Content-Transfer-Encoding: 8bit
Message-ID: <65AECDF1F89AD411900400508BFC869F9C04B9@pdw-mail-1.dera.gov.uk>
From: "Talbot Andrew" <ACTALBOT@dera.gov.uk>
To: rsgb_lf_group@blacksheep.org, 
 "Holtby Linda C" <LCHOLTBY@dera.gov.uk>
Subject: LF: RE: Wimborne talk on Wolf
Date: Mon, 21 May 2001 14:10:29 +0100
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1; format=flowed
Precedence: bulk
Reply-To: rsgb_lf_group@blacksheep.org
Sender: <majordom@post.thorcom.com>

Replying to Stuart's comments re Wolf signalling .
This is quite a long EMail so if you have no interest in real radio
comms engineering switch off now :-(

> I was thinking some more about James' talk on Wolf yesterday,
> and I have a few basic questions.
> Wolf consists of a data encoding system layered over an
> error recover system, layered over a bit synchronisation system
> finally layered over bpsk. The upper layers woud work fine
> over any transmission layer ie dfcw. Is BPSK actually the most
> optimal bit transmission system?

How is convolutional coding going to work with a fuzzy mode signalling
waveform such as DFCW ?. By Fuzzy Mode, I mean it requires human
intervention to decode the signal.   If you mean Frequency Shift Keying,
see below.

In an environment where the distortion mechanism is purely additive
noise and interference - such as seems to be the case at LF, with no
significant multipath or coherent time delayed interference - BPSK will
nearly always be the optimum solution for the signalling waveform.  This
is due to its bipolar nature ie +/-1 as compared with any orthogonal
solution such as FSK where the descision is being made between 0/1 - ie
there is a 3dB advantage immediately.   Practical matters such as clock
recovery and frequency error can make the descision less clear cut, but
on a purely mathematical basis BPSK will always win in a noise + QRM
only environment

> Am I correct in thinking that BPSK is so good because you are
> actually repeating the bit on every carrier cycle and integrating
> the result?

Integrating over multiple bits is nothing to do with it being BPSK -
that technique could be used for any signalling waveform.  The 16 times
integration - if it is applied to a coherent waveform, ie. voltage
summing, can give a 16^2 times improvement or 24dB in S/N which is the
case for Wolf   If the summing is incoherent, ie using Power (which is
VERY unlikely to be the case for any BPSK demodulator, but would be the
case for incoherent wqveforms such as FSK) the improvement would only be
16 times, ie 12dB.

> I was also thinking that you could send wolf on the divide down
> CW transmitters that some folks use. By injecting the audio
> tone from a PC into a 13.6MHz transciever operating SSB and
> then dividing down by 100 you should generate an equivelent
> signal to the linear translation approach save for the envelope

To generate a 180 degree  phase shift by dividing down cannot be done.
Phase shift scales with division ratio, and if dividing by 100, this
would require 180 * 100 = 18000 degrees phase shift at the fundamental.
This, of course, is a multiple of 360 degrees so would not give any
phase shift at all.   To generate the shift would still require a
separate modulator.    The converse is also true of course.   To
generate 180 deg on a mutiplied signal requires a lower shift at the
fundemental.  eg to generate BPSK at 1296MHz would require just a 0 / 15
degree phase shift to be applied to a 108MHz drive signal.   This is the
reason why old style VHF FM transmitters (before synthesizers were in
use) always started with a low frequency crystal and multiplied up. The
phase shift - and hence FM - could be generated in a simple circuit and
increased by multiplication.   

It would probably work if you used an odd division ratio though.

> shaping. However by keying the carrier off-on during the transition
> the normal CW wave shaping should clean up the signal.
> Does this work? Note that the error in the carrier will be the
> dominant term, and not the error in the the modulation, which may
> have some advantages.
> 
As we tried to say at the meeting, it is NOT just a case of switching
the carrier off at the phase transition point then back on.   The
amplitude has to be slowly ramped down, the phase switched, then the
amplitude ramped back up again.   The width of the sidebands is directly
related to the speed of this ramp.   Therefore, a ramp of 1ms from full
carrier to zero would result in sidebands 1/(2.pi * 1ms) Hz wide -
approx 160 Hz at some defined level.   A 2ms  ramp 80 Hz at this same
level and pro-rata. The shape of the ramp dictates how fast the higher
order sidebands roll off.  A linear ramp is poor, the high order
sidebands ones roll off slowly, whatever their initial level may be.
However, a ramp based on the shape of half a sinewave (the so-call
raised cosine shape) gives ones of the best roll offs of all.  

I suggest you fully read the article on PSK31 By G3PLX that appeared in
RadCom a few years ago.  PSK31 is the ultimate case of waveform shaping
where the complete bit interval is a half sine wave and a 0/1 repeat
cycle gives two single tones separated by half the baud rate - and
nothing else.   But in any PSK mode, there will always be a trade off of
bandwidth vs. signalling efficiency.   PSK31 throws away several dB of
Signal / Noise performance to achieve a very narrow bandwidth.   The
VE2IQ system on the other hand works best if no shaping at all is
employed.    

Where waveform shaping is used, any system works best at optimum Signal
/ Noise  when the receiver is exactly matched to the transmitted
waveform - a so called matched filter technique - even if this means the
Rx appears to take in the signal over a very wide bandwidth.   It is
collecting as much of the signal as it can and processing this correctly
to give the best Signal to Noise ratio possible - Whatever the bandwidth
of the signal may end up as.   It could well be that the energy of a
signal keyed at  1Hz is spread over 1MHz bandwidth for very sharp BPSK.
BUT if the receiver takes in every 1Hz whisker over this entire bandidth
it will give better decoding and optimise S/N than if the signal were
filtered to 1 or 2Hz bandwidth before demodulation.

For data communications you need always to take a holistic approach and
not just consider bandwidth, filtering, modulation type data rate etc.
as separate entities.   All are closely related and it may be necessary
to separate out or sacrifice one parameter for the sake of optimising
other factors such as bandwidth, resistance to interference etc.   For
example, at HF the dominant interference is often not noise but
multipath.  Here parameters such as data rate need to be optimised to
counter the several milliseconds of multipath, and often the best HF
waeforms are those that take up a whole 3kHz bandwidth and are
subsequently reduced by coding and repetition to allow data rates that
can be a slow as 70 Bits / s.   Needless to say these are not favoured
by radio Amateurs - but can often be heard all over HF these days
sounding a bit like a diesel engine chuntering away.   The repetition
rate is the repeat length needed to test and measure the multipath and
repeat data if necessary.

Whew.........!

Andy  G4JNT
> 


-- 
The Information contained in this E-Mail and any subsequent correspondence
is private and is intended solely for the intended recipient(s).
For those other than the recipient any disclosure, copying, distribution, 
or any action taken or omitted to be taken in reliance on such information is
prohibited and may be unlawful.