Investigation of Spe tral Baseline Properties of theGreen Bank Teles opeEle troni s Division Internal Report No. 312J. R. Fisher, R. D. Norrod, D. S. BalserSeptember 2, 2003Contents1 Introdu tion 21.1 System Des ription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Antenna Noise 42.1 L-band (1.4 GHz) measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 C-band (5 GHz) measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.3 X-band (9 GHz) measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.4 Antenna noise e�e ts on observations . . . . . . . . . . . . . . . . . . . . . . . . . . 123 Antenna and re eiver response to a ontinuum radio sour e 143.1 Quasi-periodi ripples in ontinuum sour e noise . . . . . . . . . . . . . . . . . . . 143.2 Feed/LNA noise model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173.3 Fine s ale stru ture due to avity resonan es . . . . . . . . . . . . . . . . . . . . . 213.4 Continuum sour e spe trum stability . . . . . . . . . . . . . . . . . . . . . . . . . . 253.5 Cal spe trum ripples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 IF system 324.1 2.4-MHz ripple from opti al modulators . . . . . . . . . . . . . . . . . . . . . . . . 324.2 Baseline Periodi ities Produ ed in the IF Ele troni s . . . . . . . . . . . . . . . . . 374.3 Changing Cable Lengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404.4 GBT Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414.5 IF Converter Ra k Ripple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415 Spe trometers 465.1 Spe tral pro essor wide-bandwidth distortions . . . . . . . . . . . . . . . . . . . . . 465.2 Auto orrelator linearity tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465.3 O�set problem in assembling omposite re eiver Tsr /Tsys spe tra . . . . . . . . . 486 RFI 497 Summary 508 Atta hments 521

1 Introdu tionOne of the primary motivations for the o�-axis design of the Green Bank Teles ope (GBT) wasto redu e the e�e ts of multi-path interferen e, often referred to as standing waves, betweenthe feed, subre e tor, and main re e tor of the antenna. This interferen e auses low-levelripples in the frequen y dependen e of the gain of the antenna, whi h an mask weak spe trallines in the presen e of ontinuum radiation from the spe tral line sour e. Re e tions in theantenna stru ture an also introdu e frequen y dependen e of the total system noise whi h maynot ompletely an el in on-sour e/o�-sour e spe tral di�eren es or may show up as se ond-ordere�e ts in the alibration arithmeti . With an o�-axis paraboloid the spe ular re e tion areas that ause the strongest gain and noise ripple are at the inner edge of the dish and subre e tor wherefeed illumination is relatively low. Hen e, the e�e ts of these re e tions should be orrespondinglylower.As will be shown in this report, re e tions within the antenna are not the only sour e ofspe tral baseline distortions. To realize the full potential of the GBT design are must be takento minimize the e�e ts of transmission line and waveguide re e tions, ampli�er and �lter insta-bility, data sampling quantization, and other more subtle sour es of spe trum distortion. Theinvestigation des ribed here fo used on separating the various e�e ts as a guide to engineeringimprovements and alibration te hniques that will follow.Be ause of the laims of better baselines with an o�-axis teles ope it is natural to ompareGBT results with spe tra from symmetri antennas. Sin e we do not have mu h dire tly orre-sponding data from other teles opes, these omparisons must be largely ane dotal. In makingthese mental omparisons from your own experien e keep in mind a few fa tors. First, thebandwidths of most of the spe tra shown here are onsiderably wider than have been used at entimeter wavelengths in the past. Se ond, the re e tion distan es in a 100-meter antennamakes the orresponding ripple periods of multi-path interferen e smaller than have been seenon smaller teles opes. Third, the spe tra shown here were spe i� ally designed to emphasizespe tral baseline distortions so they are not ne essarily representative of spe tra that one wouldexperien e under typi al observing onditions.Most of the spe tra to follow are (on� off)=off ve tor di�eren es and quotients, where thenormal assumptions are that the only di�eren e between the on and off spe tra is the e�e tbeing measured and that the frequen y dependen e of the system gain is normalized by dividingby the system noise spe trum. Spe tral baseline distortions are largely due to one or both ofthese assumptions being in orre t. Equation 1 is a partial separation of the gain and noise termsin the (on � off)=off ve tor, keeping the assumption that the only di�eren e between the onand off spe tra is the inje tion of the ontinuum radio sour e power, Psr (f).on� offoff = Gsr (f)Psr (f)G1(f)Pbknd(f) +G2(f)Pspill(f) +G3(f)Pwg(f) +G4(f)PLNA(f) (1)The noise spe tra of the radio sour e, Psr (f), ba kground, Pbknd(f), and spillover, Pspill(f), areonly weakly frequen y dependent, but the waveguide ohmi loss noise, Pwg(f), and ampli�er noise,PLNA(f), probably have a moderate to strong frequen y dependen e. Likewise, the gains of thesystem to the radio sour e and ba kground noise, Gsr (f) and G1(f), are probably nearly equal,but the frequen y dependen e of the gains to spillover, G2(f), waveguide, G3(f), and ampli�er,G4(f), noise are signi� antly di�erent sin e they only partially share signal paths. Hen e, theoff spe trum, whi h is the denominator of Equation 1, will not be simply proportional to thesystem gain to the radio sour e, Gsr (f).2

1.1 System Des riptionTypi ally, �ve to nine ryogeni re eiver front-ends are installed on the GBT and available for use.All of the urrent front-ends are built around HFET ampli�ers ooled to 15 Kelvin by losed- y lerefrigerators, and, depending on the frequen y range, have one, two, or four feed horns. Ea hfeed horn is followed by a dual-polarized re eiver with two independent hannels, so the existingfront-ends have two to eight hannels. A ommon IF system has eight hannels and mi rowaveswit hes to handle the hanges in onne tions required when a parti ular re eiver is sele tedand fa ilities to frequen y onvert the front-end signals to appropriate ranges for digitization.Figure 1 shows a very simpli�ed blo k diagram of a re eiver hannel and riti al portions of the ommon IF system. In this �gure, omponents to the left of the �ber are lo ated on the GBTfeed arm; items to the right are lo ated in the Equipment Room in the Jansky Lab. Following isa brief des ription of ea h omponent of this diagram.Feed Horn Above 600MHz all the GBT feed horns are ir ular orrugated horns. Below 4GHzthese feeds are fabri ated using a \hoop-and-washer" te hnique. The orrugations areformed by sta king hoops and washers under pressure and then spot-welds are used to holdthe assembly together followed by the addition of a �berglass shell for sti�ness. Above 4GHzthe feed horns are ma hined from aluminum in se tions eight to ten in hes long ontainingseveral orrugations per se tion.Front-end Ele troni s Following the ryogeni ampli�ers, ea h hannel is frequen y onvertedto a �rst IF enter frequen y of 1080, 3000, or 6000MHz. The �rst LO (LO1) is used fordoppler tra king and frequen y swit hing if desired. The mixer is followed by IF ampli�ersand �lters.IF Router The swit h symbol between the front-end and the Opti al Driver Module in Figure 1represents a group of mi rowave PIN diode swit hes that sele t one of several possible inputsto ea h ommon IF hannel.Noise Sour e The IF Ra k Noise Sour e onsists of a solid-state diode noise sour e, ampli�er,and �lters and may be sele ted by the IF Router as an input to any of the eight Opti alDriver Modules. The Noise Sour e Module in ludes a four-way splitter that allows thesour e to drive four IF hannels simultaneously.Opti al Driver The Opti al Driver Module (ODM) in ludes ampli�ers, a �lter bank for sele -tion of the IF bandwidth, an equalizer, and a total-power dete tor and V/F onverter with�ber onne tion (not shown) to the Digital Continuum Ba k-End (DCR). The ODM's over1-8GHz, but the user may sele t one of several more narrow bandwidths entered at either3000 or 6000MHz.Laser Modulator The ODM IF oaxial output is onne ted to a Laser Modulator whi h inten-sity modulates a laser output onne ted to a single-mode opti al �ber.IF Fiber The IF single-mode opti al �ber is approximately 2.3km long, onne ting the feedarm Re eiver Room to the Equipment Room in the Jansky Lab. Fusion spli es are usedthroughout to eliminate instabilities typi ally seen with �ber onne tors.Opti al Re eiver In the Opti al Re eiver Module, the laser signal is demodulated using a pho-todiode dete tor. A mi rowave ampli�er and four-way power splitter follow the photodiodeproviding four identi al outputs of the �rst IF signal in the Equipment Room.3

Figure 1: A simpli�ed blo k diagram of one GBT re eiver hannel and asso iated IF ele troni s.Converter Module Ea h Opti al Re eiver output is onne ted to a Converter Module (CM)whi h uses a two-step onversion s heme to supply signals with high image reje tion toany of several ba k-ends. The input 1-8GHz IF signal is up- onverted to the 9.0-10.35GHzrange, and then down- onverted to the 150-1600MHz range. The various CM outputs onne t to the IF input systems of the Spe tral Pro essor, VLBI Data A quisition Ra k,Auto orrelation Spe trometer, or other ba k-ends.2 Antenna Noise2.1 L-band (1.4 GHz) measurementsEarly in our investigations we were puzzled by a baseline ripple in frequen y-swit hed data near1.4 GHz with a ripple period of about 1.6 MHz. It evidently had something to do with the antennabe ause the phase of the ripple hanged with subre e tor position, and the phase hanged byhalf of a period with an 1/8-wavelength shift in the subre e tor toward the main dish. Thisripple period showed up again in later total power measurements where the blank sky on andoff spe tra di�ered only by a displa ement of the subre e tor.Figure 2 shows 50-MHz bandwidth di�eren e spe tra near 1.4 GHz for the two linear re eiverpolarizations at 1/8- and 1/4-wavelength subre e tor displa ements between the on and offspe tra. Several notable features an be seen in these spe tra. First, there are quasi-sinusoidalripples with periods of about 1.6 and 9 MHz. Se ond, the 1.6-MHz amplitude is strongest in the1/8-wavelength subre e tor displa ement spe tra, and the 9-MHz amplitude is strongest in the4

Figure 2: Cold sky (on - o�) / o� spe tra at an elevation of 80 degrees, where 'o�' is with thesubre e tor in its nominal fo us position and 'on' is with the subre e tor displa ed in the +Ydire tion (roughly toward the enter of the main re e tor). The red or darker urves are forre eiver hannel X and the green, lighter urves are for hannel Y. The top urves are with 1/8-wavelength (26.7mm) displa ement and the bottom urves are with 1/4-wavelength (53.3mm)displa ement. The bottom urves are o�set by -0.02 in the verti al dire tion to separate theplots.1/4-wavelength displa ement spe tra. Third, the 1.6-MHz ripple amplitude is highest in the Ylinear polarization whose E-ve tor is parallel of the plane of symmetry of the teles ope.The waviness in the spe tra in Figure 2 is a frequen y dependen e in the noise power enteringthe re eiver feed and has something to do with the position of the subre e tor with respe t tothe rest of the opti s. Another lue to the origins of the spe tral ripples an be found in Figure 3,whi h shows periodograms generated by taking the Fourier transform of the spe tra in Figure 2after removing the RFI and 1420.4{MHz radiation from Gala ti hydrogen. The period of the1.6-MHz ripple is spread over a range of periods from about 1.3 to 1.8 MHz. If we assume thatthis ripple is due to multipath interferen e, the path length di�eren e ranges from about 165to 220 meters (300 meters divided by the ripple period in MHz), whi h is roughly the range oftwi e the distan e from the Gregorian feed to the main re e tor via the subre e tor. The GBTre e tor geometry is shown in Figure 4. This leads us to the on lusion that part of the noisein the teles ope system ( osmi ba kground, atmosphere, spillover, or re eiver/waveguide noise)enters the re eiver through two paths: dire tly and after emission or re e tion from the feed areaand then s attered from the ir umferential gaps between the surfa e panels. Sin e these gapsrun along lines of onstant phase as seen from the feed, their s attered waves are oherent at thefeed. We expe t the s attering to be strongest in the linear polarization perpendi ular to thegap length ( hannel Y), whi h is what is observed. Also, sin e this multipath involves two extrare e tions from the subre e tor, the ripple amplitude measured with this te hniques should bestrongest for odd multiples of 1/8-wavelength. This, too is onsistent with the measurements.The strongest periodogram features in Figure 3 at about 9 MHz period are suggestive of amultipath distan e di�eren e of about 30 meters, whi h is twi e the distan e from the feed to the5

