Math Puzzle...

"Linear" and "differentiable" are very different concepts. Tanh is in fact highly nonlinear--a function can't have two different asymptotes otherwise--but as you say it has continuous derivatives of all orders.

Using soft limit functions as opposed to brick walls is often helpful in practical cases, even theoretically, e.g. neutralizer functions used in asymptotic analysis.

Hard limits can confuse algorithms that are sensitive to the continuity of derivatives. But that's pretty elementary stuff that could be dealt with in 2 minutes with a bit of good will and good manners.

Godzilla vs Rodan has been on late-night TV since about the 1960s, and the endlessly-iterated SED version is rather less compelling.

Circuits are fun. How about we actually talk about circuits, rather than using them for stupid dominance games?

Cheers

Phil Hobbs

You know the drill. Speak to your cohort. ...Jim Thompson

I'm not the boss of anyone here except me. Both lurkers and participants (except those whose egos are involved) appreciate on-topic posts, and we all learned some sort of manners at an early age. This sort of crap would be smacked down very rapidly in a well-run kindergarten.

Stupid point-scoring is, well, stupid.

Cheers

Phil Hobbs

I don't see you aiming your statements at Larkin. But he pays you some money, doesn't he ?:-) ...Jim Thompson

Where I come from, even the aforementioned kindergarteners know a trick worth two of that one.

Cheers

Phil Hobbs

Why don't you publicly aim the same statement at Larkin?

(You have received payment from Larkin, haven't you? :-)

I can probably go back YEARS and find you butting in and telling _me_ to knock it off, but never saying the same to Larkin.

So... butt out. Take your sanctimonious BS elsewhere. ...Jim Thompson

If you want your stupid fights to be private, take them off-line.

Nobody's impressed, and AFAICT everyone except those with ego involvement finds them about as interesting as last week's classified ads.

Cheers

Phil Hobbs

I've been posting converter circuits that include equations to compute efficiency. I've defended what they do and why I want to see the results. All Jim has done is call them stupid. He hasn't said what, specifically, he finds wrong with my math.

He also objects to my use of LIMIT to clip startup extremes, again with no rational explanation why.

Note that I generally try to drift back on topic.

Simulation of switchers does leave one with snippets of available time. Maybe I should take up knitting or something.

There does seem to be a disconnect here. Most folks on SED will disagree vigorously with each other and correct others' mistakes in such a way that the conversation continues in a technical vein, but that doesn't seem to be possible for you and Jim just now.

There's such bad blood between the two of you that it reduces the amount of conversational space for everyone, which is a shame. It also invites jackals and hyaenas in to snap at everybody's ankles and make it worse.

Unless I'm forgetting something important, it's been a long time since Jim posted a helpful clarification or even a well-supported flat denial of something of yours. (Supported by more than "because you're an idiot", I mean.) I'm finding the technical part of this discussion interesting, because as I said a week or two ago, I'm needing to come up to speed on SMPSes myself.

Maybe you could count to ten before pushing JT's buttons, though, just for the sake of the SNR. Mild ridicule IME works a lot better than name calling. Scorn and abuse that falls completely flat is obvious to everyone, and who knows, maybe the hyaenas might starve.

Cheers

Phil Hobbs

I give up. Larkin's equation is NOT correct. And He doesn't defend it, he hand waves. I'd like to see a mathematical defense of his approach, but it's never going to happen.

But Hobbs (paid supporter) likes it >:-} ...Jim Thompson

I'm computing output power over input power. The currents are lowpass filtered, as they will be in real life. The results make sense.

At any point in time, after the startup weirdness, I can check the calc by hand, and get the same number. I can check the power dissipation in individual parts and get nearly the same losses as the calc predicts.

What's *specifically* wrong with that?

He never said that he likes my circuit. I suspect he doesn't. Rob doesn't much like it either.

I like the 'HC14's. They would fit into my 1974 design. So I'm trying to develop an analog model (but not reliant on device-level models)... digital models are crap when inserted into an analog system, like LT's "model"...

