Discrete-time signals A discrete-time signal is represented as a sequence of numbers: Here n is an integer, and x[n] is the nth sample in the

sequence. Discrete-time signals are often obtained by sampling continuous-time signals. n this case the nth sample of the sequence is equal to the !alue of the analogue signal xa"t# at time t $ n%: %he sampling period is then equal to %, and the sampling frequency is fs $ &$% . x[&]

'or this reason, although x[n] is strictly the nth number in the sequence, (e often refer to it as the nth sample. )e also often refer to *the sequence x[n]+ (hen (e mean the entire sequence. Discrete-time signals are often depicted graphically as follo(s:

"%his can be plotted using the ,A%-A. function stem.# %he !alue x[n] is unde/ned for noninteger !alues of n. 0equences can be manipulated in se!eral (ays. %he sum and product of t(o sequences x[n] and y[n] are de/ned as the sample-by-sample sum and product respecti!ely. ,ultiplication of x[n] by a is de/ned as the multiplication of each sample !alue by a. A sequence y[n] is a delayed or shifted !ersion of x[n] if (ith n1 an integer. %he unit sample sequence

is

defined

as

%his sequence is often referred to as a discrete-time impulse, or 2ust impulse. t plays the same role for discrete-time signals as the Dirac delta function does for continuous-time signals. Ho(e!er, there are no mathematical complications in its defnition. An important aspect of the impulse sequence is that an arbitrary sequence can be represented as a sum of scaled, delayed impulses. 'or

example, the

sequence represented as

can

be

n general, any sequence can be expressed as

%he unit step sequence

is defined as

%he unit step is related to the impulse by Alternati!ely, this can be expressed as

3on!ersely, the unit sample sequence can be expressed as the /rst bac4(ard difference of the unit step sequence 5xponential sequences are important for analysing and representing discrete-time systems. %he general form is f A and / are real numbers then the sequence is real. f 1 6 / 6 & and A is positi!e, then the sequence !alues are positi!e and decrease (ith increasing n:

'or & 6 / 6 1

the sequence alternates in sign, but decreases in magnitude. 'or 2/2 7 & the sequence gro(s in magnitude as n increases. A sinusoidal sequence

has the form

%he frequency of this complex sinusoid is 81, and is measured in radians per sample. %he phase of the signal is . %he index n is al(ays an integer. %his leads to some important di/erences bet(een the properties of discrete-time and continuous-time complex exponentials: 3onsider the complex exponential (ith frequency %hus the sequence for the complex exponential (ith frequency is exactly the same as that for the complex exponential (ith frequency ,ore generally, complex exponential sequences (ith frequencies from one (here r is an integer, are indistinguishable another. 0imilarly, for sinusoidal sequences

n the continuous-time case, sinusoidal and complex exponential sequences are al(ays periodic. Discrete-time sequences are periodic "(ith period 9# if x[n] $ x[n : 9] for all n: %hus the discrete-time sinusoid is only periodic if (hich requires that %he same condition is required for the complex exponential sequence to be periodic. %he t(o factors 2ust described can be combined to reach the conclusion that there are only 9 distinguishable frequencies for (hich the corresponding sequences are periodic (ith period 9. ;ne such set is

Discrete-time systems

A discrete-time system is de/ned as a transformation or mapping operator that maps an input signal x[n] to an output signal y[n]. %his can be denoted as

5xample: deal delay

Memoryless systems A system is memoryless if the output y[n] depends only on x[n] at the same n. 'or example, y[n] $ "x[n]#< is memoryless, but the ideal delay Linear systems A system is linear if the principle of superposition applies. %hus if y&[n]

is the response of the system to the input x&[n], and y<[n] the response to x<[n], then linearity implies Additivity:

Scaling: %hese properties combine to form the general principle of superposition n all cases a and b are arbitrary constants. %his property generalises to many inputs, so the response of a linear system to Time-invariant systems A system is time in!ariant if a time shift or delay of the input sequence causes a corresponding shift in the output sequence. %hat is, if y[n] is the response to x[n], then y[n -n1] is the response to x[n -n1]. 'or example, the accumulator system

is time in!ariant, but the compressor system for , a positi!e integer "(hich selects e!ery ,th sample from a sequence# is not. Causality A system is causal if the output at n depends only on the input at n and earlier inputs. 'or example, the bac4(ard difference system is causal, but the for(ard difference system is not. Stability A system is stable if e!ery bounded input sequence produces a bounded output sequence:

x[n]

is an example of an unbounded system, since its response to the unit

(hich has no /nite upper bound. Linear time-invariant systems If the linearity property is combined (ith the representation of a general sequence as a linear combination of delayed impulses, then it follo(s that a linear time-in!ariant "-% # system can be completely characterised by its impulse response. 0uppose h4[n] is the response of a linear system to the impulse h[n -4] at n $ 4. 0ince

f the system is additionally time in!ariant, then the response to /[n -4] is h[n -4]. %he pre!ious equation then becomes

%his expression is called the con!olution sum. %herefore, a -% system has the property that gi!en h[n], (e can /nd y[n] for any input x[n]. Alternati!ely, y[n] is the con!olution of x[n] (ith h[n], denoted as follo(s: %he pre!ious deri!ation suggests the interpretation that the input sample at n $ 4, represented by is transformed by the system into an output sequence . 'or each 4, these sequences are superimposed to yield the o!erall output sequence: A slightly di/erent interpretation, ho(e!er, leads to a on!enient computational form: the nth !alue of the output, namely y[n], is obtained by multiplying the input sequence "expressed as a function of 4# by the sequence (ith !alues h[n-4], and then summing all the !alues of the products x[4]h[n-4]. %he 4ey to this method is in understanding ho( to form the sequence h[n -4] for all !alues of n of interest. %o this end, note that h[n -4] $ h[- "4 -n#]. %he sequence h[-4] is seen to be equi!alent to the sequence h[4] reected around the origin

