CSP1027 Voice Band Codec for
Cellular Handset and Modem Applications
Data Sheet
December 1999
Lucent Technologies Inc.
6
4 Architectural Information
(continued)
4.1 Overview
The CSP1027 is a complete analog-to-digital and digi-
tal-to-analog acquisition and conversion system (see
Figure 3 on page 5) that band limits and encodes ana-
log input signals into 16-bit PCM, and takes 16-bit PCM
inputs and reconstructs and filters the resultant analog
output signal. The selectable A/D input circuits, pro-
grammable sample rates, and digital filter options allow
the user to optimize the codec configuration for either
speech coding or voice band data communications.
The on-chip digital filters meet the ITU-T G.712 voice
band frequency response and signal to distortion plus
noise specifications and are suitable for IS-54, GSM,
and JDC digital cellular applications. In addition, the
small supply current drain, when powered down,
extends battery life in mobile communication applica-
tions.
The CSP1027 is intended for both voice band voice and
data communication systems. As a result, this codec
has a variety of features not found in standard voice
band codecs:
I
3.0 V regulated power supply for a condenser micro-
phone.
I
Microphone preamplifier with programmable input
ranges.
I
Mute control of D/A output.
I
Programmable output gain in 3 dB increments.
I
Output speaker driver.
I
Programmable master clock divider to set A/D and
D/A conversion rate.
I
Testability loopback mode.
I
High-quality dither scheme to eliminate idle channel
tones.
4.2 Description of Signal Paths
4.2.1 Sampling Frequency
The oversampling ratio of the codec is 125:1; this is the
ratio of the frequency of the oversampling clock to the
frequency of the sampling clock. Most speech applica-
tions specify a sampling frequency of 8 kHz, yielding an
oversampling frequency of 8 kHz x 125 = 1.0 MHz. The
codec will operate at sampling frequencies up to
24 kHz, with the frequency response of the digital filters
being changed proportionally. For this architectural
description, the sampling frequency, f
S
, is assumed to
be 8 kHz, with an oversampling frequency, f
OS
, of
1 MHz, unless otherwise stated.
4.2.2 Analog-to-Digital Path
The analog-to-digital (A/D) conversion signal path (see
Figure 3 on page 5) begins with the analog input driving
the input block. The signal from the input block is then
encoded by a second-order
-
Σ
modulator A/D. The
bulk of the antialiasing filtering is done in the digital
domain in two stages following the
-
Σ
modulator to
give a 16-bit result. The blocks will next be covered in
more detail.
4.2.3 Analog Input Block
The A/D input block operates in two modes: when the
external input gain select (EIGS) pin is low or left
unconnected, the input goes through a preamplifier and
is band limited by a second-order 30 kHz low-pass anti-
aliasing filter (see Figure 4 on page 7). When EIGS is
high, external resistors, Rin and Rfb, are used to set the
gain of an inverting amplifier (see Figure 5 on page 7).
These resistors, in combination with Cin and Cfb, cre-
ate a bandpass antialiasing filter. Note that EIGS is a
digital pin whose input levels are relative to digital
power and ground (V
DD
and V
SS
).
4.2.4 A/D Modulator and Digital Filters
A second-order
-
Σ
modulator quantizes the analog
signal to 1 bit (see Figure 3 on page 5). At the same
time, the resulting quantization noise is shaped such
that most of this noise lies outside of the baseband.
The modulator output is then digitally low-pass filtered
to remove the out-of-band quantization noise. After this
filtering, the output samples are decimated down to the
output sampling frequency. In the CSP1027, the filter-
ing and decimation are completed in two stages. The
first-stage low-pass filter shapes the modulator output
according to the sinc-cubic transfer function:
The output sampling frequency of the sinc-cubic filter is
reduced by a factor of 25 from 1 MHz to 40 kHz. The
sinc-cubic filter places nulls in the frequency response
at multiples of 40 kHz, and removes most of the quanti-
zation noise above 20 kHz so that very little energy is
aliased as a result of the decimation.
The sinc-cubic filter output is then processed by a
seventh-order IIR digital low-pass filter. This filter
removes the out-of-band quantization noise between
3.4 kHz and 20 kHz, compensates for the passband
droop caused by the sinc-cubic decimator, and deci-
mates the sampling frequency by a factor of five from
40 kHz to 8 kHz.
H z
25
1
1
(
z
25
z
1
–
–
–
-1
)
–
)
×
3
=