Sciencemadness Discussion Board
Not logged in [Login ]
Go To Bottom

Printable Version  
 Pages:  1    3    5  ..  9
Author: Subject: PC PSU to laboratory PSU
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 22-10-2007 at 16:01


Well as power supplies go,



Nuff said. You will not match the simplicity, parts count, performance and elegance of this design. Stop trying to think you can; you have shown time and time again that you cannot.




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
Twospoons
International Hazard
*****




Posts: 1201
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline

Mood: A trace of hope...

[*] posted on 22-10-2007 at 17:32


Quote:
Originally posted by Xenoid

Ahhh.. Twospoons, that was you was it! I had my eye on a constant current 60Amp HP supply on "Trade Me" but ummed and aaahed a little too long! ..


You snooze, you lose on Trademe. Buggered if I was going to let that one slip past!

@Rosco
Quote:

the spice simulator isn't sophisticated enough to give it a fair rendering


Or maybe you are not using it properly, or don't understand the results. You would be better with a transient analysis
- that will show up instabilites, and you can try things like line and load step responses


Quote:

The spice models for linear region operation of mosfets are pure crap too so that may be part of the simulation problem


Now thats just silly. SPICE only exists because modelling of fets, mosfets, bjts etc was needed by the semiconducter industry in order to design chips with thousands of devices. The device modelling is based on semiconductor physics, not a few textbook equations for gain. A complete BJT model has over 30 parameters IIRC.
If I get a screwy result from a simulator, my first assumption is that it is ME that's got something wrong. Garbage in - garbage out.

BTW that STM example of the fan controller looks like a PWM control at first glance (I haven't botherd to analyse it).




Helicopter: "helico" -> spiral, "pter" -> with wings
View user's profile View All Posts By User
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 22-10-2007 at 21:27


Elegance ??? Hmmmm ......

Let's look at just a few of the things that are wrong with this design for now . Plenty of time later for all the other things :P

On the 12 volt range you have voltage control pot response from a minimum of 0 volts to a maximum of 6.2 volts.

On the 5 volt range you have a voltage control pot response from a minimum of 0 volts to a maximum of 6.2 volts.

On the 3.3 volt range you have a voltage control pot response from a minimum of 0 volts to a maximum of 6.2 volts .

I suppose the theory there was one size fits all or one size fits none at all , with the latter of those two possibilities being the design outcome . I fail to see
anything satisfactory about that scaling .

Maybe you could get around needing a switchable reference and get linear proportional full range response on your control pot ....but not like that .

Your op-amp supply ground in the Constant Current Modules needs to go to true ground or else you need to move your meter shunt to the other side of the array
so you don't create power rail noise rejection issues for your modules . Actually something better is to
use a local regulated supply at 10 volts gotten from the 12 volts , which provides a cleaner supply and noise immunity for the op amps and mosfet drivers .

What about having some status indicators for current and voltage mode ? Can't really watch for the voltmeter dip
since there isn't one .

As for performance , well that's an open question for your
one op amp per module solution , using that low of a low resistance current sensing shunt with one op amp .
There are stability concerns about the voltage controlled current sink configuration since the capacitive load of the Mosfet gate is a sort of worst case scenario load for an op amp . What is stable there with a 0.1 ohm shunt isn't necessarily stable with a 0.01 ohm shunt and which mosfet is being driven has a bearing there also .

IMO it is better in a VCCS module having a low loss , low resistance shunt to use a dedicated gain amp for monitoring the shunt voltage , another dedicated error amp , a unity gain buffer low pass filter driving an NPN source follower as a dedicated driver stage for the Mosfet . This arrangement seems to be what is required for the circuit to "hunt quietly" in a self dampening and rapid settling , stable servo lock fashion .

Be very clear on something , I am definitely not trying to match your work there nor have I been trying to think I can , nor do I even want to try ....why would I , or why would you even think that ? Is this supposed to be a farting contest or a legitimate technical discussion ?
View user's profile View All Posts By User
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 22-10-2007 at 22:12


Quote:
Originally posted by Twospoons
@Rosco

Quote:
the spice simulator isn't sophisticated enough to give it a fair rendering


Or maybe you are not using it properly, or don't understand the results. You would be better with a transient analysis
- that will show up instabilites, and you can try things like line and load step responses


Been there , done that .......understand what I see .
You will sit for many long days setting and testing and noting results of different transient analysis scenarios , and still not catch the instability that an AC large signal , medium signal , signal sweep will turn up . Indeed you can NOT trust a Bode plot on small signal AC analysis when the voltages and results are contradictory , with phase being respected . I have spent many happy hours in front of real waveform analyzers and signal generators , and I know when a computer simulation is feeding me unbelievable bullshit as results . You can look at the voltage levels on the gate of a mosfet in a simulation and see when it is way below threshold , there should be no output . Just like you can look at signal levels on the inputs of an amplifier and tell
that something is wrong when the output is exactly the opposite of what it should be .

