Dumpy filter

Do you need a load resistance as low as 8R? I might have though any load to maintain the output in a 'class A' regime would work?

It is a lot simpler than that, and can be done on a cheapo microcontroller. (Still slightly more expensive than a simple analogue solution, but with digital you get more flexibility and can filter in any way you want.) I can't imagine why you want an ASIC to measure a 1 kHz PWM signal.

Yeah probably not. An 8 ohm load at the DC output voltage positive limit will likely be pushing the max power dissipation of a SOIC or DIP package, assuming the die definitely won't be at 25 degrees C and taking temperature de-rating into account - I haven't seen any datasheets for this old chip that provide temperature de-rating curves for it, they don't expect it to be used this way I suppose.

The datasheet says it can drive a 4 ohm load but I'm sure that implies AC audio signals only

g:

utput

as

tal number crunching to turn that into a number, and a D/A converter to pus h out that voltage over the next cycle. If the period of waveform being loo ked at is stable, you can use a phase-locked loop to lock the period-counti ng clock to some (high) multiple of the frequency of the waveform being loo ked at, and save yourself the digital number crunching.

Integrating into a capacitor for the appropriate portion of the duty cycle may look obvious and easy, but it's a can of worms.

org:

e:

output

t as

gital number crunching to turn that into a number, and a D/A converter to p ush out that voltage over the next cycle. If the period of waveform being l ooked at is stable, you can use a phase-locked loop to lock the period-coun ting clock to some (high) multiple of the frequency of the waveform being l ooked at, and save yourself the digital number crunching.

e may look obvious and easy, but it's a can of worms.

Thinking about it, you probably need to use one cycle of the mark-to-space waveform to zero your capacitor, the next to charge it up for the mark peri od of waveform, and the next to hold the voltage so somebody can read it.

So you end up with three capacitors, and three constant current sources, an d a three way multiplexer to present the stable voltage on each of the thre e different capacitors to the observer who needs it.

Sadly, the constant current sources need to be matched to the capacitor the y are charging, and the period over which they charge it, so you probably n eed four capacitors and four constant current sources, and four more capaci tors to hold the voltages the determine each of the four constant current.

Each of the four capacitor constant current source pairs would do a calibra tion cycle after three reading cycles, in which the capacitor would be char ged for the full period of the mark-to-space cycle, and that 100% output us ed to correct the stored voltage that defined that constant current.

This does assume that period of the mark-to-space waveform has good short t erm stability (which the digital scheme doesn't) but it could be made to wo rk.

Jim's idea of simple and easy strikes me as complicated and messy.

If it has enough of the right kind of timing registers, and integrates a de cent DAC (which some of them do). The whole point about a Turing machine is that it can do anything - or at least anything computable. The fact that a particular Turing machine could do the job doesn't make it the cheapest or best way of doing the job.

It's going to be cheaper than a general purpose microcontroller, which uses up silicon area to offer the capacity to do stuff that you don't need.

On Saturday, February 3, 2018 at 1:43:10 PM UTC+11, snipped-for-privacy@ieee.org wrote :

te:

e.org:

ote:

DC output

ost as

digital number crunching to turn that into a number, and a D/A converter to push out that voltage over the next cycle. If the period of waveform being looked at is stable, you can use a phase-locked loop to lock the period-co unting clock to some (high) multiple of the frequency of the waveform being looked at, and save yourself the digital number crunching.

cle may look obvious and easy, but it's a can of worms.

e waveform to zero your capacitor, the next to charge it up for the mark pe riod of waveform, and the next to hold the voltage so somebody can read it.

and a three way multiplexer to present the stable voltage on each of the th ree different capacitors to the observer who needs it.

hey are charging, and the period over which they charge it, so you probably need four capacitors and four constant current sources, and four more capa citors to hold the voltages the determine each of the four constant current .

ration cycle after three reading cycles, in which the capacitor would be ch arged for the full period of the mark-to-space cycle, and that 100% output used to correct the stored voltage that defined that constant current.

term stability (which the digital scheme doesn't) but it could be made to work.

And there's also the question of what happens when the mark-to-space ration goes to 0% or 100% - you haven't even got a narrow positive or negative go ing spike to clock the system.

It's easy enough to deal with at the system design stage, but it is a furth er complication of the stable-output-for-a-single-cycle approach. Jim's cle arly not a systems designer - he gets commissioned by people who have spent time and money working out what they want, in proportion to the cost of ha ving an ASIC made.

Except that you WRONG and IGNORANT as usual.

For a high proportion of my ASIC designs I COMPLETELY re-architect the customer's system design because...

(1) I AM a systems designer

AND

(2) a device-level designer

AND, THUS, I can see solutions that the pure systems designer, constrained by the use of OTS components, can not possibly envision.

