Digital Signal Processing (ECE – 2222)

List of Formulae

1 DFT pair 𝑁−1

1 2𝜋
𝑥(𝑛) = ∑ 𝑋(𝑘)𝑒 𝑗 𝑁 𝑘𝑛 , 𝑛 = 0, 1,2 … … 𝑁 − 1
N x(n) : Signal of length L
𝑘=0
X(k): N-point DFT of x(n)
𝑁−1
2𝜋 N≥L
𝑋(𝑘) = ∑ 𝑥(𝑛)𝑒 −𝑗 𝑁 𝑘𝑛 , 𝑘 = 0, 1,2 … … 𝑁 − 1
𝑛=0

2 N-point 𝑁−1

Circular 𝑥3 (𝑛) = 𝑥1 (𝑛) ⊛ 𝑥2 (𝑛) = ∑ 𝑥1 (𝑚)𝑥2 (𝑛 − 𝑚) 𝑥1 (𝑛), 𝑥2 (𝑛): signals of length N

convolution 𝑚=0 or less
𝑁 𝑀
3 Pole-zero IIR
system 𝑦(𝑛) = − ∑ 𝑎𝑘 𝑦(𝑛 − 𝑘) + ∑ 𝑏𝑘 𝑥(𝑛 − 𝑘)
𝑘=1 𝑘=0

𝑌(𝑧) ∑𝑀
𝑘=0 𝑏𝑘 𝑧
−𝑘
𝐻(𝑧) = =
𝑋(𝑧) 1 + ∑𝑁
𝑘=1 𝑎𝑘 𝑧
−𝑘

𝑁
4 All pole IIR
system 𝑦(𝑛) = ∑ 𝑎𝑘 𝑦(𝑛 − 𝑘) + 𝑏0 𝑥(𝑛) 𝑥(𝑛): 𝑖𝑛𝑝𝑢𝑡
𝑘=1
𝑦(𝑛): 𝑜𝑢𝑡𝑝𝑢𝑡
𝑌(𝑧) 𝑏0
𝐻(𝑧) = = 𝑁
𝑋(𝑧) 1 + ∑𝑘=1 𝑎𝑘 𝑧 −𝑘 𝑋(𝑧): z-transform of 𝑥(𝑛)
𝑀 𝑌(𝑧): z-transform of y(𝑛)
5 All zero FIR
system 𝑦(𝑛) = ∑ 𝑏𝑘 𝑥(𝑛 − 𝑘)
𝑘=0 𝐻(𝑧): System function
𝑀
𝑌(𝑧) 𝐾 : Integer part of (N+1)/2
𝐻(𝑧) = = ∑ 𝑏𝑘 𝑧−𝑘
𝑋(𝑧)
𝑘=0 N≥M
6 Cascade 𝐾
∑𝑀
𝑘=0 𝑏𝑘 𝑧
−𝑘 𝑏
𝐶 = 𝑎𝑁
realization for 𝐻(𝑧) = = ∏ 𝐻𝑘 (𝑧) 𝑁
IIR filters 1 + ∑𝑁
𝑘=1 𝑎𝑘 𝑧
−𝑘
𝑘=1

𝑏𝑘0 + 𝑏𝑘1 𝑧 −1 + 𝑏𝑘2 𝑧 −2

𝐻𝑘 (𝑧) =
1 + 𝑎𝑘1 𝑧 −1 + 𝑎𝑘2 𝑧 −2
 𝐾
7 Parallel ∑𝑀 −𝑘
𝑘=0 𝑏𝑘 𝑧
realization for 𝐻(𝑧) = = 𝐶 + ∑ 𝐻𝑘 (𝑧)
IIR filters 1 + ∑𝑁
𝑘=1 𝑎𝑘 𝑧
−𝑘
𝑘=1

