CASA is used for initial processing of the data to calibrate raw visibility measurements. This is followed by the creation of a model sky image from clean components. This model is then subtracted in the visibility domain to obtain residual visibilities. We use both the calibrated and residual visibilities for computing the power spectrum

Each visibility is then Fourier transformed in frequency space (Eq. (31)). This process is needed for isolation of foregrounds in the k⊥-k plane. We note the our method utilizes both the subtraction of foregrounds and their isolation. But it does not employ an external point source catalog

The procedure outlined above yields complex visibilities as a function of five variables: Vτ (u, v, w, t). For computing the power spectrum we crosscorrelate these visibilities for t = t to remove the noise bias. To weigh each cross-correlation we assume that there exist regions in k⊥-k plane which are dominated by only noise and the HI signal. This allows us to compute a weight for each crosscorrelation based on the expected HI signal. For computing these weights we take into account the impact of w-term and the distortion of intensity pattern in a tracking scan. The relevant method is elaborated in detail in sections 2, 2.1, 2.2, and 2.3 and summarized in section

4. In section 4.1, we describe the power spectrum estimator, taking into account weights given by the expected HI signal, in 3-, 2- and 1-dimension. We also discuss our method to compute the errors on the estimated power spectrum. 3.1. CASA processing MWA data were collected at 2-minute intervals with a time resolution of 0.5 seconds and frequency resolution of 40 kHz. The central frequency of these observations is 154.24 MHz. For preprocessing we have used the Cotter pipeline to average to 10 seconds of integration

we have not performed any averaging over the frequency channels. Cotter also uses the in-built AOFlagger to flag and remove radio frequency interference. The edge channels of each coarse band are flagged with Cotter due to aliasing effects. After this preprocessing the Cotter pipeline delivers the data in the CASA readable ‘Measurement set (ms)’ format for further processing. Once the ‘ms’ files are produced for each 2-minute data set, we process each of these 2-minute data in CASA to produce an image source is used to calculate the bandpass solutions which are applied to the uncalibrated data. We next construct a sky model from

In Figure 6, we present a sample image of 2 minute deconvolution. As noted above we process the data for only 2 minutes to ensure the primary beam doesn’t substantially change during the run. For a 2-minute scan we obtain an RMS of nearly 40 mJy/beam. The residual visibility Vν(uν, vν, wν, t) is a function of five variables. We compute the discrete Fourier transform of the residual visibilities in the frequency space weighted by the Blackman-Nuttall (Nuttall (1981)) window Bν to suppress leakage into the EoR window (2013, 2016)): Vτ (u, v, w, t) = ∆ exp(i2πντ)Vν(uν, vν, wν, t)Bν (31) Notice that in Eq. (31) the frequency dependence of the baseline vector bν = {uν, vν, wν} is integrated over. Therefore, the labels {u, v, w} on the LHS of Eq. (31) need further explanation. As noted above (the discussion following Eq. (11)) they can be chosen to denote a given baseline vector at a fixed frequency, ν0. We choose this frequency to be the central frequency of the band ν0 = 154 MHz. Parsons et al (a,b) provide detail implications of the frequency dependence of the baseline vector. Here ∆ = 40 kHz and 256 channels are used for our study, which correspond to total band-