Table Of Content

Mon.Not.R.Astron.Soc.000,000–000 (0000) Printed26January2009 (MNLATEXstylefilev2.2) An Interference Removal Technique for Radio Pulsar Searches 9 0 R. P. Eatough1⋆, E. F. Keane1, A. G. Lyne1 0 1 Jodrell Bank Centre for Astrophysics, Alan Turing Building, Universityof Manchester, Manchester, M13 9PL, United Kingdom. 2 n a 1May2008 J 6 2 ABSTRACT Searches for radio pulsars are becoming increasingly difficult because of a rise in im- ] pulsiveman-madeterrestrialradio-frequencyinterference.Herewepresentanewtech- A nique, zero-DM filtering, which can significantly reduce the effects of such signals in G pulsar search data. The technique has already been applied to a small portion of the . data fromthe Parkesmulti-beampulsar survey,resulting inthe discoveryoffour new h pulsars, so illustrating its efficacy. p - Key words: methods: data analysis - pulsars: general - stars: neutron o r t s a [ 1 INTRODUCTION 2 ZERO-DM FILTERING 1 v Pulsarsearchdataareoftencorruptedbythepresenceofim- A fundamental characteristic of broadband pulsar signals 3 pulsive, broadband and sometimes periodic terrestrial radi- is the frequency-dependent dispersion of their pulse arrival 9 ation. Such radio frequency interference (RFI) originates in timesas a result of traversing theionised component of the 9 unshielded electrical equipment which produces discharges, interstellar medium.Pulses observed at lower frequency ar- 3 such as automobile ignition systems, electric motors, and rivelaterthantheirhigher-frequencycounterparts.Thede- 1. fluorescent lighting, as well as discharge from high-voltage lay,t, at radio frequency f (MHz) is given by: 0 power transmission lines. It may also arise naturally from DM 9 the radio emission generated in lightning discharges. Such t=4150 sec, (1) „ f2 « 0 RFI can often be very strong and can even enter receiver : systemsthroughthefar-outsidelobesofthetelescoperecep- whereDMisthedispersion measure, theintegratedcolumn v tionpattern.Severalmethodshavebeenemployedtoreduce densityoffreeelectrons along thelineofsight,measured in i X theeffectsof RFI,suchasclippingintensespikes,removing cm−3pc(e.g.Lyne&Smith2006).Detectingapulsewitha r parts of the fluctuation spectrum, and identifying common finitebandwidthreceiverresultsinabroadenedpulseprofile a signals in the different beams of a multi-beam receiver sys- andacorrespondingreductioninpulsesignal-to-noiseratio. tem.Theseproceduresallrequirecarefullytunedalgorithms Before any search for pulses can be performed, the data toremovetheinterference,andatthesametimemustcause have to be compensated for the effects of a number of trial minimal damage to theastronomical data. values of DM. In order to do this, the bandwidth of the RFIsignals aremostly broadbandbutgenerally donot receiverisfirstsplitintoanumberofindependentfrequency display the dispersed signature of radio pulsars. Here, we channels,whichareappropriatelysampledtoproduceatwo- presentasimplealgorithm whichwecall‘zero-DMfiltering’ dimensional array of samples, S(f,t) at frequency f and i j i to selectively remove broadband undispersed signals from time t. Appropriate time delays for a given trial DM are j data prior to the application of normal pulsar search algo- thenappliedtoeachfrequencychannelsothatanypulsesat rithms.Theoutlineofthispaperisasfollows: following the thisDMarealignedintime.Thefrequencychannelsarethen description of the algorithm in section 2, in section 3 we summed together to produce a time series of dedispersed show that the procedure does indeed remove much of the data, x(t). interference in typical observing situations and in general The zero-DM filter is implemented prior to the dedis- greatly