2.3.2 Practical Specifications for Real ADCs
The SNR in a real ADC can be determined by measuring the residual error. Residual error is the
combination of quantization noise, random noise, and nonlinear distortion (i.e., all of the
undesired components of the output signal from the ADC). The residual error for an ADC is
found by using a sinusoidal input into the ADC. An estimate of the input signal is subtracted
from the output of the ADC; the remaining signal is the residual error. The mean squared (MS)
power of the residual error then is computed. The SNR then is found by dividing the mean
squared power of the input signal by the mean squared power of the residual error.
2
A specification sometimes used for real ADCs instead of the SNR is the effective number of bits
(ENOB). This specification is defined as the number of bits required in an ideal ADC so that the
mean squared noise power in the ideal ADC equals the mean squared power of the residual
error in the real ADC.
The spurious free dynamic range (SFDR) is another useful specification for ADCs. One
definition of the SFDR assumes a single tone sinusoidal input into the ADC. Measurement of
this SFDR is made by taking the Fast Fourier Transform (FFT) of the output of the ADC. This
provides the frequency spectrum of the output of the ADC and is plotted as the ADC output
power in dB vs. frequency. The SFDR is then the difference between the power in the sinusoidal
input signal and the peak power of the largest spurious signal in the ADC output spectrum. An
example of determining the SFDR from the ADC output spectrum is shown in Figure 5. In this
idealized ADC output spectrum, the input signal is a 10-MHz sinusoid. Various spurious
responses are shown. The SFDR is 50 dB.
SFDR allows one to assess how well an ADC can detect simultaneously a very small signal in
the presence of a very large signal. Hence, it is an important specification for ADCs used in radio
receiver applications. A common misconception is that the SFDR of the ADC is equivalent to the
SNR of the ADC. In fact, there is typically a large difference between the SFDR and the SNR of
an ADC. The SNR is the ratio between the signal power and the power of the residual error. The
SFDR, however, is the ratio between the signal power and the peak power of only the largest
spurious product that falls within the band of interest. Therefore, the SFDR is not a direct
function of bandwidth; it does not necessarily change with a change in bandwidth, but it may.
Since the power of the residual error includes quantization noise, random noise, and nonlinear
2
The SNR is often (and more accurately) called the signal-to-noise plus distortion ratio (SINAD) when distortion is
included with the noise (as in this case).
13
distortion within the entire 0 to f
s
/2 band, the power of the residual error can be much higher than
the peak power of the largest spurious product. Hence, the SFDR can be much larger than the
SNR [5]. A practical example of this can be seen from the specifications for the Analog Devices
AD9042 monolithic ADC. With a 19.5-MHz analog input signal, 1 dB below full scale (the full-
scale input is 1 V
p-p
), the typical SFDR specification is 81 dB while the SNR specification is
66.5 dB (from −40 - +85 °C) [12].
Figure 5. Example ADC output spectrum showing the spurious free dynamic range.
The SFDR specification is useful for applications when the desired signal bandwidth is smaller
than f
s
/2. In this case, a wide band of frequencies is digitized and results in a given SNR. The
desired signal then is obtained by using a narrowband digital bandpass filter on this entire band
of frequencies. The SNR is improved by this digital-filtering process since the power of the
residual error is decreased by filtering. The SFDR specification for the ADC is important
because a spurious component still may fall within the bandwidth of the digital filter; hence, the
SFDR, unlike the SNR, does not necessarily improve by the digital-filtering process. However,
several techniques are available to improve the SFDR. Dithering (discussed in Section 2.2)
improves the SFDR of ADCs. Additionally, postdigitization-processing techniques such as state
variable compensation [13], phase-plane compensation [14], and projection filtering [15] have
been used to improve SFDR.
10
0
-10
-10
-30
-40
-50
-60
-70
-80
-90
-100
0 2 4 6 8 10 12 14 16 18 20
SFDR=
50dB
Frequency (MHz)
P
o
w
er
(
d
B
m
)
14
For an ideal ADC, and in practical sigma-delta (ΣΔ) converters, the maximum SFDR occurs at a
full-scale input level. In other types of practical ADCs, however, the maximum SFDR occurs at
input levels at least several dB below the full-scale input level. This occurs because as the input
levels approach full-scale (within several dB), the response of the ADC becomes more nonlinear
and more distortion is exhibited. Additionally, due to random fluctuations in the amplitude of
real input signals, as the input signal level approaches the FSR of the ADC, the probability of the
signal amplitude exceeding the FSR increases. This causes additional distortion from clipping.
