TI_impact of Sampling Clock Spurs on Adc Performance
TI_impact of Sampling Clock Spurs on Adc Performance
–100
Clock Input with
Spur Present
Amplitude and
Phase-Noise –140
B Measurement
Clock –180
100 1k 10 k 100 k 1 M 10 M
Input ADC Frequency (Hz)
Digital Output
fIN
FFT Processing
Spur
Amplitude
(dBc)
FFT Analysis m m
taken to extract it with the resolution bandwidth of the instrument in that measurement:
Amplitude (dBc) = SX (dBc/Hz) + 10log(Resolution Bandwidth)
Due to the presence of the spur, the original sampling instant, or zero crossing of the clock, has shifted slightly by ∆T.
Now the sampling instant, y(t) = 0, can be solved for:
y ( t ) = A × sin[wCLK ( t + ∆T )] + B × sin[wS ( t + ∆T )] = 0
y ( t ) = A × sin(wCLK t ) × cos(wCLK ∆T ) + A × cos(wCLK t ) × sin(wCLK ∆T ) + B × sin(wS t ) × cos(wS ∆T ) + B × cos(wS t ) × sin(wS ∆T ) = 0
0 1 1 ωCLK∆T 1 1 ωS∆T
y ( t ) = A × wCLK ∆T + B × sin(wS t ) + B × wS ∆T = 0
B × sin(wS t ) B × sin(wS t)
Then ∆T can be solved for: ∆T = − . Assuming that A >> B results in ∆T = − .
A × wCLK + B × wS A × wCLK
Next, the input signal, x(t) = AIN×sin(wINt), is sampled at the zero crossing, t + ∆T, of the non-ideal clock:
x( t ) = A IN × sin(ωIN T ) = A IN × sin ωIN ( t + ∆T ) = A IN × sin(ωIN t ) × cos(ωIN ∆T ) + A IN × cos(ωIN t ) × sin ( ωIN ∆T )
1 ωIN∆T
Therefore, it can be observed that each spurious component of the sampling clock generates two spurs, S1 and S2, in the
data converter with amplitude and frequencies relative to the input signal as follows.
B wIN B fIN
S1 and S2 amplitude: × = × or, in terms of decibels,
A 2 × wCLK A 2 × fCLK
f
= B − A + 20 log IN .
2× fCLK
S1 and S2 frequencies: fS1 = − fS − fIN
fS 2 = −fS + fIN
The resulting spurs can be shifted by one clock period, 2p/T = fCLK, and considering fS – fCLK = m yields:
fS1 = − fS − fIN + fCLK = −fIN + fCLK − fS = −( fIN − fCLK + fS ) = −( fIN + m ) = fIN + m
fS 2 = − fS + fIN + fCLK = + fIN + fCLK − fS = fIN − m
These equations show that the frequencies of the gener Sometimes the fast Fourier transform (FFT) plot can be
ated spurs will be centered around the input signal and a bit misleading when one is trying to trace spurs back to
offset by the distance m, which is the difference between their origins. If the clock spur is relatively far from the
the clock frequency and the clock-spur frequency. The clock frequency, the generated spurs of the ADC can get
amplitude of the generated spurs, on the other hand, is pushed outside the plot’s boundaries—either to negative
highly dependent upon the input frequency. For every frequencies or beyond fCLK/2. The spurs then alias back in-
doubling of the input frequency (e.g., fIN = 20 MHz versus band and generate an asymmetric FFT plot, as demon-
fIN = 10 MHz), the spur amplitude increases by 6 dB! strated in Figure 3.
Hence, as system designers consider sampling in higher
Nyquist zones, this relationship becomes very important
to them.
Figure 3. Spurs pushed outside the FFT band and aliased back in-band
m m m m
–10
–20 33 dB
–30 fS = 102 MHz (–30 dBm),
m = 20-MHz Offset
–40
–50
Phase Noise (dBc/Hz)
–60
–70
–80
–90
–100
–110
–120
–130
–140
–150
–160
–170
100 1k 10 k 100 k 1M 10 M
Frequency (Hz)
0
fIN = 10-MHz sine wave
–20 fS = 102 MHz (–40 dBm)
FFT Output Plot (dBc)
62 dB
–40 fS2 = 10.9 MHz fS1 = 30.9 MHz
–60
–80
–100
–120
0 10 20 30 40 50 60
Frequency (MHz)
This was also confirmed with the FFT plot of the ADS5463 Once again, this correlated very well with the FFT out-
output (see Figure 8). put plot from the ADS5463, as illustrated in Figure 9.
