The total focusing method reconstructs an image by summing, for every pixel, the contribution of every transmit-receive pair in the full matrix capture. The image is only as good as the grid it is drawn on: if the pixels are coarser than the point spread function, the peak of a reflector response can fall between pixels and be under-reported.

The working rule is a pixel pitch no coarser than a fifth of the wavelength in the zone material, with a tenth used where amplitude accuracy matters most and a third being about the coarsest defensible. That rule exists to satisfy the amplitude fidelity requirement rather than the other way round – ASME Section V Article 4 Mandatory Appendix XI sets a 2 dB amplitude fidelity limit, and the grid has to be fine enough to meet it for the beam actually produced.

Cost rises steeply. Pixel count goes as the inverse square of the pitch, and the delay-and-sum load goes as pixels multiplied by the number of transmit-receive pairs and by the number of wave mode sets being imaged. Halving the pitch quadruples both the pixel count and the processing load, and the frame rate falls in proportion.

Use the wavelength of the mode actually imaged in the zone. Compression on plate at 5900 m/s gives a longer wavelength and a coarser permissible grid than a shear mode weld image at 3240 m/s, so a grid sized for one is not automatically valid for the other.

Compression or shear

Worked example

Probe frequency	5 MHz
Wave mode	compression
Material velocity	5900 m/s
Grid fineness rule	standard
Pixel pitch to be used	0.2 mm
Zone width	40 mm
Zone height (depth extent)	30 mm
Elements used in transmit and receive	64
Wave mode sets imaged	1
Wavelength	1.18 mm
Maximum pixel pitch	0.236 mm
Pixels across the zone	201
Pixels down the zone	151
Pixels in the zone	30351
Delay-and-sum operations per frame (millions)	124.32

At 5 MHz compression in steel λ = 5900/5000 = 1.180 mm, so the λ/5 rule gives a maximum pixel pitch of 0.236 mm. A 0.2 mm grid over a 40 by 30 mm zone is 201 by 151 = 30 351 pixels, and with 64 by 64 transmit-receive pairs that is 30 351 × 4096 = 124.32 million delay-and-sum operations per frame.