Good evening everyone!
I have a question concerning data transmission via FSK.
As far as I know, conventional methods encode symbols via matching a sequence of bits to some frequency with fixed symbol length.

Why is it not common to compose symbols by combining a number of frequencies? For example, we can have a set of 3 frequencies, numbered 0, 1, 2 and compose words of given length T like '000', '12221'.

Of course , this setup requires f0 = T/3, f1 = T/4, f2 = T/6. But it does give an increase in bitrate. What are some possible conerns in this case?
Thank you very much!

The usual case for FSK is 1 or 2 bits per symbol.  That would be 2-FSK or 4-FSK.

4-FSK with RRC shaping on the base band (before modulation) gives a very nice fall off shape.  Pretty much has to be a Software Defined Radio because not much supports custom modulation out of the box.

3 frequencies would give you 1 1/2 bits per symbol.  Might be useful with some kind for error correction scheme.  I've never seen a paper on it being done but sure it could be.

There are HAM standards with 8 or more frequencies using layered error correction.

If you are asking why multiple frequencies are not lit up at once it is a problem with the transmitters.  FSK TX almost always means constant envelop modulation.  With multiple frequencies the waveforms would constructively and destructively add.

There *is* a constant envelope OFDM modulation using FSK modulation but it is a special case.  The "negative" frequency lanes are complex conjugates of the positive ones and the result is constant envelope.  I think there is a Military standard like this.

FWIW,

Mark Napier

Thank you very much for your answer!
I'm sorry for getting back this late.

I guess my question was more avout creating distinct patterns using 3 given frequencies. We then skip the demodulation and search for patterns in the signal directly. Is there something I am missing? I'll try to make some images for better understanding of the scheme.

I think I understand.  So for instance, with 3 tones you could send 3 bits of information per symbol by simple binary encoding.

Then have a narrow FIR filter on each frequency.  Say a preamble and a sync to establish the average power and the symbol center and start on each frequency.

Would essentially be OOK on each independent frequency.  Maybe some fancier scheme with forward error encoding.

But still, the bandwidth is wider and the transmitter is no longer constant envelope.  Multiple tones even more so.

FWIW, although non-coherent FSK is ~3dB better than OOK in reception, OOK is ~3dB more power efficient at the same peak power.  So it's about a wash from a TX power/link budget point of view.  *BUT*, OOK needs more bandwidth and has bad roll off due to the pulses.  And the chip needs to support it.  But simplicity and low power means that a lot of older standard battery powered utility devices still use it.

No cheap off the shelf chip is going to support multiple tone OOK.  So it would be an SDR.  And an SDR to receive it.

With that much work there are just much better options available.

Mark Napier

Well, I certainly don't understand some things that you've said. That's my lack of training.

But you've said it right: this is all done via software. Well, the signal is analogous, it is then digitized and piped into deciphering software.

The idea was actually to skip FIR altogether. Except for selecting the signal frequency band. The decoding is supposed to be done in the passband without transforming the signal into baseband by correlation-matching all the possible templates.

If you are keen on any new idea that is good but don't go public until you can model it and see for yourself. A great obstacle is to change the current systems, or current standards, vendors and users equipment that they have invested heavily even if your solution is better.

I don't claim that my solution is better. I don't even know if it will work. Well, it seems to work for a simple setup, but the idea is so simple that I wonder why it isn't widely used. My intuition makes me sceptical because if it was really that simple, everybody would be using it.

I guess, my general questions is: why use demodulation at all?
For example, if we have BFSK and an SDR, why isn't it common to just match 2 templates to locate their positions in a signal?