Quote:

Quote:

The spice models for linear region operation of mosfets are pure crap too so that may be part of the simulation problem


Now thats just silly. SPICE only exists because modelling of fets, mosfets, bjts etc was needed by the semiconducter industry in order to design chips with thousands of devices. The device modelling is based on semiconductor physics, not a few textbook equations for gain. A complete BJT model has over 30 parameters IIRC.
If I get a screwy result from a simulator, my first assumption is that it is ME that's got something wrong. Garbage in - garbage out.


I'll guarantee you the modeling was done for "normalized"
mosfet parameters pertinent to switchmode applications ....not linear region applications .

Here's one real life factor in linear region operation of mosfets ....*variable* transconductance , and its a hell of a lot lower than the fixed figure even perhaps 20% or less of what is stated as "minimum transconductance" , but the sims don't render linear region gate voltages accurately , not even spice 3 models cover it . What the sim generates looks nothing like the data sheets , but a great oversimplification with greatly flattened response , more than a volt to 2 volts away from what the manufacturers own data sheet says about threshold voltage and linear region in comparison with the same manufacturers spice model . The models are probably authored by some third party corner cutter or they would overlap perfectly with the data sheets :P

Quote:

BTW that STM example of the fan controller looks like a PWM control at first glance (I haven't botherd to analyse it).


Nope it's pure linear .
View user's profile View All Posts By User
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 22-10-2007 at 22:34


LOL, you didn't happen to realize that the gate threshold voltage varies by perhaps 4 to 6 volts in typical units, did you? Sounds to me like that model is just fine!

Re: my circuit;
I don't recall if more than 6V was even of interest. If so, then the 20k resistor can simply be removed. The control will then be linear from 0 to 12.4V (give or take actual zener voltage), so your voltage setting remains the same if you switch the jumper.

Indicators are external and any responsible scientist would have a proper precision voltmeter and ammeter on his experiment, regardless of the power supply in use. A light could be added to indicate constant current mode, with a comparator, trimpot, LED and resistor. This too is almost external. Performance will be excellent inside the bandwidth of this circuit, which should be at least 20kHz. (Any residual switching noise from the supply can be filtered with differential and common mode chokes and some capacitors.) None of your other points have any validity so I need not address them.

Tim




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 22-10-2007 at 23:47


Quote:
Originally posted by 12AX7
LOL, you didn't happen to realize that the gate threshold voltage varies by perhaps 4 to 6 volts in typical units, did you? Sounds to me like that model is just fine!


You don't really know too much about mosfets do ya?
1 volt variation maybe , 4 to 6 , you must be kidding .
And what about the transconductance value , it is definitely generally expressed as a saturation value plateau derived figure not even resembling the true much lower and slowly increasing figure for linear region operation .

Quote:

Re: my circuit;
I don't recall if more than 6V was even of interest. If so, then the 20k resistor can simply be removed. The control will then be linear from 0 to 12.4V (give or take actual zener voltage), so your voltage setting remains the same if you switch the jumper.

Indicators are external and any responsible scientist would have a proper precision voltmeter and ammeter on his experiment, regardless of the power supply in use. A light could be added to indicate constant current mode, with a comparator, trimpot, LED and resistor. This too is almost external. Performance will be excellent inside the bandwidth of this circuit, which should be at least 20kHz. (Any residual switching noise from the supply can be filtered with differential and common mode chokes and some capacitors.) None of your other points have any validity so I need not address them.

Tim


Excuses , excuses :P

I suppose it would be pointless also to note that bandwidth is a somewhat ambiguous concept with regards to this regulation scheme , so no bandwidth figure having any general meaning can actually be cited . You see the full power bandwidth is a function of where the current limit is set in relation to how much baseline current is flowing , and as the power is increased towards the level where current limit is approached , the bandwidth decreases to zero or unity as for DC .......however you wish to characterize that zero hertz bandwidth . So you have to characterize bandwidth as a function of percentage amplitude modulation (regulation) above and below any given voltage level , like plus or minus 1% @90% full current output for example .

