3.2 Output

We can currently read out the entire correlator in 0.25s. Thus you can get $t_{\rm int}=0.25$s for any $\mathcal{N}_{\rm sta}/N_{\rm SB}/N_{\rm pol}/N_\nu$ configuration that satifies eq.(4).

Of course, one disadvantage of short $t_{\rm int}$ and/or large $N_\nu$ is that the size of your output data grows all the more quickly. To look at how big your dataset might become, start with a single board, which outputs 48kB of data per integration, 32kB of which is lag-based correlation data and 16kB of which is header information. Considering only the correlation-function data portion and converting to frequency-space ( $N_\nu = N_{\rm lag}/2$) yields 0.5MB per integration for the whole correlator. We can calculate a fraction of ``equivalent correlator usage":

$$f \quad = \quad \frac{N_{\rm sta}^2 N_{\rm pol} N_{\rm SB} N_\nu}{131072},$$ (5)

where $N_{\rm sta}$ reflects the stations you actually used (i.e., no longer $\mathcal{N}_{\rm sta}$ from eq.4). Introducing $t_{\rm int}$ to convert from per-integration to per-time units, your resulting FITS file would grow as

$$\simeq \quad \frac{1.75 f}{t_{\rm int}} \mathrm{ GB per hour of observation.}$$ (6)