sources we use th e A IPS routine s d c l n which utilises th e Steer-D ew dney-Ito alg o rith m (Steer, Dewdney & Ito 1984). T his is an im age plane alg o rith m which cleans a num ber of com ponents sim ultaneously an d successfully avoids strip in g in th e image.
For th e sta n d a rd images th e restoring beam is taken to be a gaussian beam w ith th e sam e o rien tatio n and full-w idth at half-m axim um as th e core of th e d irty beam . T his choice is often the optim um for the d a ta at han d . T h e pixel size is generally chosen so th ere are betw een four and nine pixels p er clean beam . T otal in ten sity m aps w ith this optim um resolution are m ade in this m an n er for b o th th e 1.4 GHz d a ta and th e 5 GHz d ata.
For th e wide field m aps a heavy ta p er is applied to th e uv d a ta. T his w eights th e uv d a ta by a gaussian function which has th e effect of suppressing inform ation from th e high sp atial frequencies. T he result is a m ap w ith a large b eam an d hence low resolution. In cases where wide-field m aps reveal confusing sources w ithin th e p rim ary beam m ultiple fields are cleaned to rid th e m ain field of sidelobes from th e confusing source(s). This is accom plished w ith little difficulty using th e m ultiple field capabilities of MX. For confused sources requiring tre a tm e n t w ith s d c l n it is necessary to generate and clean a large im age containing b o th the pro g ram m e source an d th e confusing source. F o rtu n ately th is situ a tio n does not occur often.
Following th e initial image deconvolution the resu lts are inspected for evi dence of errors or interference in th e d ata. T he m ost com m on source of im age d eg rad atio n is phase errors from tropospheric effects discussed above. T hese are rem edied by applying self-calibration to th e d ata. O ften th ere are a m p litu d e errors w here a p a rtic u la r a n ten n a has an abnorm ally low response. T hese are p a rtic u la rly evident in plots of uv am plitude as a function of baseline an d are p ro b ab ly due to a pointing error in the anten n a. In these cases th e d a ta for th a t p a rtic u la r a n ten n a is flagged out. In a few cases baseline based interference cause ch aracteristic ripples across th e image (see for exam ple Ekers 1986). In these cases th e offending baselines are identified an d flagged out.
T h e next step is self-calibration of th e images. Since phase erro rs are the m ost prevalent and dam aging errors in th e d a ta, only th e phases are corrected on th e first few passes th ro u g h self-calibration. T he in p u ts to th e self-calibration alg o rith m ASCAL are th e uv d a ta plus all clean com ponents up to th e first negative clean com ponent. T his serves as th e m odel for th e source. T h e alg o rith m com pares th e observed am plitudes and phases w ith those expected from th e m odel (th e clean com ponents) and calculates a n ten n a gain corrections in m uch th e sam e
m a n n er as decribed in th e calibration section above. T h e difference is th a t in stead of using an unresolved calib rato r to calculate a n ten n a gains, a m odel of the ob ject itself is used.
O nce th e d a ta has been edited and self-calibrated th e whole process of m ap m aking, cleaning, inspection and self-calibration is rep eated . For th e first few cycles only th e phases are corrected via self-calibration. A fter two or th re e cycles these phases converge to a p artic u la r value and th e am p litu d es as well as phases are corrected for th e next few cycles. These self-calibrations are im p o rta n t for increasing th e dynam ic range of th e m aps significantly. T h e final m aps ty p i cally have noise levels of betw een ~ 0.5m Jy and a few m Jy w ith corresponding dynam ic ranges of ab o u t 1000:1. These noise levels are close to those expected theoretically.