[RTTY] Re: Noise with MMTTY

[Kok Chen]

On Mar 23, 2004, at 9:07 AM, Bill Turner wrote:

> On Tue, 23 Mar 2004 10:29:59 -0600, Charles Morrison wrote:
>
>> I can't speak to this issue directly.  However, on the human factors 
>> side
>> people doing signal detection research stumbled across something that 
>> is
>> counter intuitive.  Is it easier to copy a weak CW signal on a quiet 
>> or
>> noisy band?
>> Many people will say it is easiest to copy CW on a quiet band.  
>> However,
>> under controlled testing people did a better job of copying a weak CW 
>> signal
>> in noise than when there was no noise.  A demonstration of this on NPR
>> nearly blew me away.  The played 10 seconds of what to the human ear 
>> sounded
>> like silence but actually contained a very low level CW signal.  When 
>> noise
>> was added in the CW signal was very obvious.  Of course too much 
>> noise mask
>> the signal entirely.
>> I doubt electronic circuits/software work the same as the human 
>> brain.  But,
>> the next time you operate the original digital mode don't cuss any 
>> noise for
>> it might be help your brain detect a signal that it otherwise might 
>> not
>> process.
>
> _________________________________________________________
>
> I've noticed a similar effect on CW myself.  Often I can copy CW better
> with a wide filter instead of a narrow one, even though the apparent
> noise is more.  Strange.

This might have something to do with what was was called "noise 
linearization."  The signal, which is too low in amplitude to detect is 
detectable as a modulation of the noise.

Different forms of noise linearizations exist in the signal detection 
world.  The van Vleck method allows you to hard clip a signal when it 
is immersed in noise that is much larger than the signal and yet be 
able to derive the spectrum of the original signal back with just a 
1.92 dB loss in signal to noise ratio (in Probability theory books, 
this method is known as the "arcsine theorem."  The method was invented 
in WWII for used with radars when it was cheaper to deal with the 
autocorrelation of hard clipped on-off signals than the autocorrelation 
of analog signals.  I have used this technique before to design 
practically lossless 4-bit A/D converters that have extremely large 
dynamic range back in the days when fast (20 MHz) 24-bit converters 
were not feasible.  Even with two bits, a properly designed converter 
would have less than 0.5 dB of loss in SNR.  By 4 bits, it became 
vanishingly small.

Another application is in the "dithering" of signals before being A/D 
converted, in this case the noise is much smaller than the average 
signal.  A noise whose rms value is about 1/2 to 1 least significant 
bit is injected before the signal enters the A/D converter.  Without 
the dithering noise, the signal's spectrum is completely screwed up, 
but with the noise, even if the noise floor is higher, the spectrum 
remains the original (with a slightly raised noise floor).  Modern CD 
recordings are dithered.  Some of the early CD recordings were not and 
they sound very strident.

In short, yes, adding noise is used profitably in the signal detection 
world.  But remember that you need to add noise that is not correlated 
with anything (i.e., random).  I have always welcomed a little hiss in 
the background of RTTY signals :-).

- kc