Multi-frequency, or bandwidth synthesis, is the practice of combining continuum data at several frequencies during the imaging stage to produce a single output image. u and v for each correlation are calculated from its actual frequency (rather than some mean frequency), and each correlation is convolved onto the gridded u-v plane as a separate visibility.
There are a number of advantages to this:
The ATCA is an excellent instrument for multi-frequency techniques. A single observation could result in several frequencies in three ways. Firstly the normal continuum mode is for the correlator to divide the 128-MHz bandwidth into 32 channels (although, due to bandpass effects, usually only the central 100 MHz is useable). Indeed, the natural ATCA imaging mode should be to use multi-frequency synthesis. In the 20-cm band, the 128 MHz bandwidth provides an 8% frequency spread. If all the channels are retained throughout data processing, this is enough to significantly improve the u-v coverage. Secondly, the ATCA has the ability to observe two separate frequency bands simultaneously. These two bands can be combined into the one output image. Thirdly, the wide bandwidths of the front-ends and the frequency agility of the receivers means that observing frequencies can be switched rapidly -- indeed every integration. In practice, because of problems with the hardware that measures system temperature, it is generally not advisable to shift frequencies more often than once every few minutes (see Russell Gough's note ``Transient Response of the Receiver System''). While switching frequencies with time does result in better u-v coverage, it clearly does not improve sensitivity. It also involves a trade-off between tangential and radial holes in the u-v plane.
All three of the above approaches may be used together to maximise the number of
frequencies observed in a single observation. The applicability of each
of them will depend on the objectives of the observation. Observations
where high dynamic-range images are desired will benefit from the
improved u-v
coverage that results from frequent changes of the observing
frequency. Detection experiments are unlikely to benefit from this frequency
switching, unless there are confusing sources that need to be imaged.
However, using the ATCA's ability to observe at two different bands will
double the observing bandwidth, and consequently reduce the final noise
by a factor of 
(do not observe exactly the same band -- you will
measure exactly the same signal, and noise, twice, and so you will not get
a 
improvement).