Get documents about "Paper Presention on digital signal processing and its applications
W
Description
This is the my paper Presention report on digital signal processing .in which i cove the what is mean by dsp? and application of dsp and advantages of dsp.hope it will help you all.
Document Sample
1. Introduction 2. Digital Signal processing
Signal processing is an area of electrical Digital signal processing (DSP) is
engineering, systems engineering and concerned with the representation of signals
applied mathematics that deals with by a sequence of numbers or symbols and
operations on or analysis of signals, in either the processing of these signals. Digital
discrete or continuous time to perform signal processing and analog signal
useful operations on those signals. Signals processing are subfields of signal
of interest can include sound, images, time- processing. DSP includes subfields like:
varying measurement values and sensor audio and speech signal processing, sonar
data, for example biological data such as and radar signal processing, sensor array
electrocardiograms, control system signals, processing, spectral estimation, statistical
telecommunication transmission signals signal processing, digital image processing,
such as radio signals, and many others. signal processing for communications, bio-
Signals are analog or digital electrical medical signal processing, seismic data
representations of time-varying or spatial- processing, etc.
varying physical quantities. In the context of
signal processing, arbitrary binary data
streams and on-off signals are not 2.1 Why DSP & Not ASP
considered as signals, but only analog and
digital signals that are representations of Following are the reasons why DSP is
analog physical quantities. preferred over ASP:
Signal processing is divided into following i. A digital programmable system allows
categories; flexibility in reconfiguring the digital signal
processing operations simply by changing
i. Analog signal processing [ASP] the program where as reconfiguring an
analog system usually implies a re-design of
ii. Digital signal processing [DSP] the hardware followed by testing &
verification.
Analog signal processing is for signals
that have not been digitized, as in classical ii. Tolerances in analog circuit components
radio, telephone, radar, and television make it extremely difficult to control the
systems. This involves linear electronic accuracy of an ASP. On the other hand a
circuits such as passive filters, active filters, digital system provides much better control
additive mixers, integrators and delay lines. of accuracy requirements.
It also involves non-linear circuits such as
commanders, multiplicators (frequency iii. Digital signals are easily stored on
mixers and voltage-controlled amplifiers), magnetic media without deterioration
voltage-controlled filters, voltage-controlled beyond that introduced in A to D
oscillators and phase-locked loops. The conversion. As a result the signals become
output of ASP is a processed analog signal. transportable and can be processed off- line
in a remote laboratory.
1
iv. DSP allows implementation of more convert the signal from an analog to a digital
sophisticated signal processing algorithm. form, by using an analog-to-digital converter
(ADC). Often, the required output signal is
v. It’s usually very difficult to perform another analog output signal, which requires
precise mathematical operations on signals a digital-to-analog converter (DAC). Even if
in analog form but these same operations this process is more complex than analog
can be routinely implemented on a digital processing and has a discrete value range,
computer using software. the stability of digital signal processing
thanks to error detection and correction and
2.2 DSP Domains being less vulnerable to noise makes it
advantageous over analog signal processing
In DSP, engineers usually study digital for many, though not all, applications.
signals in one of the following domains:
time domain (one-dimensional signals), 2.3.1 A/D Converter
spatial domain (multidimensional signals),
frequency domain, autocorrelation domain, An analog-to-digital converter is a device
and wavelet domains. They choose the which converts continuous signals to
domain in which to process a signal by discrete digital numbers. Typically, an ADC
making an informed guess (or by trying is an electronic device that converts an input
different possibilities) as to which domain analog voltage (or current) to a digital
best represents the essential characteristics number proportional to the magnitude of the
of the signal. A sequence of samples from a voltage or current. However, some non-
measuring device produces a time or spatial electronic or only partially electronic
domain representation, whereas a discrete devices, such as rotary encoders, can also be
Fourier transform produces the frequency considered ADCs. The digital output may
domain information, which is the frequency use different coding schemes, such as
spectrum. binary, Gray code or two's complement
binary.
2.3 Process of Digital Signal
Processing
Figure 1. Block diagram of a DSP system
Since the goal of DSP is usually to Figure 2. (4-channel stereo multiplexed
measure or filter continuous real-world analog-to-digital converter WM8775SEDS)
analog signals, the first step is usually to
2
Sampling can be described by the relation;
x[n] = xa (n Ts)
Where x[n] = Discrete time signal
Xa = input analog signal
n= samples
Ts = Sampling interval
Figure 3. Block diagram of Analog to Fs = 1/T
Digital converter
2.3.1. A Sampler
. In order to use an analog signal on a
computer it must be digitized with an
analog-to-digital converter. A continuously
varying band limited signal can be sampled
(that is, the signal values at intervals of time Figure 5. Sampled signal (discrete time,
T, the sampling time .Sampling is usually continuous values)
carried out in two stages, discretization and
quantization. Since a practical ADC cannot make an
instantaneous conversion, the input value
In the discretization stage, the space of must necessarily be held constant during the
signals is partitioned into equivalence time that the converter performs a
classes such that the sampling rate is higher conversion (called the conversion time). An
than twice the highest frequency of the input circuit called a sample and hold
signal. This is essentially what is embodied performs this task—in most cases by using a
in the Shannon-Nyquist sampling theorem. capacitor to store the analog voltage at the
input, & using an electronic switch or gate
to disconnect the capacitor from the input.
