 code leaves little room.

The other tests in D.6, each four transmissions:

• Static multipath, D.6.2, at 50 dB. No errors on any of the three channels: waveform 2 on 0, 3 and 9 ms (48 kbit), waveform 10 on 0, 2 and 4.5 ms (960 kbit), and waveform 12 on 0 and 1.5 ms (1.6 Mbit).
• Offset, D.6.4: waveform 10, long interleaver, poor channel, 30 dB. At +75 Hz, no errors in 960 kbit. At -75 Hz, 19 errors in 960 kbit. This test asks only that synchronization be kept.

At +75 Hz a transmission was first lost whole, because acquisition took the search from 0 Hz. That search read the offset 150 Hz wrong (a 16-symbol chunk cannot tell the two apart). The same preamble found from +50 Hz now wins on its far higher score.

• Doppler sweep, D.6.5. Swept at 3.5 Hz/s through the -75 Hz corner at 24 dB: decoded whole, in the test suite.