Compensated CIC-Cosine Decimation Filter 8 ECTI TRANSACTIONS ON COMPUTER AND INFORMATION TECHNOLOGY VOL.4, NO.1 MAY 2010 Gordana Jovanovic Dolecek and Fernando Javier Trejo Torres Compensated CIC-Cosine Decimation Institute INAOE, Department of Electronics, Puebla, Mexico Filter Gordana Jovanovic Dolecek and Fernando Javier Trejo Torres, Non-members structure of (1) can be used [2], [3]. More details on a of the performances of the recursive and nonThis paper presents efficient modification of the CIC-datacomparison rate resulting in a larger chip higher ABSTRACT recursive implementations are area givenand in [2]. Additionally, cosine decimation filter. The second order compensatorpower consumption especially when the decimation This paper presents efficient modification of the the magnitude response of the CIC filter has filter is introduced at low rate in order to improve thefactor is high. In order to resolve this problem thea high droop CICcosine decimation filter. The second order comin the desired passband and low stopband attenuation. passband of interest of the overall filter. The structure of (1) can be used [2], [3]. pensator filter is introduced at low rate in coefficients order to non-recursive The decimation factor Rof of next decimation stage of the compensator filter are presented in a canonical More details on a comparison the the performances of improve the passband of interest of the overall fildetermines the frequency ω , where the worst passband c are signed form,ofand be implemented usingthe recursive and nonrecursive implementations ter. digits The (CSD) coefficients thecan compensator filter are ω , where distortion occurs, and the frequency given in [2]. Additionally, the magnitude response of the worst A onlypresented adders and shifts. Consequently, the resulting filter in a canonical signed digits (CSD) form, the CIC filter has a high droop in the desired passωc/π case aliasing occurs [4]. The normalized frequencies can be implemented onlyaadders and shifts. in is a and multiplierfree filter andusing exhibits high attenuation band and low stopband attenuation. The decimation Consequently, the as resulting filter is adroop. multiplierfree and ωA/π are at the stopband, as well a low passband decimation stage determines the filter and exhibits a high attenuation in the stopband, factor Rωofc the next ω Al 1 2 1 2R − 1 π 1 the worst frequency ωc=, where passband . (2) ⋅ = = distortion − = as well as a low passband droop. Keywords: Decimation filter, Compensator, CIC filter, occurs, and π theMR π MR π theMworst MRcase MR frequency ωA , where Cosine filter. occursstructures [4]. The normalized Several have beenfrequencies proposed ωto c /πdesign CIC Keywords: Decimation filter, Compensator, CIC aliasing and filters ωA /π are at improved frequency response as for example with filter, Cosine filter. 1. INTRODUCTION [4-11]. Usually methods deal to improve either passband 1. INTRODUCTION The process of converting the given rate of a signal ωc [5-6],π or1stopband 1 ω[7-10]. 2 1 2R-1 Al = . = = work − is = . (2) a simple The goal of this to propose into a The different is called the sampling rateof conversion processrate of converting given rate a signal π MR π MR π M MR MR (SRC). reduction of a sampling sampling is called multiplierless decimation filter with no filtering at high into aThe different rate is called raterate conversion (SRC). The of a sampling is called decimation, and reduction is accomplished in two rate stages, filtering input rate, and with a desired high stopband attenuation decimation, and is accomplished in two stages, fil- Several andstructures a desired low passband ToCIC this end we have been proposeddroop. to design and downsampling. tering and downsampling. Hogenauer [1] introduced filters with improved frequency response as for examHogenauer [1] introduced a very simple decimation propose to compensate the passband droop of the very simple decimation filtercomb called(CIC) cascaded Usually methods deal to improve either cascaded CIC-cosine decimation filter [10], introducing a filtera called cascaded integrator filter,intewhichple [4-11]. grator comb (CIC) filter, which consists of cascaded passband [5-6], or stopband [7-10]. consists of cascaded integrators and differentiators simple compensator at the low rate. The rest of the paper integrators and differentiators separated by a downgoal of thisaswork is to In propose a simple mulis organized follows. Section 2 we propose the new separated by a downsampler, as shown in Fig. 1. The The sampler, as shown in Fig. 1. The transfer function of tiplierless decimation filter