Better and more meaningful experimentations

When I was an undergraduate at Iowa State in the mid-2000’s, I became fascinated with the research that was conducted under the programs or Vikram Dalal and Gary Tuttle.  My self adopted mentor, Dr. Curtis Sell whom I originally knew through the amateur radio club on campus and also recruited for my tutoring-for-pizza venture, was a very active participant in the microelectronics program.

The joke in that department was that “You get your masters for repairing the equipment and your PhD for using the equipment you repaired”  The equipment consisted of a PDP-11s and other mid-1970’s era equipment that had been likely discarded from prominent semiconductor fab firms that regularly contribute to the college.

Despite the old equipment, much of the work I had witnessed was groundbreaking glimpses into the future.  Photonic bandgap research that gave birth to on-the-die RF channels (waveguides) happened in this lab.  I was fortunate to have surrounded myself with some of the more brilliant graduate students doing dazzling work.

I also got to see the other side.  Those high on the ‘what-on-earth-is-he-trying-to-do’ quotient.  Usually confident, and somewhat self-assured, yet hiding in an office space, strategically arranged so that their desk, monitor or work area is entirely obscured from the vantage point of the doorway window.

It is from one of these people whose experiment I witnessed stuck with me to this day.

Imagine the world of radio.  Terrestrial radio signals almost universally appear deliberately from, or as a consequence of an electronic circuit.  These circuits, and the antennae that radiate signals are almost exclusively made from metals.  Metals allow for a very easy electron flow, so they are ideal conductors, but they also can cast very elaborate shadows (imagining radio waves as visible light) and diffraction patterns.  Modern human beings are well aware of this as we have adapted by learning the ‘cell phone dances’ which include steps such as the spin-around, the stretch-neck-tilt and the circling-wanderer. What you are doing is finding the sweet-spots that exist as columns and rays criss-crossing the area due to diffraction pattern due to building, steel rebar or radio interference.

With this kind of knowledge expected of an ISU Electrical and Computer Engineering Department undergraduate, I was dumbfounded to see an antenna experiment set up incredibly awkward.

Two microwave horn antennas facing each other, the 1/2″ adjustable steel rod that the antenna was attached to extended outwards and in front of the horn.  Observing the optics of the experiment, I was fairly confident that the receiving horn would be seeing diffraction patterns due to the mounting apparatus, patterns that will vary as a function of the changing distances made during the experiment.  When I asked about the validity of the experiment, I was told, “We tried warning him.”

Not a useful RF experiment.
Not a useful RF experiment.

If you are doing an experiment, do not be too confident to ask your peers if they see any flaws that may wreck your experiment.  The Journal of Irreproducible Results exists as a joke, and no, they seldom make calls for papers.  Use skilled and trusted people around you to make your work great, but if even a total goofball makes a suggestion about something being invalid or incorrect, take the criticism with a grateful appreciation and give the comment or idea a thorough analysis.

This semester, I am studying under Dr. Yadav of North Dakota State University in his course of Experimental Design.  The assumptions at this level are that your apparatus is reasonable, but approaches a scientific method in which experiments are conducted.  It is possible to destroy the validity of a perfectly good experiment by using even the wrong techniques during measurement, and your apparatus may be of a highly-elegant configuration.

Randomized ordering of experimental controls and statistical analysis of your data with multiple runs will give you a bank of data where noise can be filtered using statistical models.  It is not good enough the step through your controls in a linear manner, taking repeated data points on each pass, but instead randomize the experiments such that each pass requires changing at least one control.  Do not repeat your data points multiple times with the same control values, instead replicate your data points by performing each experiment as if it were the first pass.  Save the data aggregation for the statistical analysis use the data is collected.