Chip with simple program for Toy

[dan williams]

Fred Bloggs wrote:

See http://www.national.com/an/AN/AN-1095.pdf#page=1 and follow the
instructions exactly, using off-the-shelf flyback transformers.

the primary reason I am resisting that temptation is because I know it
will not always be an option. For example my other application required
dealing with an 80VDC supply, national dont go there. (not with simple
switchers) I'm not farmiliar with the document, so I will check it out.

dan

dan williams wrote:
This is the second time I have been called on for a smps design.
Granted, for this one I can use National's design system(s) But I'd like
to know how to do this. (Its easy to be out-of-range of NS's limits for
simple switchers)

Please tell me where I go wrong, there are an uncomfortable number of
arbitrary limits made, I dont know if their right. you may want to read
it trough before making comments.

This may also be good reading for people studdying switchmode supplies,
formulas here have taken me about 48 hours (solid hours) to work out
from many references, and almost all make sense.

I would really like to see an "out with it, already" document on smps,
come to think of it, I didn't do a google search for a smps FAQ


Lets take the case at hand:

Input: 10-15VDC
Ouput: 5VDC (30V Isolation) 3A

Just for the sake of it, lets go with a 100Khz switching frequincy, that
way we can compare to NS's SimpleSwitcher Design

Step 1) An innocent Start...
Choose a Topology.
Because of the isolated nature, we basically have forward or flyback
to choose from. (of the BASIC topologies)
The last time I saw a forward converter, it was in an apple IIe,
(There will be a reason for this, I'm happy not knowing what it is.) so
lets go with flyback.

Step 2) Things get a little harry...
First pass parts selection
we need information about our switches and diodes to run subsiquent
calculations
Switch:
As per motorola's SMPSRM (which nolonger seems to be available)
"Mosfet power switch, Vdss = 1.5Vin(max) Id = 2Pout/Vin(min)"
so we need at least: Vdss = 22.5V Id = 3A
Go for the Kill IRFZ44N Vdss = 55V Id = 49A [Rds(on) = .024 ohms]
That was easy enough... lots of those in the bin.
Diode:
As per motorola's SMPSRM "Vr = 10 Vout If = Iout"
so we need at least: Vr = 50V If = 3A
Every reference talk about doide losses, resulting in "use a
shottkey doide." ok. (I'm ignoring synchronous rectification here)
since the IRF cd is already in the drive... how about a 50SQ080 Vr
= 80V Id = 5A [Vf = .52V]
I'v never heard of the diode before, but if nobody else in my end
of Canada can get it (quite likley), I'll substitute (a process I'm
getting good at).

Step 3) Spark up the algebra.
Select a value for your inductor. I shall derive the following
formulae, to prove I do (or don't) know where it comes from, and because
almost none of the 5+ references I'm using factor EVERYTHING in.

Lets Start by drawing 'er out. (I'll save you my ascii art, anyone
who can help me knows what a basic flyback topology looks like, for
everyone else, its like a boost, but the output commes from a secondary
winding on the inductor that engauges when the switch turns off. For
those of you who don't off hand know what a boost topology looks like,
may I suggest taking a look through application note 19 from Linear's
web site.)

Next chart the voltage across the inductor vs time.

V

^
|
|
|============
| on |
--------------------------> t
| | off |
| ===========
|
|
|
V

- The Voltage across the primary side of the inductor when the switch
is on, is the supply voltage, Vin minus the drop across the switch,
being that this is a big burly mosfet, I'm approximating this to be 0.
So the charge voltage is Vin.

- The Voltage across the primary side of the inductor when the switch
is off, is the output voltage, plus the diodes forward drop, all divided
(turns ratio is 1(primary):N(secondary)) by the turns ratio. ( as you
can quite plainly see from the schematic I didn't draw)

- The Vt area of the on and off time are, for a close approximation,
the same (we have loss via the core, I can't calculate it, so I'm
ignoring it... its small! right?)

From these we get the formula:

ton Vin = toff (Vout + Vfdiode) / N

set that aside,

From any physics book, we can obtain the formula

L = V (dt/di)

so whats our charge time?

t = 1/f (ats just the way she iz)f being our switching frequincy.
Because were using PWM, t is the absolute maximum time the indutor would
get in a "cycle" at 100% duty. In order to know our REAL charge time, we
need to multiply t by our duty cycle, D. which goes from 0 to 1.

dt = D/f or tD as in the duties fraction of the maximum time.

so whats our current?

this IS a good question, the answer I find is "20% of the switches
maximum rated" but what!?
it seems to me, that this should be related to, say, the power
output? but such an approch will cause this current to be dependent on
the turns ratio, which I havn't worked out yet! to avoid simotanious
equations, I will use this 20% nonesense, erm, I mean "arbitrary rule of
thumb"

di = .2Id

scoop it all up into a pile:

L = (Vin D) / (0.2 f Id)

hey wait! you didn't say what D is!

thats easy: D = ton/(ton + toff) (ats just the way she iz)
now, previously, we worked out a formula in terms of Vin, Vout, ton,
and toff, plus a few inneficiencies.

