Skip to content
TALAS

Mains hum in a condenser microphone: the body is part of the circuit

Talas One · workshop notes · 21 November 2026

A DIY condenser microphone that hums at 50 or 60 Hz is one of the most common questions on audio forums, and the answers usually stop at “check your grounding”. Over seven months and two prototypes of the Talas One, an open-hardware dual-diaphragm microphone, we measured the hum, its cause and its cure. The short version: the metal body is not a box around the circuit, it is part of it.

In short. The input of a condenser microphone is a 1 GΩ node: it picks up the mains field like an antenna. The cure is a closed metal body, solidly and permanently tied to XLR pin 1. On our prototypes, closing the body cut the hum by about 60 dB; making its parts touch properly cut another 13–41 dB; and a body bond that came and went made the capsule’s bias voltage itself drift. If touching the body changes the sound, you have found your problem.

1Why 1 GΩ makes an antenna

The diaphragm of a condenser capsule feeds the preamp through a very high resistance, 1 GΩ on the Talas One. An ambient 50 Hz field of a few volts per metre couples capacitively about 10 pA into a nearby high-impedance node. Across 1 GΩ, 10 pA is 10 mV: the same order as the capsule’s wanted signal. A dynamic microphone, at 50 Ω, turns the same current into 0.5 nV, far below its noise. Nothing about the preamp or the capsule needs to change; only the shielding does.

2April: the buzzer that speaks

On the first prototype, boards on the bench and out of their body, more than 90 % of every recording’s energy was below 100 Hz: 50 Hz and its harmonics. With the preamp also misbehaving, the hum drove the op-amp into clipping, and the voice rode on it as a modulation: a buzzer that seemed to speak, mostly on the sibilants. A digital 50 Hz notch found a clean voice underneath about 50 dB of hum.

We tried two circuit changes, a capacitor on the bias node and a bypass of its filter: each helped a little. Then we mounted the boards inside the metal body and closed it. That fixed it, and we removed both changes: the cleanest result was the unmodified circuit in a closed metal body.

3June: when the bond comes and goes

Two months later the microphone’s sensitivity started drifting on its own, by 4 to 7 dB per minute, with collapses of about 30 dB for tens of seconds. Resting a hand on the body made it instantly stable, to 0.01 dB; removing the hand brought the drift back.

A warming hand that stabilises a circuit rules out a leaky part: heat would make a leak worse. The hand was grounding the body through the person. A continuity check showed why: tube and grille were connected to the XLR shell, pin 1 to the board ground, but the link between those two groups was a marginal mechanical contact, open on one reading and closed on the next.

The consequence went beyond hum. The capsule’s bias voltage comes from a charge pump that can only deliver microamps, through a 2 MΩ filter. Unshielded, that node collects charge from the room and the bias itself wanders, and the sensitivity with it. After soldering a permanent bond, 11 of 15 consecutive one-minute windows were stable to 0.01 dB, and all of them from the eighth on.

A trick worth stealing. A 1 GΩ node cannot be probed: any meter loads it. But the capsule’s sensitivity is proportional to its bias. Play a steady 1 kHz tone, record the microphone next to a stable reference microphone, and follow the level difference over time: every change in it is a change in bias, read acoustically, without touching anything.

4November: paint, and two capacitors that look alike

On the third prototype, the P3, the hum came back in a milder form. It followed one channel rather than the interface input, a hand on the body raised the 50 Hz by 6–8 dB, and simply re-plugging the cables raised 150 and 200 Hz by 15–20 dB: the body and the cable shields were not a solid ground. Two faults were found:

  1. The body’s parts touched badly. They are painted, and paint insulates. Re-assembled and screwed firmly, without changing any part, the 50–200 Hz hum fell by 13–41 dB, the hand on the body stopped making any difference, and an interference floor between the mains lines disappeared.
  2. Three wrong RF capacitors. The design puts 22 nF from each XLR signal pin to ground, against radio frequencies. Three of the four soldered were 220 pF: identical-looking parts. In the audio band it changes nothing, but at 1 MHz 220 pF is about 720 Ω against 7 Ω for 22 nF, and radio picked up by the cable, rectified in the op-amps, comes out as a hum that changes with the surroundings. With 22 nF on all four pins, the 100–200 Hz harmonics fell by another 8–20 dB.

At 100 Hz, channel A went from −103 to −141 dBFS at minimum gain, and the noise of the electronics measured with a dummy capsule from 23 to about 5 dB(A): see measuring self-noise without an anechoic chamber.

5A checklist

6Sources

Every figure here comes from the Talas One repository: grounding and bias stability (June), prototype status (April), P3 measurements and the bring-up journal (November).

The Talas One’s circuit is the one Jules Ryckebusch (DJJules) published in 2021 (“True Condenser OPA Mics”, CC BY 4.0). The boards, the measurements and this text are Talas’s work, under CC BY-SA 4.0.

← The Talas One