The R-390A/URR gets its reputation for selectivity from four Collins mechanical filters in the 455 kHz IF deck. They are the heart of the receiver, they are no longer made, and one wrong capacitor can kill one in seconds. This post covers what each filter is, which one to use for each kind of listening, where you can still buy them in 2026, and how to prove a filter is healthy before you trust it. Much of what follows is distilled from twenty-five years of R-390A reflector discussion; the footnotes credit the people who did the work.

What is actually in the IF deck

The BANDWIDTH switch on the front panel offers six positions. Four of them select a mechanical filter directly; the two narrowest add the 455 kHz crystal filter in front of the 2 kHz mechanical filter. In the words of the depot manual, when the switch is turned to .1, 1 or 2, the 2 kc mechanical filter FL502 is switched into the circuit, and the 0.1 kc and 1 kc positions also incorporate the crystal filter into the first IF amplifier circuit.¹

The six BANDWIDTH positions. The two red positions are not separate filters: the crystal filter is switched in series ahead of the 2 kHz mechanical filter.

The Collins type number tells you the bandwidth: John Kolb explained on the reflector that F455N-20 means a 455 kHz filter 20 hundreds of hertz wide, and that the NAVSHIPS parts list gives the four filters as 522-9163 (2 kHz), 9160 (4 kHz), 9161 (8 kHz) and 9162 (16 kHz), with loose spares carrying the same core number under a 526- prefix.² Norman Ryan added the physical layout: with the IF deck upright and the two shafts facing you, FL502 (2 kc) is at the left rear, FL503 (4 kc) right rear, FL504 (8 kc) left front and FL505 (16 kc) right front.³

All four are 455 kHz disc-resonator mechanical filters with magnetostrictive transducers at each end. They present a high impedance (nominally 100 kΩ) at input and output, and each transducer coil is resonated by a capacitor on the IF deck.⁴ That high-impedance design is why S-Line and KWM-2 filters, which are 2 kΩ in and out with a 390 pF resonating capacitor, are not drop-in substitutes even though they are also 455 kHz Collins mechanical filters.⁵ One caution when moving filters between decks: receivers on Order No. 14-Phila-56 have 8 and 16 kHz filters that need less resonating capacitance, and on those decks C507/C508 and C515/C516 are parked on ground lugs rather than in circuit, ready to be reconnected if a full-capacitance filter is fitted.⁶

Anatomy of a filter killer: C553

The claim that one capacitor can destroy a mechanical filter deserves more than a warning box, because the reasoning is simple and the evidence is on the record.

The circuit

The first IF amplifier V501 drives the mechanical filters from its plate. Section S502 (front) of the BANDWIDTH switch connects that plate circuit to the input of the selected filter, and S503 (front) connects the filter’s output to the control grid of the second IF amplifier V502. The rear sections of the same two switches short-circuit the input and output terminals of the filters not in use.¹ Between the V501 plate and the switch sits C553, the DC blocking capacitor. Its only job is to pass 455 kHz signal to the filter while holding the B+ at the V501 plate, about 195 V DC, off the filter’s input transducer coil.¹⁹ As long as C553 is a good insulator, the coil sees only a small RF voltage. If C553 conducts, the coil sees B+ through whatever resistance the leak presents.

