HdrChopper wrote on Sep 23^rd, 2008, 7:51pm:


Hi Tosei,

In principle, this should function as a lowpass filter. I remember having had to build an analog moving average filter to reduce white noise in input samples, but cannot remember any details, as I was still a greenhorn then. However, we allowed good settling at each sample.

You will surely have plenty of distortion in your case, since you are sizing your switch in a such a way that it has too large an Ron, or to put it another way, your RC timeconstant is too large, and the R is nonlinear (since a switch is used).

Why not use a passive, linear R and a C for your filter. It would appear to me that the sampling part is anyway redundant since you are relying on averaging across multiple cycles.

As for white noise, the basic truth remains - total white noise power integrated = (kT/C).

Regards,

Vivek

Thanks vivek and Tlaloc for your answers...

vivkr wrote on Sep 23^rd, 2008, 11:27pm:


If in this case you allowed for good settling therefore the LPF action is very light, almost approaching the ideal T&H circuit where full settling is allowed. I'm actually looking for the opposite: a heavy filtering action.


vivkr wrote on Sep 23^rd, 2008, 11:27pm:


I'm actually thinking of using a very low Ron switch in series with a linear R and C (option 3 in the attached figure), in such a way the LPF time constant is set by linear components, and the switch just provides the holding action, which I require in my application.

In fact I´m looking to not increase the base band white noise power and at the same time hold the incoming signal every Tck. Therefore I need to provide some anti-aliasing means before sampling and I was wondering if merging the LPF (supposedly for preventing aliasing) and the sampling circuit would fulfill both requirements...

Obviously having a simple LPF and the T&H circuit with a buffer in the middle would do it (option 1 in the attached figure), but the whole point is about saving the buffer's area and merging both the filter and the T&H circuit....

vivkr wrote on Sep 23^rd, 2008, 11:27pm:


I agree, but my problem is not the total integrated white noise, but rather how that power ends up distributed over frequency....


[/quote]

Or you could use a switched cap R into a C.  Then your pole would be more stable across process and you wouldn't have a DC voltage drop.

[/quote]

Tlaloc:

Using a switched cap is not an option since - as I described above - I 'd like to provide some antialiasing means and sample the signal in the same S&H circuit....if this were possible... that´s my question  :-/


Finally,
I think option 2 does provide antialiasing means before sampling the signal. Although the time constant of the filter changes during the sampling time (Ch in parallel with C1), the anti- aliasing action takes place....so, can this option also be reduced to option 3?

Many thanks
Tosei

Hi Tosei,

The option 3 is not feasible becaue the input and white noise have already been aliased before they are low-pass filtered. That sampling occurs preceding the filter.


Yawei

Hi Tosei,

From the relative values of the 2 resistors, I would also agree that there is no aliasing prior to sampling even with option # 3. Nonetheless, I fail to see why you would want to use it at all, since there appears to be no real advantage.

Your switch which is a nonlinear element is exposed to the largest possible signals, since there is no filtering preceding it, and so it will tend to worsen its nonlinear behavior.

Also you need to be careful if you want to neglect the kT/C noise from the smaller capacitor. Whether you can ignore it or not depends on its magnitude. Personally, I still do not see a significant advantage of this method but maybe you have found some use for it.

Regards,

Vivek

HdrChopper wrote on Sep 25^th, 2008, 6:09am: