An aperture synthesis telescope, like the GMRT, consists of a number of antennas located on the ground and the resolution of such a telescope is proportional to the maximum projected separation between the antennas. The locations of the antennas are usually specified in the Earthcentred XY Z coordinate system. The Earth centred XY Z frame is related to the xyz location of the antennas on the ground by
 (2.1) 
where is the nominal latitude of the telescope site. The xaxis points towards the geographical east, yaxis towards the north and zaxis towards the local zenith.
However, for the purpose of imaging, only the relative separation between the antennas is important. This separation between the antennas is usually referred to as the baseline and is measured in units of the wavelength λ of the incident radiation. For the purpose of the theory of synthesis imaging, the antenna positions are specified in the so called uvw frame. The geometric relationship between this XY Z frame and the uvw frame is shown in Fig. 2.1 and the coordinate transformation is given by
 (2.2) 
where H is the Hour Angle and δ the declination of the source.
The geometric relation between the observing plane represented by the uvplane and the sky plane represented by the lmplane is shown in Fig. 2.2. The lmplane in the sky is parallel to the plane in which measurements are made (the uvplane) and the separation between them is denoted by R. The waxis points towards the origin of the lmn frame given by (l = 0,m = 0) and is parallel to the naxis. The treatment of the theory of synthesis imaging given below follows that of Thompson et al. (1986) and Clark (1999).
Let E(,_{1},t) represent the electric field produced by an infinitesimal element in the sky in the direction at point _{1} in the uvw frame (Fig. 2.2). The total electric field e(_{1},t) measured at this point will be an integral of the signals from all the radiation elements and is given by

 (2.3) 
_{1} is the vector denoting the point 1 in the uvw frame. Let R_{1} be the distance between point _{1} and the point in the source plane towards the direction given by the vector _{1}. E(_{1},_{1},t) can then be written as
 (2.4) 
where ξ is the complex amplitude of the radiation at the source, ν is the frequency of the radiation, τ_{1} = R_{1}∕c and c is the speed of light. τ_{1} represents the propagation delay for the radiation of complex amplitude ξ to travel the distance R_{1}.
The source coherence function, defined as
 (2.5) 
is a measure of the coherence between radiation from two points on the source given by the directions _{1} and _{2} respectively. For a spatially incoherent source, radiation from two points on the source will be uncorrelated. For such a source, Γ(_{1},_{2},τ) = δ(_{1} _{2}) and equation 2.5 can be written as
 (2.6) 
This is the autocorrelation function of the radiation from the source Γ(,0) is a measure of the average surface brightness of the radiation from the source.
The mutual coherence function, measured at two points _{1} and _{2} on the observing plane is defined as
 (2.7) 
Using Equations 2.3 and 2.4 and the equivalent equations for the point _{2}, the above equation can be written as
 (2.8) 
where τ_{t} = τ_{2}  τ_{1} = (R_{2}  R_{1})∕c is the relative delay between the signals measured at the points _{1} and _{2} respectively.
In the above equation, the dependence on _{1} and _{2} is implicitly via τ_{t}. For a plane wavefront from a direction in the sky, τ_{t} = _{12} ⋅∕c where _{12} is the vector separating the two antennas. It is therefore clear that γ depends only on the relative separation between the antenna positions given by _{1} and _{2} and can be written as
 (2.9) 
The components of _{12} will be given by (u_{12},v_{12},w_{12})=(u_{2} u_{1},v_{2} v_{1},w_{2} w_{1}). Using direction cosines of as the (l,m,n) coordinates of a point in the sky, we get
 (2.10) 
Let the antenna located at point _{1} track a direction given by the unit vector in the sky. Due to the separation between the antennas on the ground, to track the same point in the sky, a second antenna located at point _{2} will need to point in a slightly different direction given by _{2}. The angle between and _{2} is given by tan^{1}(_{2} _{1}∕R). This difference in the pointing directions results in an extra relative delay between the signal arriving at the two antennas from a point in the sky. As a result, one of the antennas samples a slightly delayed version of the wave front. If, however, R ≫_{2} _{1}∕λ where λ is the wavelength of the radiation, the same wavefront will be sampled by the two antennas (tan^{1}(_{2} _{1}∕R) ≈ 0). This corresponds to the usual plane wavefront approximation. This approximation is used in the rest of the treatment.
In the limit of the plane wave front approximation, the denominator in Equation 2.8 can be adequately represented by R^{2} and d∕R^{2} = dldm∕. The phase of the mutual coherence function in Equation 2.7, for a spatially incoherent source, therefore, depends only on the relative separation between the points of measurement in the far field of the radiation.
Let _{∘} represent the unit vector along the waxis pointing towards (l = 0,m = 0). The components of _{∘} in the uvw frame will be (0,0,1). The relative delay between the signals received at the two antennas from a point source in this direction will be τ_{g} = _{∘}⋅_{12}∕c = w_{12}∕c. Treating _{∘} as the reference direction towards the origin of the source coordinate system, the instantaneous phase of γ(_{12}) can be measured with respect to this direction, without losing any information about the sky brightness distribution. This can be done by rotating the phase of γ by 2πνw_{12}∕c. The point in the sky in the direction _{∘} is referred to as the phase centre and τ_{g} is referred to as the geometrical delay. The Phase centre direction defines the origin of the source coordinate system. Since _{∘} is an arbitrary vector, the phase centre can be chosen at any convenient point in the sky. Usually, this is coincident with the antenna pointing centre.
The visibility phase for a point source at the phase centre, after correcting for the geometrical phase is equal to zero. With this correction, the visibility is said to be ‘phased’ for the phase centre. An interferometer phased for this reference direction will remain phased for all points on the surface of a sphere of unit radius passing through the point (l = 0,m = 0), centred at (u = 0,v = 0) and described by the equation l^{2} + m^{2} + n^{2} = 1. Under ideal conditions (i.e., with no other source of phase errors), any residual phase would be due to the relative separation between the phase centre and a source of emission located on such a sphere. In that sense, the visibility phase from a ‘phased’ interferometer provides information about the distribution of the sky brightness, relative to the phase centre. Moving one of the antennas to the origin of the uvw frame, the (u_{12},v_{12},w_{12}) coordinates are just the coordinates of the other antenna and the subscript ‘12’ can be dropped from Equation 2.10.
