Work in Progress

Singing Capacitors aka The Piezoelectric Effect

Close up photo of a PCB half populated

Just over a year ago I was experimenting with the LM2703 DC/DC converter. At the time I had tried to build a 5 V to 12 V boost converter using this chip on a breadboard without much success.

Photo of a 5V to 12V boost converter using an LM2703 laid out a breadboard

Fast forward to today and I’m now using my reflow oven to produce a custom PCB for the LM2704, which has a higher peak current limit than the LM2703, but in every other way the same.

Typical application circuit diagram for 12V regulator using the LM2704

It took me a while to design the PCB layout, this was my first design using surface mount components, but I eventually ended up with the board below.

Close up photo showing the LM2704 circuit solder to a PCB

C1 & C2 are the input and output decoupling capacitors, both 4.7 μF. R15 & R16 form a feedback potential divider to set the output to 12V and F1 is a resettable fuse I’ve added on the output rail. The switch on the left can be ignored and is not part of this circuit. This board is part of a Raspberry Pi HAT so the input voltage is 5V.

Powering up the circuit one of the things I noticed was a high pitched noise coming from the board. I connected my scope to the voltage rails and you can see there is noise on both the input and output rails, in green and red respectively.

Scope trace showing noise on the Vin and Vout rails

The ripple on the output is around 1V and 0.2V on the input. The LM2704 can be put into a sleep state and when I do this the output drops to 5V (i.e. the same as the input) and the ripple on the input disappears so this noise is being caused by my circuit.

I’ve got an Android phone and installed a spectrum analyser app. The first snapshot is of the background noise of the room when the board is powered down. The dB readings are not calibrated and I just held the bottom of the phone about 1 cm away from the board.

SPL vs frequency plot showing background audio noise

The second is with the board powered up and there are several new peaks in the spectrum with the largest occurring at around 16 kHz.

SPL vs frequency plot showing capacitor noise

My initial thoughts were that the noise was coming from the inductors and this is a common source of noise in power supplies however putting my finger on the inductor did not change the pitch of the sound so it wasn’t the source.

I used a plastic ruler and placed a corner on each component to amplify any sounds and found that the noise was coming from the output decoupling capacitor C2.

Photo of a plastic ruler being used to find the source of the audible noise

Capacitor C2 is a 0805 sized 4.7 μF multi-layer ceramic capacitor (MLCC). MLCCs are susceptible to the piezoelectric effect where changes in the voltage across the capacitor cause its structure to move or distort and if this happens repeatedly in the audible frequency range can be heard and is known as ‘singing’. This isn’t just a one-way process either – MLCCs are not often used in high-end audio equipment as sound vibrations transmitted through the capacitor affects the electrical signals going through them causing distortions – this is essentially how a piezoelectric sensor works.

I’m going to see if I can eliminate this noise through changes to the:

  • capacitance – increasing the value of the capcitors to say 47 μF to reduce the voltage ripple (and hence noise) on the input and output rails.
  • size – increasing the physical size of the capcitor from 0805 to 1210 toreduce the physical movment within the capacitor and reduce the noise.
  • type – selecting another type of capacitor (i.e. tantalum, aluminium etc.) which is not susceptible to the piezoelectric effect.
  • layout – changing the design and/or layout of the circuit.