𝑏𝑘0 + 𝑏𝑘1 𝑧 −1
𝐻𝑘 (𝑧) =
1 + 𝑎𝑘1 𝑧 −1 + 𝑎𝑘2 𝑧 −2

8 Frequency 1
𝐻(𝑧) = (1 − 𝑧 −𝑀 )𝐻𝑝 (𝑧)
sampling 𝑀
realization for (𝑀−1)/2
FIR filters 𝐻(0) 𝐴(𝑘) − 𝐵(𝑘)𝑧 −1
𝐻𝑝 (𝑧) = + ∑ 𝑀 𝑜𝑑𝑑
1 − 𝑧 −1 2𝜋𝑘 −1
𝑘=1 1 − 2𝑐𝑜𝑠 ( ) 𝑧 + 𝑧 −2
𝑀
𝐻(𝜔): Frequency response
𝑀
𝐻(0) 𝐻( )
𝐻𝑝 (𝑧) = + 2 2𝜋𝑘
1 − 𝑧 −1 1 + 𝑧 −1 𝐻(𝑘): 𝐻(𝜔) at 𝜔 = 𝜔𝑘 =
𝑀
𝑀
−1
2
𝐴(𝑘) − 𝐵(𝑘)𝑧 −1 M: Length of the FIR filter
+∑ 𝑀 𝑒𝑣𝑒𝑛
2𝜋𝑘 −1
𝑘=1 1 − 2𝑐𝑜𝑠 ( ) 𝑧 + 𝑧 −2
𝑀

𝐴(𝑘) = 𝐻(𝑘) + 𝐻(𝑀 − 𝑘)

2𝜋𝑘 2𝜋𝑘
𝐵(𝑘) = 𝐻(𝑘)𝑒 −𝑗 𝑀 + 𝐻(𝑀 − 𝑘)𝑒 𝑗 𝑀
9 Impulse N: Order of the analog filter
Invariant 𝑧 = 𝑒 𝑠𝑇
𝑁 𝑁 𝑠𝑘 : Poles of 𝐻(𝑠)
Transformation 𝐴𝑘 𝐴𝑘
𝐻(𝑠) = ∑ ≡ 𝐻(𝑧) = ∑
𝑠 − 𝑠𝑘 1 − 𝑒 𝑠𝑘 𝑇 𝑧 −1 𝑧𝑘 : Zeroes of 𝐻(𝑠)
𝑘=1 𝑘=1
−1 )
10 Bilinear 2 (1 − 𝑧
Transformation 𝑠 = 𝑇 (1 + 𝑧−1 ) T: Sampling interval

2 𝜔 Ω: Analog filter frequency variable

Ω= 𝑡𝑎𝑛 ( )
𝑇 2 ω: Digital filter frequency variable
11 Matched
z-transform ∏𝑀
𝑘=1(𝑠 − 𝑧𝑘 ) ∏𝑀
𝑘=1 1 − 𝑒
𝑧𝑘 𝑇 −1
𝑧
𝐻(𝑠) = ≡ 𝐻(𝑧) =
∏𝑁
𝑘=1(𝑠 − 𝑠𝑘 ) 𝑁
∏𝑘=1 1 − 𝑒 𝑠𝑘 𝑇 𝑧 −1

12 Butterworth 1 1
|𝐻(Ω)|2 = 2𝑁 =
analog low Ω Ω 2𝑁
1 + (Ω ) 1 + 𝜖 2 (Ω )
pass filter 𝑐 𝑝
response
1
𝑙𝑜𝑔(
𝛿2 2
−1) N: order of the filter
𝑁= Ω or N ≥ log(d)/ log(k)
2𝑙𝑜𝑔( 𝑠 )
Ω𝑐
Ω𝑝 |𝐻(Ω)|2 : Squared magnitude response
Where d=√(1/𝐴𝑝2 ) − 1/√(1/𝐴𝑠 2 ) − 1 and k = Ω𝑠
Ap and As are absolute values Ω𝑝 : Pass band edge frequency
Ω𝑝 : Stop band edge frequency
𝑠𝑘 = Ω𝑐 𝑒 𝑗∅𝑘
 Ω𝑐 : 3-dB cut-off frequency

