AadGTyzzz{{}HH $ d    daFootnote TableFootnote**. . / -  \2 \Heading T    B E                  z # % ' ) + / M C G I K M O Q S U W Y i ] ^ ` b e g k m o q s u w ~ }  {  *        N  )   O 7 9 ; = ? A Ce 3 E  IAa Kd MT O Qyz ) 9 + ) { * omm<$lastpagenum>n<$monthname> <$daynum>, <$year>o"<$monthnum>/<$daynum>/<$shortyear>p;<$monthname> <$daynum>, <$year> <$hour>:<$minute00> <$ampm>q"<$monthnum>/<$daynum>/<$shortyear>r<$monthname> <$daynum>, <$year>s"<$monthnum>/<$daynum>/<$shortyear>t <$fullfilename>u <$filename>v <$paratext[Title]>w <$paratext[Heading]>xyPagepage <$pagenum>zHeading & Page<$paratext> on page <$pagenum>{ <$curpagenum>M| <$marker1>} <$marker2>M~ (Continued)+ (Sheet <$tblsheetnum> of <$tblsheetcount>)^5q5gA6Ao7s8~9: *;..<A=00>11 )?22@339A44=B55CC66IAaD77TOE88F99 G::H;;anuI<<ohnJ== yeK>>ohnL??$orM@@;moNu, O:miPQ/daRaSne>Te>Un>/VreaWufiX fiYpaZ [Hdi\Pa]n>^ ge_nag` aM bcMd)ebhefsetghgijsklmn.op01234=56I7T89:;a<o= >o?$@;u:/anenru Hn nM    M  )  b  s.012345678  9!!:"";##<$$=%%>&&?''@(())**++,,--..//001122 H1 n  Table 1:  b 2 258  Table 2: :; #2.3  $ 3 >  ?'@)6 ,-/02 !2.1 "#$H%&'()*+,-./012: 3456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVW Ta Table 2: XY Z[ \ ]3 ^_?`'abcdefghijklm2.2 nopqr s!4 tuv Table 3: wxyz{|}~5 :  Tale    3 ?'.2  !: ]AdAydAzHHB HH5`  HHB HH6`  HUX B HUX 7e UUh *Draft -   7/31/92   2  H$ B 'H$ 8UU`  HUX B HUX 9UU`  H$ B H$ :UU`  dA{H;3K ^B H H FootnoteHE;? ^B HMHMA Single LineH'%BFootnote F HHlADf ^D HuHu Double LineHHz %B Double Line E  H1Ee 11ft h+'G GJequal[cong[over[times[char[r],char[m],char[s],id[times[char[e],char[r],char[r],char[o],char[r]]]],times[char[r],char[m],char[s],id[times[char[r],char[e],char[s],char[u],char[l],char[t]]]]],over[times[indexes[1,0,num[2,"2"],fract[id[plus[char[M],num[3,"3"]]],num[2,"2"]]],indexes[1,0,num[2,"2"],minus[char[B]]],id[num[0.3,"0.3"]]],times[char[r],char[m],char[s],id[times[char[I],char[n],char[i],char[t],char[i],char[a],char[l],char[A],char[r],char[r],char[a],char[y]]]]]],over[times[indexes[1,0,num[2,"2"],fract[id[plus[char[M],num[3,"3"]]],num[2,"2"]]],indexes[1,0,num[2,"2"],minus[char[B]]],id[num[0.3,"0.3"]]],fract[indexes[1,0,id[char[K]],fract[num[1,"1"],num[2,"2"]]],num[2,"2"]]]]^HT%  B Single Line  E   dAHZj-'Hb  HHB GHH/,h<ngtihT],n =   *`VhaNumber of years = 552 D N[m_This represents the number of measurements of 1 minute repetitions for the systematic error to [2,R@N[cequal the shot noise error. ]]l ,"bSo the systematic component of the error becomes equal to the random error after 552 years, so it z@,c@is negligible for the MDI with a maximum lifetime of six years. im` [2 Other Bounds p "3aAs a check, a few other bounds for the FFT truncation error in the literature are calculated and h,"]found to be smaller, because of the exponent 2b which is double the exponent of that used by e@ Welch [1]. ` From Weinstein [2] ` A hTotal FFT error   :` T`Where (b+1) = 15 bits n`G0a = 2 (for complex multiplication) ti`],N = 1024 = 210 `f 3Total FFT error @ 0.000163 (dimensionless) nu`tsFrom Tran-Thong [3] nshc &Total FFT error   (dimensionless)  oraIn [3] it is assumed that each multiplication performed in the FFT algorithm introduces an error rsodwhich is statistically independent of all other errors. Multiplication produces an error which can  Hun/be modeled as a noise with a variance of  . bo: uru_These upper bounds are all smaller compared to that calculated from Welchs equation Number 1, cauH@u 2:and this leads to smaller possible systematic components. b`].LWe still use Welchs equation for a conservative estimate of the FFT error. |`El Q)/˴.HwfT( h io` 7L' LGBonF 5?(bSequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)3,"3"]],num[(*s9.00s*)0.3,"0.3"]]he 7J''TH7ifO- whdA|b.` 7L' LGEonF 5?tySequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)4,"4"]],num[(*s9.00s*)0.8,"0.8"]]er HHB thHH <di `FTOInitial array X0 contains the input data on which the FFT is performed "hchwhereB = 15 bits  <` i(N = number of input data points = 1024  Vh9WAK= average modulus squared of the initial (signal) array =   rp`[1EM = number of stages in FFT = log2N = log21024 = 10 u`1,Nwe find the rms error in a 15 bit fixed point FFT is 0.00083 (dimensionless).  T(`This equation gives an approximate upper bound for the ratio of the rms error to the rms of the ar(*cresult due to truncation. This bound increases as the   or 1/2 bit per stage. In deriving the whAdabove equation it is assumed that all errors are independent and, hence, the variance of the sum is de00hthe sum of the variances.It also assumes error-free multiplications by 1 and  j, which are not @accounted for in the error. UT UT`y Digitization Level and Scale a`T OTo Calculate the Scaling of the input data appropriate to carrying out the FFT (N3` d2Input range of the Suns velocity = 4000 m/sec. M`of&RMS velocity of the Sun = 2000 m/sec. g`M-Shot noise in the input data = S = 20 m/sec. g`  u `weaDigitization error when digitizing an input velocity signal to the least significant bit (m/sec) hhan  =   ehheTotal error =   or  `!ar%To Calculate the Digitization Level: n# ld cAs D is a degree of freedom which is selected such that it does not add significantly to the total it 1@ll Jerror T, so let the digitization error be 1/10th of the shot noise error. #?Z>2G.I' S econg[prompt[],over[times[char[alpha],indexes[1,0,num[2,"2"],minus[times[num[2,"2"],char[b]]]],indexes[1,0,char[N],num[2,"2"]]],num[12,"12"]]]ieZ>2G@ alequal[cong[prompt[],over[times[char[alpha],indexes[1,0,num[2,"2"],minus[times[num[2,"2"],char[b]]]],indexes[1,0,char[N],num[2,"2"]]],num[72,"72"]]],num[2.7e-05,"0.000027"]]veTXTNH/ 2, D n}xx'mjG?<eequal[indexes[1,1,char[N],char[j],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],plus[char[j],minus[num[1,"1"]]],num[1,"1"]],indexes[0,1,char[T],char[j]]]],num[2,"2"]],indexes[1,1,char[N],char[j],num[2,"2"]]]],over[num[1,"1"],num[2,"2"]]]]eZ >2Hfr 2, d dAan~G/2txx)mjHfth h e HHB #HH <coh#im  = 2 m/sec s[1*`$s[ eDh%b]/  = 2   LSB =   = 6.93 m/sec ]]i^`G xh& FSo the number of bits required to represent the solar signal is    (m[aTherefore 9+1= 10 bits are required, including the sign bit to represent the input data, and the T@(error in the signal is then: nh*G  ` h+1,  m/sec. "1`-tr 16`.0,aSo, the error due to digitization is negligible in comparison with the shot noise in the signal. ]P`/de^10 bits are required to represent an input data that ranges from - 4000 m/sec to +4000 m/sec. vh0So,   m/sec. `2,The conversion is: 1 bit scales to 8 m/sec. mjh3th1 bit  8 m/sec. e `BhAs the data is stored and calculated as 15 bit integers, other scalings are also possible, see Table 1. ecHZH TableFootnote IG WaF equal[char[D],over[times[string["Least'"],string["Significant'"],string["Bit"]],sqrt[num[12,"12"]]],over[times[char[L],char[S],char[B]],sqrt[num[12,"12"]]]]ngEmtyG ut#u TZequal[char[T],sqrt[plus[indexes[1,0,char[D],num[2,"2"]],indexes[1,0,char[S],num[2,"2"]]]]]yI H   i tr G .5:aFermequal[char[D],times[over[num[1,"1"],num[10,"10"]],char[S]],cross[over[num[1,"1"],num[10,"10"]],num[20,"20"]]]0) ~Gmty H  i fr2h& GTEFequal[char[D],over[times[char[L],char[S],char[B]],sqrt[num[12,"12"]]]]H?k tr Hth 2,   e _"'rGis0aFlcjequal[over[num[4000,"4000"],num[6.93,"6.93"]],cong[num[577.2,"577.2"],indexes[1,0,num[2,"2"],num[9,"9"]]]]H{`4h&H ~tn   G_Geq/ŠerUequal[char[T],sqrt[plus[indexes[1,0,char[D],num[2,"2"]],indexes[1,0,char[S],num[2,"2"]]]],sqrt[plus[times[over[num[1,"1"],num[100,"100"]],indexes[1,0,char[S],num[2,"2"]]],indexes[1,0,char[S],num[2,"2"]]]],sqrt[times[over[num[101,"101"],num[100,"100"]],indexes[1,0,char[S],num[2,"2"]]]],times[over[sqrt[num[101,"101"]],num[10,"10"]],char[S]]]ma"'rH~. 2, l[H5:`tH10 2, [oH:`_H0, 2, tytGW\:Šequal[char[T],cross[over[sqrt[num[101,"101"]],num[10,"10"]],num[20,"20"]],times[num[2,"2"],sqrt[num[101,"101"]]],num[20.09,"20.09"]]H,GKaF"'Napprox[over[num[8000,"8000"],indexes[1,0,num[2,"2"],num[10,"10"]]],num[8,"8"]]dAde[9\.Hd HHB HHC.>,<rt inh6nu  ]`:[SgThis table gives the scaling factors. As the scale factor decreases, it makes the FFT error smaller. 1,06 ]]jThe data are of 10-bits precision, and about 1/2 the least significant bit is lost at each stage; so, the D"]irelative error for 10-bit data is about 0.1%; whereas, if the 10 bit data is represented as 15 bits, the 0R@ relative error is about 0.003%. t`mComparison of Errors  7^Since we want to see how the systematic component (if any?) will show up in the observations, @7nucthe following calculation is performed and we claim that this is negligible, as can be seen below. [nu`;ex6Total FFT error = 0.00083 (dimensionless for 15 bits) h8/Since 1 bit  8 m/sec in the scaling we use, h<d6  total FFT error = 0.00083 x 8 = 0.00663 m/sec. `=^Random error from the table (Tim Browns table at the FFT stage - see Table 2) = 0.351 m/sec. pGfa 8@, equiv[prompt[],prompt[]]alpGda 8@itequiv[prompt[],prompt[]] tRp H Z 6Gor -@ =s therefore[prompt[]]if }Rp HC15 VAx Gz0.@ aFequal[over[num[552,"552"],indexes[1,0,id[num[900,"900"]],fract[num[1,"1"],num[2,"2"]]]],over[num[552,"552"],num[30,"30"]],num[18.4,"18.4"]]ervQ(Z 6Hfo  ph'2h G n5Fcacong[indexes[1,0,matrix[1,1,over[num[0.351,"0.351"],over[num[0.000163,"0.000163"],cross[num[10,"10"],num[4,"4"]]]]],num[2,"2"]],num[14312,"14312"]]sca0u\T!G}d;.u Fequal[char[n],cong[indexes[1,0,matrix[1,1,over[num[0.351,"0.351"],num[0.0002075,"0.0002075"]]],num[2,"2"]],times[num[5.52,"5.52"],char[y],char[e],char[a],char[r],char[s]]]]G;'#"G%eqroUindexes[1,0,matrix[1,1,over[num[0.351,"0.351"],num[0.000663,"0.000663"]]],num[2,"2"]]tHoIv#<#H*, w$G'orBEs }equal[over[char[n],cross[num[360,"360"],num[1440,"1440"]]],times[num[0.541,"0.541"],char[y],char[e],char[a],char[r],char[s]]]F^&8='%%H'*],"", ,"Y42&G)]]K6,"cong[indexes[1,0,id[over[num[0.351,"0.351"],over[cross[num[0.00083,"0.00083"],num[4,"4"]],num[10,"10"]]]],num[2,"2"]],times[num[2.156,"2.156"],char[y],char[e],char[a],char[r],char[s]]][n[`w'H%)*3, 2,$$, 10p(G+nu 8@"]equiv[prompt[],prompt[]]=\Y62)H'+*l[&&, ,0d*A],,,75Rp +H)*.5((, arHH,B* 'HH}+{{<in1, >nubSince truncation does not lead to systematic error in this analysis, and we do not know about the >dsource content in the FFT error, let the systematic component of the error be arbitrarily 1/10th of 40$H>41the total FFT error.  h>`?