Krell had there own design on this making the amp as a source lower them there preamp
if you used there own output
this they claimed reduced greatly the effects of interconnects
IIRC, the Krell CAST system uses current-mode driver (high-impedance) and receiver (low-impedance) circuits. One idea (claim) being that current-mode interconnects pick up less noise than voltage-mode links. Essentially the output of the preamp is a current source, ideally infinite impedance, and the amplifier input a current sink which is optimally 0-ohm impedance. Current mode is often used in IC bias circuits because coupling on-chip does tend toward voltage noise moreso than current noise. A number of my (RF/mW) designs operated in current mode as it provided greater bandwidth and decently low distortion, albeit with higher noise. The current source is also sensitive to loading; you are depending upon the load's very load impedance to counter that. There are some very high-speed (up to Gb/s) signal interfaces that are current-based.
The trade is less clear (to me) for an audio link given high-current power and speaker lines that may induce noise via current coupling, and there is the argument that you've just moved the high-impedance (more sensitive) point from the load to the source. There is also the issue that you still have to convert to voltage someplace, and, while current-mode amps tend to be more intrinsically linear than pure voltage-mode amps, current amplifier circuits tend to have much less gain making it difficult to apply feedback to improve their performance. (Yes, I realize the trend toward "no feedback", I just don't agree with it. But, IIRC, Krell claimed "no feedback" in their circuits, accomplishing low distortion with lots of bias current, then -- I
think -- varied dynamically to keep from burning down the house.) The scheme seemed to work, but it has been many years since I listened to a Krell component. I have no reason to doubt their implementation.
Special cases aside, the vast majority of preamps today output a voltage, and the amplifier inputs expect a voltage, so I think the argument is valid even though you can always find an exception.
IME/IMO - Don