13 Chebyshev 1 1
|𝐻(Ω)|2 = 1+𝜖 2
: Pass band edge value of
analog low Ω
1+ 𝜖 2 𝑇𝑁 2 ( ) |𝐻(Ω)|2
pass filter Ω𝑝
𝑐𝑜𝑠(𝑁𝑐𝑜𝑠 𝑥), −1
|𝑥| ≤ 1 𝛿2 2 : Stop band edge value of
response 𝑇𝑁 (𝑥) = { |𝐻(Ω)|2
−1
𝑐𝑜𝑠ℎ(𝑁𝑐𝑜𝑠ℎ 𝑥), |𝑥| > 1
𝑠𝑘 : Poles of H(s)
√1 − 𝛿2 2 + √1 − 𝛿2 2 (1 + 𝜖 2 )
𝑙𝑜𝑔 ∅𝑘 = 𝑝𝑜𝑙𝑒 𝑎𝑛𝑔𝑙𝑒
𝜀𝛿2
𝜋 (2𝑘 + 1)𝜋
𝑁= [ ] = + ,
2 2𝑁
Ω Ω 2 k= 0, 1, 2, ……N-1
𝑙𝑜𝑔 [(Ω 𝑠 ) + √(Ω 𝑠 ) − 1]
𝑝 𝑝

N ≥ cosh-1(1/d)/ cosh-1(1/k)
Ω𝑝
Where d=√(1 − 𝐴𝑝2 ) − 1/√(1 − 𝐴𝑠 2 ) − 1 and k = Ω𝑠

𝑠𝑘 = 𝑟2 𝑐𝑜𝑠(∅𝑘 ) + 𝑗𝑟1 𝑠𝑖𝑛(∅𝑘 )

1/𝑁
𝛽 2 +1 𝛽 2 −1 √1+𝜖2 +1
𝑟1 = Ω𝑝 2𝛽
; 𝑟2 = Ω𝑝 2𝛽
; 𝛽=[ 𝜀
]

14 Frequency Prototype Low pass filter has band edge frequency Ω𝒑

Transformation
for analog Type of Transformation Transformation Band edge frequency of new
filters filter

Low pass Ω𝑝 Ω𝑝𝑛

𝑠→ 𝑠
Ω𝑝𝑛

High pass Ω𝑝 Ω𝑝𝑛 Ω𝑝𝑛

𝑠→
𝑠

Band pass 𝑠 2 + Ω𝑙 Ω𝑢 Ω𝑙 , Ω𝑢
𝑠→ Ω𝑝
𝑠(Ω𝑢 − Ω𝑙 )

Band stop 𝑠(Ω𝑢 − Ω𝑙 ) Ω𝑙 , Ω𝑢

𝑠→ Ω𝑝
𝑠 2 + Ω𝑙 Ω𝑢
15 Frequency Prototype Low pass filter has band edge frequency 𝜔𝑝
Transformation
for digital Type of Transformation Parameters
filters Transformation
𝜔𝑝𝑛 = 𝑏𝑎𝑛𝑑 𝑒𝑑𝑔𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑜𝑓
𝑛𝑒𝑤 𝑓𝑖𝑙𝑡𝑒𝑟
−1
𝑧 − 𝑎
Low pass 𝑧 −1 →
1 − 𝑎𝑧 −1 𝑠𝑖𝑛[(𝜔𝑝 − 𝜔𝑝𝑛 )/2]
𝑎=
𝑠𝑖𝑛[(𝜔𝑝 + 𝜔𝑝𝑛 )/2]