Many ADC integrated circuits include the
sample and hold subsystem internally.
2.3.1. B Quantizer
Quantization is carried out by replacing the
sampled signal with representative signal of
the corresponding equivalence class. In the
quantization stage the representative signal
values are approximated by values from a
finite set. For e.g. rounding a real number in
the interval [0,100] to an integer 0, 1,
Figure 4. Sampler
3
2...100. In other words, quantization can be converter, for instance, must take its signal
described as a mapping that represents a samples often enough to catch all the
finite continuous interval I = [a,b] of the relevant fluctuations. If the ADC is too
range of a continuous valued signal, with a slow, it misses some of the action. Imagine
single number c, which is also on that trying to film a football game with a movie
interval. For example, rounding to the camera running at one frame per minute.
The film would be incoherent, missing
nearest integer replaces the interval [c −
entire plays in the intervals between frames.
.5,c + .5) with the number c, for integer The DSP, too, must keep pace, churning out
calculations as fast as the signal data is
received from the ADC. The pace gets
progressively more demanding as the signal
gets faster. Stereo equipment handles sound
signals of up to 20 kilohertz (20,000 cycles
per second, the upper limit of human
hearing), requiring a DSP to perform
Figure 6. Quantized signal: continuous time, hundreds of millions of operations per
discrete values. second. Other signals, such as satellite
transmissions, are even faster, reaching up
into the Gigahertz (billions of cycles per
second) range.
2.3.2. B DSPs versus Microprocessors
Figure 7. Digital signal (sampled, quantized: DSPs differ from microprocessors in a
discrete time, discrete values.) number of ways. Microprocessors are
typically built for a range of general purpose
2.3.2 Digital Signal Processors functions, and normally run large blocks of
software, such as operating systems like
2.3.2. A Blinding Speed Windows or UNIX. Although today's
microprocessors, including the popular and
At its heart, digital signal processing is well-known Pentium family, are extremely
highly numerical and very repetitive. As fast--as fast or faster than some DSPs--they
each new piece of signal data arrives, it must are still not often called upon to perform
be multiplied, summed, and otherwise real-time computation or signal processing.
transformed according to complex formulas. Usually, their bulk processing power is
What makes this such a keen technological directed more at handling many tasks at
challenge is the speed requirement. DSP once, and controlling huge amounts of
systems must work in real time, capturing memory and data, and controlling a wide
and processing information as it happens. variety of computer peripherals (disk drive,
Like a worker on a fast-moving assembly modem, video display, etc). However,
line, Analog-to-Digital converters and DSPs microprocessors such as Pentiums are
must keep up with the work flow. If they fall notorious for their size, cost, and power
behind, information is lost and the signal consumption to achieve their muscular
gets distorted. The Analog-to-Digital performance, whereas DSPs are more
4
dedicated, racing through a smaller range of hand, hi-fidelity stereo sound has a wider
functions at lightning speed, yet less costly range, calling for a 16-bit ADC or 24-bit
and requiring much less space (size) and ADC, and a 24-bit fixed-point DSP like the
power consumption to achieve their purpose. Motorola DSP563xx series. In this case, the
DSPs are often used in "embedded systems", ADC's 16-bit or 24-bit width is needed to
where they are accompanied by all capture the complete high-fidelity signal
necessary software (stored in onchip ROM (i.e. much better than a phone); the DSP thus
or offchip EEPROM), built deep into a piece must be 24 bits to accommodate the larger
of equipment, and dedicated to a group of values resulting when the signal data is
related tasks. In computer systems, DSPs manipulated.) Applications requiring still
may be employed as attached processors, greater dynamic range include image
assisting a general purpose host processing, 3-D graphics, and scientific and
microprocessor. research simulations; such applications
typically a 32-bit floating-point processor.
2.3.2. C Different DSPs for Different Jobs
2.3.2. D DSP Evolution
One way to classify DSP devices and
applications is by their dynamic range. The Around 30 years ago, digital signal
dynamic range is the spread of numbers, processing was more theory than practice.
from small to large that must be processed in The only systems capable of doing signal
the course of an application. It takes a processing were massive mainframes and
certain range of values, for instance, to supercomputers and even then, much of the
describe the entire waveform of a particular processing was done not in real time, but
signal, from deepest valley to highest peak. off-line in batches. For example, seismic
The range may get even wider as data was collected in the field, stored on
calculations are performed, generating larger magnetic tapes and then taken to a
and smaller numbers through multiplication computing centre, where a mainframe might
and division. The DSP device must have the take hours or days to digest the information.
capacity to handle the numbers so generated. The first practical real-time DSP systems
If it doesn't, the numbers may "overflow," emerged in the late 1970s and used bipolar
producing invalid results. The processor's "bit-slice" components. Large quantities of
capacity is a function of its data width (i.e. these building-block chips were needed to
the number of bits it manipulates) and the design a system, at considerable effort and
type of arithmetic it performs (i.e., fixed or expense. Uses were limited to esoteric high-
floating point). A 32-bit processor has a end technology, such as military and space
wider dynamic range than a 24-bit systems. The economics began to change in
Processor, which has a wider range than 16- the early 80s with the advent of single-chip
bit processor. And floating-point chips have MOS (Metal-Oxide Semiconductor) DSPs.
