1 / 30

ELEN E4810: Digital Signal Processing Topic 9: Filter Design: FIR

ELEN E4810: Digital Signal Processing Topic 9: Filter Design: FIR. Windowed Impulse Response Window Shapes Design by Iterative Optimization. 1. FIR Filter Design. FIR filters no poles (just zeros) no precedent in analog filter design Approaches windowing ideal impulse response

terri
Download Presentation

ELEN E4810: Digital Signal Processing Topic 9: Filter Design: FIR

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. ELEN E4810: Digital Signal ProcessingTopic 9: Filter Design: FIR • Windowed Impulse Response • Window Shapes • Design by Iterative Optimization Dan Ellis

  2. 1. FIR Filter Design • FIR filters • no poles (just zeros) • no precedent in analog filter design • Approaches • windowing ideal impulse response • iterative (computer-aided) design Dan Ellis

  3. Least Integral-Squared Error • Given desired FR Hd(ejw), what is the best finite ht[n] to approximate it? • Can try to minimize Integral Squared Error (ISE) of frequency responses: best in what sense? = DTFT{ht[n]} Dan Ellis

  4. Least Integral-Squared Error • Ideal IR is hd[n] = IDTFT{Hd(ejw)}, (usually infinite-extent) • By Parseval, ISE • But: ht[n] only exists for n = -M..M , minimized by making ht[n] = hd[n], -M ≤ n ≤ M not altered byht[n] Dan Ellis

  5. Least Integral-Squared Error • Thus, minimum mean-squared error approximation in 2M+1 point FIR is truncated IDFT: • Make causal by delaying by M points h't[n] = 0 forn < 0 Dan Ellis

  6. w -p -wc wc p Approximating Ideal Filters • From topic 6,ideal lowpasshas:and: (doubly infinite) Dan Ellis

  7. Approximating Ideal Filters • Thus, minimum ISEcausal approximation to an ideal lowpass Causal shift 2M+1 points Dan Ellis

  8. Freq. Resp. (FR) Arithmetic • Ideal LPF has pure-real FR i.e.q(w) = 0, H(ejw) = |H(ejw)|  Can build piecewise-constant FRs by combining ideal responses, e.g. HPF: d[n] i.e. H(ejw) = 1 1 w – HLP(ejw) = 1 for |w| < wc wouldn’t work if phases were nonzero! hLP[n] 1 w = hHP[n] 1 = d[n] - (sinwcn)/pn w -p -wc wc p Dan Ellis

  9. Gibbs Phenomenon • Truncated ideal filters have Gibbs’ Ears: Increasing filter lengthnarrower ears(reduces ISE)but height the same (11% overshoot) not optimal by minimax criterion Dan Ellis

  10. Where Gibbs comes from • Truncation of hd[n] to 2M+1 points is multiplication by a rectangular window: • Multiplication in time domain is convolution in frequency domain: wR[n] Dan Ellis

  11. Where Gibbs comes from • Thus, FR of truncated response is convolution of ideal FR and FR of rectangual window (pdc.sinc): Hd(ejw) Ht(ejw) DTFT{wR[n]}periodic sinc... Dan Ellis

  12. Where Gibbs comes from • Rectangular window: • Mainlobe width (1/L) determinestransition band • Sidelobe heightdetermines ripples  doesn’t vary with length “periodic sinc” Dan Ellis

  13. 2. Window Shapes for Filters • Windowing (infinite) ideal response FIR filter: • Rectangular window has best ISE error • Other “tapered windows” vary in: • mainlobe transition band width • sidelobes size of ripples near transition • Variety of ‘classic’ windows... Dan Ellis

  14. Window Shapes for FIR Filters bigsidelobes! • Rectangular: • Hann: • Hamming: • Blackman: double widthmainlobe reduced 1stsidelobe triple width mainlobe Dan Ellis

  15. 2p 2M+1 Window Shapes for FIR Filters • Comparison on dB scale: Dan Ellis

  16. Adjustable Windows • Have discrete main-sidelobe tradeoffs... • Kaiser window = parametric, continuous tradeoff: • Empirically, for min. SB atten. of a dB: modified zero-order Bessel function requiredorder transitionwidth Dan Ellis

