A True One-Bit Power D/A Converter

A method is presented to convert 1-bit digital audio signals into an analogue signal with sufficient current and voltage to drive loudspeakers. For this goal a novel non-PWM class D power stage is constructed that performs this function with very low distortion and very high efficiency, without the use of feedback or other analogue processing. Results of the prototype development are detailed. INTRODUCTION A continuous theme in audio amplification the past decade has been the desire to merge class D power stages with digital modulators, forming a “digital amplifier” or a “power DAC”. While the authors believe this desire to be subjective in origin, the question is accepted “as is” without further elaboration. The most obvious setup is to generate a PWM (-alike) signal through DSP or other digital means. Another approach often looked into is directly to use 1-bit deltasigma signals such as what can be found in many digitalto-analogue conversion systems. Although the latter is fraught with difficulties normally considered insoluble, it has resurged due to the introduction of a 1-bit deltasigma based release format called DSD. The present method is capable of surmounting the audio quality and efficiency related problems of 1-bit power conversion to such an extent that it consistently outperforms PWM based methods. PROBLEMS FOUND WITH 1-BIT SIGNALS AND REAL POWER STAGES For practicality we shall use the term “1-bit” whenever a 1-bit noiseshaped signal at typically 2.8224MHz sampling rate is meant, such as DSD. Switching Frequency Spectral and mathematical analysis of a 1-bit signal shows that a strong discrete tone is present which is at idle exactly half the sampling rate (corresponding to frequent and long ...101010... patterns) and which decreases linearly with modulation index. This means that the effective repetition rate is around 1.4MHz, improving down to below 1MHz for strong modulation. In class D power circuits such high frequencies are to be avoided in order to keep switching losses acceptable and to reduce the contribution of timing errors to output distortion. Switching Losses In one-quadrant power systems such as DC/DC converters, switching losses can generally be minimized by the use of zerovoltage and/or zero-current switching techniques. These recover energy stored in parasitic circuit elements. In four-quadrant mode these energies can no longer always be recovered and must be dissipated inside the active elements (MOSFETs). Additionally, the internal diodes of the power MOSFETs will sometimes be forward biased, incurring very significant losses during forced turnoff. These issues dictate a need for minimizing switching rates. PUTZEYS AND DE SAINT MOULIN ONE-BIT POWER D/A CONVERTER AES 112 CONVENTION, MUNICH, GERMANY, 2002 MAY 10–13 2 Distortion When the output stage has settled, it has firm control over the output voltage. While transitioning this is not the case and the actual voltage waveform found will depend upon the output current delivered. Attempts to correct for these errors in a digital control circuit without feedback must inexorably fail as the load impedance and hence the output current is unknown. The error committed under large signal conditions by an open-loop class D power stage due to a timing uncertainty td (roughly equal to dead time) is approximately a square wave of amplitude td·fsw·Vcc i.e. the supply voltage times the fraction of time spent in transition. The harmonic content of a square wave is 0.41 so the error can be estimated as