B Y MARK HO L DAWAY • X I GN A L TE CH NO L O G I ES AG

Designing antialias
filters for ADCs
CONTINUOUS-TIME ADCs CAN BENEFIT SIGNAL-CHAIN
DESIGN. AN OVERVIEW OF DISCRETE- AND CONTINU-
OUS-TIME SYSTEMS DETAILS THE DIFFERENCES.

yquist-sampling theory lies at the heart of problem: Once you sample the signal, you have no way of

N
today’s digital-communications systems. It determining which resulting signal components originate
requires that data-conversion systems include from the desired signal band and which ones are aliased
antialiasing input filters. Designers need to errors. Figure 1 shows two alias signals, A, a single tone,
understand the requirements for antialiasing and B, a spectrum, each folding down into the first Nyquist
filters and examine the consequences of filter zone. Note that A originates in Nyquist Zone 4, and B
application. They must also consider the benefits of a new is from Zone 3. Also note that, in a communications ap-
class of ADC that uses a low-power, high-speed, continuous- plication, this folding may allow interference signals to
time-sampling method. These devices claim the ability to completely obscure information-bearing Signal A.
achieve a first Nyquist-zone-sampling capability without the You should bandlimit a signal for digitization to eliminate
aid of external filters. any signal power beyond the frequency range of interest. The
You can reconstruct a time-continuous signal from discrete- design of a suitable antialiasing-filter network may seem fairly
time-sampled data if the original sampling rate is twice that trivial; however, as ADC linearity and performance improve,
of the highest frequency component in the sampled signal. these filters become a significant part of the total system
The Nyquist-sampling theory states that data clocked with design.
a sample rate of fS (sampling frequency) samples/sec can
effectively represent a signal of bandwidth as high as 0.5
fS IDEAL AND PRACTICAL FILTERS
Hz. The Nyquist theory places demands on the sampling Ideal baseband, lowpass antialiasing filters should have a
function, time, and amplitude precision. Sampling signals steep transition band, excellent gain flatness, and low dis-
with signal content greater than a 0.5
fS-Hz bandwidth tortion in the passband—difficult goals to achieve. Further-
cause aliasing, a nonlinear process that results in frequency more, the stopband attenuation should be enough to reduce
shifting. Signal content at frequencies greater than 0.5
fS any residual out-of-band signal power to a level invisible
Hz folds around 0.5
fS Hz—the Nyquist frequency—and to the ADC. You achieve this performance by employing
alias back into the baseband. This aliasing creates a serious stopband attenuation in excess of the dynamic range of the
ADC (Figure 2). Assume that the stopband
NYQUIST extends to infinity. Applications encountering
FREQUENCY high noise levels, especially those with high
levels of interference occurring close to the edge
A A A of the first Nyquist zone, require filters with
SIGNAL aggressive falloff. You achieve this performance
OBSCURED
BY A using high-order filters that typically exhibit
SIGNAL
poor phase performance and result in dispersion
POWER or large group delay. In antialiasing filters, filter-
ing takes place before the time-sampling point,
or quantizer; these filters consequently require
the use of an analog filter. This requirement is
B B B B B B B
unfortunate because you can more easily and
fS 2f S 3f S cost-effectively implement aggressive filters in
NYQUIST ZONE
NO.
1 2 3 4 5 6 the digital domain. High-order analog filters
SAMPLING FREQUENCY (Hz) provide low harmonic distortion and gain flat-
ness to in-band signals. However, the design of
Figure 1 Alias signals A, a single tone, and B, a spectrum, can reside in any these filters is complex because they are too sen-
Nyquist zone if no antialias filter exists in a sampled system, but you can find sitive to gain matching to be practical at more
both in Zone 1, where A now obscures an information-bearing tone. A origi- than a few orders of attenuation magnitude.
nates in Nyquist Zone 4, and B is from Zone 3. Furthermore, any passband harmonic distortion
the filter introduces also produces undesirable

NOVEMBER 23, 2006 | EDN 65

signals in the output spectrum of the ADC.
Insertion loss might also be important when
using passive filters, which increase system
⫺3 dB
noise.
An ideal antialiasing filter features 0-dB DESIRED ALIAS
PASSBAND
unity gain in the passband with little or no GAIN FLATNESS
ATTENUATION

gain variation and a level of alias attenuation ATTENUATION

that matches the theoretical dynamic range of (dB)
PASSBAND STOPBAND
the data-conversion system in use. You derive TRANSITION
BAND
a first approximation of this value from the
theoretical SNR (signal-to-noise ratio) for an
N-bit ADC: SNR⫽6.02⫻N⫹1.76 dB. For
fC f STOP
a 14-bit ADC, this approximation requires
FREQUENCY (Hz)
80- to 86-dB attenuation with an ideal SNR of
approximately 86 dB. Figure 2 Ideal lowpass antialiasing filters should have a steep transition band,
A number of standardized filter-transfer excellent gain flatness, and low distortion in the passband.
functions, including Bessel, Butterworth,
Chebyshev, and elliptic, exist. Each has spe-
cific characteristics in the passband, transition 0
FOUR-POLE BUTTERWORTH
band, and stopband. Selecting the appropri- ⫺10
ate topology depends on the most critical ⫺20
EIGHT-POLE BUTTERWORTH
performance aspects of a design. Butterworth
⫺30
filters have the flattest passband region and
minimal group delays. Chebyshev filters have AMPLITUDE⫺40 EIGHT-POLE CHEBYSHEV
steeper roll-offs but more passband ripple. (dB) ⫺50
Elliptic filters feature the steepest roll-off ⫺60 EIGHT-POLE ELLIPTIC
(Figure 3). The figure does not show a Bessel ⫺70
filter, which has a more gradual roll-off but
⫺80
has the key advantage of a linear, or constant,
phase response. A number of public-domain ⫺90