with no filtering at high second order compensator filter which is illustrated with transfer function decimation of the resulting the resulting filter decimation is given as filter is giveninput rate, and with a desired high stopband attenutwo examples. Section 3 briefly introduces the cascaded as ation and a desired low passband droop. To this end CIC-cosine decimation from [10]. The proposed we propose to compensate the filter passband droop of the KK -M modified compensated CIC-cosine decimation filter is 1 1 − z −M ⎤ cascaded CIC-cosine decimation filter [10], introduc⎞ . (1) H (z) = ⎡ 1 ⎛ 1 − z−1 in section 3, illustrated with one simple compensator at and the low rate. The rest of example. (1)ing adescribed H (z ) = ⎢M ⎜1 − z −1 ⎟⎥ , ⎜ ⎟ M the paper is organized as follows. In Section 2 we pro1 − z ⎠⎦⎥ ⎣⎢ ⎝ ABSTRACT where M is the decimation factor and K represents the number of cascaded filters. pose the new second order compensator filter which is illustrated with two examples. Section FILTER 3 briefly 2. PROPOSED COMPENSATION where M is the decimation factor and K represents theintroduces the cascaded CIC-cosine decimation filter TheThe transfer function of compensated the proposedCICcompensation number of cascaded filters. from [10]. proposed modified filter is given by cosine decimation filter is described in section 3, and M illustrated with H one example. = a + bz − M + az −2 M , (3) COMP z ( ) where a and b are real valued constants, and M is decimation factor. Using a multirate identity this filter The function therate proposed cantransfer be moved to a of low which compensais M times less than Fig.1: CIC decimation filter. Fig.1: CIC decimation filter. tion high filter input is given by rate becoming a second order filter. response of (3) is expressed as The above decimation filter is attractive in many The magnitude The above decimation filter is attractive in many HCOMP zM = a + bz-M + az-2M . (3) applications because of its very low complexity. It applications because of its very low complexity. It is is usually used at the first stage of decimation [1]. H COMP e jMω = 2a cos(Mω ) + b . (4) usually used at thedecimation first stagefilter of decimation [1]. However, However, this exhibits several drawthis backs. decimation filter exhibits several drawbacks. The integrator section works at the high inputThewhere a and b are real valued constants, and M is factor. a multirate identity this fil- b. To this integrator section works at the high input data ratedecimation Next issue is Using to define the coefficients a and The authors are with the Institute INAOE, Deter can be moved to a low rate which is M times less resulting in a larger chip Puebla, area and higher power end we consider magnitude response of Eq. (4) in two partment of Electronics, Mexico. Email:gordana.dolecek@gmail.com, gordana@ieee.org, and isthan high input rate becoming a second order filter. consumption especially when the decimation factor points ωof=0, and ω=ωc,aswhere the worst gordana@inaoep.mx The frequency magnitude response (3) is expressed high. In order to resolve this problem the non-recursive passband distortion occurs, defined in Eq. (2). In ω=0 the 2. PROPOSED COMPENSATION FILTER ( ) Compensated CIC-Cosine Decimation Filter HCOMP ejMω 9 = |2acos (Mω) + b| . 2aq + bq = 1. (4) (11) Next issue is to define the coefficients a and b. To this end we consider magnitude response of Eq. (4) in two frequency points ω=0, and ω = ωc , where the worst passband distortion occurs, defined in Eq. (2). In ω=0 the desired magnitude characteristic has the value 1, resulting in the following equation Starting with k=2 the values of k are increased until the condition (11) is satisfied. If the passband droop is within the desired limit, the corresponding quantized coefficients aq and bq are presented in canonical signed digit (CSD) form [12-15]. 2a + b = 1. HCOMP-CSD zM = aCSD + bCSD z-M + aCSD z-2M . (12) (5) desired magnitude characteristic has the value 1, resulting in the following equation START In order to compensate the passband droop δc at the frequency ωc , we have from Eq. (4) where aCSD , and bCSD are the CSD representations of the quantized coefficients aq , bq of the proposed 2a + b = 1 . (5) compensation filter (3). Compute a and b Otherwise, the procedure is continued (Eq. (9))until the deIn order to compensate at the 2acos (Mωc )the + bpassband = 1/δc . droop δc(6) sired passband compensation is obtained. This prob the correa frequency ωc, we have from Eq. (4) cedure is programmed in MATLAB, and sponding block diagram is shown in Fig. 2. 