In reality it is at the moment, but, for example, I would reckon that
discrete frequencies from GPS satellites that Dr Shuch quotes as examples of
this should be about 30dB below the total signal power. To get better
signal to noise than implied by this you need to know the spreading code
and chipping rate. If the signal is strong enough to pass a pre-qualification
test, you might be able to get enough clues to the units of time used to
restrict the number of chipping rates to try and the number of good spreading
codes, for non-encrypted systems like this, may be small enough to allow
the code space to be searched. This is for direct sequence SS.
An ideal Shannon signal would be indistinguishable from noise unless you
had prior knowledge of its structure. Real ones won't be because an ideal
Shannon signal needs infinite time to transmit and engineering constraints
means their has to be some redundancy to allow a receiver to lock on to the
Incidentally, although modems aren't really spread spectrum but rather have
the spectrum of non-random data evened out within the same passband, for
alternative 00 and FF characters they do behave like that. Listen to the
modem speaker on a modern modem and you will hear a noise like sound, but,
if the signal weren't scrambled, you should hear some sort of tone. Old
FM modems, or RTTY, are much more obvious in this respect, as they don't
scramble and only use two level codes. Spreading makes things worse, but
optimum communication also favours evening out the power spectrum, e.g.
the modems above, or Clover onr HF radio.
Note that modems also point out that there may be small times during a
transmission when there are easy to detect signals (listen to the initial
training sequence, or even a retrain sequence).