Some months ago, a colleague and I ran up against an interesting obstacle in the development of a whole-home music system amplifier. This particular product contained three stereo Class D amplifiers. We were using integrated IC devices from a leading manufacturer that had plenty of experience in this kind of application, so my colleague, Jon O’Donnell, and I were very comfortable following the company’s reference design pretty much as documented.
At a point when about 50 pre-production units had been produced, someone noticed acoustic noise from the board when he happened to have no speakers connected (i.e., he was using line outputs) and the volume was cranked. The acoustic noise was loud enough to be audible in a moderately quiet room when the board was mounted in a rack unit enclosure. It had a particularly annoying sound and was unquestionably following the music signal, but there was no bass at all.
In listening closely to it, I perceived sounds with a relatively narrowband acoustic spectrum somewhere in the range of 2kHz to 5kHz. The presence of an 8-ohm dummy load didn’t seem to affect the “singing” from the board much. Pure tones could not be heard as much as clicks, buzzes, and pops. Since we had two installed cases where a user could be in a quiet room with amp channels operating, the acoustic noise was obviously unacceptable, so Jon set out to track down the cause with me watching and contributing ideas from over the cube wall.
We knew the offending circuitry was confined to the output filters of our Class D amplifier, because it could be reproduced with a pure digital music source. The relevant part of these filters look like this:
Schematic of the Class-D output filter (between the IC and the speakers)
The left and right differential PWM outputs of the amplifier ICs are on the left side. Each signal pair drove mutually coupled inductors in series with 0.68 uF capacitors to ground. This formed a pair of two-pole LC filters that provided significant attenuation to the 384kHz PWM frequency and its harmonics, and passed the audio frequency energy to the connected speakers. It’s a well established filter recipe for a Class D amp. Components between the outputs shown and the bridge-tied speaker terminals form a high-frequency EMI filter.
When Jon first approached me with this anomaly, I suspected the inductors. I figured those ferrite-core inductors were somehow “rattling” with induced force from the audio-frequency currents. Jon came up with the ingenious idea of using a ¼” electret condenser measurement microphone as a near-field acoustic probe and soon proved the inductors innocent. The sound was emanating from the 1210 package multilayer ceramic capacitors (C112 through C115 in Figure 1).
It is well known that some multilayer ceramic capacitors have piezoelectric qualities. They can generate a small signal voltage when subjected to acoustic or vibrational energy. For this reason, care must be taken when using MLCCs in low-level analog signal paths, because this “microphonic” behavior can contaminate the audio signal. In this case, a complementary effect was the problem; the signal voltage was creating vibrational energy.
According to an online technical article from Kemet Electronics, “The complementary effect also occurs, in that electrical stimulation of ferroelectric compounds can result in mechanical deformation. In circuits which operate at acoustic frequencies, the capacitors will tend to respond and may emit acoustic noise.” The audio-modulated 384kHz pulse currents shunted to ground in our first stage filters were enough to excite mechanical deformation across a range of body-resonant frequencies -- in effect, turning the capacitors into very small speakers.
The referenced Kemet article details how this unwanted effect can be minimized with ceramic capacitors, but our solution was to replace the ceramic capacitors with polarized surface-mount tantalum electrolytic devices. Problem solved: The amplifier no longer tried to sing along with the connected speakers.