Vs is the switches voltage drop...

ton (Vin - Vs) = toff (Vout + Vfdiode) / N

mulled becomes:

(Vin - Vs)/((Vout + Vfdiode) / N) = (toff / ton)

set aside...

D = ton / (ton + toff)

therefore

1/D = (ton + toff) / ton
or
1/D = (ton / ton) + (toff / ton)
or
1/D = 1 + (toff / ton)

so

(1/D) - 1 = (toff / ton) <- third to last step
or
(1/D) - (D/D) = (toff / ton)
or
(1-D)/D = (toff / ton)

and look! we have the same term for our Duty calc and the deriviation
of duty cycle!

(Vin - Vs)/((Vout + Vfdiode) / N) = (toff / ton) = (1-D)/D

crunch crunch crunch!

(Vin - Vs)/((Vout + Vfdiode) / N) = (1/D) - 1 (from third-to-last
step)
or
N (Vin - Vs)/(Vout + Vfdiode) = (1/D) - 1

so

N (Vin - Vs)/(Vout + Vfdiode) + 1 = (1/D)
or
N (Vin - Vs)/(Vout + Vfdiode) + ((Vout + Vfdiode) / (Vout + Vfdiode))
= (1/D)
or
(N (Vin - Vs)+(Vout + Vfdiode))/(Vout + Vfdiode) = (1/D)

so

D = (Vout + Vfdiode)/(N (Vin - Vs)+(Vout + Vfdiode))

now we know the duty, we can calculate the iductor:

L = (Vin / (0.2 f Id)) * ((Vout + Vfdiode)/(N (Vin - Vs)+(Vout +
Vfdiode)))
or
L = (Vin (Vout + Vfdiode)) / (0.2 f Id N ((Vin - Vs)+(Vout +
Vfdiode)))

step 4) lets get arbitrary...

So what is our inductor size?

Vin = 10 (this is a worst case thing)
Vout = 5
Vfdoide = Vf = .52V as per spec sheet
f = 100Khz
Id = 49A
N = we really havn't a clue, do we?
Vs = 0 (approximation, looking back on this, our iductor current
is 9.8A... which is awefully high, and even across .024 ohms starts to
become significant(.2V))

Now this is a stumbling block... the best references I can find for
choosing N are based on the limitations of the voltage on the switch
when its off (Vswitch_off = (Vout + Vdiode)/N) of which you want to stay
under its rating, which motorola had us choose as 1.5Vin_max
Linear provides a formula for "optimum value" based on Vsnub, which
they calculate from N (oh, thats helpfull)

Its also not logical to be running nearly 10A through the primary,
not for a 15W supply...

It would seem logical to use a formula for the primary current like
(Pout/Vin(min))which seems to result in a much more reasonable current
of 1.5A

somewhere(forgotton now) I read "bipolar transistors stressed 75%"
something "their maximum rating, are subject to crowding, which is an
instentanious failure mode"

so I'd say 70% of the mosfets Vdss is a safe goal

this puts 0.2Id = 1.5 A
and N = (Vout+Vf)/(0.7Vdss + Vin) = .103 = about 10:1 ratio, awefully
small....

but wait, Ip = Is * N (the secondary current is N times the primary)
so N = Ip/Is resulting in a ratio of 0.5 ... this is much more
reasonable, concidering the simple switcher software suggests about .333
for an equivilent supply.

I'm going to go with the sane numbers:
0.2Id = 1.5A
N = .5

Vin = 10 (this is a worst case thing)
Vout = 5
Vfdoide = Vf = .52V as per spec sheet
f = 100Khz

so

L = (Vin (Vout + Vfdiode)) / (0.2 Id f N ((Vin - Vs)+(Vout +
Vfdiode)))

= (10 (5 + .52)) / (1.5 100,000 0.5 ((10 - 0)+(5 + .52)))

= 47.42 uH

which, is, atleast, in the same order of magnitude as the
simpleswitcher software comes up with.

Step 5) missing essentials...

