Spatial filtering in the Fourier plane
Two lenses, four focal lengths: a grating placed in front is split into its diffraction orders, each focused at its own point between the lenses, where a mask decides which of them reach the image.
Open in the canvas →Background
A fine pattern lit by a laser beam sends light into several directions: a grating of period d sends it into the orders m = 0, ±1, ±2…, at angles given by sin θ = mλ/d. The finer the pattern, the larger the angles. Each order is one spatial frequency of the pattern, m/d lines per millimetre.
A lens one focal length behind the pattern brings each direction to its own point in its back focal plane, where the field is the Fourier transform of the field in the front focal plane[1]. A second lens, placed so that the two share that plane, puts the light together again in its own back focal plane, as an inverted image of the pattern. Input, first lens, Fourier plane, second lens and output are each one focal length apart: the 4f arrangement[1].
The plane between the lenses is where the image can be edited. A mask there acts on the spatial frequencies one by one. An opening that passes only the centre removes the high frequencies, and the image loses its fine detail: it is blurred[1]. The pinhole spatial filter is the extreme case, where the mask keeps the centre alone.
What this setup demonstrates
A 633 nm laser, 5 mW in a 4 mm beam, lights a transmission grating of 50 lines per millimetre, set to send the orders −2 to +2. The first orders leave at 1.81° and the second at 3.63°. Two lenses of f = 100 mm follow: the first 100 mm after the grating, the second 300 mm after it, and the output plane lies at 400 mm.
In the Fourier plane, 200 mm after the grating, the five orders come to five points: on the axis, 3.17 mm either side of it, and 6.34 mm either side (f tan θ; the enlarged view shows them). The mask there is a Slit with an 8 mm gap. It passes the orders 0 and ±1 and stops ±2. Behind the second lens the three remaining beams cross again in the output plane, on a power meter that reads 3.00 mW.
Things to try. Narrow the mask to 6 mm or less: only the zero order passes and the meter reads 1.00 mW. Open it to 13 mm or more: all five pass, 5.00 mW. Set the mask back to 8 mm and the grating to 100 lines per millimetre: its first orders now land where the second ones did, 6.34 mm out, and the mask stops them: 1.00 mW. A pattern twice as fine needs an opening twice as wide to reach the image.
The Imaging with lenses calculator treats the same pair of lenses as a 4f relay: where the Fourier plane lies and where the image forms, for thin lenses and for real ones.
The tracer follows rays, so it shows which orders pass the mask and how much power they carry, and nothing more. It does not compute the image: the pattern of lines that the orders rebuild by interference where they cross, and how that pattern changes when orders are removed (with the zero order alone, uniform light with no lines at all). The power meter in the output plane adds the power of the beams.
The grating sends the same power into each order it is set to trace, a fifth each here. A real grating's orders are unequal, and their strengths are what encodes the shape of its lines. The mask is a slit, so it can pass a band around the axis but not block the centre alone; and, as everywhere in this top view, only spatial frequencies in the plane of the drawing exist.