Signal Processing and Linear Systems-B.P.Lathi copy

Let ykj b e t he linear convolution of t hese

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ler processing delay. We shall now discuss two such m ethod o f block processing, o verlap a nd add m ethod a nd o verlap a nd s ave m ethod. E ither o f t he m ethods r equires t he s ame n umber o f c omputations; hence, which m ethod is used is a m atter o f choice. Overlap a nd Add M ethod C onsider a filtering operation, where t he o utput y[k] is t he c onvolution of t he i nput J[k] a nd a F IR filter impulse response h[k] o f l ength M . I n p ractice, usually t It c an b e shown t hat i n convolution o f t wo sequences, each o f l ength No, t he n umber o f computations r equired is o n t he o rder o f N6, w hereas for O FT, u sing F FT a lgorithm, t he n umber o f c omputations r equired is only o n t he o rder o f No log2 No. 6 56 10.6 10 Fourier Analysis o f D iscrete- Time S ignals Nonoveriapping data blocks L Zero-padded L L ( a) I [k) JL L data blocks N o.1 L o 0 12345678 N o.2 I k_ 1M-II L 3 to 1M-II o=r\-I'-I-I• ; =1 \ 61 10 1 10 • ( c) L -M+I N o.2 1M-II L -M+I N o.3 Final output 012345678 M is m uch s maller t han t he l ength o f t he i nput. F igure 10.15 i llustrates a l ong i nput s equence sectioned i nto n onoverlapping blocks o f a m anageable l ength L . T he n umber i nside each block i ndicates i ts l ength. L et u s a ssume t hat L » M . We n ow p rocess each block o f t he i nput d ata in sequence. T o b e a ble t o u se t he c ircular c onvolution for performing t he l inear convolution, we n eed t o p ad M - 1 z eros a t t he e nd o f e ach d ata block.:} F igure 10.15 shows e ach d ata b lock a ugmented b y M - 1 zeros. T he a ugmented p ortion o f t he b lock is s hown s haded. O bserve t hat t he a ugmented ( zero-padded) blocks o f t he i nput, e ach o f l ength L + M - 1, now o verlap. T he o utput s equence c orresponding t o e ach block also has a l ength L + M - 1 ( recall t hat t he l ength o f t he c ircular c onvolution o f t wo sequences, e ach o f l ength L +M - 1, is also L +M - 1). T he o utput sequences, therefore, also overlap, a s s hown i n F ig. 10.15. T he t otal o utput is g iven by t he s um o f a ll t hese o verlapping o utput b locks o f l ength L + M - 1. T he c ontents o f t he t wo successive blocks a re a dded w herever t hey o verlap. T his m ethod is known a s t he overlap a nd a dd m ethod. • E xample 1 0.13 Using overlap and add method of block filtering, find the response y[k] of a n LTID system, whose impulse response h[k] and the input f [k] are shown in Fig. 10.16. T he o utput y[k] is a linear convolution of f [k] a nd h[k]. Let us use L = 3 for t he block convolution. Also M = 2. Hence, we need to break t he i nput sequence in blocks of 3 digits and pad each block with M - 1 = 1 zero, as depicted in Fig. 10.16. We convolve each of these blocks with h[k] using DFT, as demonstrated in Example 10.11. I n this case, No = 4 and 110 = -rr/4. The DFTs Fr a nd Hr of the zero-padded sequences f [k] a nd h[k] are given by 2 = LJ[k]e-ir~k k=O I and Hr IIII2G1II . . . .:--2-1 ---.4...
View Full Document

This note was uploaded on 04/14/2013 for the course ENG 350 taught by Professor Bayliss during the Spring '13 term at Northwestern.

Ask a homework question - tutors are online