The length of time for which a signal remains coherent is given by the inverse of the bandwidth Δν. After correcting for the geometrical delay, the relative delay between the signals is τ_{c} = τ_{t} τ_{g}. Replacing τ_{t} by τ_{c} in Equation 2.8 corresponds to the geometrical delay corrected version of this equation. If τ_{c} ≪ Δν^{1}, two versions of a time series displaced with respect to each other in time by τ_{c} will remain coherent. In the limit of the validity of the assumption that τ_{c} ≪ Δν^{1}, ⟨ξ(,t)ξ^{⋆}(,t  τ_{c})⟩ in Equation 2.8 can be replaced by ⟨ξ(,t)ξ^{⋆}(,t)⟩ = ⟨ξ()^{2}⟩. This is the same as the source autocorrelation function at zero lag, Γ(,0) (Equation 2.6) and is equal to the two dimensional source surface brightness, denoted by I(l,m)
Therefore, under the approximation that:
using the relation between l,m,n in Equation 2.10 and the above mentioned phase rotation, Equation 2.8 can be rewritten as
 (2.11) 
where u, v and w are now the coordinates in the uvw frame measured in units of the wavelength. γ is usually referred to as the visibility function and denoted by V (u,v,w).
The integrals in the above equation are over the entire sky (limits of the integral from 1 to 1). However, the antenna primary beam limits the part of the sky from which the antenna can receive radiation to ~ λ∕D, where D is the diameter of a circular aperture. Assuming that the response close to the centre of the primary beam is ≈ 1, with an additional approximation that the field of view is small (l^{2} + m^{2} ≪ 1), Equation 2.11 can be written as
 (2.12) 
This is a 2D Fourier transform relation between the mutual coherence function (visibility) and the source surface brightness. This statement is referred to as the van CittertZernike theorem (Born & Wolf 1959 [and later eds]; Thompson et al. 1986) and forms the basis of imaging with interferometric radio telescopes.
The source surface brightness I(l,m) is described in the lmplane while u and v are the equivalent conjugate variables in Fourier space. u and v can therefore be interpreted as the spatial frequencies and the visibility function as the spatial frequency spectrum of the source surface brightness distribution. Synthesis radio telescopes like the GMRT measure the visibility function at several points in the uvw frame using an array of N antennas which instantaneously produce N(N  1)∕2 pairs of interferometers. Due to the rotation of the earth, the projected separations between the antennas change as the antennas track a source in the sky. Over time, each antenna pair therefore measures the visibility at several points in the uvw frame. As a result, over time, the array of antennas partially covers the uvw volume, up to a maximum projected antenna separation of B_{max} corresponding to a spatial resolution of 1∕B_{max}. Also, an interferometric array does not measure V for baselines smaller than a minimum value. Just as the maximum baseline length for which V is measured corresponds to the highest spatial resolution the telescope can provide, the smallest baseline for which V is measured corresponds to largest angular scale that can be represented reliably in the final image. Hence, an interferometric observation will be insensitive to scales larger than 1∕B_{min}; this problem of missing short spacing measurements is referred to as the problem of “missing short spacing”. When imaging extended objects in the sky, like the ones imaged for this dissertation, the shortest spacing for which a reliable measurement exist has for reaching implications – missing short spacings result in missing extended emission in the image.
If γ(u,v) is completely sampled, at least at the Nyquist rate, it can be inverse Fourier transformed to recover the source brightness distribution I(l,m). With a finite number of antennas located at discrete locations on the ground, the visibility function is sampled at a discreet set of points. The observed visibility can, therefore, be thought of as the true visibility, multiplied by the sampling function given by
 (2.13) 
where (u(t),v(t)) represents the points sampled at time t. Using the convolution theorem for Fourier transform (Bracewell 1986 [and later eds]), it can be shown that the function I, recovered by the inverse Fourier transform of S(u,v)γ(u,v), will be convolved by the inverse Fourier transform of S. The inverse Fourier transform of S, denoted by B, is usually referred to as the dirty beam or the point spread function or the telescope transfer function. I^{d} = I ⋆B is referred to as the dirty map, where ‘⋆’ represents the convolution operator. I is recovered from I^{d} using nonlinear deconvolution techniques (Högbom 1974; Clark 1980; Cornwell & Evans 1985).
The approximation that l^{2} + m^{2} ≪ 1 breaks down at low frequencies and Equation 2.11 accurately described the measured visibility function. However, this is not a Fourier transform relation. Techniques used to recover I in such cases are described in Chapter 4.

In practice, γ is measured by a number of two element interferometers, one of which is shown in Fig. 2.3. The signal from antenna 1 lags behind the signal from the antenna 2 by a phase equal to 2πb_{12}⋅ c where is the reference unit vector mentioned in the previous section. Under the plane wave approximation, the signals from the two antennas can be written as
 (2.14) 
where ϵ_{1} and ϵ_{2} are independent additive random noises from the two antennas. These signals are multiplied and the product averaged in the complex correlator to form the visibilities as
 (2.15) 
Since ϵ_{1} and ϵ_{2} are independent zero mean gaussian random variables, the last three terms will reduce to zero.
The response of the interferometer to a point source of flux density I, located at the phase centre (l = 0,m = 0) is therefore
 (2.16) 
where τ_{g} is the geometrical delay mentioned earlier. It is equal to the time delay due to the extra path length (equal to the projected separation between the antennas along ) which the radiation has to travel to reach one of the antennas. With reference to Fig. 2.3, ντ_{g} = D λ sin(θ) = w_{λ} where D = _{12} and w_{λ} is the instantaneous component of the separation between the antennas along the waxis measured in units of the wavelength of the incident radiation. Equation 2.16 can therefore be rewritten as
Antennas receive radiation from a finite part of the sky, defined by the antenna primary beam. For an extended source, within the antenna primary beam, V (u,v,w) will be an integral over the source (integral over l and m;Equation 2.12) and can be thought of as a superposition of plane waves from the individual infinitesimal elements constituting the extended source given by Equation 2.17.
