Power requirements, moving mass and audio sampling and processing.

Hearing aids run on tiny voltages (typically ~1.4 Volts). They are designed to amplify speech (a small frequency range) at moderate volumes. An IEM is designed to handle the massive energy of a live drum kit without distortion. To do this, the internal amplifier needs Headroom. It likely steps up the battery voltage significantly (internally converting to higher voltage rails) to ensure that when a snare drum hits 120dB, the amplifier has enough electrical height to reproduce that spike without clipping.

Hearing aids use microscopic balanced armature receivers that require almost zero power to move because they are only moving a tiny amount of air near your eardrum. IEMs use dual-driver miniature subwoofers and tweeters that are physically larger and heavier. It takes significantly more electrical current to push these drivers back and forth.

Hearing aids often use aggressive battery-saving tricks, such as lowering the sampling rate or "sleeping" processes when silence is detected. The processor of an IEM is running wide open 100% of the time. It is constantly digitizing the world at a high sampling rate to ensure zero latency. If it tried to save battery by "sleeping" between notes, you would hear a delay (latency), which would make it impossible to play in time.

Let’s try some actual numbers.

A fancy passive IEM quotes:

> Sensitivity: 114 dB-SPL (@ 0.1V)

That’s on the high side for an IEM. Let’s be conservative and say it’s actually 100dB SPL at 1mW (which is on the low side) and that the user has hearing loss and the IEM is actually outputting 90dB continuously (I’m not an expert but this seems high. Certainly I would not set an active IEM anywhere near that loud, even at a concert). That’s 0.1mW of electrical power to each IEM, for 0.2mW total. This part could run for days on a small battery pack.

A modern amplifier might as well be 100% efficient, although that body pack could easily be some wildly efficient Class A design.

Let’s suppose the DSP is processing 96kSPS (might as well minimize latency and the need for a complex antialiasing setup and let’s assume it’s using a duper-high-fidelity FIR filter that’s a whole 100ms long, i.e. 9600 taps, and that the FIR implementation is pure brute force, so there are 9600 times 96k FMAs per second. That’s about 2 billion FMAs per second for both ears. (Again, this is a ridiculous way to do this.) The fancy nonlinear compression stuff will be negligible in comparison.

From some quick Googling, you could easily spend a whole watt on the DSP. And there’s where all that battery charge goes :)

I bet that someone who really cared about optimizing this could put some software engineering into getting DSP power down below 50mW. Either use an FFT to optimize the filter or use a much lower complexity filter bank.