Hey , I know ...you can just say it regulates good , or it regulates poorly ....and for bandwidth , well whatever :D

You will be lucky to get a stable sweep from zero to full current at rated voltage to even 1 kHz driving the power mosfet directly with an op-amp . But you might get it up to 2 or 3 kHz using the strategy I suggested ....but thats probably an outside figure , and half of that is more likely for either scenario . You won't see much bandwidth in linear operation for these power mosfets which have a lot of capacitance , because of the peaking / overshoot which appears as you try to drive them faster . They are sluggish unless hit with a hard and fast rising drive signal and then they overshoot the mark so you end up having to tradeoff speed for accuracy and stability for linear operation . Otherwise the mosfet will go into the porpoising routine of oscillation unless you slow it down . A ripple generator is easily achieved here , as in big ripples .

Thinking yours won't oscillate may be just so much wishful thinking .
View user's profile View All Posts By User
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 23-10-2007 at 09:13


Oops, worded misleadingly: should read "varies from perhaps 4 to 6 volts". Example: STW11NB80 Vgs(th) (Id = 250uA, Vds = Vgs) ranges from 3 to 5V, for a range of 2, or 4 +/- 1 V.

FETs are notorious for irregular Id and Gm parameters, both due to manufacture and variation with temperature.

Your further objections are again invalid and needn't be addressed.

Tim




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 23-10-2007 at 10:09


Yeah a range of 2 to 4 for threshold is more like it .
And it's not just that the Gm varies from unit to unit ,
but it varies in the same unit depending on the level of saturation .

You say my other objections have no validity , maybe you would expound on that and say exactly where .

Since there seems to be little agreement so far ,
how about this ....would you agree that in this particular
and peculiar sort of circuit configuration , that really no spice AC small signal simulation is trustworthy as any substitute for actual breadboarding and a real instrument analysis of the actual physical circuit ?

Otherwise we could talk about theory forever here and both be wrong . I have put some of these things through their paces in spice sims , and gotten good DC solutions , but garbage results for AC and transient response .
View user's profile View All Posts By User
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 23-10-2007 at 12:23


I don't even use SPICE.

Many people do use SPICE, and get excellent results from it. It's definetly the professional tool to explore things without a dusty old breadboard. Winfield Hill, author of Art Of Electronics, uses it, among other professionals I know. Others use it and get excellent results, only to discover that their circuits never work in reality. Still others never get any simulated results from a circuit which works just fine in reality. Legendary IC designer Bob Widlar has shaken a breadboard in one hand, yelling "according to SPICE, this circuit doesn't work!". You can talk to them yourself if you need hints and tips on SPICE.

I can analyze this circuit quite fine from here, anyway.

At worst, three phase shifts -- two op-amps and the 220 ohm gate resistor and the gate capacitance -- conspire to create a phase-shift oscillator, probably in the 50 to 200kHz range. Just where depends on the op-amps used. If the voltage error amp is fast, I can guarantee you this circuit will oscillate. That's your fault for irresponsibly using an op-amp too fast, or too undercompensated for the loop it's in.

A typical compensation scheme would be a output-to-(-)in capacitor at the op-amp, adding a series resistor to that input if its impedance isn't high enough (as in the CCM) or constant enough (as at the reference pot) to assure compensation. Another approach is to add loop NFB to the op-amp locally, via resistor, to limit the amount of NFB delivered by the external loop.

The error amp should be about 5 times slower than the CCM, so the CCM should use a reasonably fast op-amp with good drive capability and slew rate. Probably, a regular TL074 could be used, with an emitter follower to provide suitable gate drive. A more powerful op-amp with the same general specs would be more suitable, I just don't know any offhand. Full signal bandwidth can be expected from DC to 100kHz or so.

If the MOSFETs chosen are IRFZ44N, the "miller charge" is 23nC, or over 2 microseconds, a constant current of less than 15mA (since I == dQ/dt). At 10V, a typical gate resistor would then be about 10/0.01 = 1kohm or less, so I wrote 220 ohms, which will supply ample gate current to the MOSFET in less time than the op-amp will notice. Some compensation may still be needed, which as I've indicated is a simple adjustment.