As the interferometer tracks the source in the sky, the projected separation and hence the geometrical delay changes with time. θ changes slowly going from θ_{max} to θ_{max} when the antennas track a source from rise to set over several hours. θ_{max} depends on the declination of the source and the minimum elevation limits of the antennas. Over short intervals of time, θ changes almost linearly with time. The interferometer response to a point source (the real and imaginary parts) therefore varies quasisinusoidally over short periods of time with a frequency proportional to the separation between the antennas. This quasisinusoidal variation of the interferometer output, due to the changing geometrical delay, is referred to as the fringe pattern. The amplitude of this fringe pattern is directly proportional to the power emitted by the source in the sky. The phase of these fringes also changes sinusoidally with time with a period of 24 hours. The phase of this fringe pattern due to the geometrical delay however carries no astronomically useful information about the sky (it only carries information about the direction in the sky being tracked). In practice therefore, the geometrical delay is continuously compensated by introducing a time variable compensating delay τ_{i} in the signal from one of the antennas. This operations is referred to as delay tracking. The radiofrequency (RF) signals from the antenna feeds, centred at a frequency ν_{RF }^{∘}, are converted to an intermediate frequency band centred at ν_{IF }^{∘} to be transported to the correlator (in the case of GMRT, over optical fiber cables). The path length for the signals from the antenna to the correlator introduces a time invariant fixed delay τ_{Fix} suffered by the signals at ν_{IF }. The compensating delay τ_{i} must therefore compensate for τ_{Fix} as well.
For the GMRT, the compensating delay τ_{i} = τ_{g} + τ_{Fix} is applied to the signals at baseband frequencies in the correlator. However, the signals suffer the delays τ_{g} and τ_{Fix} at the RF and IF frequencies respectively. Delay compensation at the baseband therefore leaves a residual phase given by
 (2.18) 
where ν_{i} is the centre frequency of the i^{th} frequency channel and τ_{Int} is the time delay, applied in the correlator Delay/DPC system (see 2.5.2 for details). This delay compensated signal is fed to the FFT section of the correlator to be split into a number of frequency channels. Δϕ^{∘} is applied at the very first stage of the FFT butterfly network in the FFT section. The remaining phase gradient, denoted by Δϕ_{i}, is applied as a phase gradient in the last stage of the FFT network.
Effectively, the phase of the visibilities is rotated by Δϕ, equivalent to the residual delay due to the differences in the RF, IF and base band frequencies. The phase of the visibility for a point source located at the phase centre is thus reduced to zero at all times. This final rotation of the fringe phase is referred to as fringe stopping or fringe rotation.
Note that the total phase applied to the visibilities via delay tracking and fringe rotation is equal to 2πw and this effectively phases the interferometer for a point in the sky (the phase centre). Application of this total phase is achieved in the GMRT correlator in two stages, as explained in Section 2.5.2.
The response of the interferometer is integrated over the signal bandwidth as seen by the multiplier in the correlator. If H(ν) represents the voltage response of the antennas as seen by the multiplier, the effect of a finite bandwidth is to modulate the output of the interferometer by the Fourier transform of H. If the receiver has a flat response for a range of frequencies ν_{o} ± Δν, the output V , for a source of unit flux density, will be given by
w is the relative antenna separation along the waxis. As is clear from Equation 2.2, the geometrical delay involves the measurement of the relative (X,Y,Z) coordinates of different antennas. It is therefore important to measure the relative antenna coordinates as well as the fixed delays for phasing the array (reducing the fringe phase to zero for the phase centre). The procedure used for measuring the relative antenna coordinates (referred to as baseline calibration) and fixed delays (referred to as fixed delay calibration) for the GMRT is described in Sections 2.6.1 and 2.6.2 respectively.
As mentioned earlier, V (u,v,w) is measured for several values of u,v and w by tracking the source in the sky for several hours, during which the projected baseline of a single interferometer changes with time. In practice, a number of antennas are used and the output of all the N(N  1)∕2 interferometer pairs made by an array of N antennas provide instantaneous measurements at several values of u,v and w. The instantaneous set of points measured in the uvw frame by an array of antennas is referred to as the snapshot uvcoverage. As the array tracks a source in the sky, each interferometer generates a track in the uvplane, dramatically increasing the uvcoverage of the array. The set of u,v,w points, measured by an array of antennas over several hours of source tracking, is referred to as the full synthesis uvcoverage. The shape and density of the uvcoverage of the array determines the telescope transfer function and the geometry of the array is usually optimized to maximize the uvcoverage (Mathur 1969). The configuration of the 30antenna GMRT array and resulting uvcoverage are described in Section 2.2.
The GMRT antennas are fully steerable parabolic dishes each of 45m diameter each with an Altazimuth mount. A turret at the prime focus of the antenna is supported by a quadripod and holds a sealed metallic cube which houses the broadband low noise amplifier (LNA), RF filters, the polarizer and the RF switch for swapping the polarization channels for each RF band (see Section 2.3). The feeds for 150, 327, 233/610, and 1420 MHz are mounted on the four faces of the turret which can be rotated to bring the desired feed in focus using the feed positioning system (FPS). The feed bandwidths and the range of frequencies covered by the various feeds are listed in Table 2.1. The feed bandwidths correspond to a Standing wave ratio of ≤ 2. The 233MHz feed has the smallest bandwidth while the crosspolar characteristic of the 150MHz feed is the worst. The measured crosspolar power at 327 and 610 MHz bands is ~42 and ~47 dB respectively while that for the 150MHz feed is ~22 dB. The value away from the centre of the radiation pattern is ~22 dB for 327 and 610 MHz bands while that for the 150MHz feed is ~20 dB (see Sankar (2000) for more details). Since the GMRT is predominantly a low frequency instrument, the reflecting surface of the antennas is a light weight wire mesh. The wire mesh is held in place by a network of steel ropes, creating patches of flat surfaces which approximate a parabola. This design is referred to as the Stretched Mesh Attached to Rope Trusses design or the SMART design (Swarup et al. 1991). Three sizes of wire mesh are used, going from smallest sized mesh (10 mm) in the inner region of the parabola to the largest size (20 mm) towards the edge to produce a tapered illumination pattern which reduces the sidelobe levels of the antenna radiation pattern. Mesh of different sizes each covers one third of the total surface area. The reflecting surface for each antenna was measured by theodolite, to determine the surface accuracy compared to an ideal parabola. An RMS deviation of 8, 9 and 14 mm from an ideal parabola was measured for the three regions of the surface (Sankar 2000).