2a cos( Mωc ) + b = 1 / δ c . Denoting k=2 (6) desired magnitude characteristic has the value 1, resulting Denoting in the following equation δcomp = 1/δc . k=k +1 (7) δ comp2a=+1 /bδ=c 1 . (7) (5) order to compensate the passband droop δc at the we In have frequency ωc, we have from Eq. (4) 1 a ⎡ 1 ⎤ ⎡ 22 2a cos( ⎤ ⎡ 1)⎤+ b = 1 / δ . 1 δ⎢comp c ⎥ =. ⎢ ⎥c . (8) (6)(8) ⎢ 2cos (Mc )1 =M⎥ω ⎣2 cos(Mωc ) 1⎦ ⎣b ⎦ Compute a and bq START q (Eq. (10)) ⎣δ comp ⎦ Denoting No k=2 No k=k +1 and bb as Solving (8) we get the Solving (8) we get the coefficients coefficients aa and as δ comp = 1 / δ c (7) 1 1 ⎤ ⎡ − we ⎢have 2(cos(Mω c ) − 1) 2(cos(Mω c ) − 1) ⎥ ⎡ 1 ⎤ ⎥ ⎡a ⎤ ⎢ ⎥ . (9) ⎥⎢ ⎢b ⎥ = ⎢ 1⎤ ⎡a ⎤1 ⎡ 1⎥ ⎣δ⎤ comp ⎦ ⎣ ⎦ ⎢ cos⎡(Mω c )2 (8) ⎢2 cos(Mω ) − 1⎥ ⎢b ⎥ = ⎢δ ⎥ ⎥ . ⎢ cos(M ⎣ ω c ) − 1 c cos ⎦ ⎣(M⎦ω c )⎣− 1comp ⎦ ⎦ ⎣ and b asfrom (9) are (8) we the coefficients a aand In Solving following, thegetcoefficients (9) quantized as shown in the next relations Compute a and b (Eq.(11) (9)) satisfied? Is Eq. b a No Yes Is the passband compensation Compute aOK? q and bq (Eq. (10)) Yes CSD Is Eq. (11) satisfied? END Yes No Is the passband compensation Fig.2: Block diagram of OK? the compensator filter design. Yes 1 1 The method is illustratedCSD in the following example. ⎤ ⎡ − ⎢ 2(cos− k(Mω ) −−k1) 2(cos(Mω − k ) − 1) ⎥− k 2 a /c2 b = 2 c b / 2 ⎡ 1, ⎤ (10) ⎡a ⎤ a⎢ q = the In following, coefficients aq and b from⎥ ⎢(9) are Example 1: Consider the design of compensator for END ⎥ . (9) ⎥ δ ⎢b ⎥ = ⎢ quantized ⎣ ⎦ ⎢ as shown cascaded CIC decimation filter with M=4, K=4, and cos(Mωinc )the next relations 1 ⎥ ⎣ comp ⎦ − factor, aq and where k ⎢is cos the(Mquantization decimation factor of stage is R=filter 8. The ω c ) −-k1 ω c )-k−1 ⎥⎦ bq are the cos-k(M Fig.2: Block diagram of second the compensator de- passband ⎣ -k aq = 2quantized a/2 bq = 2 b/2and .⎣x⎦ represents (10) sign. droop of CIC filter at normalized frequency ωc/π is corresponding coefficients, Fig.2: Block diagram of the compensator filter design. the integer part of x. The relation (5) must also hold for 0.2095792dB. a and(5)b must from also (9) are In following, theof coefficients the integer part x. The relation athe and b resulting in the condition the quantized quantized values However, we want to have a resulting passband droop shownofinvalues next hold for the as quantized ofrelations a and b resulting in less Thethan method is illustrated in the following example. The method is0.01dB. illustrated in the following example. the condition ⎣ ⎦ ⎣ ⎦ a q = 2 −k 2 aa/ q2 −+k bq =b1q .= 2 − k b / 2 − k , ⎣ ⎦ ⎣ ⎦ (11) (10) Starting k=2quantization the values of k are aincreased until the where kwith is the factor, q and bq are the condition (11) is satisfied. If the passband droop is within corresponding quantized coefficients, and ⎣x⎦ represents the the desired limit, the corresponding quantized coefficients integer part of x. The relation (5) must also hold for From Eq. (7), it follows Example 1: Consider the design of compensator for (13) δcomp=1.02442214823189. cascaded CIC decimation filter with M=4, K=4, and decimation factor of second stage is R= 8. The passband From (9) we have droop of CIC filter at normalized frequency ωc/π is Fig.2: Block diagram of the compensator filter design. 11) the hin ents SD) 12) the sed red e is ock The method is illustrated in the following example. Example 1: Consider the the design compensatorforfor Example 1: Consider designofof compensator cascaded decimation filterwith with M=4, M=4, K=4, K=4,and and cascaded CIC CIC decimation filter decimation factor of second stage is R= 8. The passdecimation factor of second stage is R= 8. The passband droopfilter of CIC filter at normalized frequency droop band of CIC at normalized frequency ωc/π is ωc /π is 0.2095792dB. 0.2095792dB. However, we want to have a resulting passband However, we want to have a resulting passband droop droop less than 0.01dB. less thanFrom 0.01dB. Eq. (7), it follows From Eq. (7), it follows δcomp=1.02442214823189. (13) Next example compares the proposed compensation filter with that recently proposed in literature [5]. Example 2: Consider the design of a CIC decimation filter for the design parameters M=11, K=5, [5]. From [5], the normalized passband and stopband frequencies of interest are ωc /π=0.035455 and ωAl /π=0.146364. The corresponding passband droop δcomp = 1.02442214823189. (13) and stopband attenuation of the CIC filter are 2.7288dB and AAl =-63.9441dB, respectively. From (9) we have The compensation filter from [5] has the correFrom (9) we have sponding passband droop 0.01597dB, and stopband attenuation - 61.199338dB, respectively. ⎡a ⎤ ⎡- 0.16041774990093⎤ Using the proposed compensator for the quantized . (14) ⎢b ⎥ = ⎢ 1.32083549980185 ⎥ factor k=8, we got the passband droop of -0.01475dB ⎣ ⎦ ⎣ ⎦ the stopband attenuation of -61.215311dB. corresponding CSD terms for quantized coefficients and and compensation filter with the corresponding CSD The terms of compensation filter with the by corresponding CSD the transfer function is given by quantized coefficients is given corresponding terms quantized coefficients and filter with the corresponding CSD terms of Starting with kCSD =2 we findforthe quantized values shown (14) compensation terms of quantized coefficients is given by the transfer function is given by quantized coefficients is given by in Table I. −2 5 5 Starting withz 4k=2 H COMP _ CSD11z11 = − 2−2 − 2 −−5 + 2 −8 −8 1 + z −22 −22 H COMP _ CSD = −we 2 −3find − 2 −the 1 +quantized z −8 + 20 values + 2 −2 + shown 2 − 4 z −4 . Thein condition (11) is satisfied only for the values of (16) H z = −2 − 2 + 2 1 + .z COMP-CSD 11 Table I. 4 −5 5 −8 1 −1z H COMP =− 4 −− 22−−27 −z2−11 + 2 −8 1 + z −22 . H COMP z = bold − 2 −3 −in2 −Table 1 + z I. + 20 + 2 −2 + 2 −4 z −4(15) + 2 2 + 2 CSD _− _ CSD k=3 and kThe =5, shown condition (11) is satisfied only for the values (16) + 21 − 2−1 + 2−4 − 2−7 . z−11 .(16) =3 the passband not (15) satisfy + 21 − 2−1 + 2− 4 − 2− 7 z −11 However, for kk=5, of k=3 and shown bold indroop Tabledoes I. The passband details for CIC filter, the proposed the desired specifications. For k =5 the desired However, k=3 the passband droop does passband not satTable Ifor : Quantized Factors and Coefficients. compensated CIC, and compensated from [5] are isfy the is desired specifications. For k k=5 the desired =5 we find the specification satisfied. Therefore using The passband details forfor CIC the proposed proposed The passband details CIC filter, filter, CIC the k I: Quantized aq bq and Coefficients. Ac(dB) Table given in Fig.4. passband specificationFactors is satisfied. Therefore using k compensated CIC, and compensated CIC from [5] are compensated CIC, and compensated CIC from [5] are k 2 find athe b1.25 Ac(terms dB) for quan- given in Fig.4. q 0 corresponding q =5 we CSD given in Fig.4. -0.0458388 3 -0.125 1.25 0 and the 1.25 tized2 coefficients transfer function is given by 0.25 -0. 125 1.25 1.3125 -0.0458388 3 4 -0.125 4 −3 −5 −8 0.25 =1.3125 −2 − 2-0.005381495 1+z 4H5COMP-CSD -0. 125 z 1.3125 -0.15625 0.2 −4 0 −2 −4 -0.005381495 5 -0.15625 1.3125 0.2 + 2 +2 +2 z . (15) ( ) ( ( ) ( )( )( ) ( ) ( ) ) ( ) ( ( ) ( ) ) ( ( )( )( ) ) 0.15 The passband passband details details of the The the CIC CIC and andthe theproposed proposed compensated CIC filters are shown in Fig. 3. compensated filters shown inand Fig. 3. proposed The passband CIC details of are the CIC the There exists a trade-off between the compensated CIC filters are shown in Fig. 3. desired pass0.01 0.15 Gain (dB) the ents for on ECTI TRANSACTIONS ON COMPUTER AND INFORMATION TECHNOLOGY VOL.4, NO.1 MAY 2010 Gain (dB) 0) 10 0.1 0.1 0.05 0.05 0 0.01 0 -0.05 0 0 -0.01 -0.01 Gain(dB) Gain(dB) are -0.02 -0.02 CIC CICCompensated CIC, k=5 Compensated CIC, k=5 CIC Filter CIC Filter+Roll-Off Compensator, [5] CIC Filter CIC Filter+Proposed Compensator, -0.05 CIC Filter+Roll-Off Compensator, [5] k=8 -0.1 CIC Filter+Proposed Compensator, k=8 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 -0.1 Frequency (ω/π )0.03 0.035 0 0.005 0.01 Normalized 0.015 0.02 0.025 Normalized Frequency (ω/π ) Fig. 4: Example 2. Fig.4: Example2.2. Fig. 4: Example Note that the proposed compensator results in a slightly Note that proposed compensator in a Note that thethe proposed compensator resultsresults in a slightly -0.04 better passband characteristic. -0.04 slightly better passband characteristic. better passband characteristic. -0.05 3. CIC-COSINE DECIMATION FILTER 0.005 0.01 0.015 0.02 0.025 0.03 -0.05 0 . CIC-COSINE DECIMATION FILTER 3 Normalized ) 0 0.005 0.01 0.015Frequency 0.02 (ω /π0.025 0.03 3. CIC-COSINE FILTER factor Normalized Frequency (ω /π) Considering the DECIMATION case when the down-sampling Considering the case when the down-sampling can be expressed Fig.3: Considering the as case when the down-samplingfactor facFig. 3:Example Example1. 1. can be expressed as Fig. 3: Example 1. M = M1M 2 M 3...M N tor can be expressed as (17) M = M1M 2 M 3...M N band droop the number of thethe CSD termspassband of co(17) There existsand a trade-off between desired we can rewrite Eq. (1) as There exists anumber trade-off desired passband a we can rewrite M efficients atheand b. of between (17) Eq.