Why the filter cannot survive it

• The transducer coil is tiny. Graham Baxter G8OAD, who has opened and rewound these filters, found the coil wound with 0.003 inch enamelled copper wire, about 610 turns per section with the phase reversing at the centre partition.⁷ A good coil measures only about 45 to 60 Ω.^3,7 Wire that fine, with that little thermal mass, has no tolerance for a steady DC current it was never designed to carry.
• The coil is a direct DC path to ground. The far end of the input coil returns to the deck ground (via the ground of the filter trimmer capacitors on decks that have them²⁰). A leaky C553 therefore puts a DC current through the coil continuously, not just during a transient. Nelson’s warning in Hollow State Newsletter #5 was blunt: the 0.01 µF paper C553 has a low voltage rating, frequently shorts, and takes the mechanical filter in use out with it; he knew of one owner who, finding the set dead, stepped through every bandwidth position looking for audio and popped all four filters in turn.¹⁹ A leak of a few hundred kilohms produces only a fraction of a milliamp, which does no immediate harm but is already enough to bias the transducer and shift the filter’s response. Paper dielectrics fail progressively, and faster once they warm, so the current rises until the winding opens or the enamel breaks down and the coil shorts to its copper cup.
• The filter’s own insulation is rated at only 300 V DC. G8OAD notes the filters were originally rated at 300 V DC to case, and that as they age they develop leakage paths of their own: conductive build-up on the glass feedthrough insulators, and a conductive film between the coil winding and the copper cup.⁷ A filter that is already marginal has even less margin against a leaky blocking capacitor.
• The 2 kHz filter is the one at risk most of the time. Because FL502 is in circuit for the 0.1, 1 and 2 kHz positions,¹ it is the filter most often connected to the V501 plate. It is also the scarcest and most expensive of the four.
• The output side is not innocent either. The AGC voltage for V502 is applied through the output coils of all four filters, so a filter that develops leakage to its case drags the AGC line down. Tatum traced an R-390A that distorted and blocked after ten minutes of warm-up to an 8 kHz filter reading 1.5 MΩ to case cold and 300 kΩ when warmed with a hair dryer.²¹ That filter was not killed by C553; it is a reminder that the coils live in a DC circuit at both ends.

What the original part was

The Y2K R-390A reference lists C553 as a 0.01 µF, 300 WVDC, 20 % paper capacitor, and adds that this is not a location for a cheap replacement.⁸ Paper capacitors of this era are the ones that fail leakage tests: one reflector member who put nineteen capacitors pulled from his set on a Sencore LC75 found about two-thirds with significant leakage and several with cracked cases.⁹ The original C553 in an unrestored R-390A is now sixty to seventy years old. One trap when you go looking for it: Figure 41 of TM 11-5820-358-35 mislabels C553 as C533, and the error was carried into the first Hollow State Newsletter article on the subject before Dallas Lankford corrected it; C533 is in the BFO circuit and is not the part you want.²⁰

The fix

1. Replace C553 with a 0.01 µF film capacitor rated at 600 V DC. Norman Ryan’s standing instruction on the reflector, delivered as an order, was to fit a 600 V Orange Drop because C553 is the B+ blocking cap that eats mechanical filters.³ A film-and-foil type is preferable to metallised film in this one position, since a self-healing clearing event in a metallised capacitor is exactly the kind of momentary leak you are trying to prevent.¹⁰ Lankford’s own preference was a 0.01 µF, 1000 V disc ceramic, with anything from 0.01 to 0.1 µF at 600 V or better acceptable.²⁰
2. If you want belt and braces, Lankford’s two-capacitor modification reroutes the filter ground return to the trimmer-capacitor ground and inserts a second 600 V-plus capacitor between that point and chassis, so that a short in either capacitor still leaves the filters protected.²⁰
3. While the deck is on the bench, replace C549 as well; the same advice that names C553 as the filter killer recommends doing both together.³
4. Before fitting any filter, with the deck powered, measure DC volts from the filter input terminal to chassis. It must read zero.
5. Test the old C553 you removed, if only for curiosity. A leakage tester, or a high-value series resistor and a voltmeter, will usually show why the advice exists.

Choosing the filter for each mode

The right filter is the narrowest one that still passes the signal you want without making it fatiguing to listen to. A few rules of thumb, followed by the reasoning.

AM

An AM signal occupies twice its highest audio frequency, so 8 kHz passes about 4 kHz of audio and 16 kHz passes about 8 kHz. On a quiet band the 16 kHz filter is the reason people say the R-390A sounds like a broadcast receiver. As soon as neighbouring stations appear, drop to 8 kHz; the loss of highs is minor and the cleanup is dramatic. The 4 kHz filter is the DXer’s AM filter: it clips the audio at about 2 kHz but keeps everything the ear needs for intelligibility.

Exalted-carrier and SSB

For a weak AM station being chewed up on one side, switch to 2 kHz, turn the BFO on and tune so that the carrier sits at the edge of the passband with the clean sideband inside it. This is exalted-carrier or ECSS reception, and the R-390A does it well because the 2 kHz filter has steep skirts. The same setting is the standard SSB filter: the mechanical filter is symmetrical about 455 kHz, so for SSB you offset the BFO by roughly 1.3 to 1.5 kHz to place the audio passband where you want it. With a product-detector modification (see the earlier Hermes Lite 2 / Thetis post for the SB-390, DashDist and 6BE6 options) this becomes very comfortable listening.