wider ranges than fixed-point devices. Each Cheaper and easier to design-in than
type of processor is suited for a particular building blocks, these "monolithic"
range of applications. 16-bit fixed-point processors meant that digital signal
DSPs such as typically used for voice-grade processing could be cost-effectively
and telecom systems (such as cell-phones), integrated into an array of ordinary products.
since they work with a relatively narrow The early single-chip processors were
range of sound frequencies. On the other relatively simple 16-bit devices, which,
5
teamed with 8- or 10-bit ADCs, were complete processor requires combining the
suitable for low-speed applications, general- core with memory and interfaces to the
purpose coders such as talking toys, simple outside world. While the core and these
controllers, and vocoders; (voice encoding peripheral sections are designed separately,
devices used in telecommunications). they will be fabricated on the same piece of
silicon, making the processor a single
2.3.2. E The Digital Signal Processor integrated circuit.
Market
Suppose you build cellular telephones and
The DSP market is very large and growing want to include a DSP in the design. You
rapidly. As shown in Fig.8 it will be about will probably want to purchase the DSP as a
8-10 billion dollars/year at the turn of the processor, that is, an integrated circuit
century, and growing at a rate of 30-40% ("chip") that contains the core, memory and
each year. This is being fueled by the other internal features. To incorporate this
incessant IC in your product, you design a printed
circuit board where it will be soldered in
next to your other electronics. This is the
most common way that DSPs are used.
There are several dozen companies that will sell
you DSPs already mounted on a printed circuit
board. These have such features as extra
memory, A/D and D/A converters, EPROM
sockets, multiple processors on the same board,
and so on. While some of these boards are
intended to be used as stand alone computers,
most are configured to be plugged into a host,
such as a personal computer. Following are
Figure 8. Graph of increasing demand of some of the companies that dominate
DSP today’s DSP market:
DSP marketdemand for better and cheaper
consumer products, such as: cellular
telephones, multimedia computers, and
high-fidelity music reproduction. These
high-revenue applications are shaping the
field, while less profitable areas, such as
scientific instrumentation, are just riding the
wave of technology.
DSPs can be purchased in three forms, as a
core, as a processor, and as a board level
product. In DSP, the term "core" refers to
the section of the processor where the key
tasks are carried out, including the data
registers, multiplier, ALU, address
generator, and program sequencer. A
6
2.3.2.F Things that have DSPs
Figure 9. (8-channel digital-to-analog
converter Cirrus Logic CS4382 placed on
Sound Blaster X-Fi Fatal1ty)
A DAC converts an abstract finite-
precision number (usually a fixed-point
binary number) into a concrete physical
quantity (e.g., a voltage or a pressure). In
particular, DACs are often used to convert
finite-precision time series data to a
Some typical and well-known items which continually varying physical signal.
contain one (or many) embedded DSPs:
∑ the biggie: cell phones
∑ fax machines
∑ DVD players and other home audio
equipment
∑ your car (for example: the anti-lock
braking system)
∑ computer disk drives Figure 10. Reconstructed analog signal
∑ satellites (they have a lot)
∑ the "switch" at your local telephone A typical DAC converts the abstract
company (more than a lot) numbers into a concrete sequence of
∑ digital radios impulses that are then processed by a
∑ high-resolution printers reconstruction filter using some form of
∑ digital cameras interpolation to fill in data between the
impulses. Other DAC methods (e.g.,
2.3.3 Digital To Analog Converter methods based on Delta-sigma modulation)
produce a pulse-density modulated signal
In electronics, a digital-to-analog that can then be filtered in a similar way to
converter (DAC or D-to-A) is a device for produce a smoothly varying signal.
converting a digital (usually binary) code to
an analog signal (current, voltage or electric By the Nyquist–Shannon sampling
charge). theorem, sampled data can be reconstructed
perfectly provided that its bandwidth meets
certain requirements (e.g., a baseband signal
with bandwidth less than the Nyquist
7
frequency). However, even with an ideal rather than a voltage or current as in a
reconstruction filter, digital sampling analog filter.
introduces quantization error that makes
perfect reconstruction practically 3.1 Advantages of using digital
impossible. Increasing the digital resolution filters
(i.e., increasing the number of bits used in
each sample) or introducing sampling dither The following list gives some of the main
can reduce this error. advantages of digital over analog filters.
1. A digital filter is programmable, i.e. its
3. Digital Filters operation is determined by a program stored
in the processor's memory. This means the
In signal processing, the function of a digital filter can easily be changed without
filter is to remove unwanted parts of the affecting the circuitry (hardware).
signal, such as random noise, or to extract An analog filter can only be changed by
useful parts of the signal, such as the redesigning the filter circuit.
components lying within a certain frequency
range. 2. Digital filters are easily designed, tested
There are two main kinds of filter, analog and implemented on a general-purpose
and digital. They are quite different in their computer or workstation.
physical makeup and in 3. The characteristics of analog filter circuits
how they work. are subject to drift and are dependent on
An analog filter uses analog electronic temperature. Digital filters do not suffer
circuits made up from components such as from these problems, and so are extremely
resistors, capacitors and op-amps to produce stable with respect both to time and
the required filtering effect. Such filter temperature.
