Forum Springer Astron. Astrophys.
Forum Whats New Search Orders

Astron. Astrophys. 358, 169-176 (2000)

Previous Section Next Section Title Page Table of Contents

2. Survey observations and data reduction

All observations reported in this paper were carried out at a centre frequency of 1402 MHz on a number of separate sessions between 1998 June and 1999 April using the 100-m Effelsberg radio telescope operated by the Max-Planck-Institut für Radioastronomie. Although 1400-MHz timing observations at Effelsberg are routinely made with typical bandwidths of 40 MHz or more (see e.g. Kramer et al. 1998), the search hardware available to us has a maximum bandwidth of 16 MHz in each of the two orthogonal, circular polarisation channels. Nonetheless, the large forward gain of the telescope at 1400 MHz (1.5 K Jy-1), the relatively low system temperature of the receiver (35 K) and long integration times employed in the survey (35 min per pointing) means that the system achieves a sensitivity which represents a threefold improvement over that achieved by Clifton et al. (1992) during their survey.

The main aim of the observations reported here was to test the feasibility of a larger search with a 100-MHz bandwidth system which is presently being commissioned. Given the limited amount of telescope time available for this pilot project, we chose to restrict our search area to a 2 deg2 patch of the Galactic plane defined by [FORMULA] and [FORMULA]. The rationale for this choice is simple - this line-of-sight is close to the Scutum spiral arm and as a result passes through one of the most pulsar-rich parts of the northern Galactic plane. In addition, since this part of the sky is not visible from Arecibo, Effelsberg is presently the largest radio telescope in the world capable of surveying it. The survey region was divided up into a grid of 126 positions consisting of 9 strips of 14 positions along lines of constant galactic latitude ([FORMULA]). This choice of spacing ensured some overlap between the 3-dB width of the telescope beam (9´). The [FORMULA] strip was centred on [FORMULA]. Beam centres on adjacent strips were alternately offset by half a beam width to ensure the most efficient coverage on the sky.

At the start of each observing run, we carried out a 5-min observation of PSR B2011+38. This relatively luminous 230-ms pulsar has a dispersion measure of 239 cm-3 pc and is known not to be prone to significant intensity variations due to interstellar scintillation (Lorimer et al. 1995). The fact that the search code detected this pulsar with consistently high signal-to-noise ratios (consistent with its 1400-MHz flux density - [FORMULA] mJy; Lorimer et al. 1995) gave us confidence that the individual filterbank channels were functioning normally, and that the nominal system sensitivity was being achieved.

The search field is visible from Effelsberg for about 8.5 hr per day. Since each grid position in the field was observed for 35 min, we typically observed up to 14 separate positions on the sky during a given transit. In search mode, the incoming signals of each polarisation are fed into a pair of [FORMULA]-MHz filterbanks. The outputs from the filterbanks are subsequently detected and digitised every 500 µs using 2-MHz voltage-to-frequency converters, resulting in an effective 10-bit quantisation of the signals. This is the fastest sustainable data rate using this system. Signals from the orthogonal polarisations were combined to form a total power time series for each 4-MHz frequency channel over the band. These four frequency channels are then passed to the standard Effelsberg Pulsar Observing System (see Kramer 1995) which stored contiguous blocks of data to disk every 1024 samples (0.512 s).

The four-channel search system used for this survey results in refreshingly low data rates compared to most other searches where the backends routinely sample 256 channels or more (see e.g. Manchester et al. 1996). The main advantage of such a simple system is that, as soon as each 35-min integration was complete, a preliminary analysis of the data could be carried out well within the time that the telescope was observing the next grid position. This quasi on-the-fly processing scheme allowed rapid re-observation of any pulsar candidates found during the search.

The data analysis procedure was optimised to search Fourier spectra of each 35-min time series for dispersed periodic signals. The software for this purpose was developed largely from scratch, taking advantage of ideas used to process our on-going search of the Galactic centre from Effelsberg (Kramer et al. 1996; Kramer et al. 2000), as well as previous experience gained by one of us (DRL) during the Parkes [FORMULA] 70-cm Southern Sky Survey (Manchester et al. 1996; Lyne et al. 1998). We also made use of several "standard" pulsar search techniques described in detail by a number of authors (Hankins & Rickett 1975; Lyne 1988; Nice 1992). In what follows we give a brief overview of our analysis procedure.

We adopted a two-stage data analysis procedure whereby the data were quickly analysed after the observation at Effelsberg and then stored on magnetic tape for more detailed off-line analyses at Bonn. The three differences in the analyses are the range of dispersion measures searched, the signal-to-noise thresholds, and the method used to excise radio frequency interference (see below).

