Hi Neil,

I agree with what Ron said. Also, note that even in FSK the phase is
continuous, so I wouldn't call the instant frequency jump a "sharp
transition" (but this is arguing about names, anyway).

It's true that the GFSK version includes amplitude ramp-up and ramp-down
at the start and end of the transmission to prevent the keying click
associated with going to full TX power instantly. However, the shape of
this ramp-up/down is not very critical, so even if this ramp is sent
through a non-linear amplifier, it's OK. The amplifier would need to be
really bad to transfrom the ramp into a large keying click.

I believe that the following is true (at least mathematically):

If a constant amplitude waveform is amplified non-linearly, then the
spectrum of the amplified signal is very similar to the spectrum of the
original signal (disregarding harmonic distortion that makes energy
apppear near the integer multiples of the carrier frequency).

This means that there is no noticeable spectral regrowth by sending an
FSK or GFSK waveform through a non-linear PA.

I don't know how the proof of this statement goes exactly, but I just
simulated it with GNU Radio, comparing distorting a GMSK signal with
BT=0.35 and another with large BT so that it looks like unfiltered MSK.

So I don't aggree with "With a nonlinear amplifier well into saturation,
the level of the energy away from the main tone will be boosted, so
there is distortion of the relative amplitudes of the tone and sidebands."



El 13/1/20 a las 21:35, Neil escribió:
> OK, agreed that the envelope in even a sharply-switched FSK is sort-of
> constant at baseband (except during tone startup, when it goes from zero
> to full amplitude in the first half-cycle), but with the sharp
> transitions, some energy goes into the modulation sidebands created by
> those transitions. The level of the main tone during a sharp transition
> must drop as energy is being put into the sidebands generated by the
> abrupt frequency change. With a nonlinear amplifier well into
> saturation, the level of the energy away from the main tone will be
> boosted, so there is distortion of the relative amplitudes of the tone
> and sidebands.  In a linear amplifier, the changes in level of the main
> tones during frequency transitions will be reproduced faithfully and the
> overall bandwidth will remain as per design.  Also when operating close
> to the edge of a filter passband, a fair amount of the energy in those
> sidebands is removed during tone transitions. The amplitude variations
> only happen around the time of a tone transition, so it depends on how
> you define the granularity of your amplitude measurements.  I should run
> a proper test on the GFSK version.
> Anyway, none of this really matters unless you happen to be in line of
> sight to a 30kW ERP station on 144MHz and you are trying to resolve a
> signal 400Hz away that is 80dB weaker.
> The change to GFSK is *hugely* welcome as it has reduced the occupied
> bandwidth of FT8 signals (even those using over-driven or
> not-very-linear amplifiers) to a much more neighbourly level. 
> Nice work by the OP.
> Neil G4DBN
> Average amplitude spectrum of an old-style FT8 signal at baseband.
> Spectrum after going through an SDR transmitter.  Blue is a 15-second
> average of the envelope, yellow is instantaneous. Some of the sidebands
> have gone missing, so there is amplitude variation, at least very
> briefly during tone transitions.
> On 13/01/2020 19:48, Ron Economos wrote:
>> There are no amplitude variations. Take a look at the figure from
>> Daniel's link. What's being filtered is the *frequency* transitions.
>> 4fsk
>> Ron
>> On 1/13/20 11:23, Neil wrote:
>>> The time-constant length (attack/sustain/decay times) is the root of
>>> the original problem with ALC of course.  Is there a way to store the
>>> averaged envelope and remember it in a variable used to set the gain,
>>> so you don't have to recalibrate every 15 seconds with each new
>>> transmission, or syllabically with SSB voice?
>>> GFSK will work fine through a non-linear amplifier, but you lose the
>>> benefits of the new smooth transitions.  The amplitude variations are
>>> only there during start, end and transitions.  The old way used to
>>> look like this in the frequency domain.  That startup splat is more
>>> than 2kHz wide.  This is audio direct from WSJT-X to Spectrum Lab
>>> within the same PC.
>>> I don't have an image of the new GFSK, but it was much cleaner when I
>>> checked.  The FSK transitions in this are done at zero crossing, but
>>> even so, that represents a step change in slope and generates a
>>> spread of frequencies.
>>> Neil
>>> On 13/01/2020 19:05, Daniel Estévez wrote:
>>>> Hi Neil,
>>>> As far as I know the new GFSK of FT8 and FT4 are still constant
>>>> envelope, so they are tolerant to non-linear amplification (but don't
>>>> read this as "immune to all sorts of terrible clipping and distortion").
>>>> See the bottom of page 4 in
>>>> https://physics.princeton.edu/pulsar/k1jt/FT4_Protocol.pdf
>>>> You raise a fair point that nowadays with SDR it is often enough to
>>>> adjust the gain in an open-loop fashion, by monitoring the output power
>>>> and changing the gain until an appropriate level is found. This can
>>>> often be done once in a set an forget fashion. I do that for my QO-100
>>>> groundstation.
>>>> However it's true that the gain of PAs can vary somewhat with
>>>> temperature and frequency, so sometimes some sort of close-loop
>>>> adjustment of gain (of an appropriately large time constant and
>>>> everything to prevent distortion) would be better.
>>>> Best,
>>>> Daniel.
>>> -- 
>>> Neil
>>> <a href="http://g4dbn.uk/";><small>g4dbn.uk</small></a>
> -- 
> Neil
> <a href="http://g4dbn.uk/";><small>g4dbn.uk</small></a>

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to