The new beta version of WSJT-X by K1JT gives us access to another new digital mode called FST4.
There are some similarities to FT8: it consumes a similar bandwidth, it’s also a single-carrier FSK (GFSK-4, to be exact), transmission period is (by default) 15 seconds.
I’ve created two wav files using WSJT-X: one containing FT8 transmission (12.5 seconds of actual audio) and FST4 (just 9.5 (!) seconds – leaving over 5 seconds out of 15 second period to the operator). FT8 on top, FST4 at the bottom:
Test setup
Just so you know – it’s not a laboratory setup, just a quick compare of an old, well-known mode with new and.. better?
I’ve created the following flowgraph in Gnuradio 3.8:
It can’t be simpler. It plays a .wav file, multiplies all of the samples with some predefined value – lowering the amplitude, then we add some random noise, and the noisy audio goes back out to an actual audio output, monitored by an instance of WSJT-X 2.3.
I’ve also written a simple Linux shell script to take care of running the flowgraph in the right time:
while [ 1 ]; do if [ date +%S == "29" ] || [ date +%S == "14" ] || [ date +%S == "44" ] || [ date +%S == "59" ]; then ./ft8_vs_fst4.py; break; fi; sleep 0.05; done
It starts playing the audio a second before period begins, so :59, :14, :29, :44. Convenient for testing.
Results
Let’s start with the new guy first, FST4. I’m playing the recorded message round and round with constant noise level, and gradually increase the signal level by a factor of 0.001. Finally, the message gets decoded with signal level of 0.041.
It’s almost invisible in the waterfall – look at 1000 Hz:
Let’s see how FT8 decoder deals with such a weak transmission. I increase the amplitude to 0.050, then 0.055.. and still no decode. Even though some people would call the signal a booming one, by the way it looks:
Only when I reach amplitudes of 0.059 and 0.060, I get some decodes:
I’ve spotted an interesting difference between the built-in decoders: with FST4, WSJT-X gives me the decoded message about 10 seconds into the transmission period – lot of time for me to take some action before the next period. With FT8, decodes appear only during ~14th second. And it’s not that I’m running a Pentium III under the bonnet – it’s a semi decent i7-4702MQ.
Conclusions
The big one? FST4 is better. It’s quicker and it decodes with much weaker signal than FT8.
Obviously it’s just a simulation and it may not translate well to the real life conditions, but in theory, FST4 can decode a message transmitted with 30% less power than FT8 needs.
The difference between 100 W and 70 W is about 1.5 dB. Not much, huh? Still, enough to finish an FT8 QSO, where you missed it by just 0.5 dB 😉
Morning update: we’re talking about amplitudes (aka voltages) here, so the difference in dB is 20*log(0.059/0.041) = ~3.16 dB. ~30% amplitude reduction is equal to a drop in power of a half. That’s even more impressive. A difference between 100W (FT8) and 50W (FST4).
You can check it out on your own – I’ve posted Gnuradio 3.8 flowgraph, wav files (feel free to try your own!) and the terrible runner script in my Github repository: https://github.com/olgierd/ft8_vs_fst4. Get WSJT-X 2.3 from https://physics.princeton.edu/pulsar/K1JT/wsjtx.html.
Olgierd SQ3SWF