An interesting probe of the nature of dark energy is the measure of its sound speed, $c_s$. We review the significance for constraining sound speed models of dark energy using large neutral hydrogen (HI) surveys with the Square Kilometre Array (SKA). Our analysis considers the effect on the sound speed measurement that arises from the covariance of $c_s$ with the dark energy density, $\Omega_\LLambda$, and a time-varying equation of state, $w(a)=w_0+(1-a)w_a$. We find that the approximate degeneracy between dark energy parameters that arises in power spectrum observations is lifted through redshift tomography of the HI-galaxy angular power spectrum, resulting in sound speed constraints that are not severely degraded. The cross-correlation of the galaxy and the integrated Sachs-Wolfe (ISW) effect spectra contributes approximately 10 percent of the information that is needed to distinguish variations in the dark energy parameters, and most of the discriminating signal comes from the galaxy auto-correlation spectrum. We also find that the sound speed constraints are weakly sensitive to the HI bias model. These constraints do not improve substantially for a significantly deeper HI survey since most of the clustering sensitivity to sound speed variations arises from $z \lsim 1.5$. A detection of models with sound speeds close to zero, $c_s \lsim 0.01,$ is possible for dark energy models with $w\gsim -0.9$.

We investigate the potential of the Square Kilometer Array Telescope (SKA) to constrain the sound speed of dark energy. The Integrated Sachs Wolfe (ISW) effect results in a significant power spectrum signal when CMB temperature anisotropies are cross-correlated with galaxies detectable with the SKA in HI. We consider using this measurement, the autocorrelation of HI galaxies and the CMB temperature power spectrum to derive constraints on the sound speed. We study the contributions to the cross-correlation signal made by galaxies at different redshifts and use redshift tomography to improve the signal-to-noise. We use a chi-square analysis to estimate the significance of detecting a sound speed different from that expected in quintessence models, finding that there is potential to distinguish very low sound speeds from the quintessence value.