A mechanical filter is a signal-processing device that converts an electrical signal into mechanical vibration, passes that vibration through a chain of coupled metal resonators tuned to specific frequencies, then converts the filtered vibration back into an electrical signal. The Collins 455 kHz IF filter in classic Collins KWM-2 amateur radios is the textbook example. The mechanical resonators have far higher Q factors than LC circuits — typically 5,000 to 25,000 — so the filter cuts off adjacent channels with a steep skirt that an inductor-capacitor design cannot match. The result is cleaner SSB reception in crowded HF bands.
Mechanical Filter Interactive Calculator
Vary resonator machining error and tuning sensitivity to see the resulting frequency shift in a 455 kHz-style mechanical filter.
Equation Used
This calculator uses the article example as a linear tuning rule. A negative diameter error means the resonator disc is undersized, so the natural frequency shifts upward by k times the machining error magnitude.
- Linearized near the nominal resonator diameter
- Negative diameter error means the disc is undersized
- Sensitivity is taken from the article example: 10 um undersize gives about +200 Hz
Inside the Mechanical Filter
A mechanical filter takes the same job an LC bandpass filter does — pass a narrow band of frequencies, reject everything else — but it does it with metal bars vibrating in mechanical resonance instead of inductors and capacitors swapping energy. The input transducer converts the electrical signal into mechanical motion. That motion travels through a stack of precision-machined resonator discs or rods coupled by thin wires. Each resonator is tuned to the centre frequency of the passband. The output transducer converts the mechanical vibration back into an electrical signal. Because steel and nickel-iron alloys have intrinsic Q values of 10,000 or higher — versus 100-300 for a good toroidal inductor — the filter shape factor (ratio of -60 dB bandwidth to -6 dB bandwidth) gets down near 1.5, which an LC filter physically cannot reach.
The geometry has to be right or the filter does not work. Each resonator disc is typically machined to a diameter tolerance of ±5 µm, with the coupling wire diameter held to ±2 µm. If you grind a disc 10 µm undersized, that resonator's natural frequency shifts up by roughly 200 Hz at 455 kHz centre — enough to put a notch inside what was supposed to be a flat passband. The coupling wires set the bandwidth: thicker wire couples more energy and widens the passband, thinner wire narrows it. Get the wire length wrong by 0.1 mm and you'll see passband ripple climb past 3 dB.
The common failure modes are mechanical, not electrical. Drop a Collins F455 filter on a concrete floor and you can crack the brazed coupling joints — the filter still passes signal but the shape factor degrades and you get ringing on CW keying. Temperature swings matter too: nickel-iron resonators drift around 1.5 ppm/°C, so a filter calibrated at 25 °C running in a hot transmitter cabinet at 55 °C shifts about 20 Hz, which you'll hear as a slight pitch change on a steady carrier.
Key Components
- Input Transducer: Converts the incoming electrical signal into mechanical vibration. Most filters use a magnetostrictive nickel-iron rod wound with a coil, or a piezoelectric ceramic disc bonded to the first resonator. Conversion efficiency typically runs 5-15%, with the rest of the input energy dissipated or reflected — that's why mechanical filters need 10-20 dB of insertion loss budget.
- Resonator Discs or Rods: The frequency-selective elements. Usually 5 to 9 discs of nickel-iron alloy (Ni-Span-C or similar low-temperature-coefficient material), each machined to a diameter that sets its natural resonant frequency. A 455 kHz disc runs about 8 mm diameter and 1.5 mm thick. Disc-to-disc frequency match must hold within ±50 Hz.
- Coupling Wires: Thin steel wires (typically 0.3-0.5 mm diameter) brazed between adjacent resonators. The wire diameter and length determine how much vibrational energy passes from one disc to the next, which sets the filter's bandwidth. A 2.5 kHz SSB filter uses thinner coupling than a 6 kHz AM filter.
- Output Transducer: Mirror image of the input transducer — converts the filtered mechanical vibration back into an electrical signal. Source impedance is typically a few hundred ohms to a few kΩ, requiring an impedance-matching transformer for most receiver IF stages.
- Hermetic Housing: A sealed metal can, often gas-filled or evacuated, that protects the resonator stack from humidity and contamination. Moisture loading on the resonators alone can shift centre frequency 30-100 Hz and cut Q by half, so the seal integrity matters as much as the mechanical tuning.
Where the Mechanical Filter Is Used
Mechanical filters dominated narrow-band signal selection from the 1950s through the 1990s in any application that needed steep skirts at frequencies between 50 kHz and 600 kHz. They turn up wherever the engineer needed to pull one signal out of a crowded spectrum without the bulk and drift of crystal lattice filters or the wide skirts of LC tank circuits. They are still in use in legacy radio gear, in some sonar receivers, and in specialised telecom test equipment where the alternatives — crystal filters, SAW devices, or digital filtering — either cost more, drift more, or aren't power-efficient enough.