improves the visibility of celestial dispersed pulses. persion described above, by simply calculating the mean of In section 4 we present four new pulsars discovered in the all frequency channels in each time sample and subtracting Parkesmulti-beampulsarsurvey(PMPS)afterapplyingthe thisfromeachindividualfrequencychannelinthetimesam- new technique. Finally section 5 is a discussion of the ben- ple. Hence the adjusted values of the samples, S′(f,t) are i j efits and limitations of theprocedure. calculated as: ′ 1 nchans S(f,t)=S(f,t) S(f,t). (2) ⋆ E-mail:[email protected] i j i j − nchans Xi=1 i j f a) B f 0 f t b) B f 0 t Figure 2.Thepulsedistortioncaused byzero-DMfilteronide- Bdt/df alised box-car pulses of varying width for the PMPS observing w(t) system.ThetimescaleshouldbemultipliedbytheDMincm−3pc c) togivethetimeinmilliseconds. t Figure1.Theeffectofthethezero-DMfilteronanidealisednar- row linearly-dispersed pulse in the frequency-time domain. The toppanel(a)showsthedispersiondriftofthepulseBdt/df over thefullbandwidth Bofthereceiver. Subtractionof themean in verticalstripstoremovezero-DMsignals(Equation2)leavesthe non-pulse area (shaded grey) negative. Panel (b) shows the re- sultofdedispersingthedata atthecorrectvalueofDMandthe lower panel (c) shows the resultant pulse shape after adding all thefrequencychannels together. Clearly,anybroadbandundispersedsignalwillresultin asimultaneousriseacrossallfrequencychannelsandwillbe Figure 3. Two example effective zero-DM filters for removedbythisprocess.Wenowinvestigatetheeffectofthis DM=500 cm−3pc and DM=100 cm−3pc for the PMPS, procedureondispersedcosmicpulses.Consideranidealised showing variation in relative spectral amplitude with the dispersedpulsarsignalwithverynarrowpulsesandapprox- fluctuationfrequency. imatethedispersiondriftacrossthefullreceiverbandwidth Btoalinearslope,dt/df,whichisgivenapproximatelyby: In the frequency domain, equivalently, the fluctuation dt DM = 8300 sec/MHz, (3) spectrumofthededispersedtimesseriesismultipliedbythe df − „f 3 « 0 Fouriertransform, W(ν),ofw(t) which canbeshown tobe wheref (MHz)isthecentralobservingfrequency.Figure1a of theform: 0 showssuchapulsesweepingthroughthefrequencybandofa dt W(ν)=1 sinc2(πB ν) receiver,traversingthefullbandwidthinatimeofBdt/df. − df Aftersubtractionofthemeaninverticalstripsasdescribed (4) 8300πBDM by Equation 2, the non-pulse grey area in the Figure 1a =1 sinc2 ν , − „ f3 « is negative. Dedispersion then distorts this pattern in the 0 manner shown in Figure 1b, so that adding vertically after where ν is the fluctuation frequency of the signal. Figure 3 the dedispersion then gives the convolving function, w(t) shows the effect of the zero-DM filter on the spectral am- showninFigure1c.Notethenegativetriangularareawhich plitudeofapulseattwospecifiedDM’sforPMPS-styleob- is adjacent to, and equals the area of, the pulse. Thus the servations. The zero-DM process effectively acts as a high- zero-DM process results in the dedispersed time series x(t) passfilteronthedispersedpulse.The‘low-frequencycutoff’, being convolved by this function, w(t). Of course any real ν canbeshownusingEquation4tolieatafluctuation cutoff celestial pulse will have a finite width. In Figure 2 we show frequency 500/DMHz.ThusforDM=100cm−3pc,allfre- ∼ the effects of the zero-DM filter on idealised box-car pulses quencies<5Hzwillbelost.Thereductioninamplitudeand of different widths for the PMPS observing system at 1.4 pulse distortion we see in Figure 2 can be explained by the GHz (Manchester et al. 2001). In this case, f =1394 MHz loss of thelow frequency components of the pulseprofile. 