With the CCMs in the 100kHz range, the error amp should have compensation down to the 10-20kHz range. A jellybean LM358 could be used, which is slow enough as-is that it will probably be stable with no changes.

Tim




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 23-10-2007 at 15:12


Quote:
Originally posted by 12AX7
I don't even use SPICE.

Many people do use SPICE, and get excellent results from it. It's definetly the professional tool to explore things without a dusty old breadboard. Winfield Hill, author of Art Of Electronics, uses it, among other professionals I know. Others use it and get excellent results, only to discover that their circuits never work in reality. Still others never get any simulated results from a circuit which works just fine in reality. Legendary IC designer Bob Widlar has shaken a breadboard in one hand, yelling "according to SPICE, this circuit doesn't work!". You can talk to them yourself if you need hints and tips on SPICE.


From what I have seen of it so far , spice is good for initial layout and component selection and confirming DC solutions for various selected conditions , like a snapshot of control loop instantaneous values ...but it is inept at simulations which are dynamic across time and frequency sweep changes ......unless it involves a particular templated circuit arrangement for which it has been specifically debugged and tested in advance . It gets lost easily and doesn't have any automatic self-check on the integrity of the lies it starts telling .

Quote:

I can analyze this circuit quite fine from here, anyway.

At worst, three phase shifts -- two op-amps and the 220 ohm gate resistor and the gate capacitance -- conspire to create a phase-shift oscillator, probably in the 50 to 200kHz range. Just where depends on the op-amps used. If the voltage error amp is fast, I can guarantee you this circuit will oscillate. That's your fault for irresponsibly using an op-amp too fast, or too undercompensated for the loop it's in.


Below a few kHz the phase shift is negligible for lightly loaded op-amps where the load is purely resistive , but you are right about the gate capacitance being a problem , and it is only one of several problems which aggravate the situation there with the mosfet . There is other capacitance about that mosfet and the way it responds is current dependent , so what compensation is worked out for one operating point isn't effective at a different operating point .

You have it exactly backwards with regards to the speed requirement for the voltage error amplifier as compared to the power element . The error amp needs to be fast so it can sample and correct on the fly several times while the power element is still in transition . Then a buffer stage having a low pass characteristic averages and derives a correction signal at speeds which can be followed by the power element . That buffer response then establishes your maximum bandwidth .
Quote:

A typical compensation scheme would be a output-to-(-)in capacitor at the op-amp, adding a series resistor to that input if its impedance isn't high enough (as in the CCM) or constant enough (as at the reference pot) to assure compensation.

Yeah that was a first approach tried for compensation ....but the problem there is you already have about the limit of a capacitive load on the op amp with the mosfet , and now you are going to add *more* capacitance for filtering ...which has
an effect that is quite different from what you may have wanted ......especially for different operating points .
That is precisely what leads to having to use as a buffer an active filter driving an NPN source follower for the gate drive of the mosfet . It is really very frustrating to see what works so perfectly for one load and voltage , go completely to hell for a different load and voltage because of the mosfet behavior changing beyond the capability of compensation
intended to tame it .

Quote:

Another approach is to add loop NFB to the op-amp locally, via resistor, to limit the amount of NFB delivered by the external loop.

It's more than just another approach , you probably have to do both that and the earlier compensation strategy as well .
That is partly why I added a gain stage , early on . But more to the point of what you are saying is perhaps setting a fixed
gain , high but fixed , perhaps 10,000 ? ....for the current error amplifier in the VCCS module ? Maybe .....I think I already looked at that as a limiter possibility , and left it undecided , unresolved . That could do some good .
Quote:

The error amp should be about 5 times slower than the CCM, so the CCM should use a reasonably fast op-amp with good drive capability and slew rate. Probably, a regular TL074 could be used, with an emitter follower to provide suitable gate drive. A more powerful op-amp with the same general specs would be more suitable, I just don't know any offhand. Full signal bandwidth can be expected from DC to 100kHz or so.


You need to give that some more thought :P The usual semantics for bandwidth simply do not apply to this circuit . Maybe if there was a 4 or 5 msec overcurrent delay before current limiting was invoked .......but not so long as it has hardwired instant current limiting .
Quote:

If the MOSFETs chosen are IRFZ44N, the "miller charge" is 23nC, or over 2 microseconds, a constant current of less than 15mA (since I == dQ/dt). At 10V, a typical gate resistor would then be about 10/0.01 = 1kohm or less, so I wrote 220 ohms, which will supply ample gate current to the MOSFET in less time than the op-amp will notice. Some compensation may still be needed, which as I've indicated is a simple adjustment.