- Amateur Radio: Collins F455FA-21 mechanical filter in the Collins KWM-2 transceiver, providing 2.1 kHz SSB selectivity at 455 kHz IF.
- Military Communications: URC-32 and R-390A/URR HF receivers used Collins mechanical filters for the secondary IF, giving the R-390A its famous selectivity in crowded HF bands.
- Naval Sonar: Narrow-band mechanical filters in mid-frequency hull-mounted sonar processors, separating the active ping return from broadband sea noise.
- Telecom Test Equipment: Channel-spacing filters in HP 3551A transmission test sets used for FDM telephony measurements at carrier frequencies in the 60-108 kHz group band.
- Marine Radio: Furuno and Icom marine SSB radios used 455 kHz mechanical filters through the 1980s for distress channel selectivity on 2182 kHz.
- Broadcast Monitoring: Rohde & Schwarz EK-070 and EK-890 monitoring receivers employed mechanical filters for narrowband HF surveillance work.
The Formula Behind the Mechanical Filter
The shape factor tells you how steep the filter skirts are — the ratio of the -60 dB bandwidth to the -6 dB bandwidth. This is the single most useful number for deciding whether a mechanical filter beats an LC or crystal alternative for your application. At the low end of what's practical, a shape factor near 4 is what a good LC filter delivers and tells you a mechanical filter would be wasted. Around 2.0 to 2.5 sits the sweet spot for a 5- to 7-disc mechanical filter — sharp enough to reject an adjacent SSB channel 3 kHz away, simple enough to manufacture without exotic tolerances. Push to 1.5 with 9 or more discs and you get the cleanest skirt available short of digital filtering, but insertion loss climbs past 15 dB and the unit gets fragile.
Variables
| Symbol | Meaning | Unit (SI) | Unit (Imperial) |
|---|---|---|---|
| SF | Shape factor (dimensionless ratio — lower is sharper) | dimensionless | dimensionless |
| BW-60dB | Bandwidth measured at -60 dB attenuation points | Hz | Hz |
| BW-6dB | Bandwidth measured at -6 dB attenuation points (the nominal passband) | Hz | Hz |
| Q | Quality factor of an individual resonator (sets achievable filter sharpness) | dimensionless | dimensionless |
Worked Example: Mechanical Filter in a vintage Collins R-390A receiver restoration
A retired naval communications technician in Halifax is restoring a 1956 Collins R-390A/URR receiver and measuring the selectivity of the original 4 kHz mechanical filter on the 455 kHz IF. He sweeps the filter on a tracking generator and needs to know whether the measured skirt steepness still matches Collins' published spec or whether the filter has drifted enough to justify replacement.
Given
- BW-6dB = 4.0 kHz
- BW-60dB = 8.4 kHz (measured)
- fcentre = 455 kHz
Solution
Step 1 — at the nominal measured condition, compute the shape factor directly from the swept response:
That's right on the original Collins published spec of 2.1 for this filter — the 7-disc resonator stack is still tuned correctly and the coupling wires have not fatigued.
Step 2 — at the low end of the typical operating range for a mechanical filter, a fresh 9-disc Collins F455J unit:
A shape factor of 1.5 is what you'd see in a top-tier 9-disc filter straight from the factory. The skirts drop almost vertically — adjacent-channel rejection improves by roughly 20 dB versus the 7-disc unit, but insertion loss runs 15-18 dB instead of 10 dB.
Step 3 — at the high end of the range, a degraded or low-disc-count filter:
A shape factor of 3.5 means the filter has either lost a resonator (cracked coupling wire), suffered humidity contamination that dropped Q on multiple discs, or was a budget 5-disc design to begin with. At this point you might as well use the LC roofing filter — the mechanical filter is no longer earning its insertion loss.
Result
The R-390A filter measures a shape factor of 2. 10, which matches the original Collins specification exactly — no replacement needed. Compared to the 1.5 sweet spot of a 9-disc filter and the 3.5 degraded-unit figure, this puts the technician's filter solidly in the middle of the healthy operating range. If a different unit measured 2.8 instead of 2.1, the three most likely causes are: (1) one cracked coupling wire from mechanical shock, which shows up as a dip 1-2 kHz off centre on the swept response; (2) moisture ingress past a failed hermetic seal, which broadens the passband symmetrically and drops insertion loss by 2-3 dB; or (3) thermal drift of a single resonator disc whose Ni-Span-C alloy has aged, putting that disc 100-200 Hz off the others and creating asymmetric skirts.