[sUSystematic error = 1/10 x total FFT error = 1/10 x 0.00663" = 0.000663 m/sec. X A6oThe contribution of the random error is reduced by the (number of measurements)"1/2 = n"1/2, the ]],fAnuWsystematic component of the error remains constant. So we can solve for the number of tA,\measurements n, for which the effect of the systematic error (SE) and random error (RE) are \@Aequal. `,0#Set RE = SE after n measurements. h@75     hC.5n =    FZNumber of years (for handling 10 bit input data stored as 15 bits) for one minute cadence $HFtiobservations =   atJ`KnaENumber of years (for handling 10 bit input data stored as 11 bits) = tph9 l   `Ot  t`GareIf we take l bit scales to 1/4 m/sec, corresponding to handling 10 bit input data stored as 15 bits, =hPerH1 bit   1/4 m/sec for handling 10 bit input data stored as 15 bits. co`Qra2Total FFT error = 0.00083 x 1/4 =0.0002075 m/sec. `R"5Systematic error = 1/10 FFT error =0.00002075 m/sec. eM-G/ S'l f\indexes[1,0,id[(*i1i*)over[(*n*)num[0.351,"0.351"],num[2.075e-05,"0.00002075"]]],num[2,"2"]]icH` .H1 arH` ; ` a rm^=O/H75 2,-- ` 0H ` = (g  ifWs  b)r(1H.2 r(> K X#Numbers of bits of input data in a red@X16 bit sequence  Dr(2H13 Dr(? /srr`Yin Scale factor dh5it1 bit  ? m/sec r3H24 ng@ `ZT 10 = 0D4H35 DA er.0`[8 5H46 B 0..3`\,"11 020]D6H57 HDC `]8/2 = 4 a7H68 D H`^12 D8H79 iDE 2`_4/2 = 2 9H8: F a ed``13 bitqD:H9; 3DG `a2/2 = 1 inc;H:< 1 H 4`b14 D<H;= DI 5`c 1/2 = 1/2 !=H<> [!J `d15 D!>H= ,"D!K `e1/2  1/2 = 1/4 H& ?Hf aH& Lh ` a33& @Hf a33& Mh )f  pAGC4/ 8@equiv[prompt[],prompt[]]7:J0'TBH;^f |HRp CHAA2 ` 7L' LDGGF 5?Sequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)5,"5"]],num[(*s9.00s*)0.5,"0.5"]]7^J0'TEH=`f ` 7L' LFGI=F 5?Sequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)6,"6"]],num[(*s9.00s*)7.9,"7.9"]]7J0'TGH?bfDD ` 7L' LHGKF 5?Sequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)7,"7"]],num[(*s9.00s*)0.1,"0.1"]]7J0'TIHAef FF  7L,L LJGMF 5?Uequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)8,"8"]],num[(*s9.00s*)0.25,"0.25"]]7J0'TKHCgfHH ` 7L' LLGOpF 5?Sequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)9,"9"]],num[(*s9.00s*)0.1,"0.1"]]7J6'TMHEkfpJJ  7L/ LNGQ 5?Wequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)10,"10"]],num[(*s9.00s*)0.03,"0.03"]]*s97J0'TOHImfTLL  +& PGSL 5?Eequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)11,"11"]],prompt[]])76J; 'TQHKofNN" b` 7L4L LRGU 5?LYequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)12,"12"]],num[(*s9.00s*)0.007,"0.007"]]77ZJ, 'TSHMqfPP' ` 7L4L LTGW 5?GYequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)13,"13"]],num[(*s9.00s*)0.005,"0.005"]],7~JA 'TUHOsfJRR, ` ~AY VGY7L' aequal[indexes[0,1,char[(*s9.00s*)N],num[(*s7.00s*)2,"2"]],cross[num[(*s9.00s*)3.2,"3.2"],indexes[1,0,num[(*s9.00s*)10,"10"],num[(*s7.00s*)2,"2"]]]]p7JA 'TWHQufTT1  rXGi1,:j)Nequal[indexes[1,1,char[N],num[1,"1"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[0,"0"],num[1,"1"]],indexes[0,1,char[T],num[1,"1"]]]],num[2,"2"]],indexes[1,1,char[N],num[1,"1"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[0,"0"]]7xNNYH9]fLVV  UV{ZG]1,>B)Nequal[indexes[1,1,char[N],num[2,"2"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[1,"1"],num[1,"1"]],indexes[0,1,char[T],num[2,"2"]]]],num[2,"2"]],indexes[1,1,char[N],num[2,"2"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[320,"320"]]H ~![G^@ Yequal[indexes[1,1,char[N],num[3,"3"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[2,"2"],num[1,"1"]],indexes[0,1,char[T],num[3,"3"]]]],num[2,"2"]],indexes[1,1,char[N],num[3,"3"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[6.41,"6.41"]]  \G`s[BJm[equal[indexes[1,1,char[N],num[4,"4"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[3,"3"],num[1,"1"]],indexes[0,1,char[T],num[4,"4"]]]],num[2,"2"]],indexes[1,1,char[N],num[4,"4"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[17.96,"17.96"]]fe}՞#]HY;f ZZ fe0$^HB=far[[ 1" _GbexBJs[equal[indexes[1,1,char[N],num[5,"5"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[4,"4"],num[1,"1"]],indexes[0,1,char[T],num[5,"5"]]]],num[2,"2"]],indexes[1,1,char[N],num[5,"5"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[11.87,"11.87"]]ifeT$`HE?f[1\\ 1,  aGeBJ,nequal[indexes[1,1,char[N],num[6,"6"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[5,"5"],num[1,"1"]],indexes[0,1,char[T],num[6,"6"]]]],num[2,"2"]],indexes[1,1,char[N],num[6,"6"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[14.26,"14.26"]],few$bHGAfex__ 4" cGkN]BJtm[equal[indexes[1,1,char[N],num[8,"8"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[7,"7"],num[1,"1"]],indexes[0,1,char[T],num[8,"8"]]]],num[2,"2"]],indexes[1,1,char[N],num[8,"8"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[12.98,"12.98"]]i dGg[1BJtdeequal[indexes[1,1,char[N],num[7,"7"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[6,"6"],num[1,"1"]],indexes[0,1,char[T],num[7,"7"]]]],num[2,"2"]],indexes[1,1,char[N],num[7,"7"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[14.26,"14.26"]]fe$eHICf],aa ,idfA[1hh1,fe gHKEfex ldd 6"HHhBf N]HHu?2x<,n2"h26.  l = 500 ;   *h   fet՞#aiH9f lXX 1, jGmexBJt1,equal[indexes[1,1,char[N],num[9,"9"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[8,"8"],num[1,"1"]],indexes[0,1,char[T],num[9,"9"]]]],num[2,"2"]],indexes[1,1,char[N],num[9,"9"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[12.98,"12.98"]],fe kHMIfex lcc [i!lGoarG_t,nequal[indexes[1,1,char[N],num[10,"10"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[9,"9"],num[1,"1"]],indexes[0,1,char[T],num[10,"10"]]]],num[2,"2"]],indexes[1,1,char[N],num[10,"10"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[11.68,"11.68"]]gHfe mHOKf ljj BnGqIGt!equal[indexes[1,1,char[N],num[11,"11"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[10,"10"],num[1,"1"]],indexes[0,1,char[T],num[11,"11"]]]],num[2,"2"]],indexes[1,1,char[N],num[11,"11"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[0.351,"0.351"]]ife0 oHQMfde lll# "8pGs,nIGtum!equal[indexes[1,1,char[N],num[12,"12"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[11,"11"],num[1,"1"]],indexes[0,1,char[T],num[12,"12"]]]],num[2,"2"]],indexes[1,1,char[N],num[12,"12"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[0.014,"0.014"]],feT qHSOf1, lnn( umrGu0,IGt,"!equal[indexes[1,1,char[N],num[13,"13"],num[1,"1"]],indexes[1,0,matrix[1,1,plus[indexes[1,0,id[times[indexes[1,1,char[N],num[12,"12"],num[1,"1"]],indexes[0,1,char[T],num[13,"13"]]]],num[2,"2"]],indexes[1,1,char[N],num[13,"13"],num[2,"2"]]]],fract[num[1,"1"],num[2,"2"]]],num[0.015,"0.015"]],fex sHUQf1, lpp- ,ntGw[1U@ u[2+id[indexes[1,1,char[N],char[j],num[1,"1"]]],"2fe uHWf], lrr2 fedvAxxqJ wH7fGYtt deHHxBv ,"HHzzh<,0me UT UT` [N3Propagation of FFT Error Through the Inner Product ,"1# I,ifIn reality we are doing a 2-D calculation, that is, we first do a 1-D FFT along the latitude and then 1I_project it on the Legendre function along the longitude, so in effect we want to study the FFT es[?I,"_error propagating through the dot product. The dot product adds noise which is independent of ],nM@I[0-the input noise, and is not considered here. 1g "1cThe final