Some optimal designs for repeated measurements with autoregressive errors