=(1)M as1 M2 M3 ...MN . K droop and the CSDthe terms of coefficients i −1 droop and the number of the CSD terms of coefficients a ⎡ and b. i −1 ∏ M jK ⎤ ⎡ N⎢ N and b. Mj =j 0 ⎤ ⎥ H ( z )⎢= H i∏ ) j( = 0z H ( z ) H ( z )⎥ ⎥ = ⎢ i =1i Next example compares the proposed compensation filter ⎥ ⎢ Next compares the proposed compensation filter ⎥⎦ ⎣1 i =⎢ withexample that recently proposed in literature [5]. (18) ⎦⎥ ⎣⎢ with that recently proposed in literature [5]. (18) where -0.03 -0.03 ∏∏ 5 better passband characteristic. 0.03 3. CIC-COSINE DECIMATION FILTER Compensated CIC-Cosine Decimation Filter d passband efficients a ation filter decimation stopband 455 and droop and 288dB and esponding enuation - quantized 475dB and dB. The 11 Considering the case when the down-sampling factor can be expressed as M = M1M 2 M 3...M N (17) we can rewrite Eq. (1) as we can rewrite Eq. (1) as K i −1 ⎡" N ∏ M j #⎤ N Q j = 0 ⎥K Y ⎢ HH((z) z) = Hi i (z z M ) . = ⎢∏H ⎥ i =1 ⎥⎦ ⎢⎣ i=1 At first we design the modified CIC-cosine filter H (z) [10] in order to satisfy the desired stopmCCOS i-1 j j=0 (18) band attenuation. Next, we find the passband droop δ in the frequency of interest ω = ωc of the designed (18) c modified CIC-cosine filter. Using (7) we compute where δcomp and solving equations (8) we get the coefficients i ⎛ a and b of the compensator. Finally, the coefficients i −1 −∏ M j ⎞ ⎟ ⎜ j =1 ∏M j a and b are expressed in CSD form as explained in 1 ⎜1− z ⎟; M = 1 H i ( z j =0 ) = Section 2. 0 i −1 . (19) Mi ⎜ −∏M j ⎟ The method is illustrated in Example 3. ⎜ 1 − z j =0 ⎟ Example 3: We design decimation filter for ⎠ ⎝ M=32, which has passband droop at the frequency of interest less than 0.002dB, and stopband attenuation Using Eqs. (18) and (19) we express the modified (19)CICin the frequency of interest at least of 110 dB. First we filter H m (z ) as, [10] design CIC-cosine filter choosing M1 =2, M2 =4 and Using Eqs. (18) and (19) we express the modified M3 =4 yielding k2 M1 H1k1 ( zas, ) ⋅ H[10] ) ⋅ ⋅ ⋅ H Nk N ( z M 1 ⋅⋅⋅M N −1 ), (20) CICHfilter Hm(z) m ( z) = 2 (z 1 1 − z−8 1 1 − z−2 2 , H , z = H (z) = where ki is the number of the cascaded filters H i. 2 1 2 1 − z−1 4 1 − z−2 Hm (z) = Hk11 (z) .Hk22 zM1 ...HkNN zM1 ...MN-1 . (20) −32 We use the cosine prefilters introduced in [16] to improve H 3 z8 = 41 1−z . (26) 1−z−8 In the next, we cascade filters H1 (z), H2 (z2 ) and where ki is the number of the cascaded filters Hi. H3 (z8 ) as shown in the following equation We use the cosine prefilters introduced in [16] to (27) Hm (z) = H41 (z) H22 z2 H23 z8 . improve the frequency characteristic of the modified CIC filter (20). HCCOS (z) = K Y KCOS z Ni . (21) i=1 The stopband attenuation of this filter is further improved introducing cascade of the expanded cosine filters HCCOS (z) = H2COS z8 H4COS z4 H4COS z2 . (28) where [16]. 2 HCOS zN = 0.125 1 + z-2N 1 + z-N . the frequency characteristic of the modified CIC filter (20). In general we have [10] K H CCOS ( z ) = ∏ H COS ( z i =1 Ni = Ni ), M . 2i+1 (22) (21) (23) ( ) ( ) ( )( ( ) 2 4 4 H CCOS (z ) = H COS z 8 H COS z 4 H COS z2 , 2 N -2N 1 + z-N . whereHCOS z = 0.125 1 + z where ( ) H COS z N = 0.125 1 + z −2 N 1 + z ). (28) (29) −N 2 (29) From (2) using R=8 the passband and stopband where [16]. From (2) using R=8 are the passband and stopband frequencies of interest N −2 N −N 2 H COSThe ( z transfer ) = 0.125 (1 + z of)(the 1 + zcascaded ) . modified (22) CIC function frequencies of interest are and cosine prefilters is 0.00390625 ω c / πωc=/π0.= 00390625 . (30) In general we have [10] K = 0.05859375. (30) ω A / πωA=/π 0.05859375 Y M= Hm (z) Hni zNi . HmCCOS (z) (24) COS N i = i +1 . (23) Stopband attenuation in the frequency of interest is Stopband attenuation in the frequency of interi=1 2 AAl=-116.7993dB, i.e., it is satisfied. However the est is AAl =-116.7993dB, i.e., it is satisfied. However The transfer function of the cascaded modified CIC and passband droop isdroop -0.3162dB. In order where ni is the number of cascaded cosine prefilters, the passband is -0.3162dB. In ordertoto improve improve cosine prefilters is in passband, we design compensation filterfilter as described and NK = M1 . passband, we design compensation as described K ni Section 2. Ni For more details see [10]. in Section 2. H mCCOS ( z ) = H m ( z )∏ H COS ( z ) , (24) Solving (9) we Solving (9)get wethe getcoefficients the coefficients i =1 DECIMATION FILTER where ni is 4. the PROPOSED number of cascaded cosine prefilters, and ⎡a ⎤ ⎡- 0.2435389959475⎤ NK=M1. The proposed filter is the cascade of the modified ⎢b ⎥ = ⎢ 1.487077991895 ⎥ . (31) For more CICcosine details seefilter [10].HmCCOS (z) [10] and the compensator ⎣ ⎦ ⎣ ⎦ filter introduced in Section 2. . . 