CPEN 304 Digital Signal Processing

Lecture 1
Introduction to Digital Signal Processing
September 20, 2021
Dr. Godfrey A. Mills
Email: [email protected]
Phone: 020-549-6944
1 CPEN 304 Lecture 1
 LECTURE #1 OUTLINE
— What is digital signal processing
— Why digital signal processing
— Digital signal processing tasks
— Advantages and limitation of analog processing
— Advantages and limitations of digital processing
— Components of DSP systems and functions
— Architecture of DSP processor and functions
— Hardware DSP implementation
— Software DSP implementation
— Application areas and scope of DSP
2 CPEN 304 Lecture 1
 What is Signal Processing
— Signal >> physical phenomenon that varies with respect to
time, space or other independent variable that contains
information about the behavior or nature of a phenomenon.
— System >> an operator (physical device or software) that
performs an operation on a signals applied to its terminal
based on well-defined rules to produce an output response
— Signal processing >> operation of processing signals to
extract information content for various applications.
— When we pass a signal through a system (as in filtering), we
refer to the process as “processing of the signals”
— Why perform signal processing >> One key reason is
to eliminate or reduce the effect of undesirable signals (noise
and interference) on the information bearing signals.
3 CPEN 304 Lecture 1
 What is Signal Processing
— Cont’d……
— Many signals encountered in practice are analog in nature >>
to process these signals digitally requires converting the analog
signals into a digital form for processing
— Analog signals can be processed directly using appropriate
analog systems (such as amplifier, filters, frequency analyzers
or frequency multipliers) to change the signal characteristics
or extract the desired data

4 CPEN 304 Lecture 1

 What is Digital signal processing
— Digital signal processing >> application of mathematical
operations to manipulate analog signals (such as voice, audio,
video, pressure, temperature, position), that are represented
in digital form (sequences of samples) using a program or a
computer (or special purpose processor);
— We refer to the method or rules for implementing a system to
perform signal processing operation as the DSP algorithm.
— The algorithm must have large mathematical operations that
must be performed quickly and must repeat on series of data
— Example, program a computer to perform digital filtering
— The general goal of a digital signal processing is to measure,
filter and/or compress digital signals to desired form

5 CPEN 304 Lecture 1

 What is Digital signal processing
— Cont’d….
— Processing analog signals in digital form require an interface
between the analog signal and digital processor >> ADC
— Once signal has been reduced to discrete samples/numbers,
we can isolate, analyze, and rearrange the components more
easily than if it were in analog form.
— Digital processors used for processing the discrete data may
be a programmable digital computer or a microprocessor
programmed to perform desired operation on signal

6 CPEN 304 Lecture 1

 What is Digital signal processing
— Cont’d….
— The digital processor may be hardwired processor that may
be configured to perform specified set of instructions
— With programmable digital computers (DSP processors),
there is flexibility to change the signal processing operations
via changes in the software
— Hardwired implementation are useful in situations where
the signal processing operations are well defined
— When output is required in analog form (example speech
communication), another interface must be required
between the digital processor and the output >> DAC
— In situations where output is not required in another form,
then there is no need for requirement of a DAC
7 CPEN 304 Lecture 1
 Why we need digital processor
— We need a digital signal processor to perform the following
digital signal processing tasks in real time:
— Signal modelling
— Difference equation
— Convolution
— Transfer function
— Frequency response
— Signal processing
— Data manipulation
— Algorithms
— Filtering
8 — Estimation
CPEN 304 Lecture 1
 Task of DSP systems
— DSP system tasks require :
— Repetitive numeric computations
— Attention to numeric fidelity
— High memory bandwidth, mostly via array accesses
— Real-time processing
— To achieve the above task, a DSP design should minimize :
— Cost
— Power
— Memory use
— Development time

9 CPEN 304 Lecture 1

 Advantages of Analog Processing
— Advantages of analog signal processing :
— Low cost and simple in implementation ;
— Analog filters for operation are simple to build >>
attenuators and amplifiers >> active lowpass RC circuit
— Low signal levels ;
— Infinite effective sampling rate ;
— Infinite resolution in frequency
— No aliasing and reconstruction issues
— Infinite resolution in amplitude ;
— No quantization noise

