Forum

snovotill · Joined: Feb 8, 2013 Posts: 80

Hola, I'm experimenting with FFT>Subvector>VectorToScalar for measuring band power, and so I fired up a 10Hz 28uVp-p sinewave in SimSource -> ScalarTimeTransform(RMS) -> NumericDisplay, and I get a reading of 9.5uV RMS, although I was expecting 9.9uV.

Confidence up, I then patched the SimSource -> FFT -> VectorToScalar(RMS) -> NumericDisplay, and this time I get a reading of 183 whereas I was expecting 9.9 again. I then tried SumOfSquares and got a huge number, yet this dataset contains one little spectral peak at 10Hz.

In the end I'm trying to measure power in various frequency bands. Not sure where to look next, who to ask for, or what to see, so would appreciate a tip or small square of chocolate. Thanks

jarek · Joined: Oct 22, 2007 Posts: 1073

See this snippet, the RMS value is 9.9 as you expected.

http://proatech.com/design/rms_snippet.bpd

The RMS out of VectorToScalar element will not give the same result. It calculates RMS of vector bins.

If you use operations on VECTORS, use the float elements only (not scalars). Scaling is fully implemented only for one dimensional stream scalar elements.

snovotill · Joined: Feb 8, 2013 Posts: 80

There doesn't seem to be any BioEra element to transform VectorToFloat, which is why I used the RMS function in VectorToScalar instead. Does that not give me the RMS of Sine&Cos elements of the bins, via some kind of sum of squares algorithm or similar? Of course if the RMS calculation works primarily across bins rather than across time, then I'll need an intermediate transform prior to running Average or SumOfSquares etc.
A bit of clarification could be useful.

jarek · Joined: Oct 22, 2007 Posts: 1073

Yes, VectorToScalar but select float option.

Vector elements do not know about sines or cosines. They are not only for FFT processing. Vector is simply an array of values. And RMS (and all other functions) calculate it for all those values in that array.

snovotill · Joined: Feb 8, 2013 Posts: 80

Aha! I thought it was much more complicated than that.

The GOOD NEWS is that FFT-vector feeding into VectorToScalar does yield fairly close results now that I know how these work, and I can always do a full vector-to-vector preprocessing.

Thanks again for success.

There is a lingering question: I can't find an element which transforms an FFT-vector into simple Float out via Average, RMS, Peak etc... and I'd rather not use the FFT-floatvector block due to processing overhead. Is this planned?

jarek · Joined: Oct 22, 2007 Posts: 1073

I don't think I understand. The VectorToScalar does that, doesn't it? And what processing overhead?

snovotill · Joined: Feb 8, 2013 Posts: 80

No such luck. VectorToScalar only comes with a Scalar (yellow) output. Maybe because the incoming vectors consist of Scalar values.

The processing overhead I'm referring to FFT. According to the manual, using Float for FFT doubles the CPU useage, which is why I'd rather stay away from FloatVector.

jarek · Joined: Oct 22, 2007 Posts: 1073

You need to select FLOATVECTOR option, not FLOAT. It does come with float output.

I wouldn't worry about performance difference between integer and float FFT. Yes, they may be twice slower, but it doesn't mean you will notice any difference.

snovotill · Joined: Feb 8, 2013 Posts: 80

That does work if I also use Float-FFT. Thanks. I'll give it a shot and if I have performance issues I'll go see my doctor.