Online Calibration

The noise source in the cal coupler injects a signal that is about 5% of the level of the system temperature at a rate of 30Hz. This noise signal is synchronously demodulated and the following relationship holds:

$$T_{sys} = T_{cal} \times \frac{P_{on}}{P_{on} - P_{off}}$$ (5)

Here, $T_{cal}$ is the equivalent temperature of the noise source, $P_{on}$ is the power received while the noise diode is on, and $P_{off}$ is the power received while the noise diode is off. $T_{sys}$ is measured accurately once by placing a thermal radiator of known temperature (a microwave absorber at 300K, giving 300K + $T_{sys}$) directly above the feeds. The power received with this load in place is then compared with the power output with the antenna just looking at the sky ($\sim T_{sys}$ + 3K). This measurement is used to establish the level of $T_{cal}$, which is assumed not to vary with time. The measurement of system and antenna temperature is described by Sinclair and Gough (1991).

We now rejoin the signal path below the cal coupler, located at the bottom of the feed-horn. The cal coupler is mounted on another coupler that provides a vacuum seal between the cal coupler and the subsequent receiver system, which is housed in an evacuated, cryogenic cylinder. An air gap thermally isolates the cylinder, the contents of which are cooled by a helium pump to 20K. The second coupler is mounted on the polariser.

The dual function of the polariser is to select the two linear polarisations and the desired band, so it is also known as a band splitter. The polariser consists of four strips of metal inside a conical waveguide that `guide' two orthogonal linear polarisations into probes at the narrow end of the polariser. The way in which the metal strips select the linear polarisations can be thought of as analogous to the effect of a ridged waveguide, which constrains a particular mode to propagate along the region between the ridge and the roof of the waveguide. The probes are simply short (with respect to the wavelength) lengths of the cores of the coaxial cables that take the signal into the first amplifiers. Thus the polariser converts a wave into a voltage: the impedance of the circuit is effectively the same as a waveguide impedance, so it works like a well-matched load. The polariser's official title is quad-ridged orthomode transducer, or OMT.

Separate electronics exist for both orthogonal linear polarisations, which are measured simultaneously. The position angle of the polariser is fixed with respect to the antenna. As the antennas have altitude-azimuth mounts, the position angle of the linear probes rotate on the sky during the course of an observation (imagine a circumpolar line on the sky orbiting around a stationary antenna). Measurement of both polarisations is therefore required for polarised sources. The two linear polarisations can be combined to give an accurate total intensity: the Stokes I parameter. Without additional calibration, the other Stokes parameters (Q, U and V) can be in error by about 2% of the value of I. Measurements of the phase difference of the two linear polarisations (referred to as the X and Y (or A and B) polarisations) at the receiver by on-line hardware may need further off-line calibration.

For an unpolarised source, observing both orthogonal polarisations offers a $\sqrt 2$ improvement in signal to noise ratio for an I image over an image generated from only one polarisation. The two polarised outputs from the polariser go by way of coaxial cables into diplexers which filter off the 6cm band and 3cm band signals. Separate IF chains exist for each of the subsequent four signals. For simplicity, the path of one signal (say, 3cm band, X polarisation) is followed in the rest of this section. In fact there are eight such paths (four frequencies, two polarisations), and provision has been made at each stage for the addition of more frequencies.

Coaxial cable takes the signal from the diplexer into a low noise, broadband amplifier. The 20/13cm receivers use four gallium-arsenide field effect transistors (GaAs FETs), while the 6/3cm receivers each use four gallium-arsenide high electron mobility transistors (GaAs HEMTs). The low noise amplifiers (LNAs), amplify the signal by 30-40dB. The power supply and bias voltages of these amplifiers are monitored, as is the temperature and pressure in the cryogenic cylinder.

