Thanks for the additional tip, Paul. I remember reading that same tip elsewhere (though I can't remember where. an MRL paper?).
Yes, it is discussed in one of Jay McKnight's papers available on the MRL site.
One thing that I'm puzzled by with regard to this tip, however:
Wouldn't I be better off biasing / calibrating by using the larger voltage / db swing? My intuition tells me calibrating to a 0.2 db difference would be more susceptible to noise margins than calibrating to a 3.5 db difference.
You have correctly identified the one advantage of using 10KHz: the larger swing. I have never really heard or read any argument in favor of using 10K, but I assume that's the reason. It's theoretically possible to dial it in more precisely because of the wider swing.
I say "theoretically" for several reasons. If you are using a VU meter, it's probably more accurate within 1/2 dB of zero than it is at minus 3.5, and if you're using a meter like a Fluke with a good digital readout then you can get 0.2 without any difficulty. But still, let's say you have a way of metering 3.5 dB accurately. That 3.5 dB target may or may not be accurate. It is based on an average sample of the tape, and the optimum bias for a given piece of that formulation will vary a lot more at 10K than it does at 1K. More than that, the figure is based on the length of the record head gap. And not the physical length, but the effective length, which varies as the head wears. Now with your Studer, you can be pretty sure that the gap (when the head was new) is what they said it was...unless it has been replaced. Studer always made their own heads, and the tolerances were very tight. However some manufacturers didn't make their own heads and sourced them from a variety of OEM's. I know of some instances of MCI machines of the same model number with heads sourced from different OEM's, and with different head gaps.
Anyone who wants to use 10K and rely on the tables may do so. I won't argue. But I want to bias my machines critically, and I have chosen to do it at 1K. To condense a very long story to a few words, I will say that I arrived at this after having biased my machines by many methods, including by ear for a long time, and proving to myself that I could get it just as accurate by doing 1K a quarter dB over.
Remember what we're trying to accomplish here. We aren't optimising 1K. And we certainly aren't optimising 10K, or we wouldn't be shooting for a point 3.5 dB away from the peak. We are trying to get it into the notch that represents the minimum modulation noise and the minimum distortion. Any method that hits that minimum is a good method.