A machined aluminum enclosure is a good industrial design decision and a difficult RF decision. The two facts are independent, which is why the conflict surfaces late. We worked through this on a handheld remote control for a high-end audio manufacturer: a sealed aluminum body, an existing 433 MHz control protocol that had to stay backward compatible with fielded receivers, and a sleep current budget in the microamp range. The engineering below is general. The physics does not care what the product is.

What aluminum actually does at 433 MHz

The first thing to establish is that a metal wall is no partial attenuator at this frequency. Skin depth is given by the standard expression, one over the square root of pi times frequency times permeability times conductivity. For aluminum at 433.92 MHz, taking conductivity as 3.5 x 10^7 S/m, skin depth works out to roughly 4.1 micrometers. A 2 mm machined wall is about 490 skin depths thick. Nothing propagates through it, and no increase in transmit power recovers it.

Energy leaves a metal box through apertures. For a single opening the classical estimate for shielding effectiveness is 20 log (lambda / 2d), where d is the longest dimension of the aperture. At 433.92 MHz the free-space wavelength is 691 mm, so a 10 mm seam gives roughly 31 dB of attenuation and a 40 mm slot gives roughly 19 dB. The governing dimension is the longest one, including a diagonal. This is the slot-antenna behavior that EMC engineers spend their careers suppressing. In a product that has to transmit, it becomes the intended path.

Enclosure architecture therefore decides the link budget before any schematic exists. Useful apertures include a plastic end cap, a plastic bottom plate, an IR window, a battery door, or a deliberate dielectric insert. In the remote above the top was aluminum and the bottom was plastic, so the radiating aperture was the entire bottom face. That decision was made in mechanical CAD, and it was worth more than any later component change.

Why sub-GHz, and what it costs

Sub-GHz is the right band for this class of product. From the Friis relation, free-space path loss scales with the square of frequency. The ratio between 2.45 GHz and 433.92 MHz is 20 log (2450 / 433.92), or 15.0 dB in favor of 433 MHz at equal distance and antenna gain. Texas Instruments puts the same result operationally: halving the RF frequency doubles free-space range for equal antenna performance. Longer wavelengths also diffract more readily around furniture and bodies, which matters in a listening room.

The cost is size. A quarter wave at 433.92 MHz is 172.7 mm and a half wave is 345.4 mm. Neither fits in a handheld remote. Every antenna that does fit is electrically small, with low radiation resistance and high stored reactive energy. That is the whole problem in one sentence.

The radio side is well served. The TI CC1101 covers 300 to 348 MHz, 387 to 464 MHz and 779 to 928 MHz, delivers up to +10 dBm at 433 MHz, and reaches -116 dBm sensitivity at 0.6 kBaud with 1 percent packet error rate. It draws 29.2 mA transmitting at +10 dBm and 16.0 mA receiving at 1.2 kBaud, with 200 nA in sleep and 240 microseconds from sleep to TX. Its optimum differential load at 433 MHz is 116 + j41 ohms, so a balun and matching network come before any 50 ohm reference plane exists. One caution: the datasheet radiated second harmonic figure of -49 dBm at 433 MHz was measured with a real external whip, and that antenna contributes to the attenuation. A small internal antenna does not.

The antenna ladder

We ran the usual iteration ladder: PCB trace antenna, then chip antenna, then wire antenna, then a dedicated external transmitter with a proper whip. TI application note AN058 sets out the tradeoffs and the ordering is consistent. A trace antenna is free in bill of materials terms and highly sensitive to ground plane geometry and enclosure proximity. A chip antenna buys predictability and a specified footprint at the cost of efficiency and a mandatory keepout. A wire or whip reaches efficiency and gain no printed option matches. An external SMA whip for this band, for example the Ebyte TX433-JWG-7, is 75 mm long and specifies 2.5 dBi gain, SWR below 1.5 and a 50 ohm input. It is also 75 mm of exposed rod, which a luxury handheld product will not accept.

The ground plane is part of the antenna in every case. A quarter-wave monopole on an undersized ground plane behaves as a poorly balanced dipole whose counterpoise is whatever conductor sits nearby: the case, the battery pack, the user's hand. Keepout is the boundary of the radiating structure.

Detuning, and the difference between match and radiation

Two bench measurements are worth reporting. First, the aluminum case shifted the matching network resonance by about 8.8 MHz, while the bare antenna element moved only about 1.7 MHz. The case detunes the network more than it detunes the radiator, so any tuning done on a bare board is invalid the moment the lid goes on. Tune in the assembled product.

Second, and more instructive: an EM simulation predicted a feed impedance of about 1.9 - j62 ohms. The fabricated antenna measured about 2.2 - j50 ohms. That 12 ohm reactance error looks small. Matching a 2 ohm radiation resistance to 50 ohms requires a loaded Q near 22, which tolerates roughly 3 ohms of drift before half the power is lost. The as-built network delivered 8 to 10 percent of available power, close to -10 dB, and that matched the over-air deficit at the receiver.

Retuning improved the return loss and made the product worse. A series 22 nH inductor with a Q of about 30 contributes roughly 2 ohms of equivalent series resistance. Against a 2.2 ohm radiation resistance, half the delivered power heats the inductor. At low radiation resistance, S11 is a poor proxy for radiated power. The fix is to raise radiation resistance so the match stops being knife-edge, or to re-optimize the network against the measured impedance with component Q in the model.

Measure the right quantity

Two measurement errors cost real time. An SDR placed next to the transmitter reads near-field coupling. In our case it reported a board 7 dB high that was in fact 19 dB down at the receiver. Radiated performance must be measured beside the receiver, at operational distance, in the assembled product.

The VNA fixture must also be electrically short. An 8 cm pigtail rotated a calibration short to +83j ohms and made every reading after it meaningless. Shortened to 1 to 2 cm, the fixture reduced to a clean 7.56 nH series inductance that de-embeds reliably.

The regulatory ceiling

Under FCC 47 CFR 15.231(b), a periodic-operation intentional radiator between 260 and 470 MHz is limited to a fundamental field strength between 3,750 and 12,500 microvolts per meter at 3 meters, by linear interpolation, with spurious emissions between 375 and 1,250 microvolts per meter. A manually operated transmitter must deactivate within 5 seconds of button release. Under 15.231(e) the fundamental limit falls to a range of 1,500 to 5,000 microvolts per meter, each transmission is capped at one second, and the silent period must be at least 30 times the transmission length and never less than 10 seconds. In Europe, ETSI EN 300 220-2 and ERC Recommendation 70-03 allow 433.050 to 434.790 MHz at 10 mW e.r.p. with a duty cycle at or below 10 percent.

Every one of those limits is stated as radiated field strength. An inefficient antenna therefore gives away link margin and gains no regulatory headroom in return. The enclosure aperture, the ground plane and the match are where the margin lives.

References