PSB21473FV13XT Infineon Technologies, PSB21473FV13XT Datasheet - Page 386

• Manufacturer Part Number: PSB21473FV13XT
• Manufacturer: Infineon Technologies

Specifications of PSB21473FV13XT

• Operating Supply Voltage (typ): 3.3V
• Operating Supply Voltage (min): 3.135V
• Operating Supply Voltage (max): 3.6V
• Operating Temp Range: 0C to 70C
• Operating Temperature Classification: Commercial
• Mounting: Surface Mount
• Pin Count: 144
• Lead Free Status / RoHS Status: Compliant

INCA-D
PSB 21473
Digital Signal Processor
speech: voice signals consist of short peaks with high power (bursts). In contrast,
background noise can be regarded approximately stationary from its average power.
Low-Pass Filter 2 provides different time constants for noise (non-detected speech) and
speech. It determines the average of the noise reference level. In case of background
noise the level at the output of LP2 is approximately the level of the input. As in the other
branch an additional offset OFF is added to the signal, the comparator signals noise. At
speech bursts the digital signals arriving at the comparator via the offset branch change
faster than those via the LP2-branch. If the difference exceeds the offset OFF, the
comparator signals speech. Therefore the output of the background noise monitor is a
digital signal indicating speech (1) or noise (0).
A small fade constant (LP2N) enables fast settling of LP2 to the average noise level after
the end of speech recognition. However, a too small time constant for LP2N can cause
rapid charging to such a high level that after recognizing speech the danger of an
unwanted switching back to noise exists. It is recommended to choose a large rising
constant (LP2S) so that speech itself charges the LP2 very slowly. Generally, it is not
recommended to choose an infinite LP2S because then approaching the noise level is
disabled. During continuous speech or tones the LP2 will be charged until the limitation
LP2L is reached. Then the value of LP2 is frozen until a break discharges the LP2. This
limitation permits transmission of continuous tones and “music on hold”.
The offset stage represents the estimated difference between the speech signal and
averaged noise.
Signal Preprocessing
As described in the preceding chapter, the background noise monitor is able to
discriminate between speech and noise. In very short speech pauses e.g. between two
words, however, it changes immediately to non-speech, which is equal to noise.
Therefore a peak detection is required in front of the Noise Monitor.
The main task of the Peak Detector (PD) is to bridge the very short speech pauses during
a monolog so that this time constant has to be long. Furthermore, the speech bursts are
stored so that a sure speech detection is guaranteed. But if no speech is recognized the
noise low-pass LP2 must be charged faster to the average noise level. In addition, the
noise edges are to be smoothed. Therefore two time constants are necessary. As the
peak detector is very sensitive to spikes, the low-pass LP1 filters the incoming signal
containing noise in a way that main spikes are eliminated. Due to the programmable time
constant it is possible to refuse high-energy sibilants and noise edges.
To compress the speech signals in their amplitudes and to ease the detection of speech,
the signals have to be companded logarithmically. Hereby, the speech detector should
not be influenced by the system noise which is always present but should discriminate
between speech and background noise. The limitation of the logarithmic amplifier can be
programmed via the parameter LIM. LIM is related to the maximum PCM level. A signal
exceeding the limitation defined by LIM is getting amplified logarithmically, while very
Data Sheet
386
2003-03-31