From: bill.sloman@ieee.org
On 2/12/2025 3:02 am, john larkin wrote:
> On Mon, 1 Dec 2025 21:35:40 +1100, Bill Sloman
> wrote:
>
>> On 1/12/2025 2:32 am, john larkin wrote:
>>> On Mon, 1 Dec 2025 00:28:19 +1100, Bill Sloman
>>> wrote:
>>>
>>>> On 30/11/2025 4:56 pm, john larkin wrote:
>>>>> On Sat, 29 Nov 2025 11:17:12 -0800, Joerg
>>>>> wrote:
>>>>>
>>>>>> On 11/29/25 3:38 AM, Bill Sloman wrote:
>>>>>>> On 29/11/2025 8:56 am, Joerg wrote:
>>>>>>>> On 11/28/25 1:32 PM, Joe Gwinn wrote:
>>>>>>>>> On Fri, 28 Nov 2025 12:52:07 -0800, Joerg
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>> On 11/28/25 12:45 PM, Joerg wrote:
>>>>>>>>
>>>>>>>> [...]
>>>>>>>>
>>>>>>>>>>> To the surprise of my clients it's the contrary. The most
>>>>>>>>>>> client-shocking redesign was an auto-align circuit for ganged
>>>>>>>>>>> ADC-channels. High speed, high phase accuracy and all that. They
>>>>>>>>>>> had an
>>>>>>>>>>> elaborate time domain method with a fat DSP, lots of code and very
>>>>>>>>>>
>>>>>>>>>> I meant they used a frequency domain method.
>>>>>>>>>>
>>>>>>>>>>> expensive chips used as programmable delay chips. The NRE alone had
>>>>>>>>>>> been
>>>>>>>>>>> humongous. It never reliably converged so the system hung a lot. I
>>>>>>>>>>> suggested to ditch all that and use time domain. This caused an
uproar
>>>>>>>>>>> because I had rocked the boat a lot and usually consultants aren't
>>>>>>>>>>> supposed to do that. "I don't think this can possibly work", "It
won't
>>>>>>>>>>> deliver the accuracy", "It won't converge either" and all that.
>>>>>>>>>>>
>>>>>>>>>>> Yet the boss let me do it. In the end the whole thing dropped from
>>>>>>>>>>> three-digit dollars in HW to under 10 bucks. Instead of expensive
>>>>>>>>>>> discrete-step time delay chips I used inductors, caps and varicap
>>>>>>>>>>> diodes
>>>>>>>>>>> for almost infinite granularity. The DSP became unemployed because
the
>>>>>>>>>>> connected PC could easily handle the computations. It converged in
>>>>>>>>>>> less
>>>>>>>>>>> than a second, always. The NRE was low because it took less than
two
>>>>>>>>>>> weeks of my time and less than a day for the programmer, and we
didn't
>>>>>>>>>>> need an expensive DSP programmer.
>>>>>>>>>
>>>>>>>>> Embarrassing. Were any of the customers design team later
>>>>>>>>> defenstrated?
>>>>>>>>
>>>>>>>> No, they were pretty good. It's the usual phenomenon where, in an old
>>>>>>>> German saying, you can't see the forest because of all the trees.
>>>>>>>
>>>>>>> Brainstorming is designed to get around that to some extent, but if you
>>>>>>> aren't used to thinking outside the box it's difficult to step back far
>>>>>>> enough to get outside the box.
>>>>>>>
>>>>>>
>>>>>> My experience with brainstorm sessions is not good. The results are
>>>>>> often encouraging but then hardly anything of it gets documented and
>>>>>> typically none of it is implemented. All I need is a large whiteboard
>>>>>> or a large piece of paper. Plus coffee or mate (having a mate right
now).
>>>>
>>>> It's a desperation measure to break up some kind of intellectual log-jam.
>>>>
>>>> It isn't going to work all that often. Documenting it is a chore, but it
>>>> does need to be done
>>>>
>>>>> Brainstorming is great, done right.
>>>>>
>>>>> We sign and date our whiteboard scribbles and photograph them and
>>>>> stash the pics in the project notes folder.
>>>>
>>>> Great for people who like visualisation.
>>>
>>> Or are literate.
>>
>> The illiterate can still look at pictures. Being able to write doesn't
>> guarantee that you can organise your ideas into a chunk of text that
>> other people will understand in the way that you want them to.
>> Whiteboard scribbles aren't structured text.
>
> A photo of an FPGA register map with notes is a serious document.
>
> Given a group of smart people, a good brainstorm session is a way to
> add (or multiply) their intelligence.
>
> But the personalities and the attitude have to be right. Some people
> can't brainstorm.
>
> You attitude that most people are dumber than you are would (no doubt
> has) poison a brainstorm session.
It didn't. I don't make a habit of telling people that they are dumb. If
they tell me stuff that is demonstrably wrong I do try to get them to
recognise that they have to change their opinion, but it's a lot easier
to do that in a collaborative fashion in face-to-face encounters.
Some people do dig their heels in, but that doesn't happen in a well-run
brainstorming session.
> We've had junior people, even
> interns, see something or say something that inspired cool ideas. The
> audience must be grateful for all ideas, to encourage ideas.
That's a well-known feature of brain-storming sessions
> "That violates conservation of energy, but there might be something
> there..."
>
> A brainstorm session is a way for a group of people to explore an
> infinite solution space by using parallel processing.
No finite brain, nor any loosely linked collection of finite brains, is
going to explore an infinite solution space.
Brainstorming is much less ambitious - it's just a mechanism for getting
people to think of less obvious potential solutions to the problem that
needs to be solved. There's a well known problem in multi-parameter
curve fitting where the search algorithm latches onto a local minimum in
the search space, and you can need to get the search process to check
out a bunch of more or less arbitrary points to get a feel for how deep
the minima may get.
--
Bill Sloman, Sydney
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