Figure 3: Averaged periodograms of several 1/8-wavelength displa ement spe tra like those inFigure 2. The red or darker lines are re eiver hannel X, and the green, lighter lines are hannelY. The verti al s ale is roughly the ripple rms amplitude in units of 0.001 of the system noisepower density (fra tion of Tsys).Table 1: Periodogram integrals in the inverse ripple period ranges of 0.6 to 0.84and 0.1 to 0.12 for the 1.4 GHz data. Periodogram IntegralsElevation se (z) Total Power 1.6-MHz(Y) 9-MHz(X) 9-MHz(Y)80.0 1.02 1.00 4.59 4.09 5.5314.5 4.00 1.25 4.44 6.93 10.569.5 6.06 1.46 5.48 4.11 13.23subre e tor. (See Figure 4.) Some of the noise in the antenna system is returned dire tly to thefeed after one re e tion o� the subre e tor. This is onsistent with the 9-MHz ripple omponentbeing strongest in the 1/4-wavelength subre e tor o�set spe tra. There is a hint in Figure 3that this re e tion is partially resolved into more than one path-length omponent, whi h willbe shown in more detail in measurements with wider bandwidths des ribed below.To determine whether the noise that is ausing the 9- and 1.6-MHz ripples is entering thesystem in the vi inity of the main beam ( osmi ba kground and atmosphere) we measured theripple amplitudes at three teles ope elevations of 80, 14.5, and 9.5 degrees to vary the amountof atmospheri noise seen by the main beam and near sidelobes. These elevations orrespond toair masses of 1, 4, and 6, whi h would add about 1.5, 6, and 9 Kelvins of noise to the system,respe tively. Table 1 shows these measurement results, whi h are the periodogram integrals inthe inverse ripple period ranges of 0.6 to 0.84 and 0.1 to 0.12 (distan e ranges of 87 to 123and 15 to 18 meters) as a fun tion of teles ope elevation. The integrals are orre ted for risingsystem temperature with se (z) so that the values are proportional to temperature, not the ratioto system noise power. The 1.6-MHz ripple amplitude does not appear to be dependent on6

Figure 4: GBT re e tor geometry. The distan es from the feed phase enter to the main re e torvia the subre e tor are 80.9 and 125.9 meters for the inner and outer edges, respe tively. Thedistan es to the inner and outer edges of the subre e tor from the feed phase enter are 15.9 and13.5 meters, respe tively.elevation so we tentatively on lude that the noise engaged in this multipath interferen e is notentering through the the forward dire tion of the GBT. Channel Y (polarization perpendi ularto plane of symmetry) shows a strong elevation dependen e in the 9-MHz ripple, but hannelX does not. These measurements may be onfused by the superposition of several multipathsthat are only partially resolved at the distan e of 15 meters. The distan e between the feed andsubre e tor hanges on the order of 15 entimeters due to fo us tra king as the teles ope movesfrom zenith to horizon so the multipath interferen e patterns are expe ted to hange somewhat.Noise omponents in the antenna system that are not expe ted to hange signi� antly withteles ope elevation are due to thermal losses in the feed/OMT/LNA and rearward main re e torspillover. Ea h of these is estimated to be about 5 K as seen by the LNA, but the edges ofthe feed and surrounding stru ture is bathed in noise at somewhat higher temperature due tospillover. The feed is designed to reje t most of this noise, but it is there, nevertheless. It is thiso�- enter spillover noise radiation that is re e ted ba k into the GBT opti s and interferes withthe dire tly-re eived spillover noise in the feed. Hen e, when omputing the return loss requiredto produ e the ripple amplitude observed one may need to assume a somewhat higher re e tedspillover noise omponent. Without a physi al opti s analysis of the spillover radiation we anonly guess at the amplitude of this omponent, but some rough al ulations are instru tive.To estimate the magnitude of the re e tion oeÆ ient from the panel gaps required to produ ethe measured 1.6-MHz ripple, let P0 be the power dire tly entering the feed and PR be the oherent power transmitted or re e ted into the opti s. Sin e we will be omputing power ratioswe an normalize to an impedan e of one ohm and take the ele tri �eld amplitudes asV0 =pP0; VR =pPR (2)Let L, in dB, be the return loss of the power, PR, s attered ba k into the feed, and let � bethe equivalent voltage re e tion oeÆ ient.L = �10 log10 (�2) (3)The measured relative powers at the peak and trough of the baseline ripple will then beP+P0 = (V0 + �VR)2V02 � 1 + 2�VRV0 (4)7

and P�P0 = (V0 � �VR)2V02 � 1� 2�VRV0 (5)The relative peak-to-peak and rms ripple amplitudes will then bePp�pP0 = P+ � P�P0 = 4�VRV0 (6)and PrmsP0 = Pp�p2p2 P0 = p2 � VRV0 (7)Let's assume that the noise due to ohmi losses in the feed and waveguide omponents istransmitted oherently in both dire tions, toward the LNA and toward the antenna re e tors.Assume, too, that its temperature is 5 K and that the system temperature is 20 K. In the notationof the equations above, Psys = 4 PR = 4 P0 and VR = V0 = pP0. Then, from Equation 7� = 2p2PrmsPsys (8)From Figure 3a we see that the highest 1.6-MHz ripple omponent for the polarization per-pendi ular to the ir umferential panel gaps is about 0:001 � Psys, but this is twi e the valueof Prms=Psys given in Equation 8 sin e two ripple patterns with a 1/2-wavelength shift weresubtra ted. Hen e, � � 2p2� 0:0005 = 1:4� 10�3 (9)and L = 57dB (10)For the linear polarization that is largely parallel to the panel gaps the highest rms ripple ampli-tude shown in Figure 3a is only about 0:0001�Psys, whi h orresponds to a return loss of about77 dB. The return losses omputed from the highest 9-MHz ripple amplitudes seen in Figure 3bare about 45 and 51 dB for the polarizations parallel and perpendi ular to the teles ope's planeof symmetry, respe tively.In a memo dated February 6, 2001 Norrod and Stennes report on re e tometry measurementsof the GBT through the L-band feed. At the distan e of the subre e tor they measured returnlosses of about 57 and 60 dB for the linear polarizations parallel and perpendi ular to the planeof symmetry, respe tively. These values are 12 and 9 dB higher than derived in the al ulationsabove, whi h would indi ate that there is more noise being inje ted into the multipath interferen ethan the 5 K assumed. This may argue that spillover has a signi� ant role in produ ing the 9-MHz baseline ripple. The re e tometry measurements resolved the returned power into two orthree omponents separated by as mu h as 1.6 meters in one-way distan e. This and the partialresolution of the 0.1/MHz feature in the periodograms suggest that several re e tions from thesubre e tor are mixed together in the 9-MHz period baseline ripples.At the distan e range of the main re e tor Norrod and Stennes measured a return loss lowerlimit of 83 dB for the perpendi ular polarization, whi h is lose to the value of 77 dB from theripple amplitude al ulations. Unfortunately, they did not make a measurement in the polariza-tion parallel to the plane of symmetry, whi h, in retrospe t, would make a better omparisonwith the 1.6-MHz ripple results.8

Table 2: Periodogram integrals in the inverse ripple period ranges of 0.56 to 0.75and 0.095 to 0.115 for the 5-GHz data. Periodogram IntegralsElevation se (z) Total Power 1.6-MHz(Y) 9-MHz(X) 9-MHz(Y)65 1.1 1.00 1.08 0.24 1.1611 5.2 1.32 0.42 0.65 0.726 9.6 1.65 0.83 0.95 1.182.2 C-band (5 GHz) measurementsFigure 5 shows wider, 200-MHz bandwidth spe tra taken at 5 GHz with the same displa edsubre e tor di�eren ing te hnique used to generate the 50-MHz bandwidth spe tra at 1.4 GHz inFigure 2. The Y hannel, whose linear polarization is parallel to the teles ope plane of symmetry,in Figure 5 again shows the stronger 9- and 1.6-MHz ripples. The periodograms of these spe traare shown in Figure 6. The highest ripple omponents in the periodograms have Prms=Psysvalues of about 0.0005 and 0.0001 (whi h are twi e the a tual values be ause of the subtra tionte hnique) at inverse periods near 0.11 and 0.6/MHz, respe tively. Using the same assumptionsof Psys = 4 PR = 4 P0 that we used at L-band and Equations 8 and 3 we ompute return lossesof 63 and 77 dB, respe tively, for hannel Y. These losses are 18 and 26 dB higher than the orresponding values found at 1.4 GHz.The re e tion from the subre e tor that auses the 9-MHz ripple period is a ombinationof an edge di�ra tion and a near-spe ular re e tion from the area of the subre e tor lose tothe parent ellipsoid's primary axis where the surfa e is almost normal to the line of sight fromthe feed. The return loss from spe ular re e tion is expe ted to hange roughly as wavelengthsquared, or about 11 dB from 1.4 to 5 GHz. A more detailed physi al opti s omputation wouldbe required to a ount for the full 18 dB di�eren e.The 26 dB di�eren e in the panel gap return loss between the two frequen ies is more ompli- ated. One might expe t the e�e tive s attering area of the gaps to hange roughly linearly withwavelength sin e the urrent interruption is in one dimension. This would suggest only a 5 dBdi�eren e. We are left to spe ulate that the rest of the di�eren e may be due to the fa t that thegap re e tions are more resolved in distan e with the larger bandwidth used at 5 GHz (hen e,the re e ted power is spread over more omponents), and that the oheren e of the re e tionsover the full ir umferential gap extents may be partially redu ed by opti s misalignments, whi hwould have a larger e�e t at the shorter wavelength. Part of the di�eren e might also be at-tributed to lower subre e tor di�ra tion at the higher frequen y whi h would inje t less spillovernoise into the multipath interferen e.One notable feature of the periodogram in Figure 6b is that the longer period ripple at aninverse period near 0.1/MHz has learly been resolved into a number of omponents. Channel Yappears to have one strongest omponent whi h hannel X does not.Table 2 shows the results of integrating the re e ted power over the two ranges of the peri-odograms that show strong ripple amplitudes. This time the 9-MHz ripple in hannel X showsan in rease in amplitude with de reasing teles ope elevation, but the other two integral rangesdo not show this trend. In the 1.4-GHz data it was hannel Y that showed the in rease in 9-MHz ripple amplitude at lower elevation. It seems likely that the hanges in integrated rippleamplitudes are largely due to fa tors other than the amount of noise entering the system in theforward dire tion of the GBT. 9

Figure 5: Cold sky (on - o�) / o� spe tra at an elevation of 65 degrees, where 'o�' is with thesubre e tor in its nominal fo us position and 'on' is with the subre e tor displa ed in the +Ydire tion. The red or darker urves are for re eiver hannel X and the green, lighter urves arefor hannel Y. The top urves are with 1/8-wavelength (7.62mm) displa ement and the bottom urves are with 1/4-wavelength (15.2mm) displa ement.

Figure 6: Periodograms of the 1/8-wavelength (a) and 1/4-wavelength (b) displa ement spe tralike those in Figure 5. The red or darker lines are re eiver hannel X, and the green, lighter linesare hannel Y. The verti al s ale is roughly the ripple rms amplitude in units of 0.001 of thesystem noise power density (fra tion of Tsys). 10

Figure 7: Periodogram of the 1/4-wavelength displa ement spe tra on old sky at 9 GHz withan 800-MHz bandwidth. The verti al s ale is roughly the ripple rms amplitude in units of 0.001of the system noise power density (fra tion of Tsys).2.3 X-band (9 GHz) measurementsThe same displa ed-subre e tor measurements were made at 9 GHz with a spe trometer band-width of 800 MHz. At this frequen y the 1.6-MHz ripple was too weak to make useful mea-surements, but the wider bandwidth provided very good resolution on the re e tions at thedistan e of the subre e tor. Figure 7 shows the periodogram of the 1/4-wavelength displa ement(on � off)=off spe trum from ir ularly polarized hannel X for ripple periods greater thanabout 8.2 MHz. Quite a few sharp features an be seen with this higher ripple period resolution.Table 3 lists the periodogram features labeled in Figure 7 with their measured lo ation and orresponding ripple period and the path length di�eren e required in multipath interferen e to ause that ripple period. Figure 8 shows the Gregorian subre e tor and feed geometry at theorientation that it has with the teles ope pointed near the horizon. The X-band feed is shown atthe se ondary fo us used in the measurement of the periodogram of Figure 7. The L-band feedseems to shadow the X-band feed in this drawing, but this is only the appearan e in pro�le. F1and F2 are the prime and se ondary fo us lo ations, respe tively. Ray path F2-A-OA goes to thefar edge of the main re e tor, and F2-B-IB goes to the edge of the main re e tor nearest the axisof the parent paraboloid. Lines B-OB and A-IA are parallel to the F2-A-OA and F2-B-IB rays.The lateral displa ement of the subre e tor in making these ripple measurements was roughlyalong the dire tion of a 45-degree angle down and to the left in this �gure. F2 is in the planeof the Re eiver Room roof. The distan e F2-B is 15.9 meters, F2-A is 13.5 meters, and theL-, S-, and C-band feeds extend about 1.4, 0.9, and 0.9 meters above the Re eiver Room roof,respe tively.Periodogram features L, K, and J are almost ertainly due to re e tions from the Re eiverRoom roof and the tops of the L-, S-, and C-band feeds sending a small portion of the antennanoise signal on an extra path distan e to subre e tor edge B and ba k. Periodogram features Ethrough H ould be due to re e tions o� the stru ture of the prime fo us re eiver boom lo atedto the top left of Figure 8. Spillover noise in a given dire tion ould interfere with itself and ause11