𝜔𝑝𝑛 = 𝑏𝑎𝑛𝑑 𝑒𝑑𝑔𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦 𝑜𝑓

𝑛𝑒𝑤 𝑓𝑖𝑙𝑡𝑒𝑟
𝑧 −1 + 𝑎
High pass 𝑧 −1 →
1 + 𝑎𝑧 −1 𝑐𝑜𝑠[(𝜔𝑝 − 𝜔𝑝𝑛 )/2]
𝑎=
𝑐𝑜𝑠[(𝜔𝑝 + 𝜔𝑝𝑛 )/2]

𝜔𝑙 = 𝑙𝑜𝑤𝑒𝑟 𝑏𝑎𝑛𝑑 𝑒𝑑𝑔𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦

𝜔𝑢 = 𝑢𝑝𝑝𝑒𝑟 𝑏𝑎𝑛𝑑 𝑒𝑑𝑔𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦
−2𝛼𝐾
𝑎1 =
(𝐾 + 1)
−2 −1
𝑧 − 𝑎1 𝑧 + 𝑎 2 (𝐾 − 1)
Band pass 𝑧 −1 → − −2 −1 𝑎2 =
𝑎2 𝑧 − 𝑎1 𝑧 + 1 (𝐾 + 1)
𝑐𝑜𝑠[(𝜔𝑢 + 𝜔𝑙 )/2]
𝛼=
𝑐𝑜𝑠[(𝜔𝑢 − 𝜔𝑙 )/2]
𝜔𝑢 − 𝜔𝑙 𝜔𝑝
𝐾 = 𝑐𝑜𝑡 ( ) 𝑡𝑎𝑛 ( )
2 2

𝜔𝑙 = 𝑙𝑜𝑤𝑒𝑟 𝑏𝑎𝑛𝑑 𝑒𝑑𝑔𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦

𝜔𝑢 = 𝑢𝑝𝑝𝑒𝑟 𝑏𝑎𝑛𝑑 𝑒𝑑𝑔𝑒 𝑓𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦
−2𝛼
𝑎1 =
(𝐾 + 1)
(1 − 𝐾)
𝑧 −2 − 𝑎1 𝑧 −1 + 𝑎2 𝑎2 =
𝑧 −1 → (1 + 𝐾)
Band stop 𝑎2 𝑧 −2 − 𝑎1 𝑧 −1 + 1 𝑐𝑜𝑠[(𝜔𝑢 + 𝜔𝑙 )/2]
𝛼=
𝑐𝑜𝑠[(𝜔𝑢 − 𝜔𝑙 )/2]
𝜔𝑢 − 𝜔𝑙 𝜔𝑝
𝐾 = 𝑡𝑎𝑛 ( ) 𝑡𝑎𝑛 ( )
2 2
16 Linear phase i) Symmetric impulse response, odd length
𝑀−3
FIR filter 2
frequency −𝑗𝜔(
𝑀−1
) 𝑀−1
𝐻(𝜔) = 𝑒 2 2 ∑ ℎ(𝑛)𝑐𝑜𝑠 {𝜔 ( − 𝑛)}
response 2
𝑛=0
[ ]

ii) Symmetric impulse response, even length

𝑀
−1
2
−𝑗𝜔(
𝑀−1
) 𝑀−1
𝐻(𝜔) = 𝑒 2 2 ∑ ℎ(𝑛)𝑐𝑜𝑠 {𝜔 ( − 𝑛)}
2 M: length of the filter
𝑛=0
[ ]
iii) Anti-symmetric impulse response, odd length
𝑀−3 𝐻(𝜔): Frequency response
2
−𝑗𝜔(
𝑀−1
) 𝑀−1
𝐻(𝜔) = 𝑗𝑒 2 2 ∑ ℎ(𝑛)𝑠𝑖𝑛 {𝜔 ( − 𝑛)}
2
𝑛=0
[ ]

iv) Anti-symmetric impulse response, even length

𝑀
−1
2
−𝑗𝜔(
𝑀−1
) 𝑀−1
𝐻(𝜔) = 𝑗𝑒 2 2 ∑ ℎ(𝑛)𝑠𝑖𝑛 {𝜔 ( − 𝑛)}
2
𝑛=0
[ ]
17 Linear phase Window functions for FIR filter design
FIR filter
design using Name of the Window function 0≤n≤M-1 Main Peak Normalized Stop band
window window lobe side transition attenuation
functions width lobe width# (dB)
(dB)

