Mester IMPERIAL COLLEGE LONDON, DEPARTMENT OF ELECTRICAL AND ELECTRONIC ENGINEERING EXAMINATIONS 2004 EEE/ISE PART lIVV: MEng, BEng and ACGI DIGITAL SIGNAL PROCESSING Wednesday, 12 May 10:00 am Time allowed: 3:00 hours There are SIX questions on this paper. Answer FOUR questions. All questions carry equal marks Any special instructions for invigilators and information for candidates are on page 1. Examiners responsible First Marker(s) P.A. Naylor Second Marker(s) : D.P. Mandic © University of London 2004 3.07 Aut 1SE3.11E3.07/AUMISE3.11 Special Instructions for Invigilators: None Information for Candidates: Sequence mata din) 1 we a™u(n) 5 (cosa a Table 1 : z-transform pi 5(m) is defined to be the unit impulse function, (1) is defined to be the unit step function. Numbers in square brackets against the right margin of the following pages are a guide to the marking scheme, Digital Signal Processing Page 1 of 6E3,07/AUWISES.11 (a) Give a summary description of the processing steps of the overlap-add block filtering 8] method for convolution of a time series x(12) with a system impulse response h(t) . Be precise in your description and include diagrams where appropriate. (b) Use the overlap-add block fitering method to calculate the output of a filter with, [12] impulse response h(n) =[1, 0,1] and input signal x(n) =[1,2,3,-3,-2,-1,0,-1 ] Use only the first 2 data blocks to calculate the output, Show your method clearly and include any relevant diagrams. {Hint: your solution must employ, and be illustrative of, the overlap-add block fitter approach but consider carefully how the internal calculations can be performed most efficiently for this particular case of (72) Let X(k) be the DFT of the real discrete-time sequence x(x) of length N. (a) Describe what is meant by the term basis function in the context of the DFT. Write (5) down the basis functions corresponding to a 4-point DFT and compute the samples of each basis function (b) Show that (k) can be real under certain conditions on x(1t). State these 8 conditions and give an illustrative example. (©) W x(n) satisfies x(n) = = x(M + n) with N = 2M , show that X(2I) = 0 for 5} 1=0.K.M-1 (a) State any conditions on the magnitude and phase of X(k) that must be satisfied for [5} x(n) to be real and draw a labelled sketch of the magnitude and phase of a non- trivial example of such a funetion X (k) Digital Signal Processing Page 2 of 6E3.07/AUMISES.11 (2) Figure 1 shows multirate processing blocks for decimation and expansion. Derive expressions for Y,,(e") and ¥,(e") in terms of X(e"), M and L. Describe the effect of each multirate processing block in the frequency domain using one or two sentences and an illustrative diagram for each xn) tu yon) x(n) tL den) M-fold decimator L-fold expander Figure 1 (b) Write down an expression for ¥(=) in terms of X(z) for the discrete-time system of Figure 2, td X(2) ta | [ Ha fe] Ya) Figure 2 Part of the magnitude spectrum corresponding to X(z) is shown in Figure 3. Draw a labelled sketch of the magnitude spectrum corresponding to Y(z) over the range of frequencies ~ 2 <@<6r . Briefly explain the main features of your sketch xi) Bis Figure 3 c) Draw and label the implementation of the structure of Figure 2 using Type 1 7 polyphase decomposition. State any advantages or disadvantages of your polyphase implementation compared to the structure of Figure 2. Digital Signal Processing Page 3 of 6E3.07/AUW/ISES.11 (2) Draw a labelled sketch covering the range ~ 2 @ ~ 70, | W, 4 “PY » 4 20 ”- % an The impulse response of the ideal fter can be found using the inverse DTFT. A simple approach is to determine the impulse response for a lowpass filter of the same bandwidth and then modulate this impulse response by e”” giving 3 sin pnt | 4) nr) = SE SAGPAT TD yy = amt ae" Itis not possible to implement this fiter in a real-time system because: (i) its impulse response is of infinite duration (id its impulse response is non-causal 5 (b) , 19} His)= 2? -2azba? +b? For zeros on the unit circle: a? +b? =2az+! ‘The magnitude response will have a single null corresponding to the zeros on the unit circle at frequency p/ 3 10) ‘magnitude response dB 8 8 02 03 04 05 06 07 08 o9 1 ‘normalized freq Digital Signal Processing Page 5 of 8‘The output sequence is found by convolution with the impulse respone h(n)={1 -1 1). a] x(n) o lo 0.866 0.866 ete. Digital Signal Processing Page 6 of 8‘+ ztransform of the impulse response contains both zeros and poles recursive structure (output depends on previous outputs and also current and previous inputs) can be unstable. This happens if any poles outside izl=1 can achieve high frequency selectivity with fewer coefficients compared to FIR non-linear phase typically (dispersive) implementation in fixed point requires careful treatment for rounding errors (limit cycles, precision) ») Starting from a prototype analogue filter H(s) normalized for a comer frequency of unity, pre-warping is applied to set the effective corner frequency @, to the specified value: Pre-warping: @, = tan| Frequency sealing Is then applied to convert the filter from a comer frequency of unity to the desired comer frequency @, : 9 is replaced by @ The bilinear transform is then applied to H(s) to obtain H(z) 9 So Pre-warp: @ oo = 0.72654 \2x8000 Scaling: replaced by —- gives @. I+(s/@) ots 05278648" H(s)= Le(sfa)#(sf@) @ +80 +3" 0.52786+0.72654s4 (1+0.52786)z? +2(0.52786-1) z+ (0.52786+1) © (14052786 +0.72654) = +2(0.52786-1) <+0.52786 1.52786 -0.944282"' + 1.527862" 44 -0.944282' + 0.8015 Digital Signal Processing Page 7 of 8¥ A xine" at ! OXEK" le vn) = Region of Convergence is that part of the z-plane for which X (z) is a convergent series. ROC must include the unit circle in z for stability. 43 re02* 745 vin) = din) + 4.6 0.2% lain = I) + 3.6 5)" Nain =D This is an unstable system because in order to be stable causal the ROC must included the unit circle. In this case the ROC is 121 > 5. ° Wo psa-q 2 qaprldtdyr 3 4) taxd,z? -qe'd, -qz Nedyc? +d, l4d,z"' +d, ye(e-get+q dz! dye +d, I4d,z'+d,) (=z This is an allpass fiter Digital Signal Processing Page 8 of 8