TMS320F2812 - FIR Filter

DSP28 - FIR - Filter 14 - 1
Introduction
This chapter looks into one of the most common applications for Digital Signal Processors:
Digital Filters. As we have seen before, there are two basic classes of computing: off line
and real time. This is also valid for digital signal processing. Because we are dealing with the
C28x controller, which is designed for embedded control, calculations are usually done in a real
time environment. For a digital filter, this means that all internal processing of the current state of
a system must be finished before a next input value is sampled. Again, computing time is most
precious! The faster we calculate the algorithm of a digital filter, the more samples we can take
from an input channel. It also means that the frequency of the input signal that is to be processed
depends directly on the efficiency of the controller.
We will start with some mathematical basics of Digital Filters, but we will not go too much into
the theoretical background. To learn more about the mathematics behind Digital Filters, you will
have to join other courses at your university. The Texas Instruments C6000- Teaching CD-ROM
is highly recommended to learn more about the design of digital filters. Beginning with chapter
14 of this CD, you will be introduced to techniques for filter coefficient estimation and to basic
approaches of windowing. Although this C6000 CD is based on the C6000 family the chapters
about Filters and Fourier Transform are also valid for the C28x.
Next we will look into the structure of Finite Impulse Response (FIR)-filters and their
properties. Because of their simplicity and stability, these types of filters are often used in digital
signal processing. We will calculate some examples for low-pass (LPF) and high-pass (HPF)
FIR-filters before we look into a C implementation of such a filter algorithm for the C28x. The
use of IQ-Math C-functions leads to a much faster execution then using standard ANSI-C
implementation. As we have already seen, IQ-Math is a unique feature of the C28x, which is
based on its internal hardware units.
If the computing speed of the IQ-Math implementation is still not fast enough, we can do better!
By switching to Assembly Language coding and by using Texas Instruments Digital Filter
Library (sprc082.zip), we can bring the C28x to its top speed. By means of two examples, we will
have a look into the implementation of a FIR in Assembly Language and into the usage of the C-
callable library function FIR16, provided by TI.
Finally we will perform a laboratory experiment using a FIR-Filter. After we generate a 2 kHz
square wave signal, we will sample it back into the C28x by means of the internal ADC. We will
use the samples to calculate a low-pass function with a 4
th
order FIR-Filter in real-time. Code
Composer Studios Graph Tool will be used to visualize the behaviour of both the unfiltered and
filtered signal in parallel.

C28x FIR - Filter
Module Topics
14 - 2 DSP28 - FIR - Filter
Module Topics
C28x FIR - Filter ......................................................................................................................................14-1
Introduction ...........................................................................................................................................14-1
Module Topics........................................................................................................................................14-2
Basics of Digital Filter Theory ..............................................................................................................14-3
Time Domain Equation .....................................................................................................................14-3
Frequency Domain Equation.............................................................................................................14-5
Finite Impulse Response Filter ..............................................................................................................14-8
Properties of a FIR - Filter.................................................................................................................14-9
FIR Examples.......................................................................................................................................14-10
FIR Implementation in C......................................................................................................................14-14
FIR Implementation in Assembly Language ........................................................................................14-16
Circular Addressing Mode...............................................................................................................14-17
FIR- Filter Code ..............................................................................................................................14-19
Texas Instruments C28x Filter Library................................................................................................14-21
MATLAB Filter Script ....................................................................................................................14-22
FIR16 Library Function...................................................................................................................14-23
Lab 14: FIR Filter for a square-wave signal ...................................................................................14-25
Objective .........................................................................................................................................14-25
Procedure.........................................................................................................................................14-26
Open Files, Create Project File........................................................................................................14-26
Project Build Options ......................................................................................................................14-27
Modify Source Code........................................................................................................................14-27
Build and Load ................................................................................................................................14-29
Test ..................................................................................................................................................14-30
Feedback the Signal into ADC........................................................................................................14-31
Set up ADC sample period (Timer 2)..............................................................................................14-31
Connect T1PWM to ADCIN2 .........................................................................................................14-32
Build, Load and Test .......................................................................................................................14-32
Inspect and Visualize the FIR..........................................................................................................14-33
CCS Graphical Tool ........................................................................................................................14-34

Basics of Digital Filter Theory
Time Domain Equation
The following equation for a Linear Time-Invariant (LTI) system is the starting point to derive a
representation of a Digital Filter:

14 14 - - 2 2
Basics of Digital Filter Theory Basics of Digital Filter Theory
Digital Filter Algorithms are probably the most used numerical Digital Filter Algorithms are probably the most used numerical
operations of a Digital Signal Processor operations of a Digital Signal Processor
Digital Filters are based on the common difference equation for Digital Filters are based on the common difference equation for
Linear Time Linear Time- -Invariant (LTI) Invariant (LTI) systems: systems:
y(n y(n) = output signal ) = output signal
x(n x(n) = input signal ) = input signal
a a
m m
, , b b
k k
= coefficients = coefficients
N = number of coefficients ( order of system) N = number of coefficients ( order of system)
Normalized to a Normalized to a
0 0
= 1 we derive the basic equation in time domain: = 1 we derive the basic equation in time domain:

=
=
1
0
1
0
] [ ] [
N
k
k
N
m
m
k n x b m n y a

=
=
1
0
1
1
] [ ] [ ) (
N
k
N
m
m k
m n y a k n x b n y

It is an equation for discrete input (x (n)) and output (y (n)) signals that are processed by a
transforming system. The properties of the transformation are expressed by coefficients (a
m
and
b
k
). Terms like x[n-k] or y[n-m] are used to express the status of the input and output signal k or
m times before the current sample time n. For causal systems, all samples before time t = 0 are
zero.
These types of equations represent the modification of the output signal y(n) on a time base we
call it Time Domain Representation. The smallest amount of time that is used in these
equations is the sampling period. After the input signal is sampled, the continuous time scale is
replaced by a sequence of numbers. According to Shannons sampling theorem, the sampling
frequency must be at least twice as fast as the highest frequency component of the input signal.
The calculation of the normalized equation for y(n) can be visualized graphically, as shown with
the next slide:


14 14 - - 3 3
Time Domain Chart of a Digital Filter Time Domain Chart of a Digital Filter
x
x
x
x x
+ +
d
d
d
d
x(n)
x(n-1)
x(n-2)
x(n-k)
y(n)
y(n-2)
y(n-1)
y(n-m)
= delay 1 of sample period
gain = 1
d
b
0
b
2
b
1
-a
1
-a
2