Simple for one operating point yeah , but effective across the entire range of output capability .....not so simple .
Quote:

With the CCMs in the 100kHz range, the error amp should have compensation down to the 10-20kHz range. A jellybean LM358 could be used, which is slow enough as-is that it will probably be stable with no changes.

Tim


Yeah an inherently slow op amp could actually have appeal for the VCCS current error amplifier in regards to stability concerns . The slower it slews the less overshoot will
be caused on the mosfet . You don't cure the overshoot
by driving the gate harder and faster , but by driving it gentler and slower .....which gives you stability , but of course kills the bandwidth . See the dilemma ?

Stop thinking microseconds if you want to see a stable control loop .....think milliseconds , and many of those for settling times .

With a 1 ohm sense resistor you might get 100 kHz range ,
but not with a 10 milliohm .....nada no way no how not today :D
View user's profile View All Posts By User
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 23-10-2007 at 17:28


Rosco,

Most of your points are backwards. The loop around the error amp, for instance, must be faster, or else phase shift will approach 180 degrees at some frequency and there it will oscillate. Two op-amps of the same speed both phase shift by almost exactly 90 degrees for any frequency over, say, 20Hz (yes, 20Hz, I'm not kidding, read the datasheets some time), making a loop containing only two completely unstable without compensation. With the cutoffs staggered and gain and feedback limited, the phase shifts do not coincide and it can be stabilized without half-assed patching.

The "hand waving" way to demonstrate this is to note that a faster error amp will see the error, then seeing it growing, and see it not responding, and the feedback signal is therefore much larger than it should be (excessive derivative in the PID term). The slow amp eventually responds by passing this amount to the output, which overcompensates for the original error, causing overshoot and so forth. If you're lucky, it will ring down. If you aren't, it will ring up and oscillate.

Learn control loops some day. I have been playing with electronic loops for years. I have more experience than you.

Tim




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 23-10-2007 at 19:29


Quote:
Originally posted by 12AX7
Rosco,

Most of your points are backwards.

Ya think ?
Quote:

The loop around the error amp, for instance, must be faster, or else phase shift will approach 180 degrees at some frequency and there it will oscillate. Two op-amps of the same speed both phase shift by almost exactly 90 degrees for any frequency over, say, 20Hz (yes, 20Hz, I'm not kidding, read the datasheets some time), making a loop containing only two completely unstable without compensation.


Phase shift is not just some magic number like 90 degrees
"per op amp" , but it is related to gain , frequency , and the sort of load .

If you have sufficient phase shift in a number of series stages , then the phase shifted signal arrives on the feedback return path in matched phase with the input signal on the other op amp input . Common mode rejection
then excludes response to the like phase signal and amplifies the difference . So all the op amp effectively sees is two differing DC level inputs . And the stability of the loop holds ..... out to as high a frequency as the phase remains matched reasonably in sync .
Quote:

With the cutoffs staggered and gain and feedback limited, the phase shifts do not coincide and it can be stabilized without half-assed patching.

Making the phase shifts coincide at the error amp inputs is precisely what you want to do , it's having them go 135-180 degrees out that causes any problem . It's keeping them in sync across a 120:1 range of operating current that is the challenge and no small one .
Quote:

The "hand waving" way to demonstrate this is to note that a faster error amp will see the error, then seeing it growing, and see it not responding, and the feedback signal is therefore much larger than it should be (excessive derivative in the PID term). The slow amp eventually responds by passing this amount to the output, which overcompensates for the original error, causing overshoot and so forth. If you're lucky, it will ring down. If you aren't, it will ring up and oscillate.

Without a buffer , that's right . Think about the way a cruise control works . The driveshaft sensor and error amplifier are very fast . The vacuum solenoid throttle linkage buffer stage is much slower , and then the throttle acellerated vehicle itself is slower responding still . Works like a charm .
The response time hierarchy there is correct for stability .
A vactrol could serve a similar purpose as a sort of DC restorer buffer stage in this control loop , since it would isolate the stages and render phase moot due to the slow response and storage time of the CdS element in the vactrol optoisolator . For a millisecond or so it acts as a sort of "sample and hold" continuously refreshing signal reference .
Quote:

Learn control loops some day. I have been playing with electronic loops for years. I have more experience than you.

