The 75S-3B C29 Capacitor: Why a Failed Part Makes the Audio Oscillate, and What the Bench Tests Show
On 13 November 2026 Dick Weber, K5IU, reported an interesting fault to the Collins Collectors Association reflector. A 75S-3B on his bench broke into audio oscillation when the volume control passed about 80 percent. The cause was C29, an 8 µF 25 V electrolytic, which had gone open. A new capacitor cured it.
Dick also noted two things that did not fit. He has added a C29 to at least a dozen 75S-1 and early 75S-3 receivers that never had one, and none of them oscillated beforehand. And he could find no Collins bulletin that mentions the oscillation. This article looks at both points, proposes a change, and then reports what Dick’s follow-up bench tests showed.
What Collins published
We read the scanned factory documents held in the Collins Collectors Association archive. None of them mentions audio oscillation in a 75S receiver.
Document | Date | What it covers | C29 or oscillation? |
|---|---|---|---|
10 May 1962 | To minimise background hum level. Adds C29, L16, C33 and C34, and changes R25 and R54. | Adds C29 for hum. States the change is for hum only. No mention of oscillation. | |
27 Feb 1967, revised twice | Strong-signal handling, mechanical filters, CW operation and headphone audio. | C29 not touched in the 90 modification steps. | |
23 Dec 1960 | 75S-1/2: fixed bias on the 6BF5 reduced by changing R54 from 27K to 22K. | Confirms R54 sets the 6BF5 bias. No mention of oscillation. | |
3 Aug 1960 | 75S-1/2: two minor component notes. | Neither. |
Collins did chase audio instability in the KWM-2 in the same period, so the silence on the 75S line looks like a fault that was never reported, not one that was ignored. A good C29 hides it completely.
The stock circuit
The key is that C29 does not sit on a simple bias supply. In the 75S-3B, and in any 75S-3 or 75S-3A with the Service Bulletin 2 wiring, one series string of resistors runs from the −65 V bias rail to ground and supplies two different stages.
- C29 node, about −12 V: this is the fixed bias for the 6BF5 output tube, V10. Its grid resistor R35 returns here.
- Detector grid, about −2.4 V: R54 continues from the C29 node to the grid of the product detector, V8A, with R25 to ground.
- Check against the schematic: the string calculates to −11.8 V and −2.4 V. Collins marks −11 V on the V10 grid and −2.7 V on V8A pin 9.
So the node that C29 bypasses is shared. It is the grid return of the last audio stage and, through R54, the bias feed of the first one.
Why an open C29 causes oscillation
With C29 working, the shared node is held at AC ground and the two stages cannot see each other. With C29 open, the node floats at audio frequencies and a loop closes around the whole audio chain.
- The path: audio at the V10 grid passes through R35 to the C29 node, then through R54 and R25 to the product detector grid. V8A amplifies it, it passes through the AF gain control and V9, and it arrives back at the V10 grid.
- The polarity: V8A and V9 each invert the signal. Two inversions make the feedback positive.
- The gain: with C29 open, about 3 percent of the V10 grid signal reaches the detector grid. We estimate the two stages have 55 to 60 dB of gain at full volume, so the loop gain exceeds unity and the receiver oscillates.
- The volume setting: the AF gain control is inside the loop. That is why the oscillation starts only when the control is well advanced.
This also answers Dick’s puzzle. In a 75S-1 or an early 75S-3 without Service Bulletin 2, R54 goes straight to ground and the detector grid has its own separate return. There is no path between the two stages, so those receivers are stable with or without a C29. Only receivers with the later wiring depend on C29 for stability.
Bench results from K5IU, 15 November 2026
Dick Weber ran four tests on a 75S-3A, which has the same bias circuit as the 75S-3B. The measurements below are his, taken with an HP 400E AC voltmeter across the speaker terminals, a 50 ohm termination on the antenna input and the audio gain at maximum.