This flow can be used to calculate y(n) from the current input sample x(n) and samples of the
input signal, taken one (x(n-1)), two (x(n-2)) or k (x(n-k)) samples before. We call this part the
forward section of the calculation. If we include the status of the output signal delayed by one
period (y(n-1)), two periods (y(n-2)) up to k periods (y(n-k)) into the calculation of the new value
of y(n), we add a feedback section to the computing scheme.
To translate this flowchart into a computer program, we would have to store not only the current
input x(n) and output y(n), but also information about their previous states. How do we code this
in a programming environment? Usually, with two arrays that are big enough to store all the
previous states of x and y. This type of array is usually called a buffer. It functions as delay-
line, hence the term Delay-Line-Buffer.
So what happens when the code has calculated a new value of y(n)? Obviously, y(n) must be
presented to the outside world as a new result of our calculation. Fine, but what is next? To
perform a calculation in real-time, the code must read the next sample from input signal x and
store it at buffer position x(n). But x(n) is still occupied by the sample from one period earlier!
Before we can store the new sample in x(n), the code must move all entries in array x to the next
position, x(n) to x(n-1), x(n-1) to x(n-2) and so on. During this procedure the oldest sample will
be discarded. At the output side, the code has to shift all y values in a similar manner.
Consider the sequence of shift operations! In practice, we have to shift the second oldest first,
followed by the next oldest. If not, we fill the entire buffer with x(n)!
Frequency Domain Equation
The second interpretation of the behavior of a LTI-system is done in terms of frequency the
Frequency Domain - Equation.
The basic operation to transfer a time discrete signal in frequency domain is called Z-
Transformation (ZT). The transformation follows these rules:

14 14 - - 4 4
Transfer Function of a Digital Filter Transfer Function of a Digital Filter
The Z The Z- -Transform of the original input signal Transform of the original input signal x(n x(n) is defined as: ) is defined as:
with with
One property of the Z One property of the Z- -Transform is that the ZT of a time Transform is that the ZT of a time- -shifted shifted
signal is equal to the ZT of the original signal except of a fac signal is equal to the ZT of the original signal except of a factor tor z z
- -k k
: :
{ } ) ( ) ( z X z k n x ZT
k
=
{ }
= =
0
) ( ) ( ) (
n
n
z n x z X n x ZT
j p e z
pT
+ = = and
p = complex angular frequency

The slide shows how the series of discrete input samples x (n) is converted into a complex series
X(z). Instead of representing the signal as sequence number over time we can represent the
signal as sequence complex number over frequency.
One important property of the ZT is that the ZT of a time shifted input signal x (n-k) is identical
to the ZT of the non-time shifted signal x(n), except for a multiplier z
-k
. This feature reduces the
workload to calculate the complex series for X(z) dramatically.
How do we use this ZT to convert the time-domain equation for an LTI-system into its frequency
representation? Well, we have to apply the ZT to both sides of the time-domain equation, shown
at the next slide:
14 14 - - 5 5
Z Z- -Transform is applied to both sides of the time domain equation o Transform is applied to both sides of the time domain equation of f
a Digital Filter : a Digital Filter :
)
`
=
)
`
=
1
0
1
1
] [ ] [ ) (
N
k
k
N
m
m
k n x b ZT m n y a n y ZT

= +
1
0
1
1
) ( ) ( ) (
N
k
k
k
N
m
m
m
z X z b z Y z a z Y
(
=
(
1
0
1
1
) ( 1 ) (
N
k
k
k
N
m
m
m
z b z X z a z Y

The final equation is called Transfer Function of the Digital Filter. It is a frequency domain
representation of the influence that is exerted to an input signal by the Digital Filter.
14 14 - - 6 6
Finally we derive the Transfer Function of a Digital Filter of o Finally we derive the Transfer Function of a Digital Filter of order rder
N in frequency domain: N in frequency domain:
+
= =
1
1
1
0
1
) (
) (
) (
N
m
m
m
N
k
k
k
z a
z b
z X
z Y
z H

For a given spectrum of input frequencies X(z) the transfer function defines the shape of the
output spectrum Y(z). Complex frequency numbers are represented as magnitude and phase per
frequency line.
In a similar way as we have seen for the time-domain flow, we can draft a calculation scheme for
the transfer function in the frequency domain. Each delay-unit in the time domain is replaced by a
multiplication by complex number z
-1
. The basic principle to calculate a new Y(z) is similar to the
time-domain flow, except the complex multiplication. The algorithm still needs two arrays to
store X(z) and Y(z), except that they have to now use complex numbers with a real and imaginary
part for each number. The Frequency Domain Calculation of the frequency response of a system
is normally used to analyse an incoming signal for its frequency components.

14 14 - - 7 7
Frequency Domain Flow of a Digital Filter Frequency Domain Flow of a Digital Filter
x
x
x
x x
+ +
d
d
d
d
X(z) Y(z)
= z
-1
d
b
0
b
2
b
1
-a
1
-a
2
Multiply by z
-1
in Frequency Domain =
Unit delay by one sample period in Time Domain

Finite Impulse Response Filter
If a Digital Filter does not have any feedback components (all a
m
= 0), we call this system a
Finite Impulse Response (FIR). It can be shown that the response of such a system to a single
input impulse will eventually vanish.
14 14 - - 8 8
Finite Impulse Response (FIR) Finite Impulse Response (FIR) - - Filter Filter
If all feedback coefficients a If all feedback coefficients a
m m
are equal to zero we derive the are equal to zero we derive the
equation system for a Finite Impulse Response (FIR) equation system for a Finite Impulse Response (FIR) Filter: Filter:
and: and:
= =
1
0
) (
) (
) (
N
k
k
k
z b
z X
z Y
z H
=
=
1
0
] [ ) (
N
k
k
k n x b n y
Frequency Domain
Time Domain

If feedback components exist, the system is called an Infinite Response Filter (IIR).
14 14 - - 9 9
Infinite Impulse Response (IIR) Infinite Impulse Response (IIR) - - Filter Filter
If coefficients a If coefficients a
m m
are present we call this type of filter Infinite are present we call this type of filter Infinite
Impulse Response (IIR). In this case the equation with feedback Impulse Response (IIR). In this case the equation with feedback
part must be used for the filter calculation. part must be used for the filter calculation.
Obviously the feedback terms a Obviously the feedback terms a
m m
*y(n *y(n- -m) deliver some amount of m) deliver some amount of
energy back into the calculation. energy back into the calculation.
Under particular circumstances this feedback system will respond Under particular circumstances this feedback system will respond
to a finite input impulse infinite in time to a finite input impulse infinite in time hence the name. hence the name.
+
= =
1
1
1
0
1
) (
) (
) (
N
m
m
m
N
k
k
k
z a
z b
z X
z Y
z H

