MFSK

Instead of 2 tones, MFSK uses 4 to 64 tones: each symbol is one tone and carries 2 to 6 bits. At the same data rate the symbols are much longer than FSK bits, which makes MFSK very robust against noise, echoes and different clocks of the sender and the receiver. It is the best choice for speakers and microphones. MFSK is an extension of TinyModem: it is not compatible with other programs.

Advantages

• Very robust against noise: mfsk64 decodes without errors when the noise is stronger than the signal.
• The receiver adapts to the clock of the sender: different clocks of ±3% don't matter.
• Long symbols are robust against echoes.
• All tones are between 1000 and 3000 Hz (telephone band).

Disadvantages

• Slower than FSK and OFDM: 23 to 125 bytes/s.
• More tones need more CPU in the decoder.
• Not compatible with other programs.

Modes

Mode	Tones	Symbols/s	Data rate	Tone spacing
mfsk4	4	500	1000 bit/s (125 bytes/s)	500 Hz
mfsk8	8	250	750 bit/s (94 bytes/s)	250 Hz
mfsk16	16	125	500 bit/s (62 bytes/s)	125 Hz
mfsk32	32	62.5	312.5 bit/s (39 bytes/s)	62.5 Hz
mfsk64	64	31.25	187.5 bit/s (23 bytes/s)	31.25 Hz

Fewer tones are faster, more tones are more robust. mfsk16 is the default.

Usage

I2SStream i2s;
MFSKStream modem(i2s, RXTX_MODE);

void setup() {
  MFSKConfig cfg = modem.defaultConfig();  // mfsk16
  cfg.sample_rate = 22050;
  ...
  modem.begin(cfg);
}

echo "Hello" | tinymodem --tx -f hello.wav mfsk16
tinymodem --rx -f hello.wav mfsk16

MFSKConfig, MFSKEncoder, MFSKDecoder, MFSKStream and MFSKStreamRTOS are the MFSK classes. See the mfsk-loopback and packet-modem examples. On the command line, -R, -v, -M (lowest tone), -q, --rx-one, --packets and --Xrxnoise are supported.

MFSKConfig field	Default
tones	16	4, 8, 16, 32 or 64
symbol_rate	125	Symbols per second
base_freq	1000	Frequency of the lowest tone (Hz)
tone_spacing	0	Distance between the tones (Hz), 0 = symbol_rate
volume	1.0	Transmit amplitude
preamble_symbols	8	Length of the preamble
min_amplitude	0.005	Receiver: minimum signal level (1.0 = full scale)
rx_one	false	Receiver: stop after the first transmission
channel	0	Receiver: input channel
sample_rate, channels, bits_per_sample	48000, 1, 16	Audio format

Call setMode("mfsk32") first and then change individual fields. The highest tone must be below half the sample rate, and a symbol needs at least 8 samples.

How it works

A transmission consists of:

1. A preamble of 8 symbols, alternating between the lowest and the highest tone. The receiver uses it to find the start of the symbols and to measure the clock of the sender.
2. Sync symbols which carry at least 8 bits (tones / 2, tones / 4, …). They mark the start of the data.
3. The data: each symbol carries log2(tones) bits, most significant bit first. The tones are Gray coded, so a confusion with a neighbouring tone only changes one bit.
4. Two symbols of silence: the receiver detects the end of the transmission when the signal is lost.

The receiver analyses each symbol with one Goertzel filter per tone (AudioTools GoertzelDetector). The symbol timing is tracked with an early/late gate and a second order loop, which also learns a constant drift. If the sender's clock differs, all tones are shifted by the same factor: the receiver measures this factor with the highest preamble tone and retunes all tones and the symbol length.

Measured performance

Loopback with the unit tests at 8000 Hz. Noise is uniform noise which is added in the receiver (rx_noise, --Xrxnoise): 0.5 corresponds to a signal-to-noise ratio of about 7.8 dB, 1.0 to 1.8 dB and 1.5 to −1.8 dB.

	Without errors up to
mfsk4 noise	0.8
mfsk16 noise	1.0
mfsk64 noise	1.5
Clock difference of sender and receiver	±3% (all modes)
Volume	0.01 (1% of full scale)

All modes also decode 50 short transmissions in a row with 2% clock difference and noise 0.5 at 8000 Hz and at 48000 Hz. On a PC at 8000 Hz, the decoder ran about 2500 (mfsk4), 1100 (mfsk16) and 300 (mfsk64) times faster than real time. Check your board with the benchmark example.