0 and B=288 MHz. The pulse width in milliseconds can be For a typical pulsar we measure a regular train of nar- calculatedbysimplymultiplyingbytheDMincm−3pc.For row pulses, the spectrum of which is a ‘picket fence.’ The example, at a DM of 100 cm−3pc the bottom pulse would zero-DMfilterwillcutoffallspectralfeaturesbelowν . cutoff haveawidthof46ms.Thepeakamplitudeclearlydecreases However,normalpulsarswithpulsewidthsafewpercentof for pulses with larger widths. their pulse period will usually have most of their power in spectral harmonics at higher frequencies and these will not beaffectedsignificantly.Sincestandardpulsarsearchesper- formharmonicsumming(seee.g.Lyne&Smith2006or,for a detailed description of pulsar search algorithms, Lorimer & Kramer 2005), the detrimental effect of the zero-DM fil- teronthefinalspectralamplitudeisreduced.Assumingthe pulsar has n P/2W harmonics of equal amplitude where ∼ P is the pulsar period and W is the pulse width, the num- berofharmonicslostbelowthelow-frequencycutoffisgiven byn Pν ,theharmonic summing procedurewill cutoff cutoff ∼ increase the spectral signal-to-noise ratio by a factor of ap- proximately (n n )1/2. cutoff − The zero-DM process causes apredictable reduction in the amplitudes of low frequencies of the fluctuation spectrum of a dispersed pulsar. It will also reduce the amplitude of any broadband undispersed impulsive signals sub- stantially,butwillhavelittleeffectuponthermalnoise.The signal-to-noiseratioofthepulsarcanthereforebeoptimised by applying an optimum matched filter to the spectrum, Figure4.Thetoppanel(a)showsagreyplotofsimulateddata of a 130-ms burst of broadband RFI with DM=0 cm−3pc fol- which reduces the spectral amplitude where the pulsar sig- lowedbya20-msdispersedpulsewithDM=150cm−3pcacrossa nal is known to be small because of the zero-DM filtering. 288MHz bandpass centered at 1374 MHz. The middle panel (b) Thus,thereisasecondsteptothefilteringprocesswhereby showsthesamedataafterapplicationofthezero-DMfilter.Note the frequency response of the zero-DM filter is re-applied the negative non-pulse area. In panel (c), the data have been aftereachdedispersiontrial.Inthetimedomainthisisbest dedispersed for a DM=150 cm−3pc and summed in frequency illustrated by considering a single-pulse search (McLaugh- givingthepulseprofile. lin et al. 2006). Standard single-pulse searches involve convolution of the dedispersed time series x(t) with box-cars b(t) of various widthsand searching for peaks in x(t) b(t), ∗ where denotes convolution. In our case, the zero-DM fil- search examples re-application of the zero-DM filter after ∗ ′ tered time series is x(t) = x(t) w(t), so that a standard each dispersion trial hasnotbeen performed becauseof the ∗ ′ single-pulsesearchlooksforpeaksinx(t) b(t).Asaresult additional computational overheads this would impose in ∗ of zero-DM filtering the optimal filter is now not a box-car oursearchalgorithms (seeSection5forafurtherdiscussion ′ but rather is given by b(t) = b(t) w(t). We now search on this subject). for peaksin x′′(t)=x′(t) b′(t)= ∗−1[X′(ν) W(ν)] b(t) Figure 5 illustrates clearly the effects of the proce- ∗ F ∗ ∗ where denotes the Fourier transformation and we have dure on a 35-minute PMPS observation in the direction of F used the commutability of convolution so the convolution PSR J1842+0257, which has a period of 3.1 seconds and canbeperformedinthefrequencydomain,whichiswhatis a DM of 148 cm−3pc, so that the zero-DM low-frequency donepractice. cutoff is at about 3.3 Hz. On the left is the normal search output,whileontherightisthatwiththezero-DMfiltering procedure in place. After application of the zero-DM filter 3 EXAMPLES therednoiseandimpulsiveinterferenceisalmostcompletely removed,andtheexpectedpulsedistortionduetothefilter- The