Rectangular 1 4π/M -13 0.9/(M-1) 21

Hanning 2𝜋𝑛 8π/M -32 3.1/(M-1) 44

0.5 − 0.5𝑐𝑜𝑠 ( )
𝑀−1

Hamming 2𝜋𝑛 8π/M -43 3.3/(M-1) 53

0.54 − 0.46𝑐𝑜𝑠 ( )
𝑀−1

Blackman 2𝜋𝑛 12π/M -58 5.5/(M-1) 75

0.42 − 0.5𝑐𝑜𝑠 ( )
𝑀−1
4𝜋𝑛
+ 0.08𝑐𝑜𝑠 ( )
𝑀−1

Keiser* 𝑀−1 2 𝑀−1 2 > 70

𝐼0 [𝛽 √( 2 ) − (𝑛 − 2 ) ]
𝑀−1
𝐼0 [𝛽 ( )]
2

*Keiser window parameters can be controlled by 𝛽. I0[.] is modified Bessel function.

 #Transition width is normalized to 2π or equivalently to sampling frequency Fs

18 Frequency
sampling (𝑀−1)
2
design of FIR 1 2𝜋𝑘 1
filter ℎ(𝑛) = 𝐺(0) + 2 ∑ 𝐺(𝑘)𝑐𝑜𝑠 (𝑛 + ) ,
𝑀 𝑀 2
𝑘=1
M: length of the filter
{ }
𝑀 𝑜𝑑𝑑
𝐻𝑑 (𝜔) : Desired frequency response
𝑀
−1
2
1 2𝜋𝑘 1 2𝜋𝑘
ℎ(𝑛) = 𝐺(0) + 2 ∑ 𝐺(𝑘)𝑐𝑜𝑠 (𝑛 + ) , 𝐺(𝑘) = (−1)𝑘 𝐻𝑟 ( )
𝑀 𝑀 2 𝑀
𝑘=1
{ }
𝑀 𝑒𝑣𝑒𝑛
𝑀−1 𝑗2𝜋𝑛𝑘
1 2
ℎ(𝑛) = 𝑀 [𝐻(0) + 2 ∑𝑘=1 𝑅𝑒 {𝐻(𝑘)𝑒 𝑀 }] M odd
𝑀 𝑗2𝜋𝑛𝑘
1 −1
2
ℎ(𝑛) = 𝑀 [𝐻(0) + 2 ∑𝑘=1 𝑅𝑒 {𝐻(𝑘)𝑒 𝑀 }] M even

𝐻𝑑 (𝜔) = 𝐻𝑟 (𝜔)𝑒 −𝑗𝜔(𝑀−1)/2

2𝜋𝑘
Where ω= 𝑀
−𝑗2𝜋𝑘(𝑀−1)
𝐻(𝑘) = 𝐻𝑟 (𝑘)𝑒 𝑀 2

19 Non-
parametric
power Estimate Quality factor Computational
spectrum requirement Quality factor
estimators {𝐸[𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒]}2
𝑁 0.9 𝑄=
𝑣𝑎𝑟[𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒]
log 2
Bartlett 𝑄𝐵 2 ∆𝑓
= 1.1𝑁∆𝑓 ∆𝑓: Frequency resolution

Welch 5.12 N: Data frame length

𝑁 log 2
(50% 𝑄𝑊 ∆𝑓
overlap) = 1.39𝑁∆𝑓

1.28
𝑁 log 2
Blackman- 𝑄𝐵𝑇 ∆𝑓
Tukey = 2.34𝑁∆𝑓