tools exist to help developers in the design of ⫺100

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 2 3 4 5 6 7 8 10
a suitable antialiasing filter.
FREQUENCY (Hz) ALIASING
A consequence of using an antialiasing GUARDBANDS
filter is the limit on available alias-free band-
width when you use it in a traditional ADC. Figure 3 Possible antialias-filter designs illustrate the different transition-band
At first glance, the Nyquist theorem seems to characteristics of example filter systems.
promise a lot. Consider an ADC that samples
at 40M samples/sec at a clock frequency of 40
MHz. It theoretically promises a 20-MHz signal bandwidth. So, if this ADC were to sample a 5-MHz signal, then this
However, aliasing with practical filter design means that system would still see frequencies all the way out to 50 MHz.
the free bandwidth is considerably less than this amount. To fully sample the 5-MHz bandwidth and eliminate aliasing,
A 14-bit converter can resolve to one part in 214—that is, the correct sample frequency using this filter would need to
one part in 16,384. To bury any alias component in the be 100 MHz. The range of 5 to 50 MHz becomes a guardband
ADC’s noise floor requires attenuation to be less than ⫾0.5 against alias errors. An obvious option is to look for higher
LSB. That amount equates to 90-dB attenuation—that performance filters. Consider an aggressive, eight-pole filter.
is, ⫾0.5 LSB⫽one part in 32,768⫽90.3 dB. In practical Inspection shows that the 80-dB-attenuation point occurs at
terms, however, this level of attenuation need exceed only a frequency that is 3.2 times the cutoff frequency, or 16 MHz,
the measured SNR of a 14-bit ADC. A more realistic level a significantly reduced alias guardband. Alias-free sampling
in the filter design is an attenuation of 80 dB. requires considerably more system bandwidth to handle the
Figure 3 shows several possible filter topologies, includ- alias-guardband needs of an application. It is also important
ing two Butterworth-transfer functions—those of four- and to note the cost trade-offs you must weigh when considering
eight-pole systems—both compared with an ideal Nyquist the severity of the antialiasing filter and the performance
filter. Note that, by convention, the cutoff frequency is the level of the ADC.
point at which the filter produces 3 dB of attenuation. The To ease antialiasing-filter design, pipeline ADCs—often
horizontal axis shows the normalized input frequency as a confusingly referred to as Nyquist converters—have been
ratio of the absolute frequency to the cutoff frequency. Note offering increased sample rates and input bandwidths.
that the four-pole curve does not drop to 80 dB until the Oversampling a signal at twice the Nyquist rate evenly
input frequency has risen to 10 times the cutoff frequency. spreads the ADC’s quantization-noise power into a two-

66 EDN | NOVEMBER 23, 2006

 0 0

⫺20 ⫺20

⫺40 ⫺40

MAGNITUDE MAGNITUDE
(dB) ⫺60 (dB) ⫺60

⫺80 ⫺80

⫺100 ⫺100

⫺120 ⫺120
0 5 10 15 20 25 30 0 20 40 60 80 100 120 140 160 180
(a) FREQUENCY (MHz) (b) FREQUENCY (MHz)

Figure 4 Discrete time sampling produces a lowpass signal-transfer function in a discrete time delta-sigma ADC (a). The graph in (a)
appears to show alias protection; however, the transfer function of (a) is wrapped around integer multiples of the sampling frequency,
as the expanded plot (b) shows. Aliasing gaps appear centered on 60, 120, and 180 MHz in this case.