effects of the zero-DM filter have been investigated ingasshowninFigure1candFigure4cisclear.Remarkably, using simulated PMPS data generated with the Sigproc- even with such a long-period pulsar, in which the funda- 4.21 pulsar analysis software suite. Figure 4a shows sim- mental and first few harmonics are completely missing, the ulated data of a 130-ms burst of broadband RFI with absence of the interference makes its detection more secure DM=0 cm−3pc followed by a 20-ms dispersed pulse with thanpreviously. AnumberofperiodicitysearchesonPMPS DM=150 cm−3pc across a 288MHz bandpass centered at beamscontainingknownpulsarswitharangeofperiodsand 1374MHz.Figure4bshowsthesamedataafterapplication dispersion measures has been performed. The fluctuation of the zero-DM filter. The broadband RFI signal has been spectra from the periodicity search are much cleaner than completelyremovedwithlittleeffectonthedispersedpulse. before (see Figure 6). The‘forest’ of RFI signals at low fre- Dedispersion at a DM=150 cm−3pc and summation in frequency are completely removed by the new technique. The quency produces the pulse profile shown in Figure 4c. The pulsarspectralsignal-to-noiseratiosbehaveasexpected,but expected pulse distortion is clearly visible. becauseofthereductionin thenumberofperiodic RFIsig- We also present examples of the zero-DM filter as ap- nals, the pulsars are easier to detect. plied to observational data from the PMPS in both peri- Figure7showsexamplesofsingle-pulsesearchdiagnos- odicity and single-pulse searches. All clipping and zapping ticplotsfromouranalysisofthePMPS.PlottedinFigure7 algorithms used previously formitigating theeffectsof RFI are the time series for each of the 325 trial DM’s (in the (Hobbs et al. 2004) have been omitted. In our periodicity range0 2200cm−3pc).Eachdetectedsingle-pulseeventis − plottedasacirclewithradiusproportionaltosignal-to-noise ratio.Thetoppanelistheresultofastandardanalysiswith- 1 http://sigproc.sourceforge.net outzero-DMfiltering.Theobservationcontainssinglepulses 1000 11000000 800 880000 time (s) 460000 time (s)time (s) 464600000000 200 220000 0 0.2 0.4 0.6 0.8 1 00 00..22 00..44 00..66 00..88 11 30000 20000 flux (arbitrary units) 1122 505050000000000 000000 flux (arbitrary units) 11- 550500000000 00000 -5000 0 0.2 0.4 0.6 0.8 1 -10000 0 0.2 0.4 0.6 0.8 1 pulse phase pulse phase Figure 5. The effects of the zero-DM filter on folding PMPS data from the 3.1-second pulsar, PSR J1842+0257, which has a DMof148cm−3pc.Theleft-handpanelsaretheresultofusinga regularanalysis,whiletheright-handpanelsareafterapplication of the zero-DM filter. On each side, at the top are grey plots of 63 temporal sub-integrations folded at the pulse period, cover- ingthefull35-minutesurveyobservation,andatthebottomisa panel showing the full integrated profile. The effect of the filter on the shape of the pulse profile can be clearly seen. The asym- metry of the depressions either side of the pulse arise because the dispersion sweep across the bandpass is not perfectly linear buthasaquadraticcomponent,althoughinmostcasesthesweep is well approximated by a linear slope. Note that the impulsive RFI present through the majority of the observation is almost completelyremovedandthebaselinewobbleinthefinalprofileis muchflatter inthefilteredcase. Figure7.Single-pulsesearchdiagnosticplotsofa35-minuteob- servation of the erratic pulsar PSR B1735−32. Top: using stan- dardsearchmethodwithoutzero-DMfilter;Broadbandterrestrial interference causes the large stripes across a wide range of DM. Middle: After using zero-DM filter, most RFI streaks have been removedwithsinglepulsesfromthepulsarnowmuchmorevisible (at DM channel≈82). Bottom: After using the filter and apply- ingtheoptimalmatchedfilter.RemnantRFIstreaksarefurther removedandpulsesfromthepulsarremain. fromPSRB1735 32atDMchannel=82(DM=50cm−3pc), − but we can see that the output is contaminated with many RFIstreaksmakingidentificationofthepulsardifficult.The middle panel shows the corresponding plot with zero-DM filteringapplied. The vast majority of theRFI hasbeen re- movedbutthepulsarisstillpresent.Obviousalsoareafew remnant RFI streaks at high DM which have not been removed by the filtering. Searching the time series with the optimal single-pulse profile, w(t), improves things even fur- therbyremovingtheremnantstreaksalmostcompletely,as Figure 6.FluctuationspectrafromaPMPSsurveyobservation can beseen in thelower panel. inthedirectionofthe0.53-secondpulsar,PSRJ1604-4718,which hasaDMof52cm−3pc.Thetoppanelshowsthespectrumwith- outanyzero-DMfiltering,thebottompanelisthesamedataafter 4 DISCOVERY OF FOUR NEW SOURCES thezero-DMfilterhasbeenapplied.Thearrowsindicatethepo- sition of the fundamental spin frequency of the pulsar and four The zero-DM filter has been incorporated into a new re- subsequent harmonics. Most of the RFI signals are completely analysis of the PMPS using acceleration searches (Eatough removedbythenewtechnique. etal.inprep).Sofarthreenewpulsarshavebeendiscovered (seeTable1& Figure8).Figure8shows folded subintegra- tionsand integrated pulseprofilesfor each of thepulsarsin bothastandardanalysisandafterusingthezero-DMfilter. Inallcases impulsiveRFIhasbeenremoved.Althoughthis demonstrates an improvement in the quality of the folded 2000 22000000 Table 1.Preliminarypropertiesofthe4newsourcesdiscovered J1539-4835 JJ11553399--44883355 ((zzeerroo--DDMM ffiilltteerreedd)) 1500 11550000 in our re-analyses of the PMPS. The nominal positions quoted time (s) 1000 time (s)time (s) 11000000 are the positions of the centre of the beam in which the source 500 550000 was discovered and have errorsof lessthan 7arcmin, the radius 0 0.2 0.4 0.6 0.8 1 00 00..22 00..44 00..66 00..88 11 12000 12000 ofthePMPSbeam(Manchester etal.2001). 10000 10000 flux (arbitrary units) - 246820000000000 000000 flux (arbitrary units) - 246820000000000 000000 Name α(J2000) δ(J2000) Period(s) DM(cm−3pc) -4000 0 0.2 0.4pulse phase 0.6 0.8 1 -4000 0 0.2 0.4pulse phase 0.6 0.8 1 J1724-3549 17:24:43.0 -35:49:18.6 1.4085 550.0 2000 22000000 J1539-4835 15:39:18.7 -48:35:14.0 1.2729 136.3 J1819-1711 JJ11881199--11771111 ((zzeerroo--DDMM ffiilltteerreedd)) 1500 11550000 JJ11881395--10711115 1188::1395::5103..39 --1071::1115::4187..00 00..30903551 49080..18 time (s) 1000 time (s)time (s) 11000000 500 550000 0 0.2 0.4 0.6 0.8 1 00 00..22 00..44 00..66 00..88 11 14000 14000 12000 12000 deiatthae,rtahereddisuccotvioenryooffRthFeIsefesaotuurrceessinwatsheprflimucatruilaytiaoindesdpebcy- flux (arbitrary units) 1-- 246842000000000000000 00000000 0 0.2 0.4 0.6 0.8 1 flux (arbitrary units) 1-- 246842000000000000000 00000000 0 0.2 0.4 0.6 0.8 1 pulse phase pulse phase tra or because wide spectral filters have not been applied aroundstrongperiodicsourcesofRFI.Ofparticularinterest 2000 22000000 J1835-0115 JJ11883355--00111155 ((zzeerroo--DDMM ffiilltteerreedd)) is PSR J1835 0115, a 5.1-millisecond pulsar in a 6-day or- 1500 11550000 bitaroundal−owmasscompanion.Allthreepulsarsarenow time (s) 1000 time (s)time (s) 11000000 500 550000 beingtimedusingradiotelescopesateitherJodrellBankor 0 0.2 0.4 0.6 0.8 1 00 00..22 00..44 00..66 00..88 11 Parkes. 