=
=
1
0
1
1
] [ ] [ ) (
N
k
N
m
m k
m n y a k n x b n y
IIR IIR Filter Filter

Properties of a FIR - Filter

14 14 - - 10 10
Simple FIR Diagram Simple FIR Diagram
y(n) = b0 y(n) = b0 x(n) x(n) + + b1 b1 x(n x(n 1) 1) + + b2 b2 x(n x(n 2) 2)

z
1 z z
1 1
z
1 z z
1 1

+ +
X0 X0 X1 X1 X2 X2
x x
in in
y y
out out
b0 b0 b1 b1 b2 b2

One typical property of the FIR-Transfer Function is its periodicity by 2:
14 14 - - 11 11
Properties of a FIR Filter Properties of a FIR Filter
Replacing z by its original definition: Replacing z by its original definition:
disregarding disregarding ( (loss loss less filter less filter) ) and normalizing to T=1: and normalizing to T=1:
Since e Since e
- -j2 j2 k k
= 1: = 1:
FIR filters have a periodic frequency response of 2 FIR filters have a periodic frequency response of 2 ! !
We need to limit the spectrum! We need to limit the spectrum!
T j pT
e e z
) ( +
= =
=
= =
1
0
) ( ) (
N
k
jk
k
j
e z
e b e H z H
j

=
+ +
= = =
1
0
1
0
2 ) 2 ( ) 2 (
) ( ) (
N
k
N
k
j k j jk
k
jk
k
j
e H e e b e b e H

FIR Examples
FIR Examples
Let us calculate the frequency response of the following filter system. As the diagram shows it
lacks feedback components it is of FIR-type. It is a first-order filter, the filter coefficients are
b
0
= +0.5
b
1
= +0.5
What will the magnitude of the output signal look like? Which frequency components will pass
the filter, which one will be damped?

14 14 - - 12 12
FIR FIR Example 1 Example 1
) 1 ( 5 . 0 ) (
) 1 ( 5 . 0 ) (
) (
2
1
1
1
0
0
A
f
f j
e j H
z z H
z b z b z H
+ =
+ =
+ =

f
1
T f; 2 ; j p ;
A
= = + = =
pT
e z
Frequency Response ?
Type of Filter ?
x(n-1)
Z
-1
x(n)
y(n)
0.5 0.5
b
0
= 0.5 b
1
= 0.5
+
f
A
= sampling frequency
2 2
Im Re ) (
)) 2 sin( ) 2 cos( 1 ( 5 . 0 ) (
+ =
+ =

j H
f
f
j
f
f
j H
A A

If we calculate the magnitude for frequencies f = 0, f = f
A
/8, f= f
A
/4, f =f
A
*3/4 and f = f
A
/2 we get:

0 and 0.382 0.707, 0.92, , 1 ) ( = j H

FIR Examples
The graph is shown next. Low frequencies are amplified by 1, the more we approach 0.5*f
A
, the
more the magnitude is damped and finally reaches 0. This is a low-pass filter!
14 14 - - 13 13
FIR FIR Example 1 (cont.) Example 1 (cont.)
2 2
Im Re ) (
)) 2 sin( ) 2 cos( 1 ( 5 . 0 ) (
+ =
+ =

j H
f
f
j
f
f
j H
A A
f/f
A
|H(j)|
1
0.5
0.5 0.25
Filter Filter Type: Low Type: Low- -pass pass
0 H(j | : f * 0.5 f 5
382 . 0 H(j | : f * 0.375 f 4
707 . 0 H(j | : f * 0.25 f 3
92 . 0 H(j | : f * 0.125 f
H(j | : 0 f . 1
A
A
A
A
= , ) = .
= , ) = .
= , ) = .
= , ) = 2.
1 = , ) =

What happens, if the input frequency f goes beyond f
A
/2, violating the SHANNON-theorem?
The magnitude rises from 0 to 1, introducing false (aliased) frequency components!
14 14 - - 14 14
2 2
Im Re ) (
)) 2 sin( ) 2 cos( 1 ( 5 . 0 ) (
+ =
+ =

j H
f
f
j
f
f
j H
A A
f/f
A
|H(j)|
1
0.5
0.5 0.25
Aliasing, if f > 0.5 f
A
Input Frequencies must be
limited to 0.5*f
A
by an
additional Low Pass input filter
1 H(j | : f f 4
92 . 0 H(j | : f * 0.875 f 3
707 . 0 H(j | : f * 0.75 f
382 . 0 H(j | : f * 0.625 f . 1
A
A
A
A
= , ) = .
= , ) = .
= , ) = 2.
= , ) =

FIR Examples
The solution to suppress any frequencies that violate the sampling theorem is to introduce an anti-
aliasing filter before the samples are taken. This way, no frequency component beyond f
A
/2 will
disturb the digital processing. The anti-alias filter is an analogue low-pass filter.
14 14 - - 15 15
FIR FIR y[n] y[n]
x[n] x[n]
ADC ADC
Analogue Analogue
Anti Anti- -Aliasing Aliasing
x(t) x(t)
Solution: Use an anti Solution: Use an anti- -aliasing filter at aliasing filter at
input to limit all input frequencies to f input to limit all input frequencies to f
A A
/2. /2.

In the next example for an FIR-Filter of order 1 we only change coefficient b
1
from +0.5 into -
0.5.
14 14 - - 16 16
Frequency Response ?
Type of Filter ?
x(n-1)
Z
-1
x(n)
y(n)
0.5 -0.5
b
0
= 0.5 b
1
= - 0.5
+
Note : We only changed b
1
from +0.5 to - -0.5 0.5 !
1
0.5
f/f
A
0.5 0.25
|H(j)|
Filter Type: High Pass Filter Type: High Pass

FIR Examples
What happens? The digital FIR-filter now damps frequencies around f=0 and amplifies an input
frequency of f = f
A
/2 by 1. The filter type has changed from low-pass into high-pass. By
modifying coefficients we can change the behaviour of the filtering system. Compare this feature
with an analogue filter, where you would have to change resistors or capacitors with a soldering
iron.
If the modification of filter coefficients is done in real-time by the controller code itself, we call
this an Adaptive Filter.
We can also use digital filters to non-technical topics. The next example of a 2
nd
order FIR-Filter
shows how the average stock price per week is calculated using the moving average
calculation:

14 14 - - 17 17
10
20
30
40
$
time
mon tue wed thu fri sat sun
Input
12
40
Assume no previous inputs
X(0) = 20; X(-1) = 0; X(-2) = 0
y(0) = 0.25*x(0) + 0.5*x(-1) + 0.25*x(-2) = 5
y(1) = 0.25*20 + 0.5*20 + 0.25*0 = 15
y(2) = 0.25*20 + 0.5*20 + 0.25*20 = 20
y(3) = 0.25* + 0.5* + 0.25* =
y(5) = 0.25*20 + 0.5*40 + 0.25*12 = 28
y(4) = 0.25* + 0.5* + 0.25* =
y(6) = 0.25*20 + 0.5*20 + 0.25*40 = 25
Moving average calculation
10
20
30
40
$
time
mon tue wed thu fri sat sun
Output
x(n-1)
x(n-2)
Z
-1
Z
-1
x(n)
y(n)
0.25 0.5 0.25
b
0
= 0.25 b
1
= 0.5 b
2
= 0.25
And let

The input is the daily stock price; the output is the moving average. Due to the nature of causally
determined systems, we assume that we do not have previous inputs when we start on Monday
(time = zero). Of course a broker would know the prices from last week and would request us to
add them to our calculation technically speaking, the stock market is a non causal system,
probably practically too.
Anyway, if our calculation advances to Wednesday everybody will be pleased with our results of
a moving average.
Lets go back to more technical issues and leave the stock market brokers without further DSP-
support.
FIR Implementation in C
Before we proceed to the implementation of a FIR filter using the C28x, let us recall one
important step between the periodic calculations. The sample buffer must be prepared to include
the latest sample value at the start of the buffer by shifting all elements by one position. Quite
often this is done in ascending order:

14 14 - - 18 18
Which movement should you perform first? Which movement should you perform first?
X X
0 0
X X
1 1
X X
2 2
Periodic FIR Periodic FIR - - Filter Calculation Filter Calculation
LOWER LOWER
MEMORY MEMORY
ADDRESS ADDRESS
HIGHER HIGHER
MEMORY MEMORY
ADDRESS ADDRESS
old X old X
0 0
becomes new X becomes new X
1 1
X X
0 0
X X
1 1
X X
2 2
old X old X
1 1
becomes new X becomes new X
2 2
before we can before we can
calculate the FIR a calculate the FIR a
second time: second time:
X X
0 0
is used for latest sample is used for latest sample
Delay Line update Delay Line update

The following code is taken from a Texas Instruments example to implement FIR-Code for the
C28x in IQ-Math-format. You have seen that this fixed-point maths is much better adapted to the
C28x than any standard ANSI-C solution in terms of computing power.
The name of the function IQssFIR relates to IQ single source FIR only one stream of
input numbers is computed. Input parameters are two pointers to the array of input samples and to
the coefficients and the number of taps thats order minus 1.
The processing is based on the data type _iq which is defined in IQmathLib.h. The return
parameter is the new output value y(k), also in _iq-format.
When you inspect the code, you will notice that it operates from back to front, placing the two
pointers to the end of the two buffers and post-decrementing them after any single multiplication.
The shift operation on the delay line is done immediately after the current tap has been processed
with the help of a temporary pointer xold.
The accumulation is done with a simple add-operation using local variable y.
C sourcecode FIR - IQmath
14 14 - - 19 19
FIR Filter Implementation in C FIR Filter Implementation in C
/***************************************************************
* Function: IQssfir()
* Description: IQmath n-tap single-sample FIR filter.
*
* y(k) = a(0)*x(k) + a(1)*x(k-1) + ... + a(n-1)*x(k-n+1)
*
* DSP: TMS320F2812, TMS320F2811, TMS320F2810
* Include files: DSP281x_Device.h, IQmathLib.h
* Function Prototype: _iq IQssfir(_iq*, _iq*, Uint16)
* Useage: y = IQssfir(x, a, n);
* Input Parameters: x = pointer to array of input samples
* a = pointer to array of coefficients
* n = number of coefficients
* Return Value: y = result
* Notes:
* 1) This is just a simple filter example, and completely
* un-optimized. The goal with the code was clarity and
* simplicity,
* not efficiency.
* 2) The filtering is done from last tap to first tap. This
* allows
* more efficient delay chain updating.
*****************************************************************/

14 14 - - 20 20
FIR Filter Implementation in C FIR Filter Implementation in C
_iq IQssfir(_iq *x, _iq *a, Uint16 n)
{
Uint16 i; // general purpose
_iq y; // result
_iq *xold; // delay line pointer
/*** Setup the pointers ***/
a = a + (n-1); // a points to last coefficient
x = x + (n-1); // x points to last buffer element
xold = x; // temporary buffer
/*** Last tap has no delay line update ***/
y = _IQmpy(*a--, *x--);
/*** Do the other taps from end to beginning ***/
for(i=0; i<n-1; i++)
{
y = y + _IQmpy(*a--, *x); // filter tap
*xold-- = *x--; // delay line update
}
return(y);
}

FIR Implementation in Assembly Language
Although the previous example of a C based implementation of a FIR algorithm used IQ-Math
function calls to calculate the next value for the output, there is still headroom to optimize the FIR
code for the C28x. As we have seen from the C example, basic mathematical operations in this
algorithm are multiply instructions for each coefficient and sample in the delay chain and add
operations to sum the partial products. To prepare the next filter calculation cycle, all sample
values have been shifted by one position after they have been processed.
DSPs have a unique group of assembly instructions that take advantage of the parallel hardware
units: Arithmetic Logic Unit (ALU) for the sum operation and Hardware Multiplier (MUL)
- for the multiplication. Thanks to the Harvard Architecture of DSPs, two operands can be read
simultaneously one from the data bus and the other from the program-bus.
The assembly language instruction set supports these types of operations in single clock cycle
with the Multiply and Accumulate (MAC) instruction. In case of the C28x, two groups of
assembly instructions are available:
MAC for 16-bit and 32-bit operands
DMAC for 16-bit operands
The DMAC Dual Mac instruction takes advantage of the 32-bit width of the internal busses
and processes four 16-bit operands (two coefficients and two samples) in a single cycle.