circuits are widely used in applications such
as noise reduction, video signal 4. Unlike their analog counterparts, digital
enhancement, graphic equalisers in hi-fi filters can handle low frequency signals
systems, and many other areas. There are accurately. As the speed of DSP technology
well-established standard techniques for continues to increase, digital filters are being
designing an analog filter circuit for a given applied to high frequency signals in the RF
requirement. At all stages, the signal being (radio frequency) domain, which in the past
filtered is an electrical voltage or current was the exclusive preserve of analog
which is the direct analogue of the physical technology.
quantity (e.g. a sound or video signal or
transducer output) involved. 5. Digital filters are very much more
A digital filter uses a digital processor to versatile in their ability to process signals in
perform numerical calculations on sampled a variety of ways; this includes the ability of
values of the signal. The processor may be a some types of digital filter to adapt to
general-purpose computer such as a PC, or a changes in the characteristics of the signal.
specialised DSP (Digital Signal Processor)
chip. 6. Fast DSP processors can handle complex
Note that in a digital filter, the signal is combinations of filters in parallel or cascade
represented by a sequence of numbers, (series), making the hardware requirements
8
relatively simple and compact in comparison Examples of simple digital filters
with the equivalent analog circuitry. The following examples illustrate the
essential features of digital filters.
3.2 Operation of digital filters
A. Unity gain filter:
In this section, we will develop the basic
theory of the operation of digital filters. It is yn = xn
essential to understand how digital filters are Each output value yn is exactly the same as
designed and used. Suppose the "raw" signal the corresponding input value xn:
which is to be digitally filtered is in the form y1= x1
of a voltage waveform described by the y2= x2
function y3= x3
V = x (t) ...etc
Where (t) is time. This is a trivial case in which the filter has
This signal is sampled at time intervals h no effect on the signal.
(the sampling interval). The sampled value
at time t = ih is B. Simple gain filter:
xi = x(ih)
Thus the digital values transferred from the yn = K xn
ADC to the processor can be represented by Where K = constant.
the sequence This simply applies a gain factor K to each
x0 , x1 , x2 , x3 , ... input value.
Corresponding to the values of the signal K > 1 makes the filter an amplifier, while 0
waveform at < K < 1 makes it an attenuator. K < 0
t = 0, h, 2h, 3h,... corresponds to an inverting amplifier.
And t = 0 is the instant at which sampling Example (1) above is simply the special case
begins. where K = 1.
At time t = nh (where n is some positive
integer), the values available to the C. Pure delay filter:
processor, stored in memory, are
x0 , x1 , x2 , x3 , ... xn yn= xn-1
Note that the sampled values xn+1, xn+2 etc. The output value at time t = nh is simply the
are not available, as they haven't happened input at time t = (n-1)h, i.e. the signal is
yet! delayed by time h:
The digital output from the processor to the y0 = x-1
DAC consists of the sequence of values y1= x0
y1, y2, y3, y4 ...yn y2= x1
In general, the value of yn is calculated from y3= x2
the values x0, x1, x2, x3 ..., xn. The way in ... etc
which the y's are calculated from the x's Note that as sampling is assumed to
determines the filtering action of the digital commence at t = 0, the input value x-1 at t =
filter. -h is undefined. It is usual to take this (and
any other values of x prior to t = 0) as zero.
9
D. Two-term difference filter: G. Central difference filter:
yn = xn – xn-1 yn = (xn – xn-2) / 2
The output value at t = nh is equal to the This is similar in its effect to example (4).
difference between the current input xn and The output is equal to half the change in the
the previous input xn-1: input signal over the previous two sampling
y0 = x0 – x-1 intervals:
y1 = x1 – x0 y0 = (x0 – x2) / 2
y2 = x2 – x1 y1 = (x1 – x-1) / 2
y3 = x3 – x2 y2 = (x2 – x0) / 2
... etc ... etc
i.e.. the output is the change in the input
over the most recent sampling interval h. 3.3 Order of a digital filter
The effect of this filter is similar to that of
an analog differentiator circuit. The order of a digital filter is the number
of previous inputs (stored in the processor's
E. Two-term average filter: memory) used to calculate the current
output. Thus:
yn= (xn + xn-1) / 2 1. Examples (1) and (2) above are zero-order
The output is the average (arithmetic mean) filters, as the current output yn depends only
of the current and previous input: on the current input xn and not on any
y0 = (x0 + x-1) / 2 previous inputs.
y1= (x1 + x0) / 2
y2 = (x2 + x1) / 2 2. Examples (3), (4) and (5) are all of first
y3 = (x3 + x2) / 2 order, as one previous input (xn-1) is required
... etc to calculate yn. (Note that the filter of
This is a simple type of low pass filter as it example (3) is classed as first-order because
tends to smooth out high-frequency it uses one previous input, even though the
variations in a signal. current input is not used).
F. Three-term average filter: 3. In examples (6) and (7), two previous
inputs (xn-1 and xn-2) are needed, so these are
yn= (xn + xn-1+ xn-2) / 3 second-order filters. Filters may be of any
This is similar to the previous example, with order from zero upwards.
the average being taken of the current and
two previous inputs: All of the digital filter examples given
y0 = (x0 + x-1+x-2) / 2 above can be written in the following
y1= (x1 + x0+x-1) / 3 general forms:
y2 = (x2 + x1+x0) / 3 Zero order: yn = a0xn
y3 = (x3 + x2+x1) / 3 First order: yn = a0xn + a1xn-1
... etc Second order: yn = a0xn + a1xn-1 + a2xn-2
Similar expressions can be developed for
filters of any order where a0, a1...are signal
coefficients.