35000 35000 30000 30000 re-anIanlyasdisdiotfiotnhethPeMzPerSo-uDsiMngfislitnegrlei-spbuelsinegseaaprcphlieesd(tKoeaonuer flux (arbitrary units) 1122- 550505000000000000 0000000 flux (arbitrary units) 1122- 550505000000000000 0000000 et al. in prep). Here we present one new source (see Table -10000 0 0.2 0.4 0.6 0.8 1 -10000 0 0.2 0.4 0.6 0.8 1 pulse phase pulse phase 1). This source, which has a period of 1.4s, shows single- Figure 8. Folded subintegrations and integrated pulse profiles pulse diagnostic plots which indicate a strong resemblance for three of the pulsars discovered. For each pulsar we show the to thenewly-discovered class of Rotating RAdioTransients standard analysis on the left and the zero-DM filtered data on (RRATs)(McLaughlinetal.2006).Interestingly,thisobject theright. showspulsesfromtwopartsofthepulserotationperiod,sep- arated by about 0.2 sec. We are continuing to monitor this sourceusing theParkesradio telescope tofurtherconstrain signals in the fluctuation spectra from independent obser- the spin period and to obtain a period derivative so it can vations at different positions on the sky. Such signals likely becomparedtothevaluesofperiodderivativemeasuredfor have a local terrestrial origin and enter the receiver system theother RRATs. through the far-out sidelobes of the telescope beam. Once theRFIsignalshavebeenidentifiedtheirharmonicsequence is removed from the fluctuation spectrum. Typical spectral 5 DISCUSSION zapping routines can remove a few percent of the fluctua- tionspectrumgivingasmallchancethatacoincidentpulsar Zero-DMfilteringreducesthenumberofspuriouscandidates frequencyoritsharmonics couldberemoved.Thezero-DM detected by pulsar search software - both in standard peri- filtering process presents a more natural way of removing odicity searches and in single-pulse searches. A major ad- such RFI signals without thedangers inherent tozapping. vantage of zero-DM filtering is that it eliminates the need Although effective at removing RFI in our analyses of toapplyarbitrarycleaningtechniquestopulsarsearchdata. thePMPS,wenownoteanumberoflimitationsandpracti- Onecommonly-usedtechniqueisthatofclippingthemulti- calities of the zero-DM filter. Firstly the technique will not channel data based upon spikes in the DM=0 time series benefit high-frequency, or small bandwidth pulsar searches (e.g. Hobbs et al 2004). For 1-bit sampled filterbank data because in this case even signals from celestial sources will like that in the PMPS, with n channels, the time samples not be highly dispersed. For example, the dispersive de- shouldhaveameanofn/2andastandarddeviationof√n/2 lay across the entire 576-MHz bandwidth in the Parkes (e.g. Lorimer & Kramer 2005). Typical clipping algorithms methanol multi-beam pulsar survey at 6.3 GHz is less than set each of the n channels to alternate zeros and ones if 2 ms for a source with DM=100 cm−3pc (e.g. O’Brien et the time sample differs from the mean by more than three al.2008.). Inthiscasethezero-DMfilterwould havealow- standarddeviations.Clippingcouldposeathreattointense frequencycutoff around 230 Hzmaking detection of all but burst-likedispersedsignalsthatarestrongenoughtobede- millisecond pulsars impossible. tectedinthelow-DMtrialsofapulsarsearch.Itisprecisely In single-pulse searches, impulsive RFI signals can oc- thesetypeoftransienteventswhichmayrevealexcitingnew casionally be strong enough to persist even after zero-DM astrophysical phenomena (Lorimer et al. 2007). Since zero- filtering.TheremnantRFIstreaksappearashighsignal-to- DMfilteringreliesontheundispersednatureofRFI,rather noise features at non-zero DM. However such remnant RFI than its sheer strength, such signals will not be removed. streakswillnothavetheexpecteddropoffinsignal-to-noise The primary RFI excision technique for pulsar period- ratio as a function of trial DM, nor will they appear dis- icitysearchesupuntilnowhasbeenfrequencydomain‘zap- persed