  17. H(ejw) w p4 kHz 2p 8 kHz 0.15 p600 Hz Windowed Filter Example • Design a 25 point FIR low-pass filter with a cutoff of 600 Hz (SR = 8 kHz) • No specific transition/ripple req’s  compromise: use Hamming window • Convert the frequency to radians/sample: Dan Ellis

  18. Windowed Filter Example • Get ideal filter impulse response: • Get window: Hamming @ N = 25 M = 12 (N = 2M+1) • Apply window: M Dan Ellis

  19. 3. Iterative FIR Filter Design • Can derive filter coefficients by iterative optimization: • Gradient descent / nonlinear optimiz’n desiredresponse H(ejw) Goodness of fit criterion  error e Filter coefs h[n] Update filterto reduce e Estimate derivatives e/h[n] Dan Ellis

  20. errormeasurement region exponent:2  least sq’s  minimax actualresponse error weighting desiredresponse Error Criteria = W(w)·[D(ejw) – H(ejw)] Dan Ellis

  21. Minimax FIR Filters • Iterative design of FIR filters with: • equiripple (minimax criterion) • linear-phase  symmetric IR h[n] = (–)h[-n] • Recall, symmetric FIR filters have FR with pure-real i.e. combo of cosines of multiples of w (type I) n M a[0] = h[M]a[k] = 2h[M - k] Dan Ellis

  22. Minimax FIR Filters • Now, cos(kw) can be expressed as a polynomial in cos(w)k and lower powers • e.g. cos(2w) = 2(cosw)2 - 1 • Thus, we can find a’s such that • Mth order polynomial in cosw • a[k]s are simply related to a[k]s Mth order polynomial in cosw Dan Ellis

  23. Minimax FIR Filters Mth order polynomial in cosw • An Mth order polynomial has at most M - 1 maxima and minima: has at most M-1 min/max (ripples) 1 5th order polynomial in cosw 3 2 4 Dan Ellis

  24. Alternation Theorum • Key ingredient to Parks-McClellan: is the unique, best, weighted-minimax order 2M approx. to D(ejw) • has at least M+2 “extremal” freqs over w subset R • error magnitude is equal at each extremal: • peak error alternates in sign:  Dan Ellis

  25. Alternation Theorum • Hence, for a frequency response: • Ife(w) reaches a peak error magnitude e at some set of extremal frequenceswi • And the sign of the peak error alternates • And we have at least M+2 of them • Thenoptimal minimax (10th order filter, M = 5) Dan Ellis

  26. Alternation Theorum • By Alternation Theorum, M+2extrema of alternating signs  optimal minimax filter • But has at most M-1 extrema  need at least 3 more from band edges • 2 bands give 4 band edges can afford to “miss” only one • Alternation rules out transition band edges, thus have 1 or 2 outer edges Dan Ellis

  27. Alternation Theorum • For M = 5 (10th order): • 8 extrema (M+3, 4 band edges) - great! • 7 extrema (M+2, 3 band edges) - OK! • 6 extrema (M+1, only 2 transition band edges)NOT OPTIMAL Dan Ellis

  28. Parks-McClellan Algorithm • To recap: • FIR CAD constraintsD(ejw), W(w)  e(w) • Zero-phase FIRH(w) = SkakcoskwM-1 min/max • Alternation theorumoptimal ≥M+2 pk errs, alter’ng sign • Hence, can spot ‘best’ filter when we see it – but how to find it? ~ Dan Ellis

  29. Parks-McClellan Algorithm ~ ~ • Alternation [H(w)-D(w)]/W(w) must = ±e at M+2 (unknown) frequencies {wi}... • Iteratively update h[n] with Remez exchange algorithm: • estimate/guess M+2 extremals {wi} • solve for a[n], e ( h[n] ) • find actual min/max in e(w)  new {wi} • repeat until |e(wi)|is constant • Converges rapidly! Dan Ellis

  30. Parks-McClellan Algorithm • In Matlab, >> h=remez(10, [0 0.4 0.6 1], [1 1 0 0], [1 2]); filter order (2M) band edges ÷ p desired magnitudeat band edges error weightsper band Dan Ellis

More Related