10
3.4 Finite Impulse Response (FIR) This equation can also be expressed as a
Filter convolution of the coefficient sequence bi
with the input signal:
A finite impulse response (FIR) filter is a
type of a digital filter. The impulse response,
the filter's response to a Kronecker delta
input, is finite because it settles to zero in a
finite number of sample intervals. This is in That is, the filter output is a weighted sum of
contrast to infinite impulse response (IIR) the current and a finite number of previous
filters, which have internal feedback and values of the input signal
may continue to respond indefinitely. The
impulse response of an Nth-order FIR filter
Properties
lasts for N+ 1 sample, and then dies to zero.
In an FIR filter the current output (yn) is
An FIR filter has a number of useful
calculated solely from the current and
properties which sometimes make it
previous input values (xn, xn-1, xn-2...).
preferable to an infinite impulse response
(IIR) filter. FIR filters:
∑ Are inherently stable. This is due to
the fact that all the poles are located
at the origin and thus are located
within the unit circle.
∑ Require no feedback. This means
that any rounding errors are not
Figure 11. Block diagram of a simple FIR compounded by summed iterations.
filter The same relative error occurs in
each calculation. This also makes
The difference equation that defines the implementation simpler.
output of an FIR filter in terms of its input ∑ They can easily be designed to be
is: linear phase by making the
coefficient sequence symmetric;
y[n]=b0x[n] + b1x[n-1] +....+bNx[n-N] linear phase, or phase change
proportional to frequency,
Where: corresponds to equal delay at all
frequencies. This property is
∑ x[n] is the input signal, sometimes desired for phase-
∑ y[n] is the output signal, sensitive applications, for example
∑ bi are the filter coefficients, and crossover filters, and mastering.
∑ N is the filter order – an Nth-order
filter has (N + 1) terms on the right- The main disadvantage of FIR filters is
hand side; these are commonly that considerably more computation power
referred to as taps. is required compared to an IIR filter with
similar sharpness or selectivity, especially
when low frequencies (relative to the sample
rate) cutoffs are needed.
11
3.5. Infinite Impulse Response yn = xn + yn-1
Filter
Infinite impulse response (IIR) is a
property of signal processing systems.
Systems with this property are known as IIR
systems or, when dealing with filter systems,
as IIR filters. IIR systems have an impulse
response function that is non-zero over an
infinite length of time. This is in contrast to
finite impulse response filters (FIR), which
have fixed-duration impulse responses. The
simplest analog IIR filter is an RC filter
made up of a single resistor (R) feeding into Figure 12. Block diagram of an IIR filter
a node shared with a single capacitor (C).
This filter has an exponential impulse In other words, this filter determines the
response characterized by an RC time current output (yn) by adding the current
constant. input (xn) to the previous output (yn-1).
IIR filters may be implemented as either 4. Applications of DSP
analog or digital filters. In digital IIR filters,
the output feedback is immediately apparent The main applications of DSP are audio
in the equations defining the output. Note signal processing, audio compression, digital
that unlike with FIR filters, in designing IIR image processing, video compression,
filters it is necessary to carefully consider speech processing, speech recognition,
"time zero" case in which the outputs of the digital communications, RADAR,SONAR,
filter have not yet been clearly defined. seismology, and biomedicine.
Design of digital IIR filters is heavily Specific examples are speech
dependent on that of their analog compression and transmission in digital
counterparts because there are plenty of mobile phones, room matching equalization
resources, works and straightforward design of sound in Hifi and sound reinforcement
methods concerning analog feedback filter applications, weather forecasting, economic
design while there are hardly any for digital forecasting, seismic data processing,
IIR filters. As a result, usually, when a analysis and control of industrial processes,
digital IIR filter is going to be implemented, computer-generated animations in movies,
an analog filter (e.g. Chebyshev filter, medical imaging such as CAT scans and
Butterworth filter, Elliptic filter) is first MRI, MP3 compression, image
designed and then is converted to a digital manipulation, high fidelity loudspeaker
filter by applying discretization techniques crossovers and equalization, and audio
such as Bilinear transform or Impulse effects for use with electric guitar
invariance. amplifiers.
A simple example of an IIR is given by
12
4.1 Speech Processing ADC quantizes input signal & provides
digital output.
Speech processing is the study of speech
signals and the processing methods of these
signals are regarded as a special case of
digital signal processing, applied to speech
signal.
Speech processing can be divided into the
following categories:
A. Speech recognition, which deals with
analysis of the linguistic content of a
speech signal.
B. Speaker recognition, where the aim
is to recognize the identity of the Figure 13. Block Diagram Of A Speech
speaker. Recognition System
C. Enhancement of speech signals, e.g.
audio noise reduction. The output given is stored in memory.
D. Speech coding, a specialized form of After storage, recognition process starts. In
data compression, is important in the that, spoken word is again digitised & its
telecommunication area. template compared template of memory.
E. Voice analysis for medical purposes, When match occurs, word has been
such as analysis of vocal loading and recognised & system informs user, about the
dysfunction of the vocal cords. match.