The TLSI project that I mentioned in a recent post is a good example.

For those readers with a real project in mind, contact me directly.

I will provide references who can verify my abilities, success rate, and economics.

Have gun, will travel >:-}

(Sorry for responding to a Slowman post... I shouldn't waste my time, or yours :-( ...Jim Thompson

But he doesn't say how or why. As usual.

Perhaps, but somebody else did the system design for you, and you got to offer alternative realisations. The business of sorting out the data available and how to use it is always outside your scope.

I wonder what that acronym stands for. Google offers Telephonic Large Scale Integration, and the telephone system is one where the basic system design was done a long time ago.

Or advertise the sloppines of your thinking.

The LM386 design is little changed from the older LM380 - I got two samples LM380 in 1974 but AN-69 describing it was dated Dec 1972. So basically a 45+ year old design still selling well.

piglet

I like to use it for all sorts of things other than driving a speaker

As most microcontrollers do, even the cheapo ones.

Some do. Even if they don't, a simple serial DAC is small and cheap (many, many orders of magnitude cheaper than making an ASIC, and vastly more flexible and responsive than an analogue filter). And even a PWM output with a simple analogue filter will give you a better quality filter with more flexibility than you'd get from pure analogue.

That is entirely true - and utterly pointless.

No, an ASIC is not going to be cheaper - even if you are making millions of these things. Even the simplest ASICs have a minimum cost for the packaging, testing, etc., and I think you'd have a hard time beating the price of the smallest and cheapest microcontroller that could do the job. For 100K units, you'd be talking about perhaps $0.30 for the microcontroller. For an ASIC production to get down close to that unit price, you would need to be investing millions in NRE costs. Do you really think Bitrex is interested in the absurdly high numbers needed to make an ASIC the cheap choice here?

This is a "simple analog filter" being fed with a PWM output from an DAC, lol

I've forgotten what we were talking about!

Confirmed. It's right there on the yellowed pages of my National Semiconductor _Linear Applications Handbook_ that sits on a nearby shelf. (It's always sat on a nearby shelf within arm's reach.) OTOH the pages of my National Semiconductor _Audio/Radio Handbook_ remain white. The _Audio/Radio Handbook_ starts to talk about the LM380 on page 4.5 and uses (without attribution) a lot of the material presented in AN-69. In the end, Wikipedia's wrong and we're right. It's my sad duty to proclaim caveat emptor Wikipedia.

Thank you,

For some definitions of "little changed" that is to say there are several significant differences between the LM380 and LM386.

Exactly as pointless as your point that you could do the job in a microcont roller rather than an ASIC.

Bitrex was prefectly happy with his Dumpy Filter. It was Jim Thompson who t hought that a cycle by cycle staircase representation of the mark-to-space ratio would be better, and hard to design (so that only half-a-dozen of us could do it). I sketched out a digital solution - and because Jim designs A SICs I couched in it terms of an ASIC.

Jim then went on to complain that a solution with a hgher analog content wo uld be better (but didn't produce one). I sketched out something for that, but found it messy. Jim didn't come up with anything simpler.

Neither have you.

My own design history would bias me towards as digital solution mostly real ised in a cheap programmable logic device, but microcontrollers sell in muc h larger volumes, and I agree that the right microcontroller would probably be the cheapest solution in most situations.

Jim Thompson's branch of the thread wasn't about cheap but rather about alt ernative approaches.

It doesn't make sense to worry about price until you've worked out what you need and want to do. You may end up compromisng on what you want to do if your first thoughts leave you stuck with a too-expensive solution, but syst em design is all about working out what you can and should do, followed by thinking about how you might do that.

piglet threw a curve ball at me when the LM380 morphed into a LM386 "under the radar" so to speak. (No worries piglet.) My yellowed _Linear Applications Handbook_ also contains Linear Brief 29 "Low Cost AM Radio System using the LM1820 and LM386" on page LB29-1. Papanicolaou and and Mortensen published LB29 on February, 1975. By my math that's almost a decade earlier than the LM386 release date cited in Wikipedia. In the end, despite any curve balls, Wikipedia's wrong and we (piglet and me) are still right. So it remains my sad duty to proclaim caveat emptor Wikipedia.

Thank you,

Why not try editing the page? "Encyclopedia anyone can edit" and all that.

ok_

to

Years ago I did that, fixing some really silly errors on a page. An editor reversed them all right away, insisting that because he had found some barm y book claiming that nearly all electronic profucts were based on valves in the 1980s, they must be and he 'corrected' that point and similar others, for some deluded value of corrected. Wiki really needs to grow a pair and b an idiots like that. It needs some resident subject experts who can ban the very out there & destructive individuals, but not overreach so as to inter fere with the legitimate editing process.

NT