4. PROPOSED DECIMATION FILTER G (z) = HmCCOS (z) HCOMP-CSD zM . (25) Table II shows the corresponding quantized values(31) aq The proposed filter is the cascade of the modified CIC- and bq. cosine filter HmCCOS(z) [10] and the compensator filter Table II: Quantized Factors and Coefficients. introduced in Section 2. ( ). G ( z ) = H mCCOS (z )H COMP _ CSD z M (25) k aq bq coefficients can control the desired passband droop of the overall decimation filter. The overall filter exhibits a low passband droop, and high stopband attenuation at the frequencies of interest. Additionally, using the - 0.24353899 59475 ⎡ECTI TRANSACTIONS ON⎤ COMPUTERpolyphase AND INFORMATION TECHNOLOGY 2010 decomposition, the filtersVOL.4, at theNO.1 firstMAY stage ⎢ 1.487077991895 ⎥ . (31) can be moved at the lower rate. Therefore, there is no ⎣ ⎦ filtering at high input rate. Section 2. Solving (9) we get the coefficients (24) ters, and ⎡a ⎤ ⎢b ⎥ = ⎣ ⎦ 12 . TableIIIIshows showsthe thecorresponding correspondingquantized quantized values Table values aq a and b . The condition (11) is satisfied only for q q and b . ied CICtor filter 6. ACKN ne filter stopband δc in the modified d solving b of the d b are aq 0 -0.125 -0.1875 -0.21875 -0.234375 -0.2421875 bq 1.25 1.375 1.4375 1.46875 1.484375 1.484375 -100 Gain(dB) k 2 3 4 5 6 7 CIC-cosine Compensated CIC-cosine, k=7 -50 Table II: Quantized Factors and Coefficients. (25) This No.49640 -150 7. REFER -200 -250 -300 0 0.2 a. 0.4 0.6 Normalized Frequency (ω /π) 0.8 1 Overall gain responses. 0.01 The condition (11) is satisfied only for k=7 resulting r M=32, erest less requency C-cosine ( ) ( ) )( ) H COMP _ CSD z 32 = − 2 −2 + 2 −7 ⋅ 1 + z −64 Using multirate identities we have efficient struc+ 21 − 2 −1 − 2 − 6 z − 32 . (32) ture shown in Fig. 5. ( 0 -0.01 Gain(dB) in thepassband corresponding droop ink=7 the resulting corresponding droop passband of -0.001723dB, of -0.001723dB, which satisfies the desired specificawhich satisfies the desired specification. Consequently, Consequently, we present for quantized coefficients wetion. present quantized coefficients k=7 in CSD terms for k=7 in CSD terms -0.02 CIC-cosine Compensated CIC-cosine, k=7 -0.03 -0.04 Overall magnitude response is shown in Fig. 6.a Using multirate identities we have efficient structure along with that of the CIC-cosine filter. -0.05 0 0.5 1 1.5 2 2.5 3 3.5 -3 x 10 Gain(dB) Normalized Frequency (ω/π) . (26) shown in Fig. 5. Corresponding passband is shown Overall magnitude responsezoom is shown in Fig.in6.aFig. b. Passband zooms. Fig. 5: Efficient structure for Example 3. 6.b. with that of the CIC-cosine filter. along Corresponding passband zoom is shown Fig. 6.b. Fig. 6.c illustrates attenuation in theinneighborequations, and expressed in CSD terms which can be Fig. 6.c illustrates attenuation in the neighborhood of hood of the first zero. implemented the firstusing zero.only adders and shifts. As a result (z2) and CIC-cosine -120 the proposed filter is a multiplier free filter. Compensated CIC-cosine, k=7 There is a trade-off between the desired -140 5. CONCLUSIONS 5. CONCLUSIONS compensation of the passband droop, and the -160 (27) complexity We of proposed the CSD here representation. Number of the multistage decimation filter s further adders based mayWe beonproposed further reduced using sub-expression here the multistage decimation the CIC-cosine decimation filter and fil-the -180 d cosine elimination for CSD represented coefficients ter (CSE) basedsecond onthe theorder CIC-cosine decimation filter andof symmetric compensator. The coefficients [15]. In contrast to methods [5-6],compensator. thesolving proposed the symmetric second The coef-200 the compensator are order obtained two simple compensation design is simpler, the filter ficientsfilter of the compensator are and obtained solving two -220 coefficients can equations, control theand desired passband droop of which simple expressed in CSD terms the overall filter. The overall filter exhibits candecimation be implemented using only adders and shifts. As -240 a low passband and high stopband at free fila resultdroop, the proposed filter is attenuation a multiplier 0.059 0.06 0.061 0.062 0.063 0.064 0.065 the frequencies of interest. Additionally, the comNormalized Frequency (ω/π) ter. There is a trade-off between using the desired polyphase decomposition, the filtersdroop, at theand firstthe stage pensation of the passband complexity c. Stopbands. can be moved the lower rate. Therefore, thereofisadders no of the atCSD representation. Number may Fig.6: Example3.3. filtering at high input rate. be further reduced using sub-expression elimination Fig. 6: Example Gain(dB) 0.066 (CSE) for the CSD represented coefficients [15]. In contrast to methodsCIC-cosine [5-6], the proposed compensaCompensated CIC-cosine, k=7 5. 