F. Speech synthesis: the artificial
synthesis of speech, which usually It is but obvious that performance of
means computer-generated speech. system is greatly affected by:
A. Speech Recognition 1) Background noise
Speech recognition is a broad term which 2) Speaker characteristics (microphone
means it can recognize almost anybody's characteristics)
speech - such as a call-centre system
designed to recognize many voices. Voice 3) Pause taken between two words.
recognition is a system trained to a
particular user, where it recognizes their 4) How carefully words are pronounced.
speech based on their unique vocal sound.
Now here DSP will be in the picture.
Basically speech recognition involves Considering all above problems, one has to
inputting of information into a computer extract important parameter (template) from
using human voice; & the computer spoken word. After that matching it with
quantizes it & then recognizes human standard template, this operation is
speech. As shown in the fig. Using accurately, done by DSP processors
microphone one can input speech / voice.
13
Speech recognition applications include has earned speaker recognition its
voice dialing (e.g., "Call home"), call classification as a "behavioral biometric."
routing (e.g., "I would like to make a collect
call"), demotic appliance control and C. Speech Coding
content-based spoken audio search (e.g.,
find a podcast where particular words were Speech coding is the application of data
spoken), simple data entry (e.g., entering a compression of digital audio signals
credit card number), preparation of containing speech. Speech coding uses
structured documents (e.g., a radiology speech-specific parameter estimation using
report), speech-to-text processing (e.g., word audio signal processing techniques to model
processors or emails), and in aircraft the speech signal, combined with generic
cockpits (usually termed Direct Voice data compression algorithms to represent the
Input). resulting modeled parameters in a compact
bit stream.
B. Speaker Recognition
The two most important applications of
Speaker recognition is the computing task speech coding are mobile telephony and
of validating a user's claimed identity using Voice over IP.
characteristics extracted from their voices.
The techniques used in speech coding are
There is a difference between speaker similar to that in audio data compression and
recognition (recognizing who is speaking) audio coding where knowledge in
and speech recognition (recognizing what is psychoacoustics is used to transmit only data
being said). These two terms are frequently that is relevant to the human auditory
confused, as is voice recognition. Voice system. For example, in narrowband speech
recognition is combination of the two where coding, only information in the frequency
it uses learned aspects of a speaker’s voice band 400 Hz to 3500 Hz is transmitted but
to determine what is being said - such a the reconstructed signal is still adequate for
system cannot recognize speech from intelligibility.
random speakers very accurately, but it can
reach high accuracy for individual voices it Speech coding differs from other forms of
has been trained with. In addition, there is a audio coding in that speech is a much
difference between the act of authentication simpler signal than most other audio signals,
(commonly referred to as speaker and that there is a lot more statistical
verification or speaker authentication) and information available about the properties of
identification. speech. As a result, some auditory
information which is relevant in audio
Speaker recognition has a history dating coding can be unnecessary in the speech
back some four decades and uses the coding context. In speech coding, the most
acoustic features of speech that have been important criterion is preservation of
found to differ between individuals. These intelligibility and "pleasantness" of speech,
acoustic patterns reflect both anatomy (e.g., with a constrained amount of transmitted
size and shape of the throat and mouth) and data.
learned behavioral patterns (e.g., voice
pitch, speaking style). Speaker verification
14
It should be emphasized that the There are three major components in
intelligibility of speech includes, besides the Radar system.
actual literal content, also speaker identity,
emotions, intonation, timbre etc. that are all i. Antenna
important for perfect intelligibility. The
more abstract concept of pleasantness of ii. Tracking computer
degraded speech is a different property than
intelligibility, since it is possible that iii. Signal processor
degraded speech is completely intelligible,
but subjectively annoying to the listener. Antenna is used to transmit analog signal.
4.2 Application in Radar Tracking computer schedules the
appropriate antenna positions & transmitted
Radar is used to detect stationary/moving signals as a function of time. It also keeps
objects. Radar has a transmitter & receiver. track of important targets & controls the
From transmitter, the signals are generated display in radar.
& transmitted through antenna. If the object
is present signals will hit the target & due to Signal processor performs following
this, portion of the signal is echoed back. functions,
Receiver receives echoed signal, will cancel
noise and amplify the signal. Depending i. Matched filtering
upon the time duration between the
transmitted & received signals, the distance ii. Removal of useless information-threshold
at which the target is located can be detection.
identified.
4.3 DTMF Signal Detection
Dual-tone multi-frequency (DTMF)
signaling is used for telecommunication
signaling over analog telephone lines in the
voice-frequency band between telephone
handsets and other communications devices
and the switching center.