Table 3: Tentative identi� ations of multipath interferen e features in the periodogram ofFigure 7. See Figure 8 for subre e tor/feed geometry referen es.Inverse Path LengthPeriod Period Di� fromFeature (MHz�1) MHz Period (m) Remarks or possible auseA 0.0175 57 5.25B 0.0200 50 6.00 interf. of rays IA-A-F2 & IB-B-F2 ??C 0.0250 40 7.49D 0.0350 28.6 10.49E 0.0525 19.0 15.74F 0.0688 14.5 20.61G 0.0850 11.8 25.48H 0.0925 10.8 27.73J 0.0975 10.3 29.23 S.R. edge B to L-band feed (29.0m)K 0.1000 10.0 29.98 S.R. edge B to S or C-band feed (30.0m)L 0.1062 9.4 31.85 S.R. edge B to r vr room roof (31.8m)baseline ripple, if it has two ray paths to the feed. For example, spillover noise from the inneredge of the dish enters most strongly through ray path IB-B-F2, but it an also s atter fromsubre e tor edge A and enter through path IA-A-F2. The di�eren e in these two path lengths isabout 5.9 meters, whi h might a ount for periodogram feature B. Positive identi� ation of allof the observed ripple periods would require further measurements. Whatever the explanations,most of these features must depend on hanges in relative lengths of interfering paths withsubre e tor displa ement. All of the peaks in Figure 7 are stronger with a 1/4-wavelengthsubre e tor displa ement than with an 1/8-wavelength displa ement.2.4 Antenna noise e�e ts on observationsIn total power, position swit hing observations, where the on and off positions are taken bytra king the antenna over the same hour angle or elevation tra k, the antenna noise ripples should an el in the on�off di�eren e spe trum. Changes in antenna elevation will ause both hangesin the amount of antenna noise due to the atmosphere and spillover and the geometry of theteles ope due to gravitational deformations and ompensating fo us tra king of the subre e tor.Both of these will a�e t the antenna noise ripple pattern. If on � off are done by tilting thesubre e tor or swit hing between two feed horns, then the antenna noise ripple an ellation willbe in omplete.Frequen y swit hing is mu h more problemati sin e the whole antenna noise ripple patternwill shift between the on and off spe tra. If the frequen y swit hing interval is equal to half ofone of the ripple hara teristi periods its amplitude will be doubled in the on � off di�eren espe trum.One might try intentionally displa ing the subre e tor or frequen y swit hing by a full rippleperiod between the on and off or between two on � off pairs in a way that an els the antennanoise ripple. The problem with this is that it works well for only one period or usually onlyone multipath route. The panel gap re e tions ause ripple periods from about 1.2 to 1.8 MHz,and these re e tions arrive from di�erent dire tions relative to any subre e tor displa ementaxis. The same period and dire tion of arrival spread is true for the longer ripple periods due tofeed area re e tions. One also needs to be aware that the phases of the panel gap ripples move12

Figure 8: Subre e tor and Gregorian feed geometry with teles ope pointed near horizon andX-band feed in position. B-IB and A-OA are ray lines to the edges of the main re e tor.twi e as fast with subre e tor displa ement as do the ripples due to multipaths that involve onlyone additional subre e tor re e tion. A few simple experiments with subre e tor displa ementindi ate that the limit on the amount of ripple amplitude redu tion to be expe ted is on theorder of a fa tor of three and often onsiderably less. At a spe i� frequen y and bandwidthone might be able to �nd a parti ular frequen y swit hing subre e tor displa ement pattern thatdoes better, but it will require some experimentation. There may also be some merit in usinga mu h wider bandwidth than the astronomi al spe trum requires to model the antenna noisebaseline ripples over the part of the spe trum without spe tral lines in hopes that the model willprovide a good interpolation in the spe tral line region of the spe trum.A more subtle e�e t to wat h out for is that the antenna noise portion of Tsys will have afrequen y dependen e. As mentioned in onne tion with Equation 1, this will show up in the(on � off) = off spe trum of a sour e with signi� ant ontinuum radiation.Finally, a rough estimate of the amplitude of the antenna noise ripple to be expe ted fromthe GBT ranges from about 5 to 40 mK rms at 1.4 GHz to about 2 to 8 mK at 9 GHz, with the9-MHz ripple being roughly three times stronger than the 1.6-MHz ripple. More detailed values an be found in the �gures in this se tion. Careful total power, position swit hing an redu ethese amplitudes by a fa tor of 30 or better.One antenna noise sour e of baseline ripples that we did not investigate is the sun. Thisadds the additional parameter of the position of the sun with respe t to the teles ope pointingdire tion, and it was more than we had time for in this investigation. It is ertainly an aspe t ofthe teles ope that needs better quanti� ation.13

3 Antenna and re eiver response to a ontinuum radiosour eFor a good understanding of the spe tral hara teristi s of the antenna, feed, and re eiver front-ends of the GBT we need to separate the frequen y dependen ies of gain and noise power. Thesetwo quantities must then be hara terized for di�erent parts of the antenna/front-end systemto the extent possible. The previous se tion des ribed the far-o�-axis antenna gain e�e ts onenvironmental noise power entering the system. This se tion deals with the main beam system hara teristi s and the signal path through the feed and low-noise ampli�ers.There is no laboratory or absorber-type noise sour e that an be fed into the re eiver or feedinput and assumed to be spe trally at to the levels of interest to radio astronomy. Return-lossmismat hes smaller than -60 dB an be signi� ant, and it is impra ti al to onstru t test noisesour es to this a ura y for the wide variety of re eivers and test points in the GBT. Hen e, weare left to tease apart the various gain and noise spe trum e�e ts beginning with the assumptionthat the spe trum of a ontinuum noise sour e in the main beam of the teles ope is smooth andvaries with a something like a simple power of frequen y a ross the measured passband.As was done in the investigation of antenna noise, the antenna opti s geometry an be variedto un over the e�e t of the antenna's gain on the radio sour e's signal. The major re e torantenna gain e�e t that we expe t is multipath interferen e, so the periodograms of on � offspe tra are again useful.For ea h re eiver the antenna response to a ontinuum radio sour e was investigated byperforming on � off observations toward ontinuum sour es with varying intensities. Typi ally,5-minutes were spent on and then 5-minutes off the ontinuum sour e. Figure 9 shows 200-MHz bandwidth observations toward the 5.7 Jy ontinuum sour e 2232+1143 at S-band (1990MHz). Three, 5-minute on � off pairs have been averaged together. The se ondary re e torwas positioned at the nominal fo us for both the on and off observations. Note the quasi-periodi , large s ale ripple with a period of � 100 MHz and an amplitude of � 3% of the systemtemperature. Similar baseline stru tures are observed in all the Gregorian re eivers and will bedis ussed below. Also present in the hannel Y data is the 9-MHz ripple thought to be aused byre e tions from the sub-re e tor and Gregorian feeds and the 1.6-MHz ripple aused by re e tionsfrom ir umferential gaps between surfa e panels.A new feature at 0.435 MHz�1 (1/2.3 MHz) is also visible in both X and Y hannels ofthe on � off spe tra. This ripple is learly dete ted in the periodogram shown in Figure 10.This frequen y is lose to the 2.4-MHz ripple dete ted from the opti al �ber modulators (seeSe tion 4.1). However, the ripple phase hanges by half of a period when the sub-re e tor isshifted by �/4, indi ating a single re e tion involving the sub-re e tor. The 2.3-MHz rippleperiod orresponds to a one-way distan e of � 65 meters whi h is the distan e between the sub-re e tor and the primary re e tor near the axis of the parent paraboloid (see Figure 4). Thisripple has also been dete ted at L-band (1400 MHz) and C-band (5000 MHz). The relative powerof the ripple in reases as �2.3.1 Quasi-periodi ripples in ontinuum sour e noiseWhen observing strong ontinuum sour es with the GBT Gregorian re eivers, quasi-periodi ripples are dete ted that appear to reside upstream of the �rst LO mixer and downstream of theopti s. The phase of the ripples are �xed to sky frequen y and do not hange when the opti alelements are adjusted; thus these features appear to reside in the feed, waveguide omponentsand LNA omponents.Figure 11 plots on � off spe tra for three ontinuum sour es at L-band. Be ause of signi� antRFI only 50-MHz bandwidths were used. Two urves are shown for ea h sour e orresponding to14

Figure 9: (on - o�) / o� spe tra toward the ontinuum sour e 2232+1143, entered at a skyfrequen y of 1990 MHz. Three on � off pairs have been averaged. The red or darker urveis for re eiver hannel X and the green, lighter urve is for hannel Y. The hannel Y data havebeen o�set by �0:1. The spe tra have been smoothed by trun ating the auto orrelation fun tionsto 2048 lags and Hanning onvolved.

Figure 10: Periodograms of the spe tra in Figure 9. The red or darker line is re eiver hannelX, and the green, lighter line is hannel Y. The spike at 0.435 MHz�1 orresponds to a rippleperiod of 2.3 MHz and a one-way re e tion distan e of 65 meters.15

Figure 11: (on - o�) / o� spe tra toward the ontinuum sour e 2316+0405 (4.68 Jy, red andorange urves in the middle), 0054-0333 (2.21 Jy, dark blue and blue-green urves at the bottom),and 0120-1520 (5.08 Jy, green and green- yan at the top). The intensity s ale of 0054-0333 and0120-1520 were s aled by 1.9 and 0.98, respe tively to put their spe tra lose to that of the�rst sour e, 2316+0405, for easy omparison. All spe tra have been smoothed by trun ating theauto orrelation fun tions to 4096 lags and Hanning onvolved.Table 4: Front-end Ripple Periods.Re eiver BandL S C XFrequen y (GHz) 1.4 2 5 9Ripple Period (MHz) 30 100 65 | enter frequen ies of 1395 and 1405 MHz. Note the � 30-MHz ripple in all spe tra that are �xedin sky frequen y and losely s ale with sour e ontinuum intensity. Larger bandwidths an besynthesized by on atenating several 50-MHz pass-bands. Figure 12 illustrates su h a ompositespe trum toward 2316+0405 that spans 190 MHz. Ea h 50-MHz pass-band has been overlappedby 10 MHz. Unless the system temperature has signi� antly hanged between su essive spe trait is puzzling why the spe tra do not mat h up in the overlapped region (this is dis ussed in moredetail below). The verti al lines orrespond to RFI. There is a hint of a � 85-MHz ripple in thesespe tra.At S-band (1990 MHz) a wider bandwidth of 200-MHz an be used. Figure 9 shows on � offspe tra toward 2232+1143 as dis ussed above. The largest wavelength ripple dete ted is at � 100MHz. These features are �xed to sky frequen y, they do not involve multi-path re e tions fromthe teles ope stru ture, and the results are generally repeatable.Similar observations were made at C-band (5000 MHz) and X-band (9000 MHz) using thelargest available Spe trometer bandwidth of 800-MHz. These re eivers also have large-s alespe tral baseline stru ture that appears to be lo ated in the front-end. The hara ter of the16

Figure 12: (on - o�) / o� omposite spe tra toward the ontinuum sour e 2316+0405 that hasa ux density of 4.68 Jy at a frequen y of 1400 MHz. The red or darker line is re eiver hannelX, and the green, lighter line is hannel Y. The spe tra have been orre ted for a sour e spe tralindex of �1:006 derived from the NVSS (Condon et al, 1998, AJ, 115, 1693) ux densities at1.4 and 5 GHz.baseline stru ture is somewhat di�erent, however. Figure 13 shows on � off spe tra toward0137+3309 at C-band. A 65-MHz ripple is learly present with an amplitude of a few per entof the total system temperature. The baseline stru ture appears more sinusoidal than the L orS-band stru ture. Additional investigation has revealed that these smooth features were at leastin part due to water on the S-band feed. In ontrast, the results of on � off spe tra toward3C48 at X-band appears less sinusoidal (see Figure 14). The baseline stru ture is irregularwith very sharp features. The hara ter of these X-band spe tra is not onsistent with typi alre e tions between di�erent front-end omponents (e.g., the feed and waveguide window). Su hbaseline stru ture has been observed before whereby weak avity resonan es are formed in thedewar resulting in signi� ant stru ture in system temperature spe tra. Possible lo ations forthis radiation are irregularities in the waveguide joints or the thermal gap lo ated in the dewar.Copper tape pla ed around the waveguide joints near the feed throat had essentially no e�e t onthe baseline stru ture. Further investigation of the sharp features in the X-band spe trum aredes ribed in Se tion 3.3.Ea h re eiver appears to have unique baseline stru ture that is lo ated in the front-end om-ponents. This has been on�rmed by hanging the �rst LO mixer and the teles ope opti s. Thephase of these stru tures is �xed to sky frequen y and is not a�e ted by sub-re e tor motion. Insome ases the baseline stru ture appears quasi-sinusoidal, suggesting re e tions between various omponents in the front-end su h as the LNA, OMT, or feed horn. Table 4 summarizes the mainripple frequen ies observed for ea h re eiver band.3.2 Feed/LNA noise modelWe know that there are re e tions within the feed and waveguide system of the GBT re eiversdue to small impedan e mismat hes with return losses in the -30 to -60 dB range. Also, the17

Figure 13: (on - o�) / o� spe tra toward the ontinuum sour e 0137+3309 that has a ux densityof 6.6 Jy at a frequen y of 5000 MHz. The red or darker line is re eiver hannel X, and the green,lighter line is hannel Y. The spe tra have been orre ted for a sour e spe tral index of �0:81derived from the NVSS ux densities at 1.4 and 5 GHz.

