Defining the folding ambiguities
The method described above contains two inherent folding ambiguities, which are explained in the diagrams of Figure 1 .
Figure 1: Two types of folding ambiguities — trigger folding and sine-fit folding — are present in the described measurement method.
In the figure, ti and tj are the phases from the sine fit converted into time. D is the sampling-clock delay and si the sampling interval. Red crosses represent the samples from the reference channel, with A marking the first sample. B marks the first sample of the measured channel, represented as blue circles. T is the period of the sine wave.
The first ambiguity, called trigger folding, is due to the trigger instant falling between B and the subsequent red cross (one sampling interval after A), or between A and the subsequent blue circle (one sampling interval after B). The current assumption is that the digitizer’s first sample in an acquisition always precedes the trigger instant by less than a sampling interval.
Therefore, if the trigger instant (which is common to both digitizers) arrives between A and the subsequent blue circle (second diagram), it means the hypothesis of A preceding B is false and it is necessary to subtract one sampling interval from the difference ti– tj to obtain the sampling-clock delay D.
The second ambiguity, called the sine-fi t folding, is due to the sine fit, which returns a phase of ±p. The trigger can fall at a time such that channel A has a negative phase and channel B a positive phase (Figure 1, bottom diagram). In such cases, the period T of the sine wave must be added to the ti– tj difference to obtain the sampling clock delay D.
Examining measurement results
A series of actual measurements shows the precise synchronization that is possible with the methods described above. The three key parameters were the synchronization states, the sampling-clock delay and the stability of the sampling-clock delay.
Configuring the measurement
The test setup used for the measurements reported below was composed of two 12-bit digitizers  (Figure 2).
Figure 2: With the Agilent U1066A Digitizers, this histogram of the measured sampling-clock delay has a mean value of 1.115 ns and a standard deviation of 3.475 ps.
The sine wave signal providing the time reference was produced by a synthesized signal generator. The signal amplitude was adjusted to be about 80 percent of full-scale input. To ensure the same signal was being fed into both channels, a 50 Ohm passive splitter was used to connect two cables of identical length. A similar splitter and identical length cables were used with a function/arbitrary waveform generator  to provide a trigger pulse.
The digitizers were controlled through a MATLAB® script. Care must be taken to ensure that both digitizers are armed and ready before a trigger pulse is sent, preferably by controlling the function generator with the MATLAB script.