As a method of in-band signaling, DTMF
tones were also used by cable television
broadcasters to indicate the start and stop
times of local commercial insertion points
during station breaks for the benefit of cable
companies. Until better out-of-band
signaling equipment was developed in the
1990s, fast, unacknowledged, and loud
Figure 14. Block Diagram of Modern Radar
DTMF tone sequences could be heard
System.
during the commercial breaks of cable
channels
15
Figure 16. Graph of generated DTMF signal
The original keypads had levers inside, so
each button activated two contacts. The
Figure 15. Telephone keypad in DTMF
multiple tones are the reason for calling the
dialing
system multi frequency. These tones are
The DTMF keypad is laid out in a 4×4 then decoded by the switching center to
matrix, with each row representing a low determine which key was pressed.
frequency, and each column representing a
high frequency. For decoding we use DSP, to be very
specific, we use DFT. Following are the
steps to be followed.
i. Sample DTMF signal
1209 1336 1477 1633 The minimum duration of DTMF signal is
40 msec. So if we sample it with 8 kHz, there
697 Hz 1 2 3 A are at most 0.04 x 8000 = 3200 samples.
4 5 6 B ii. Compute N Point DFT
770 Hz
Now we have to compute n point DFT
852 Hz 7 8 9 C values, of sampled DTMF signals.
Normally, the actual number of samples
941 Hz * 0 # D should be in such a way that it should
minimize the difference between the actual
location of the sinusoid & the nearest integer
Pressing a single key (such as 1) will send value of DFT index K.
a sinusoidal tone for each of the two
frequencies (697 and 1209 hertz ). iii. Compute DFT of 8 frequency tones
We know that in totality we have eight
frequency tones. The DTMF decoder
computes DFT samples closest in frequency
to the 8 DTMF fundamental tones & their
respective second harmonics. After that we
get energy spectrum.
16
In this spectrum, we get very high energy 4.4 Removing vocals from
at index K depending time frequency. The commercial tracks
index K, at high energy level is taken and
compared with standard one. Since the advent of commercial music,
After comparison, we can come out with songs have commonly been identified by the
unique digit or code. singer’s voice. Popular tracks are
distinguishable by an extraordinary lead
F k = K fs / N (K= 1....N-1) voice. Meanwhile, songs with great
instrumental backgrounds can be marred by
an unfortunate selection of vocalists and
lyrics. The goal of this project is to remove
iv. Choosing N
vocals from commercial tracks in order to
In this application N is supposed to be appreciate the underlying instrumental
carefully chosen. We know that, N background. Moreover, the removal of
determines frequency spacing between the vocals has several applications. Karaoke is a
locations of DFT samples. N also common pastime in which the removal of
determines time taken to compute DFT vocals permits the participants to effectively
samples. N is large, time required will be interact with the song. Additionally,
large, but provides resolution in frequency removing vocals makes the production of
domain. One more important parameter i.e. ringtones and remixes easier. Finally, people
spectral leakage, one has to consider while may just wish to hear the track without the
choosing N. The spectral leakage has to be lead singer’s voice. In order to remove
as minimum as possible. vocals from commercial tracks, several
techniques are used. These techniques
include filtering the known (average)
frequency range of the human voice (band
v. Considering error sources stop filtering), cancelling common
frequencies between stereo channels (stereo
If you observe tone frequencies you will
cancellation), and masking a time frequency
find that, spectrum of human voice contains
spectrogram (audio blind source separation).
all these frequencies. Therefore it is very
These techniques are described in detail
important to distinguish between human
below.
voice & DTMF signal. The problem is very
simple, because DTMF signal generates 4.4.A Band stop filtering
pure sine wave with negligible power of
second harmonics. This is not true for The simplest technique for removing
human voice; it will contain second or third vocals is band stop filtering. The human
harmonics. Therefore practical DTMF voice has a distinct frequency range between
decoder computes DFT closest in frequency 300 Hz and 3 kHz. By applying a band stop
to the second harmonics corresponding to filter (figure 16) at these frequencies, most
each of the fundamental tone frequencies. vocals can be removed. Software filtering
This will distinguish between DTMF signal allows us the luxury of implementing a very
& human voice high order filter. Unfortunately, however,
this technique has the side-effect of also
removing any instruments that occupy the
17
same frequency range, such as strings and 4.4.C Audio Blind Source Separation
guitars. This is undesirable.
This technique consists of “extracting
from an input audio signal a set of audio
signals whose mix is perceived similarly to
the original audio signal”. In our case, we
focused on extracting the vocals track from
the mix consisting of the rest of the
instruments. In order to extract the vocals
from the mix, the following steps were
followed:
Figure 17. The result of applying a band i. Selection of song
stop filter A stereo mix has to be chosen, preferably
4.4.B Stereo Cancellation without any reverberation. If the selected
mix has reverberation, the procedure is
Stereo cancellation, as the name implies, complicated considerably because the mono
requires using a stereo sample, and involves tracks overlap with other tracks. Good
subtracting common frequencies between candidate mixes for this technique are old
the two channels (figure 17). This works stereo songs, where complex post-recording
most of the time, because the lead singer’s audio effects are not as frequent.
voice is mostly centre-panned, and therefore
has common frequencies in both channels. The song that we chose is “Let it be” by
the Beatles. In figure 18, we observe both
However, since one channel is subtracted channels of our song. This clearly depicts
from the other, the result of stereo that the left channel (l) does not contain the
cancellation is a mono track. Another same information as the right channel (r).
unwelcome result is the lowering in volume Hence, the song was recorded in stereo
of other sounds that are common between mode.
both channels, such as drums or bass. Still,
the results of this technique are better than
that of simple band stop cancellation
.
Figure 19. The chosen stereo clip
Figure 18. The left channel, before and after
stereo cancellation
18
ii. Short Time Fourier Transform (STFT) The number of DFT points for our STFT
and spectrogram (N) was set to 8192 while the offset among
frames was set to 2048 (N/4). This yields, in
∗236 DFT points (for our specific clip).
