Re: Crest Factor measurements + 80 meter MT63
That seems to be my emphical observation as well. I adjust my three
watt warbler on either PSK31 or MFSK31, by peaking the power, and
backing it off slightly. With this method, I don't have any problems
working anyone, and reports of signal quality are good.
Now, if I go to MT63, I noticed the power (average) is at least 10db
lower then what I am seeing on MFSK16 for the exact same settings.
This would tend to confirm the high crest factor of MT63.
I'm assuming this is the correct way to set up MT63, lacking an ALC
meter. It should be noted, the one time I tried coming back to a MT63
CQ, I had adjusted MT63 with the "peak and backoff" method, and
needless to say, they didn't come back to me. Now I am first tuning
up on MFSK16, then using those settings for MT63. Just haven't seen
anyone since then to respond to!
Anyone interested in a 80 meter MT63 QSO later tonight? I can operate
3581-3582 with my warbler, so 500hz bandwidth is best.
--- In MT63@yahoogroups.com, Tomi Manninen <oh2bns@s...> wrote:
> On Wed, 2003-09-10 at 17:08, Robert McGwier wrote:
> > Making the assumption (which I will check) that these files
> > of "perfect" collection (recordings) were not already clipped,
> > the crest factor is nowhere near 13 dB but is more like
> > 6 or 6.5 dB (again, as Tomi said, depending on the data).
> My data was calculated from numerical sample data directly
> from the modulator code. In fact Pawels own code also has tools
> to do this:
> tpm@oh2bns ~/src/mt63 $ ./mt63tx -tsound.h -sout.raw
> Multitone MT63 modem for Linux - transmitter, (c) 1999 Pawel
> Made to be compatible with the MT63ASC for the EVM56K/DSPCARD4
> No audio device selected
> We write the audio to out.raw
> Requested rate = 8000 Hz, device reported rate = 8000 Hz
> Modem bandwidth is 1000 Hz with SINGLE (32) interleave
> MT63 transmitting text from sound.h ...
> // definition file for sound.cc, (c) 1999 Pawel Jalocha
> #ifdef __linux__
> ... clip clip ...
> #endif // __linux__
> Flushing data interleaver ...
> Sending jamming waveform ...
> Closing audio device ...
> Stopped OK.
> 1280800 samples sent out = 160.1 sec
> tpm@oh2bns ~/src/mt63 $ ./peakrms out.raw
> out.raw: 1st pass, 1280800 samples
> DC = -0.000, RMS = 0.167, Peak = -0.872, Peak/RMS = 14.4dB
> out.raw: 2nd pass, 1280800 samples
> Samples above RMS: 31.39%, above 2*RMS: 4.64%, above 3*RMS: 0.32%
> Power above RMS: 48.79%, above 2*RMS: 8.55%, above 3*RMS: 0.65%
> That figure is even bigger (14.4dB) than the one I gave. That is of
> course because the peak value is pretty much random. If I use my
> software to calculate the crest factor from the same data in blocks
> of 65536 samples, I get 20 figures, most of them are between 12.5
> and 13.5, one hits that 14.4.
> I suppose those might be a bit theoretical and pessimistic values.
> With real soundcards and rigs I think you're bound to clip the
> waveform without even noticing.
> I while ago I did some testing with MT63 and clipping. Simply hard
> limiting the waveform to about 7-8 dB crest factor raised the
> level to about -40 dB (is this acceptable? remember it's only inside
> the audio bandwidth). Any attempt to filter the sidebands only
> in the crest factor growing back up (surprise...). I didn't do very
> thorough testing however and might have screwed up something as
> Tomi Manninen / oh2bns@s... / KP20ME04
- On Fri, 2004-01-09 at 21:08, wb8wka wrote:
> That seems to be my emphical observation as well. I adjust my threeWell, at least in this way you can be pretty sure you are not over-
> watt warbler on either PSK31 or MFSK31, by peaking the power, and
> backing it off slightly. With this method, I don't have any problems
> working anyone, and reports of signal quality are good.
> Now, if I go to MT63, I noticed the power (average) is at least 10db
> lower then what I am seeing on MFSK16 for the exact same settings.
> This would tend to confirm the high crest factor of MT63.
> I'm assuming this is the correct way to set up MT63, lacking an ALC
modulating. You might be a bit on the safe side, but that is only
There are several complications in this issue. One of them is that
the modem code needs to be set somewhat experimentally. If you
assume that the sound card uses signed 16 bit samples, you can
write sample values between -32768 and +32767. With modes like
MFSK16 (where the envelope is constant) and PSK31 (where the envelope
is not constant but well defined) it's easy to write a modulator
that makes the best use of the available sample value range without
any danger of overflows.
The stochastic nature of the MT63 envelope makes this much harder.
One can calculate the theoretical maximum value of the envelope
but this value is bound to be *very* rare. In fact it's likely to
never occur. So in practise the output range needs to be experimented.
What this means is that it is likely that the output level from a
MT63 modem is a bit lower than the optimum. Which in turn means that
your method results in being on the safe side.
Another issue is that most power meters read somewhat random values
with MT63. Most meters are by nature peak detectors (detecting the
peak RF value and then "calculating" the RMS value). They show
correctly only with a pure carrier. With the complex envelope of
MT63 they show "something" that depends on the time constants
involved (filtering + meter time constant). I guess typically they
read something between the real peak and average values.
I think adjusting levels is the most common problem with MT63
(and Q15X25 which is similar in this respect).
Tomi Manninen / OH2BNS / KP20ME04