The GMRT is composed of 30 such antennas. The antennas are located in a roughly ‘Y’ shaped geometry (Mathur 1969), as shown in Fig. 2.4. Fourteen antennas are located randomly in a region of size ≈ 1 km×1 km, referred to as the Central Square (see Fig. 2.5). These antennas provide the short spacing coverage of the uvplane, which is essential for mapping large scale structures. The rest of the antennas are located in ~ 14 km long Western, Eastern, and Southernarms. The Eastern and Southernarms each have 5 antennas, roughly along a straight line, while the Westernarm has 6 antennas. This arrangement of antennas was designed to maximize the telescope sensitivity to large scale emission as well as to provide high resolution. The shortest spacing of ~ 60 m in the Central Square is provided by the antennas C05, C06 and C09 while the largest spacing of ~ 25 km is provided by the arm antennas. The Central Square antennas together provide a uvcoverage up to 1 km. The uvcoverage for a full 12^{h} synthesis observation using the full array is shown in Fig. 2.6. The uvcoverage due to Central Square antennas alone, for a full 12^{h} synthesis at a few declinations is shown in Fig. 2.7.
A detailed description and analysis of the analog signal flow for the GMRT has been described elsewhere (Praveen Kumar & Srinivas 1996; Praveen Kumar 2000). This section briefly describes the aspects of the GMRT analog receivers which are relevant from the point of view of astronomical observations.
The GMRT operates at the 150, 233, 327, 610 and Lband extending from 1000  1450 MHz. The Lband is split into four subbands centred at 1060, 1170, 1280 and 1390 MHz with a bandwidth of 120 MHz while the bandwidth at other bands is ~ 40 MHz. Fig. 2.8 shows a simplified schematic diagram, with the major components of

the GMRT receiver. The Lband feeds measure two orthogonal linear polarizations while the feeds at all other bands measure circular polarizations. The 233 and 610 MHz feeds are coaxial feeds. The frontend low noise amplifiers have been designed to either receive two polarization signals from a single feed or the same polarization signal from two different feeds. This, for the case of 233/610 MHz feeds, allows simultaneous single polarization observations at these bands. An RF switch after the frontend box allows swapping of the two polarizations signals. Solar attenuators of 14, 30 and 44 dB are also available. Four calibrated noise sources, named Low, Medium, High and Extrahigh cal can be used to calibrate the receiver system. However, the planned periodic injection of calibrated noise has not yet been implemented. The Frontend electronics also has the facility for Walsh switching of the two polarization channels using orthogonal Walsh functions to minimize polarization leakage. However this is also not yet operational at the GMRT.
The RF band is first converted to an IF band centred at 70 MHz using what is referred to as the First LO (First Local Oscillator). Here, bandwidths of 5.5, 16 or the full RF bandwidths can be selected. The two IF signals corresponding to the two polarizations are then converted to 130 and 175 MHz with a maximum bandwidth of 32 MHz using the Second LO. These signals are then transported over optical fibers to the Central Electronics Building (CEB). Variable attenuators in the range of 0  30 dB, which can be varied in steps of 2 dB are available separately for the two IF signals. An automatic level controller (ALC) at the output of the IF can be bypassed if required (e.g. for observations requiring measurement of variations at high time resolution). At the CEB, the 130 and 175MHz signals, recovered from the output of the optical fibers, are fed to the Base Band system for conversion to baseband signals. These signals are first converted to 70 MHz band and then split into Upper and LowerSidebands using the tunable Fourth LO. This LO can be tuned between 50  90 MHz in steps of 100 Hz. At the base band, bandwidths of the 62.5, 125, 250, 500 KHz or 1, 2, 4, 8, 16 MHz can be selected. Another ALC is provided at the output of the baseband. Signal levels are kept at 0 dBm by this ALC before being fed to the correlator (see Section 2.5). This ALC also can, however, be bypassed.
For the purpose of astronomical observations, settings of the attenuators and the First and Fourth LOs are important. The combination of these two LOs determine the exact RF frequency used for observations.
The output of the individual antennas is a random signal composed of the signals from the source of interest in the sky, the sky background emission and the thermal noise generated by the various electronic components through which the signal flows. The output of the antennas and the receiver is usually expressed in terms of the temperature of a blackbody which will emit equivalent power. The system temperature (T^{sys}) of an antenna at a particular frequency, is equivalent to the total power from the antenna when it points towards a blank sky. The power received by an antenna due to the celestial source alone is represented by the antenna temperature (T^{a}).
The normalized visibility amplitude from antennas 1 and 2 is given by
 (2.20) 
where T^{sys} = T^{r} + T^{bg} + T^{spill}, T^{r} is the receiver temperature, T^{bg} is the temperature of the sky background emission, and T^{spill} is antenna spillover temperature. An unpolarized point source of flux density S (in units of Wm^{2}Hz^{1}) will result in T^{a} = A_{e}S∕2k = ηS where A_{e} is the effective collecting area of the antenna, k is the Boltzman’s constant and η is the antenna sensitivity. For identical antennas and for the range of S where T^{a} ≪ T^{sys}, V _{12}≈ T^{a}∕T^{sys} ∝ S. T^{r}, T^{sys} and η are therefore important telescope parameters, which together determine the sensitivity of the telescope.
These and other astronomically relevant parameters of the GMRT are listed in Table 2.2^{1} . Fifth column is listed the angular resolution for both Central Square and full array.

The prototype 4 and 8 antenna correlator, all of the initial design and most of the implementation of the final 30 antenna single sideband correlator currently in use and which was used for the observations for this dissertation were designed and built by the team consisting of C.R.Subramanya, A.Dutta, V.M.Tatke, A.Dikshit, R.K.Malik, U.Puranik, M.Burse and others. Further development for the second sideband correlator and maintenance is now done by Y.Gupta, M.Burse, S.Sirothia, C.P.Kanade, K.H.Dahimiwal, I.Halagalli and others, with contributions from A.Roshi. This section describes aspects of the GMRT correlator relevant from the point of view of astronomical observations. A more detailed description of the current correlator hardware design is given by Tatke (1998). The correlator control and data acquisition software for potentially multiple subarray mode of operation was designed and implemented by R.K.Singh and C.R.Subrahmanya. A detailed description of the current correlator control and data acquisition software is given by Singh (2000). The subarray mode of operation is described by Chengalur (2000).