After the song was chosen and the the spectrogram, an array of 4097∗ 2+1
channels separated, a spectrogram was
generated for each one of the channels. In Moreover, the window used for the frames is
order to get the spectrogram, we used the the Hamming window, although a Blackman
STFT. window will also yield the same or better
L= STFT (1) (1) results due to a higher attenuation at the stop
band. Figure 19. Shows the spectrograms for
R= STFT (r) (2) both the left and the right channels.
Where, iii. Track Identification
L = Left channel STFT Once we have both spectrograms, we
proceed to recognize similarities between
R = Right Channel STFT both channels. We do this with the purpose
of getting one or more of the tracks that the
The STFT has three important factors to channels share. In our case, the technique
consider: The first factor is the number of that we used is to divide the magnitude of
DFT points (N) that will be generated per
frame; the second factor is the offset among
frames; and third factor is the type of
window that each frame will use.
The first two factors will create a large
number of samples in our spectrogram. This
large number of samples will let us extract
areas were the voice is concentrated with
more accuracy. Also, the window in each
frame is of extreme importance because the
offset among frames is really low. So, if the Figure 21 - Frequency of channel ratio
data is not well confined in each frame, we
each DFT coefficient in the left channel by
will have noise coming from other frames
its counterpart in the right channel
due to overlap. spectrogram. This division gives us a
channel ratio (CR) between both channels.
This is expressed as follows:
CR = |L| / |R| (3)
When you divide a coefficient from the
left channel (that represents a single track)
by a coefficient from the right channel (that
also represents the same single track), the
result will be a constant value no matter
Figure 20. Spectrogram of left and right where we are located in the spectrogram [1].
channel However, if you divide coefficients that
19
represent two or more tracks, your result to zero (blue). Now, it becomes clear why a
will not be constant throughout the greater array of coefficients would yield
spectrogram anymore. In figure 20. We can better results. This is mainly because we can
see the frequency for each ratio in the be more selective in the areas we want to
spectrogram. At a ratio of 1, we found a reduce.
peak. This peak represents a mono track that
was inserted evenly on both channels (if the v. Time-signal recovery–Inverse STFT
track is different to one, this means that the (ISTFT)
mono track in one of the channels was
After the binary mask has been applied,
attenuated or amplified).
the signal has to be transformed back to the
iv. Time Frequency Mask –Binary time domain. In order to do this, the ISTFT
Method is used on each one of the channels.
Once the coefficients that represent mono l= istft (L) (5)
tracks in the channels are identified, we
r= istft (R)
proceed to substitute them by zeros. This
method is usually known as binary masking Where
because the coefficients are multiplied by
either one or by zero (in other, more L= Left channel masked signal before istft
advanced, techniques, the coefficients can
be weighted). This can be seen l = RECOVERED left channel signal
mathematically as follows: R= Right channel masked signal before istft
r = RECOVERED right channel signal
In figures 22 & 21, the original signal and
the recovered signal are compared. It can be
seen that each one of the signals has been
altered from the original one. When the two
recovered signals are mixed again, we get a
stereo clip, but this time without vocals.
Figure 22. Spectrogram with binary mask
M=0 if a < CR < b (4)
1 otherwise
Where, M = Binary mask
Figure 23. Left channel with its recovered
In figure 21, it can be observed that some
counterpart
parts of the spectrograms have been reduced
20
The five steps explained above are one
way to remove the vocals from the song.
However, this technique is not limited to
extracting the vocals from a song; for
instance, we could extract the instruments
and leave the vocals in the song. Therefore,
this technique permits a greater flexibility
compared to the other techniques explained Figure 25. Use of DSP in MP3 player
in this paper.
The DSP performs the MP3 encoding and
Nonetheless, the audio source separation saves the file to memory. During the
technique used in this project is just one of playback phase, the file is taken from
many audio source separation approaches. memory, decoded by the DSP and then
This is mainly because different mixes of converted back to an analog signal through
instruments and new sound effects the digital-to-analog converter so it can be
intermingles frequencies in more complex output through the speaker system. In a
ways. Due to this added complexity, a more complex example, the DSP would
binary mask approach will not be enough to perform other functions such as volume
separate the sources from the song. control, equalization and user interface.
Conclusion
Hence we can conclude by saying that,
DSP forms an alternative to ASP taking into
consideration both their merits & demerits.
In this paper we have presented the basic
elements of a Digital Processing System &
defined the operations required to process a
signal digitally. I have also discussed the
Figure 24. Right channel with its recovered
benefits of using a specialized digital signal
counterpart
processor rather than using a general
microprocessor.
4.5 DSP in MP3 Audio Player
Digital filters operate upon a signal and
The diagram below shows how a DSP is depending on the users requirements change
used in an MP3 audio player. During the the Amplitude-frequency & Phase-
recording phase, analog audio is input frequency characteristics of a signal, so as to
through a receiver or other source. This improve the quality of the signal.
analog signal is then converted to a digital
signal by an analog-to-digital converter and DSP has it’s applications in almost every
passed to the DSP. field related to electronics today. This paper
treats applications to speech processing,
RADAR, DTMF signal detection, Removing
vocals from commercial tracks & MP3
players.
21
22