Tim


Hell , I'm trying to learn everything I can , but it's hard to teach an old dog new tricks :P

I have about four sample packages of PerkinElmer Vactrols
sitting in front of me , and they are probably older than you are . The ones I have used before came in very handy for
situations just like this where it became necessary to remedy phase and voltage incompatability issues in a control loop with a proven performer from way back .

I think they were originally developed as noise free potentiometers for use as volume control faders in tube audio theater sound systems . No static , no snap crackle pop . Smooooth :D

They probably even have an honest injun spice model for them ......unlike the situation for power mosfets :D

[Edited on 23-10-2007 by Rosco Bodine]
View user's profile View All Posts By User
DerAlte
International Hazard
*****




Posts: 779
Registered: 14-5-2007
Location: Erehwon
Member Is Offline

Mood: Disgusted

[*] posted on 23-10-2007 at 20:57


If any of you are seriously considering making a constant current or a constant voltage regulated supply, ignore all the ranting above and search for manufacturer's application notes on the web. Here's a starter (a modicum of electronic knowledge is needed):

http://www.national.com/appinfo/power/files/f4.pdf

I suggest the above posts between the two antagonists be put into Whimsy - pure whimsy. If you are going to design a feedback circuit please learn the fundamentals as explained in any Elecronics 100 course for engineers, who have to make things actually work. Stop wasting our time.

Regards,

Der Alte
View user's profile View All Posts By User
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 23-10-2007 at 21:16


Quote:
Originally posted by Rosco Bodine
Without a buffer , that's right . Think about the way a cruise control works . The driveshaft sensor and error amplifier are very fast .


I suspect you should go work for Chrysler, then.

Mom's minivan, a 2000-something Voyager or whatever it is, has a cruise control with a response time of about 1 second. It really is quite slow, and I can bet that the engineers at Chrysler played with it and decided that there was enough windage in the slowly-responding engine (computer controlled, bah!), automatic transmission and vehicle weight that it would've oscillated horribly without a severely low pole to let it work out. Now there's something you don't want to see, velocity oscillation on the highway. That would be weird, annoying and in traffic, dangerous. Or even worse, the gas pedal twitches up and down, gotta be bad for the engine.

On the other hand, Dad's Nissan Pathfinder, with manual transmission, has very little windage in the drivetrain, so its cruise control is much faster -- still slower than the vehicle, but because there are fewer phase shifts in the loop, it is able to respond and regulate much better in turn.

Tim




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
DerAlte
International Hazard
*****




Posts: 779
Registered: 14-5-2007
Location: Erehwon
Member Is Offline

Mood: Disgusted

[*] posted on 23-10-2007 at 21:35


@Twospoons - re your post 22-10-2007:

Quote:
Now thats just silly. SPICE only exists because modelling of fets, mosfets, bjts etc was needed by the semiconducter industry in order to design chips with thousands of devices. The device modelling is based on semiconductor physics, not a few textbook equations for gain. A complete BJT model has over 30 parameters IIRC.
If I get a screwy result from a simulator, my first assumption is that it is ME that's got something wrong. Garbage in - garbage out.


I must say up front I am not a great fan of SIMS, but they do have uses, especially for the tedious business of tolerancing, etc - once you are sure the nominal circuit is making sense. GIGO - sure! But sometimes the garbage is created internally by a progam fauly causing out of range values, divide by zero etc.

As to junction transistors using 30 parameters, I doubt it. As for physics, well, most SIMS merely use the equivalent circuit parameters as engineers do by hand - from the data sheet. I'd find it hard to define 30 parameters. And for some uses that may be inadequate. Avalanche conditions, e.g.

Most of the junction transistor theory was around in the 1950s, even before my time! The Early effect was reported in 1952; Ebers and Moll considered large signal behavior in around 1953, IIRC.


Even transistor design itself has been reduced to fairly elementary considerations. It is not handled by the type of program we are talking about here.

What I used to object to is the attempted use of the simulations to actually design circuits rather than merely analyze them. Too many scientists and engineers today use a computer program and not their own brains. A computer program, at best, represents some one else's brainwork and reduces the person using it to a mere operator of a machine, without any talent of his(her) own. And that where GIGO comes in, in force...

Regards, Der Alte
View user's profile View All Posts By User
Twospoons
International Hazard
*****




Posts: 1201
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline

Mood: A trace of hope...

