AD1878/AD1879

REV. 0

–6–

The AD1878/AD1879’s patented

∑∆ architectures “shape” the

quantization noise-transfer function in a nonuniform manner.

Through careful design, this transfer function can be specified to

high-pass filter the quantization noise out of the audio band into

higher frequency regions. See Figure 27. The Analog Devices’

AD1878/AD1879 also incorporates feedback resonators from

the third integrator’s output to the second integrator’s input and

from the fifth integrator’s output to the fourth integrators’ input.

These resonators do not affect the signal transfer function but

allow flexible placement of zeros in the noise transfer function.

For the AD1878/AD1879, these zeros were placed near the high

frequency end of the audio passband, reducing the quantization

noise in a region where it otherwise would have been increasing.

Oversampling by 64 simplifies the implementation of a high per-

formance audio analog-to-digital conversion system. Antialias

requirements are minimal; a single pole of filtering will usually

A fifth-order architecture was chosen both to strongly shape the

noise out of the audio band and to help break up the idle tones

produced in all

∑∆ architectures. These architectures have a ten-

dency to generate periodic patterns with a constant dc input, a

response that looks like a tone in the frequency domain. These

idle tones have a direct frequency dependence on the input dc

offset and indirect dependence on temperature and time as it

affects dc offset. The human ear operates effectively like a spec-

trum analyzer and can be sensitive to tones below the integrated

noise floor, depending on frequency and level. The AD1878/

AD1879 suppresses idle tones typically 110 dB or better below

full-scale input levels.

Previously it was thought that higher-order modulators could

not be designed to be globally stable. However, the AD1878/

AD1879’s modulator was designed, simulated, and exhaustively

tested to remain stable for any input within a wide tolerance of

its rated input range. The AD1878/AD1879 was designed to

reset itself should it ever be overdriven and go unstable. It will

reset itself within 5

µs at a 48 kHz sampling frequency. Any such

reset events will be invisible to the user since overdriving the in-

puts will produce a “clipped” waveform at the output.

The AD1878/AD1879 modulator architecture has been imple-

mented using switched-capacitors. A systems benefit is that ex-

ternal sample-and-hold amplifiers are unnecessary since the

capacitors perform the sample-and-hold function Coefficient

weights are created out of varying capacitor sizes. The dominant

noise source in this design is kT/C noise, and the input capaci-

tors are accordingly very large to achieve the AD1878/AD1879’s

performance levels. (Each 6 dB improvement in dynamic range

requires a quadrupling of input capacitor size, as well as an

increase in size of the op amps to drive them.) This AD1878/

AD1879 thermal noise has been controlled to properly dither the

input to an 18-bit level. (Note that 16-bit results from either the

AD1878 or AD1879 will be underdithered.)

With capacitors of adequate size and op amps of adequate drive,

a well-designed switched-capacitor modulator will be relatively

insensitive to jitter on the sampling clock. The key issue is

whether the capacitors have had sufficient time to charge or

discharge during the clock period. A properly designed switched

capacitor modulator should be no more sensitive to clock jitter

than are traditional nonoversampled ADCs. This contrasts with

continuous-time modulators, which are very sensitive to the

exact location of sampling clock edges.

See Figures 20–23 for illustrations of the AD1878/AD1879’s

typical analog performance resulting from this design. Signal-

to-noise+distortion is shown under a range of conditions. Note

the very good linearity performance of the AD1878/AD1879 as

a consequence of its single-bit

∑∆ architecture in Figure 24.

The common-mode rejection (Figure 25) graph illustrates the

benefits of the AD1878/AD1879’s differential architecture. The

excellent channel separation shown in Figure 26 is the result of

careful chip design and layout. The relatively small change in

gain over temperature (Figure 31) results from a robust refer-

ence design.

The output of the AD1878/AD1879 modulators is a stereo

bitstream at 64

× F

analysis of these bits would show that they contain a high qual-

ity replica of the input in the audio band and an enormous

amount of quantization noise at higher frequencies. The input

signal can be recreated directly if these bits are fed into a prop-

erly designed analog low-pass filter.

Digital Filter Characteristics

The digital decimator accepts the modulators’ stereo bitstream

and simultaneously performs two operations on it. First, the

decimator low-pass filters the quantization noise that the modu-

lator shaped to high frequencies and filters any other out-of-

audio-band input signals. Second, it reduces the data rate to an

reduced to stereo 16-/18-bit words at 48 kHz (or other desired

ponents of the bitstream, and analog signals in the stopband are

attenuated by at least 115 dB.

The AD1878/AD1879 decimator implements a symmetric Finite

Impulse Response (FIR) filter, resulting in its linear phase re-

sponse. This filter achieves a narrow transition band (0.0923

×

ripple (< 0.001 dB). The narrow transition band allows the

as 44.1 kHz. The stopband attenuation is sufficient to eliminate

modulator quantization noise from affecting the output. Low

passband ripple prevents the digital filter from coloring the

audio signal. For this level of performance, 4095 22-bit coeffic-

ients (taps) were required in each channel of this filter. The

AD1878/AD1879’s decimator employs a proprietary single-

stage, multiplier-free structure developed in conjunction with

Ensoniq Corporation. See Figures 28 and 29 for the digital

filter’s characteristics.

The output from the decimator is available as a single serial

output, multiplexed between left and right channels.

Note that the digital filter itself is operating at 64

× F

consequence, Nyquist images of the passband, transition band,

and stopband will be repeated in the frequency spectrum at

multiples of 64

× F

115 dB across the frequency spectrum except for a window

±0.5458 × F

× F

signals, clock noise, or digital noise in these frequency windows

will not be attenuated to the full 115 dB. If the high frequency

signals or noise appear within the passband images within these

windows, they will not be digitally attenuated at all.