AD9879
tSU
MCLK
TxSYNC
TxIQ
tHD
TxI[11:6]
TxI[5:0] TxQ[11:6] TxQ[5:0] TxI[11:6]
TxI[5:0] TxQ[11:6] TxQ[5:0] TxI[11:6]
TxI[5:0]
Figure 11. Timing Diagram for Register Read
signals having a bandwidth of no more than about 60% of fNYQ
Half-Band Filters (HBFs)
.
HBF 1 and HBF 2 are both interpolating filters, each of which
doubles the sampling rate. Together, HBF 1 and HBF 2 have
26 taps and provide a factor-of-four increase in the sampling
rate (4 ꢁ fIQCLK or 8 ꢁ fNYQ).
Thus, in order to keep the bandwidth of the data in the flat portion
of the filter pass band, the user must oversample the baseband
data by at least a factor of two prior to representing it to the
AD9879. Note that without oversampling, the Nyquist bandwidth
of the baseband data corresponds to the fNYQ. As such, the upper
end of the data bandwidth will suffer 6 dB or more of attenuation
due to the frequency response of the digital filters. Furthermore,
if the baseband data applied to the AD9879 has been pulse
shaped, there is an additional concern. Typically, pulse shaping
is applied to the baseband data via a filter having a raised cosine
response. In such cases, an ꢅ value is used to modify the band-
width of the data where the value of ꢅ is such that 0 < ꢅ < 1.
A value of 0 causes the data bandwidth to correspond to the
Nyquist bandwidth. A value of 1 causes the data bandwidth to
be extended to twice the Nyquist bandwith. Thus, with 2ꢁ over-
sampling of the baseband data and ꢅ =1, the Nyquist bandwidth
of the data will correspond with the I/Q Nyquist bandwidth. As
stated earlier, this results in problems near the upper edge of the
data bandwidth due to the frequency response of the filters. The
maximum value of ꢅ that can be implemented is 0.45. This is
because the data bandwidth becomes:
In relation to phase response, both HBFs are linear phase filters.
As such, virtually no phase distortion is introduced within the
pass band of the filters. This is an important feature as phase
distortion is generally intolerable in a data transmission system.
Cascaded Integrator-COMB (CIC) Filter
The CIC filter is configured as a programmable interpolator and
provides a sample rate increase by a factor of 4. The frequency
response of the CIC filter is given by:
−j(2πf(4)) 3
3
1 1− e
1 sin(4πf)
H(f) =
=
1− ej2πf
4
4
sin(πf)
The frequency response in this form is such that f is scaled to the
output sample rate of the CIC filter. That is, f = 1 corresponds
to the frequency of the output sample rate of the CIC filter.
H(f/R) will yield the frequency response with respect to the input
sample of the CIC filter.
1 2 1+ α f
= 0.725 fNYQ
(
)
NYQ
Combined Filter Response
which puts the data bandwidth at the extreme edge of the flat
portion of the filter response.
The combined frequency response of HBF 1, HBF 2, and CIC
puts a limit on the input signal bandwidth that can be propagated
through the AD9879.
If a particular application requires an ꢅ value between 0.45 and 1,
then the user must oversample the baseband data by at least a
factor of four.
The usable bandwidth of the filter chain puts a limit on the
maximum data rate that can be propagated through the AD9879.
A look at the pass-band detail of the combined filter response
(Figure 12 and Figure 13) indicates that in order to maintain an
amplitude error of no more than 1 dB, we are restricted to
The combined HB1, HB2, and CIC filter introduces, over the
frequency range of the data to be transmitted, a worst-case droop
of less than 0.2 dB.
1
1
0
–1
–2
0
–1
–2
–3
–4
–5
–6
–3
–4
–5
–6
0
0.1
0.2
0.3
0.4
0.5
0.6 0.7
0.8
0.9
1.0
0
0.1
0.2
0.3
0.4
0.5
0.6 0.7
0.8
0.9
1.0
FREQUENCY RELATIVE TO I/Q NYQUIST BW
FREQUENCY RELATIVE TO I/Q NYQUIST BW
Figure 12. Cascaded Filter Pass-Band Detail (N = 4)
Figure 13. Cascaded Filter Pass-Band Detail (N = 3)
REV. 0
–18–