[*] posted on 24-10-2007 at 19:13


In defense of SPICE :

SPICE Model parameters - BERKELEY

I appologise - I was wrong. There are in fact 41 parameters that can be applied in the latest versions of SPICE BJT models.

The model now used is the Gummel-Poon model, an extension of the Ebers-Moll model to cope with high bias levels.
Have a look at the link.

I agree that you can't use SPICE alone to design, but you can use it to verify. Using spice allowed me to succesfully design a laser pulse reciever, that picked up 1pJ, 16MHz pulses, and only needed 150uA Iq. I would never have got the bastard temperature stable without the simulations (at least, not in 3 days!).
I like it, use it, and find it a handy tool in my engineering arsenal.

Not all simulators are built equal either. There exists a set of standard test circuits for evaluating simulator performance. I use Simetrix - which could achieve convergence on a ckt with 17,000 mosfets in it, IIRC.




Helicopter: "helico" -> spiral, "pter" -> with wings
View user's profile View All Posts By User
DerAlte
International Hazard
*****




Posts: 779
Registered: 14-5-2007
Location: Erehwon
Member Is Offline

Mood: Disgusted

[*] posted on 24-10-2007 at 21:58


@Twospoons

Read your reference and stand corrected! But, if you are going to use, say, a generic 2n222 how many of those 41 parameters really apply, and how many are guess work or default values? I am sure manufacturers don't supply models with all 41! I am even more sure that less than 1% of engineers use them either, apart from device designers.

True it's a hell of a long time since I designed anything truly analog (25 years) having become a theoretical type working digital concepts for things like DSP processors, so I have no idea of the versatility of the modern SPICE program. I grew up with h, y, and z parameters and used them if I could get them, otherwise the usual equivalent cct parameters. Matrix algebra was hell using Fortran (I shouldn't even mention I used Algol also!).

But no program beats a good brain, except for speed.

Regards,
Der Alte
View user's profile View All Posts By User
12AX7
Post Harlot
*****




Posts: 4803
Registered: 8-3-2005
Location: oscillating
Member Is Offline

Mood: informative

[*] posted on 25-10-2007 at 08:16


Lots of Onsemi datasheets provide copious graphs of parameters, even all the rarely seen h-parameters besides hfe.

Some, like the 2N2222 and MJE350, among hundreds of others, are never specified for some parameters. Things like fT, hFE vs. Ic, etc. simply may not be in their specification. The 2N2222 is from a wider-tolerance silicon era, after all.

Tim




Seven Transistor Labs LLC http://seventransistorlabs.com/
Electronic Design, from Concept to Layout.
Need engineering assistance? Drop me a message!
View user's profile Visit user's homepage View All Posts By User This user has MSN Messenger
Twospoons
International Hazard
*****




Posts: 1201
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline

Mood: A trace of hope...

[*] posted on 25-10-2007 at 12:48


I find the fancy models tend to be more available for the fancy transistors - low noise RF transistors, for example. Some manufacturers supply more detailed models than others. Its easy to see which - the model file is just a text file.
12AX7 is quite correct about the garden variety semis - the process spread is so huge, and so many different fabs are making these generic parts, that the models are quite simple.

Appologies for hijacking this thread into a discussion on SPICE. I'll say no more.

@12AX7 Looking at your CCM module, you have the neg supply to the opamp connected to Vout. Since this is operating off a PC supply, would it not make more sense to use the -12V rail for the neg supply of the amps, since it is there for free? All my experience suggests the opamp will work better with stable split rails.

[Edited on 26-10-2007 by Twospoons]




Helicopter: "helico" -> spiral, "pter" -> with wings
View user's profile View All Posts By User
Xenoid
International Hazard
*****




Posts: 775
Registered: 14-6-2007
Location: Springs Junction, New Zealand
Member Is Offline

Mood: Comfortably Numb

[*] posted on 25-10-2007 at 14:14


@Rosco Bodine
@12AX7
@DerAlte
@Twospoons

Hi guys, sorry to interrupt your arguments, but I have a question about modifying AT/ATX supplies.

I have several supplies which I have modified in the usual way of removing all the surplus wiring and fitting terminals for the +5V supply and a switch if required. I use these for chlorate cells etc. and adjust the current with adjustable carbon rod resistors as mentioned elsewhere.