Both analyses begin by computing the Fast Fourier Transform of the [FORMULA] point time series in each of the four frequency channels. Since the dispersion measure (DM) of any pulsar is a priori unknown we need to de-disperse the data for a number of trial DM values before the periodicity search begins. For our purposes this is most readily achieved by applying the shift theorem (see e.g. Bracewell 1965) to the Fourier components of each channel before summing appropriately to produce a number of de-dispersed amplitude spectra. Our on-line analysis in Effelsberg produced 18 amplitude spectra for each beam corresponding to a DM range between zero and 1,500 cm-3 pc. Subsequent analyses in Bonn produced, in addition to this, a further 18 spectra per beam which increased the range of DMs out to 10,000 cm-3 pc. Each amplitude spectrum was then searched for harmonically related spikes in the Fourier domain - the characteristic signature of any periodic signal. Since pulsar signals have generally short duty cycles, and therefore many harmonics, we summed the spectra over 2, 4, 8, and 16 harmonics using an algorithm described by Lyne (1988) and repeated the search for significant spectral features.

Having completed the search of all the amplitude spectra for a given beam, we then compiled a list of all non-harmonically-related spectral features with a signal-to-noise ratio greater than 8 in the Effelsberg analyses and 7 in the Bonn analyses. Typically, depending on the amount of interference present in the data, there are of order five to ten such "pulsar candidates" in each beam. For each candidate, the analysis described so far resulted in a period P and dispersion measure DM; the latter quantity is based upon the maximum spectral signal-to-noise ratio found as a function of all the DM trials. Working now in the time domain, we fold the filterbank channels at the nominal period of each candidate to produce one pulse profile per channel. These profiles are then de-dispersed at the nominal DM to produce an integrated profile over the 16-MHz band.

The results of this analysis are summarised in the plot of the form shown in Fig. 1 which is the output from the discovery observation of PSR J1842-0415. This plot serves as a good example showing the characteristics of a strong pulsar candidate. The high signal-to-noise integrated profile (top left panel) can be seen as a function of time and radio frequency in the grey scales (lower left and right panels). In addition, the dispersed nature of the signal is immediately evident in the upper right hand panel which shows the signal-to-noise ratio as a function of trial DM. This combination of diagnostics proved extremely useful in differentiating between a good pulsar candidate and spurious interference.

[FIGURE] Fig. 1. Sample search code output showing the discovery observation of PSR J1842-0415 - the first of the four pulsars found during the survey. This plot shows how a typical pulsar appears to the search code and summarises the various diagnostics we used to identify the best pulsar candidates from the survey (see text).

The most significant difference between our two data reduction strategies concerns the methods employed to eliminate radio frequency interference. Since the radio frequency environment in Effelsberg is pervaded by a number of man-made signals with fluctuation frequencies predominantly between 10-2000-Hz, both modes of data reduction required some means of excising these unwanted signals. The "on-line" data reduction mode in Effelsberg achieved this by simply clipping all spectral features above 10-Hz whose amplitudes exceeded five times the spectral rms! Whilst this simple-minded approach was sufficient to detect and confirm all the pulsars finally discovered in the survey, we were aware that it significantly compromised our sensitivity to pulsars with periods below 0.1 s.

To address this important issue, our data analysis procedure in Bonn made use of the fact that the vast majority of man-made interfering signals are not dispersed and occur predominantly at a constant fluctuation frequency at any given epoch. These signals are immediately apparent in a compilation of a large number of zero-DM amplitude spectra for different beam positions. Based on the statistics of over 60 individual spectra, we constructed a "spectral mask" which contains the frequencies of those spectral features which occur more than 5 times above a signal-to-noise threshold of 7. We found 611 such frequencies between 30 and 2000-Hz - 0.06% of the total number of spectral bins. Most of these are in fact related to the 50-Hz mains power line. By masking (i.e. ignoring) just these frequencies in our analysis, it was then possible to detect short-period pulsars with fundamental frequencies outside the masked frequency bins in our data. We verified the validity of this approach by a analysing number of test observations on millisecond and short-period pulsars which were essentially undetectable without the use of the spectral mask, simply because of the dominating effect of the interfering signals. Thus, although we did not detect any short-period pulsars in this survey, we are confident that no potentially detectable pulsars with fundamental frequencies outside the masked frequency bins were missed because of radio frequency interference.

Previous Section Next Section Title Page Table of Contents

© European Southern Observatory (ESO) 2000

Online publication: June 26, 2000