according to Equation 1 like a true celestial source. ping’. This method requires the identification of common Theyarealsolikelytoappearinmultiplebeamsofamulti- beam receiver. Each of these additional properties can be Lorimer D. R., Bailes M., McLaughlin M. A., Narkevic D. J., used to identify such spurious signals. CrawfordF.,2007,Science,318,777 An assumption made in the analysis of the implemen- LorimerD.R.,KramerM.,2005,HandbookofPulsarAstronomy. tation of the zero-DM filter is that of a linear dispersion CambridgeUniversityPress LyneA.G.,SmithF.G.,2006,PulsarAstronomy,3rded..Cam- slope when the true slope is in fact quadratic. A complete bridgeUniversityPress,Cambridge description of the dispersion slope increases the difficulty ManchesterR.N.,LyneA.G.,CamiloF.,BellJ.F.,KaspiV.M., of implementing the algorithm, with little benefit. The ef- D’AmicoN.,McKayN.P.F.,CrawfordF.,StairsI.H.,Pos- fectsofthisassumptionmanifestthemselvesinphase-folded senti A., Morris D. J., Sheppard D. C., 2001, MNRAS, 328, pulse profiles which have asymmetric dips as shown in Fig- 17 ure 5 for PSR J1842+0257, unlike the symmetric dips we McLaughlin M. A., Lyne A. G., Lorimer D. R., Kramer M., haveconsidered in Figure 1. Faulkner A. J., Manchester R. N., Cordes J. M., Camilo F., Future pulsar surveys will be performed with higher Possenti A.,StairsI.H.,HobbsG.,D’AmicoN.,BurgayM., timeandfrequencyresolution2.Fast dedispersion codes are O’BrienJ.T.,2006,Nature,439,817 beingdeveloped to copewith thelarge volumeof data that O’Brien J. T., Johnston S., Kramer M., Lyne A. G., Bailes M., Possenti A., Burgay M., Lorimer D. R., McLaughlin M. A., will be produced (Bailes, private communication). These HobbsG.,ParentD.,GuillemotL.,2008,MNRAS,388,L1 dedispersionalgorithmsoperateon1-bitor2-bitdatabefore thesamples havebeen unpackedas floating point numbers. Inthesealgorithms,calculationofthemeanandsubsequent subtractionfromeachfrequencychannelsampleasinEqua- tion 2 may not bepossible at thepre-dedispersion stage. The factor of two increase in the number of compu- tations imposed by re-application of the zero-DM filter after every dispersion trial might not be practical in searches where large numbers of Fast Fourier Transformations are performed, viz. searches for highly accelerated binary pulsars (e.g. Eatough et al. in prep).Anotherpractical consid- eration is for data sampled with a larger number of bits, i.e. a large dynamic range. In this scenario any RFI with intensity structure across the band may ‘leak’ through the zero-DM filter. Wehaveimplementedthezero-DMfilterintheSigproc- 4.2pulsaranalysissoftwaresuiteandarecurrentlyapplying theprocessinare-analysisofthePMPSinbothacceleration searches (Eatough et al. in prep) and single-pulse searches (Keaneet al. in prep). ACKNOWLEDGEMENTS This research was partly funded by grants from the Sci- ence&TechnologyFacilities Council.EvanKeaneacknowl- edgesthesupportofaMarie-CurieESTFellowshipwiththe FP6 Network “ESTRELA” under contract number MEST- CT-2005-19669. The Australia Telescope is funded by the Commonwealth ofAustraliaforoperationasaNationalFa- cility managed by the CSIRO. We would like to thank M. Kramer, B. Stappers, and C. Jordan for useful discussions and manuscript reading. We also acknowledge A. Forti and TheUniversityofManchesterParticlePhysicsgroupforuse of the Tier2 computingfacility3. REFERENCES HobbsG.,FaulknerA.,StairsI.H.,CamiloF.,ManchesterR.N., Lyne A. G.,KramerM.,D’AmicoN.,KaspiV. M.,Possenti A.,McLaughlinM.A.,LorimerD.R.,BurgayM.,JoshiB.C., CrawfordF.,2004,MNRAS,352,1439 2 http://astronomy.swin.edu.au/pulsar/?topic=hlsurvey 3 http://www.hep.manchester.ac.uk/computing/tier2