Several aspects of audio converter design are generally not well understood but have a dramatic impact on performance, both measurable and audible. One of these is interfacing to the sampling networks of highly oversampled multibit delta-sigma A/D converters, and it is probably the least understood topic in A/D systems design.
The requirements to effectively interface to the high-speed, switched-capacitor sampling network, minimize distortion and provide the appropriate anti-alias filtering are often conflicting. This interface remains one of the “magical” areas of audio design.
Sampling network operation
The operation of the sampling network consists of two phases. During the capture phase, the sampling capacitor is charged to the voltage presented at the analog inputs, and during the conversion phase, this charge is transferred to an integrator for conversion. The transient loading created by this switched-capacitor causes the buffer amplifier to operate in a nonlinear mode and exhibit ringing and/or oscillation when driving these circuits directly.
A required buffer topology provides isolation from this loading. In addition to the linear sampling currents, there are non-linear currents that can be a source of distortion. The magnitude of these currents varies among architectures, but source resistance recommendations are generally less than 100 W, and in many cases much less, to meet distortion specifications.
It is also interesting to consider the anti-alias filter requirements for highly oversampled audio A/D converters. The oversampling frequency for a 48-kHz sample rate is typically 6.144 MHz, and the Nyquist frequency is 3.072 MHz. However, what is not obvious is that the digital decimation filter will remove the majority of the aliased components and the frequencies that will alias exist at N times the oversampling frequency plus or minus the digital filter passband. With the limited-frequency bandwidth of audio signal sources, the most critical requirement for anti-aliasing in audio is simply to avoid aliasing noise that will degrade dynamic range, and a single-pole RC filter with a corner frequency approximately a decade below the sampling frequency will suffice.
The circuitry that converter suppliers have recommended for many years simply inserts an RC low-pass filter between the amplifier and the switching network, as shown in part (a) of the figure below. The addition of series resistor (R1) and load capacitor (CL) addresses several of the requirements. In this configuration, the instantaneous current for the sampling capacitor is provided by CL while the average current is supplied by the op amp. The resistor isolates the amplifier from the transient switching currents and also provides the stability required to drive the large capacitive load presented by CL. In addition, the combination of R1 and CL provides the required anti-alias filter.
However, with the relatively recent performance increases of A/D converters, the compromises of this topology have become a limiting factor. The required capacitance of CL has increased while the upper limit for R1 has decreased to meet the dynamic range and distortion capabilities of high-performance converters. These requirements, while maintaining amplifier stability, are conflicting. Adoption of an improved approach is required to match the capabilities of modern high-performance converters.
An ideal buffer:
— Provides zero-ohm source impedance over the audio bandwidth to minimize distortion
— Offers sufficiently high output impedance at the sample frequency to provide isolation
— Has a single-pole anti-alias filter response
— Can drive the large capacitance of CL while maintaining stability.
The topology shown in part (b) of the figure was originally developed to stabilize an amplifier driving a large capacitive load. This circuit works very well for the intended purpose and has attributes that make it nearly ideal for sampling networks. Within the audio passband, C1 can be considered an open circuit and R1 is within the feedback loop of the amplifier. In this configuration, the output impedance of the amplifier is R1 divided by the loop gain of the amplifier, which provides a very low source impedance to minimize distortion.
At the oversampling frequency, C1 can be considered a short circuit and R1 is no longer within the feedback loop. The output impedance of the circuit becomes R1, which provides isolation at the sampling frequency. In addition, the component values may be selected to provide the required anti-alias filtering. The compensated buffer has the advantages of negligible source impedance within the audio band, stability for driving large capacitive loads, providing isolation at the oversampling frequencies and including the anti-alias filter. In addition, the circuit requires low resistance values that minimize noise contributions and high input impedance.
Steve Green has more than 25 years' experience in the audio business, including more than 14 years with Crystal Semiconductor and Cirrus Logic as an applications/technical marketing engineer.