Filter Selection: Taming Spurious Emissions¶
The filter is where physics imposes a hard trade-off: every dB of rejection costs you insertion loss, board space, or both. There is no free lunch.
After the mixer generates your desired RF signal — along with a constellation of unwanted spurs, harmonics, and LO leakage — a filter has to clean up the mess. The filter you choose determines your spectral compliance margin, your link budget loss, and often the physical size of your transmitter.
Why transmitter filtering is different from receiver filtering
In a receiver, the main concern is selectivity — rejecting out-of-band interferers to protect a sensitive LNA and mixer. In a transmitter, the concern is emission compliance. You're not protecting your own circuitry — you're protecting everyone else's spectrum.
Regulatory bodies (FCC, ETSI, 3GPP) specify emission masks that define how much power your transmitter can radiate outside its allocated channel and band. Violating these limits means your product doesn't ship. The filter is your primary tool for meeting these limits.
The critical difference: receiver filtering happens at low signal levels (microvolts to millivolts), while transmitter filtering often happens at moderate to high power levels (milliwatts to watts). This changes the filter technology options significantly. A thin-film BAW filter rated for +10 dBm maximum input power works fine after a mixer but can't be placed after a PA delivering +30 dBm.
Filter technologies and where they fit
SAW filters (Surface Acoustic Wave). Operate from roughly 50 MHz to 3 GHz. Small (typically 2x2 mm to 3x3 mm packages). Insertion loss of 1-3 dB. Typical rejection of 30-50 dB at 2-3x the bandwidth offset. Temperature coefficient of frequency (TCF) is -30 to -40 ppm/C for standard SAW, which means the passband shifts with temperature. Temperature-compensated SAW (TC-SAW) reduces this to -5 to -10 ppm/C at a cost premium. Power handling is limited — typically +10 to +15 dBm maximum for standard devices, +25 dBm for high-power variants.
BAW filters (Bulk Acoustic Wave). Operate from 1.5 GHz to about 6 GHz, with FBAR variants reaching higher. Better power handling than SAW — typically +25 to +30 dBm. Steeper skirts than SAW at similar sizes. Lower temperature sensitivity. The go-to technology for 5G sub-6 GHz band-select filtering. Insertion loss of 1-2.5 dB with 40-55 dB of rejection. Higher cost than SAW.
Ceramic filters. Dielectric resonator filters using high-Q ceramic materials. Available from 400 MHz to 10+ GHz. Moderate size (5x5 mm to 15x15 mm depending on frequency). Insertion loss of 1-3 dB. Good power handling — +30 dBm or more. Commonly used in base station and infrastructure applications where a bit more board space is acceptable for better performance.
Cavity filters. Metal enclosure with resonant cavities. The highest Q-factor of any filter technology — 1,000 to 10,000 or more. Extremely low insertion loss (0.1-0.5 dB) and steep rejection (60-80 dB achievable). But physically large — a 2 GHz cavity filter might be 30x50 mm or larger. Used in base stations, repeaters, and military applications where size is secondary to performance.
LC filters (lumped element). Inductors and capacitors on the PCB. No frequency lower limit. Practical up to about 6 GHz using 0201 or 01005 components. Custom bandwidth — you design exactly what you need. No insertion loss floor from a resonator — but inductor Q-factors limit practical performance to 15-25 dB rejection per section. Multiple sections increase rejection but add loss. Best suited for harmonic filtering after the PA, where the harmonics are far from the fundamental and moderate rejection is sufficient.
LTCC filters (Low Temperature Co-fired Ceramic). Multi-layer ceramic filters that integrate resonators and matching networks. Compact packages. Used for integrated front-end modules. Insertion loss of 1.5-3 dB. Good for mass-production applications where a custom filter response is needed in a small footprint.
The insertion loss vs. rejection trade-off
This is the central design tension in filter selection. Every filter technology faces it.