times-wider frequency band. Applying decimation to sub- ultrasound systems in which the received-signal phase carries
sample the resultant output samples yields a 3-dB/octave reflection information.
conversion gain. This technique is useful for deployment
in delta-sigma converters because it not only produces DELTA-SIGMA CONVERTERS
dynamic-range improvements, but also reduces the pressure Delta-sigma techniques place lower demands on antialias-
on the antialiasing filter by relaxing filter roll-off. Lower ing filters. Delta-sigma converters exploit oversampling. In
order antialiasing filters are easier to match across multiple the past, designers improved dynamic range by using high
channels than higher order ones. Oversampling techniques oversampling rates and a simple low-resolution quantizer.
reduce the demands on the filter networks, but higher- However, simple oversampling produces minimal conver-
sample-rate ADCs and faster digital processing use more sion-gain improvements. Applying feedback provides a faster
power and increase cost. route to conversion-gain improvements.
You must also consider the phase response of the antialias- Delta-sigma modulators apply feedback to shape the quan-
ing filters. A filtered signal should not see any significant tization noise in the frequency domain by pushing most noise
phase alteration. This alteration becomes even worse if phase power into frequencies beyond the signal band of interest.
varies according to input frequency. You normally measure Filtering can reduce the noise power in this band. Employing
phase variation in a filter in terms of group delay—that is, the oversampled systems, which provide free frequency space
derivative of phase with respect to frequency. For a noncon- beyond the signal band of interest, accomplishes this goal.
stant group delay, a signal spreads out in time, causing poor Conventional Nyquist converters achieve a 3-dB/octave
impulse response. Dispersion may be an additional worry for conversion gain through 2⫻ oversampling. Delta-sigma con-
system performance. This factor is important in the design of verters more efficiently build conversion gain, which the
order of the applied feedback loop determines. First-, second-,
20
DECIMATOR
or third-order loops can provide 9-, 15-, or 21-dB/octave
0
MODULATOR
COMBINED conversion gain, respectively.
Most delta-sigma-converter implementations are discrete-
⫺20
time systems in which designers build the loop-filter compo-
⫺40 nents from simple switched-capacitor filters. The signal-trans-
fer function of a delta-sigma modulator is an important factor
⫺60

GAIN
in such a design. Signal-transfer performance looks promising
(dB) ⫺80
in traditional discrete-time systems. Digital-decimation filters
⫺100
define the effective passband and provide a sharp transition
band. Unfortunately, switched-capacitor-filter networks,
⫺120
which define the input bandwidth, add a discrete-sampling
⫺140
effect to the modulator structure. This discrete sampling causes
a lowpass signal-transfer function (Figure 4a). Although this
⫺160
0 80 160 240 320 400 480 560 640 720 800 880 960 1040 1120 1200 1280 1360 1440 1520 1600 function seems acceptable, a closer inspection of a wideband-
FREQUENCY (MHz)
frequency plot reveals a problem: The passband of the digital
Figure 5 An aliasing-mitigation system ensures the analog-loop filter wraps around integer multiples of the sample frequency at
filter provides maximum stopband attenuation at the oversam- 60, 120, and 180 MHz (Figure 4b). No alias attenuation what-
pling frequency of the modulator. soever exists at these points, and this characteristic extends to
infinity. Preventing high-level, out-of-band noise at multiples

68 EDN | NOVEMBER 23, 2006

of the oversample rate is a challenge and loring this filter system for a specific beyond the oversampling frequency
a downside of such designs. product, the maximum-loop-filter at- can enter the first Nyquist zone. The
tenuation coincides with the minimum back-end digital filter provides a sharp
CONTINUOUS TIME attenuation that the decimation filter stopband attenuation, limiting the
In a continuous-time modulator, you offers. An aliasing-mitigation system maximum effective input bandwidth of
implement noise shaping using conven- ensures that the analog-loop filter pro- the ADC (blue line). Through this ar-
tional analog active filters. The benefit vides maximum stopband attenuation rangement, the maximum analog-loop-
of the continuous-time approach is that at the oversampling frequency of the filter attenuation always coincides with
you can design the loop filter to handle modulator (Figure 5, green line). This the folded-digital-filter minimum to
alias filtering of the input signal. Tai- attenuation ensures that no noise power maintain a high level of wideband at-
tenuation. The maximum attenuation
of the analog-loop filter coincides with
the alias passband of the digital filter.
The purple line shows the composite
transfer function.
The specific implementation of a
given delta-sigma topology determines
the performance of the antialias system.
For example, the 14-bit-resolution,
20M- to 40M-sample/sec Xignal (www.
xignal.com) XT11400 ADC achieves
76-dB SNR and provides a 20-MHz
analog-input bandwidth. The passband
gain flatness is ⫾0.002 dB, the transition
band is approximately 2.5 MHz wide,
and the unit achieves alias attenuation
of 80 dB beyond 22.5 MHz, all without
any external filtering. A digital allpass-
filter stage, which reduces dispersion to
0.3 samples, minimizes group delay. Such
approaches have benefits in reducing de-
sign complexity, especially in multiple-
channel designs in which cross-channel
filter matching is a major issue.
In summary, delta-sigma modulators
use oversampling to help simplify anti-
aliasing-filter design. For discrete-time
systems, you must use caution in design-
ing antialiasing filters because of the
potential occurrence of high-frequency
noise, which can couple and fold di-
rectly into the baseband. A continu-
ous-time alternative can eliminate the
need for all external antialiasing filters.
The maximum attenuation of the ana-
log-loop filter aims successfully to inter-
cept the passband frequency of the digi-
tal-decimation filter at the oversampling
frequency.EDN

AU T H O R ’ S B I O G R A P H Y
Mark Holdaway is product-marketing di-
rector at Xignal Technologies AG (Unter-
haching, Germany). He holds a bachelor’s
degree in electronics engineering from the
University of Salford (England). His in-
terests include cycling, skiing, gardening,
reading, and writing. You can reach him at
mark.holdaway@xignal.com.

70 EDN | NOVEMBER 23, 2006