14 14 - - 21 21
FIR Filter Implementation in ASM FIR Filter Implementation in ASM
FIR Filter Optimization:
Previous C solution is a generic one, coded in standard
ANSI-C, can be compiled for every microprocessor or
embedded microcomputer. Works well.
But is not optimized for Digital Signal Processors like the
C28x. In case more computing power for the real time
calculation of a FIR is needed, one can take advantage of
internal parallel hardware resources of a DSP.
ASM-coding of a FIR allows to reduce the number of clock
cycles needed to calculate one loop of the FIR algorithm.
A new Addressing Mode is used to avoid the shift operations
of the delay-line for input samples: Circular Addressing
Mode
Describe the Circular Addressing function
Implement FIR filters using Circular Addressing Mode

Circular Addressing Mode
Knowing that MAC or DMAC will accelerate FIR-code the last portion to be optimized is the
shift operation of samples, after they have been processed. Assembly language addressing modes
of operands include one particular mode that is used to avoid these shift operations altogether. We
wont dive too deep into assembly language programming yet, but to explain this addressing
mode, let us take an example.
14 14 - - 22 22
Circular Addressing Usage Circular Addressing Usage

+
Xin = X0 X1 X2
Yout = A0*Xin + A1X1 + A2X2
A0 A1 A2
x[n-2]
x(n-2)
x[n-1]
x(n-1)
x[n]
x(n)
X2 X1 X0
(T = 4)
X2 X1 X0
(T = 2)
Delay
z
1
z
1
x(0) x(1) x(2) x(3) x(4)
Time (T) Samples coming in: Xin
MAC P,*AR6%++,*XAR7++ Sum Of Product + Delay
Sum Of Product
MAC P,*XAR6++,*XAR7++
(T = 3)
? ? ?
On recursion,
X2 = X1
X1 = X0
X0 = new Xin
Finite Impulse Response (FIR) Filter
X0
X1
X2
Delay line
*AR6%++
(T = 3)
X2 X1 X0

The new addressing mode is called Circular Addressing Mode and it is coded with the
percentage (%) sign in front of the pointer register name:
The instruction
MAC P,*XAR6++,*XAR7++
Adds the previous product (stored in P) to the pre-accumulated sum in register ACC
Multiplies the data memory operand, pointed to by XAR6, by the program memory
operand, pointed to by XAR7
Post-increments the two pointers XAR6 and XAR7.

If we introduce a new syntax:
MAC P,*AR6%++,*XAR7++
The first pointer XAR6, which points to the sample array in data memory, is used in a circular
fashion. Once the pointer has reached the end of the array it will store the next value at start of the
buffer. We call this Circular Buffer.
The next slide explains this with a six element buffer, which is used to store the name of month.
If the last space in buffer was used, the month July will replace the oldest entry January

14 14 - - 23 23
JAN
FEB
MAR
APR
MAY
JUN
JUL
FEB
MAR
APR
MAY
JUN
Delay Line with a Circular Buffer Delay Line with a Circular Buffer
*AR6%++
*AR6%++
JUL
@input
AUG
@input
JAN JUN
MAR APR
FEB MAY
oldest
JUL JUN
MAR APR
FEB MAY
oldest
JUL JUN
MAR APR
AUG MAY
oldest
*AR6%++
JAN
FEB
MAR
APR
MAY
JUN
Linear Memory storage

Before we use this circular addressing mode, register XAR7 must be aligned to point to the first
element of the coefficient array.
XAR7 must be initialized to point to the start of the circular buffer. This start address of the
circular buffer must be aligned to a 256-word boundary (8 LSBs = 0000 0000), which is usually
done with the help of a linker command file instruction (see next slide).
The lower 8 bits of register XAR1 are used to specify the size of the circular buffer. This is an
implicit usage of register XAR1 by the circular addressing mode; it is not shown in the assembly
code!


14 14 - - 24 24
Circular Addressing Hardware Circular Addressing Hardware
AR1 Low is set to buffer size - 1
All 32 bits of XAR6 are used
AR1 Low (16)
end of buffer ---- ----
start of buffer
(align on 256 word boundary)
AAAA AAAA AAAA AAAA
SECTIONS
{ D_LINE: align(256) { } > RAM PAGE 1
. . .
}
LINKER.CMD
circular
buffer
range
Element 0
Element N-1
Buffer Size N
XAR6 (32) access pointer
AAAA AAAA AAAA AAAA xxxx xxxx
N-1
0000 0000
MAC P,*AR6%++,*XAR7++

FIR- Filter Code
The next slide is an assembly language implementation of a 3
rd
order FIR filter (4 taps). It can be
adapted to higher orders by changing the constant TAPS and the number of coefficients in
array tbl. All operands are 16-bit wide in Q15-format.
The directive .usect defines un-initialized data memory of length TAPS and assigns it to
symbol xn. The linker command file will connect this section D_LINE to physical memory.
The directive .data defines initialized code memory and assigns it to symbol tbl. The four
coefficients in this array are in I1Q15-format.
The directive .text opens the code-section for assembly instructions. After setting core op-mode
bits (SXM, OVM, PM) the two pointers XAR6 and XAR7 are initialized to point to symbol xn
and tbl respective. AR1 is loaded with the buffer size.
Next, a new sample is read from the external ADC at data memory address 0:adc and stored at
first place in the circular buffer. We also could have used the internal ADC. The %++ operator
will set the circular buffer pointer to its next element.
After clearing register ACC, P and OV (ZAPA) the following two instructions will do the
entire filter work. The repeat instruction (RPT) tells the C28x to repeat the following
instruction #number plus 1 times, in our case twice. The instruction DMAC will perform two
16x16-bit MAC operations, reading and processing two members of xn and tbl per cycle. The
3
rd
order FIR filter will be calculated in two clock cycles.
Finally the two halves of the result are added using the instruction ADDL ACC:P and the new
result is loaded in I1Q15-format to an external DAC at address 0:dac.

14 14 - - 25 25
FIR Filter FIR Filter Dual MAC Dual MAC - - Operation Operation
TAPS .set 4 ; FIR Order +1
xn .usect D_LINE,TAPS ; sample array in I1Q15
.text
FIR: SETC SXM ; 2s complement math
CLRC OVM ; no clipping mode
SPM 1 ; fractional math
MOVL XAR7,#tbl ; coefficient pointer
MOVL XAR6,#xn ; circular buffer pointer
MOV AR1,#TAPS-1 ; buffer offset
MOV *XAR6%++,*(0:adc) ; get new sample ( x(n) )
ZAPA ; clear ACC,P,OVC
RPT #(TAPS/2)-1 ; RPT next instr.(#+1)times
|| DMAC ACC:P,*XAR6%++,*XAR7++ ; multiply & accumulate 2pairs
ADDL ACC:P ; add even & odd pair-sums
MOV *(0:dac),AH ; update output ( y(n) )
RET
.data ; FIR Coeffs in I1Q15
tbl .word 32768*707/1000 ; 0.707
.word 32768*123/1000 ; 0.123
.word 32768*(-175)/1000 ; -0.175
.word 32768*345/1000 ; 0.345

14 14 - - 26 26
Circular Addressing Summary Circular Addressing Summary
Buffer Size
Up to 256 words
Break larger arrays into <= 256 word blocks.
Buffer Alignment
Always align on 256-word boundaries, regardless of size. Unused
space can be used for other purposes.
Let the linker assign addresses. Link largest blocks first.
Usage
XAR6 is the only circular pointer.
AR1 must be set to the size minus one (0 - 255).
Pointer update is post-increment by one (*XAR6%++) .
32-bit access causes post-increment by two. Make sure XAR6
and AR1 are even to avoid jumping past end of buffer.

