Sounds like they need Taguchi methods.
Then of course, show them what Pease thought of it.
John
That fad passed through Harman many years ago and one HR lady even managed to get the State of Calif. to pay for a series of seminars. A good deal of the pitch was that experts were no longer needed or even desirable. I challenged the lecturer and got a pitying look from him---I and others were obvious relics and should be put out to pasture as soon as possible, replaced by legions of people who didn't know Jack but could sit around and do Design of Experiments and a lot of other things.
Yes, Pease used a wonderful example of a voltage regulator that had no line regulation! But hey if you had a constant-voltage source it worked better by taking parts out. Proof positive you don't have to know anything! He got a lot of flack for those columns from staunch defenders of Taguchi and others.
In another case that I think I related, again legions of no-nothings did experiments with thermal compound amounts, type of insulating material, screw torques---all directed to optimizing thermal contact of power ICs that were failing prematurely. But the root cause was a dictatorial bastard's insistence that a plate be driven down between two heatsink sections with blows from a hammer so it would be tight and not rattle (the plate replaced a label that bothered this little fascist as it seemed too insubstantial). The PCBA carrying the power ICs located them for attachment to each half of the heatsink and flatness was fine; then end plates were screwed on, which didn't disturb the contact. When the plate was driven down the contact became about a line contact. No one had the nerve, or the desire to have their careers abruptly limited by calling attention to the actual problem!
Eventually the guy flew too close to the sun, but it took years, during which time he tried to get rid of me repeatedly, and engaged in lots of slander against me and many others. Since then, I understand that he has gone from company to company complaining that they just don't get it.
I don't believe you can have too much PSRR. The more the better. The problem is that how you achieve it begins to have negative consequences at some point.
Typical minimum design is 5% ripple. So with a 15 volt supply that would be .75 volts. At 2 volts CD level output for 16 1/2 bits noise should be -102. So between the circuit and power supply regulator -94 dB should reduce power supply noise below the resolution of the source media.
With a CRC filter of 4700 uF 100 ohms 10,000 uF you could get away with as little as 37 dB for a preamp. Then you need to be sure your regulator doesn't add noise of it's own.
As the harmonics of a rectified AC source drop off at about the same rate Fletcher-Munson says hearing threshold improves, you really don't get to do any frequency sensitive weighting.
New product I have trouble understanding the need for this in a real professional sound system.
D-fend Loudspeaker Protection | Never worry again about blowing your speakers.
D-fend Loudspeaker Protection | Never worry again about blowing your speakers.
Pretty easy to get PSRR below 1 part in (2 to the 18th power) = -108 dB. Just cascade two regulators that each give -55dB. Whichever one of them has lower noise and better HF response, put him last (nearest the payload). Example: John Curl's LM337 series mode voltage regulator IC (-66dB) in first position, followed by shunt mode discrete voltage regulator card (-60dB) using a power MOSFET as the shunt element, in second position. Wallah, more than 18 bits of negative power supply voltage purity.
I lived in a design/manufacturing/test environment through Taguchi, JIT, and SPC.
It forced me to be wary of the engineers that didn't understand the processes, the methods, the materials, yet thought that applying Taguchi would solve the problems of the world. JIT simply required I squirrel away parts and materials that, if gone, would shut down the production line. (Ford learned that big time with the brake pad fiasco). SPC forced me to actually teach the QA people what testing was and how to do it repeatably.
Every so often, somebody comes along with the magic "dejour" that allows the further dumbing down of the technical staff. Sometimes, it will work. Other times, it ends up shutting the doors.
John
It forced me to be wary of the engineers that didn't understand the processes, the methods, the materials, yet thought that applying Taguchi would solve the problems of the world. JIT simply required I squirrel away parts and materials that, if gone, would shut down the production line. (Ford learned that big time with the brake pad fiasco). SPC forced me to actually teach the QA people what testing was and how to do it repeatably.
