>>I finished UCSD in 2 1/2 years at age 19 with a degree in Molecular Biology. The following year I was attending grad school at SDSU in pathogenic bacteriology. I decided to do something different. I’ve spent thousands of dollars at bookstores. That netted an Extra Class ham license, First Class Radiotelephone license, a teaching credential, a pilot’s license and extensive experience in electrical engineering and computer science. My colleagues today are people with PhD’s in math.<<
Wow, you can’t be a US citizen if you are that smart! (Just kidding.) As I was trying to tell Top Quark, the story of Robert Noyce (which sounds like your story) is also worth reading. It is found in the Tom Wolfe book “Hooking Up” in the essay “Two Young Men who Went West.” Both your story and Noyce’s are great examples!
I got two undergraduate degrees and never got a postgraduate degree, but for me, working at IBM and Apple on operating systems and device drivers was just as “educational.”
I redesigned the CPU scheduler for the UNISYS/7000 (Tahoe), fixed the MPCC drivers, X.25 and X.29 link layers and the async device drivers. Previous to that I had fixed up the segment swapper on the UNIX-1100 platform as well as fixing the interfaces to the GCS for async dialup users.
When I moved to my current employer, I was drafted to port the Mentat SYSVr4 STREAMS into the HPUX-7 kernel on the 68040. Mentat provided the original source to the OSF. I also created a tunnel driver between the BSD and STREAMS stacks to allow native applications to pass onto the new STREAMS stack. I back-ported the HPUX-9.0 multi-LUN SCSI driver to work on the HPUX-7 kernel to permit installing the Spider DLPI (mapping X.25 to IP), X.29 and X.25 drivers and ultimately a proprietary device driver for the military to interface to a GE X.25 switch. That was a 6 week marathon effort. Lots of fun, but very tedious to single step the kernel through the device driver code. I had a particularly interesting conflict between the Mentat scheduler and what Spider expected in a STREAMS scheduler. Tracking that down was a 1600 step process to identify an order of execution problem and modify it to be fully re-entrant. Running kdb on a kernel is primitive.
My DSP work today is focused on special compression techniques. The railroad contract includes DSP aimed at vibration signature analysis on railcar bearings. The key objective on the railroad task is to drive all the electronics down to board level custom embedded systems that are dirt cheap and rugged enough to handle the pounding that comes with the rail environment. The vertical accelerometers are rated +/- 80g @ 50 KHz. I'm going to try to save money by sampling 2 channels with a stereo DAC connected with the SPI bus to a TI DSP chip. That approach will be much cheaper than the current Diamond Systems DMM32 at $750 a pop. The DSP chip will do the FFT of the samples far more efficiently than the general purpose PII on the CPU board. Cheaper too.
Don't fret about missing the postgraduate degree. Every green college grad has to learn the ropes at a given employer. Everyone has a different focus and set of expectations. It is important to keep your skill set current so you are marketable as the next big processing fad comes by. AJAX and SOA are hot buttons right now. They useful arrows in the quiver, but my customers are paying for quality DSP ahead of flashy "pop" technology right now.