Texas Instruments C28x Filter Library
So for some types of applications, it seems to make sense to code in Assembly Language. But, as
usual, there is no need to re-invent the wheel. Better than developing your own code is use a
library. Texas Instruments is offering a variety of libraries for free, one of them us dedicated to
Digital Filters.

14 14 - - 27 27
Texas Instruments C28x Filter Library Texas Instruments C28x Filter Library
MATLAB script to calculate Filter Coefficients for FIR and IIR,
includes windowing
Filter Modules:
FIR16: 16-Bit FIR-Filter
IIR5BIQ16: Cascaded IIR-Filter (16bit-biquad)
IIR5BIQ32: Cascaded IIR-Filter (32bit-biquad)
C-callable Assembly (CcA) Functions
Adapted to internal Hardware features of the C28x
Uses the Dual MAC instruction
Interface according to ANSI-C standard
Available from TI-web as document sprc082.zip

To make it easier to use this library in a C-based program environment, all library functions are
equipped with an interface structure. Thus any library function can be called like an ordinary C
subroutine.
An important step in designing a Digital Filter is to calculate Filter Coefficients. This task
involves a lot of theoretical background. Without this knowledge, you wont be able to profile the
set of coefficients for a given transfer function. As recommended earlier, join additional courses
at your university to understand the math behind Digital Signal Processing.
The library package includes also a MATLAB script to calculate filter coefficients, including
windowing techniques.

MATLAB Filter Script
The MATLAB filter script allows you to initialize essential parts of the transfer function, like
sampling frequency, filter type, order, type of window and corner frequency.
14 14 - - 28 28
MATLAB Filter Script MATLAB Filter Script
FIR Filter Design Example: Low-pass Filter of Order 50
LPF Specification:
FIR Filter Order : 50
Type of Window : Hamming
Sampling frequency : 20KHz
Filter Corner Frequency : 3000Hz
ezFIR FILTER DESIGN SCRIPT
Input FIR Filter order(EVEN for BS and HP Filter) : 50
Low Pass : 1
High Pass : 2
Band Pass : 3
Band Stop : 4
Select Any one of the above Response : 1
Hamming : 1
Hanning : 2
Bartlett : 3
Blackman : 4
Select Any one of the above window : 1
Enter the Sampling frequency : 20000
Enter the corner frequency(Fc) : 3000
Enter the name of the file for coeff storage : lpf50.dat

The output of the MATLAB calculation is a list of coefficients and a graph of magnitude and
phase of the filter response.
14 14 - - 29 29
MATLAB Filter Script MATLAB Filter Script
#define FIR16_COEFF {\
9839,-2219809,-1436900,853008,3340889,3668111,-896,\
-5963392,-8977456,-3669326,8585216,18152991,13041193,\
-8257663,-30867258,-31522540,131,45285320,64028535,\
25231269,-58654721,-124846025,-94830542,68157453,\
320667626,551550942}
MATLAB Output File for Filter Coefficients:

FIR16 Library Function
FIR16 is one of the library functions of sprc082. It processes a single stream of input samples in
Q15-format into a new output value of the same format. One instance of this function occupies 52
words of code memory and processing time is 350 ns with a 150 MHz C28x.

14 14 - - 30 30
FIR16 Library Function FIR16 Library Function
FIR16
input output

The format of the FIR16 functions object structure is shown next. Interface parameters are
Input:
2 pointers to coefficients and samples
Order of filter
1 pointer to an initialize function of the filter
1 pointer to the calculation function
Output:
1 new filter output value.
To guarantee the ability to operate with more than 1 instance of the filter, you should call the
function by its function parameters (init, calc) only.

14 14 - - 31 31
FIR16 Library Function FIR16 Library Function
Object Definition:
typedef struct {
long *coeff_ptr; /* Pointer to Filter coeffs */
long *dbuffer_ptr; /* Delay buffer pointer */
int cbindex; /* Circular Buffer Index */
int order; /* Order of the Filter */
int input; /* Latest Input sample */
int output; /* Filter Output */
void (*init)(void *); /* Pointer to Init function */
void (*calc)(void *); /* Pointer to calc function */
}FIR16;
coeff_ptr: Pointer to the Filter coefficient array.
dbuffer_ptr: Pointer to the Delay buffer.
cbindex: Circular buffer index, computed internally by initialization function
based on the order of the filter.
order: Order of the Filter. Q0-Format, range 1 255
input: Latest input sample to the Filter. Q15-Format (8000-7FFF)
output: Filter output value. Q15-Format (8000-7FFF)

The next slide is an example for the usage of the filter function. An instance of FIR16, called
lpf has been defined in a DATA_SECTION firfilt. All accesses to parameters and functions
are made by this instance.
14 14 - - 32 32
FIR16 Library Usage Example FIR16 Library Usage Example
#define FIR_ORDER 50 /* Filter Order */
#pragma DATA_SECTION(lpf, "firfilt");
FIR16 lpf = FIR16_DEFAULTS;
#pragma DATA_SECTION(dbuffer,"firldb");
long dbuffer[(FIR_ORDER+2)/2];
const long coeff[(FIR_ORDER+2)/2]= FIR16_LPF50;
main()
{
lpf.dbuffer_ptr=dbuffer;
lpf.coeff_ptr=(long *)coeff;
lpf.order=FIR_ORDER;
lpf.init(&lpf);
}
void interrupt isr20khz()
{
lpf.input=xn;
lpf.calc(&lpf);
yn=lpf.output;
}