The GMRT backend for interferometric observations is an FXtype digital correlator. The block diagram in Fig. 2.9 shows the various stages of the correlator. The first stage of this correlator consists of a set of four samplers per antenna  one for each of the two IFs per sideband of the antenna. Each antenna produces two IFs which can carry either the two orthogonal polarization signals or the same polarization from two different RF bands (as in the case of 233∕610 MHz dual frequency feed). The analog to digital converter (ADC) of the samplers converts the input analog signal to 8bit unsigned numbers. These 8bit samples are converted to 6bit unsigned numbers before being fed to the next stage of the correlator, referred to as the Delay and Data Preparation Cards, or the Delay/DPC system. These samples are further converted to 4bit signed numbers in the Delay/DPC system (see the section on Samplers below), before being fed to the next stage of the correlator. This conversion from 8bit to 4bit samples is required since the VLBA FFT chips used in the correlator work with 4bit signed numbers. Part of the compensating delay equivalent to integral units of correlator clock is applied in the Delay/DPC stage. Any residual delay (corresponding to a fraction of the correlator clock) is applied as a phase ramp across the band, after the signals have been Fourier transformed (see Section 2.5.2). The Delay/DPC stage is followed by the Fast Fourier Transform or the FFT stage which performs a 512point FFT corresponding to 256 point complex spectra (the other 256 points being the complex conjugate spectra). The output complex numbers are represented in 4,4,4bit format consisting of two 4bit numbers for the mantissa of the real and imaginary parts and a common 4bit exponent.

The correlation between signals from two antennas with the same polarization is referred to as the copolar visibilities. This correlation is done by multiplying the antenna signals and averaging the result in the correlator. The output of the FFT stage is therefore fed to the Multiplier and Accumulation (MAC) stage of the correlator. Each MAC chip accepts 4 input data streams – 2 IFs from the two antennas. However each MAC can only handle a total of 256 channels per antenna. Therefore, either the number of channels per IF must be reduced from 256 to 128 or, only one IF with all the 256 channels must be fed to the MAC. This reduction of 256 channels for each IF to 128 channels to produce the copolar visibilities is done by averaging two adjacent frequency channels. However, the GMRT correlator can also be used with a resolution of 256 frequency channels by sacrificing one of the polarization signals from the antennas.
The MAC output format consists of two 15bit mantissa for real and imaginary parts and a common 6bit exponent (this format is referred to as 15,15,6bit format). The accumulators in the MAC stage provide the Short Term Accumulation (STA). Each STA cycle corresponds to 512 correlator clock cycles, required for reading the 512 points into the FFT section plus 4 clock cycles required to setup the FFT cards before it can perform an FFT. Thus, each FFT cycle requires 516 correlator clock cycles. The correlator runs on an effective clock of 32 MHz and the STA accumulates the data for 4096 FFT cycles. Thus, the STA corresponds to an integration of (512 + 4) * 4096∕(32MHz) = 66 ms of time. (Strictly speaking, the samplers run on a 32 MHz clock. This is the shortest time resolution that the GMRT correlator can provide. The Delay/DPC system reads the input data at the rate of 32 MHz but output data rate is 32.125 MHz. The rest of the correlator operates at 32.125 MHz to effectively take care of the dead time corresponding to the 4 clock cycles required for FFT setup).
The data rates corresponding to a STA time interval is large (8 MBytes/sec) and difficult to sustain for any data recording system. The STA data therefore needs to be integrated further to reduce the data rate by accumulating in the Long Term Accumulation (LTA) stage of the correlator. This stage has not yet been implemented in the hardware and therefore, the STA stage has been set to integrate for 8192 FFT cycles corresponding to a minimum integration time of 132 ms. This brings down the data rates to 4 MBytes/sec and the data is further integrated in the software for an integration time than can be set by the user to produce the final LTA data. The output of the STA stage is the final product of the correlator and is produced in the form of three data streams from the three MAC racks per sideband.
All the three MAC output streams are fed to the Data Acquisition System (DAS), via a 16bit bus. The DAS consists of a dual CPU PC running under the GNU/Linux Operating System (OS) with a data acquisition card on its PCI bus. The STA data from all MACs is acquired on this card at the rate of 4 MBytes/sec consisting of all the 256 channels (256 frequency channels with only one polarization, or 128 channels with both polarizations). The software device driver (which runs in the kernel space of the OS) and the supporting program (running in the user space of the OS) convert the input complex numbers from the correlator representation to the IEEE floating point representation. The software LTA is done at this stage for a time interval set by the user and the data rate further reduced. Through a series of network software, this data is transfered to another machine over the fast Ethernet to be recorded on the disk in what is referred to as the LTAformat (Singh (2000)). The last stage in the chain of software also provide facilities for further integration in time and frequency and for defining subarrays, etc. (the usage of the subarray mode of the software is described by Chengalur (2000)).
The analogtodigital converter (ADC) used in the GMRT samplers is an 8bit flash ADC (Tatke 1998). The input to these samplers is the band limited gaussian random noise from the antennas, converted to baseband (BB) (corresponding to a frequency range of 0 to 16 MHz). The output power of the baseband system is proportional to the antenna temperature which is a function of the sky temperature. The sky temperature, can vary by as much as a factor of 4  5 between the Galactic plane and away from it. However, the samplers are designed to sample a fixed range of input signal. The BB output signals are therefore maintained at 0 dbm by the Automatic Level Controller (ALC) before being fed to the samplers. This corresponds to a peaktopeak variation of ±1 V spanning the input random signal up to the six sigma level. In order to preserve the statistical properties of the input, the 2^{8} comparator voltage levels (in the ADC) must be optimally set such that the samplers neither saturate when the input voltage deviates by six sigma from the mean, nor sample the weaker signals too coarsely (corresponding to much smaller deviations from the mean). As of now, the 2^{8} voltage levels corresponding to 8bit sampling are uniformly distributed across this range of voltage, corresponding to a voltage resolution of 7.8 mV (provisions however exist for a more sophisticated scheme to distribute the levels across the range of input signal). However, before being fed to the next stage of the correlator (namely the Delay/DPC stage), the 8bits samples are converted to 6bit samples by dropping the 2 least significant bits (LSB) of the samples. Effectively, this corresponds to 2^{6} levels covering the full range of ±1 V, with a resolution of 31.25 mV, but which still sample the full range of the input signal up to six sigma, although at a coarser resolution. These 6bit samples are unsigned. Inside the Delay/DPC system these samples are further converted to 4bit signed values via a lookup table. The lookup table is made such that the 2^{4} levels are again uniformly distributed for the entire range of the input signal (in this case, the 6bit samples). Hence, effectively, the input signal with 6bit resolution is further sampled at a coarser resolution of 4bits without statistically altering the signal.
For the purpose of astronomical observations, this entire operation of converting the 8bit unsigned samples to 4bit signed samples can be ignored and the GMRT samplers can be treated as an effective 4bit samplers.
This section describes the mechanisms used in the correlator data acquisition software for deriving accurate time information. This is essentially a description of the work done by Singh (2000).
