A new method of wireless data telemetry in oil well services uses compressional acoustic waves to transmit data along the drill string. The compressional acoustic waves are produced as coded wave trains by an acoustic transducer, travel through the drill string and subsequently decoded to recover the data. Normal drilling operations produce in-band acoustic noise from multiple sources at intensities comparable to the transducer output. Based on a theoretical channel model, we predict that the drill string acoustic channel has a capacity of several hundreds bits per second in such noisy drilling conditions. In making these calculations, the surface noise is shown to be the limiting factor. We explore methods of improving channel capacity by exploiting upward and downward propagation modes. Simulations based on our acoustic model show that substantial (an order of magnitude or more) capacity increases can result if we can cancel the noise-laden downward mode. We show that our two-receiver cancellation algorithm can be derived with a nonparametric, training-based approach that assumes little about the acoustic model. Finally, we investigate the effects of noise attenuators on capacity.