Lab 14: FIR Filter for a square-wave signal
Objective
14 14 - - 33 33
Lab 14: LP Lab 14: LP - -Filter of a square wave Filter of a square wave
Objective:
Generate a square wave signal of 2 KHz at EVA-T1PWM
Asymmetric PWM , duty cycle 50%
Use T1-Compare Interrupt Service to serve the watchdog
Wire - Connect T1PWM to ADC-input ADCIN2
Sample the square wave signal at 50KHz
Sample period generated by EVA-Timer 2
Store samples in buffer AdcBuf
Filter the input samples with a FIR Low pass 4
th
order
Store filtered output samples in buffer AdcBufFiltered
Visualize AdcBuf and AdcBufFiltered graphically by Code
Composer Studios Graph Tool

The lab experiment consists of four parts:
First we will generate a 2 kHz square wave signal at output T1PWM. At the
Zwickau adapter board this signal can be connected to a loudspeaker (JP3
closed). Or, use a scope to visualize the signal.
Second we will feedback this signal back into one channel of the internal ADC
and store the digital samples in a data memory buffer AdcBuf.
Next, we call a Low-Pass Filter of FIR-Type with order 4 to wave-shape the
signal edges. The coefficients were calculated with MATLAB as:
o 1/16, 4/16, 6/16, 4/16 and 1/16
o The sampling frequency is set to 50 kHz
The filtered numbers will be stored in AdcBufFiltered
Finally we will use Code Composer Studios graphical tool to visualize the
contents of AdcBuf and AdcBufFiltered. We will take advantage of the real
time debug capabilities to display the data without interrupting or delaying the
C28x while it is running!

Procedure
Open Files, Create Project File
1. Create a new project, called Lab14.pjt in E:\C281x\Labs.
2. Open the file Lab5.c from E:\C281x\Labs\Lab5 and save it as Lab14.c in
E:\C281x\Labs\Lab14.
3. Add the source code file to your project:
Lab14.c
4. From C:\tidcs\c28\dsp281x\v100\DSP281x_headers\source add:
DSP281x_GlobalVariableDefs.c
From C:\tidcs\c28\dsp281x\v100\DSP281x_common\cmd add:
F2812_EzDSP_RAM_lnk.cmd
From C:\tidcs\c28\dsp281x\v100\DSP281x_headers\cmd add:
F2812_Headers_nonBIOS.cmd
From C:\tidcs\c28\dsp281x\v100\DSP281x_common\source add to project:
DSP281x_PieCtrl.c
DSP281x_PieVect.c
DSP281x_DefaultIsr.c
From C:\ti\c2000\cgtoolslib add:
rts2800_ml.lib
Project Build Options
5. Setup the search path to include the peripheral register header files. Click:
Project Build Options
Select the Compiler tab. In the preprocessor Category, find the Include Search
Path (-i) box and enter:
C:\tidcs\C28\dsp281x\v100\DSP281x_headers\include;
..\include; C:\tidcs\C28\IQmath\cIQmath\include

6. Setup the stack size: Inside Build Options select the Linker tab and enter in the Stack
Size (-stack) box:
400

Close the Build Options Menu by Clicking <OK>.
Modify Source Code
7. Open Lab14.c to edit: double click on Lab14.c inside the project window. First we
have to remove the parts of the code that we do not need any longer. We will not use
the CPU core timer 0 in this exercise; therefore we do not need the prototype of
interrupt service routine cpu_timer0_isr(). Instead, we need a new ISR for EVA-
Timer1-Compare-Interrupt. Add a new prototype interrupt function: interrupt void
T1_Compare_isr(void).
8. We do not need the variables i,time_stamp and frequency[8] from Lab5 - delete
their definition lines at the beginning of the function main.
9. Next, modify the re-map lines for the PIE entry. Instead of PieVectTable.TINT0 =
& cpu_timer_isr we need to re-map:
PieVectTable.T1CINT = &T1_Compare_isr;
10. Delete the next two function calls: InitCpuTimers(); and
ConfigCpuTimer(&CpuTimer0, 150, 50000); and add an instruction to enable the
EVA-Timer1-Compare interrupt. According to the interrupt chapter this source is
wired to PIE-group 2 , interrupt 5:
PieCtrlRegs.PIEIER2.bit.INTx5 = 1;
Also modify the set up for register IER into:
IER | = 2;
11. Next we have to initialize the Event Manager Timer 1 to produce a PWM signal. This
involves the registers GPTCONA, T1CON, T1CMPR and T1PR.
For register GPTCONA it is recommended to use the bit-member of this
predefined union to set bit TCMPOE to 1 and bit field T1PIN to active low.
For register T1CON set
The TMODE-field to counting up mode;
Field TPS to divide by 1;
Bit TENABLE to disable timer;
Field TCLKS to internal clock
Field TCLD to reload on underflow
Bit TECMPR to enable compare operation
12. Remove the 3 lines before the while(1)-loop in main:
CpuTimer0Regs.TCR.bit.TSS = 0;
i = 0;
time_stamp = 0;
and add 4 new lines to initialise T1PR, T1CMPR, to enable GP Timer1 Compare
interrupt and to start GP Timer 1:
EvaRegs.T1PR = 37500;
EvaRegs.T1CMPR = EvaRegs.T1PR/2;
EvaRegs.EVAIMRA.bit.T1CINT = 1;
EvaRegs.T1CON.bit.TENABLE = 1;
What is this number 37500 for? Well, it defines the length of a PWM period:
HISCP TPS PR T
f
f
T
CPU
PWM

=
1
1

with TPS
T1
=1, HISCP = 2, f
CPU
= 150MHz and a desired f
PWM
= 2 kHz
we derive: T1PR = 37500!
T1CMPR is preloaded with half of T1PR. Whys that? Well, in general T1CMPR
defines the width of the PWM-pulse. Our start-up value obviously defines a pulse
width of 50%.
13. Modify the endless while(1) loop of main! We will perform all activities out of GP
Timer 1 Compare Interrupt Service. Therefore we can delete almost all lines of this
main background loop, we only have to keep the watchdog service:
while(1)
{
EALLOW;
SysCtrlRegs.WDKEY = 0xAA;
EDIS;
}
14. Rename the interrupt service routine cpu_timer0_isr into T1_Compare_isr.
Remove the line CpuTimer0.InterruptCount++; and replace the last line of this
routine by:
PieCtrlRegs.PIEACK.all = PIEACK_GROUP2;
Before this line add another one to acknowledge the GP Timer 1 Compare Interrupt
Service is done. Remember how? The Event Manager has 3 interrupt flag registers
EVAIFRA,EVAIFRB and EVAIFRC. We have to clear the T1CINT bit (done
by setting of the bit):
EvaRegs.EVAIFRA.bit.T1CINT = 1;
Build and Load
15. Click the Rebuild All button or perform:

Project Build

and watch the tools run in the build window. If you get syntax errors or warnings
debug as necessary.