When the power switch turns off, before the output diode engauges,
there's a really big voltage spike on the coil 'o wire. quite capable of
destroying the mosfet. so we add something across (debatable) the coil
to catch the spikes, there are the various types, used in various times.

snubber ---/\/\/\---||-----

soft clamp ----->|----\/\/\/------
| |
----||----

and Zener clamp ----->|-----Z<----

I know that national designs always use zener clamps, I also know that
every other smps I have seen uses a soft clamp. (computer supplies,
computer monitors (the guys get annoyed with me continiously
disassembing the computers "Yup, see, thats a soft clamp too", "put it
back togethor, I need to check my email!" ), industrial supplies,
printers, VCR's)

snubbers always dissipate power, so I'll slide them aside...

Zener's are used to limit voltage, soft clamps and snubbers to limit
voltage and dv/dt, which will break down switches given enough.

I dont know how to calculate limits for dv/dt, so I'll assume its not
a problem, and go for a zener clamp.

I gather from the linear app note (19) that you want to stay atleast
10V away from the switches max rated voltage. we also have to concider
we need to be above the normal flyback voltage which is
Vp = Vs / N

Vs = Vout + Vdiode_forward

= 5.52/.5 = 11.04V

Based on Vdss = 55V

= 45V

so we want 11.04 > Vsecondary > 45

this is a pretty wide margin! I'll go closer to the 45V, which I
recall an application note from AIC (I think) suggesting to use the
highest possable voltage....

We just happen to have 33V zeners around the shop, I'll go with them
(this design is not quite optimal....)

as a side note I have the following formula for calculating soft
clamp components (no , I dont know where I got it from)

Lp = primary inductance
Ip = peak drive current
Vc = clamping voltage (e.g. 33V)
Vi = supply voltage
Vs = voltage across switch when off (load engauged)

C = (0.2LpIp^2)/(Vc^2 - Vs^2)
R = ((Vc+Vs-Vi)/2)^2 (.0192/(LpIp^2))

I'm not going to work that out, but I would probably end up using a
soft clamp, I have seen a lot of zeners fail over the years (short out)
I see a soft clamp as something much more rugged.

I'm not going to mention PWM control systems, We have some rails of
TL494's at the shop, they work fine.

Step 6) follow through.

ok, but how do we BUILD it. actually I'm only reffering to the
inductor.

This is where the linear app note accels, how to make an inductor

- Calculate a suitable core by volume, from

V = ( Ip ^ 2 L u 0.4 Pi 100000000) / Bo ^ 2

where L = inductance (H)
Ip = peak inductor current
u = permeability (from material datasheet e.g. material #77 =
2000)
Bo = max flux density (from material datasheet e.g. material #77 =
4600)
V = core volume(cm^3)

At the shop we have a set of E cores from CWS bytemark which are 2540
mH/1000T, I know there big enough for this application, so I just need
to put wire on them...

- Apply wire to selected core

Where
Npri is the primary turns
L is primary inductance
Al is from core datasheets

Npri = 1000 * square root of ((L/Al)

so for
Al = 2540mH/1000T
L = 47.42uH

N = .14 turns... ok, well maybe that core is a little big... thats
how I calculate it anyhow.... I suppose we need some new cores.....
darn. (Turns must be an interger, prefferably larger than 0)

anyhow, the secondary turns would be the primary turns * N (the
winding ratio N...)

well, thats it. how did I do? I would quite like suggestions on
places I went wrong or just was wrong. Its 12:30am, so I'm not going to
list the references I used for this documents formulea, if you want,
ask, I'll tell.

provided this process can be nicely nailed down, a minor bit of
software (something I'm actually good at) can quickly take care of the
calculator labor. which I will be happy to share with anyone.
I dont want to have to labor over designing a smps, I dont see why
anyone should after so many years of them being around. NS's
simpleswitcher stuff is nice, but cant help if you need something over
about 30W or 40V input (like an off-line running at 170V)
Descriptions of control circuits could probably easily double the
size of this letter, but as its mearly feedback and stability
calculations, and most controller chips come with one-size-fits-all
sugested filters, and resonably good formulae to boot, I'm not worried
about it.

Thanks for the feedback course, Rod.

Dan Williams

--
Dan Williams, Owner
Electronic Device Services
(604) 741 8431
RR8 855 Oshea rd
Gibsons BC Canada
V0N 1V8

 

[dan williams]

"R.Legg" wrote:

dan williams <dan_williams@sunshine.net> wrote in message news:<3F2B680A.8554570@sunshine.net>...
This is the second time I have been called on for a smps design.
Granted, for this one I can use National's design system(s) But I'd like
to know how to do this. (Its easy to be out-of-range of NS's limits for
simple switchers)

Please tell me where I go wrong, there are an uncomfortable number of
arbitrary limits made, I dont know if their right. you may want to read
it trough before making comments.