For the last experiment, a comparison of spur frequen-
cies was made with the clock frequency set at 102 MHz Practical example
and at 132 MHz. The spur amplitude was set to –30 dBm, Let’s go back and analyze the case of the CDCE72010,
and the input signal was set to 10 MHz. These settings mentioned earlier under “Introduction.” This device’s low-
caused the spur-frequency offset (m) to change from jitter phase-locked loop was configured to drive the TI
about 20.9 MHz to about 9.1 MHz, respectively. Two new ADS5483 with LVDS outputs at 122.88 MSPS. No filter
spur frequencies resulted: was placed between the outputs of the CDCE72010 and
the clock input of the ADS5483. This way the full effect of
fS1 = fIN + m = 10 MHz + 9.1 MHz = 19.1 MHz
the clock spurs in a real-world design can be observed.
fS 2 = fIN − m = 10 MHz − 9.1 MHz = 0.9 MHz
0
fIN = 100-MHz sine wave
32 dB fS = 102 MHz (–30 dBm)
–20
FFT Output Plot (dBc)
–60
–80
–100
–120
0 10 20 30 40 50 60
Frequency (MHz)
0
Spurs at Spurs at fIN = 10-MHz sine wave
–20 0.9 and 10.9 and fS = 102 MHz (–30 dBm)
19.1 MHz 30.9 MHz
FFT Output Plot (dBc)
–60
–80
–100
–120
0 10 20 30 40 50 60
Frequency (MHz)
10
= −81.6 dBc .
–150
For S2, − 144 dBc / Hz = −144 dBc + 10 log(3 MHz × 1%)
= −144 dBc + 45 dB –160
= −99 dBc .
–170
These results can be used to calculate the spur ampli-
tudes of the ADC output spectrum: –180
100 1k 10 k 100 k 1M 10 M
100 MHz
S1 = 81.6 dBc + 20 log Frequency (Hz)
2 × 122.88 MHz
= −81.6 dBc − 7.8 dB
= −89.4 dBc
Figure 11. FFT output with 100-MHz input and a
100 MHz 122.88-MHz LVDS clock
S2 = −99 dBc + 20 log
2 × 122.88 MHz
= −99 dBc − 7.8 dB 0
S1 = – 87.2 dBc fIN = 100-MHz sine wave
= −106.8 dBc –20 mS1 offset = 27 MHz Clock Driver: CDCE72010
These amplitudes match the measured spur
FFT Output Plot (dB)
–100
–120
0 10 20 30 40 50 60
Frequency (MHz)
11
Figure 12. FFT output with 122.88-MHz clock and a 102-MHz, –30-dBm spur
Spurs 1 and 2
SNR 50 dBFS
Impact of clock spurs on SNR For calculating the SNR of the data converter, the DJ
Besides reducing the SFDR, spurs on the clock also impact needs to be added to the phase noise of the clock and the
the SNR of the data converter. Since the spurs are at a aperture jitter of the ADC. However, in this case, the DJ
fixed frequency, they are considered deterministic jitter far exceeds the other two jitter components. Therefore,
(DJ); and they contribute to the overall clock jitter, which the resulting SNR can be calculated with a jitter of about
in turn highly impacts the SNR. 52 ps (fIN = 10 MHz), which is approximately 50.5 dBFS.
The peak-to-peak DJ from the clock spur can be The resulting FFT plot of this setup with the ADS5463
approximated by is shown in Figure 12. The plot clearly shows the two
resulting spurs with an amplitude of –52 dBc and an SFDR
SX ( dBc )
of about –52 dBc. The SNR ≈ 50 dBFS, which matches the
2 × 10 20 calculated value very well.
DJPP ≈ ,
p × fCLK
Conclusion
where SX (dBc) is the spur amplitude in dBc. The RMS This article has shown that spurs on the ADC sampling
jitter can be calculated as clock can significantly degrade the overall system SFDR
DJPP as well as the SNR. This effect gets amplified even more
DJRMS ≈ . in undersampling applications where the signal input is
14
moved to higher frequencies than those traditionally used
As in the first experiment, with the measured amplitude for baseband input. Therefore, it can be concluded that a
of the spurs at –33 dBm and that of the clock at about filtered, high-quality sampling clock is necessary for system
–10 dBm, the relative spur amplitude is roughly engineers who are trying to achieve maximum data-
–33 dBm – (–10 dBm) = –23 dBc. converter performance.
Substituting –23 dBc into the formula for DJRMS yields Related Web sites
−23
dataconverter.ti.com
DJ 1 2 × 10 20 www.ti.com/sc/device/partnumber
DJRMS ≈ PP = × = 26 ps.
14 14 p × 122.88 MHz Replace partnumber with ADS5463, ADS5483, or
CDCE72010
Since there are two spurs with a 20-MHz offset, the 26-ps
DJ of each spur can be summed together for a total DJ of
about 52 ps.
12
Internet
TI Semiconductor Product Information Center Home
Page
support.ti.com