The following information was from a thread on FSK31, posted by Peter, G3PLX, on the FSK31 reflector.
Posted 18 November, 2000:
Subject: FSK31 Explained
>From Peter G3PLX
FSK31 has been mentioned a few times again recently on this reflector. Perhaps this is a good time to describe, for those who have not met it before, what FSK31, how it relates to PSK31, and put it into perspective.
FSK31 was not part of my original PSK31 system but was introduced by UT2UZ and has been carried forward into Digipan. It should really be called MSK31, since it uses Minimum Shift Keying, a well-established mode commercially. You will hear a lot of MSK in the 300-400kHz band transmitting DGPS telemetry. It's basically FSK but with the shift set to half the baudrate - the lowest shift you can use without the performance dropping off. For 31.25 baud this means the two tones are 7.8125Hz either side of the centre. If you think of the phase of a carrier that is 7.8125Hz low of centre, it lags by 90 degrees after 32mS, so MSK is actually equivalent to BPSK but using +90 and -90 degree shifts instead of 0 and 180. The beauty of MSK is that, like all forms of FSK, the amplitude is constant and so a linear transmitter is not needed. This is why FSK31 has been designed into the balloon experiments recently announced on this reflector, and why PA0OCD chose to use it on 2m with a particularly bad transmitter.
With BPSK the phase scope display shows two vertical lines, top and bottom, and with FSK31 it shows two horizontal lines left and right. If you think about it, since the noise is the same in all directions, the noise performance of FSK31 should therefore be exactly the same as BPSK. Given the advantage of constant amplitude, does this mean that FSK31 is a good idea?
To answer this we need to consider two things. Firstly there is a need to synchronise the receiver timing to the distant transmitter. Also, since our equipment (and indeed the ionosphere) is only marginally stable enough to work at bandwidths of 31 Hz on HF, we need to have AFC.
Let's first see how these things affect PSK31. If you picture the spectrum of a BPSK signal you can see that it's either a tone in the centre or two tones either side of centre, or somewhere between the two, but at all times it's obvious from the natural symmetry that there is going to be a way to locate the middle of the signal. Notice that the AFC system doesn't need to know anything about the data content or the timing. To get in sync. the PSK31 receiver derives it's timing from the 31Hz amplitude modulation on the signal. The Varicode alphabet has been specially designed to make sure there's always enough AM to keep the receiver in sync. Notice that we can extract the AM from the incoming signal even if it's not quite on tune. In PSK31 therefore, the AFC and the synchronisation are completely independent of each other.
Now lets consider MSK. The spectrum can be either a tone 7.8Hz high of the centre or a tone 7.8Hz low of centre, or somewhere in between. There is no natural symmetry now. If we want to design an AFC system that will do it's job right, it WILL need to know something about the decoded data. On the timing side, there is no amplitude modulation so the only way to get the timing is to decode the data and line up on the edges, as is done in other FSK systems like AMTOR and Pactor. But there's a catch-22 situation here: to decode the data we must be on--tune, but to be on-tune we must have AFC, but to have AFC we must decode the data.
This doesn't mean that FSK31 can never work at all, but in practice it means that both the AFC and the synchonisation will be more critical and prone to malfunction. This is inherent in any MSK system. It's not a problem where the frequency stability is good and we don't need AFC, but in amateur HF (and particularly VHF) working, it's going to give trouble. It's interesting to note that although MSK is used on 300kHz for DGPS, in Europe at least the DGPS systems in the 1.6-4MHz region are all PSK.
There's another problem that relates specifically to FSK31 as currently implemented. To make sure that there are always enough transitions in the data to keep the timing in sync, the data is differentially encoded, just like in AX25 packet. The idle signal becomes a 50:50 mark/space squarewave rather than a constant mark or space tone. But it does mean that one corruption on the radio link causes two corruptions in the differentially-decoded data. In AX25 this isn't a problem - even one error causes the whole packet to be re-sent, so a second error is no worse - but in the FSK31 scenario the second bit error could easily cause a second character error.
I am saying all this to warn people away from thinking that FSK31 might be the way to go. This isn't just a bit of "not invented here" on my part: I have already been this way and rejected it. I had all but abandoned the idea of a narrow-band HF RTTY replacement using MSK when I saw the BPSK work of SP9VRC.
I therefore suggest that we don't tread the FSK31 path, at least not for the kind of things we are doing currently with PSK31. There are ways of generating high-quality high-efficiency PSK31 which do not require linear transmitters, and one has already been described in this reflector using hard-keyed PSK and an auxiliary AM modulator. I would even suggest that for applications where linearity is difficult but band congestion is not a problem, we simply tolerate the extra sidebands!
Thanks for your time. 73 Peter
Followup, posted: 28 November, 2000:
>From Peter G3PLX
Since my last message about FSK31, in which I explained how it worked and compared it's performance with PSK31, I have learned that my explanation was wrong in one detail. I said that FSK31 was constant-amplitude and I now know that this is not correct, at least not in the implementation of FSK31 implemented by UT2UZ in Digipan (I will call it FSK31d here).
The keying in FSK31d does involve a change in amplitude, in fact a dip down to 70% of the maximum as the signal advances or retards in phase by 90 degrees. This is the same dip that occurs in QPSK when a 90 degree change occurs but in FSK31d there is a dip in every symbol so the power meter doesn't kick up as you type like it does in QPSK or BPSK. The spectrum width of FSK31d is exactly the same as that of BPSK and QPSK. If the amplitude had been constant, the spectrum would have been slightly wider. I suggested earlier that FSK31 should really be described as MSK31 but this is only correct for constant-amplitude FSK31. FSK31d is effectively BPSK with a phase offset of 90 degrees so it could be described as OBPSK (Offset Bipolar Phase Shift Keying).
The advantages of constant-amplitude FSK31 (MSK) over BPSK is that it doesn't get wider when the transmitter is overdriven. The 30% amplitude modulation present in FSK31d means that it WILL get wider when the transmitter is overdriven, but not by very much. In fact if you overdrive an FSK31d transmission, it never gets wider than an MSK signal. FSK31, as implemented for the high-altitude balloon experiments described in this reflector recently, is constant-amplitude FSK31 and therefore true MSK.
The performance comparisons I made between PSK31 and FSK31 remain valid. Although FSK31d has amplitude modulation, it not used in the synchronisation process. If it had done, the timing would suffer if the transmitter was overdriven and constant-amplitude FSK31 could never work at all! My point about the interaction between AFC and timing still applies and this was the fundamental reason why I discarded MSK myself in the early days. Using BPSK means that the receiver can pull into tune and into sync quicker and more accurately and I thought that this advantage of BPSK over MSK for live two-way QSO's was important. I knew that non-linearity in the transmitter would make the signal wider, but the typical performance of present-day SSB transmitters was enough to keep the signal as narrow as MSK would have been. In any case, Clover (which is effectively four simultaneous PSK signals) and Pactor-2 (two simultaneous PSK signals) had already been well established. I don't accept the argument that PSK31 is too difficult to set up: the ITU definition of the amateur radio service says that we are capable of training ourselves to get it right. It's just unfortunate that a typical soundcard will overdrive a typical mike input by 50dB!
I can see that MSK could have applications for VLF bands where the frequency stability is not a problem. It would also be suitable for beacons and broadcasts where it doesn't matter if it takes a bit longer to pull-in at the start, but in these cases there would be also a good case for using strong error-correction, something which was deliberately omitted in PSK31 because of the way it slows-down the live two-way QSO.