AD9853 Analog Devices, AD9853 Datasheet - Page 17

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

the next contains the message information and is used to de-
modulate the signal instead of the absolute phase of the signal.
The transmitter and receiver must use the same symbol deriva-
tion scheme.
Differential encoding in the AD9853 occurs while data still
exists as a serial data stream. When in straight QPSK or 16-QAM,
the serial data stream passes to the symbol mapper/format en-
coder stage without modification. When differential encoding is
engaged, the serial data stream is modified prior to the symbol
mapper/format stage according to Table VI. Only I1 and Q1 are
modified, even in the D16-QAM mode whose symbols are com-
posed of Q1, I1, Q0, I0. In D16-QAM, only the two MSBs of
the 4-bit symbol are modified; furthermore, the “previously
transmitted symbol” referred to in Table VI are the two MSBs
of the previous 4-bit symbol.
Symbol mapping for QPSK and DQPSK are identical. Symbol
mapping for 16-QAM and D16-QAM are slightly different (see
Figure 37) in accordance with MCNS (DOCSIS) specifications.
Special Note: For most modulation modes, a minimum pre-
amble is required. For DQPSK the minimum preamble is one
symbol (2 bits) and for either 16-QAM or D16-QAM the mini-
mum preamble is one symbol (4 bits). For FSK or QPSK, no
preamble is required.
User should be additionally aware that in the DQPSK mode,
the preamble is not differentially encoded in accordance with
MCNS (DOCSIS) specifications. If the preamble must be dif-
ferentially encoded, it can “pre-encoded” using the derivation in
Table VI. In D16-QAM, the preamble is always differentially
encoded as is the “payload” data.
When initiating a new differentially encoded transmission, the
“previously transmitted symbol” is always the last symbol of the
preamble.
PROGRAMMABLE PULSE-SHAPING FIR FILTERS
The I and Q data paths of the modulator each contain a pulse
shaping filter. Each is a 41-tap, linear phase FIR. They are used
to provide bandwidth containment and pulse shaping of the data
in order to minimize intersymbol interference. The filter coeffi-
cients are programmable, so any realizable linear phase response
characteristic may be implemented. The linear phase restriction
is due to the fact that the user may only define the center coeffi-
cient and the lower 20 coefficients. The hardware fills in the
upper 20 coefficients as a mirror image of the lower 20. This
forces a linear phase response. It should also be noted that the
pulse shaping filter upsamples the symbol rate by a factor of
four.
Normally, a square-root raised cosine (SRRC) response is desired.
In fact, the AD9853 Evaluation Board software driver implements
an SRRC response. When using the SRRC response, an excess
bandwidth factor ( ) is defined that affects the low pass roll-off
characteristic of the filter (where 0
SRRC is an ideal low-pass filter with a “brick wall” at one-half
of the symbol rate (the Nyquist bandwidth of the data). Although
this provides maximum bandwidth containment, it has the ad-
verse affect of causing the tails of the time domain response to
be large, which increases intersymbol interference (ISI). On the
other hand, when
characteristic that significantly reduces the time domain tails,
which improves ISI. Unfortunately, the cost of this benefit is a
doubling of the bandwidth of the data signal. Values of
REV. C
= 1, the SRRC yields a smooth roll-off
1). When
= 0, the
between
–17–
0 and 1 yield a tradeoff between excess bandwidth in the fre-
quency domain and tail suppression in the time domain.
The FIR filter coefficients for the SRRC response may be calcu-
lated using a variety of methods. One such method uses the
Inverse Fourier Transform Integral to calculate the impulse re-
sponse (time domain) from the SRRC frequency response (fre-
quency domain). An example of this method is shown in Figure
33. Of course, this method requires that the SRRC frequency
response be known beforehand.
The FIR filters in the AD9853 are implemented in hardware
using a fixed point architecture of 10-bit, twos complement
integers. Thus, each of the filter coefficients, a
such that:
PROGRAMMABLE INTERPOLATION FILTERS
The AD9853 employs two stages of interpolation filters in each
of the I and Q channels of the modulator. These filters are
implemented as Cascaded Integrator-Comb (CIC) filters. CIC
filters are unique in that they not only provide a low-pass fre-
quency response characteristic, but also provide the ability to
have one sampling rate at the input and another sampling rate at
the output. In general, a CIC filter may either be used as an
interpolator (low-to-high sample rate conversion) or as a
decimator (high-to-low sample rate conversion). In the case of
the AD9853, the CIC filters are configured as interpolators,
only. Furthermore, the interpolation is done in two separate
stages with each stage designed so that the rate change is pro-
grammable. The first interpolator stage offers rate change ratios
of 3 to 31, while the second stage offers rate change ratios of 2
to 63.
As stated in the previous section, the data coming out of the
FIR filters is oversampled by four. Spectral images appear at
their output (a direct result of the sampling process). These
images are replicas of the baseband spectrum which are re-
peated at intervals of four times the symbol rate (the rate at
which the FIR filters sample the data). The images are an un-
wanted byproduct of the sampling process and effectively repre-
sent a source of noise.
Normally, the output of the FIR filters would be fed directly to
the input of the I and Q modulator. This means that the spectral
images produced by the FIRs would become part of the modu-
lated signal—definitely not a desirable consequence. This is
where the CIC filters play their role. Since they have a low-pass
characteristic, they can be used to eliminate the spectral images
produced by the FIRs.
Frequency Response of the CIC Filters
The frequency response of a CIC filter is predictable. It can be
shown that the system function of a CIC filter is:
Where N is the number of cascaded integrator (or comb) sec-
tions, R is the rate change ratio, and M is the number of unit
delays in each integrator/comb stage. For the AD9853, two of
these variables are fixed as a result of the hardware implementa-
tion; specifically, N = 4 and M = 1. As mentioned earlier, R (the
rate change ratio) is programmable.
–512
a
i
511
H z
( )
[i = 0, 1, … , 40]
R M
k
0
1
z
k
N
i
, is an integer
AD9853