Successful pulsar timing requires optimal TOA precision which largely depends on the signal-to-noise ratio (S/N) of the pulse profile. Since the TOA uncertainty is roughly the pulse width divided by the S/N, using Equation (3) we may write the fractional error as
One of the main problems of employing large bandwidths is pulse dispersion. As discussed in Section 2.4, pulses emitted at lower radio frequencies travel slower and arrive later than those emitted at higher frequencies. This process has the effect of “stretching” the pulse across a finite receiver bandwidth, increasing and therefore increasing . For normal pulsars, dispersion can largely be compensated for by the incoherent de-dispersion process outlined in Section 3.1.
The short periods of millisecond pulsars offer the ultimate in timing precision. In order to fully exploit this, a better method of dispersion removal is required. Technical difficulties in building devices with very narrow channel bandwidths require another dispersion removal technique. In the process of coherent de-dispersion [110, 185] the incoming signals are de-dispersed over the whole bandwidth using a filter which has the inverse transfer function to that of the interstellar medium. The signal processing can be done on-line either using finite impulse response filter devices  or completely in software [290, 297]. Off-line data reduction, while disk-space limited, allows for more flexible analysis schemes .
The maximum time resolution obtainable via coherent dedispersion is the inverse of the total receiver bandwidth. The current state of the art is the detection  of features on nanosecond timescales in pulses from the pulsar B0531+21 in the Crab nebula shown in Figure 21. Simple light travel-time arguments can be made to show that, in the absence of relativistic beaming effects , these incredibly bright bursts originate from regions less than in size.
© Max Planck Society and the author(s)