CW and the crystal-filter positions

In the 1 kHz and 0.1 kHz positions the crystal filter is doing the narrowing and the 2 kHz mechanical filter is only cleaning up what gets past it.¹ The crystal filter’s response is not symmetrical and its centre shifts slightly relative to the mechanical filters, so do not be surprised if a CW signal moves in pitch as you step through 2, 1 and 0.1 kHz. The 0.1 kHz position rings on anything but slow, steady sending and is genuinely useful only for pulling one signal out of a pile-up. For most operating, 1 kHz is the working CW filter.

Where to buy R-390A mechanical filters in 2026

No one manufactures F455N filters today. What is on the market is new-old-stock from contract spares, pulls from parted-out receivers, and the occasional estate lot. Prices reflect scarcity: a 16 kHz filter can be had for the price of a lunch, a 2 kHz filter for several times that. Sources that were active as of September 2026:

If you cannot find an original

You have three realistic options, and none of them is a plug-in ceramic filter. First, repair. Many “dead” filters are not electrically dead at all: the foam doughnuts that suspend the filter body inside the can have turned to goo, letting the mechanism sag against the barrel and stressing the lead-out wires. G8OAD’s repair walk-through, with photographs, applies to the later style of filter with a matt barrel and diametrically opposed coil connections (the early shiny-barrel type is built differently) and is on r-390a.net; it involves venting and unsoldering the end cap, cutting new supporting rings from closed-cell foam, cleaning the barrel and the feedthrough insulators, and re-seating the filter at the mid-point of the barrel.⁷ If the coils read correctly, try this before buying.

Second, a third-party mechanical filter. In 2005 a side-by-side study of the Collins filters against replacement filters made by Curry was run in an R-390A and written up as the Collins vs Curry filter study;¹⁵ read it before spending money on a non-Collins unit.

Third, ceramic. There is precedent: Jim N1MAA reported an early EAC deck (serial 24) fitted from the factory with Clevite ceramic filters in the 8, 4 and 2 kc positions and a Collins mechanical filter only in the 16 kc slot, and thought it sounded good, with better skirts than an R-390 but not as sharp as the Collins mechanicals.¹⁶ Gerald Johnson’s assessment was that the ceramics probably just met the specification while the mechanical filters exceeded it by a wide margin in skirt steepness and ultimate rejection, and that each ceramic filter needs its own set of BFO frequencies.¹⁴ The Army’s own instruction in the Y2K reference is blunt: faulty ceramic filters should be replaced with mechanical types.⁶ If you want to experiment anyway, Murata’s current 455 kHz CERAFIL families are catalogued here;¹⁷ you will need to design the transformation from the deck’s 100 kΩ terminations to the filter’s low impedance yourself, and the resonating capacitors for that position come out. Whatever route you take, keep any original 2 kHz filter for the 2 kHz slot; nothing else matches it.

Testing a mechanical filter

Test in this order. The first two steps take five minutes, cost nothing, and catch most dead filters. Do them before you buy, before you install, and before you blame the filter for a problem that is really in the deck.

1. Static checks (filter out of circuit)

1. Coil continuity. With an ohmmeter, measure the DC resistance of the input coil pair and the output coil pair. Each should read roughly 45 to 60 Ω (Ryan gives 45–55, G8OAD 50–60 for a good rewound coil), and the two coils of one filter should read about the same.^3,7 An open reading on either coil is a dead transducer. The four filters in a set should read similar to each other; a coil reading far higher than its mate has probably been cooked.
2. Insulation to case. Measure each terminal to the filter can. It should read infinity.³ A leak of even a few megohms matters: G8OAD found that a leaky filter disturbs the AGC and causes overloading on strong signals,⁷ and Tatum’s leaky 8 kHz filter read 1.5 MΩ cold and worsened with heat, so warm the filter with a hair dryer and re-check.²¹ In-circuit, the quick screen is resistance from J512 pin 6 (the AGC line) to ground with the deck unplugged: it should read infinity, and a finite reading points at a filter or capacitor on the AGC side.^3,21
3. Mechanical. Shake the filter gently next to your ear. Loose material rattling inside points to a broken disc or to deteriorated foam supports; a filter whose foam has failed may still pass signal but its response will be wrong, and G8OAD’s symptom for one such filter was a vague twittering behind SSB caused by blow-by.⁷