error after the dot product is proportional to the 1-D FFT error and the proportionality 1,u@=factor is (Number of points in the dot product)"1/2. d`,ha+Number of points in the dot product = 900. f`r.For 1-D FFT, the time calculated = 552 years. hM1Error after the dot product, time =   years. tUR UT`s+Comparison of the Various FFT Error Bounds  teTable 3 gives the time in years for handling 10-bit input data stored as 15 bits for the various FFT Ite berror bounds in the literature, taking truncation error into account. From this table we see the -Htit\assumption that the systematic error is 1/10th the total FFT error is quite reasonable.  +UU #r [This table shows the error in an integer FFT for assumptions about Systematic Error in the M7UU#th]input signal. It gives the number of years for handling 10 bit input data stored as 15 bits CUU#ioWfor one minute cadence observations. It shows the number of observations required for berOUU#do\the Systematic Error to equal random error (mainly shot noise) in the signal at this stage [UU@#FoYin the calculation. It compares several estimates of the FFT error from the literature. c.yG~ Z /ar$times[num[2,"2"],sqrt[num[12,"12"]]]VAx zHvs x haHp_{A* 1Hp_, (terUU  linIf systematic error is considered equal to the total FFT error rather than 1/10th of the FFT error, and 1 bit UU@ rr?scales to 1/4 m/sec, for 10 bit input data stored as 15 bits, UU` t e*UUh n   g6UU` io aBUU`h ro#Time for the 1-D FFT = 5.52 years. ut NUU`D th uZUUhn ha0Error after the dot product, time =  years. I!9|Gce%  pequal[over[num[552,"552"],sqrt[num[900,"900"]]],over[num[5.52,"5.52"],num[30,"30"]],over[num[1,"1"],num[6,"6"]]]roZ|:w[T}H*l %!!{ Щh.~H Iyy im5[':Gerʭ:pequal[char[n],indexes[1,0,matrix[1,1,over[indexes[0,1,char[sigma],char[v]],times[char[S],char[E]]]],num[2,"2"]]]K!9H}#* 0||{ HP%2G%Rtimes[over[num[1,"1"],char[N]],sum[indexes[1,0,abs[times[indexes[0,1,char[X],num[0,"0"]],id[char[j]]]],num[2,"2"]],equal[char[j],num[0,"0"]],plus[char[N],minus[num[1,"1"]]]]]vC7[):H#%*s , UUGv!<G# $ZUUgequal[over[indexes[0,1,char[sigma],char[v]],sqrt[char[n]]],Rightarrow[times[char[S],char[E]],prompt[]]]UU@JP'2HucA  rGceaF %over[num[14312,"14312"],num[30,"30"]]0j.3j'4hHm[    1"dAH%"rHH 2, HHB GHHx<arde UT UT`~,o Conclusion ,1,# ]]bFrom the above analysis we see that the FFT error is negligible in comparison to other sources of 1Gderror at that stage in the calculation. As the DFT is a linear operation, we can do the detrending ,c?@,ibefore or after the FFT. ucUR UT` References N],} ]]][1]P.D. Welch, A fixed-point fast Fourier trnasform error analysis, Trans. IEEE, Vol. AU- @eq17, 1969, pp. 151-157. [si [ca[2]C.J. Weinstein, Quantization effects in digital filters, MIT Lincoln Lab. Tech. Rep. 468, @*ASTIA Doc. DDC AD-706862, Nov. 21, 1969.  12^[3]B. Liu and Tran-Thong, Fixed-point fast Fourier transform error analysis, Trans. IEEE, @!Vol. ASSP-24, 1976, pp. 563-573. `G[4] L. Bacon, Fixed-Point Fast Fourier Transform, SOI-TN-009, 1991. HUP UT` Appendix 1 ar?hi+From Weinstein [2], number of years =   bFrqhsi5Error after the dot product, time =   = 477 years hof ,From Tran-Thong [3], number of years =   hehpth8Error after the dot product, time =   = 17387 years. `qbeHThe above is the time for handling 10 bit input data stored as 15 bits. n'2hGnt8Ftrcong[indexes[1,0,matrix[1,1,over[num[0.351,"0.351"],over[num[2.7e-05,"0.000027"],cross[num[10,"10"],num[4,"4"]]]]],num[2,"2"]],num[521605,"521605"]]n &rGLiaFh.'over[num[521605,"521605"],num[30,"30"]]Doc^.3p'4hH  B.H#A faH#  ns, UU r jOne bit is used for overflow checking, and this is an extra overhead for the algorithm. If detrending is UUr 1.ldone on the ground with the ramp present, then there is always overflow, so a division by 2 is performed to UU@r Eroremove the overflow, and there is no need for this extra overflow checking bit. Thereby the data is truncated.  (rHte 2, e H Hv H Nx t `vas s33 Hv 33 Ox  riovfw1" vn'33 Hv 0,'33 Px  5,5"fxr Go33 Hv [no33 Qx  .3hfy  zffO33BHv AzffO33BRx $`z UHHv flHSx  f afen g YHv grYTx alve`onSE = total FFT error HHv loHUx  rfecfth aizffO33Hv HzffO33Vx `{Welch (years) HHv HWx `18.4 YHHv YHXx H`5.52 HHv HYx `1/6 zffO33(Hv zffO33(Zx A | Weinstein @|(years) zH(Hv H([x `477 nYH(Hv YH(\x xal`143 onEH(Hv H(]x `4.77 )GG f$Lfract[indexes[1,0,num[2,"2"],minus[times[num[2,"2"],char[b]]]],num[12,"12"]]Wezff6O33(Hv Hzff6O33(^x x } Tran-Thong @}(years) H [+GH 6H(Hv 6H(_x `x17387 Y6H(Hv Y6H(`x `x5216 6H(Hv n 6H(ax `173.86 HzffO33(Hv zffO33(bx  Hf H(Hv H(cx   Error after @x dot product YH(Hv YH(dx t[`"21-D FFT imnH(Hv WeH(ex  x Error after x@ dot product 33 Hv 33 fx  f HəHv əgx 17`SE = 1/10 total FFT error əH(Hv əH(hx `1-D FFT əHHv əHix ff33`552 ffəH(Hv əH(jx `14312 ə6H(Hv ə6H(kx `uc521605 YzffO33Hv zffO33lx t[`FF mnə Hv Weə mx `x1Assumption of magnitude of Systematic Error (SE) HHv Hnx  f HYHHv YHox  17f E1HHv Hpx  f 48G 5?H sqrt[char[N]]& Hf x& h )fff  S*48H  1& Hf 1& h )f  fe& Hf fe& h )fx [H33-H?f nH33-h `"xj a33@-Hf mpa33@-h  `'Name -Hf x-h `)NTRANS (T!j) `Noise Transfer Function 1Vff-Hf 1Vff-h x`1 Additive `48Noise (N!j) 5?feK-Hf feK-h `4 Output Noise h  8HH33$Hf H33$h &*UU`B#0 a33@$Hf a33@$h fe&*UU`H$- $Hf $h 33`J- 1Vff$Hf 1Vff$h 33`L- feK$Hf NafeK$h x`S- hH33$Hf H33$h No*UU`U#io1 a33@$Hf a33@$h x*UU`f# Telescope $Hf j$h fehg  h1Vff$Hf 1Vff$h  8h  feK$Hf hfeK$h h  3H 33$Hf H 33$h *UU`#2 a33 @$Hf a33 @$h ff*UU`# CCD Expose Vff $Hf  $h h  1 Vff$Hf 1 Vff$h Hh33  fe K$Hf fe K$h h  H-33$Hf UUH-33$h H*UU`#3 a33-@$Hf a33-@$h *UU`# CCD Read -$Hf -$h HhK  1-Vff$Hf 1-Vff$h Hh33  fe-K$Hf fe-K$h h  HQ33$Hf UUHQ33$h @*UU`#4 a33Q@$Hf a33Q@$h ffH*UU`# CCD Sums Q$Hf Q$h h  1QVff$Hf 1QVff$h h  feQK$Hf *feQK$h Hh  Hu33$Hf Hu33$h *UU`#5 a33u@$Hf a33u@$h ffH*UU`#V Calc $u$H f u$h h  1uVff$ H f 1uVff$h h  feuK$ H  f *feuK$h Hh  H33$ H  f H33$h *UU`#6 a33@$ H  f a33@$h ffH*UU`# Fix bad pix $ H f $h HhK  1Vff$H f 1Vff$h Hh33  feK$Hf feK$h h  H33$Hf UUH33$h H*UU`#7 a33@$Hf a33@$h  *UU`# Sub V%o $Hf $h @h  1Vff$Hf 1Vff$h  h  feK$Hf UUfeK$h h  H33$Hf H33$h H*UU`#8 a33@$Hf a33@$h *UU`#Remap $Hf $h h  1Vff$Hf 1Vff$h h  feK$Hf *feK$h Hh  H33$Hf H33$h H*UU`#9 a33@$Hf a33@$h *UU`$W# Subtract $Hf $h HhK  1Vff$Hf 1Vff$h Hh33  feK$Hf feK$h h  H)33$H f UUH)33$h *UU`#10 a33)@$ H!f a33)@$h *UU`#Apodize )$!H "f )$h *h  1)Vff$"H!#f 1)Vff$h h  fe)K$#H"$f fe)K$h @33h  hHM33$$H#%f HM33$h H*UU`#11 a33M@$%H$&f a33M@$h H*UU`#ffFFT M$&H%'f M$h HhK  1MVff$'H&(f 1MVff$h Hh33  feMK$(H')f feMK$h Hh  Hq33$)H(*f Hq33$h "*UU`#12 $a33q@$*H)+f a33q@$h #*UU`# Dot Read q$+H*,f q$h HhK  1qVff$,H+-f 1qVff$h Hh33  feqK$-H,.f feqK$h Hh  H33$.H-/f H33$h H*UU`#13 a33@$/H.0f a33@$h H*UU`#ff Data Comp. $0H/1f $h h  1Vff$1H02f 1Vff$h h  feK$2H1f feK$h h   _r?x4G9BŇf equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)2,"2"]],sin[times[char[(*s9.00s*)c],matrix[(*n*)1,1,over[times[(*n*)char[(*s9.00s*)pi],char[(*f"MT"fs9.00s*)l]],num[(*s9.00s*)3000,"3000"]]]]],num[(*s9.00s*)0.955,"0.955"]] rAfx6G7B@f equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)1,"1"]],sin[times[indexes[1,0,char[(*s9.00s*)c],num[(*s7.00s*)2,"2"]],matrix[(*n*)1,1,over[times[(*n*)char[(*s9.00s*)pi],char[(*f"MT"fs9.00s*)l]],num[(*s9.00s*)1500,"1500"]]]]],num[(*s9.00s*)0.6899999999999999,"0.69"]]巚6&a7HwfUUL66 x+QQ 8G;- 5?cequal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)3,"3"]],id[prompt[]],num[(*s9.00sn*)0.02,"0.02"]] &a9HiYfL44 5x+MQ :G=* 5?`equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)4,"4"]],id[prompt[]],num[(*s9.00s*)2.8,"2.8"]]_:J[AT;H]BfŇ88 dex+QQ <G?.0- 5?n[bequal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)5,"5"]],id[prompt[]],num[(*s9.00s*)0.66,"0.66"]]^JUAT=H^Ef[(:: ]]x+FVQ >GA&@ 5?\equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)6,"6"]],id[prompt[]],num[(*s9.00s*)1,"1"]]s*J[AT?H`Gf*n<< chx+FVQ @GC0s&@ 5?.0\equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)7,"7"]],id[prompt[]],num[(*s9.00s*)1,"1"]]JLATAHbIf+>>  vfxBGEs[E@f00equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)8,"8"]],sin[times[indexes[1,0,char[(*s9.00s*)c],num[(*s7.00s*)2,"2"]],matrix[(*n*)1,1,over[times[(*n*)char[(*s9.00s*)pi],char[(*f"MT"fs9.00s*)l]],num[(*s9.00s*)3000,"3000"]]]]],num[(*s9.00s*)0.955,"0.955"]]umJLATCHeKfA@@ Ňᷚ6&aEHgMfQLBB .0x+FVQ GGI1,&@ 5?)T\equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)9,"9"]],id[prompt[]],num[(*s9.00s*)1,"1"]]x+PQ HGK+- 5?bequal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)10,"10"]],id[prompt[]],num[(*s9.00s*)0.9,"0.9"]]JLATIHkOfAGG *nux+UQ JGMVQ0 5?eequal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)11,"11"]],id[prompt[]],num[(*s9.00sn*)0.03,"0.03"]]s6JZATKHmQfHHH! 5x+YQ LGOG3 5?s[fequal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)12,"12"]],id[prompt[]],num[(*s9.00s*)0.035,"0.035"]]ZJ`ATMHoSfmaJJ& esx+IQ NGQMT( 5?nu^equal[indexes[0,1,char[(*s9.00s*)T],num[(*s7.00s*)13,"13"]],id[prompt[]],num[(*s9.00s*)1,"1"]]~JfATOHqUf6LL+ QJQATQHsWfVQNN0 y 5?dyLeftdz.0Right9],d{00 Referencexd|}~d}|eqd~|00d~"]d*s*dTdfdvQd.f[ 6J$HQHHFormulaG fPum[(Title0Bodyf ^eq$s[1,*s00nuFormula)idfQQHHeading Bodyf# ereCellBodyf *Footnotef   f Footerf H   HeaderffT y TableTitleT:Table : fq $Bodyf$ dinCellBodyfQereHeadingy Body f tno$Body f  Bodyf  CellHeadingdf $TaeTe Formula f BodyfT  TableTitleT:Table : f, CellBody f $$Body$f $BoFormulafT  TableTitleT:Table : f e CellBody f+PTitleBody f,P TaTitleBody+>f  Footnote@   $ Hdy l   f    D  h    Normal f 5TaBodyTa f.PTitle Body f $TiBodyfQ  SubHeading . Bodyf CellBodyfT l TableTitleT:Table : f $fFormulaf $Formula.f $FormulafPAppendix BodyfPfAppendix Bodyf CellBodyfQa f SubHeading . Bodyf CellBody*f CellBody f $Body666 f#  $captionBo666 f  $Apnd caption$fQHeading Body 9    ing    Superscript    Subscript   yEmphasis       Subscript    Bo ! Subscript " Superscript # $ % Subscript+He, . AQCABDEFGHThinMediumDoubleThick@ Very ThinSingle% Hxp0p p ` !% HHHHHFormat A HHHHHFormat B HHHHHFormat A HxHHH0HFormat Bp HHHHHFormat AM]OD:M o33nkN FoO33HHHH&?O 33@VffhKH` M.0rD(M1H2DMmat34DM56HDMH780DM9:DM;H<!DM=>HH& Oh?@o33NH NxHzffo33Nx!Kzffo33Nx(Mzffo33(NxMmatzff6o33(NxH MHzffo33(NxMMzffo33NxM!HH&--Oho33NH&$$OhHH&$$OhKH &$$Oh(MH-&$$OhMmatHQ&$$OhHMHHu&$$OhM  MH&$$Oh  M HH&$$Oho33NH&$$OhHH&$$OhKH)&$$Oh !("M#HM&$$Oh$M%mat&'(Hq&$$Oh)H*+M,H-H&$$OhM./012MyzyCommentO Helvetica MTSymbolTimesRegularExtraRomanMediumRegularBold RegularItalicKX8-̡0ܖlx;]hc*ZaDtE}RǑ!Tx:֖FppǮSb. Gjl"8zgb⻁ vkUdY'JXfi -wi< AZ q|nDP+I7%R.9 t//agYƉ'yZ=Mֿ{6D-MUR?