Future directions

• More and more sound cards have on-board programmable DSPs, but unfortunately the sound card manufacturers are not willing to release programming information about them. If this information ever becomes available, it will be possible to offload much of the modem's signal processing from the main CPU to the sound card.

• Increase the bit rate by using a better hybrid circuit.

• Increase the bit rate by implementing sequence detection in the PR-PAM receiver.

• Alternatively, increase the bit rate by switching to QAM (and optimizing it enough to run in real time).

• Double the bit rate by by using both channels of a stereo sound card simultaneously with a four-wire leased line.

• Alternatively, use the two channels of the stereo sound card to serve two separate AuDSL connections simultaneously.

• Transmit each IP packet as a self-contained audio burst with its own synchronization preamble, instead of the current continuous bit stream. This would have several advantages. First, this would save CPU cycles by not having to generate and process an idle bit stream when no data is being transmitted. Second, it could avoid many of the problems with the poor real-time scheduling under Unix because the audio signal for each transmit packet could be calculated in non-real-time and then placed in the sound card's playback DMA buffer all at once.

• Reduce CPU usage by making more aggressive use of Intel's MMX and/or AMD's 3DNow instruction set.

• Turn this technology into a hardware modem product, either a PCI card or an external unit. By taking advantage of low-cost consumer audio DSPs and converters, the product could be quite inexpensive.