convert_i16_f adds a portional of original data to the end #57
Description
Thank you for this useful DSP tool which I am hoping to use. But I notice something strange when converting data types.
I recorded a 1 second IQ sine wave at 48kHz, and used csdr convert_i16_f to convert to a float file. The int16 file is 185.k where as the float32 file is 176k, which is 1k bigger than the expected 187.5*2.
Using python to read the int16 data to array yy and the float32 data to zz.
The beginnings are the same
yy[:5] = array([ 0.+250.j, -812.+134.j, -824. +18.j, -818. -98.j, -800.-213.j])
np.round(zz[:5]*(1<<15))
array([ 0.+250.j, -812.+134.j, -824. +18.j, -818. -98.j, -800.-213.j], dtype=complex64)
The end of the int16 data is
yy[-5:] = array([ 298.+834.j, 195.+868.j, 88.+885.j, -19.+889.j, -129.+876.j])
While the float32 data extends 128 samples beyond sample 48,000
np.round(zz[-131:-126]*(1<<15))
array([ 88.+885.j, -19.+889.j, -129.+876.j, -564.-658.j, -478.-730.j], dtype=complex64)
So the last sample in the int16 data, shown in bold, is at sample [-129] which corresponds to the expected 48000 th sample of the float32. But there are 128 more samples after 48000.
Then, by scanning through the int data, I found the start of the 128 sample extension to be sample 512 from the end of the original
yy[-513:-508].
array([-637.-573.j, -564.-658.j, -478.-730.j, -389.-789.j, -290.-837.j])
np.round(zz[-129:-124]*(1<<15))
array([-129.+876.j, -564.-658.j, -478.-730.j, -389.-789.j, -290.-837.j], dtype=complex64)
In summary the converted data is 128 samples longer than the original with the added data being a 128 sample section taken 512 samples from the end of the original data. I suspect some buffer under/over run but cannot see where this might occur.
Help in understanding and fixing this would be appreciated. R
PS this is running on a raspberry pi using the branch recommended in the QTCSDR installation instructions.