A SAW filter with 40 dB of rejection at a given offset might have 2 dB of insertion loss. A similar SAW with 50 dB of rejection needs more resonator sections, increasing insertion loss to 3 dB. That extra 1 dB of loss means your PA needs to deliver 1 dB more power — which increases DC power consumption by roughly 25% (at typical PA efficiency levels).
Alternatively, you can achieve higher rejection with the same insertion loss by using a higher-Q technology — moving from SAW to BAW, or from BAW to cavity. But each step up in Q comes with a step up in cost and usually size.
The practical question is: how much rejection do you actually need? The answer comes from the spur analysis (Part 3) and the emission mask requirements. If your mixer produces an image spur at -30 dBc relative to the desired signal, and the emission mask requires -60 dBc at that frequency, you need 30 dB of filter rejection at the image offset. Not 60 dB. Over-specifying rejection wastes insertion loss, board space, and cost.
Bandwidth considerations
Fractional bandwidth. Filter bandwidth as a percentage of center frequency. A 100 MHz bandwidth filter at 2.4 GHz has 4.2% fractional bandwidth. The same 100 MHz bandwidth at 28 GHz has 0.36% fractional bandwidth — which is extremely narrow and limits technology options. SAW and BAW filters can achieve fractional bandwidths of 1-5%. Cavity filters can go below 0.5%. LC filters can achieve very wide or very narrow bandwidths but with limited rejection.
Group delay variation. Filters introduce frequency-dependent delay. A steep-skirted filter with sharp cutoff produces significant group delay variation near the passband edges. For wideband modulated signals (like 100 MHz 5G NR channels), this group delay variation can distort the signal and increase EVM. Group delay equalization — either through filter design or digital pre-compensation — may be necessary.
Passband ripple. Chebyshev-type filter responses achieve steeper roll-off than Butterworth for a given order, but at the cost of amplitude ripple within the passband. For a transmitter, ripple directly creates amplitude variation across the signal bandwidth. Typical acceptable ripple is 0.5-1 dB for most modulated waveforms.
Filter placement in the transmitter chain
Where you place the filter in the chain matters as much as which filter you choose.
After the mixer (channel/image filter). The most common position. Signal levels are moderate (typically -10 to +10 dBm), so SAW, BAW, and ceramic filters all work. The filter needs to reject mixer spurs and the image frequency. This position has the widest technology selection.
Between driver and PA (interstage filter). Used when the driver amplifier's harmonics need to be cleaned up before reaching the PA. Signal levels are higher (+10 to +20 dBm), which limits options — BAW, ceramic, or LC filters. SAW filters may be at their power handling limit.
After the PA (harmonic filter). The signal is at full output power — potentially +30 to +43 dBm. Only LC filters, ceramic filters, and cavity filters can handle this. SAW and BAW are excluded by power handling. The primary targets are second and third harmonics, which are far from the fundamental, so moderate rejection (20-30 dB) from a simple LC low-pass is often sufficient.
Temperature and aging effects
Filters are physical resonators, and their response shifts with temperature. A SAW filter with -35 ppm/C TCF shifts its center frequency by 3.5 MHz at 2.4 GHz over a 40C temperature change. If your passband margin is only 5 MHz on each side, that shift can push the band edge into your signal bandwidth.
Specify filters with temperature-compensated variants (TC-SAW, temperature-stable BAW) for applications with wide operating temperature ranges (-40 to +85C industrial, or -55 to +125C military). The cost premium is typically 20-40% over standard variants, but the alternative — designing extra passband margin into the system — wastes bandwidth and may not be possible if the channel allocation is fixed.
What filter technology do you default to, and what forces you to move up to the next tier? I find that SAW covers most designs below 3 GHz, but the power handling and temperature stability limits push you to BAW sooner than expected.
This is Part 4 of an 8-part series on RF transmitter design. Previously: "Mixer and Upconversion." Next: "Driver Amplifiers: The Stage Everyone Underestimates."