DEV Community

Yang Fei
Yang Fei

Posted on

Antenna Switching: The dB You Lose Before the LNA

Every dB of loss placed between an antenna and a low-noise amplifier costs the same as a dB of noise figure, except you cannot buy your way out of it later with a better LNA. That single sentence explains most of the bad antenna-switching designs I have seen reviewed: the switch gets treated as plumbing, sized on power handling and connector type, and then the link budget quietly loses half a dB that nobody accounts for.

Antenna switching is where receiver design is decided, not where it is implemented.

The arithmetic that decides the part

Cascaded noise figure for two stages is:

F_total = F1 + (F2 - 1) / G1
Enter fullscreen mode Exit fullscreen mode

A 0.5 dB insertion loss is F1 = 1.122, G1 = 0.891. Follow it with an LNA of 1.0 dB noise figure, F2 = 1.259:

F_total = 1.122 + (1.259 - 1) / 0.891 = 1.413  ->  1.50 dB
Enter fullscreen mode Exit fullscreen mode

The system noise figure went from 1.0 dB to 1.5 dB. Sensitivity in a 1 MHz bandwidth at 290 K degrades by roughly 0.5 dB, which on a link with 3 dB of fade margin is 17 percent of your margin spent on a component that appears on no budget line.

Now do the same calculation with a 1.2 dB switch, which is what you get when the switch is specified for 200 W instead of 20 W:

F1 = 1.318, G1 = 0.759
F_total = 1.318 + 0.259/0.759 = 1.659  ->  2.20 dB
Enter fullscreen mode Exit fullscreen mode

An over-specified power rating just cost 1.2 dB of sensitivity. This is the single most common antenna-switch mistake: the part is chosen for the transmitter's power because it sits near the transmitter, when the constraint that matters is what happens on receive.

If the switch must handle high power and low loss simultaneously, that is a different, more expensive component, and it is worth naming in the requirement instead of discovering it in the BOM.

Isolation is not a number, it is a budget

In a duplex or multi-antenna system the receive port sees a fraction of the transmit power through the finite isolation of the switch and through antenna-to-antenna coupling. Do the sum in dBm, never in watts:

Term Value
Transmit power at antenna port +47 dBm (50 W)
Antenna-to-antenna coupling, 3 m separation -30 dB
Switch isolation, worst case over band -55 dB
Power arriving at LNA input -38 dBm

That is comfortable for most LNAs. Now shrink the antenna spacing to 0.5 m, coupling rises to about -18 dB, and the same switch delivers -26 dBm. Still survivable, but the LNA is no longer in its linear region for weak signals in the presence of the strong one — cross-modulation and desensitisation show up as "interference from nowhere".

The mistake here is treating isolation as a specification to meet rather than a term in a sum. Ask for isolation at the frequency where it is worst, not the typical value: a switch specified at 80 dB typical and 62 dB minimum over the band is a 62 dB part.

Absorptive or reflective, decided by what is on the unused port

A reflective switch opens the unused path. Whatever was connected to it — an antenna, a filter, a length of cable — now sees a mismatch and reflects. That reflection does not vanish. It couples back through the finite isolation of the switch into the selected path, and it changes with the impedance of whatever is connected, which changes with the antenna's environment.

In a two-antenna diversity front end this shows up as a receive level that shifts by a few tenths of a dB when the other antenna is disconnected, and as a phase that is not repeatable between switch states. Absorptive switches terminate the unused port in 50 ohms and the problem disappears. They cost insertion loss on the selected path and they dissipate power in the internal load, so they are not free — but for any system where the two ports are not identical, absorptive is the correct default.

Reflective is the right choice when the unused port is genuinely open by design, when loss matters more than repeatability, and when the switch is carrying high power on transmit where an internal termination would be a heat source.

Switching speed is usually the least important specification

Relays switch in 5 to 20 ms, PIN diode switches in under a microsecond, FET switches in nanoseconds. Almost every antenna application has a dwell time measured in milliseconds to seconds, so all three are fast enough. The specification that actually matters is what happens during the transition:

  • Hot switching — changing state while RF is present — arcs relay contacts and progressively destroys them. A relay rated for 10 million cold operations may survive 100,000 hot ones. If the system can be sequenced so RF is removed first, do that; it is free.
  • Make-before-break vs break-before-make — a break-before-break path momentarily leaves the receiver open, and in a high-Q front end that produces a transient that takes longer to settle than the switching time itself.
  • Settling, not switching — the specification to request is time from control edge to the point where insertion loss and phase are within tolerance, not the datasheet's switching time.

For TDD systems where the switch flips every frame, PIN diodes and FETs dominate for exactly this reason. For a ground station that reconfigures between passes, a relay is better: lower loss, no bias network, no control noise.

Where the design actually fails

Ranked by how often I have seen each one cause a field problem:

  1. Loss budget built from typical values. Typical 0.3 dB, maximum 0.8 dB over temperature. The link margin was computed with 0.3.
  2. Isolation specified at one frequency. The worst case is at the band edge, and the band edge is where the interferer lives.
  3. Control lines treated as DC. A relay coil driven by a long unshielded pair injects switching transients into the receive path. Opto-isolate and add a local reservoir capacitor.
  4. Connector interfaces that cost more than the switch. A 0.2 dB switch between two adapters and a damaged N connector is a 1 dB path. Torque matters; see the note on connector care.
  5. Power handling assumed to be average. A 100 W CW rating says nothing about a 1 kW pulse with a 10 percent duty cycle, and less about the peak field at the contact.

What to put in the requirement

Parameter Ask for
Insertion loss Maximum over band and over temperature, not typical
Isolation Minimum over band, at the worst frequency
Configuration Absorptive or reflective, stated explicitly
Power Average, peak, and whether hot switching is required
Switching Time to settle within tolerance, plus control interface
Life Operations at rated power, not just cold cycles
Return loss On the selected path, in both states

Two of those seven — maximum insertion loss and minimum isolation — decide whether the system works. The other five decide whether it keeps working.

The switch is a small part of the BOM and a large part of the link budget. Sizing it on power handling alone is how a receiver ends up 1 dB worse than the spreadsheet said, with nothing to blame.

Notes on coaxial switches, waveguide switches and PIN diode switches for the components referred to here, with terminations for the unused ports.

Top comments (0)