Frank A. Vorstenbosch wrote:

> If you use N transistors in parallel in your frontend, then noise will
> go down by SQRT(N) -- same as having a transistor of N times the die
> area.

  Yes, but if your input is impedance-matched, your input SIGNAL will be
divided by N, so your signal-to-noise ratio will be SQRT(N) WORSE for
the exercise.

  You remind me of something I saw somewhere about using power output
devices as front-end amplifiers but I think the advantage of that was
actually in overload capability!  Fact is, we DON'T use large die area
devices but rather tiny little GaAsFETs in RF work.

  Now, another mail to me personally suggested that the amplifiers in
question were so high in impedance compared to the transducer that
paralleling them would not alter the signal voltage!  Fine!  If you are
using such a bad impedance match, the MOST effective way to improve
signal-to-noise ratio is stunningly easy - use a step-UP transformer to
correctly match the impedance!

  Peter van Hoof writes again:

> This is quite untrue if you understand the math behind it
>  -The output signals from the amps are added together nicely because
>  they are all in phase
>  -The noise generated in the preamps you can compare to random numbers
>  rangin from positive to negative **add them all up and voila** the
>  average will be less than the peak to peak noise of a single preamp

  Yes, Peter, that is the significance of the SQRT(N) in the above
discussion and applies to what happens at the common OUTPUT, but you are
still failing to understand what is happening at the INPUT where you
will be LOSING signal much faster!  The term is "signal-to-noise ratio".

  Cheers,
        Paul B.