10 CPEN 304 Lecture 1

 Limitations of Analog Processing
— Challenges of analog signal processing :
— Accuracy limitations due to component tolerances,
undesired nonlinearities, etc.
— Difficulty associated with repeatability due to tolerance of
components, changes in environmental conditions such as
temperature and vibration effects
— High sensitivity to electrical noise;
— Inflexibility to changes ;
— Difficulty of implementing certain operations such as:
— Nonlinear operations; time-varying operations, etc.
— Difficulty of storing data ; issues with programmability
11 CPEN 304 Lecture 1
 Advantages of Digital Processing
— Advantages of digital signal processing are :
— DSP operations can easily be changed through changes in
the program (in digital programmable processors) >>
this gives a flexible systems via software implementation.
— Better control of accuracy can be obtained by changing
(increasing) the number of bits for data representation
— Digital signals are easily stored on storage media without
loss of quality of reproduction of signal.
— Digital signals sensitivity to electrical noise is minimal in
comparison with the analog signal
— Digital circuits are less sensitive to tolerances of the circuit
component values.
CPEN 304 Lecture 1
12
 Advantages of Digital Processing
— Cont’d……
— Digital signals can be easily processed offline and thus, are
easily transported
— Digital information can be easily encrypted for security
— Digital systems are independent of external factors such as
temperature, ageing and other external parameters.
— Flexibility can be achieve with software implementations;
— The DSPs can be cascaded without any loading problems
as exist in analog processing systems (example cascading
of multiple analog filters vrs cascading of digital systems)
13 CPEN 304 Lecture 1
 Limitations of Digital Processing
— Limitations of digital signal processing :
— Sampling of signals are required >> lead to loss of data;
— ADC and DAC devices are required >> requirements of
mixed-signal hardware;
— Limited speed of processors
— Quantization and round-off errors introduced due to the
operations >> lead to loss of information

14 CPEN 304 Lecture 1

 Components of DSP System
— Architecture of a DSP system comprises 3 main components
— Analog filters (anti-aliasing + reconstruction filters);
— ADC and DAC (mixed signal circuits or elements); and
— Digital processor (microprocessor unit)

15 CPEN 304 Lecture 1

 Anti-aliasing filter systems
— Practically, most continuous time signals contain some other
frequency components (noise signals) whose frequencies are
higher than the folding frequency (Nyquist sampling rate) or
the maximum frequency content of signal
— These higher frequencies must be removed before we satisfy
sampling theorem condition >> band-limit the input signal
so that all the frequencies will be less then folding frequency
— To accurately describe an audio signal with frequencies up
to 20 kHz, ADC must sample at a minimum of 40 kHz
— We remove the higher frequencies by passing the analog
signals through a low-pass filter system (anti-aliasing
filter), before passing the filtered signals to the ADC.

16 CPEN 304 Lecture 1

 Reconstruction filter systems
— Output signals from the DAC system usually takes the form
of stair-case like waveform which requires smoothing;
— Smoothening in the signal is achieved via use of a low-pass
analog filter (interpolator) called reconstruction filter
>> refer to CPEN 301 on sampling and reconstruction

17 CPEN 304 Lecture 1

 ADC and DAC of DSP System
— The ADC operation can be viewed as a 3 stage process:
— Sampling >> discretization of analog signals by taking
samples at discrete instances of time
— Quantization >> conversion of discrete time sampled
signals into a discrete valued signal
— Value of each signal sample is represented by a value
selected from a finite set of integer values
— Difference between the unquantized sample and the
quantized output is called the quantization error
— Coding >> representing each of the discrete time values
by a b-bits of binary sequence

18 CPEN 304 Lecture 1

 ADC and DAC of DSP System
— Cont’d…..
— ADC sampling rate depends on bandwidth of the analog
signals being sampled >> this rate sets the pace at which
the samples are made available to processor for processing.
— The sampling rate is determined by the amount of signal
required for processing for a given application.

19 CPEN 304 Lecture 1

 ADC and DAC of DSP System
— Cont’d…..