2. Check the deck before installing

1. With the IF deck powered and no filter fitted, measure DC voltage from each filter input socket terminal to chassis. It must be zero. Any DC here means C553 is leaking and will destroy the filter you are about to plug in.
2. Inspect, and if necessary replace, the resonating capacitors on the input and output of each filter position. The TM procedure for installing a replacement filter selects these values individually, by finding the capacitance that gives maximum output, and specifies silvered mica; the input and output values are not necessarily the same.⁶ A drifted or open resonating capacitor detunes the transducer and produces a lopsided passband and high insertion loss that looks exactly like a bad filter.

3. Dynamic test: sweeping the passband

This is the test that tells you whether a filter is good, and it is the only way to sort a “works but sounds odd” filter from a healthy one.

1. Fit the filter in the IF deck and select its BANDWIDTH position. Set the RF GAIN fully up, the AGC switch to MGC, and connect an AC voltmeter or oscilloscope across the DIODE LOAD terminals on the rear panel (or an RF probe at the IF output).
2. Inject a 455.000 kHz unmodulated signal from a signal generator into the IF deck input (or into the antenna jack on a known frequency, using the calibrator to confirm the dial). Reduce the generator level until the receiver is well clear of limiting.
3. Record the diode load level at 455.000 kHz. Then step the generator in 250 Hz increments (100 Hz for the 2 kHz filter, 50 Hz for the crystal positions) up and down from centre, writing down the level at each step until you are more than 40 dB below the peak.
4. Plot the results. From the plot, read off:
   • the actual centre frequency (midpoint of the 6 dB points), which should be close to 455 kHz;
   • the 6 dB bandwidth, which should be near the nominal figure;
   • the 60 dB bandwidth, and from it the shape factor (60 dB BW divided by 6 dB BW). Gerald Johnson’s benchmark was that modern 3 to 6 kHz mechanical filters run about 2:1, and the older R-390A filters are sharper at the bottom of the skirts;¹⁴
   • passband ripple, the variation in level between the 6 dB points. A few dB is normal; deep notches inside the passband indicate a broken or detuned disc.
5. Repeat for each filter and compare. A filter whose peak level is markedly lower than the others, whose passband is lopsided, or whose centre is well off 455 kHz should be treated as suspect even if it works.

4. Confirming alignment after installation

Once a filter is in and tested, align the IF deck as described in the TM, and in Chuck Rippel’s IF deck alignment notes on r-390a.net,¹⁸ using the 2 kHz position for the initial IF transformer peaking and then checking each wider position for a symmetrical response. Two cautions from the Hollow State Newsletter. Cornelius advised against attempting alignment of the 455 kHz IF transformers at all, and Lankford explained why: the transformers are stagger-tuned and fitted with Q-spoiling resistors to give a flat response in the 8 and 16 kHz positions, so peaking them all at 455 kHz, as one published product-detector mod suggested, degrades the flatness and raises the level at the detector.²² Touch them only with the TM procedure in front of you and a reason to. And if the IF transformers are peaked against a filter that is itself off-centre, the whole deck ends up aligned to the wrong frequency. Finish with a listening check on a broadcast signal, stepping through all six positions: the audio should narrow in even steps with no sudden drop in level and no change in the perceived tuning of the station.

Looking after the ones you have

• Replace C553 and the other original paper capacitors in the IF deck. This is worth repeating.
• Never apply a DC test voltage, such as a high-voltage insulation tester, across a transducer coil.
• Store spares upright, dry and away from strong magnets; the transducer bias magnets are part of the calibration, and the filters are mounted vertically in the deck for a reason.¹¹
• Record the passband plot of every filter you own. When a filter changes, the plot is what tells you.
• When buying, pay for a documented sweep. An untested “should be good” 2 kHz filter is a gamble at any price.