1 ACKNOWLEDGEMENT -50tion filter design is simpler, and the filter coefficients can control the desired passband droop of the over- 6. ACKNOWLEDGEMENT This work is part of the CONACYT project -100all decimation filter. The overall filter exhibits a low This work is part of the CONACYT project No.49640. passband droop, and high stopband attenuation at No.49640. -150the frequencies of interest. Additionally, using the polyphase decomposition, the filters at the first stage References 7. REFERENCES -200can be moved at the lower rate. Therefore, there is [1] E. B. Hogenauer, An economical class of [1] E. digital B. Hogenauer, “Andecimation economicaland classinterpolation, of digital no filtering at high input rate. filters for -250 filters forTransactions decimation on and interpolation,” IEEE Acoustics, Speech, IEEE and SigTransactions on Acoustics, Speech, and Signal -300 Processing, vol. ASSP-29, No.2, pp.155-162, April 0 0.2 0.4 0.6 0.8 1 Normalized Frequency (ω /π) 1981. a. Overall gain responses. [2] Y. Gao, L. Jia, J. Isoaho and H. Tenhunen, “A comparison design of comb decimators for sigma0.01 0 -2 0 q Table 1: Quantized Factors and Coefficients. -2 [1] E. B. filters Transa Proces 1981. [2] Y. G comp delta Integ 22, p [3] H. Ab H. M of co conve Syste Proce [4] A. K “Appl integra on Sig 1997. Compensated CIC-Cosine Decimation Filter 13 Fig.5: structurefor forExample Example Fig. 5:Efficient Efficient structure 3. 3. Gain(dB) Gain(dB) ter, -120 International IEEE Conference ISCAS 2005, CIC-cosine CompensatedMay CIC-cosine, k=7 Kobe, Japan, pp.3733-3736, 2005. [11] G. -140 Jovanovic Dolecek, and S.K. Mitra, A New Two-stage Sharpened Comb Decimator, -160 IEEE Transactions on Circuits and Systems I: Regular -180 Papers, vol.52, No.7, pp.1416-1420, July 2005. [12] Y. -200C. Lim, Design of discrete-coefficientvalue-220 linear phase FIR filters with optimum normalized peak ripple magnitude, -240Transactions on Circuits and Systems, IEEE 0.059 0.06 0.061 0.062 0.063 0.064 0.065 0.066 vol. CAS-37, pp.1480-1486, Dec.(ω/1990. π) Normalized Frequency [13] Y. C. Lim and Liu, Design of cascade c. B. Stopbands. form FIR filters with discrete valued coefficients, IEEE Transactions, Speech, Signal PrFig. 6:Acoustic, Example 3. ocessing, vol. ASSP-36, pp.1735-1739, Nov. 1988. [14] Y. C. Lim, J. B. Evans, and B. Liu, Decomposition of binary inte6. ACKNOWLEDGEMENT gers into signed powers-of-two terms IEEE Transactions and Systems, This work is part on of Circuits the CONACYT project vol. CAS-38, pp. 667-672, June 1991. No.49640. [15] R. I. Hartley, Subexpression sharing in filters using canonic signed digit multipliers,” 7. REFERENCES Transactions Circuits and [1] IEEE E. B. Hogenauer, “Anoneconomical classSystems:II, of digital vol.43, Issue 10, pp.677-688, October 1996. filters for decimation and interpolation,” IEEE [16]Transactions Y. Lian andonY.Acoustics, C. Lim, New prefilter Speech, and structure Signal for designing FIR filters, No.2, Electronic letters,April vol.29, Processing, vol. ASSP-29, pp.155-162, 1981. No. 11, pp.1034-1035, May 1993. [2] Y. Gao, L. Jia, J. Isoaho and H. Tenhunen, “A comparison design of comb decimators for sigmadelta analog-to-digital converters,” Analog Integrated Circuits and Signal Processing, vol. Gordana Jovanovic Dolecek re22, pp. 51-60, 1999. ceived a BS degree from the Depart[3] H. Aboushady, Y. Dumonteix, M. Engineering, M. Loerat, and ment of Electrical University of Sarajevo, MSc degree from H. Mehrezz, “Efficient polyphaseandecomposition University of Belgrade, and a PhD deof comb decimation filters in Σ-Δ analog-to-digital gree from the Faculty of Electrical Enconverters,” IEEE Transactions onofCircuits gineering, University Sarajevo.& She professorand at theDigital Faculty ofSignal Electrical Systems – II:wasAnalog University of Sarajevo Processing, vol. Engineering, 48, pp. 898-903, October 2001. until 1993, and 1993-1995 she was with the [4] A. Kwentus, Z. Jiang,Mihailo and A. Willson, Jr., In Institute Pupin, Belgrade. 