[en] When a continuous signal f(t) is digitized and then spectrally analysed, the resultant energy spectral density R(ω) is given as R(ω) = |F(ω) * D(ω)|², where F(ω) is the exact Fourier transform of f(t), D(ω) is the exact Fourier transform of the digitization process and * denotes convolution operation. A notable practical problem in spectral analysis is how to adequately decouple D(ω) from R(ω) and hence obtain the exact energy spectral density of f(t), i.e. |F(ω)|², since R(ω) → |F(ω)|² only if D(ω) → delta(ω) or (under certain conditions) when D(ω) → delta(ω-ω₀) or if D(ω) → Σsub(n) delta(ω-ωsub(n)), where the latter is a sufficiently spaced series of delta functions and ωsub(j) is constant for a given j. A solution to this problem requires, among others, thorough understanding of D(ω), how it relates to F(ω) and hence the manner or degree to which D(ω) distorts or contaminates F(ω) to form R(ω). In this paper, we have developed exact analytical expressions of D(ω) that are well related to the corresponding F(ω) in the cases when f(t) is a simple sinusoid as well as when it is in the form of a more complex function. It is established that in either of these cases, D(ω) is a clear function of the salient parameters of both f(t) and F(ω). The contents of this paper are used in Part II to examine the manner and extent to which D(ω) causes distortions in R(ω) under given conditions, and also to establish a procedure by which such distortions may be decoupled from a practically computed R(ω). Other related issues such as frequency shifts in computed power spectra are also discussed therein. (author)