Figure 14: (on - o�) / o� spe tra toward the ontinuum sour e 3C48 that has a ux density of3.1 Jy at a frequen y of 9000 MHz. The red or darker line is re eiver hannel X, and the green,lighter line is hannel Y. The hannel Y data have been o�set by �0:022 for larity. The spe trahave been orre ted for a sour e spe tral index of �0:9 derived from Ott et al, 1994, A&A, 284,331 18

optimum noise impedan e of the LNA's is often at a point where its input return loss is on theorder of -10 to -20 dB. These re e tions an ause gain ripples in the feed and waveguide systemin the same way as multipath interferen e in the antenna system.We know, too, that noise is generated in the waveguide omponents due to ohmi losses and inthe LNA input stage. These noise sour es transmit oherently in both dire tions, in the intendedsignal ow dire tion toward later re eiver stages and toward the feed aperture and re e torantenna. The ba kward-traveling noise omponents an be re e ted by the small mismat hesand returned to the LNA input where they interfere with their forward-traveling omponents.Again, the extra path length of the ba kward-traveling power introdu es a ripple on the noisespe tra. If more than one re e tion and path length are signi� ant, the spe trum an be fairly omplex.Figure 15 is a s hemati representation of a re eiver feed and waveguide system. Equations 3and 6 an be used to ompute the expe ted peak-to-peak ripple amplitude from a given re e tionreturn loss in the feed-waveguide system by setting VR = V0 and P0 equal to the noise omponenttemperature. With 5 K of thermal noise traveling in both dire tions in the waveguide and returnlosses of 30, 40, and 60 dB, the peak-to-peak ripples will be 0.63, 0.2, and 0.02 K, respe tively. Ifwe assume a 20 K system temperature, these numbers would represent a peak-to-peak variationin system temperature ranging from 3.2 to 0.1%.Noise from the LNA input stage is a bit more ompli ated be ause VR 6= V0, but the ratio ofthese two values is probably less than two, so the ripple amplitude from this sour e and the waveguide re e tions should be omparable to or somewhat less than those of the ohmi loss noisesour es.Noise from the antenna ( ontinuum radio sour e, osmi ba kground, atmosphere, and spillover)will also be modi�ed by re e tions in the LNA-waveguide-feed stru ture. Be ause this begins asforward-traveling noise the e�e tive return loss to use in Equations 3 and 6 is the sum of thereturn loss that reverses its dire tion (typi ally at the LNA input) and the return loss where itis re e ted to the forward dire tion again. If we assume a total antenna noise temperature of10 K and an LNA return loss of 10 dB, then the expe ted peak-to-peak ripple amplitude fromwaveguide return losses of 30, 40, and 60 dB will be 0.4, 0.13, and 0.013 K, respe tively. With a20 K system temperature, these number orrespond to a peak-to-peak variation in system tem-perature ranging from 2 to 0.06%. A radio sour e with a 10 K antenna temperature would havea peak-to-peak variation in its re eiver power ranging from 4 to 0.13%.The period of the noise spe trum ripples due to waveguide and feed re e tions will dependon the e�e tive distan e between the noise sour es and re e tions or between two re e tions.The longest omponent in the system is generally the feed. Table 5 lists the lengths of the GBTfeeds and the orresponding ripple periods assuming a wave velo ity equal to the speed of light.The total e�e tive signal path length through the waveguide and oax line to the LNA maybe on the order of 50% longer, and the ripple periods may be s aled down by about 30%. Thelargest re e tions at the LNA, OMT, waveguide window, and feed throat are separated by smallerdistan es than the feed length so the highest ripple amplitudes would be expe ted at somewhatlonger ripple periods.| Tentative on lusion: From the ripple amplitude al ulations above and the ripple periodsgiven in Table 5 it seems unlikely that most of the observed variation in Tsr =Tsys, su h as inFigures 9, 12, 13, and 14, an be explained by feed-waveguide-LNA re e tions alone. For a moredetailed analysis of the noise and re e tion properties of the system shown in Figure 15 see theatta hed memos by R. F. Bradley (January 30, 2003) and M. W. Pospieszalski (January 28,2003).19

LNAOMT

GapWGwindow

Feed

Dewar

15K

300K

−35 dB −50 dB−35 dB −40 dB

up to 3.3 meters

3K 1K1K2K

−10 dB

10K

Figure 15: GBT re eiver feed-waveguide-LNA diagram. Thermal noise sour es and re e tionpoints are labeled with typi al temperatures and return losses, respe tively.

Table 5: Feed lengths and orresponding ripple periods implied byre e tions at this distan e. Re eiver BandL S C X Ku KFrequen y (GHz) 1.4 2 5 9 14 20Feed Length (m) 3.3 2.2 1.2 0.56 0.37 0.25Ripple Period (MHz) 45 68 125 268 405 60020

Figure 16: The normalized antenna temperature toward 3C48 (red) and 3C123 (blue). Ea hspe trum onsists of one 5 minute on-o� pair. The antenna temperature, in units of the systemtemperature, were divided by the median value in the spe trum. The spe tra are entered at asky frequen y of 9000 MHz.3.3 Fine s ale stru ture due to avity resonan esThe baselines when observing strong ontinuum sour es with the GBT tend to show obje tionablestru ture, as mentioned elsewhere in this report. The X-band (8-10GHz) re eiver in parti ularexhibited a great deal of relatively �ne-grain stru ture that makes even narrow-band observationsdiÆ ult.Figure 14 shows the results of two 5 minute, on-o� pair observations toward 3C48 (3.4 Jyat 8000 MHz). The very irregular stru ture is similar to what was reported earlier and appearsin both polarizations. The amplitude of these features s ales with sour e ontinuum intensity.Figure 16 show the normalized antenna temperature for 3C48 and 3C123 (10.6 Jy at 9000 MHz).The baseline stru ture is very similar.These baseline features appear to be upstream of the �rst mixer. Figure 17 shows the antennatemperature versus sky frequen y for two spe tra entered at a sky frequen y of 9000 MHz and9010 MHz. Noti e that most of the baseline stru ture is �xed with sky frequen y. Similar toother Gregorian re eivers there appears to be broad band frequen y stru ture whi h is upstreamof the �rst mixer but not related to the opti s.Independent analysis by Marian Pospieszalski and Ri hard Bradley showed that re e tionsbetween the ryogeni LNA (whi h has quite poor input return loss when mat hed for optimumnoise), and other input omponents su h as the feed horn aperture, an introdu e ripple in there eiver noise. As shown earlier, the (on � off)=off data pro essing auses any non-uniformre eiver noise spe trum (as opposed to gain ripples) to show in the baseline spe trum. However,21

Figure 17: The antenna temperature toward 3C48, in units of the system temperature, enteredat a sky frequen y of 9000 MHz (red) and 9010 MHz (blue). The lower panel shows an expandedview.22

the LNA-Feed Horn re e tion should be quite well-de�ned and dependent on the path length,whi h does not �t the observed hara teristi s.Figure 15 shows in some detail a typi al GBT front-end in the area between the feed hornand the LNA. The ir ular waveguide throat se tion of the feed horn onne ts through a low-lossva uum window to a thermal transition assembly, whi h provides high thermal impedan e bya small gap in the ir ular waveguide wall, and thereby allows the following omponents to be ooled to ryogeni temperatures. Following the gap is a polarizer or OMT to separate the twopolarizations, and then ea h hannel onne ts to a LNA. Not shown are stripline al ouplersbetween the OMT and the LNA; some re eivers also have ooled isolators before the LNA. Dueto ir umstan es, the Ku-band (12-15.4GHz) front-end was �rst investigated in depth, to try tounderstand the baselines evidently due to �ne-grain re eiver noise stru ture. This re eiver waspla ed in the Equipment Room and onne ted dire tly into two of the Converter Modules.The Ku-band re eiver is one that has ooled isolators in front of the LNA's, so a tight waveg-uide short over the dewar input waveguide a ts like a old load of about 20K (the isolator loadphysi al temperature, plus the e�e tive temperature of any input losses). Figure 18 shows thetotal power spe tra for the re eiver with the thermal gap as normal, and with the gap temporar-ily overed with opper tape. Based on these results, it is likely that the several large noisespikes seen in the normal state are due to resonan es or leakage related to the thermal gap. This on lusion was strengthened by S. Srikanth using the EM modeling program Qui kwave. TheKu-band thermal gap uses a simple hoke ring to provide RF isolation, and A. Kerr suggested analternative broadband hoke design. A new gap using Kerr's suggestion has been designed andis now in fabri ation to be evaluated.Sin e the X-band baselines on ontinuum sour es seemed to be parti ularly bad, when theteles ope s hedule allowed this front-end was removed and brought to the Equipment Room forsimilar testing. A shorting plate was pla ed over the input waveguide. The total power spe trumshowed one spike near 9.7GHz that seemed similar to those seen in the Ku-band total power andwere identi�ed with the thermal gap. Nothing was seen that obviously would ause the stru tureseen in Figures 14, 16, and 17. In order to simulate a ontinuum sour e observation, we de ided totry using the internal alibration noise sour es. While the noise al is inje ted after the polarizer,part of the inje ted noise will be re e ted o� the LNA isolators, travel ba k to the waveguideshort, and ba k into the re eiver. This ex ess noise should show the frequen y stru ture seen onthe teles ope, if it is generated in the re eiver. For this re eiver the low al is about 2.5K andthe high al is 15-30K.A series of s ans were taken while manually ontrolling the re eiver noise sour es. During the�rst s an pair, both the low al and the high al sour es were o�. During the se ond pair, thelow al was on. During the third pair, the high al was on. During the fourth and �nal pair,both als were again o�. Figure 19 shows the resulting s an ratios for the LCP hannel for thissequen e of s ans; the RCP hannel results are similar. While there is large ripple when s anpairs with the noise sour es on and o� are ratioed (expe ted be ause of the standing wave dueto the waveguide short), there is little eviden e of extremely sharp frequen y stru ture like thatseen on the antenna.The negative result led to a lose inspe tion of the feed horn assembly. Several quality issueswere identi�ed with the assembly:� One joint in ir ular waveguide between the feed horn and the va uum window was foundto have a mistake that resulted in a 0.015 in h gap in the waveguide wall. The joint wasre-ma hined.� Several (10-12) metal hips were found in the feed horn orrugations. These were evidentlyleft from the feed fabri ation pro ess. 23

Figure 18: The total power spe tra of the Ku-band re eiver with a shorting plate over the dewarinput waveguide. Ea h panel overs 11.7 (lower left) to 15.2 (upper right) GHz, by �ve tra eso�set verti ally for larity and in frequen y by 700MHz. The upper panel has the waveguidethermal gap as normal; the lower panel has the gap temporarily overed by opper tape.24