The next step is the conversion down to a frequency that can be digitised by the samplers. The 3cm and 6cm signals undergo four conversions to successively lower frequencies. After the first conversion, the 3cm band is converted down to the 13cm band and the 6cm band is converted down to the 20cm band. Therefore it can be seen that the 13cm and 20cm signals require only three conversions.

The local oscillator signals have a relatively narrow tuning range, lying in the regions between the radio frequency bands (this reduces the likelihood of self-generated interference). The conversion system is flexible, allowing the input signals to be within the same band (e.g., both in the 3-cm centred band), or one from each of the two bands simultaneously received by the wide-band feed-horn.

The final output from the conversion system is sent to the samplers for digitisation. A switch immediately before the samplers allows you to select (this is done in software) whether you want 4, 8, 16, 32, 64 or 128MHz output. The samplers do two things. First the analog (continuously variable in time and amplitude) signal is sampled into a discrete-time sequence of values. This process does not degrade the signal as it is sampled at the Nyquist rate (twice the inverse frequency of the bandwidth of the analog signal). Subsequently, each of the discrete-time, continuously variable values are converted to one of a finite set of values: this is called quantising. In order to keep the data rate at a manageable level, the quantisation is fairly coarse: in the case of one-bit sampling only the sign of the signal is retained. Two-bit sampling detects which of four states the signal is in when sampled, and so on.

Quantisation noise is added to the signal as a result of quantisation, and the result is degradation of signal to noise ratio. The sensitivity lost when quantised signals are correlated, however, is remarkably little: one-bit sampling recovers 64% of the information contained in the signal (Van Vleck and Middleton 1966, but see, e.g., Thompson, Moran & Swenson 1986, pp. 210-230). Two-bit sampling recovers 88%.

The digitised signals from the sampler are sent from each antenna to the correlator along optic fibres at a rate of 512 million bits per second for each IF. Four optic fibres run from the samplers in each antenna to the correlator, which is housed in a central building. Synchronising code is added to the data stream at the start of each integration cycle to allow each bit to be correctly identified at the correlator irrespective of temporal changes in the length of the fibre.

The correlator is a device that compares simultaneous signals from two samplers: if the signal from both antennas is in the same state, then the signals are correlated. A plane wavefront approaching the ATCA will, in general, arrive at different antennas at different times. The signals from the antennas therefore need to be delayed before being presented to the correlator so as to simulate a wavefront arriving at each antenna simultaneously. This is achieved by the delay units, which are located in the screened room or Correlator Room in the Control Building. There is one delay unit for each antenna, and each delay unit deals with the four IFs coming from its antenna.

The delay units also remove other delays caused by return path length differences, instrumental delay variations and the like, so that wavefront samples are presented to the correlator synchronously. The delay units are told what to do by the Correlator Control Computer (CACCC). The CACCC gets its information about the physical location of the antennas and the direction to the source from the principal online computer program CAOBS (from Compact Array observations).

From this point on we necessarily lose track of our single 3cm, X polarisation signal. The signals from the delay units go to the correlator by way of 32 bit parallel cables. The correlator is made up of a number of blocks. There are 21 blocks: one per baseline (for n antennas there are $\frac{n(n - 1)}{2}$ baselines), plus six additional blocks. Each block consists of eight modules, as one module is required for each IF of each antenna. Correlation is carried out by the blocks. Each set of 3 blocks is controlled by Block Control Computers (BCCs), microcomputers which can be re-booted just like a personal computer. The blocks are automatically reconfigured when the observing mode changes from continuum mode to spectral line mode or vice versa.

The output from the correlator is the Fourier transform of the frequency spectrum: the lag spectrum or cross-correlation function. The cross-correlation function of two signals is a measure of the correlation of the signals as a function of the time delay (lag) between them. Online software (SPD) allows you to display this spectrum.

The output from each block goes to CACCC (with display on PERICLES), which performs the Fourier transform to recover the cross-power spectra and then writes the data to hard disk. Data are averaged for the period of one integration, which can be up to 30 seconds. Integration times are further discussed.