The input data to the DAS is averaged for a length of time set by the user to produce the long term accumulated (LTA) data. This interval is typically set to 10  20 sec. The LTA operation is currently done in the software and may be moved to hardware in future (if required).
As mentioned earlier, a synthesis radio telescope like the GMRT utilizes the earth rotation to synthesize an effective large aperture. To keep the array instantaneously phased, the time varying geometrical delay must be continually compensated. The phasing of the array is therefore a time critical operation requiring accurate time keeping. Accurate time information is also required for proper pointing of the antennas and tracking of an astronomical source. Stamping of astronomical data with accurate time is therefore very important.
Absolute time for the GMRT is kept using the real time clock (RTC) of the GMRT correlator control computer. This clock is synchronized with the GPS time using a GPS receiver. The GPS and correlator system clocks are used to remove any time jumps and/or slow drifts of this real time computer clock as described below. This time stamping of the visibility data in the GMRT correlator/data acquisition achieves an accuracy of ~ 100 μsec.
The correlator provides a synchronization pulse to the DAS to indicate the start of each STA cycle via an STA interrupt. This information is also supplied to the DAS via one of the bits of the data bus, referred to as the data synchronization bit. The DAS software uses this information to (1) do a consistency check (2) determine the start of the STA data in a stream of continuous data flowing on the data bus from the MAC. The DAS card also receives an interrupt signal at the start of each minute from the GPS receiver. The interrupt handler (the software which does time critical operations when a hardware interrupt is received) for both these interrupts (STA and GPS interrupts) reads the PC time from the PC RTC at each interrupt and maintains a list of corresponding PC time stamps. The list of PC time stamps recorded at each GPS interrupt is then examined for a constant difference of 1 minute. A continuous series is selected such that the time difference between successive samples is accurate to 100 μsec. This series of PC time stamps is referred to as a series of “good” GPSinterrupted time stamps. This series is used to set up a linear equation between the GPS time and the PC time. Since GPS time can be easily converted to the local time standard (Indian Standard Time or IST), this equation essentially converts the PC time to accurate IST time. This equation provides a first order calibration of the PC time, and takes care of any short glitch in the PC time (which can happen due to variety of reasons including overloading of the PC bus). Similarly, the PC time is recorded at each STA interrupt. This list of STAinterrupted PC time stamps is also examined and a continuous series selected such that the time difference between successive PC time stamps is equal to one STA cycle, to an accuracy of 100 μsec. This series of “good” PC time stamps is used to set up a linear relation between the PC time and STA time. The PC time to IST time equation is then used to convert the STA time to IST time, and is also used to tag each STA data block with a sequence number.
The three sources of timing information to the DAS (1) the STA interrupts (2) the GPS interrupts, and (3) the data synchronization bit, all provide time information with respect to the PC RTC. In the above scheme, on a short time scale, unless all these three sources produce drifts and/or glitches, time stamping of the data will remain accurate to 100 μsec. Since all these are independently running clocks, the probability of all three producing error at the same time is very low and it has been shown that the above time stamping scheme is quite robust to short time scale glitches and time drifts.
To account for slow drifts in the PC time, an attempt is made to set up a new linear equation between the STA sequence number and PC time after every 16 STA cycles. The new relation is not accepted unless it matches with the previous one within 100 μsec. Similarly, a new relation is set up between GPS and PC time at each GPS interrupt.
On a longer time scale, an error can occur due to some catastrophic error in either of these sources (e.g. the interrupts not arriving at the PC due to hardware failures). If the DAS software cannot set up the new equations for more than a threshold time interval, it is assumed that an irrecoverable error has occurred rendering the current equations invalid. In such a case, the DAS sets up fresh equations and the data between such an event and the time interval required to set up valid equations is unreliable and is flagged. In practice, however, such events are very infrequent.
However, a slow drift in the PC RTC, with respect to which all three sources of timing information are measured, can produce a drift in the final IST time stamps in the data. Data blocks are expected to arrive at the DAS at a predetermined constant rate. Data blocks themselves therefore can be effectively used as yet another clock, which is independent of the PC time. Any drift of the PC clock will show up as a consistent shift in the equations between the GPS and PC time and between STA and PC time with respect to the arrival of the data blocks. Hence, any consistent error in the arrival of the data blocks measured with respect to the IST time, derived from the two linear equations, will indicate a drift in the PC time. This fact is used to monitor a slow drift in the PC time over the time scale of setting up new equations.
The PC time is also synchronized to the minute pulse from the GPS. Errors in the PC RTC less than half a minute can therefore be corrected and PC clock synchronized to the GPS clock. However if the PC RTC is in error by more than half a minute, this synchronization will result into minimum of 1 min in the PC RTC. It is therefore important to set the PC RTC to within 1∕2 min of the GPS time.
As mentioned earlier, the compensating delay (equal to τ_{g} + τ_{Fix}; see Section 2.1.1, page 86) which needs to be applied to one of the antennas of an interferometer is applied in the GMRT correlator in two stages. A delay equivalent to an integral multiple of the correlator clock (τ_{Int}) is applied in the Delay/DPC section of the correlator. The input data stream is copied into a dual port RAM at the location given by the writecounter. Data is read simultaneously from this RAM from the location given by the readcounter. The offset between the read and the writecounters for this RAM is maintained such that the readcounter lags behind the writecounter. Since the output samples are read with a time delay (equal to the difference between the counters multiplied by the correlator clock cycle), the output data stream corresponds to the input data stream with a time delay. The time delay, however, can only be in multiples of the correlator clock cycle. The residual delay, corresponding to a fraction of the correlator clock cycle is applied in the fractional sample time correction or FSTC operation. The last stage of the FFT butterfly network provides a facility to apply a channel dependent phase. The frequency channel dependent time varying, phase corresponding to this residual delay, represented by τ_{FSTC}, is applied as a linear phase gradient across the band as
 (2.21) 
Note that τ_{FSTC} includes the residual delay due to delay resolution in the Delay/DPC system of the correlator plus the channel dependent residual delay of Equation 2.18.
The output of the FFT stage of the correlator is a set of 256 independent complex numbers. These complex numbers are represented by three 4bit numbers; one each for the mantissa of real and imaginary parts and one for a common exponent. The complex gains, applicable at the last stage of the FFT butterfly network, are represented by 5 bits each for the real and imaginary parts but has no exponent bit. Thus the channel dependent phase gradient for the FSTC is also limited to a 5bit representation.
