Larry Desjardin

's profile
image
President

Larry Desjardin is President of Modular Methods LLC, a consulting company focused on the rapidly growing modular instrument industry. Larry joined Hewlett Packard (now Agilent Technologies), serving in several R&D and executive management positions. As an R&D Manager, Larry received the John Fluke Sr. Memorial Award in recognition of his contribution to the creation of the VXIbus. Most recently, he was General Manager of Agilent’s Modular Product Operation before retiring in 2011. Larry holds a BS Engineering from CalTech, and an MS Electrical Engineering from Stanford University. Larry also writes a regular column “Outside the Box” for Test and Measurement World.


Larry Desjardin

's contributions
  • 01.20.2016
  • For error analysis, it’s hip to be squared
  • All true. If you have just a couple or even a handful of components with tolerances, it may be best to do the detailed convolution of the probability densities of each and calculate the confidence limits of the total as shown above with two uniform distribution resistors. This is particularly true if your major contributor to error has a non-normal distribution. However, there are still advantages with the approach of RSS-ing the standard deviations. The most obvious is that it is simpler, and errors on the side of being conservative. If there are a multitude of small tolerances, it may also be advantageous, because when summed,these will eventually approach a normal distribution where the math is fairly accurate, though still conservative.
  • 01.20.2016
  • For error analysis, it’s hip to be squared
  • I received a response from a reader pointing out that using RSS and the standard deviations on devices that don't have a normal distribution is not technically accurate. I agree, and will place my comments in the Reply sections. Here is the reader's comments: When considering uniformly distributed tolerance errors the author correctly mentions that the standard deviation sigma is the maximum deviation from nominal divided by sqrt(3): sigma = max_dev / sqrt(3) However, he incorrectly considers 3*sigma is equivalent to tolerance. This is only true for a normal distribution --not a uniform one--. As he states earlier in the article, the reason for 3*sigma in the case of a normal distribution is that it guaranetees that 99 % of the resistors lie within +/- 3*sigma, i.e., this is the 99 % confidence interval. But in the case of a uniform distribution, 100 % of the items are contained within the limits of the deviation, i.e., sqrt(3)*sigma and not 3*sigma (of course +/- 3*sigma also contains the whole population, but it is unnecessarily too conservative). The application of RSS yields, for 2 resistors in series, sigma = sqrt(2/3)*max_dev = 0.82*max_dev. Application of the 3*sigma rule would yield, again, a grossly oversized confidence interval: 3*sqrt(2/3)*max_dev = 2.45*max_dev while +/- 2*max_dev encompasses the whole population of possible series combinations of two randomly selected resistors. In this case, the probability density function (PDF) is triangular, starting at -2*max_dev, reaching its maximum at 0 error, and falling down again to 0 at 2*max_dev. (NOTE: the PDF of a sum of random variables is the convolution of the PDF's of both random variables) The 99 % confidence interval can be proven to be +/- 1.8*max_dev (just integrate the triangular PDF until an unknown value such that the probability equals 0.99). This is 2.2*sigma instead of 3*sigma.
  • 10.03.2017
  • New optical interface standard aims at 5G
  • Please see my reply to another question, as I have more details from Guzik. The port can, indeed, be used for instrument control. This requires putting an ODI device into the control PC, such as the Guzik FOBC v2 (Fiber Optic Bridge Card).
  • 10.03.2017
  • New optical interface standard aims at 5G
  • I spoke to Guzik about this. Yes, indeed, this port may be used as overall control. However, it doesn't use PCIe as it's protocol, so it's not a technical substitute for MXI. It uses a not-yet-announced- protocol. The purpose is to allow the equipment to work in environments where there is no PC, embedded or nearby. The advantages are electrical isolation, distance, and speed.
  • 10.03.2017
  • New optical interface standard aims at 5G
  • The digitizer still has a normal control plane interface through the PCI Express fabric on the AXIe backplane to the controller. Guzik put another optical port for optional control between ODI devices. That is the extra optical port you see. It is driven from FPGA code internally on both ends, and allows the user an extra axis of freedom in customizing the FPGAs within the digitizer and the processor. This is a custom port, and outside the scope of ODI.
  • 12.18.2016
  • EDN Exclusive: North Pole to host first commercial deployment of 5G
  • zarechem, Welcome back. The article is a collage of different 5G technologies. 5G is simply 5th generation cellular technology, the next wave after 4G LTE. The goal is to get 1Gb/s to the handset, low latency for automation, and much higher total capacity. mmWave is a candidate technology, referring to any spectrum above 30GHz. The advantage to mmWave is simply the huge amount of spectrum available. The disadvantage is that the reflections create a very ill-behaved channel, and mobile communication can become problematic. Massive MIMO is another candidate, and is assumed to occur in today's cellular spectrum or in the unlicensed WiFi bands. I've written extensively this year, questioning what applications will drive 5G, particularly mmWave. I concluded that point-to-multipoint substitution of fixed internet access was the only reasonable match, at least short term. Strictly speaking, this isn't cellular at all. Today, this is the accepted introductory application. Another is very high density applications, such as in a stadium. But this could also be achieved by WiGig, a 60GHz WiFi technology. Anyway, this story combines all the 5G candidates together. You can read some of my 5G articles in the Test Cafe blog for more info: http://www.edn.com/electronics-blogs/4376426/Test-Cafe
  • 11.24.2016
  • Why do users buy test gear? – Part 3: Manufacturing
  • Hello MWagner_MA, Yes, that's true. Think of a pure software product like MS Office. You don't have to load the software and actually edit a Word document with it, which would be a functional test. You merely need good error detection as you mentioned, using CRC or Checksum, to know the disk is good. A good example of selectively loading firmware is automotive ECU (engine control unit) testing. Pure functional testing is no longer used in manufacturing. That is, the tester is not emulating an engine to test the ECU, as is design validation testing. Instead, special test software is used that isolates each component, which then is subject to a parametric test. The testing can be performed much faster that way. The final step is to load the final ECU firmware.