Test | Condition | Result |
|---|---|---|
Oscillation | 25 µF C29 fitted, volume at maximum | No oscillation |
Oscillation | C29 removed | Oscillation from about 75 percent of full volume |
C29 value, RF gain at minimum | 10 µF | 12.1 mV rms |
C29 value, RF gain at minimum | 25 µF | 12.3 mV rms |
C29 value, RF gain at minimum | 100 µF | 12.2 mV rms |
Parallel bypass, RF gain at minimum | 25 µF plus 0.1 µF film | 12.1 mV rms, no improvement |
Parallel bypass, RF gain at minimum | 25 µF plus 0.01 µF ceramic disc | 12.2 mV rms, no improvement |
RF gain at maximum | 25 µF C29 | 65 mV rms, about 14.7 dB higher |
- The oscillation is real and repeatable: it appeared on a second receiver as soon as C29 was removed. Dick reports that this verifies the prediction made here.
- A 10 µF C29 is enough: a larger value made no measurable difference to the noise.
- Bypass capacitors do nothing measurable: a scope on C29 showed no 455 kHz, and neither a film nor a ceramic capacitor in parallel changed the reading. An earlier version of this article suggested ceramic bypasses and a ferrite bead. We have withdrawn both.
- The bias network is not the noise source: with the RF gain turned up, the noise is dominated by the stages ahead of the product detector. Dick concludes that no change to the bias network will lower it, and we agree.
A proposed change for fault tolerance
The stock design depends on one electrolytic capacitor to keep the audio chain stable. When it dries out, the receiver oscillates. The change below adds a second filter section between the shared node and the detector grid, so that no single capacitor failure can close the loop. In the light of Dick’s measurements it should not be expected to lower noise or hum.
- Split R54: replace the 22K resistor with two 11K resistors in series.
- Bypass the midpoint: fit a 10 µF 63 V film capacitor from the junction of the two resistors to ground. Use 22 µF if space allows, for more margin at low frequencies.
- Replace C29: a film capacitor will not dry out. Dick’s data shows 10 µF is sufficient; the 22 µF shown is optional.
Stock, good C29 | Modified | |
|---|---|---|
If C29 fails open | audio oscillates at high volume (measured) | stays stable, by our estimate |
If the new capacitor fails open | not applicable | reverts to stock behaviour |
Audio noise and hum | baseline | no measurable change expected |
6BF5 and detector bias voltages | −11.8 V and −2.4 V | unchanged |
By our calculation the stability margin with C29 open is smallest at low frequencies, about 10 dB at 100 Hz with a 10 µF midpoint capacitor. These figures are estimates and have not been measured. For most owners, simply replacing an old C29 with a fresh 10 µF part is all that is needed. The split-R54 change is for those who want the receiver to stay stable even if that part fails again.
Build notes
- Check your values first: the October 1963 schematic shows R54 as 22K and R53 as 56K. Other sources show 15K and 47K. If your receiver has a 15K R54, use two 7.5K resistors.
- Polarity: film capacitors are non-polar. If you fit an electrolytic in either position, the positive end goes to ground, because the bias string is negative.
- Leave C8 alone: its node also feeds the mode switch, and we have not traced where that goes.
- Measure before and after: the C29 node should read about −12 V and V8A pin 9 about −2.4 to −2.7 V both before and after the change.
- Reversible: no holes are needed and the original parts can go back in.
Where the remaining noise comes from
Dick’s last test raises an obvious question: can the noise from the stages ahead of the product detector be reduced? Mostly it does not need to be. With a 50 ohm termination and both gain controls at maximum, the receiver is amplifying thermal noise together with the noise of its own RF amplifier and mixers. That floor cannot be removed. What matters on the air is whether noise from the antenna is stronger.
On HF it almost always is. Adam Farson, VA7OJ/AB4OJ, summarising the ITU data, gives the quietest external noise between 4 and 30 MHz as roughly 52 minus 23 times the logarithm of the frequency in megahertz, in decibels above thermal noise. That is about 32 dB at 7.1 MHz and about 19 dB at 28 MHz. He puts the receiver noise figure needed at a quiet site at the top of HF at under 16 dB, which a healthy 75S-3B should meet comfortably.
