CMX983 Datasheet, PDF (38/93 Pages) CML Microcircuits – Programmable Channel Filter

Analogue Front End (AFE) for Digital Radio
CMX983
otherwise the original signal remains undistorted. The first upsample factor can be set to any value up to 8.
Alternatively, the first upsampler, FIR filter and bit selector can be disabled and bypassed.
Interpolation filter
The interpolation filter is a 128-tap FIR type that takes the zero padded samples from the first upsampler at a
frequency equal to CT2. There are two banks of programmable coefficients. The purpose of the low-pass
interpolation filter is to attenuate the unwanted spectral images caused by the first upsampler, perform any
transfer function shaping that is required by the transmission standard and, if necessary, provide
compensation for the droop caused by the second upsampler or the DAC reconstruction filter.
Bit selector
The bit selector at the output of the interpolation filter selects which 16 bits of the 38-bit FIR filter accumulator
are passed to the following upsample and hold stage.
Second upsample and hold
The second upsampler includes a zero-order hold function, so for an upsample factor of L2 (= f_CT3/f_CT2) there
are L2-1 repeated sample values inserted after each sample value from the bit selector. This upsampling
process results in unwanted spectral images at multiples of the CT2 frequency, but these lie near the nulls in
transfer function of the zero-order hold so they are usefully attenuated. The transfer function of the zero-order
hold is H(z)=(1-z^-L2)/(z-1), where the sample period T=1/f_CT2. This transfer function also causes droop in the
wanted signal, but if the CT2 clock rate is high enough this can be made insignificant. If necessary, droop
compensation can be performed in the preceding FIR interpolation filter. The second upsample factor can be
set to any value up to 32.
Sigma-delta modulator and reconstruction filter
The sigma-delta modulator is a 2^nd-order type whose output has a characteristic high-pass filtered noise
profile, with the quantisation noise rising at 12dB per octave. The reconstruction filter attenuates this
quantisation noise, along with any spectral remnants from the upsamplers. The reconstruction filter comprises
a linear-phase switched capacitor filter (with a selectable bandwidth) followed by a linear-phase continuous
time filter. These have the following nominal transfer characteristics, with a sample period T=1/f_CT3:
Switched capacitor section:

^Low

bandwidth
:
H
(z)

0.0413510
1 0.958649z
1

0.00206664z ^1
11.93323z ^1  0.935301z
2


_{High bandwidth : H (z) }
0.0794179 _
0.00826662z ^1

1 0.920582z ^1 11.86233z ^1  0.870601z ^2
Continuous time section:
H(s) 
1

1
1.868410^12 s ² 1.978010^6 s 1 1_.4_96__10_^6_s _1
Recommended
off chip filter
With a sigma-delta clock frequency of 2.4MHz, the switched capacitor section has a -3dB cutoff frequency of
approximately 12.2kHz (low b/w) or 23.9kHz (high b/w). The continuous time filter has a -3dB cutoff frequency
of approximately 80kHz.
Output gain stage
 2015 CML Microsystems Plc
38
D/983/6