0ince the sequences are nono!erlapping for all negati!e n, the output must be =ero y[n] $ 1> n 6 1:

%he Discrete 'ourier %ransform %he discrete-time 'ourier transform "D%'%# of a sequence is a continuous function of 8, and repeats (ith period </. n practice (e usually (ant to obtain the 'ourier components using digital computation, and can only e!aluate them for a discrete set of frequencies. %he discrete 'ourier transform "D'%# pro!ides a means for achie!ing this. %he D'% is itself a sequence, and it corresponds roughly to samples, equally spaced in frequency, of the 'ourier transform of the signal. %he discrete 'ourier transform of a length 9 signal x[n], n $ 1> &> : : : >9 -& is gi!en by

An important property of the D'% is that it is cyclic, (ith period 9, both in the discretetime and discrete-frequency domains. 'or example, for any integer r,

since 0imilarly, it is easy to sho( that x[n : r9] $ x[n], implying periodicity of the synthesis equation. %his is important ? e!en though the D'% only depends on samples in the inter!al 1 to 9 -&, it is implicitly assumed that the signals repeat (ith period 9 in both the time and frequency domains. %o this end, it is sometimes useful to de/ne the periodic extension of the signal x[n] to be %o this end, it is sometimes useful to de/ne the periodic extension of the signal x[n] to be x[n] $ x[n mod 9] $ x[""n##9]: Here n mod 9 and ""n##9 are ta4en to mean n modulo 9, (hich has the !alue of the remainder after n is di!ided by 9. Alternati!ely, if n is (ritten in the form n $ 49 : l for 1 6 l 6 9, then n mod 9 $ ""n##9 $ l:

t is sometimes better to reason in terms of these periodic extensions (hen dealing (ith the D'%. 0pecifically, if @[4] is the D'% of x[n], then the in!erse D'% of @[4] is Ax[n]. %he signals x[n] and Ax[n] are identical o!er the inter!al 1 to 9 &, but may differ outside of this range. 0imilar statements can be made regarding the transform @f[4]. Broperties of the D'% ,any of the properties of the D'% are analogous to those of the discrete-time 'ourier transform, (ith the notable exception that all shifts in!ol!ed must be considered to be circular, or modulo 9. Defining the D'% pairs and

-inear con!olution of t(o finite-length sequences 3onsider a sequence x&[n] (ith length - points, and x<[n] (ith length B points. %he linear con!olution of the sequences,

%herefore - : B & is the maximum length of xC[n] resulting from the linear con!olution. %he 9-point circular con!olution of x&[n] and x<[n] is

t is easy to see that the circular con!olution product (ill be equal to the linear on!olution product on the inter!al 1 to 9 & as long as (e choose 9 - - : B :&. %he process of augmenting a sequence (ith =eros to ma4e it of a required length is called =ero padding. 'ast 'ourier transforms %he (idespread application of the D'% to con!olution and spectrum analysis is due to the existence of fast algorithms for its implementation. %he class of methods are referred to as fast 'ourier transforms "''%s#. 3onsider a direct implementation of an D-point D'%:

f the factors ha!e been calculated in ad!ance "and perhaps stored in a loo4up table#, then the calculation of @[4] for each !alue of 4 requires D complex multiplications

and E complex additions. %he D-point D'% therefore requires D F D multiplications and DF E additions. 'or an 9-point D'% these become 9< and 9"9 - &# respecti!ely. f 9 $ &1<G, then approximately one million complex multiplications and one million complex additions are required. %he 4ey to reducing the computational complexity lies in the obser!ation that the same !alues of x[n] are efecti!ely calculated many times as the computation proceeds ? particularly if the transform is long. %he con!entional decomposition in!ol!es decimation-in-time, (here at each stage a 9-point transform is decomposed into t(o 9$<-point transforms. %hat is, @[4] can be (ritten as @[4] $9

%he original 9-point D'% can therefore be expressed in terms of t(o 9$<-point D'%s. %he 9$<-point transforms can again be decomposed, and the process repeated until only <-point transforms remain. n general this requires log<9 stages of decomposition. 0ince each stage requires approximately 9 complex multiplications, the complexity of the resulting algorithm is of the order of 9 log< 9. %he difference bet(een 9< and 9 log< 9 complex multiplications can become considerable for large !alues of 9. 'or example, if 9 $ <1GD then 9<$"9 log< 9# / <11. %here are numerous !ariations of ''% algorithms, and all exploit the basic redundancy in the computation of the D'%. n almost all cases an ;f the shelf implementation of the ''% (ill be sufficient ? there is seldom any reason to implement a ''% yourself. 0 ome forms of digital filters are more appropriate than others (hen real-(orld effects are considered. %his article loo4s at the effects of finite (ord length and suggests that some implementation forms are less susceptible to the errors that finite (ord length effects introduce. n articles about digital signal processing "D0B# and digital filter design, one thing H!e noticed is that after an in-depth de!elopment of the filter design, the implementation is often 2ust gi!en a passing nod. Ieferences abound concerning digital filter design, but surprisingly fe( deal (ith implementation. %he implementation of a digital filter can ta4e many forms. 0ome forms are more appropriate than others (hen !arious real-(orld effects are considered. %his article examines the effects of finite (ord length. t suggests that certain implementation forms are less susceptible than others to the errors introduced by finite (ord length effects.