| H (ω)| in dB
Table. 1 Optimal FIR lowpass filter
Filter Length = 51
Impulse Response
h(0) -0.0053 h(50)
h(1) 0.0081 h(49)
h(2) 0.0070 h(48)
h(3) -0.0121 h(47)
h(4) -0.0068 h(46)
h(5) 0.0035 h(45) ω in rad/s
h(6) 0.0107 h(44)
Fig. 2: Optimal FIR filter response.
h(7) 0.0064 h(43)
h(8) -0.0061 h(42)
h(9) -0.0137 h(41)
III. ANALYTICAL DESIGN OF LINEAR PHASE FIR FILTER
h(10) -0.0067 h(40) USING BLACKMAN WINDOW
h(11) 0.0099 h(39)
h(12) 0.0186 h(38) Here we begin to determine the impulse response hd (n) [1]
h(13) 0.0073 h(37) from the desired frequency response H d (ω ) . The impulse
h(14) -0.0158 h(36) response sequence is obtained from
h(15) -0.0263 h(35)
sin ω c (n − τ )
h(16) -0.0080 h(34) hd (n ) = n ≠τ
h(17) 0.0260 h(33) π (n − τ )
(5)
h(18) 0.0398 h(32) ωc
= n =τ
h(19) 0.0086 h(31) π
h(20) -0.0479 h(30) N −1
h(21) -0.0711 h(29) where τ = = 25
2
h(22) -0.0090 h(28)
The Blackman window [1] is defined by
h(23) 0.1321 h(27)
2π n 4π n
h(24) 0.2797 h(26) w b (n ) = 0 . 42 − 0 . 5 cos + 0 . 08 cos
N −1 (6) N −1
h(25) 0.3425 h(25)
for 0 ≤ n ≤ N − 1 .
The values of filter impulse response coefficients, h(n) for
Blackman window are following:
Table. 2 Lowpass Filter (Blackman Window)
Filter Length = 51
Blackman Window
Impulse Response
h(0) 0.0000 h(50)
| H(ω) |
h(15) -0.0142 h(35) coefficients. The filter coefficients are then determined as the
h(16) -0.0164 h(34) inverse discrete Fourier transform (IDFT) [3] of this set of
h(17) 0.0058 h(33) samples. For this design technique, the frequency samples are
h(18) 0.0323 h(32) chosen to be
h(19) 0.0242 h(31) 2π k k = 0, 1, 2, …………, N-1
ω k =
h(20) -0.0275 h(30) N
h(21) -0.0712 h(29) Sampling the desired frequency response we get
h(22) -0.0304 h(28) N −1 (7)
H (k ) = ∑ h (n ) e − 2 π kn / N
h(23) 0.1156 h(27) n = 0
h(24) 0.2887 h(26) and finally the impulse response, h(n) is obtained [3] by taking
h(25) 0.3661 h(25) IDFT of Eq. (7).
1 ⎧ ( N −1) / 2 ⎛ 2πk ⎛ 1 ⎞ ⎞⎫
h(n ) = ⎨G (0) + 2 ∑ G (k )cos⎜⎜ ⎜ n + ⎟ ⎟⎟⎬
N⎩ k =1 ⎝ N ⎝ 2 ⎠ ⎠⎭
⎛ 2πk ⎞ N −1
where G (k ) = (− 1)k H ⎜ ⎟ k = 0, 1, ………
⎝ N ⎠ 2
The coefficients of impulse response, h(n ) for frequency
sampling technique are listed in Table 3.
| H (ω)|
and samples positions, one can design the desired FIR filter. on the availability of the design [5].
| H (ω)|
| H (ω)|
ω in rad/s ω in rad/s
Fig. 5: Data for FIR lowpass filter using frequency sampling
Fig. 7: Comparison of frequency responses of FIR lowpass
method.
filters.
| H (ω)| in dB
| H (ω)| in dB
ω in rad/s
Fig. 8: Comparison of frequency responses of FIR lowpass
ω in rad/s filters.
Fig. 6: Frequency response of FIR lowpass filter using
frequency sampling method.
The Remez exchange algorithm is used to find the frequency Fig. 12: Frequency response of FIR lowpass filter using
response of FIR lowpass filter using optimal method. optimal method and Remez Exchange Algorithm (when pass
band has more weight than stop band).
| H (ω)| in dB
| H (ω)|
ω in rad/s
Fig. 10: Frequency response of FIR lowpass filter using
optimal method and Remez Exchange Algorithm (when stop
band and stop band have the same weight). ω in rad/s
Fig. 13: Frequency response of FIR lowpass filter using
optimal method and Remez Exchange Algorithm (when stop
band has more weight than pass band).
| H (ω)|
| H (ω)| in dB
ω in rad/s
Fig. 11: Frequency response of FIR lowpass filter using
optimal method and Remez Exchange Algorithm (when pass ω in rad/s
band has more weight than stop band ). Fig. 14 Frequency response of FIR lowpass filter using
optimal method and Remez Exchange Algorithm (when stop
band has more weight than pass band).
But this broadens the transition band, which reduces the ideal good solution to the filter design problem. The major
properties of a filter. disadvantage of the window design method is the lack of
precise control of the critical frequencies, such as ω p and ω s ,
in the design of a low pass FIR filter and its value depends on
the type of window and the filter length N [3]. Fig. 8 shows
the relative performance of the different filtering methods. The
frequency sampling method provides an improvement over the
window design method, since Hr(ω) is specified at the
frequencies ω k = 2πk / N and transition band is a multiple of
2π / N [3]. This filter design method is particularly attractive
| H (ω)|
REFERENCES
[1] J. R. Johnson, “Introduction to Digital Signal Processing,”4th ed.,
ω in rad/s Prentice-Hall of India Private Limited, 1997, pp. 85-95.
Fig. 15: Frequency response of FIR lowpass filter using [2] J. H. McClellan., T. W. Parks, and L. R. Rabiner, “A Computer
Program for Designing Optimum FIR linear Phase Digital Filters,”
optimal method and Remez Exchange Algorithm (when pass IEEE trans.Audio Electroacoust., AU-21, December 1973, pp. 506-526.
band and stop have same weight but transition band is more). [3] J. G. proakis., D. G. Manolakis., Digital Signal Processing principles,
Algorithms, and Application.” 3rd ed., Prentice-Hall of India Private
Limited, 1997, pp. 620-662.
[4] E. C. Ifeachor, B. W. Jervis, “Digital Signal Processing A Practical
Approach,” 2nd ed., Pearson Education Pte. Ltd. 2003, pp. 342-398.
[5] L. R. Rabiner, B. Gold, and C. A. McGonegal, “An approach to the
approximation problem for nonrecursive digital filters,” IEEE trans.
Audio Electroacoustics, Vol.18, Iss. 2, pp. 83- 106.
| H (ω)| in dB
ω in rad/s
Fig. 16: Frequency response of FIR lowpass filter using
optimal method and Remez Exchange Algorithm (when pass
band and stop have same weight but transition band is more).
VI. CONCLUSION
The design of linear phase optimal FIR filters, with very flat
passbands, can be done by solving the frequency response
equations for different frequency components. To solve the
filter parameters, one can use an iterative algorithm like
Remez exchange algorithm [3]. The response of Remez
algorithm is shown in figure 9. We can also do this, by
sampling the desired frequency response with non uniform
frequency spacing as shown the response in Fig. 1. Here we
see that the degree of flatness in passbands and ripples in
stopbans depend on the transition bandwidth. If we set this
bandwidth low then ripples increases and vice versa. We can
get better stopband attenuation using Blackman window
method. In summary, the optimal solution is not always a