AD9853 Analog Devices, AD9853 Datasheet - Page 19

• Manufacturer Part Number: AD9853
• Description: Programmable Digital OPSK/16-QAM Modulator
• Manufacturer: Analog Devices

The frequency response, H(f), of a CIC filter is found by evalu-
ating H(z) at z = e
where f is relative to the input sample rate of the CIC filter.
With this formula, we can accurately predict the frequency
response of the CIC filters.
Compensating for CIC Roll-Off
As discussed previously, the CIC filters offer a low-pass charac-
teristic that can be used to eliminate the spectral images pro-
duced by the FIR filters. Unfortunately, the CIC response is not
flat over the frequency range of the baseband signal. Thus, the
inherent attenuation (or roll-off) of the CIC filters distorts the
baseband data signal. So even though the CIC filters help to
eliminate the images described earlier, they introduce another
form of error to the baseband signal—frequency-dependent
amplitude distortion. This ultimately manifests itself as a higher
level of Error Vector Magnitude (EVM) at the output of the
I and Q modulator. Also, the larger the bandwidth of the
baseband signal, the more pronounced the CIC roll-off, the
greater the amplitude distortion and the worse the EVM perfor-
mance. This is a serious problem because if a value of
used for the SRRC response of the FIR filters, a doubling of the
bandwidth of the baseband signal results and hence, a degrada-
tion in EVM performance.
Fortunately, there is a way to compensate for the effects of CIC
roll-off. Since the frequency response of the CIC filters is pre-
dictable, it is possible to compensate for the CIC roll-off charac-
teristic by adjusting the response of the FIR filters accordingly.
The adjustment is accomplished by modifying the FIR filter
response with a response that is the inverse of that of the CIC
filters. This is done by precompensating the FIR filters.
To perform CIC compensation, we simply define a function
(H
response. Specifically,
By multiplying the original FIR filter frequency response by
H
Unfortunately, it’s not quite this simple. Recall that the coeffi-
cients of the baseband filter were computed using an inverse
Fourier transform integral which included the SRRC function.
In order to compensate for the CIC filter response, the SRRC
function must be multiplied by the H
frequency scale of the SRRC response is computed based on
frequencies relative to the symbol rate, while the H
tion is computed relative to the input sampling rate of the CIC
filter. The input CIC sampling rate happens to be the same as
the sample rate of the FIR filter (see Figure 36), or four times
the symbol rate. Thus, we have a frequency scaling problem.
This problem is easily corrected by introducing a frequency
scaling factor (FreqScale = 4) into the H
REV. C
COMP
COMP
, we obtain the necessary compensation.
) that has a response which is the inverse of the CIC
j(2 f/R)
H f
( )
H
:
COMP
R M
k
f
0
1
e
j
H f
2
COMP
1
f R k
/
COMP
function. But the
N
function so that
COMP
=1 is
func-
–19–
the frequency scales of the two functions match. Thus, the
actual H
It should be noted that in compensating for the CIC roll-off,
only the first stage CIC filter need be considered. This is due to
the fact that at the output of the first stage CIC filter the
bandwidth of the signal is reduced to the point that the roll-off
introduced by the second stage is negligible in the region of the
baseband signal.
The CIC compensation method is demonstrated by example
(using MathCad) in Figures 34 and 35. An interpolation rate
(R) of 6 is used in the example. The improvement obtained by
compensating for the CIC response is graphically demonstrated
in Figure 35 which shows:
• the SRRC filter response (which is the desired overall response)
• the composite response of the SRRC in series with the CIC
• the composite response of the compensated SRRC in series
Note that the ideal SRRC response and the compensated com-
posite response are virtually identical in the region of the pass-
band. Thus, the goal of correcting for the CIC filter response
has been accomplished.
There is one subtlety to be noted in the example. The CIC
compensation is only applied to the first 90% of the bandwidth
of the baseband signal (note the variable
It was found that compensation over the full 100% of the band-
width produced a reduction in the suppression of signals in the
stopband region of the SRRC. This resulted in creating more
distortion than by not correcting for the CIC roll-off in the first
place. However, by slightly reducing the bandwidth over which
correction is applied, the stopband suppression is once again
restored and a significant improvement in EVM performance is
obtained.
Determining the Necessary Interpolator Rate Change Ratio
The AD9853 contains three stages of digital interpolation:
1) Fixed 4 Pulse Shaping FIR Filter.
2) Programmable 3 to 31 First Interpolation Filter.
3) Programmable 2 to 63 Second Interpolation Filter.
After the serial input data stream has been encoded into QPSK
or 16-QAM symbols, the symbol interpolation rate of the AD9853
is determined by the product of the three interpolating stages
listed above. In QPSK mode, the minimum symbol interpolation
rate that will work is 4 3
is 4
4
QPSK is equal to 1/2 the bit rate of the data and for 16-QAM it
is 1/4 the bit rate. Figure 36 is a partial block diagram of the
AD9853 and follows the path of the data stream from the input
of the I and Q encoder block to the output of the DAC.
filter (distorted response)
with the CIC (corrected response)
31
4
COMP
63 = 7812. The symbol rate at the encoder output for
3 = 48. The maximum symbol interpolation rate is
function required is given by:
H
COMP
2 = 24; for 16-QAM the minimum
H
FreqScale
1
f
inside the integral).
AD9853