1 Introduction

Eta Carinae ($\eta$ Car) is one of the most interesting stellar objects known. It is disconcerting that after 150 years of research, the real nature of the central star remains unknown. A review of the many problems involved can be found in Davidson & Humphreys (1997). The recent discovery of a 5.52 year periodicity (Damineli 1996) opened a new road to explore this object. Moreover, its spectrum seems to be understandable in light of the binary nature proposed by Damineli et al. (1997, hereafther DCL); also see Davidson (1997).

Since Le Sueur reported the first visual observation in 1870, $\eta$Car has been the subject of many publications. During the last century, its spectrum has been characterized by strong H I, Fe II, and [Fe II] lines (Cannon 1916; Hoffleit 1933; Whitney 1952; Thackeray 1953; Walborn & Liller 1977). For brief time intervals in 1948, 1965, 1981, 1987, and 1992, the spectrum has shown remarkable fading of the high excitation lines, such as [Ne III] and [Fe III] (Gaviola 1953; Rodgers & Searle 1967; Thackeray 1967; Whitelock et al. 1983; Ruiz et al. 1984; Zanella et al. 1984; Allen et al. 1985; Bandiera et al. 1989; Altamore et al. 1994; Damineli 1995; Damineli 1996). Such behavior led to the idea of a normal spectrum that occasionally is replaced by an event spectrum during shell episodes (Zanella et al. 1984). Long term monitoring carried out since 1989 at Laboratório Nacional de Astrofísica (LNA/CNPq - Brazil) and 1992 at European Southern Observatory (ESO - Chile) has shown that the spectrum varies continuously from a state in which the high excitation lines are at maximum intensity, which we call a high state spectrum, to a complete disappearance of the high excitation lines, which we call a low state spectrum. We prefer such a purely phenomenological description of the spectrum because it is not proven that the mechanism that governs the spectroscopic events is the ejection of a shell by the star.

The prediction and successful observation of a spectroscopic event in 1998.0 (Damineli 1996; Abraham & Damineli 1997; Lopes & Damineli 1997; Jablonski et al. 1998) allowed the collection of huge amount of data in all electromagnetic energy windows. Critical constraints to models have been derived from observations ranging from X-rays to radio wavelengths. The presently known data favors the existence of a long period binary with wind-wind collision in the heart of the Homunculus (DCL; Stevens et al. 1997; Corcoran et al. 1997). Optical and near-infrared spectroscopy, however, hasn't been exploited at the level permitted by the techniques available in the last twenty years. Some published papers include high resolution or large spectral coverage but both characteristics have not been used at the same time (Hillier & Allen 1992; Meaburn et al. 1993; Damineli et al. 1993; Hamann et al. 1994). In particular, a systematic study of line parameters, separating the narrow and broad line components, has not yet been performed, although it is known that they are formed in regions of different physical characteristics. Davidson et al. (1995, 1997) have shown that the narrow components are formed mainly in slow-moving condensations about 0$.\!\!^{\prime\prime}$3 from the star; see Sect. 6 below. In some important works, unfortunately, information from spectra of different dates have been analyzed together (Davidson et al. 1986; Hamann et al. 1994; Johansson et al. 1995), in spite of the known spectral variability. Even line identification (Aller & Dunham 1966; Viotti 1968; Thackeray 1969; Hamann et al. 1994) merits revision on the basis of higher resolution observations and new atomic transition calculations.

The purpose of our long-term survey is to improve the knowledge of $\eta$ Car's spectrum at wavelengths accessible by CCDs through ground-based telescopes. In this particular work, we analyze the wealth of information that can be derived by comparing the spectrum on two extreme situations, the states of highest and lowest excitation, that occurred mid-1995 and mid-1992, respectively. A detailed study of the temporal variability will be reported in a separate paper.