Fig. 7.

(a) Fiducial error for different source distributions in noiseless and SNR = 0 dB conditions; note that it becomes easier to define the head location when more sources are active, but this plateaus at around 4 mm and has little dependence on SNR. (b) Effect of noise for the three source simulation. At low SNR the head location is bounded by the prior uncertainty (20 mm) consequently source localisation error is even larger. Above 0 dB SNR however, the FID localisation error plateaus at around 4 mm; likewise the error in source localisation falls to zero. This plateau in fiducial error and unrealistically good source localisation performance are due to the fact that in the MSP sources are modelled as discrete non-overlapping patches and this leads to these quantisation effects. (c) An example of the confidence interval it is possible to place on peak source location (in this case for a single simulated source). This confidence interval takes account of both measurement and co-registration noise (in this case around 6 mm).