1 Introduction

We consider here planetary-search or asteroseismology programs. With CCD-equipped crossed-dispersion spectrographs, photon noise permits in principle to measure the radial velocity (RV) changes of many stars using relatively small telescopes with a precision of the order of 1 m/s (Connes 1985; Brown 1990). In order to achieve this precision in practice, steps must be taken to control many sources of instrumental errors. A major one is due to the motions of the star image across the spectrometer input due to telescope-guiding or seeing fluctuations. The slit width in a high-resolution echelle spectrograph, in RV units, is typically a few km s^-1. Thus, if the photocenter of the star image suffers a shift equal to 10^-3 slit width, errors larger than 1 m/s will occur. Since the slit width projected on the sky is typically a few arcsec, it seems that milli-arcsec guiding precision is required. One very successful method makes use of a molecular absorption cell to impress lines of stable wavelength on the incoming starlight (Butler et al. 1996; Cochran & Hatzes 1994; Walker et al. 1995). The stellar and reference spectra are then recorded from the same beam, thereby circumventing the problem of the photocenter shift. However this approach increases considerably the photon-noise RV errors, hence requires larger telescopes for a given star. Another method, which avoids this drawback, makes use of a fiber-optic feed for the starlight, plus a second fiber carrying light from a stable wavelength source (Mayor & Queloz 1995; Brown et al. 1994). One may also use a single fiber alternately for both beams, as intended for the "absolute accelerometry'' (AAA), proposed by Connes (1985) and developed by Schmitt (1997). Unfortunately, the scrambling action of a fiber reduces but does not cancel the stellar-beam geometrical fluctuations within the spectrograph. The use of a double-fiber scrambler is a current partial cure. The present paper is devoted to a study of these effects, and to the description of another partial cure: a fiber-locked auto guider (called FLAG), in which the fiber-input plays the role of guiding detector.

Several fast-guiding (or so-called "tip-tilt'') systems have recently been described (e.g. ISIS, DISCO, HRCam, AOC, FASTTRAC, UH/Ifa). In all cases, the goal is reduction of stellar image size as recorded over "long'' exposures (i.e., seconds or minutes) through elimination of fast photocenter wandering. The overall improvement may be as large as 2, for small values of the FRIED ratio $D/r_{\rm o}$. An even greater gain is realized from adaptive optics (AO), which eliminate not only photocenter motion, but also instantaneous image blurring from wavefront distortion; this is achieved at the cost of many servo channels, hence far greater complexity.

The goal of FLAG is different. Image size reduction by itself will be welcome, as leading to either more photons in the spectrometer, or higher resolution or both. However, with meter-size telescopes in the visible, the improvement may be small at best: e.g., in our Sect. 5.1.2 tests, D = 152 cm, $r_{\rm o} = 6$ cm, and Hecquet & Coupinot (1985) predict a reduction of 1.25 in image diameter. Furthermore, except for the fainter stars within the RV program, the results will not be limited by photon noise but by "scrambler noise''. Hence, stabilisation of the fiber-ouput beam is our prime goal. Moreover, a planetary-search program involves a large number of "exposures'', stretching over years; hence, it is not for minutes but for years that we want to keep our beam stable irrespective of drifts in the autoguider, possibly even of guiding-detector changes. For the same reason, we also want to correct focus position.

A last difference: image size improvements from the above tip-tilt devices may be predicted from atmospheric turbulence models, and actual agreement is not too bad. Here, the situation is far more complex: the beam is fed not to a plain imaging camera, but to a camera through a fiber first, and a spectrograph second. There is no practical way of modeling the overall optical system with the precision required here. Presently, all fiber-plus-spectrograph stellar-RV programs (including ours) are stuck at an error level of a few m/s: this state of affairs could not have been predicted from any theoretical calculation, and still cannot be accounted for quantitatively. This situation may change drastically in the future, should AO devices progress enough to give acceptable STREHL ratios down to 400 nm for a 1 or 2-m telescope. As pointed out in Sect. 2, we may then feed the spectrometer with a single-mode (SM) fiber, which acts as a perfect scrambler; the output-beam (easily computable) will then be fully stabilized, irrespective of any input-beam fluctuation, and even of residual AO errors.

Returning to the present day, the intent and limitations of FLAG may now be better understood: Unlike the builders of the current tip-tilt systems, we cannot put a number to the expected improvement; worse, we cannot even hope to demonstrate a large one within the short time range of the observations reported here. Like them, we are doing the best we can with a simple device, which risks obsolescence from future AO progress.

A brief review of the scrambling properties of step-index fibers and their use in astronomical spectrographs is given in Sect. 2. In Sect. 3, we present our own measurements of stability of the stellar beam within the ELODIE fiber-fed spectrograph, using an available but inadequate standard autoguider based on a CCD detector. In Sect. 4, we describe our prototype FLAG, built for the new spectrograph EMILIE (under construction, Bouchy et al. 1999), and the results are discussed in Sect. 5. The system is fully compatible with a double-scrambler, but has not yet been used with one.