White paper: Logitech Squeezebox Boom audio design

Typical desktop speaker systems will be a 2.0, or occasionally a 2.1 system. Very few desktop speaker systems use true tweeters, and thus the high end will either be nonexistent or it will 'beam' with much more energy coming from the front of the system than off axis. This is a fundamental property of sound propagation. For the best quality sound, it's critical that loudspeakers be as omnidirectional as possible.¹ The result is more unified and balanced sound than can be achieved with other architectures.

Figure 7: Comparison of different desktop speaker architectures.

In the figure above (Figure 7), Section A shows a typical 2.0 desktop speaker system with single drivers for each channel, and two power amplifiers. This design will generally compromise low and high-frequency response.

Section B shows a desktop 2.1 system with stereo satellite speakers plus a subwoofer. This type of system will make up the low-end and sound full, but it will lack a flat frequency response to 20 kHz.

Section C shows the design of Squeezebox Boom. Four separate speakers, four separate amplifiers, and six DACs all digitally controlled with a high powered DSP provide the ultimate in acoustic signal integrity and can produce great sound through the entire audio spectrum. Without a subwoofer, the Squeezebox Boom goes from a -3 dB response at 50 Hz (at low volume settings) to about 85 Hz (at high volume settings), all the way to 20 kHz. With the addition of a subwoofer, the entire audio band, from 20 Hz to 20 kHz, is covered.

The Squeezebox Boom DSP
The Squeezebox Boom uses a 48-bit x 24-bit, 135-MIPS (million instructions per second) DSP core. With such significant DSP horsepower there are many ways to improve the sound quality; good crossovers are just the beginning.

One may wonder why we need a 48-bit data path, providing 289 dB of dynamic range, when 16 bits and 24 bits (96 dB to 144 dB) is good enough for most sound systems. The answer is that when performing mathematical operations (e.g., DSP) on a signal, the signal is often multiplied, divided, and added to many times. With a 16-bit data signal, any 16-bit signal processing will inevitably result in loss of signal fidelity, either by causing overflow, saturation, or extra quantization noise. Quantization noise, saturation and overflow are undesirable.

In order to perform DSP effectively, the DSP processing must have significantly more precision than the signal being operated on. Of course, a 16-bit processor can perform 32-bit, 48-bit, or higher precision math, but with corresponding CPU performance penalties. The 48 x 24-bit data path of the Boom DSP processor allows us to simply design audio processing algorithms, and minimize excess digital noise into the system " we can easily maintain greater than 100 dB of dynamic range in the digital domain and thus guarantee we don't degrade the original signal beyond the noise floor of the DACs. The large bit-depth of the DSP we chose makes DSP programming relatively straightforward.

Figure 8: Block diagram of the Squeezebox Boom DSP flow (with other components for reference).

¹A Multiple Regression Model for Predicting Loudspeaker Preference Using Objective Measurements, Sean E. Olive, AES Convention 116, paper numbers 6113 and 6190. http://aes.org/e-lib/browse.cfm?elib=12847