Development of a 23cm patch antenna

Looking for a robust design for a portable antenna to use with my new DJ-G7E, I decided to try a microstrip patch.
This is just a precisely-shaped conductive plate suspended above a ground-plane; but by using PCB techniques, this becomes a simple rectangular "flat-panel" antenna. Simple estimates suggest that gains between 7 and 9dBi may be achieved, with a beamwidth in the order of 60 degrees.

The prototype 23cm patch Cross-section through the antenna

My researches uncovered a couple of design utilities:
em: Talk's Microstrip Patch Antenna Calculator
Patch design v1.6 by WB0DGF
but on comparing them, their results differed by a few percent - enough to shift the resonance by tens of Megahertz at this frequency!

Another "variable" in the design is the dielectric constant of the PCB substrate. This is a key parameter linking geometry of the patch to its resonant frequency, and so must be precisely known for the frequency in use. Now there are a number of expensive microwave substrates which are well-characterised, but I intended to try an "economy version" using ordinary FR4 fibre-glass board. The only option in this case is to design around a "guess-timate" value, and then measure the result to obtain a more precise value for use in the next iteration of the design, so I started off with a value of 4.5.

Given that my design could be some way off the desired frequency, I wanted to give it as wide a bandwidth as possible, and having read that this is increased by making the patch rectangular (rather than square) and using a thicker substrate, I came up with my "double-decker sandwich" idea. Instead of using a double-sided 1.6mm copper-clad FR4 board, I'd use a 3.2mm two-ply composite of two single-sided boards with their plain sides innermost. And as I'd be peeling the excess copper laminate off the upper board, I could make this easier by using a smaller board than is needed for the lower ground-plane.

Comparing the two design utilities, MPACalc suggested a patch of 53.8 x 69.7mm, whereas the WB0DGF program produced a smaller patch of 51.9 x 68.3mm. So erring on the large side (I could always cut it down a bit) I made my prototype patch 54.5 x 69mm.
Now the simplest way to feed the patch from 50-ohms is to use a "probe" fed through the ground-plane to a particular distance along the centre-line of the patch. Here the WB0DGF program came in useful, suggesting a distance of 17.5mm from the edge would provide the best match. This program also suggested the minimum size of the ground-plane required on the back of the board.

So armed with my "design", I ordered up a couple of sheets of FR4, and set to work with my scalpel and soldering iron (gentle heating with the iron melts the glue securing the copper laminate to the substrate, allowing it to be peeled away - but don't hold the iron on too long or you'll burn the fibre-glass). Then I fastened the two FR4 sheets together with four nylon bolts, and finally installed the "feed-probe", by drilling through from the rear for an SMA socket.

The rear of the prototype patch, showing the two attachment points The prototype patch mounted for horizontal polarisation

I did some field-testing of this antenna in February 2010, operating from the slopes of Winter Hill. I had it clipped to my walking-stick "mast" and found that it certainly had some gain over the Alinco flexi-whip, and still provided a broad beamwidth. I'd fitted two clips on the rear, so it was easy to rotate the patch to switch between horizontal polarisation (where the shorter edge of the patch is horizontal) and vertical. And it's certainly a convenient and robust enough antenna to stow in my backpack without the worries of damaging any "elements" which I had with my small 6-ele yagi.

I finally got the opportunity to "close the loop" on this design when Darren G7LWT brought his network-analyser system along to a meeting of the Bolton Wireless Club:

VSWR response of prototype 23cm patch

As can be seen from the network-analyser plot, my "guesstimates" had been quite close: the resonance was measured at 1303MHz, rather than 1297 - an error of around 0.5%! So if I use a dielectric constant of 4.345 next time it should be spot on...