Every so often, somebody comes along with the magic "dejour" that allows the further dumbing down of the technical staff. Sometimes, it will work. Other times, it ends up shutting the doors.
John
I had a, to me at the time, jaw droping experience with a "Quality Engineer" wanting to know the "process capability" of a thermal noise limited measurement - he didn't even blink when I mentioned Johnson Noise, Gaussian distribution... just insisted that he had to have an absolute number for the worst case measurement error
really made me doubt his engineering background when I was unable to pry any hint of the standard's required statistical confidence level out of him
really made me doubt his engineering background when I was unable to pry any hint of the standard's required statistical confidence level out of him
Last edited:
I had one QA guy insist that I install a known bad part to prove that it caused the effect. The line was taking boards with still-charged bulk caps from a previous test and setting them on a conductive surface and making contact to a small-signal PNP, which would fail shorted.
I told the guy he could find someone else to do it. Perhaps I was being unreasonable, but I drew an analogy to inserting a broken light bulb back into a socket and powering it up to prove it was broken.
The guys who led one of the seminar things at Harman were out on the street and starving (figuratively speaking) in a couple of years. I still have the ugly blue coat with a logo that I was given for those days, better than the proverbial T shirt
I told the guy he could find someone else to do it. Perhaps I was being unreasonable, but I drew an analogy to inserting a broken light bulb back into a socket and powering it up to prove it was broken.
The guys who led one of the seminar things at Harman were out on the street and starving (figuratively speaking) in a couple of years. I still have the ugly blue coat with a logo that I was given for those days, better than the proverbial T shirt
With the incredible bonus that current can be delivered right where it is needed, so that inductance can´t ruin ripple performance.
The other problem with QA is they want a manufacturer of a part to show a distribution function for, say, base-emitter voltage. Whatever the fashionable metric is at that point, 6 sigma etc., the assumption is that with enough samples you might find that it was 100mV or 1 volt.
The other difficulty is explaining CYA specs and how some are related to test time. You can't spend the time measuring 2 cent transistors for reverse base-emitter leakage current when it is typically in the low-pA range, so the spec often says something like 1nA max. But if I design something for closer-to-reality behavior, I can be excoriated for being dependent on happenstance.
The other difficulty is explaining CYA specs and how some are related to test time. You can't spend the time measuring 2 cent transistors for reverse base-emitter leakage current when it is typically in the low-pA range, so the spec often says something like 1nA max. But if I design something for closer-to-reality behavior, I can be excoriated for being dependent on happenstance.
There is a bit more to do with shielding and layout to make sure you actually get there. Then there is the difference between the specs on the LM337. The ON data sheet shows more like 42 dB in the critical mid audio bands of interest.
http://www.onsemi.com/pub_link/Collateral/LM337-D.PDF
But CR2C Series pass regulator followed by a decent shunt should get you there with low noise and control of the current paths.
We did have one outstanding QA senior guy and I respected him deeply. He didn't need the money as he'd inherited several million, and with that cushion he refused to lie for the aforementioned dictator, much to the latter's extreme consternation.
I asked Dennis if Harman was the worst place he'd ever worked. No, he reckoned it third. The previous one, number two, was a capacitor company J, who hired him to assemble QA window dressing and ignored real issues. The worst was a company who falsified reliability data, again on capacitors, which resulted in injuries and deaths. He said "I helped put those executives in prison".
It was wonderful to hear Dennis argue with the Harman Electronics VP of Quality and criticize his "goal-post mentality". Having F*ck You money is a marvelous thing.
I asked Dennis if Harman was the worst place he'd ever worked. No, he reckoned it third. The previous one, number two, was a capacitor company J, who hired him to assemble QA window dressing and ignored real issues. The worst was a company who falsified reliability data, again on capacitors, which resulted in injuries and deaths. He said "I helped put those executives in prison".
It was wonderful to hear Dennis argue with the Harman Electronics VP of Quality and criticize his "goal-post mentality". Having F*ck You money is a marvelous thing.
- Status
- Not open for further replies.