� The hromate surfa e �nish on the aluminum feed horn is generally non- ondu tive, and webe ame on erned that this thin diele tri layer on the ange fa es ould be a detrimentalfa tor. Hen e, we me hani ally removed the hromate oating from all the joint angefa es. (The horn is fabri ated in four se tions with bolted joints.)� A se ond waveguide joint had a noti eable oily �lm between the anges. An investigation on luded this was utting oil from the installation of some heli oil threads when the re eiverwas o� in De ember 2002, and obviously the joint was not properly leaned.After these items were orre ted, the front-end was reinstalled and observations of ontinuumsour es repeated to evaluate possible hanges in the performan e.Figure 20 shows the results of an on-o� observation of 3C218; ompare with Figure 14,on 3C48. The verti al ranges of these two �gures are the same, as a fra tion of the sour etemperatures. The baselines after the lab work, while ertainly not at, are signi� antly smootherthan earlier ones. Note also that there is little ommonality between the spe tra of the twopolarizations, unlike what was seen before orre ting the feed and waveguide defe ts. The mostlikely explanation for the improvement in ontinuum sour e baselines is that the lean-up of thewaveguide joints eliminated highly irregular losses at these joints (or one ould think of it as noisewith lots of frequen y stru ture leaking into the re eiver). The (on�off)=off alibration ausesany irregularities (even at the fra tional kelvin level is signi� ant) in the system noise to appearin the baseline response. Sin e it is almost impossible to measure feed horn loss with the degreeof pre ision needed to see these e�e ts, there is no substitute for extreme are in fabri ation,quality ontrol, leanliness, and assembly of the re eiver input waveguide and feed horn.3.4 Continuum sour e spe trum stabilityOne of the results of this investigation will be to improve the noise spe trum hara teristi s of theGBT re eivers with modi� ations to urrent re eivers and design onsiderations in new re eivers.However, it is not pra ti al to eliminate all of the distortions to ontinuum sour e spe tra so theremaining e�e ts must be alibrated. The best alibration te hnique is to measure the systemresponse to a ontinuum sour e that is as mu h like the ontinuum sour e with expe ted spe tralline radiation as possible, as is ni ely des ribed by Ghosh and Salter in the book \Single-DishRadio Astronomy: Te hniques and Appli ations," ASP Conferen e Series, Vol. 278, p521. Sin ethis generally requires moving the teles ope to a di�erent dire tion in the sky and involves a timedelay between alibration and observation of the sour e of interest, we need to know how farin time and angular distan e one an arry the alibration. This se tion presents a number ofmeasurements that address these questions.Figure 21 shows the result of using one ontinuum radio sour e to alibrate the spe trum ofanother one using the C-band (5 GHz) re eiver with a 200-MHz spe trometer bandwidth. Theleft-hand panel of this �gure shows the (on - o�) / o� or Tsr =Tsys spe tra for NGC7027 forthe two re eiver hannels. In the entral 90% of the passband the variation in sour e to systemtemperature ratio is on the order of 5% on a frequen y s ale of about 50 MHz. Mu h smallervariations an be seen at smaller frequen y s ales. The blip near 5100 MHz is probably RFI. Theintegration time for ea h of the on and off spe tra is �ve minutes.The right-hand panel of Figure 21 shows the ratio of Tsr =Tsys spe tra for 3C48 and NGC7027in the top red (dark) and green (light) urves. At this frequen y the two sour es have nearly thesame ux density so the values in the urves are lose to one. Aside from a di�eren e in spe tralindex of the two ontinuum radio sour es, the urve for re eiver hannel X is reasonably at,whi h indi ates that the ontinuum response of the re eiver an be arried between two sour esthat are at least 45 degrees apart in teles ope elevation. 3C48 was at 76 degrees elevation and25

Figure 19: (on � off)=off , for the X-band LCP hannel with the input waveguide shorted, forvarious states of the re eiver noise al sour es. Cases are: Cals O� for both s ans (red); Low alo� for 'o�' and on for 'on' (green); Low al on for both (blue); Low al on for 'o�' and high alon for 'on' ( yan); high al on for both (magenta); High al on for 'o�', and both als o� for 'on'(yellow); and both als o� for both 'o�' and 'on' (orange - the top at tra e). The s an timeswere 300 se onds. The entral portion of the upper panel is expanded in the lower panel.26

Figure 20: (on � off)=off , for the X-band LCP (red) and RCP (green) hannels, measuredon-o� 3C218 after feed horn quality problems were orre ted. Compare with Figure 14.

Figure 21: (on - o�) / o� spe tra for NGC7027 (a) and the ratio of measured spe tral powerdensities of 3C48 to NGC7027 (b) using a 200 MHz bandwidth. The verti al s ale is the fra tionof system noise power density. The red (dark) urve is for re eiver hannel X and the green(light) urve for hannel Y. In panel (b) the top two spe tra are ratio averages of (on - o�) / o�spe tra; the middle two spe tra are ratio averages of (on - o�) raw spe tra for same s an pairs;and the bottom two spe tra are ratio averages of (on - o�) raw spe tra using a di�erent s an pairfor NGC7027 with the bottom two plots o�set verti ally by -0.05 for larity.27

Figure 22: Ratio of NGC7027 ontinuum sour e (on - o�) / o� spe tra at about 11-minuteintervals extending to 3 hours 5 minutes. Panel (a) is re eiver hannel X and (b) is hannel Y.The top plot is the ratio of the spe trum from the se ond s an pair to the �rst pair. The nextplot down is the ratio of spe tra for s an pairs 3 and 1, and so forth. The plots are assignedarbitrary o�sets to reate the time progression from top to bottom. The green plot at the bottomis a repeat of the �rst plot at the top for omparison of the �rst and last ratio spe trum.NGC7027 at 31 degrees. The spe trum ratio for re eiver hannel Y varies by about 1% peak-to-peak on a s ale of about 30 MHz. This ripple period is not hara teristi of the dis overedantenna multipaths or any IF system re e tions so it requires further investigation.The lower two pairs of urves in Figure 21b show the ratios of spe tral powers from 3C48and NGC7027 using only the (on � off) di�eren e spe tra rather than Tsr =Tsys spe tra. Theresults are essentially the same. This is a bit remarkable sin e Tsys is expe ted to be a somewhathigher at lower elevation. The verti al o�set of the top pair of urves from the middle pair is dueto this di�eren e in Tsys at lower elevation for NGC7027, but the shape of the spe tral powerratio appears to be una�e ted. The re eiver is suÆ iently stable between the measurements ofthe two sour es to permit the ratio of un orre ted spe trum values to be omputed dire tly.Figure 22 shows the results of a test of system stability of the two hannels of the C-band (5GHz) re eiver with a 200-MHz bandwidth over a three hour period. In this �gure are plotted theratios of the (on - o�) / o� spe tra of NGC7027 for the two re eiver hannels at 11-minute intervalsto the �rst (on - o�) / o� spe trum at the beginning of the three hours. The spe tra for hannelX show very little hange over the three-hour period, but hannel Y shows stru ture at roughly9- and 60-MHz ripple periods. These periods are hara teristi of multipath interferen e fromre e tion from the re eiver room and IF system instabilities, respe tively, as explained elsewherein this report. Even the ratio of the �rst two (on - o�) / o� spe tra show these ripples so these hanges happen on times s ales of 10 minutes or less. Work is in progress to improve the IFstability, and the hanges in the 9-MHz ripple period from the opti s needs further investigation.One of the questions that needs to be answered in onne tion with ontinuum sour e response alibration is whether the teles ope has a di�erent response to an extended sour e than to a sour ethat is small ompared to the beam size. Figure 23 shows a test to determine the teles operesponse near the half-power points of the beam ompared to the enter of the beam. There is abit of stru ture in the ratios of the east and west o�set spe tra to the beam enter spe trum, butgenerally the ratios are what is expe ted from the de rease in beam size with in reasing frequen y.28

Figure 23: Ratios of (on - o�) spe tra of 1042+1203 taken with the teles ope pointed about 4ar minutes o� beam enter and at the beam enter. Panel (a) is for re eiver hannel X and panel(b) is for hannel Y. The red, green, blue-green, and violet lines orrespond to 4-ar minute north,south, east, and west beam o�sets. The blue lines show the expe ted ratio slope due to hangingbeam size with frequen y.However, the north and south o�set spe trum ratios show a signi� ant amount of ripple with aperiod of 2.3 MHz, whi h is the ripple period asso iated with the multipath re e tion between theedges of the main re e tor and subre e tor nearest the axis of the parent paraboloid. ChannelY also shows some 9-MHz ripple asso iated with multipath re e tions between the re eiver roomroof and the subre e tor.The data displayed in Figure 23 were taken near the meridian so the north-south o�setswere nearly in the dire tions of in reasing and de reasing elevation. Hen e, the area around thespe ular re e tion from the two surfa es moved onto or o� the edges of the two re e tors with thenorth and south o�sets so one would expe t the intensity of the multipath interferen e to hangemost in these dire tions. If a reasonably symmetri extended sour e is entered in the beam onemight expe t the ripples from the two o�set dire tion to partially an el, but these measurementsdo make the point that some aution must be exer ised when alibrating the teles ope's responseto extended ontinuum sour es.Not all ontinuum sour e spe tra were as stable as those shown in Figures 22 and 23, and eventhe best of these alibrated spe tral baselines are not perfe tly at. More extensive measurementswith longer integration times need to be done in onne tion with spe i� astronomi al problemsto more fully explore the limits of this alibration te hnique. We know, for example, that thegeometry of the GBT hanges with elevation due to gravitational deformations and intentionalmotion of the subre e tor to tra k the best fo us position. Hen e, we expe t the pattern ofthe multipath interferen e onne ted with opti s re e tions to hange with teles ope elevation.This is not evident in the 5-GHz data in Figure 22, but it an be seen in similar tests, shown inFigure 24, at 1.4 GHz where the re e tions are generally stronger. During the two hours of thiss an series the radio sour e, 2316+0405, went from an elevation of 47 degrees to 27 degrees. Themain feature that appears after some time is the 2.3-MHz ripple period from the subre e tor tomain re e tor multipath. The wave in the middle spe trum of Figure 24b may have been ausedby a drift in the ele troni s, but this remains to be tra ked down.Be ause of the vagaries of the weather and IF ele troni s baseline problems that tend to29

Figure 24: Ratio of 2316+0405 ontinuum sour e (on - o�) spe tra at intervals of 23, 47, 70, 93,and 116 minutes. Panel (a) is re eiver hannel X and (b) is hannel Y. The plots are assignedarbitrary o�sets to reate the time progression from top to bottom.dominate the spe tra using wide bandwidths we were unable to make a useful assessment of ontinuum alibration stability in the higher frequen y re eivers. This is something that needsto be revisited after improvements have been made to the IF ele troni s.3.5 Cal spe trum ripplesDuring most observations a alibration noise diode (CAL) inje ts noise into the system with aperiod on the order of a se ond. The goal is to provide a rough ux alibration and to removeele troni drift. The CAL signal does ontain frequen y stru ture, however, and urrently theintensity s ale is only known to an a ura y of about 10% with a sampling in frequen y of � 1%of the front-end bandwidth. When spe tra are generated using the CAL signal the frequen ystru ture from the CAL itself will be folded into the �nal spe trum. It is diÆ ult to de ouplefrequen y stru ture arising from the CAL and other omponents generating system noise.Examples of T al=Tsys ratios are shown in Figure 25 toward old sky at L (top left), S (topright), C (bottom left), and X-band (bottom right). Note that the bandwidth varies dependingon the frequen y. Ea h of these plots an be ompared to (on � off)=off spe tra shown inSe tion 3.1. In general the spe tral baseline stru tures are similar.Although the CAL was �ring during all of the observations dis ussed thus far, be ause boththe on and o� phases are stored separately, the CAL an e�e tively be removed by only usingthe CAL o� data when pro essing spe tra (e.g., produ ing (on � off)=off spe tra). Figure 26shows on � off spe tra toward 0137+3309 at C-band (5000 MHz) that have been produ edusing both on and o� CAL phases (panel a) and only the o� CAL phase (panel b). Plotted inboth ases are (on � off) and (on � off)=off spe tra. General inspe tion of these plotsindi ates that the baseline stru ture is not dominated by stru ture in the CAL as both �gures arevery similar. The stru ture in the (on � off) spe tra omes primarily from the gain frequen ydependen e (see Figure 27). This is typi ally why these data are divided by the off spe trum.Also plotted is the antenna temperature, Ta, in units of the CAL (e�e tively (on � off)=offtimes the o� system temperature evaluated per hannel). This spe trum is noisier sin e we aredividing by a noisy T al spe trum. 30

Figure 25: T al=Tsys ratios toward old sky for L, S, C, and X-band. The red or darker line isre eiver hannel X, and the green, lighter line is hannel Y. The L and S-band spe tra have beensmoothed by trun ating the auto orrelation fun tions to 2048 lags and Hanning onvolved.

