Re: [Discuss-gnuradio] Sample size and data rates

[George Nychis]

> 1) 1 000 000 samples writes a file 8 000 000 bytes (7.6Mb)
> This suggests each sample is 8bytes; 4bytes I and 4bytes Q.
> Is my assumption correct?

This is how complex samples are represented in GNU Radio, as gr_complex 
which is of type std::complex where I and Q are 4 bytes each, yes.

> Considering that the ADC produces 12bit samples, I would expect that
> each I and Q sample pair would be 4bytes. 2bytes I and 2bytes Q.
> Please can someone advise where i've missed some understanding.

When the samples are generated in the USRP, they are packed (by default) 
as 32-bit pairs, 16-bits for I and 16-bits for Q.  This reduces the 
amount of data across the bus rather than use full 32 bits for each. 
You can set the format register to transfer 8-bit samples though.

Once these samples hit the host, GNU Radio extracts the samples in 
whatever format they are and stores them as gr_complex.  This simplifies 
processing and makes everything standard.  It also takes care of byte 
ordering issues for you.

> 2) With reference to the above question, where can I find further
> details regarding the function of the gr.sizeof_item identifiers, in
> particular gr.sizeof_gr_complex.

What I do is find where the types are defined, i.e.:
gnuradio-core/src/lib/runtime/gr_complex.h
typedef std::complex<float>     gr_complex;

- George