I would like to be able to vary the +5V supply from say about 2-3 Volts up to about 6-7 Volts this would allow me to do away with the resistors for (per)chlorate cells. I am NOT interested in converting the supplies to full laboratory supplies or even constant current at the present time.

I have attached the circuit for a Seventeam AT supply, which I obtained here;

http://www.users.on.net/~endsodds/smps.htm

Is it possible to place a potentiometer in the +5V feedback input for the TL494 (pin 1)(shown in red). This appeared to be alluded to in the garbled reference mentioned earlier by Dann2;

http://www.fieldlines.com/story/2007/10/18/16953/116

This would necessitate changing the overvoltage sense input of the LM339 (pin 7)(shown in green) presumably by tying it to the +5V regulated line so it no longer has any effect.

At the same time as varying the +5V output the other output voltages will vary also, but this doesn't matter as I am not interested in them. The 12 Volt auxiliary supply, (marked in purple) will also vary significantly. If this variation is too great, this supply could be replaced with a small transformer and a 7812 regulator etc. which could be mounted upside down inside the power supply lid.

I gather some supplies already have a trimpot for adjusting the +5V output exactly (the Seventeam doesn't) wouldn't it just be an expansion of this feature?

Would this work, or is it a bit more complex than this. Can you suggest any other simple way of varying the +5V output up and down by a few volts.

Regards, Xenoid

230cct.gif - 27kB
View user's profile View All Posts By User
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 25-10-2007 at 15:40


I'm thinking replace R38 with a 2.2K fixed and 5K pot added in series to pin 2 of the TL494 may be all you need to do .

Connect the wiper to one end of the pot resistance attached
to either the pin , or the end of the 2.2K fixed , so that as
the pot is adjusted , your substituted adjustable value for the old R38 varies from 2.2K to 7.2K .

Set the pot so the total value of the string is the same 4.7K as before , and watch a voltmeter reading on the output as you try to vary it a little bit up or down .

The meter will tell you if I am guessing right or not .

[Edited on 25-10-2007 by Rosco Bodine]
View user's profile View All Posts By User
Xenoid
International Hazard
*****




Posts: 775
Registered: 14-6-2007
Location: Springs Junction, New Zealand
Member Is Offline

Mood: Comfortably Numb

[*] posted on 25-10-2007 at 16:24


Well Rosco, that's essentially what I was going to do until I read the thread on the page that Dann2 posted, (see my previous post). These guys do the voltage adjustment on pin 1 of the TL494. They also have a circuit for current limiting.
See here;

http://www.anotherpower.com/gallery/dinges/PC_PSU_schematic?...

Unfortunately it's a bit of a rats nest pencil sketch!

Regards, Xenoid

[Edited on 25-10-2007 by Xenoid]
View user's profile View All Posts By User
Rosco Bodine
Banned





Posts: 6370
Registered: 29-9-2004
Member Is Offline

Mood: analytical

[*] posted on 25-10-2007 at 16:42


You could try the same setup described above as a substitute for the R41 3.9K .

One or the other should work ...or it may required
a ganged pair to do both simultaneously .

I looked at that pin 1 and wondered about it , but figured
pin 2 more likely .

I'll see if I can find a data sheet for the TL494 and
maybe I'll know more then if I can see what's in it ,
and where the internal reference is .
View user's profile View All Posts By User
Twospoons
International Hazard
*****




Posts: 1201
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline

Mood: A trace of hope...

[*] posted on 25-10-2007 at 16:52


It's really hard to see but it looks like there is fb from +5 AND +12 ? Which to me suggests changing R47(?) to be variable. I can't read the values.
Or use Rosco's method, but don't tweak it too far, or you might have to muck about with the compensation network on pin3.

You should be OK with the Aux rail. data sheet says the '494 can handle quite a bit of vcc ( up to 40V).

TL494 datasheet

[Edited on 26-10-2007 by Twospoons]




Helicopter: "helico" -> spiral, "pter" -> with wings
View user's profile View All Posts By User
Xenoid
International Hazard
*****




Posts: 775
Registered: 14-6-2007
Location: Springs Junction, New Zealand
Member Is Offline

Mood: Comfortably Numb

[*] posted on 25-10-2007 at 16:59


@ Twospoons

The original circuit is here;

http://www.users.on.net/~endsodds/230cct.jpg

It's a little easier to see.

Regards, Xenoid
View user's profile View All Posts By User
 Pages:  1    3    5  ..  9

  Go To Top