Figure 26: Total power spe tra toward the ontinuum sour e 0137+3309 for hannel X at C-band(5000 MHz). Panel (a): (on � off), (on � off)=off , and the antenna temperature (Ta) inunits of the CAL. For the on and off data the CAL phases have been averaged. The Ta spe trumhas been o�set by �2:5 for larity. Panel (b): (on � off) and (on � off)=off where only theo� CAL phase has been used. 31

Figure 27: Total power spe tra toward old sky at C-band (5000 MHz). Plotted in red or thedarker line is the pass-band power. Plotted in green or the lighter line is the T al=Tsys ratio. Only hannel X is shown.4 IF system4.1 2.4-MHz ripple from opti al modulatorsThe GBT IF signal is routed from the front-end to the ba k-end via opti al �ber. There is atotal of 8 opti al driver modules (ODM's) lo ated at the front-end and an equivalent number ofopti al re eiver modules (ORM's) at the ba k-end.Measurements of �ber transmission stability early in GBT onstru tion showed that twist-ing or bending of the �ber hanges the polarization of the laser light (whi h is almost linearlypolarized) at the dete tor end. Sin e all of the photodete tors are polarization sensitive (the"gain" depends on the light polarization). Hen e, the polarization rotation gets onverted to aamplitude modulation so the e�e tive gain of the opti al link is unstable. A servo system forstabilizing gain of the �ber link was developed whi h required the use of externally modulatedlasers whose intensity an be ontrolled through feedba k of the dete ted opti al power at there eive end of the �ber.Unfortunately, a 2.4-MHz period gain ripple is produ ed by these external modulators. Fig-ure 28 plots the total power and the auto orrelation fun tion for observations using the externalmodulator lo ated in ODM 6. The IF noise sour e lo ated in the IF ra k at the GBT Re eiverRoom on the feed arm was used. The 2.4-MHz ripple is lear in both plots. The spike nearlag 667 orresponds to a ripple period of 2.3988 MHz. The spikes near lags 1333 and 1999 areharmoni s and indi ate that the ripple is not purely sinusoidal.An approximate quantitative measure of the amplitude of the gain ripple relative to the totalnoise spe tral density was omputed by summing the three highest absolute values of the ACFnear ea h ACF ripple spike and taking its ratio to the zero-lag ACF value. Figure 29 plots theresults using only the fundamental. ODM's 1, 2, 3, 5, and 6 orrespond to di�erent externalmodulators. ODM's 4 and 7 were out of servi e. ODM 8 used a dire t modulator whi h isknown to not have any gain ripple; therefore it provides a measure of the dete tion limit of this32

Figure 28: IF noise sour e spe tra of an original external modulator. Panel (a): total powerpass-band averaging over 60, 30-se ond re ords. Panel (b): the auto orrelation fun tion.algorithm. Three di�erent IF frequen ies were explored: 1500, 3000, and 6000 MHz. The spreadin ripple amplitude for the �ve external modulators is about 5 dB, and the amplitude is about 8dB higher at an IF of 1500 MHz than at 6000 MHz. The �rst and se ond harmoni s are about6 and 10 dB weaker than the fundamental.The manufa turer has sin e modi�ed the external modulators and laims to have redu ed theamplitude of the gain ripple. Figure 30 summarizes the results of similar tests using the IF noisesour e for the new unit. The 2.4-MHz ripple is now not visible in the total power spe trum. TheACF plot does reveal the fundamental and the �rst two harmoni s, but the amplitude of the gainripple is down by � 17 dB. Figure 31 shows a omparison between the original and new externalmodulators.Although the 2.4-MHz gain ripple has been signi� antly redu ed it may appear above the noiselevel for very sensitive spe tral line experiments. The ripple frequen y of 2.4 MHz orrespondsto 24� km s�1, where � is the observing wavelength in entimeters. Unfortunately, this is similarto the line widths of many astronomi al sour e transitions. (For example, at K-band (1.5 m)this orresponds to 36 km s�1). The magnitude of this problem will depend on the amplitude ofthe gain ripple in raw spe tra and the stability of the ripple in time. If the gain ripple is verystable then it should an el when pro essing the data.Tests were performed to quantitatively measure the stability of the opti al �ber ripple. Threedi�erent ODM's were used onsisting of an original external modulator, a new modi�ed externalmodulator, and a dire t modulator. The IF noise sour e was used to simulate 5 minute on � offtotal power pairs by al ulating (on � off)=off for onse utive groups of data re ords. A totalof 400, 30-se ond re ords (200 minutes) were generated with the IF noise sour e. The data weredivided into four groups of 100 re ords (50 minutes), ea h onsisting of �ve on � off pairs.Figures 32, 33, 34 summarize the results for the original external modulator, the new modi�edexternal modulator, and the dire t modulator, respe tively. For ea h plot the (on � off)=offspe tra are shown along with the periodogram. Noti e that a � 60-MHz ripple is quite strongin some spe tra and is related to other se tions of the IF system (see Se tion 4.2). The 2.4-MHzripple is only dete ted for the original external modulator. In fa t, additional averaging of thedata de reases the amplitude of the ripple for this unit. Based on these results, eight modi�edexternal modulators have been ordered.33

Figure 29: Approximate amplitude of fundamental periodi ity omponent of the 2.4-MHz gainripple measured at IF enter frequen ies 1.5, 3.0, and 6.0 GHz. The amplitude is relative to thezero-lag ACF value. The red, green, dark-blue, blue-green, violet, and yellow lines orrespond to�bers 1, 2, 3, 5, 6, and 8, respe tively. ODM 8 uses a dire tly modulated laser, whi h is knownto not have any gain ripple.

Figure 30: IF noise sour e spe tra of the new external modulator. Panel (a): total power pass-band averaging over 60, 30-se ond re ords. Panel (b): the auto orrelation fun tion. Comparewith Figure 2834

Figure 31: Auto orrelation fun tion of the IF noise sour e. The red or darker line is the result forthe original external modulator, while the green, or lighter line is the result for the new externalmodulator. A total of 60, 30-se ond re ords have been averaged.

Figure 32: Syntheti spe tra using the IF noise sour e and an original externally modulated laser.Panel (a): (on - o�) / o� spe tra using 5-minute s an lengths. Ea h spe trum onsists of 5 pairs(50 minutes). Panel (b): Periodogram for ea h spe trum.35

Figure 33: Syntheti spe tra using the IF noise sour e and a new externally modulated laser.Panel (a): (on - o�) / o� spe tra using 5-minute s an lengths. Ea h spe trum onsists of 5 pairs(50 minutes). Panel (b): Periodogram for ea h spe trum.

Figure 34: Syntheti spe tra using the IF noise sour e and a dire tly modulated laser. Panel(a): (on - o�) / o� spe tra using 5-minute s an lengths. Ea h spe trum onsists of 5 pairs (50minutes). Panel (b): Periodogram for ea h spe trum.36

Figure 35: Di�eren e spe trum, (spe trum2 - spe trum1) / spe trum1, of two 30-se ond integra-tions on the IF noise sour e taken 5 minutes apart (a) and its periodogram (b).4.2 Baseline Periodi ities Produ ed in the IF Ele troni sIn the pro ess of disentangling the various baseline distortion e�e ts we frequently saw a ripplewith a period in the range of 60 to 80 MHz when using the 800 MHz spe trometer bandwidth.The amplitude of this ripple varied irregularly on time s ales of minutes, and it was eventually orrelated with small temperature hanges in the GBT ontrol room where mu h of the IFele troni s is housed. As of this writing the major ause of the problem has been tra ed toele troni omponents in one or two of the IF onversion and �lter modules, but it appears thatthere is more than one temperature-sensitive omponent in the IF hain ontributing to baselineinstability. Here is what we know thus far.Figure 35a shows an example of an 800-MHz spe trum when the IF baseline instability isquite prominent. The noise power in this spe trum is from the IF noise sour e substituted for theGBT re eiver's IF output in the GBT Re eiver Room. The frequen y s ale in this �gure refers tothe IF passband at the input to the spe trometer sampler. The periodogram of this spe trum isshown in Figure 35b. The most notable feature in the periodogram is the spike at 0.016MHz�1(62-MHz ripple period), but there are other signi� ant periodi ities in this spe trum. The largestspike at 0.005 MHz�1 (200-MHz ripple period) is hard to interpret sin e it is a substantialfra tion of the total spe trum width.The orrelation of the amplitude of ripples, su h as those in Figure 35, with GBT equipmentroom temperature an be illustrated by plotting the strength of the dominant spike in a spe trum'speriodogram as a fun tion of time along with a plot of the measured temperatures in the room andin the ra k ontaining many of the IF ele troni s modules. This is shown in Figure 36. There is a lear orrelation of 62-MHz ripple amplitude with the temperature of the ra k ontaining some ofthe IF ele troni s used in this measurement. However, the orrelation between temperature andripple amplitude for other periods seen in the periodograms and between amplitudes of di�erentripple periods is not always obvious. This indi ates that there is more than one ause of thebaseline instabilities in the IF system.Figure 37 shows a summary of the various ripple periods found in the periodograms of(re ordN � re ord1) = re ord1 spe tra over roughly a �ve-hour period, where re ordN isthe Nth 30-se ond re ord in the run. The open boxes in this �gure refer to periodogram fea-tures that o urred o asionally or weakly, and the �lled boxes are for strong features that werepresent mu h of the time. Most of the IF hannels showed a dominant ripple period around 0.016MHz�1 (63 MHz), but there was signi� ant variation in this period from hannel to hannel,and these periodogram features tended to be omposed of several losely spa ed periods whose37

Figure 36: Correlation of the 62-MHz ripple period amplitude in IF hannel A1 (top urve) with onverter ra k A internal air temperature (middle) and GBT equipment room air temperature(bottom). The temperature values have been multiplied by 0.2 and given an arbitrary zero o�setto allow easier visual omparison with the ripple amplitude urve.amplitudes were not ne essarily well orrelated in time. Channels A2 and C2 were onsiderablymore stable than the others. Most hannels had quite a variety of periodi ities that ame andwent with time with only partial orrelation. This, too, indi ates that there is more than one ause of baseline ripple in ea h IF hannel.The three rows at the bottom of Figure 37 are the equivalent periodogram distributions forthe spe tra obtained by di�eren ing spe tra taken before and after intentional hanges to IF ablelengths at three di�erent pla es in the IF system. Figure 38 is a s hemati diagram of the GBT IFsystem between the front-end and the spe trometer. There are four ables with lengths between2.4 and 7.1 meters between the di�erent IF modules. As will be explained in the next se tion,if there are mismat hes at the ends of these ables that ause part of the signal to traverse thelengths of the ables twi e again, the multipath interferen e will ause frequen y-dependent gainripples. Very small hanges in the ele tri al lengths of these ables will ause the gain ripplesto hange in phase, whi h an manifest themselves as temperature-dependent gain ripples in thesystem noise spe trum.The gain ripple periods orresponding to the 2.4-, 6.5-, and 7.1-meter able lengths are shownat the bottom of Figure 37 as measured with the able length hange experiment. There may bea strong e�e t asso iated with the 2.4-meter able orresponding to a ripple near 0.016 MHz�1(63 MHz) as seen by the period orrelation in Figure 37, but there are no periods evident inthe IF noise periodograms orresponding to the other two able lengths, even though one mightexpe t a omparable or larger temperature sensitivity of the longer ables. As of this writing itis not lear how large a role the temperature dependen e of the ele tri al lengths of these ablesplays in the observed IF baseline instability. It is likely that this e�e t is a signi� ant problemso the pro urement of more stable oax is under way.38

Figure 37: Ripple periods found in the periodograms of eight of the sixteen GBT IF hannels forthe �ve-hour run used in Figure 36. The rows are for di�erent IF hannels labeled by spe trometeroutput hannel. The bottom three rows show equivalent results for the FFT's of spe tra obtainedby hanging IF able lengths as reported below. Rows \63", \23", and \21" refer to the 63-, 23-,and 21-MHz ripple periods that are hara teristi of the able lengths given in Figure 38.

