a little filter

I might've got the "inb4" on that thread, if it's the one I'm thinking of (the common initials don't help, eh? :-p ). But I was merely recalling that D-S works just the same for any arbitrary number of bits. A comparator is merely a 1-bit DAC.

Tim

The delta-sigma keeps the error between the instantaneous running average and the desired value to +- 0.5 LSB. Yours doesn't, I don't think.

Cheers

Phil Hobbs

larly, give the constraints. If you are prepared to sacrifice the top and b ottom fifteen slots - 0.3% of a 10-bit range - you can insist on having a c onstant number of transitions, and the PWM process makes works by making on e of the 16 blocks one clock cycle wider (or narrower) as you increment or decrement your output.

oth waveform, my version compared the same fixed level with sixteen progres sively minimally offset sawtooths. How would you make the consequent output "more regular"?

edited to the people who invented them.

pulation -- ?? Modulation", IRE Trans on Space Electronics and Te lemetry, Sep. 1962, pp. 204-209.

le had started putting the hardware into integrated circuits. I do mention using a 20-bit CS5508 sigma-delta A/D converter in the 1996 paper - it's li sted as one of the eight key features of the system described. I did balk a t trying to implement a 10-bit sigma-delta D/A converter for the power outp ut, in part because the extra switching transitions might have cooked my sw itching transistors, and in part because I didn't understand the theory wel l enough (and didn't have the time to try to master it).

nk.

No, it doesn't. It's a different compromise. With real-life filters, keepin g the instantaneous running average to to with +/- 0.5 LSB of the desired v alue is something of an over-kill. And the imperfections of the switches us ually mean that that running average is rather further from the desired out put than 0.5 LSB. They will in John Larkin's application, which doesn't all ow for much switch-settling time

No, they don't. There are exactly two transitions per cycle, so there's no opportunity for the switch to cause nonlinearity except very close to the extremes.

And it isn't overkill, it's what keeps the low frequency birdies down. I understand you're fond of your baby (as well all tend to be), but it's clearly inferior as a fix for the quadratic slowdown of PWM.

Cheers

Phil Hobbs

As if you would know. ...Jim Thompson

ics

If the "on" and "off" times of a switch are different, that introduces a no n-linearity. If the switching time is long enough that it takes more than c lock period for the output level to hit it's final value then you've got a further source of non-linearity (usually quite a bit smaller) which is more obvious at the extremes.

My point was that in John Larkin's application sigma-delta waveforms would be more likely to introduce extra non-linearity than they would be for peop le working at lower frequnecies.

It's clearly inferior - as I've said all along - if you can live with a lot of transitions. It's a half-way house between the simplest possible form o f PWM with two transitions per cycle, and sigma-delta, with lots more, whic h can be inconvenient in specific applications.

It may be my baby, but I'm as aware of it's weaknesses as I am of it's stre ngths. Engineering is all about finding the right tool for a specific appli cation, and knowing when to use a hammer and when to use a wrench.

ping the instantaneous running average to to with +/- 0.5 LSB of the desire d value is something of an over-kill. And the imperfections of the switches usually mean that that running average is rather further from the desired output than 0.5 LSB. They will in John Larkin's application, which doesn't allow for much switch-settling time

Jim-out-of-touch-with-reality-Thompson doesn't seem to have noticed that Jo hn Larkin specified his operating conditions in his original post.

Jim isn't as stupid as krw, but he can still post total nonsense.

No, it doesn't. All it does is create a tiny offset, because the on and off times don't change with the code.

It's just like a piece of coax: If its capacitance is 30 pF per foot, the capacitance of a 20-foot piece is 600 pF plus an additional contribution from fringing fields at the ends. However, if you chop off a foot, the fringing stays the same, so you lose exactly 30 pF. Lots of standard capacitors have been built on this principle.

Cheers

Phil Hobbs

Which isn't there at zero and 100%, so it isn't an offset, but a non-linearity. It also changes a bit - mostly a lot less - as the interval between transitions changes. Fewer transitions minimises this effect too.

That's why the calculable Thompson-Lampard cross-capacitor typically comes with a laser-interferometer to measure how long it is, and your calculable capacitance is the difference between capacitances at two different lengths.

Sadly for your reputation of omniscience, that offset doesn't go away at either end of the travel, so it's a false analogy.

I said "except very close to the extremes". In the absence of supply sag, which nowadays can be made pretty well as small as you like, the instantaneous error of a D-S extended PWM is within +- 0.5 LSB of the PWM. The nonlinearity sets in 1 code from each edge, where the transition goes away. Even there, it's not very large compared with your average DAC.

Right, my point.

No, it's pretty close to an exact analogy, because as you pointed out earlier, any mismatch of the rise and fall times causes an offset. A minor difference is that the fringing is positive on both ends of the coax, whereas rising and falling edges contribute oppositely, but that isn't the essential point.

And if anybody is silly enough to actually think that any mere human being is omniscient, the sooner my example disabuses them of the notion, the better. (We're on the same team there at least.) ;)

But you're still pissing into the wind.

Cheers

Phil Hobbs

That's how Slowman keeps himself hydrated >:-} ...Jim Thompson

In a slightly different context. This is getting tedious.

My first PWM DAC dates from 1974 or 1975 - early enough that the parts cost for the entire circuit dropped from 50 UK pounds to 15 UK pounds from the time I first costed the design to the time I finalised it, because RCA and Motorola started under-cutting one another's CMOS prices. The word's "grandmother" and "eggs" come to mind.

Made in a slightly more pretentious way.

Pretty close but not equal to an exact analogy. It misses the crucial difference between an offset and a non-linearity, so I claim my pedant of the thread prize.

We mostly are.

Which pretty much describes anything posted here. John Larkin isn't going to act on any of it. Not Invented Here rules.

.JPG

ter.JPG

p.JPG

After all your alligator mouthing i have no sympathy for you.

?-)

"a bunch of design" is not in Larkin's forte >:-} ...Jim Thompson

And you matter?

I am always impressd by the enormous range of subjects in which you have no interest.

You're slo-o-o-ow this morning... took you a long time to come up with a retort... and not a very good one at that >:-} ...Jim Thompson

ch.JPG

ilter.JPG

tep.JPG

gives

Of course i do, obviously more than you. Not that you can handle it. Your attempted denigration of me clearly did not work. Tuff petunias.

?-)