The FFT stage of the correlator performs a realtime Fourier transform of the input signal effectively splitting the input bandwidth of Δν into 128 frequency channels, each of width Δν∕128 (the FFT stage of the correlator can be treated as a set of 128 filter banks). The time series from each of the 128 frequency channels from an antenna, are then multiplied with the corresponding channels from all other antennas, before being integrated in the MAC section of the correlator. The effective input bandwidth, as seen by the multiplier is therefore Δν∕128. For the GMRT, the maximum value of Δν is 16 MHz corresponding to a maximum channel width of 125 KHz, giving a fringe washing function of width ~ 8 μsec. Therefore, for the GMRT correlator, the residual delay must remain less than a fraction of this width (typically less then few micro seconds).
In the final double sideband correlator, in the continuum, nonpolar mode, the correlator can be used to record the two copolar visibilities (the nominal RR and LL correlations) per antenna with a full bandwidth of 32 MHz (both sidebands). In the current single sideband correlator only one sideband of 16 MHz can be processed. In the full polar mode, which will be available with the dual sideband correlator, the cross polar visibilities (namely RL and LR) cal also be recorded. Since the correlator can handle only 256 channels per antenna at the MAC stage, the cross polar mode can be used with only one of the sidebands with a maximum bandwidth of 16 MHz.
The GMRT correlator naturally produces spectral visibilities with a maximum of 256 channels across the entire observing band, if only one polarization is used, or with 128 channels, if both polarizations are used. For normal continuum observations, the 128 channels across a 16 MHz band provide a channel resolution of 125 kHz. For observations where bandwidth smearing is not important (and in the absence of RFI), these channels can be collapsed online to produce continuum visibilities. At lower frequencies where primary beams are large, bandwidth smearing can be important for sources away from the phase centre. Also, at frequencies below 610 MHz, intermittent RFI is always a possibility. To reduce bandwidth smearing and to detect and flag narrow band RFIcontaminated data, all the 128 channels are usually recorded. Provision for frequency averaging before recording does however exist and can be used where RFI and bandwidth smearing problems are not perceived to be serious.
For spectral line observations, channel resolution can be increased by controlling the sampling rate of the input signal. The sampling rate can be changed between 32 MHz and 0.125 MHz in steps of two. This corresponds to an effective bandwidth between 16 MHz and 0.25 MHz with a frequency resolution between 125 kHz and ~ 1.95 kHz respectively.
A synthesis radio telescope measures the twodimensional spatial frequency spectrum of the source structure. Instantaneous projected separations between the elements of the interferometer in the uvwframe determine the spatial frequency being sampled by the antenna pairs. The projected separation between the antennas, given by w, gives rise to the geometrical delay which needs to be compensated by delay tracking before the signals are multiplied. The value of w is given by
 (2.22) 
where (X_{ij},Y _{ij},Z_{ij})=(X_{i} X_{j},Y _{i} Y _{j},Z_{i} Z_{j}). The geometrical delay is therefore a function of the relative antenna positions. Precise measurement of the relative locations of the antennas is hence important for phasing the array, which is necessary for imaging with such a telescope.
The radio frequency (RF) signals from the two orthogonal feeds of the antenna are converted to two separate intermediate frequencies (IF) and brought to the Central Electronics building (CEB) for correlation via the optical fiber cables. The fixed delay suffered by the signals from each antenna varies from antenna to antenna since the physical length of the optical fibers to various antennas differ. This fixed delay is largely constant, showing small and slow variations due to changing ground temperature. Since the optical fiber cables are dispersive, any relative fixed delay between antenna pairs results in a time invariant phase ramp across the band in the visibility data. This phase ramp must also be removed or calibrated before averaging the channels to produce the continuum visibilities or before being used for imaging.
The following two sections describe the methodology used and results of baseline and fixed delay calibration.
Historically, the GMRT antenna positions were first determined using theodolite measurements. These were refined using Pband observations following the procedure laid down by Bhatnagar & Rao (1996). Later, dual frequency GPS measurements (Kulkarni 1997) were done to independently determine the antenna coordinates.
The GPS measurements provided the antenna coordinates relative to the reference GPS antenna, to a quoted accuracy of ~ 2 cm. However, monitoring the phase as a function of hour angle at 327 MHz indicated errors in excess of few meters for some antennas. These GPS coordinates were then further refined using Pband observations of celestial sources. Antenna based phases were determined using the GMRT offline program rantsol (Bhatnagar 1999) from a continuous (~ 7 hours) tracking of 3C48 at 327 MHz. The antenna based phases were modeled as a function of hour angle to measure the relative antenna coordinates (see below and Bhatnagar & Rao (1996)). The set of coordinates from these measurements were accurate to ~ 1 m for the arm antennas and ~ 0.1 m for the central square antennas. The phase variations at 327 MHz due to other factors (~ 30 arcsec error in antenna pointing which was later corrected, ionospheric contributions, etc.) did not allow determination of X and Y to better accuracies. The Z coordinate could not be calibrated at 327 MHz since not enough 327MHz calibrators could be found which could cover a large declination range but were close enough in hour angle. The proximity of calibrators in hour angle for Z calibrations was necessary to eliminate any unwanted scatter in the phase as a function of declination due to residual errors in the X and Y coordinates (see page 135).
These coordinates were then further refined using Lband observations (Chengalur & Bhatnagar 2001). Since the residual phase is proportional to the projected baseline length measured in units of the wavelength of the incident radiation, Lband is the best frequency to determine the residual coordinate errors in the antenna positions. For this purpose, VLA Lband calibrators 0217+738 and 1125+261 were tracked for several hours at 1280 MHz to compute the antenna based phases. Corrections of 0.5  1 m, consistent with the residual coordinate errors from the 327MHz measurements, were found for a few antennas. The final coordinates for the Central Square antennas are now known to an accuracy of ≈ 0.05 m. The residual phases for arm antennas correspond to position errors of 0.02  0.36 m. These errors do not repeat from daytoday and are also different for different sources indicating that the residual phases are probably not due to antenna position errors.

Assuming no errors in the antenna pointing (ΔH = Δδ = 0), the antenna based phase for an antenna with a position error of (ΔX,ΔY,ΔZ) with respect to a reference antenna located at the origin ((X,Y,Z) = (0,0,0)), as a function of the hour angle and declination is given by
 (2.23) 
where ν_{RF }^{o} is the frequency of the centre of the RF band, H is the Hour Angle, δ is the declination of the phase center, τ_{Fix} is the fixed delay suffered by the signal at the IF frequency ν_{IF } and c is the speed of light. The orientation of the X, Y, and Z axis is shown in Fig. 2.1.