• Schmitt-inverter gate

• tpd 41n/15n/12n

• tr 19n/7n/6n

• VT+ 1.18/2.38/3.14

• VT- 0.52/1.4/1.89 .SUBCKT 74HC14 A Y VCC VGND vcc1={vcc} speed1={speed} tripdt1={tripdt} .param td1=1e-9*(15-3-3)*4.0/({vcc1}-0.5)*{speed1}

• XIN A Ai VCC VGND 74HC_IN_S_1 vcc2={vcc1} speed2={speed1} tripdt2={tripdt1}

• A1 Ai 0 0 0 0 Yi 0 0 BUF tripdt={tripdt1} td={td1}

• XOUT Yi Y VCC VGND 74HC_OUT_1X vcc2={vcc1} speed2={speed1} tripdt2={tripdt1} .ends

Nary an analog element in sight :-(

I think I can do it with my TANH-bounded G-sources. ...Jim Thompson

nk.

at's full of

sure of

llowing the

gression.

nce

mit

Spice

sing

t all

s in

tes

ers.

ful in

ed in

nuity

dealt

ps.

and

er

pic

garten.

Perhaps Hobbs has a more or less neutral input in this discussion.

I don't have a MIT degree or whatever Larkin has (so I will probably be fla med down now), but I have run many simulations of switchers that resembles the topology discussed here.

In some cases I have had the same situation as Larkin in which I used the c onverter to boost to a capacitor and draw load intermittent from that node and in that case Larkin is right, that what is important is the efficiency during charging of the capacitor. Low load efficiency is irrelevant and sho rtime, more than 100% efficiency from charge and discharging effects can be ignored in some cases.

On the other hand, Jim is right that the efficiency for a "normal" converte r is calculated by energy in and energy out, so the integration trick is us ed. (the integrated trick can also be used for the charge waveform, but LTS pice does not have the handles for that)

Why not just agree that both Larkin and Jim has a separate points and move on in life? (so the rest of us can read usenet posts without spending time on reading flamewars between two otherwise very knowledgeable guys)

My last input to the discussion would be that why not just use the common b oost converter instead of this new topology instead of inventing (or re-inv enting) a new one (I may have missed a critical point in between the "flame s", excuse me for that)?

Cheers

Klaus

I found the point of peak v(eff) (at around 170us). I expanded that area to a convenient size but including 20 or 30 on/off cycles of switching. I then used the power feature (hover over the target and press alt) to display the power in the source and then the power in the capacitor. I then averaged those two waveforms (using ctrl-waveformName) and recorded their numbers. I divided one by the other and got 85.5% while v(eff) said 107%.

On numerous occasions, using PSpice that is, I have found that I would get different results from a avg(W(device)) equation in the Probe window and avg(I(device)*V(device)). I never had time to investigate deeper in the root cause.

Maybe something similar is going on in the LTSpice engine.

Regards

Klaus

In that case, "efficiency" can be stated as:

Energy delivered to the capacitor / energy taken from the supply.

integrated over the charging time.

Settles to just under 90% in 3ms

The above function in a B source does that.

You can do that with integrals, and behavioral sources. No need for resistors and capacitors.

Integrators and lowpass filters are sisters under the skin:

Integral(sin wt)dt = -1/w cos wt +C

Note the 1/w, that's a 6db per octave rolloff.

Tulane, BSEE, B- overall average. Great party school and nobody checked IDs.

but I have run many simulations of switchers that resembles the topology discussed here.

I could easily do integration, but that will have to be time gated for my various operating modes: powerup, idle/topoff, and firing shots. It's easier for me to graph realtime efficiency and apply that to different situations. The sims are really slow, so the more data I can get in each run, the better. I get a better feel for what's happening with the realtime efficiency graph, too. One big integral hides the details. I want to see battery current vs time, too, so I need to lowpass filter the input current anyhow.

We're considering using LT3420-1, the photoflash thing, but it has concerns. It's only about 70% efficient and its voltage tolerance is mediocre. It has a nice DONE output (fully charged indicator) but that blips during topoffs. If I use that one, I may need to surround it with band-aids. Every part counts when the board has to be 1.5 sq inches.

And I'm a circuit designer. Exploring alternate circuits is part of the process. If all you want is links to manufacturer's reference designs, that's available on EDN or EEWEB.

I like to play with ideas and circuits, and it's fun to do with other people. Some people insist on completed perfection and slap down partially-formed ideas instead of helping to evolve them, or suggesting alternates. It's hard to brainstorm when anyone in the group is mostly critical, and in a public forum you get a lot of that.

Noting Helmut Sennewald's comments...

Subject: Re: Math Puzzle... Here's what to use... Date: Sat, 06 Jul 2013 08:54:13 -0700 Message-ID: References:

You're not supposed to use data compression while doing such a math operation. John Larkin had both compression boxes checked.

I don't know what calculation PSpice uses for "W"... I don't use it. I do know that AVGX(product) gives the result one would expect from eyeballing the circuit, so I've forever avoid AVGX(W) >:-} ...Jim Thompson