Intel 2920 Design Handbook page 74

CHAPTER 6
ADVANCED TECHNIQUES
6.0 ADVANCED TECHNIQUES
This chapter includes examples illustrating the
implementation of some special functions with the 2920
Signal Processor. In particular, time variable filters,
pseudo random noise generation, and digital I/O are
discussed.
6.1 Time Variable Filters
In some applications, filters do not remain fixed in their
characteristics, but instead are varied with time. Such
filters may be used for the tracking of time-varying
signals, for synthesis of voice and music, etc. The digital
nature of the 2920 also allows several filters to be varied
together.
To realize a variable filter, the frequency characteristic
controlling parameters (Bl and B2 in Figure 6-1) are
made variables rather than fixed values. While Figure
6-1 represents only one second order filter section, more
complicated variable filters can be made by simul-
taneously varying all stages of a multi-section filter,
such as that shown in Figure 6-2. Although 2920 pro-
grams are sequential in nature, events are essentially
simultaneous if they are completely processed within
one sample interval, e.g., within one pass of a 2920
program.
The coefficients Bl and B2 in Figure 6-1 both control
center frequency, while B2 alone affects bandwidth.
Therefore, if only center frequency needs to be varied,
only Bl need be made a variable.
x - - - - - - - - - - ,
Figure
6-1.
Basic Second Order Section
6-1
For a single stage, the following approximations may be
used:
B2
~ -e
-2n(b/fsl
Bl ~
2e- nb/f s
cos
(2nfo/fs)
where fs is the sampling frequency, fo the filter center
frequency, and b is the filter bandwidth. Note that the
relationship between center frequency and controlling
parameter B2 is nonlinear, following a cosine curve. In
some cases, a nonlinear transformation may be used to
compensate for this nonlinearity (see section 4.6 on
nonlinear transformations).
If only bandwidth is to be controlled, and no shift of
center frequency can be tolerated, then both coefficients
must be made variable, and two variable by variable
multiples must be performed for each variable stage.
More complex filters can be made variable by determin-
ing the relationships needed for each coefficient. If a
single controlling parameter is to be used, the interven-
ing relationships must be used to generate the B 1 and B2
values needed for each stage.
When filter coefficients are varied, the changes can pro-
duce transients within the filter, whose values are also a
function of the instantaneous values in the filter. It may
be desirable to take steps to ensure that coefficients
change slowly to minimize such transients.
Design Example-Consider a filter which must have a
fixed bandwidth of 80 Hz ±5 Hz, while its center fre-
quency is variable from approximately 800 Hz to 1450
Hz. A sample rate of 8000 Hz is to be used.
Using 75
~b~85
Hz for bandwidth gives a range of
possible values for B2:
0.9428~
1 B21
~
0.9354
A value of B2=-0.9375=-(1-1/16) can be realized in two
2920 program steps, and gives a bandwidth of 82 Hz.
With this value for B 1, the range for B 1 can be found as
1.5667~Bl ~0.8101