ReceiverOptical Convertor

Module

AnalogFilter

Module2.4 m 7.1 m

SpectrometerSampler

ReceiverModulator

Optical1.5, 3, or 6 GHz

1.08, 1.5, 3or 6 GHz 0.8−1.6 GHz 0.8−1.6 GHz

6.5 m

Optical Fiber

~3.5 m

Figure 38: S hemati diagram of one GBT IF hannel. The �gures in rounded re tangles are able ele tri al length in meters, and above ea h is the frequen y range arried by this able whenusing the 800 MHz spe trometer bandwidth.39

4.3 Changing Cable LengthsWe performed tests on the GBT Equipment Room IF ele troni s in an attempt to isolate thesour e of near sinusoidal baseline ripples that show up from time-to-time. We inje ted a stablenoise sour e into Converter Modules or into Sampler/Filter Modules, and simulated positionswit hed observations by running a series of s ans using the spe trometer's 200-MHz bandwidth.During the ourse of these tests, we indu ed small hanges in the length of oaxial ables onne t-ing various pie es of equipment by inserting short oaxial adapters. This introdu ed dramati ripples in the baselines (Figure 39), al ulated by the usual methods.In order to understand this ripple-produ ing me hanism, it is helpful to revisit some two-portnetwork theory. Figure 40 shows a signal- ow representation of two two-port networks, Net Aand Net B, onne ted by a lossless, mat hed transmission line. The s attering parameters s11 ands22 are input and output re e tion oeÆ ients of the networks. Parameters s21 and s12 representthe forward and reverse transmission oeÆ ients. The s-parameters are omplex values that, ingeneral, vary with frequen y. The delay of the lossless, mat hed transmission line is representedby a transmission oeÆ ient in phasor notation: ej�l, where � = 2�=� and l is the ele tri allength of the transmission line. In Figure 40, omplex voltage signals owing toward the rightare represented by ai, and signals owing toward the left by bi.We are interested in a4=a1. The square of the magnitude of this quantity is the transdu erpower gain of the as aded networks. The signal- ow diagram allows us to write equations forthe signals leaving and re e ted from ea h port, and after some algebrai manipulation, we arriveat: a3 = a1s21A(1� s11A)ej�l + a3s11Bs22Aej2�l (11)and, a4a1 = s21As21B(1� s11A)(1� s11B)ej�l1� s22As11Bej2�l (12)Equation 11 shows that a3, the input signal to Net B, onsists of the input signal a1 modi�ed byNet A and delayed by the transmission line, plus a3 (from an earlier time) re e ted by s11B and bys22A and delayed by two passes through the transmission line. It may not be immediately obviousfrom equation 12, but this dual re e tion sets up an interferen e pattern that introdu es gainripple in the as aded network frequen y response, a well-known result. Numeri al evaluationsof equation 12 show that:� The transmission line embedded between non-zero re e tion oeÆ ients introdu e passbandripple with hara teristi frequen y of 2l , where is the velo ity of propagation in the line.The ripple amplitude is set by the produ t of s22A and s11B.� If the line length hanges, the passband ripple shifts in frequen y. If this happens, say,between (or during) two position-swit hed s ans, the frequen y shift in ripple pattern in-trodu es ripple in the baseline obtained by the ratio of the two s ans. The amplitude ofthe baseline ripple is proportional to the length hange (for small hanges), as a fra tionof the wavelength of the signal on the line. For example, simulations show that in orderto keep baseline ripple below 0.01% when the transmission line is operating between 20dBreturn losses, the line ele tri al length must be stable to better than 0.144Æ (�=2500) at thehighest passband frequen y.These results may be used to set spe i� ations on the required phase stability of inter on-ne ting ables, based on the return loss seen at ea h end and on the frequen ies present on the40

able. Changes in ele tri al length of ables may be indu ed by temperature hanges, by exing,and by poor onne tions. We an see that the GBT system design has made these requirementsquite hallenging be ause of the relatively high frequen ies and broad bandwidths used betweenvarious subsystems. The following se tion dis usses the types of ables used for long signal runson the GBT.4.4 GBT CablesThree types of oaxial ables are mainly used for IF and RF signals in the GBT re eiving systems.141 Semi-rigid The solid te on diele tri in these ables exhibits a relatively strong negativetemperature oeÆ ient, giving able assemblies a delay oeÆ ient of about � = 70ppm=Cnear room temperature. Cables made from this material are generally less than 1 meterlong. To a hieve �=2500 stability at 8GHz and 1 meter length, the oeÆ ient implies atemperature stability of just over 0:1ÆC. The equation for �C is:�C � �l�l (13)where �l is the maximum hange in able length that will a hieve the required baselinestability.Heliax FSJ1-50A The polyethylene foam diele tri in this able has a temperature oeÆ- ient mu h smaller than te on. Cable assembly temperature oeÆ ients are typi ally7 � 12ppm=C. This type is used in the GBT Re eiver Room for IF ables between thefront-ends and the IF Ra k, and lengths an be 4 meters. To a hieve �=2500 stability at8GHz on a 4 meter length of FSJ1-50A, temperature stability of just over 0:3ÆC is required.Belden 1673A Conformable Coax This able uses solid te on diele tri , and we measuretemperature oeÆ ients of 50� 70ppm=C. This type is used for jumpers in the EquipmentRoom. For example, IF jumpers between the Analog Filter Ra k and the Spe trometerSampler Ra k are of the 1673A type. These ables are about 6 meters long (ele tri al) and arry up to 1.6GHz signals. To a hieve �=2500 stability at 1.6GHz on a 6 meter length of1673A, temperature stability of 0:09ÆC is required.We have measured a few other able types, and found that one must be autious about a ept-ing published manufa turer's spe i� ations. For example, type LMR ables by Times Mi rowaveare spe i�ed to a hieve \less than 10ppm=C", but our measurements on several samples obtained oeÆ ients of 20� 25ppm=C. Tests of a relatively new type, F057A Heliax by Andrews, yieldedpromising results of �4 to +5ppm=C near room temperature for several assemblies. This type isbeing adopted for use at several lo ations in the system.4.5 IF Converter Ra k RippleIn light of the tight requirements for able phase stability dis ussed above, there was expe tationthat the near 60MHz ripple often seen in the IF system was related to able stability. In parti ular,it was found that the ables between the Opti al Re eivers and the Converter Modules exhibited hara teristi ripple frequen y near 60MHz. However, extensive testing has shown the situationto be mu h more subtle and omplex. The following summarizes the �ndings to date:� The dominant ripple is tightly orrelated with the Equipment Room HVAC plenum tem-perature. By inje ting noise sour es at various points in the system, the instability hasbeen isolated to the Converter Ra ks. 41

Figure 39: Data taken with the GBT Spe trometer. Baselines are shown for two pairs of s ans.Between the s ans used to generate the green tra e, about 4 m was added to the able betweenthe Sampler Filter module in the Analog Filter ra k and the input to the Spe trometer samplerra k.

42

Figure 40: S hemati representation of two networks onne ted by a lossless, mat hed transmis-sion line.43

� The major ripple period varies by Converter Module between about 55MHz and 65MHz,but is quite stable with time. The ripple period does not exa tly mat h that due to thelength of the Opti al Re eiver to Converter Module able length.� The ripple is somehow tied to the 4-way power splitters within the Opti al Re eivers. Wehave found that if three of the four OR outputs are terminated, there is no eviden e of theripple for quite long periods. See Figure 41. The 4-way splitters have output-to-outputisolation of about 20dB, so another fa t is not surprising: a hange in the able length onne ted to any one of the four outputs, a�e ts gain ripple in all four hannels.A feature of the Converter Module design that may also be a ontributing fa tor has beenidenti�ed. During lab testing, we found that a hange in the LO2 able length, onne ting aConverter Module to the LO2 power splitter, perturbs the module gain response, and introdu esa ripple when the gain responses before and after the able hange are ratioed. The ripple period,whi h in the system is near 60MHz, depends on the length of the LO2 able. Figure 42 showsthe input se tions of the Converter Modules, and the asso iated LO2 onne tions, whi h willhelp explain the e�e t. MX1 is a triple-balan ed broadband mixer whi h is used to up- onvertthe IF input signal. Note in this up- onversion appli ation the module input signal is onne tedto the mixer I port, and the R port is used as the output. The mixer type used has I-L portisolation of only about 19dB at 3GHz. The apparent explanation for the observed behavior isthat the module input signal leaks out the mixer L port, ows through the LO isolator (whi hhas less than 10dB loss in both dire tions at 3GHz), down the LO2 able, and is re e ted atthe LO2 power splitter with -7dB return loss. It then reverses the path, and interferes with theoriginal signal. The round trip loss through the path is approximately -55dB. The magnitude ofthe gain ripple introdu ed by the small return signal an be al ulated by noting that it doesa voltage ve tor addition to the main signal, and the relative phase of the two ve tors rotateswith frequen y. Therefore, the peak-peak gain ripple expressed as a fra tion of the main signalis given by: Ripplep�p = 2 � (10RoundTripLoss20 ) (14)For -55dB round-trip loss this value be omes 0.4%, near the level seen in the system as theEquipment Room temperature varies.While we have been su essful in identifying several sensitivities in the Converter Ra kswhi h an indu e ripple, it is still not apparent why the system is so sensitive to temperature.Measurements have failed to identify any of the passive or a tive omponents that are parti ularlysensitive to temperature. Nevertheless, the sensitivities are not good, and be ause of the hannel-to- hannel oupling, it is important to redu e the sensitivities in the whole Converter Ra k inorder to evaluate the e�e t of any improvement. The plan for pro eeding at this point is toinstall phase stable ables in the Opti al Re eiver to Converter Module path, and to in rease theisolation to the LO2 ables from the signal path. A di�erent type of mixer is being sought, buthigh isolation and broad bandwidths are generally not mutually ompatible. The best way maybe to add a LO2 bu�er amp near the mixer LO port whi h should add 30dB or so isolation atthe signal frequen y. A reasonably pri ed MMIC ampli�er that should work has been identi�edand will be assembled for evaluation.44

Figure 41: Baselines for two sets of one-minute s ans, with the �rst of ea h set taken as thereferen e for the subsequent s ans of that set. The IF Ra k Noise Sour e and the OR1/CM3/SF3IF path were used. The upper panel shows results with opti al re eiver module 1 (OR1) onne tedto four Converter Modules as normal. The lower panel shows results immediately afterward whenthree of the four OR1 outputs were terminated with 50 ohm loads.45

Figure 42: A blo k diagram of the input portion of the 1-8GHz Converter Modules, and theasso iated LO2 abling.5 Spe trometers5.1 Spe tral pro essor wide-bandwidth distortionsInitial ommissioning observations of the C-band (4{6 GHz) re eiver using the spe tral pro essorba k-end showed a variation in measured Tsys=T al a ross the 40-MHz pass-bands whi h anappear to be a ripple in the Tsys=T al spe trum of a re eiver if spe tra are on atenated (seeFigure 43). The amplitude an be as large as 10% of the system temperature and the rippleappears in both polarizations. This ripple also exists in (on � off)=off spe tra. This probably olored some of the early baseline measurements made on several re eivers. Using both thespe tral pro essor and the auto orrelation spe trometer has revealed that this ripple resides inthe spe tral pro essor.The magnitude of the ripple is a fun tion of input power level whi h may indi ate that thise�e t is related to sampler quantization. The wavelength of the ripple mat hes the spe tralpro essor bandwidth and the amplitude appears to in rease with bandwidth. Figure 44 showsspe tra using the spe tral pro essor with bandwidths of 40 and 10 MHz. For ea h panel severalspe tra are plotted with di�erent input power levels. The narrower bandwidths produ e atterspe tra and input power levels around -6 dBm appear to be optimum. This phenomena wasprobably not dete ted earlier in astronomi al measurements sin e the spe tral pro essor hasprimarily been used with narrow bandwidths in spe tral line mode.5.2 Auto orrelator linearity testsSin e the GBT auto orrelation spe trometer is the key measurement instrument for this spe tralbaseline investigation one would like to have a measure of its stability and linearity. Thus far wehave no eviden e that spe trometer instability (sampler level drift, for example) is ontributing tothe baseline distortions, but this needs to be revisited after some of the other system instabilitieshave been improved.As a �rst-order he k on spe trometer linearity we measured the ratio of Tsr =Tsys for the ontinuum sour e 2052+3635 with sampler input levels that di�ered by 3 dB. The noise power into46

Figure 43: T al=Tsys ratio toward old sky a ross the C-band re eiver. Only hannel X is shown.The spe tral pro essor was used in 2� 1024 mode with a bandwidth of 40 MHz. Panel (a) plotsthe entire band, while panel (b) shows an expanded view.