As can be seen from Equation 2.23, an error in Z will contribute a constant phase for a fixed value of δ. It is therefore necessary to observe a number of sources at different declinations to distinguish between the phase contributed by the terms involving ΔZ and τ_{Fix} (both of which are independent of H).
The terms involving ΔX and ΔY in Equation 2.23 have a sinusoidal dependence on H. Hence, by tracking a point source for several hours (to get coverage in H while keeping δ constant), the left hand side of the above equation can be modeled as
 (2.24) 
where C is a constant and H is in unit of hours. Equating this to the RHS of Equation 2.23 the expressions for ΔX and ΔY (in meters) can be written as
A and ψ can be measured by fitting Equation 2.24 to the antenna based phases, with A, C and ψ as the parameters of the fit. Equations 2.25 and 2.26 then gives the antenna position errors along the X and Yaxis. ϕ, which is the antenna based phase computed with respect to the reference antenna located at the origin, was computed as a function of Hour Angle using the program rantsol. Equation 2.24 was then fitted to the phases from individual antennas to derive the values of C, A and ψ and Equations 2.25 and 2.26 used to derive the corrections for the X and Y coordinates of the antennas.

The ΔZ correction requires observation of point sources at various declinations. A set of VLA LBand calibrators with Lband flux densities > 1 Jy and covering a declination range of ≈±40^{∘} were used to measure ΔZ. Since the observations of each of these calibrators were separated in time, any residual error in the X and Y coordinates would reflect in the data as an undesirable spread of the phases as a function of declination. To minimize the effects due to residual errors in X and Y coordinates, these calibrators were chosen to be within ~ 1^{h} of each other in Right Ascension. The Z coordinate can then be found by fitting ΔZ sin(δ) to the antenna based phase as a function of declination. The final antenna coordinates, measured using this procedure, are tabulated in Table 2.3. The list of calibrators used for this purpose is given in Table 2.4. A typical phase variation as a function of declination, before and after correcting for the Z coordinate, is shown in Fig. 2.10. The variation of phase as a function of HA derived from ~ 5h long observation of 3C286 (28 Jy at 327 MHz), for a representative set of Central Square and arm antennas is shown in Figs. 2.11 and 2.12 respectively. The source of the short term drifts of ≈ 5  10^{∘} is likely to be the ionosphere. Phase variations over similar time scales and magnitude have been seen at the VLA and have been, to some extent, corrected using GPS based measurements of the variation in the ionospheric total electron content (Erickson et al. 2001).
The estimated accuracy of the antenna coordinates, derived from the variations in the phase as a function of hour angle and declination is ~ 5 cm. This will produce a maximum phase variation of ~ 0^{∘}.2 per degree in declination and ~ 5^{∘} per hour in hourangle at 1420 MHz due to errors in the antenna coordinates. Thus, observations of a calibrator located 20^{∘} away in the sky from the target source every halfanhour, would result in a phase error of a few degrees when the antenna based complex gains derived from the calibrator scans are transferred to the data on the target source. These figures will be scaled down by a factor of ~ 4 at 327 MHz. At this level, phases of many antennas drift in an unsystematic manner and constitute the source of dominant phase errors. The residual phase errors due to errors in the antenna coordinates after fringe stopping are therefore negligible.



The geometrical delay τ_{g} is suffered by radiation at the RF frequency ν_{RF } while the dominant fixed delay is suffered at the IF frequency ν_{IF }. The compensating delay is applied at the baseband (BB) frequency ν_{BB} at the correlator. The total time delay applied at the correlator, (τ_{corr}) is therefore related to τ_{g} and τ_{Fix} as
 (2.27) 
where ν_{i} is the centre frequency of the i^{th} frequency channel and equal to iν where Δν is the channel width. ν_{RF }^{o},ν_{IF }^{o} and ν_{BB}^{o} are the centre frequencies of the RF, IF and the baseband respectively. Noting that ν_{BB}^{o} = 0 the above equation can be written as
 (2.28) 
τ_{Int} corresponds to the time delay applied at the Delay/DPC stage while τ_{FSTC} corresponds to the residual delay applied in the FSTC operation (see Section 2.5.2).
In terms of the various frequency conversions in the signal flow, the RF band, for the purposes of delay compensation, can be written in terms of the first and the second local oscillator frequencies (which convert the RF signal to IF signal and the IF signal to baseband signal respectively) as ν_{RF }^{o} = ν_{LO1} + ν_{LO2} where, ν_{LO1} and ν_{LO2} are the the first and the second local oscillator frequencies respectively.


The residual phase across the band after delay compensation can therefore be written as
Assuming that τ_{Int} + τ_{FSTC} accurately compensates for τ_{g}, τ_{Fix} can be written as
 (2.30) 
The largest contribution to the fixed delay is due to the propagation delay in the optical fibers from the antenna base to the Central Electronics Building (CEB). These delays were measured by computing the antenna based complex gains for each of the 128 channels using the offline program rantsol, for a strong point source observed at 327 MHz using the full band 16 MHz. This essentially provided the antenna based amplitude and phase across the band. The slope of the phase across the band was measured and τ_{Fix} derived using the Equation 2.30 (the slope of phase as a function of frequency). The fixed instrumental delays thus measured are also listed in Table 2.3. The phase across the band before and after the fixed delay correction is shown in Figs. 2.13 and 2.14 respectively for a representative set of antennas.
For interferometric observations, frequency averaging (to produce continuum or multichannel continuum visibilities) is done after correlation and only the relative fixed delays are important. Since the same set of reasonably identical filters are selected at all antennas, the relative change in the fixed delays due to change of filters is not large. Also, since the ground temperature variation is similar from antenna to antenna and slowly varying over the scale of the array, its contribution to change in delays is not large. Although the delay component due to the filter selection and that due to diurnal variations must ultimately be tabulated and used for the delay compensation software, it is not a limiting factor and can be easily calibrated. The phase calibrators used for the continuum mapping observations used in this dissertation were strong (> 5 Jy) and were observed every halfanhour. The antenna band shapes were computed from the data on these calibrator scans and the interpolated gains for each channel applied to the visibilities of the target source. This provided the calibration for changes in the delay due to filters (small and fixed as function of time) and diurnal variations (small and slowly varying in time).