For VLBI observations the individual elements of an array are not connected in real time; 
cannot be measured as the observations are taking place. It is the task of the correlator to recreate the conditions of an observation after the event, so as to allow the data recorded at each array element to be combined, thus forming (e.g. Romney 1995).
The Mark II data recorded on VHS video tapes at each of the SHEVE antennae were correlated at the Caltech/JPL Block II VLBI processor at Caltech, in Pasadena, California.
The Block II processor is a lag correlator. The digital data streams recorded on tape are aligned by the correlator using recorded time signals. The correlator calculates the geometric and tropospheric delay and the phase rotation as a function of time, according to a correlator model which incorporates accurate source positions, antennae positions, observing frequencies, observation epochs, and sidebands in a correlator control file. The correlator applies the appropriate delay and phase rotation to the data streams and forms the correlation of pairs of data streams for a number of values of delay. A 4 MHz sampling rate is used to fulfill the Nyquist theorem for the 2 MHz recorded bandwidth of the Mark II system. Each increment in delay, or lag, at the Block II processor is therefore 1/4 
s.
To correlate the SHEVE observations the correlator was used in two modes. First, to find the interferometer fringes on each baseline, the correlation was examined over a wide range in delay (64 s = 256 delay lags) on one baseline at a time. The fringes were found at different times during an experiment, allowing drifts in the fringe delay due to changes in the time standards at each of the antennae to be followed. This pass through the data at the correlator is known as ``determining the clocks''. During this step rough residual fringe rates were also determined. Once the clocks were determined that information was added to the correlator model, in the correlator control file.
``Determining the clocks'' is a very important step in the SHEVE data reduction process since Rubidium clocks are used as frequency standards at some of the SHEVE stations. Rubidium clocks are not as stable as Hydrogen masers and can undergo fast drifts. Also, significant jumps are not uncommon. Therefore, for each SHEVE experiment, fringes were found at two or three points per 24 hour period for the purpose of ``determining the clocks''. If clock jumps existed, the precise points in time that the jumps took place were determined.
The second correlator mode was then used to correlate the data streams from all antenna pairs, N(N-1)/2 baselines, simultaneously (with some redundancy). Because the clocks had been determined it was only necessary to form the correlations over a small range in delay (2 s = 8 delay lags), enough to contain the fringes at all times during the correlation. The correlator integrated the correlated signal and output to a file data for each of the eight values of delay every two seconds.