20 CPEN 304 Lecture 1

 ADC and DAC of DSP System
— Cont’d….
— For ADC to give accurate output, input analog signals to the
ADC must remain constant in amplitude during the
read-in time until the conversion is complete.
— To account for this, a sample-and-hold (S/H) circuit is
usually used in conjunction with the ADC.
— The S/H circuit consists of analog switch, a shunt
capacitor, and a voltage follower >> the circuit is
supplied by a source voltage Vin with internal resistor, Rs.
— If the analog signal is changing too fast, the S/H circuit can
be used to freeze the input voltage during the conversion.
21 CPEN 304 Lecture 1
 ADC and DAC of DSP System
— Cont’d….
— Performance of signal processing system depends to a large
extent on the ADC.
— ADC is specified by number of bits which defines its
resolution and conversion time decides the sampling time.
— The errors in ADC are due to the finite number of bits and
finite conversion time.
— Some times noise may be introduced by the switching
circuits.
— Similarly, a DAC is represented by number of bits and the
settling time at the output
22 CPEN 304 Lecture 1
 The DSP Processor
— Digital processor of DSP system is special type of processor
designed for performing the mathematics involved in DSP
operations >> most of the processors are programmable
>> can be used for manipulating different types of
information such as sound, images, video, audio, speech, etc
— Variety of processors available for DSP implementation :
— General Purpose Processors (GPP) >> microprocessors;
— Specialized purpose processors >> DSP processors;
— Purpose-built processors or hardware >> Application-
Specific Integrated Circuit (ASICs), Field-programmable
gate arrays (FPGAs);
— Difference between DSP processor and others ???

23 CPEN 304 Lecture 1

 The DSP Processor
— Cont’d….
— Processors used in a wireless cellular communication system

24 CPEN 304 Lecture 1

 The DSP Processor
— Cont’d….
— Mapping of the the processors with the functions of the
components of the wireless cellular system on a chip

25 CPEN 304 Lecture 1

 DSP algorithm format and notation
— DSP is mainly designed to perform mathematical functions
like "add", "subtract", "multiply" and "divide" quickly.
— Consider a simple filter system described by equation below

— Recall output of discrete time system using convolution sum

— We observe from the format above that the output y[n] of the
system (for N input samples) above is combination of sum
of products of coefficients and past input values
— DSP culture has graphical representation to describe the math
— Sketch the implementation diagram for the case of N = 4 >>
26
write a Lecture
CPEN 304 C/C++ 1 code to implement case for N=10 samples
 DSP algorithm format and notation
— Cont’d……

27 CPEN 304 Lecture 1

 DSP algorithm format and notation
— Cont’d……
— We also also from the expression a flowchart for formulas,
we have inner loops, and not programs.
— We have summation for add, multiplication, and other
variable that denote delay/storage (z-1)
— DSP algorithms such as filter above spends most of the their
execution time in loops >> same set of program instructions
will continually pass from program memory to CPU
— In DSPs, algorithms are kings while software is not in DSPs
>> specification of DSP is on 4 key algorithms >> IIR
filters, FIR filters, FFT (for the DFT), DCT, convolution
28 CPEN 304 Lecture 1
 DSP algorithm format and notation
— Cont’d….
— FIR filters or discrete time systems
— FIR filters reduce signal noise and enhance image or signal
quality by removing unwanted frequencies.
— FIR compute output that depends only on the input sequence
and the impulse response.
— The algorithm has the following structure :
— N most recent samples in the delay line (Xi)
— New sample moves data down delay line
— “Tap” is a multiply-add
— Each tap nominally requires: Two data fetches, Multiply,
29 Accumulate,
CPEN 304 Lecture 1 Memory write-back to update delay line
 DSP algorithm format and notation
— Cont’d……
— To implement the FIR discrete time system (filter), the DSP
processor will perform the following operations:
— Initialize accumulator;
— Fetch coefficients and other data;
— Do multiply-accumulate (MAC); >> mul + add
— Memory write back to update line
— Repeat fetch & MAC until done

30 CPEN 304 Lecture 1

 Types of DSP Processors
— DSP multiprocessors on a die
— TMS320C80
— TMS320C6000
— 32-bit floating point DSPs
— TI TMS320C4X
— Motorola 96000
— AT&T DSP32C
— Analog devices ADSP21000
— 16-bit fixed point DSP
— TI TMS320C2X
— Motorola 56000
— AT&T DSP16
31 Analog
—CPEN Devices
304 Lecture 1 ADSP2100
 Architecture of DSP Processor
— Architectural diagram of a typical DSP processor shows four
key elements that play independent role in the execution of
the program (refer to your computer architecture course)

