14 December 2011

App Note 70 part 2

Appendix A discusses the history of low-noise power conversion. He starts with the high-voltage CRT supplies in the Tek 454 and the Tek 7904 (Figures A1 and A2). As he says, "Designing a 10,000V output DC/DC converter that does not disrupt a 500MHz, high sensitivity vertical amplifier is challenging." These designs use sine-wave drive to the transformers to keep the harmonic energy low. He also revisits his low-noise designs from App Note 29 (Figures A3 and A4).

Appendix B is "Specifying and Measuring Something Called Noise". The appendix starts with a discussion of the definition of noise. As he says,
Actually switching regulator output "noise" isn’t really noise at all, but coherent, high frequency residue directly related to the regulator's switching. Unfortunately, it is almost universal practice to refer to these parasitics as "noise," and this publication maintains this common, albeit inaccurate, terminology. (Footnote: Less genteelly, "If you can't beat 'em, join 'em.")
It seems out-of-character for Jim to accept this inaccurate terminology. Usually, Jim would define the correct terms, and refuse to use any imprecise verbiage. I imagine that he felt that this battle was uphill too steep.

Personally, I am willing to fight this battle. The word "noise" should be reserved for aperiodic, structureless, random sources. True noise is the result of some random quantum-mechanical effect (like thermal noise, shot noise, flicker noise, popcorn noise, generation-recombination noise, avalanche noise, etc.). Other interfering signals (such as hum, ringing, pick-up, hash, oscillations, ripple, spikes, etc.) are undesirable, but they're not noise. Calling all unwanted signal components "noise" is like calling everything that falls out of the sky "snow". (As any Eskimo will tell you, there's a big difference between snow, sleet, hail, freezing rain, grapule, volcanic ash, etc.)

Perhaps he is just saving his wrath, not for bad terminology, but for bad measurements. The field of "noise" measurement is littered with bad practice. His quote from the second paragraph is worth repeating (with the footnotes inline),
It is common industrial practice to specify peak-to-peak noise in a 20MHz bandpass (One DC/DC converter manufacturer specifies RMS noise in a 20MHz bandwidth. This is beyond deviousness and unworthy of comment.). Realistically, electronic systems are readily upset by spectral energy beyond 20MHz, and this specification restriction benefits no one (except, of course, eager purveyors of power sources who specify them in this manner).
(I wish he named names here. Figures B6, B7, and B8 show how insidious some of the lies are!)

The rest of the appendix discusses instrumentation, calibration, and measurement of the noise floor. The preferred measurement chain is shown in Figures B1 and B2, and a table of useful preamplifiers is list in Figure B11.

Also, he again heaps significant praise upon his favorite instruments, "In our work we have found Tektronix types 454, 454A, 547 and 556 excellent choices. Their pristine trace presentation is ideal for discerning small signals of interest against a noise floor limited background." He concludes with a warning for people over-reliant on modern scopes: "The digitizing uncertainties and raster scan limitations of DSOs impose display resolution penalties. Many DSO displays will not even register the small levels of switching-based noise."



Joe Sousa said...

In my practice designing ADC's I can't use "noise" for all unwanted signals, because it would be too confusing. I use "noise" for random intrinsic sources like thermal noise, shot noise and 1/f noise, and use "interference" for other coherent disturbances in or out of the circuit. Harmonic distortion occupies yet a distinct category. Very separate design methodologies are use to attack each of these pests. -Joe Sousa

kdv said...

Thank you. Reading this is a pleasure.