After much forum reading, testing and head scratching, I have finally come up with a plan for the output devices for the BA5k. In addition to my usual molasses-like plodding along, I also suffered a pretty severe case of "paralysis by analysis" as I pondered how to reconcile the optimum choices with being pragmatic.
At some point, the eureka moment occurred when I realized that, based upon the instructions in the service manual, the Sansui engineers must have heavily relied upon the emitter resistors in this design to force current sharing among the output devices. The specification for replacing any given output device was simply that the new device be in the same gain grouping as the original, and it had become obvious that the gain parameters within these groupings was certainly not stringent and that I should allow myself some fairly wide latitude in matching.
Initially I had leaned toward using contemporary MJ21193/4 devices, but this project is ultimately only for my use and I felt I would take a chance with reliability and try to use the Sanken devices that I had on hand. Using the rig shown in an earlier post, modified somewhat, I eventually devised a set of matching tests that I felt I could apply consistently. This is the final version, with two different pots to allow switching between lower and higher resistance between base and ground resulting in 45ma and 2A levels at 50v.
There seems to be some consensus across many diy forums that Vbe at quiescent current is the best method of matching parallel outputs, and with the fact that I did not feel I could reliably measure Vbe at higher currents, this has been the primary value that I am comparing. In this case, I tested the transistors at 50V and 45ma of collector current. However, the first test I would run was for hfe at 50V and 2A collector current, allowing the device to heat up somewhat. The collector current would stabilize within 15-20 seconds and I would take the reading of base current, and then switch between the two different potentiometers on the rig to drop the collector current to the 45ma level. The Vbe is very temperature sensitive, so once I had the collector current dialed in to approximately 45ma, I would power down for 3-4 seconds to let it cool and then bring it back on line, do a final tweak for 45ma, then wait 30 seconds to read the Vbe. At that point there is still some gradual drift at the 4th decimal point but the millivolt level is fairly stable.
With this approach, I measured all of the 2sa908/9 and 2sc1585/6 devices that I have. Based upon the best matching I could find, my plan is to use the 908/1585 combination on one channel and the 909/1586 combination on the other. Since there is as much variance in Vbe within specific gain groupings of the 908's and 909's as there is between the 908 and 909, I am not too concerned about channel mismatch. Same for the 2sc devices. With this approach, I can have fairly stringent matching within a channel, with the Vbe specs within 2-3% and the hfe within 10%. And considering the wildly variable Vbe and hfe levels I found within any given gain grouping of these devices, if the service manual prescribes only maintaining the same gain grouping, I am probably somewhat ahead of the game. Not perfect, but it seems that things will fall within a range that would have been acceptable at the time it was built. And if it turns out that I have been wildly optimistic, I can start over with contemporary MJ devices.
Now, it will be time to check the driver transistors...