Figure 44: T al=Tsys ratio toward old sky for the C-band re eiver. Only hannel X is shown.Panel (a): the spe tral pro essor was used in mode 2� 1024 mode with a bandwidth of 40 MHz.The three urves ompare di�erent spe tral pro essor attenuator settings. The labels in the plotare the measured power levels. Panel (b): the spe tral pro essor was used in 2� 1024 mode witha bandwidth of 10 MHz. The two urves ompare di�erent spe tral pro essor attenuator settings.The labels in the plot are the measured power levels. The green or lighter line orresponding to�6 dBm has been o�set by 0.01 for larity.47

Figure 45: Panel (a): Tsr = Tsys for ontinuum sour e 2052+3635 using di�erent spe trometersampler input levels. Ea h spe trum is 5 minutes on, 5 minutes o� integrations. The red (dark)spe tra are for normal sampler input levels and the green (light) spe tra are for input levels 3dB below normal. The lowest two spe tra orrespond to IF/re eiver hannel 1, and the top twospe tra are for hannel 2. Panel (b): ratio of Tsr = Tsys spe tra measured with 3 dB lowersampler input levels to spe tra with normal sampler levels. The red (dark) urve is for hannel1, and the green (light) urve is for hannel 2.the sampler was hanged by adding 3 dB to the omputer- ontrolled attenuators in the onvertermodule shown in Figure 38. The results of this measurement are shown in Figure 45. Simply hanging the attenuator settings is enough to hange the impedan e mat h to the ampli�ersatta hed to it so one expe ts some hange in the noise spe trum presented to the spe trometersamplers. Nevertheless, the ripple and 2% level hange seen in Figure 45 are larger than one wouldlike. The level hange ould be due to an error in the sampler linearization done in pro essing theauto orrelation fun tion, or it ould indi ate that the ampli�ers following the hanged attenuatorare ontributing more noise than expe ted. The ripple suggests that the omponents surroundingthe attenuator are adding a bit of stru ture to the noise spe trum on frequen y s ales of 20 MHzand larger.5.3 O�set problem in assembling omposite re eiver Tsr /Tsys spe traComposite re eiver Tsr =Tsys spe tra have been produ ed to investigate spe tral baseline e�e tsover the entire re eiver bandwidth. The results of these tests are summarized in Figure 46 for L,S, C, and X-band. The spe trometer was used with bandwidths of 50 MHz (L-band), 200 MHz(S-band), and 800 MHz (C and X-band). Composite spe tra were produ ed by overlapping thebandpasses over the re eiver's frequen y span. The spe tra should be reasonably at sin e theyhave been orre ted for a measured sour e spe tral index (with the ex eption of the observationsof NGC7027 at C-band). There appear to be signi� ant variations a ross the band, although thisis ompli ated by the spe tral baseline stru ture.For all re eivers there is an o�set between adja ent bands as mu h as 10% of the systemnoise. Possible explanations in lude hanges in the system temperature, gain drift, and pointing.Be ause the sour e intensity is being plotted in units of the system temperature, any hanges inthe system temperature would result in an o�set between spe tra. Typi ally the spe tra weretaken onse utively; although hanges in the atmosphere ould signi� antly alter the systemtemperatures at C and X-band this is unlikely at L and S-band. It seems unlikely that the gainwould drift between su essive observations. During these observations the pointing was notalways arefully monitored. Again, while this might be signi� ant at C and X-band, where the48

Figure 46: Composite re eiver Tsr =Tsys spe tra. The red, darker lines are hannel X (LCP),while the green, lighter lines are hannel Y (RCP). Panel (a): omposite spe tra for L-band withbandwidths of 50 MHz. Panel (b): omposite spe tra for S-band with bandwidths of 200 MHz.Panel ( ): omposite spe tra for C-band with bandwidths of 800 MHz. Channel Y was was notusable. Panel (d): omposite spe tra for X-band with bandwidths of 800 MHz. Spe tra in panels(a), (b), and (d) have been orre ted for a sour e spe tral indi es of �1:006, �0:7, and �0:9,respe tively. For the spe tra in panel ( ) the thermal spe trum of NGC7027 peaks between 5000and 8000 MHz.beam-size an be as small as 80 ar se , it is unlikely to e�e t the observations at L and S-band.Further investigation is required.6 RFIBroadband RFI is another sour e of spe tral baseline distortions. The RFI noise sour e itselfmay have a rough spe trum, and the path-loss through whi h it enters the GBT feed will usuallybe quite frequen y dependent. Figure 47 shows two examples of RFI spe tra re orded near 1.4GHz in the ourse of observing HI in external galaxies. Figure 47a shows a burst that o ursirregularly about on e per hour and lasts only a few se onds. A few of these bursts have beenseen with the digital ontinuum re eiver using a dete tor at the output of the IF Ra k where thesignal enters the opti al �ber. Hen e, this is not a problem in the IF ele troni s or spe trometer.The short duration and re urren e of these bursts suggest something like a thermostat onta tar . Figure 47b shows a burst that lasted about two minutes. In the �ve, 30-se ond integration49

Figure 47: Broadband RFI spe tra. Panel (a): (re ord2 - re ord1) / re ord1 for two adja ent30-se ond re ords. This RFI burst lasted less than a few se onds. The red (dark) spe trumis for re eiver hannel X and green (light) is hannel Y. Panel (b): Five su essive hannel XTsys � (on � off)=off 30-se ond spe tra. The spe trum time sequen e is red, green, dark-blue,bla k, and violet.spe tra one an see a gradual in rease and de line in the RFI noise power and hange in thespe trum shape. The origins of these two RFI signals remain to be determined.7 SummaryThe spe tral baseline properties of the Green Bank Teles ope (GBT) have been explored througha series of experiments with the goal of providing a guide to engineering improvements and alibration te hniques. The individual reports an be found at:http://www.gb.nrao.edu/GBT/baseline/index.html.This memo summarizes the urrent status of the investigation of the GBT spe tral baselineproperties. The main results are dis ussed below.� Antenna: The main ulprit in produ ing poor spe tral baselines for traditional on-axissingle-dish teles opes is multi-path re e tions between the feed, sub-re e tor, and main re e torof the antenna. The o�-axis design of the GBT was primarily motivated to minimize these e�e ts,and, indeed, it does have signi� antly redu ed multi-path interferen e. Nevertheless, severalspe tral baseline ripples were dete ted with the GBT involving the antenna. Quasi-sinusoidalripples were observed toward old sky with periods of � 1:6 MHz and 9 MHz. The 1.6-MHz ripplewas spread over a range of periods from 1.3{1.8 MHz and is onsistent with noise in the teles opesystem entering the re eiver through two paths: dire tly and after emission or re e tion from thefeed area and then s attered from the ir umferential gaps between surfa e panels. The 9-MHzripple is onsistent with the distan e between the sub-re e tor and the Gregorian feed-hornsinvolving a single re e tion. Higher frequen y observations have resolved the 9-MHz ripple intoseveral omponents onsistent with multi-path re e tions from the sub-re e tor to the Re eiverRoom roof or one of the L, S, or C-band feeds. The amplitude of these ripples vary from 5{50mK rms at L-band to 2{8 mK rms at X-band. The 9-MHz ripple is roughly three times strongerthan the 1.6-MHz ripple. Careful total power, position swit hing an redu e these amplitudes bya fa tor of 30.The antenna response to a ontinuum radio sour e reveals a 2.3-MHz ripple period feature thatis onsistent with a re e tion between the sub-re e tor and the primary re e tor near the axis of50

the parent paraboloid. This phenomena is similar to the dominant baseline ripple of traditionalon-axis teles opes but is redu ed in amplitude for the GBT sin e the feed under-illuminates thisregion of the primary surfa e, and the enter of the spe ular re e tion point is o� the edge of there e tors. The ripple amplitude is a fun tion of sour e ontinuum intensity and in reases as �2.� Front-end Re eiver: When observing strong ontinuum sour es quasi-periodi ripples aredete ted that are found to reside upstream of the �rst LO mixer and downstream of the opti s;that is, in the front-end re eiver omponents. The properties of these ripples are di�erent forea h re eiver, although they appear relatively stable. Ripple periods range from 30{100 MHzwith amplitudes as high as 5% of the system noise. Re e tions within the feed and waveguidesystem of the GBT re eivers, due to small impedan e mismat hes, produ e ripples in the sameway as multi-path interferen e in the antenna system. Models predi t that at least part of theobserved spe tral baseline stru ture an be explained by these impedan e mismat hes.The baseline ripple observed at C-band was more sinusoidal than for other re eivers andhad a period of 65 MHz. Experiments revealed that water on the C-band feed radome was themain sour e of these ripples. This was more prominent at C-band be ause of the radome design;nonetheless this e�e t will add a variable omponent to the baseline stru ture for all re eivers.An e�e tive blower/heater system is required.In ontrast, at X-band the spe tral baselines exhibited irregular, narrow features that weremore hara teristi of resonan es than ripples from re e tions. Similar stru tures have beenobserved before in other re eivers. EM modeling along with experiments in the lab indi atedthat leakage via the thermal gap an produ e resonan es in the dewar. However, for X-bandonly one spike near 9.7 GHz was identi�ed with the thermal gap. Close inspe tion of the X-bandfeed-horn led to the dis overy of several problems: a 0.015 in h gap in the waveguide wall of onejoint; several metal hips in the feed-horn orrugations; a non- ondu tive �nish on the aluminumfeed-horn joint ange fa es; and an oily �lm between anges. Corre tions to these problemssigni� antly improved the X-band spe tral baselines.Sin e it is not pra ti al to eliminate all of the distortions to ontinuum sour e spe tra theremaining e�e ts must be alibrated. One te hnique is to use an astronomi al alibrator similarto the target sour e to measure the system response. Initial tests have been en ouraging butmore work is required.� IF System: The IF system is ommon to all re eivers and it is therefore riti al to eitherremove or hara terize any spe tral baseline problems from this omponent of the GBT. Duringthe early stages of the GBT ommissioning a 2.4-MHz gain ripple was found in the externalopti al driver modules. The amplitude varied by 5 dB between the di�erent units and was about8 dB higher at an IF frequen y of 1500 MHz than at 6000 MHz. A modi�ed unit from themanufa turer has redu ed the ripple amplitude by 17 dB. New units have been ordered.During the initial baseline tests we frequently dete ted a ripple with a period between 60{80MHz. The amplitude of this ripple varied irregularly with time-s ales of minutes, and it waseventually orrelated with small temperature hanges in the ra ks housing the IF ele troni s inthe GBT equipment room in the lab. Early investigations pointed to the GBT IF ables. Changesin able length with temperature an indu e ripples with periods similar to those observed. Testsof a relatively new able, F057A Heliax by Andrews, yielded promising results and is beingadopted for use at several lo ations in the system.However, installation of the new ables did not signi� antly improve the spe tral baselines.The dominant ripple has been isolated to the onverter ra ks with a period between 55{65 MHzthat does not exa tly mat h the ripple due to hanges in the opti al re eiver{ onverter module able. The ripple is somehow tied to the four-way splitters within the opti al re eivers. If three ofthe four opti al re eiver outputs are terminated there is no eviden e of the ripple for long periodsof time. A ontributing fa tor appears to be the LO2 able whereby the input signal leaks outof the mixer port down the LO2 able and is re e ted at the LO2 power splitter. Methods of51

providing higher isolation while retaining the required broad bandwidth are being explored.� Ba k-end Dete tors: Spe tra taken with both the spe tral pro essor and the spe trometerrevealed a 40-MHz ripple in the spe tral pro essor with an amplitude of 10% of the systemtemperature. The ripple was observed in Tsys=T al and (on � off)=off spe tra. The rippleamplitude is a fun tion of input power level and the ripple period mat hes the spe tral pro essorbandwidth. This may indi ate that the e�e t is related to the sampler quantization.Sin e the auto orrelation spe trometer is a key measurement instrument for many of thesespe tral baselines tests it is important to measure its stability and linearity. There is no eviden ethat spe trometer instability is ontributing to the baseline distortions. A 3 dB hange in samplerinput level does produ e a ripple and a 2% hange in level, however. The ripple suggests thatthe omponents surrounding the attenuator are adding a bit of stru ture to the noise spe trumon frequen y s ales of 20 MHz and larger. The level hange ould be due to an error in thesampler linearization or it ould indi ate that the ampli�ers following the hanged attenuatorare ontributing more noise than expe ted. The spe trometer stability and linearity need to berevisited after some of the other system baseline problems have been improved.8 Atta hmentsThe following memos are atta hed to this report:� \A Note on a Possible Explanation of GBT S-Band Re eiver Baseline Variability", MarianW. Pospieszalski, 28 January 2003.� \A First-Order Noise Analyis of the GBT L-Band Re eiver Front-End", Ri hard F. Bradley,30 January 2003.

52