Mechanical Filter vs Alternatives
You have three real choices for narrow-band IF filtering between 50 kHz and 600 kHz: a mechanical filter, a crystal lattice filter, or a modern DSP implementation. Each wins on different axes. Pick the wrong one and you either pay for performance you can't use or you blow the noise floor.
| Property | Mechanical Filter | Crystal Lattice Filter | DSP / Digital Filter |
|---|---|---|---|
| Shape factor (typical) | 1.5 - 2.5 | 1.8 - 3.0 | 1.05 - 1.2 |
| Resonator Q factor | 5,000 - 25,000 | 10,000 - 100,000 | N/A (numerical) |
| Insertion loss | 8 - 18 dB | 2 - 6 dB | 0 dB (digital) |
| Frequency range | 50 - 600 kHz | 1 - 30 MHz typical | DC to Nyquist limit |
| Temperature drift | 1.5 ppm/°C | 0.5 ppm/°C (AT-cut) | 0 (clock-locked) |
| Shock sensitivity | High (brazed joints) | Medium (crystal blanks) | None |
| Unit cost (2024) | $80 - $300 (NOS) | $30 - $150 | $5 (MCU) + dev time |
| Best fit | 455 kHz IF in legacy gear | HF first IF, narrow CW | Anything new-design |
Frequently Asked Questions About Mechanical Filter
A discrete notch inside or just outside the passband almost always means one resonator disc has shifted frequency relative to the others, or one coupling wire has fractured. Mechanical shock is the usual cause — dropping the receiver, or even a hard chassis bump during transport, can crack a brazed coupling joint without breaking it cleanly. The filter still passes signal because the remaining resonators carry the load, but the broken disc reflects energy at its own resonant frequency and you see it as a notch.
Quick diagnostic: tap the filter housing gently with a plastic screwdriver handle while the sweep is running. If the notch jumps around or disappears momentarily, you have a cracked coupling wire making intermittent contact. That filter is finished — it cannot be repaired without unbrazing the entire resonator stack.
The most common cause is impedance mismatch at the input or output transducer. Mechanical filters are picky about source and load impedance — a Collins F455 wants to see 2 kΩ on each side, ±10%. If you've replaced the IF transformer or built the filter into a circuit with a different impedance, the transducer doesn't transfer energy efficiently and the extra loss shows up as flat attenuation across the passband.
Check the terminating resistors and the IF transformer turns ratio first. The second suspect is the DC blocking and tuning capacitors across the transducers — these are usually 100-680 pF silver mica, and if someone substituted ceramic disc capacitors, the higher dissipation factor adds 1-2 dB on its own.
For voice SSB on amateur HF bands, the 7-disc filter (shape factor around 2.1) is the right call. The 9-disc unit gets you a 1.5 shape factor, but you pay for it with 5-7 dB more insertion loss, which means you have to add another IF amplifier stage or accept a degraded noise figure. On a crowded 40 m band the 9-disc unit lets you copy a station 2.5 kHz away that the 7-disc filter can't separate — that's a real benefit for contesting.
For general listening, the 7-disc gives cleaner audio because the skirts aren't quite as steep and the group delay distortion is lower. Group delay matters for SSB intelligibility — a too-sharp filter rings on consonants and makes voices sound nasal.
No. There is no field adjustment. Each resonator disc was machined to a precise diameter at the factory and the coupling wires brazed in fixed positions. The only tuning element is the small trimmer capacitor across each transducer, and that only adjusts the impedance match — it does not move the resonator frequencies.
If the filter has drifted more than 200-300 Hz from its rated centre frequency, the cause is usually moisture contamination from a failed hermetic seal. Some technicians have had limited success baking the filter at 60-70 °C for 24 hours to drive moisture out, but the recovery is partial and temporary. Replacement is the realistic answer.
Nickel-iron alloy resonators drift roughly 1.5 ppm per °C. At 455 kHz centre frequency, a 30 °C swing between a cold January morning at 5 °C and a warm August afternoon at 35 °C shifts the centre by about 20 Hz. That's small but audible — a steady carrier on the BFO will pitch-shift by 20 Hz, which the ear hears as a tonal change.
More importantly, all the resonators don't drift identically because they were machined from different bar stock with slightly different alloy composition. Differential drift between discs broadens the passband and softens the skirts — by maybe 0.1-0.2 in shape factor over a 30 °C swing. If your selectivity feels worse in cold weather, this is why. Letting the receiver warm up for 30 minutes brings everything back into alignment.
Mechanical filters have non-flat group delay — energy near the band edges is delayed more than energy at the centre. For SSB voice this is mostly inaudible. For phase-shift-keyed digital modes it matters because the receiver decoder uses phase transitions to recover bits, and a 1-2 ms group delay variation across the passband smears those transitions.
For PSK31 at 31 baud the symbol period is 32 ms, so a 1 ms delay variation costs maybe 3% of the symbol window — tolerable. For faster modes like PSK125 or MFSK16 the relative impact is 4× worse and you'll see the decoder fail at lower noise levels than a flatter LC or DSP filter would. If you're optimising for digital modes, an SSB-shaped DSP filter beats any mechanical filter.
References & Further Reading
- Wikipedia contributors. Mechanical filter. Wikipedia
Building or designing a mechanism like this?
Explore the precision-engineered motion control hardware used by mechanical engineers, makers, and product designers.