32 CPEN 304 Lecture 1

 Architecture of DSP Processor
— Cont’d….
— DSP processors are specialized microprocessors with
architectures that are optimized for fast operational needs:
— real time processing capabilities >> have to process
data in real time >> correctness of the operation depends
heavily on the time when the data processing is complete
— high throughput
— DSP processors architectures used are of three types:
— Von Neumann >> single memory and single bus
— Harvard >> separate memories for data and program code
— Super Harvard Architecture (SHARC) >> Harvard with
33
addition
CPEN 304 Lectureof
1 an instruction cache and an I/O controller
 Features of DSP processor
— As a special microprocessor system, architectural features of
DSP processor >> CPU, ALU, memory, I/O, accumulators,
barrel shifter, multiplier unit, compare select and store unit,
DMA controller, and data and instruction buses
— These DSP features are organized into the following:
— Datapath configured for DSP
— Fixed point arithmetic and MAC
— Multiple memory banks and buses
— Specialized peripherals (I/Os) for DSP
— Specialized addressing modes
— Specialized instruction set and execution control
34 CPEN 304 Lecture 1
 Features of DSP processor
— Cont’d….
— Computational engine of DSPs (central ALU)
— This part of DSP performs major arithmetic operations
or processing instructions like multiplication, addition,,
— The ALU is centered around the MAC unit, which is the
basic computational block of computing devices
— MAC unit consists of multiplier, adder, and accumulator
which performs multiplication and accumulation process
— The accumulator register holds intermediate results
— MAC unit of DSP determines the speed of overall system
>> part that makes the DSP so fast compared with other
traditional processors
CPEN 304 Lecture 1
35
 Features of DSP processor
— Cont’d….
— Multiplier-Accumulator Unit
— DSP operations typically involves many time consuming
operations of multiplication and addition
— To make real-time operation faster, a MAC unit using
either fixed point or floating point arithmetic is mandatory
— MAC unit consists of a multiplier that has a pair or
two input registers (x and y) that hold the inputs to
the multiplier and a k-bit product register that holds
results of multiplication
— Output of the product register is connected to a double
precision accumulator where products are accumulated
CPEN 304 Lecture 1
36
 Features of DSP processor
— Cont’d….
— Output of the product register is connected to a double
precision accumulator where products are accumulated
— Perform memory write back to update line

37 CPEN 304 Lecture 1

 Features of DSP processor
— Cont’d….
— Since we do not want an overflow or scale the accumulator,
two options are possible to avoid such overflow conditions
— Option 1:
— When the accumulator is wider than the product bit, we
“guard bit”
— Option 2:
— We shift right and round the product before adder
— 3 DSP approaches for rounding
— Truncation >> chop off the results
— Round to nearest >> for < ½, round down
38 — Convergent
CPEN 304 Lecture 1 >> for < ½, round down, > ½ round up
 Features of DSP processor
— Cont’d….
— Implementation of the “guard bit” and shift round product

39 CPEN 304 Lecture 1

 Features of DSP processor
— Cont’d…..
— For example, we can map the discrete time system shown in
the implementation block diagram onto a DSP execution
unit as indicated on diagram below

40 CPEN 304 Lecture 1

 Features of DSP processor
— Cont’d…..
— We observe from the implementation diagram that the
critical hardware unit in the DSP is the multiplier
— >> much of architecture is organized around allowing
use of the multiplier on every cycle
— This means we provide two operands on every cycle,
through multiple data and address buses, multiple address
units, and local accumulator feedback

