[Discuss-gnuradio] Log power FFT block - bug?

[Urban Kuhar]

Hi everyone!

I've been using log power FFT block for signal power estimation. I've fed the output of Log Power block to my custom block in which I search for peak in power spectrum and then I sum up bins left and right from that peak until the turn is reached. C++ code for power estimation:

    // move right from the max

    iCount = iMax_index + 1;

    while (in[iCount] < in[iCount-1])

fTotalPower += in[iCount++];

    

    // move left from the max

    iCount = iMax_index;

    while (in[iCount] > in[iCount - 1])

fTotalPower += in[iCount--];

I've set the frame rate property to 10 frames per second (each 32768 samples) and I use sample rate of 200kSPS. What happens is that every fourth power estimate sample has magnitude for about 1dB lower than the other three. I've looked deeper into situation and I found out that every fourth power spectrum from log power FFT block has about three of four bins thinner signal and that's why the total magnitude of the signal is lower. I've tried lower frame rate (2 frames per second) but situation doesn't change.

Next I tried the FFT block instead of log power FFT and the problem disappears. So I am guessing that there is some kind of problem in Log Power FFT block. Has anyone experienced similar problem?

Regards,

Urban

samples.png
Description: PNG image