In a simple linear regression $Y = \beta_0 + \beta_1 X + \epsilon$, if the errors are heteroscedastic with $Var(\epsilon_i) = \sigma^2 x_i^2$, which transformation could be used to return the model to a homoscedastic state?