16. Load the output file down to the DSP Click:

File Load Program and choose the desired output file.
Test

17. Reset the DSP by clicking on:

Debug Reset CPU followed by
Debug Restart and
Debug Go main.

18. When you now run the code the DSP should generate a 2 kHz PWM signal with a
duty cycle of 50% on T1PWM. If you have an oscilloscope you can use jumper JP7
(in front of the loudspeaker) of the Zwickau Adapter board to measure the signal.
If your laboratory cant provide a scope, you can set a breakpoint into the interrupt
service routine of T1 Compare at line PieCtrlRegs.PIEACK.all =
PIEACK_GROUP2; Verify that your breakpoint is hit periodically, that register
T1PR holds 37500 and register T1CMPR is initialized with 18750. Use the Watch
Window to do so.
Do not continue with the next steps until this point is reached successfully! Instead,
go back and try to find out what went wrong during the modification of your source
code.

End of Lab 14 Part 1
Feedback the Signal into ADC
19. Three files have been provided to this lab to add the ADC functionality. Add the two
files Adc.c, Adc_isr.c and filter.c to your project.
20. In function InitSystem of Lab14.c enable the ADC-clock:
SysCtrlRegs.PCLKCR.bit.ADCENCLK = 1;
21. In main, just after the call of InitPieVectTable() add a call to initialize the ADC:
InitAdc();
This function will setup the ADC to one conversion per trigger. ADCIN2 will be
converted by SEQ1 out of Event Manager A trigger. An interrupt will be requested
with every end of sequence. Inspect the code of InitAdc().
22. Next, we have to connect the ADC interrupt to a new function:
ADC_FIR_INT_ISR(). This function is defined in the new source code file
Adc_isr. All we have to do is to replace the entry in the PieVectTable by this new
address. Look in main and locate the line, where we already overload the
PieVectTable with T1_Compare_isr. Add a new line:
PieVectTable.ADCINT = &ADC_FIR_INT_ISR;
The new function ADC_FIR_INT_ISR is not declared yet in Lab14.c. Therefore
we have to add a new prototype statement at the beginning of Lab14.c:
interrupt void ADC_FIR_INT_ISR(void);
Register IER must be modified to enable INT1 (ADC) and INT2 (T1-Compare):
IER | = 3;
Set up ADC sample period (Timer 2)
23. EVA-Timer 2 will be used to generate the sample period for the ADC. Each period
event of T2 will trigger a start of an ADC sequence automatically, if we enable this
option:
EvaRegs.GPTCONA.bit.T2TOADC = 2;
Add this line in front of the while(1)-loop of main.
Before the code enters the while(1)-loop we have to initialize EVA-Timer2 to
produce a sample period of 50 kHz. Register T2CON defines the operating mode.
Lets select:
Continuous up Mode
Timer Prescaler : 1
Enable Timer (TENABLE = 1)
No Timer Compare Operation Enable
Register T2PR must define the time period. According to:
HISCP TPS PR T
f
f
T
CPU
PWM

=
2
2

and with a given 150MHz CPU frequency, HISCP =2, TPST2 = 1 and 50 KHz as
output frequency we derive:
T2PR = 1500
Add the necessary instructions for T2PR and T2CON!
Connect T1PWM to ADCIN2
24. Connect T1PWM (eZdsp Pin P8 -15) to ADCIN2 (eZdsp Pin P9 -6) with a wire or a
1000 Ohm resistor provided by your laboratory technician.
Caution: Be careful when connecting pins while the eZdsp is
powered on. By plugging the wire into wrong pins you can
damage the board!
To be safe, ask your technician for assistance before you connect
anything!
Build, Load and Test
25. Finalize the Project and prepare a test: :

Project Build
File Load Program
Debug Reset CPU
Debug Restart
Debug Go main.
If all code was modified correct, the 2 kHz-Signal should still be auditable at the
loudspeaker.
To verify that our ADC sampling is operating as expected, place a breakpoint at the
beginning of the ADCs interrupt service routine (ADC_FIR_INT_ISR) in file
Adc_isr.c. If you run the code in real time (F5), it should hit the breakpoint
periodically.

Inspect and Visualize the FIR
26. Let us inspect the code in this ISR. The sample (AdcRegs.ADCRESULT0) is stored
in a buffer AdcBuf after it is scaled to 3.0V and converted into an IQ-number. The
default value is I8Q24, defined in IQmathLib.h.
We can display the content of this buffer in a memory window:
View Memory
Address: AdcBuf
Q-Value: 24
Format: 32_Bit Signed Int
Page: Data

27. Back to ADC_FIR_INT_ISR. After the sample is stored in AdcBuf, it is also
placed as latest sample in a filter-array xDelay. Then a function IQssfir is called
and its return value is stored in a new buffer AdcBufFiltered. Obviously, the return
value is the output signal from the FIR-Filter.
28. Now inspect the filter-function IQssfir in filter.c. Input parameters are the
samples, the coefficients and number of taps. The filter implementation is a C-based
solution with no optimization. The difference to the solution that was presented at the
beginning of the chapter is that it is a tailored solution for the C28x IQ-Math Library,
running much faster than any ANSI-C solution with float variables.
You have also learned about Texas Instruments Filter Library. Knowing that these
library functions are based on an assembly language implementation we could move
on and increase the speed of the FIR-calculation further by replacing the IQssfir-
function with one from the library.
CCS Graphical Tool
29. Now lets visualize both the square wave and the filtered signal. CCS has a build in
tool to visualize the content of an area of code or data memory graphically. We can
use this tool to plot the content of AdcBuf and AdcBufFiltered:
View Graph Time/Frequency
Select the properties:
Display Type : Dual Time
Start Address upper display: AdcBuf
Start Address lower display: AdcBufFiltered
Page: Data
Acquisition Buffer Size: 50
Display Data Size: 50
DSP Data Type: 32-bit signed integer
Q-Value 24
Sampling Rate: 50000
Time Display Unit: s
When you close the property window with <OK> a Graphical Display with a yellow
background should pop up.
30. Enable Real Time Mode:
Debug Real Time Mode
Click right in the graph window and select Continuous Refresh
31. Run the code in Real Time (F5)

The Graph window should look like this:

End of Lab 14
This page has intentionally been left blank.

TMS320F2812 - FIR Filter

Uploaded by

Copyright:

Available Formats

TMS320F2812 - FIR Filter

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

TMS320F2812 - FIR Filter

Uploaded by

Copyright:

Available Formats

DSP28 - FIR - Filter 14 - 1

You might also like