[RTTY] 300hz or 500hz IF filter?

[Kok Chen]

(Warning... as often true with my postings, this is long and technical.  Please don't hesitate to use your Delete key).

On Aug 23, 2013, at 8:19 AM, Jay WS7I wrote:

> Joe-
> 
> You are just simply wrong. 


When Joe threw out his red herring, he also caused the original thread to splinter, in a non-obvious fashion for those who do not intimately follow RTTY.  I am sure people who build demodulators know full well what I am going to say, but I want to clear it up a little for the general audience.

The original poster had asked for what receiving filter to use, which has to do with the demodulated bit stream, and not of some bureaucratic definition of the "bandwidth" of a transmitted signal, which the ITU chooses to use as a standard.  The former has to do with what *receiving* filter to use and the latter has to do with some standard to classify different emission modes (i.e., transmitter bandwidth).

As a few of us (including Andy K0SM and myself) have shown, especially in the past year, a transmitted RTTY signal can occupy very different transmitted bandwidths.  And all are different from what the ITU states.  

The widest form of transmitted RTTY is the so called "phase coherent" AFSK or FSK generation.  With this method, the Mark and Space tones come from two different coherent sources (which can be individually phase locked at the receiver).  And to bit modulate for RTTY, you simply switch between these two oscillators.  The Mark and Space are treated as two separate OOK (on-off keying) signals.

(This, by the way, is probably closest to what the ITU consider to be FSK "bandwidth.")

The result is a very broad sin(x)/x spectrum for each Mark and Space signal.

We know that the best narrow band receiving filter for RTTY is a Raised Cosine that goes to zero outside a single fundamental of the keying signal (i.e., 45.45 Hz divided by 2), so all those extra keying sidebands from a phase coherent FSK signal is completely wasted.  Not only are you QRM'ing other stations, but you are also wasting transmit power that you can otherwise focus on to the main Mark and Space lobes (not much wasted power, but the extra SNR can probably affect your contest score -- so it behooves a contester to use as narrow a transmitted signal as possible -- thus my RTTY TRansmit Filter article which I mentioned a day ago).  When you clean up your signal, not only are you a better neighbor, but you are also spending all your transmit power where it matters to the receiving end.  This probably outweighs any "millisecond shaving" efforts.

The next form of RTTY generation is called the "phase continuous" FSK/AFSK.  In this form, a single oscillator is used to represent both the Mark and Space carriers.  However, when you switch between Mark and Space, you do it in a manner so that the instantaneous phase does not change during a transition.  

Back in the 1960s, Irv Hoff had suggested a form of AFSK generation where you switch between Mark and Space right when they cross zero voltage.  Mathematically, it is absolutely identical to the phas continuous case, but suffers from the fact that you need to choose Mark and Space frequencies that perfectly divides 22.0 milliseconds.  I.e., you no longer have the standard 2125/2995 Mark/Space tones.  This method is so identical to the phase-continuoys method that it has the same precise higher order discontinuities. No one uses Hoff's method in software, since phase continuous is just as easy and much cleaner to implement.  

(Switching at zero crossing is used in PSK31, by the way as a mechanism to reduce bandwidth; but they only have a single carrier frequency to worry about, not the two we have in RTTY.)

The EE's among you will immediately see how much the keying sidebands are reduced by going to phase continuity.  However, just because instantaneous phase is continuous, the higher order derivatives of phase cannot also be continuous (otherwise the signal will be output as a single unmodulated tone, HI HI).  These higher order derivatives will also generate unwanted and unneeded keying sidebands, albeit, very much attenuated compare to the phase coherent case above.

The next step is to get rid of the unwanted keying sidebands from the phase continuous RTTY.  This can be done in software/firmware by using waveshaping, or can be done in software/hardware by applying a narrow bandpass filter.  Examples of this: 2Tone uses waveshaping and cocoaModem uses filtering.

This extra filtering can be implemented both in the modem or in the FSK transmitter, as the K3 has shown when Elecraft modified their firmware after Andy showed how dirty an FSK signal from a K3 was before the modification.

OK, none of this is new.  I am just trying to refresh everyone's memory on what transmit occupancy of an RTTY signal means.  Indeed, I had written the "FSK Sidebands" article back in 2005 that shows the spectra of the three methods (phase coherent, phase continuous and filtered phase continuous) mentioned above:

http://www.w7ay.net/site/Technical/RTTY%20Sidebands/sidebands.html

Note that although the ITU assigns a single "bandwidth" number to all three cases of RTTY (ITU does not distinguish between them), they do not have the same actual bandwidths.  

Now, as to receiving filters, a Raised Cosine filter will behave identically to all the three transmitted methods.  An optimal Raised Cosine will indeed lop off lots of the keying sidebands from any of the transmitted signal above.  The transmitted keying sidebands are neither used by the demodulator, nor are they helpful for RTTY demodulation when a Raised Cosine is used at the receiver.

K6STI's RITTY and cocoaModem both use Matched Filters instead of Raised Cosine filtering.   As such, they have a slight edge with both phase coherent and phase continuous signals over the filtered RTTY transmitters.  By how much?  Take a look at Figure 4 here:

http://w7ay.net/site/Technical/Extended%20Nyquist%20Filters/index.html

For the signals that we encounter in weak signal RTTY (about 3% character error rate), the Matched Filter beats out a Raised Cosine by perhaps 0.25 dB.

However, a Matched Filter is very wide and really only suitable for digging out your call sign from that elusive weak DX.  There will be too much QRM under contest conditions to use a Matched Filter, where what you want is a Raised Cosine (a Raised Cosine is way more narrow than any crystal filters than can be manufactured without group delays).

Indeed, it is the quest for some filter whose bandwidth that is in between a Raised Cosine and a Matched Filter that lead to my work which resulted in this "Extended Nyquist Filter" article.  It is probably the crown jewel of my RTTY work.  The recursion in both the time domain (impulse response) and frequency domain (transfer function) was especially satisfying :-).

In conclusion: you can use the ITU bandwidth as a *regulatory* number (like when you need it as an argument to present to the FCC when the ARRL again tries to force bandwidth numbers on ham bands), but they don't really represent the real world of RTTY emissions.   Look at the spectrum that K0SM recorded and which Jay displayed in this past ARRL Roundup results.  The bandwidths are all different, and I think all of them have equivalent bandwidths that are narrower than the ITU number :-).

Furthermore, that ITU transmitted (occupancy) bandwidth has *nothing* to do with optimal receiving filters.  Narrow, optimal receive filters that presents no intersymbol interference has been with us since Harry Nyquist's 1928 paper.  Nyquist really developed it for landline telegraphy (OOK).

73
Chen, W7AY