The researchers appear to have topped their prior "personal best" and produced an even better circuit, which they presented in their later paper.As I've already pointed out, I didn't "jump" to what I thought was a "conclusion", I specifically labeled it as a speculation (thus the word "appear", as most children would have noticed).
However, contrary to Andrew's snideness, it turns out that my speculation was a good one after all. I emailed the authors of the articles in question (gee, why didn't anyone else think of that?), and asked them about the differing circuits in the two articles. Although he's in the midst of traveling at the current time, Matt Streeter was still kind enough to email me back. On that question, he responded:
We definitely ran the cubic problem twice and got different results, and some of the earlier publications have the earlier result. [...] The SciAm article I believe has the latest one, which is better (I believe about twice the accuracy of the patented circuit on our fitness cases, vs. comparable accuracy for the earlier result).So there you have it.
As for your unsupported implication that they had only used 2N3904 transistors in the high-frequency cubic function generator circuit, his reply is as follows:
We ran the cubic problem both with 2N3904/2N3906 and with higher-frequency transistors. I believe we got similar accuracy in both cases.Gosh, Andrew, the researchers were way ahead of you. Who'da thunk it? And it's not surprising that they got similar accuracy in both cases, since while the 2N3904 family of transisters is not recommended for amplifier use over 100MhZ, the same stat sheet you referred to earlier showed a chart of its frequency response, and it was shown as still pretty linear up into the gigahertz range.
Well let's see the real data.
hfe of 0db at 600 Mhz is not response in the gigahertz range.