We discuss the detection of large-scale H i intensity fluctuations using a single dish approach with the ultimate objective of measuring the baryonic acoustic oscillations (BAO) and constraining the properties of dark energy. To characterize the signal we present 3D power spectra, 2D angular power spectra for individual redshift slices and also individual line-of-sight spectra computed using the S3 simulated H i catalogue which is based on the Millennium Simulation. We consider optimal instrument design and survey strategies for a single dish observation at low and high redshift for a fixed sensitivity. For a survey corresponding to an instrument with Tsys = 50 K, 50 feedhorns and 1 year of observations, we find that at low redshift (z ≈ 0.3), a resolution of ∼40 arcmin and a survey of ∼5000 deg2 is close to optimal, whereas at higher redshift (z ≈ 0.9) a resolution of ∼10 arcmin and ∼500 deg2 would be necessary – something which would be difficult to achieve cheaply using a single dish. Continuum foreground emission from the Galaxy and extragalactic radio sources are potentially a problem. In particular, we suggest that it could be that the dominant extragalactic foreground comes from the clustering of very weak sources. We assess its amplitude and discuss ways by which it might be mitigated. We then introduce our concept for a dedicated single dish telescope designed to detect BAO at low redshifts. It involves an underillumintated static ∼40 m dish and a ∼60 element receiver array held ∼90 m above the underilluminated dish. Correlation receivers will be used with each main science beam referenced against an antenna pointing at one of the celestial poles for stability and control of systematics. We make sensitivity estimates for our proposed system and projections for the uncertainties on the power spectrum after 1 year of observations. We find that it is possible to measure the acoustic scale at z ≈ 0.3 with an accuracy ∼2.4 per cent and that w can be measured to an accuracy of 16 per cent.