On Fri, 11 Jan 2008 15:57:50 -0800 (PST), MooseFET
<k...@rahul.net> wrote:

>On Jan 11, 7:49 am, John Larkin
><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
>[... multiple references ...
>> I thought of that, but the lowpass filters must include at least one
>> opamp per, and those opamps will have ballpark 1 nV/rthz noise too.
>
>You can still sum in front of the filter:
>
>
>  REF1 ---/\/\----
>  REF2 ---/\/\----+
>  REF3 ---/\/\----+
>        "      "
>  REFN ---/\/\----+-----+------!+\
>                        !      !  >-----+----More filter if needed
>                       ===   --!-/      !
>                        !    !          !
>                       GND    ----------
>
>

But THAT opamp                ^^^^^^^

has 1 nv/rthz noise!

John

On Jan 11, 4:17 pm, John Larkin
<jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> On Fri, 11 Jan 2008 15:57:50 -0800 (PST), MooseFET
>
>
>
> <kensm...@rahul.net> wrote:
> >On Jan 11, 7:49 am, John Larkin
> ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote:
> >[... multiple references ...
> >> I thought of that, but the lowpass filters must include at least one
> >> opamp per, and those opamps will have ballpark 1 nV/rthz noise too.
>
> >You can still sum in front of the filter:
>
> >  REF1 ---/\/\----
> >  REF2 ---/\/\----+
> >  REF3 ---/\/\----+
> >        "      "
> >  REFN ---/\/\----+-----+------!+\
> >                        !      !  >-----+----More filter if needed
> >                       ===   --!-/      !
> >                        !    !          !
> >                       GND    ----------
>
> But THAT opamp                ^^^^^^^
>
> has 1 nv/rthz noise!

Yes it does but the number of references would have to be:

    500^2 = 250000

before the filter before the op-amp doesn't help.

The "more filter if needed" can further reduce the noise from that
point.  Since there aren't many op-amps below 1nV/sqrt(Hz), for most
practical situations the extra filter won't be the determining noise
level of the whole system.

I suspect that the circuit could be made better if each input was
soemthing like this:


                                  Others
 REFN ----------!+\                 !
                !  >--+---\/\/------++-------!+\
              --!-/   !              !       !  >---+--- To more
filter
             !        !             ===   ---!-/    !
          R2 +--/\/\--+              !   !          !
             !        !             GND   ----------+
          C1 +---!!---+                             !
             !        !                             !
          D2 +--!<----+                             !
             !        !                             !
          D1 +-->!----                              !
             !                                      !
              ---/\/\-------------------------------
                  R1

The outliers would be given less weight when the diodes D1 and D2
conduct. The circuit as drawn assumes an exact match between the
references.  Since this is never the case, a more complex circuit
would be needed but I think this clearly shows the idea.  The noise
from a reference is not perfectly uniformly distributed.  There tend
to be noise events.  These could be ct off.