Context : The RV Tauri stars constitute a small group of classical pulsating stars with some dozen known members in the Milky Way .
The light variation is caused predominantly by pulsations , but these alone do not explain the full complexity of the light curves .
High quality photometry of RV Tau-type stars is very rare .
DF Cygni is the only member of this class of stars in the original Kepler field , hence allowing the most accurate photometric investigation of an RV Tauri star to date .

Aims : The main goal is to analyse the periodicities of the RV Tauri-type star DF Cygni by combining four years of high-quality Kepler photometry with almost half a century of visual data collected by the American Association of Variable Star Observers .

Methods : Kepler quarters of data have been stitched together to minimize the systematic effects of the space data .
The mean levels have been matched with the AAVSO visual data .
Both datasets have been submitted to Fourier and wavelet analyses , while the stability of the main pulsations has been studied with the O–C method and the analysis of the time-dependent amplitudes .

Results : DF Cygni shows a very rich behaviour on all time-scales .
The slow variation has a period of 779.606 d and it has been remarkably coherent during the whole time-span of the combined data .
On top of the long-term cycles the pulsations appear with a period of 24.925 d ( or the double period of 49.85 d if we take the RV Tau-type alternation of the cycles into account ) .
Both types of light variation significantly fluctuate in time , with a constantly changing interplay of amplitude and phase modulations .
The long-period change ( i.e .
the RVb signature ) somewhat resembles the Long Secondary Period ( LSP ) phenomenon of the pulsating red giants , whereas the short-period pulsations are very similar to those of the Cepheid variables .
Comparing the pulsation patterns with the latest models of Type-II Cepheids , we found evidence of strong non-linear effects directly observable in the Kepler light curve .

Conclusions :