1995“Application she joined Institute INAOE, Department Electronics, of filter sharpening to forcascaded Puebla, Mexico, wheredecimation she works asfilters,” a professor and researcher. integrator-comb IEEE Trans. Gordana2001-2002 Jovanovic she Dolecek received a BS of Electrical ComDuring was at Department onEngineering, Signal Processing, vol.45, pp.457-467, puter University SantaFebruary Barbara, as degree from the Department ofof California, Electrical 1997. visiting researcher on of a sabbatical Engineering, University Sarajevo, anleave. MSc She is the author of Gain(dB) equations, and expressed in CSD terms which can be implemented using only adders and shifts. As pp.155-162, a result nal Processing, vol. ASSP-29, No.2, the proposed filter is a multiplier free filter. April 1981. There is aL. trade-off betweenandtheH. desired [2] Y. Gao, Jia, J. Isoaho Tenhunen, compensation of the passband the A comparison design of droop, comb and decimators complexity of the CSD representation. Number of for sigmadelta analog-to-digital converters, adders may be further reduced using sub-expression Analog Integrated Circuits and Signal Processing, elimination (CSE) for the CSD represented coefficients vol. 22, pp. 51-60, 1999. [15]. In contrast to methods [5-6], the proposed [3] H. Aboushady, Y. Dumonteix, M. M. Locompensation filter design is simpler, and the filter erat, can andcontrol H. Mehrezz, Efficientdroop polyphase coefficients the desired passband of decomposition of comb decimation the overall decimation P filter. The overall filter exhibits filin droop, −∆and analog-to-digital converters, a lowters passband high stopband attenuation at IEEE Transactions on Additionally, Circuits & Systems the frequencies of interest. using theII: Analog and Digital Signal Processing, vol. 48, polyphase decomposition, the filters at the first stage pp.moved 898-903, 2001. can be at theOctober lower rate. Therefore, there is no filtering high input rate. [4] A. atKwentus, Z. Jiang, and A. Willson, Jr., Application of filter sharpening to cas0 CIC-cosinedecimation caded integrator-comb filters, Compensated CIC-cosine, k=7 IEEE Trans. on Signal Processing, vol.45, -50 pp.457-467, February 1997. [5] S.-100Kim, W. Lee, S. Ahn, and S. Choi, Design of CIC roll-off compensation filter in a W-CDMA -150 digital IF receiver, Digital Signal Processing, vol.16, Issue 6, pp.846-854, July 2006. -200 [6] K. S. Yeung and S. C. Chan, The design -250 and multiplier-less realization of software radio receivers with reduced system delay, -300 IEEE and Regular 0 Trans. 0.2 on Circuits 0.4 0.6 Systems-I: 0.8 1 (ω /π) papers, vol.51,Normalized No.12,Frequency pp.2444-2459, December a. Overall gain responses. 2004. [7] M. 0.01 Laddomada, Generalized comb decimation filters for sigma delta A/D 0 converters: Analysis and Design, IEEE Transactions on Circuits and Systems,Part -0.01 I: Regular papers, vol.54, No.5, pp.994-1005, May 2007. -0.02 [8] L. L Presti, Efficient filCIC-cosinemodified-sinc Compensated CIC-cosine, k=7 ters for sigma delta A/D converters, -0.03 IEEE Transactions on Circuits & Systems II, vol.-0.04 47, pp.1204-1213, Nov. 2000. [9] G. Jovanovic Dolecek and S.K. Mitra, A new -0.05 multistage (RS) dec0 0.5comb-modified 1 1.5 2 rotated 2.5 3sinc 3.5 -3 Normalized Frequency (ω/π) imator with sharpened magnitude response, x 10 zooms. b. Passband IEICE Transactions Special Issue on Recent Advances in Circuits and Systems, vol.E88-D, No.7, pp.1331-1339, 2005. [10] G. Jovanovic Dolecek and J. Diaz Carmona, A new cascaded modified CIC-Cosine decimation fil- three editor ofof one book,and and author of more than 200 degree books, from University Belgrade, a PhD papers. Hertheresearch digital signal processing degree from Faculty ofinterests Electrical include Engineering, and digital communications. She is a Senior member of IEEE, University of Sarajevo. She was professor at the the member of Mexican Academy of Science, and the member Faculty of Electrical Engineering, of of National Researcher SystemUniversity (SNI) Mexico. Sarajevo until 1993, and 1993-1995 she was with the Institute Mihailo Pupin, Belgrade. In 1995 she joined Institute INAOE, Department for Electronics, Puebla, Mexico, where she works as a professor and researcher. During 2001-2002 she was at Department of Electrical & Computer 14 ECTI TRANSACTIONS ON COMPUTER AND INFORMATION TECHNOLOGY VOL.4, NO.1 MAY 2010 Fernando Javier Torres received a BS degree from the Department of Electrical Engineering, University of Las Americas, Puebla, 2005, Mexico and an MSc degree from Institute INAOE, Puebla, Mexico, 2007. His research interest is digital signal processing and communications. Biography: Fernando Javier Torres received a BS degree from the Department of Electrical Engineering, University of Las Americas, Puebla, 2005, Mexico and an MSc degree from Institute INAOE, Puebla, Mexico, 2007. His research interest is digital signal processing and communications.