Change the sample rate after filter

[Kevin_Price]

1. I was mainly just giving an example for why someone might *want* to oversample by a large factor.  It wasn't really meant as direct advice.

2. Besides that I guessed wrong anyway.  I was talking about a scenario where one cared mainly about the shape and behavior of the signal in the *time domain*.  It now sounds like you're mainly concerned with the *frequency* domain.  There, a final sample rate of 6x the highest freq of interest is fine, perhaps even overkill.  2x is the minimum standard.

    I haven't personally studied it so I don't know if or how much better 6x will be than 2x.  My educated guess is that it won't make a dramatic difference but may help a little.  (Or not, it'd probably depend quite a bit on the nature of the signal.)

3. Filtering can still be helpful as it'll help reduce the overall noise floor so that low level frequency contant is more distinct.

4. The median filter acts like a sliding window.  At each sample point, the output will be the median value from among the sample you're at and its immediate neighbors.  The "narrow" one I mentioned, would prevent any noise spikes that are only 1 sample wide from passing through.

   In general, a median filter is a little CPU-expensive with non-linear effects so I don't use them habitually.  Just in particular situations like I described.

5. Random noise tends to skew toward looking like high frequency content.  The standard Butterworth filter will suppress quite a bit of it.  If you need to, you can take out a little more if you also do the oversampling followed by decimation-by-averaging.  But keep in mind that it also drives up your data bandwidth and CPU demands, so you'll have to see whether it's worthwhile overall.

-Kevin P

[mcduff]

---

@Kevin_Price wrote:

2. Besides that I guessed wrong anyway.  I was talking about a scenario where one cared mainly about the shape and behavior of the signal in the *time domain*.  It now sounds like you're mainly concerned with the *frequency* domain.  There, a final sample rate of 6x the highest freq of interest is fine, perhaps even overkill.  2x is the minimum standard.

    I haven't personally studied it so I don't know if or how much better 6x will be than 2x.  My educated guess is that it won't make a dramatic difference but may help a little.  (Or not, it'd probably depend quite a bit on the nature of the signal.)

---

A DSP guru at work has told me 2x the highest freq of interest is fine if you are only interested in frequency content. To get accurate measurements of phase, then 4x the highest freq of interest is sufficient for a good phase measurements.