So Class D is highly efficient. But what about that business of chopping up the musical waveform? How can transistors that are turning fully on and fully off reproduce a continuous analog waveform? Isn’t that like trying to turn hamburger back into steak?
To understand how a continuously variable analog signal can be represented by a stream of on/off pulses, we need to look a technique called pulse-width modulation (PWM). In pulse-width modulation, the audio signal’s amplitude is encoded in the pulse-stream’s on/off duty cycle—that is, the ratio of the time spent between the on and off states.
The illustration above shows the relationship between an analog waveform and its PWM representation. Full-scale positive is represented as long streams of the “on” state; full-scale negative is represented by long streams of the “off ” state. In the absence of a signal, the pulse stream has a 50/50 duty cycle, alternating evenly between on and off. The audio signal is thus encoded in the pulse widths. The pulse train is remarkably analog-like; you can actually see the sinewave’s shape in the PWM pulses.
In a Class D amplifier, the audio signal is converted into a PWM signal after the input buffer, and the PWM stream drives the output transistors to turn them fully on or fully off (Class D output stages use MOSFETs almost exclusively). The switching frequency is several hundred kilohertz. The transistors in a Class D amplifier must be able to turn on and off very quickly, and precise timing circuitry is required to make the whole thing behave correctly.
The output from the switching transistors is then put through a crucial element of Class D design—the low-pass output filter. The passive low-pass filter smoothes the waveform and removes the modulation (switching) noise, leaving only the original waveform. The reconstruction filter in a PCM digital-audio system (CD player) performs the same function. The output filter in a switching amplifier can be as simple as an inductor in series and a capacitor in parallel with the loudspeaker load.
The output filter’s design is of paramount importance. The filter must remove the switching noise (typically several hundred kilohertz) while not introducing amplitude rolloff or phase shift in the audioband. The steeper the filter, and the closer its cutoff frequency to 20kHz, the greater the audioband phase shift. A gentler filter, or one whose cut-off frequency is far above the audioband, will introduce less audioband phase shift but allow more switching noise to reach the output.
In practice, all Class D amplifiers put out some high-frequency noise at their binding posts. This is why many Class D amplifiers interfere with radio reception; the loudspeaker cables act as an antenna, broadcasting this switching noise into the local vicinity. This is another reason that Class D amplifiers can sound different in different systems; ancillary products vary in their susceptible to pollution by this radiated noise.
The output filter interacts with the loudspeaker in ways that are unpredictable to the designer of the Class D amplifier. The loudspeaker’s impedance magnitude affects the filter’s cut-off frequency (and thus the amount of audioband phase shift), and the loudspeaker’s phase angle (how inductive or capacitive a load the loudspeaker presents to the amplifier—which changes as a function of frequency) interacts with the Class D amplifier’s output filter. In essence, the loudspeaker’s inductance and capacitance become part of the filter, modifying the filter’s characteristics. This is perhaps why Class D amplifiers vary so much in their sound quality when used with different loudspeakers, and might explain the widely disparate views of certain Class D amplifiers (see the review of the NuForce Reference 9 monoblocks and the associated comments, for example).
Back to the top | Previous | Next
Advertisement
All documents notede with file size are in PDF format.
Download Acrobat Reader