Approach | Effect | Our view |
|---|---|---|
Tested low-noise RF amplifier tube and full alignment | Restores the factory noise figure | Do this first |
Set the IF gain control by the factory procedure | Removes excess gain that only raises noise at full volume | Do this first |
Audio low-pass filter or external DSP noise reduction | Cuts hiss above the speech band | Audible, and needs no change to the receiver |
L16 and C34 at the detector cathode | Dick Weber’s 2009 article in The Signal replaced them with an API Delevan 1641R-104K choke and a 1000 pF X7R capacitor to reduce product detector hash | Worth doing if hash is the complaint |
External low-noise preamplifier | Lowers the system noise figure | Only worthwhile on 15 and 10 metres at a quiet site |
Replacing the tube mixers with diode mixers | Lower mixer noise | A redesign. We would not do it to this receiver |
For anyone who wants to try a preamplifier on 10 metres, two current Mini-Circuits parts suit the job. The PHA-13LN+ is a surface-mount device covering 1 MHz to 1 GHz with about 20 dB of gain, very high intercept and a 3 to 5 volt supply. Its noise figure is specified at 500 MHz; the datasheet for its sibling, the PHA-13HLN+, lists 1.2 dB at 20 MHz. The ZFL-500LN+ is a connectorised module covering 0.1 to 500 MHz with a 2.9 dB noise figure and 24 dB of gain, but a much lower intercept, so it suits bench tests better than on-air use.
Either has too much gain for a tube front end. Follow it with a 10 dB attenuator, put a band-pass filter ahead of it, and bypass it on transmit if the receiver shares an antenna relay with a transmitter. We have not built or tested a preamplifier with this receiver.
What is new here, and what is not
- Not new: C29 itself. Collins added it in 1962 for hum, and Dick Weber has described fitting it to earlier receivers in his articles in the CCA magazine, The Signal.
- Not new: the observation that a failed C29 causes oscillation. That is Dick Weber’s finding, reported in November 2026 and now repeated by him on a second receiver.
- Possibly new: the explanation of the positive feedback loop through the shared bias node, and the split-R54 modification. We did not find either in the factory bulletins or in the published material we searched. We have not searched every back issue of The Signal or the full reflector archive, so we cannot say they have never been described.
- The loop gain figures are calculated, not measured.
- The R35 return to the C29 node was confirmed from the marked voltages, not by tracing the wiring in a receiver.
- The AM-position test has not been run.
- The split-R54 modification has not yet been built, so its stability with C29 open is an estimate.
Sources
- Dick Weber, K5IU, C29 Testing, 15 November 2026. Bench measurements quoted with credit.
- Dick Weber, K5IU, message to the CCA reflector, 13 November 2026.
- Dick Weber, K5IU, The Signal, Issue 53, 1st Quarter 2009
- Collins Collectors Association: Service Bulletins and Service Information Letters
- Collins Collectors Association: Amateur Service Agency Bulletins
- 75S-3 Service Bulletin 2, 10 May 1962
- 75S-3B/3C Service Bulletin 1, 27 February 1967
- Amateur Service Agency Bulletin 1004, 3 August 1960
- Amateur Service Agency Bulletin 1005, 23 December 1960
- Collins 75S-3B full-size schematic, 15 October 1963.
- Adam Farson, VA7OJ/AB4OJ, Antenna and Receiver Noise Figure
- Mini-Circuits PHA-13LN+ datasheet
- Mini-Circuits PHA-13HLN+ datasheet
- Mini-Circuits ZFL-500LN+ datasheet
Acknowledgement: the fault report, the observation about earlier receivers and all of the bench measurements are Dick Weber’s. The analysis and the proposed modification are ours, and any errors in them are ours too.