Re: Generation of data and file extensions

[Marcus D. Leech]

On 2021-11-28 08:57, Aditya Arun Kumar wrote:
> So, I have been trying to get a hold of some of the DSP libraries used 
> out there, for that, I felt like I need a data set with which I can 
> test and train myself on these libraries. To this end, I wanted to use 
> GRC to generate some signals samples. I have some questions regarding 
> this font.
>
> 1. What are the possible file formats that I can save the data in, I 
> understand this depends on what data type is being used in the 
> flowgraph blocks, but getting a better understanding on this front 
> would be helpful in understanding GRC and numbers from them a bit better.
> 2. Some tutorials online say and use a dataset with .dat file 
> extension, how is this going to vary from .bin, .cf32, or .cfile which 
> is better, and how each data point inside the data-set will look like? 
> (A link or some reference would be good)
> 3. If anyone can provide me with some links that can help me in this 
> effort like reference materials or some tutorials, will be really 
> helpful (I am planning on starting with fftw, kfrlib and liquid-dsp)
>
> --
> S. Aditya Arun Kumar
Gnu Radio doesn't inherently "understand" too many file formats. It has 
sources and sinks for .WAV format, and also has its own data-saving 
format based on
  HDF5 called SIgMF.

The "ordinary" file source/sinks just record the raw samples in whatever 
the native CPU format is.  For floats, that's 32-bit IEEE-754.  For 
integer formats, it's
  whatever the host CPU natively uses, so you need to be careful when 
transporting those to CPUs with different "endianness" than the CPU that 
recorded it.

Having conversion routines and complex formatting in the file sinks and 
sources would *dramatically* reduce recording performance, which is why
  "raw" formats are preferred for high-bandwidth recordings of 
real-time sample streams.