Therefore, it is extremely important to avoid input signal levels that closely approach the full-
scale level in ADCs. Prediction of the SFDR for practical ADCs is difficult, therefore
measurements are usually required to characterize the SFDR.
In the preceding discussion on SFDR, a sinusoidal ADC input signal was assumed. However,
intermodulation distortion (IMD) due to multitone inputs is important in ADCs used for
wideband radio receiver applications. To characterize this IMD due to multitone inputs, another
definition of the SFDR could be used. In this case, the SFDR is the ratio of the combined signal
power of all of the multitone inputs to the peak power of the largest spurious signal in the ADC
output spectrum. A current example of test equipment to generate multitone inputs produces up
to 48 tones.
The noise power ratio (NPR) specification is useful in applications such as mobile cellular radio,
where the spectrum of a signal to be digitized consists of many narrowband channels and where
adjacent channel interference can degrade system performance. Particularly, the NPR provides
information on the effectiveness of an ADC in limiting crosstalk between channels [13].
The NPR is measured by using a noise input signal into the ADC. This noise signal has a flat
spectrum that is bandlimited to a frequency that is less than one-half the sampling frequency.
Additionally, a narrow band of frequencies is removed from the noise signal using a notch filter.
This noise spectrum is used as the input signal to the ADC. The frequency spectrum of the output
of the ADC then is determined. The NPR then is computed by dividing the power spectral
density of the noise outside the frequency band of the notch filter by the power spectral density
of the noise inside the frequency band of the notch filter [5].
When using an ADC in a bandpass-sampling application where the maximum input frequency
into the ADC is actually higher than one-half the sampling frequency, the full-power analog
input bandwidth is an important specification. A common definition (although not universal) of
full-power analog input bandwidth is the range from DC to the frequency where the amplitude of
the output of the ADC falls to 3 dB below the maximum output level. This assumes a full-scale
input signal to the ADC. Typically, the ADC is operated at input frequencies below this
bandwidth. Aside from full-power analog input bandwidth, it is important to examine the
behavior of the other specifications such as SNR, SFDR, and NPR at the desired operating
frequencies since these specifications typically vary with frequency. In addition to the SNR,
SFDR, and NPR of real ADCs being a function of frequency, they are also a function of input
signal amplitude. Table 1 provides a summary of the important ADC specifications for radio
receiver applications.
15
Table 1. Summary of ADC Specifications for Radio Receiver Applications
Specification
Application
Definition
Signal-to-Noise
Ratio (SNR)
Desired Signal BW
Equal to f
s
/2
MS Signal Power
MS Power of Residual Error
Spurious Free
Dynamic Range
(SFDR)
Desired Signal BW
Less Than f
s
/2
MS Signal Power
Peak Power of the Largest Spurious Product
Noise Power Ratio
(NPR)
Desired Signal
Spectrum Contains
Many Narrowband
Channels
Power Spectral Density of Noise
*
Outside Freq. Band of Notch Filter
Power Spectral Density of Noise
Inside Freq. Band of Notch Filter
Full-Power Analog
Input BW
Bandpass Sampling
Range from DC to Frequency Where
Output Amplitude Falls to 3 dB Less Than
Maximum**
* With an input signal having a bandlimited, flat noise spectrum and a narrow band of frequencies removed by a
notch filter.
**For a full-scale input signal.
When testing an ADC, it is important to ensure that all quantization levels are tested. For single
tone inputs, the relationship between the input signal frequency and the sampling rate must be
chosen so that the same small set of quantization levels is not tested repeatedly. In other words,
the samples should not always occur at the same amplitude levels of the input signal. For
example, using an input frequency of f
s
/8 is a poor choice since the same eight amplitude levels
are sampled every period of the input signal (assuming that the input signal and the sampling
clock are phase coherent) [16]. The histogram test can be used to ensure that all quantization
levels are tested. In the histogram test, an input signal is applied to the ADC and the number of
samples that are taken at each of the 2
B
quantization levels are recorded. In an ideal ADC this
histogram is identical to the probability density function of the amplitude values of the input
signal. Comparing the histogram to the probability density function of the input signal gives an
indication of the nonlinearity of the ADC. An examination of the histogram reveals whether all
of the different quantization levels are being tested. When no samples are recorded for a given
quantization level, this level is either not being tested by the testing procedure (input signal and
sampling rate) or the ADC is exhibiting a missing code. A missing code is a quantization level
that is not present in the output of a real ADC that is present in the output of an ideal ADC.
Missing codes are fairly rare in currently available ADCs in general and do not occur in ΣΔ
converters.
Chia sẻ với bạn bè của bạn: |