Re: LF: Fwd: CMSK Beta released

[Andy Talbot <andy.g4jnt@googlemail.com>]

Re DDS Derived MSK...

This approach is one of the all time bodges that just about every 'average' implementation of MSK for normal links makes use of.    It is to asynchronously switch between two frequencies separated by as close as possible half the baud rate.

 

The CMX589 chip is a classic implemantation of this, 'pretending' to generate Gaussian filtered GMSK by applying a filtered waveform to an FM generator.  It needs careful setting up.  Very careful or the result is just filtered FSK.

 

It is not true MSK.  The generic term, and the correct one, is Coherent Phase  Keying  CPK as the phase glides in one direction then the other without being hard switched as in PSK.   But the change in phase is not coherent with the modulating signal using this approach.    Going back to the DDS version in the reference given,  the two frequencies programmed into the DDS are derived from the DDS clock, but they are toggled by the data line at its own clock rate, which doesn't have to be derived from that controlling the DDS. in fact it probably isn't except perhaps indirectly via a PLL   If the same master oscillator were to be used it could perhaps be argued that true MSK could result.

 

The approximate-but-good-enough, implementation of GMSK in the CMX589 chip has just about driven the whole VHF/UHF low cost data communications area over the last couple of decades,  and listening on those bands you will probably find more 9600 baud CPK than FSK or FM voice now - certainly with all the Utilities' SCADA links and remote monitoring going on.    The shipping AIS system is built around the standard, and every AIS box I've ever looked inside has one or two CMX589 chips doing the work.

 

True theoretical proper MSK is rare.  Many of the LF submarine broadcasts are a superb example, and if you get the chance, capture one of these, mix down to zero frequency and display on an X-Y vector scope.   If you can slow the plotting rate right down, watch how the vector rotates round first in one direction for one data polarity, then back again for the other, speeding up and slowing down its rotation rate.  Its a rare sight to behold :-)    The Old GBR signal on 16kHz now defunct was fascinating to watch like this..  

 

For optimum Viterbi decoding. proper MSK is essential,, as the search back through the trellis looks at the history of the phase transitions to best match what could have been sent with what it has just received.  It can't do that if the difference is an undefined "almost half a bit rate" frequency shift.

 

So, conclusion:  the DDS / CMX589 Approximation  route will work quite adequately for average S/N links, , but you can't get the benefit of a properly constructed Soft Decision FEC decoder.  S0 that solution is not a weak signal mode.   Proper true coherent MSK can be.

 

Andy
www.g4jnt.com

On 6 August 2010 15:13, Michel F5WK <f5wk@mterrier.net> wrote:

Andy,

I found on the web a few interesting pages of the book:  "Research on the DDS to Generate MSK Signal"
<
http://books.google.fr/books?id=ITOlbhvbAbQC&pg=PA36&lpg=PA36&dq=MSK+DDS&source=bl&ots=wcQBVR20gz&sig=ETkZuXXeYofIAxL7HvqtrVULclU&hl=fr&ei=mg9cTPe-I5K94ga-hdicAg&sa=X&oi=book_result&ct=result&resnum=3&ved=0CCgQ6AEwAg#v=onepage&q=MSK%20DDS&f=false
>

Michel - F5WK

> MSK does not lend itself to generation on a PIC.   It is NOT a two tone
> mode, but a waveform that transitions back and forth  around a circle in
> phase space.  Do not think of MSK as frequency Shift Keying
>
> It may be possible to do it on a on a PIC.  By not touching the frequency
> register, but reprogramming the phase offset register in real time at a fast-enough sampling rate.
>
> But it would need a high sampling rate  to get it right while keeping
> alias sidebands at the sampling rate at a low enough level (sin(x) / x and all that)
>
> PSK31 on a PIC, sampled at 7812.5Hz (250 samples per symbol) gives sidebands at +/- N.Fs of < 60dBc.
>

> Andy
> www.g4jnt.com


> On 6 August 2010 09:22, Michel F5WK <f5wk@mterrier.net> wrote:
> Hi Graham, LF,

>> As  far as I know , only wspr  and  ros-mf have  the  ability to  drive  a
>> pic/dds  vfo , with  only    mf-1/7  giving  a  tailored mf  qso  mode ?

> What about asking for a mod allowing support of a DDS ?

> MSK/CMSK has only two tones, whose frequencies are known. Therefore only one
> bit of information is required for one symbol.
> The big question is how much accurate the deltaF has to be ...

> Graham, since I swapped the link between the PC and the LF/MF transmitter
> from USB to BlueTooth, the timing has slightly degraded, and as a result
> ROSmodem is not working anymore. I have to resync the timing at the PIC
> level, definitely the correct way ...

> 73,

> Michel - F5WK