41 CPEN 304 Lecture 1

 Features of DSP processor
— Cont’d…..
— Memory (Program memory + Data memory)
— Program memory that stores the program or instructions
the DSP will use to process the data;
— Data memory that stores the information and data to be
processed
— Memory holds series of instructions which tell DSP
which operations to perform on data.
— In most cases the CPU of the DSP reads some data,
operates on it, and writes it back.
— Complete many memory accesses in a single clock cycle
— Almost all DSPs have some form of memory device –
whether on-chip memory or off-chip memory.
42 CPEN 304 Lecture 1
 Features of DSP processor
— Cont’d…..
— Many memory access is completed in a single clock cycle
— Processor can fetch instructions while also fetching the
operands or storing to memory
— Example, in the simple filter expression shown earlier,
the system can operate a multiply and accumulate while
loading the operands and coefficients for the next cycle
— May perform three reads and one or two writes per cycle
— This operation may require multiple memory buses on the
same chip rather than simple address and data bus >>
Harvard architecture

43 CPEN 304 Lecture 1

 Features of DSP processor
— Cont’d…..
— Specialized I/O peripherals for DSPs
(Synchronous serial ports, parallel ports, Timers,
on-chip ADC and DAC, D/A, host ports, bit I/O
port, on-chip DMA controller, clock generator)
— DSPs normally have internal ports that provide high
speed communication with other DSPs and converters.
— SHARC DSP >> at a 40 MHz clock speed, two serial
ports operate at 40 Mbits/s each while six parallel
ports with each providing 40 Mbytes/s data transfer
>> with all six parallel ports are used, we have data
transfer rate of 240Mbytes/s
— Ports are directly connected to internal bus to improve
44 CPENperformance
304 Lecture 1 and reduce external address decoding
 DSP Hardware Implementation
— DSP processors characterized by a number of parameters
and the choice of a DSP processor is application dependent
— Consideration for DSP processor selection include following:
— Arithmetic format >> fixed point or floating point
— Bus widths >> fixed (usually 16 bits) or float (32 bits)
— Speed >> clock frequency (MHz), MIPS, MACs per sec,,,
— Memory/Bus/ISA architecture >> Harvard, SHARC,,,
— Development tools >> S/W tools (assemblers, linkers
simulators, debuggers, compliers, code library, RTOS),
H/W tools (emulators, development boards), JTAG
— Power consumption >> minimize Vcc to reduce power,
sleep mode operation, on-chip clock distribution, etc
— Cost >> limiting packaging cost, real design cost, etc
45 CPEN 304 Lecture 1
 DSP Software Implementation
— DSP algorithms are programmed in the same languages as
any other Engineering tasks using programming tools such
as Basic, C, C++, Assembly, etc.
— C is often the preference of engineers due to its versatility.
— DSP programming can be divided into 3 levels:
— Assembly
— Compiled
— Application specific.
— Common DSP algorithms usually implemented are:
— Frequency domain filtering – FIR and IIR digital filters
— Frequency-time transformation – FFT

46 CPEN 304 Lecture 1

 DSP application areas and scope
— DSP application area and scope by market summarized as:
— Telecommunication application
— Wireless >> cellular phones, base stations, gateways, fax
— Networking >> cable modems, ADSL, VDSL
— Audio application
— MPEG audio, portable audio, speech synthesis, voice
recognition, speech quality enhancement, speech encoder,
speech decoder, demodulator and synchronizer, equalizer,
— Music application
— Synthetic instruments, audio effects, noise reduction
— Consumer electronic
— DVD, digital camera, toaster, wireless headset, audio, etc
47 CPEN 304 Lecture 1
 DSP application areas and scope
— Cont’d……
— Medical diagnostic application
— MRI, Ultrasonic imaging, computer tomography, ECG,
EEG, audiology, etc
— Aviation application
— Radar, radio navigation
— Geophysics application
— Seismology, oil exploration
— Security application
— Digital watermarking, biometric identification,
surveillance system, signal intelligence, steganography,
electronic warfare, information security
48 CPEN 304 Lecture 1
 DSP application areas and scope
— Cont’d……
— Other related fields
— Electrical engineering, industrial control, data
compression, machine learning, real-time computing,
feature extraction for pattern recognition

49 CPEN 304 Lecture 1

 DSP application areas and scope
— Cont’d……
— Other related fields
— Electrical engineering, industrial control, dataflow
engineering, data compression, machine learning, real-
time computing, vision, image compression, image
compositing, beamforming, spectral estimation, feature
extraction for pattern recognition

50 CPEN 304 Lecture 1