This may also be good reading for people studdying switchmode supplies,
formulas here have taken me about 48 hours (solid hours) to work out
from many references, and almost all make sense.

I would really like to see an "out with it, already" document on smps,
come to think of it, I didn't do a google search for a smps FAQ


Step2 should be calculation of currents and voltages. You should
really select semiconductors based on a known requirement. If a
junkbox part suffices, fine.

This is a good point as pointed out by Kevin. My process was derived
from motorola, national, and linear documents, none of which give
excelent lead-throughs of design. Linear's document definitly in a
leauge ahead of the others, but their recursive equations left much to
be desired.

You may want to recognise some mechanical realities before actual part
numbers are selected. How is the thing to be fabricated? All
through-hole or some SMD?

all we do is through-hole, so far anyhow.

Hv schottky is IR, available in small quantities from Digikey. Click
the maple leaf and pay in Canadian dollars from the Canadian division
without worrying about monetary conversions, duty or border crossings.

Lets not get arbitrary, calculate the requirements with margin. Then
fit in parts you may want to use to see if limits are maintained.

I completely agree, by arbitrary I was reffering to places where all the
documents I referenced left out key ingreadients.

http://focus.ti.com/docs/training/catalog/events/eventsbycategory.jhtml?templateId=5517&navigationId=8455

SEM400 topic2 flyback transformer.
SEM400 topic6 Flback power supply.
MAG100A basic magnetics design.

RL

I'll check it out.

dan

--
Dan Williams, Owner
Electronic Device Services
(604) 741 8431
RR8 855 Oshea rd
Gibsons BC Canada
V0N 1V8

 

[R.Legg]

Dave Baker <dpbaker@streamyx.com> wrote in message news:<b2uoivop5ick98h3ljcu84lr4lpj709b29@4ax.com>...

While we are on this subject, I'm looking for a mould/box/whatever that I can
put my small circuit board into, and fill it with epoxy, then remove the
mould/box/whatever, and just leave the epoxy cube. What material can I use,
so the epoxy won't stick to it?

Dave

A re-usable mold, made of any material, will benefit from a non-stick
film of releasing agent. This can be as simple as oil or wax, but
silicone is probably used most often nowadays. You would have to check
for chemical compatability. A popular hobby mold material is
thick-walled latex or silicone rubber, flexible enough to peel off the
set material, but dimensional stability may be a problem, particularly
with repeated use.

Alternatives are one-shot molds that you either destroy to recover the
molded object, or that you leave permanently applied. The latter can
give a more cosmetically attractive appearance. A flow-mark and
void-free epoxy surface finish is not always easy to achieve,
particularly if filled epoxy is necessary for thermal or other
reasons.

Bulk cured epoxy is also prone to mechanical instability, particularly
when foreign material or relatively large objects are imbedded, and
the reagent is either too strong or poorly mixed. A permanent shell
can protect it from shock that might cause fracture along flow lines,
due to violent handling. Well-formulated epoxy, on the other hand, can
be machined to a certain extent.

RL

 

[DarkMatter]

On Sun, 03 Aug 2003 10:51:50 +0800, Dave Baker <dpbaker@streamyx.com>
Gave us:

While we are on this subject, I'm looking for a mould/box/whatever that I can
put my small circuit board into, and fill it with epoxy, then remove the
mould/box/whatever, and just leave the epoxy cube. What material can I use,
so the epoxy won't stick to it?

Dry mold release agent. Sold in spray cans. For plastics, and even
for the food processing industry. Wet mold release will inhibit epoxy
curing, or can.

 

[DarkMatter]

On Sat, 02 Aug 2003 20:11:29 -0700, John Larkin
<jjlarkin@highlandSNIPtechTHISnologyPLEASE.com> Gave us:

On Sun, 03 Aug 2003 10:51:50 +0800, Dave Baker <dpbaker@streamyx.com
wrote:

While we are on this subject, I'm looking for a mould/box/whatever that I can
put my small circuit board into, and fill it with epoxy, then remove the
mould/box/whatever, and just leave the epoxy cube. What material can I use,
so the epoxy won't stick to it?

Dave

Use a potting shell if you possibly can; just stick the board in the
shell, fill, and you're done. Robison makes a jillion sizes and shapes
of potting shells.

Molds, especially hard molds, are a nasty mess. If you do want to use
a mold, use silicone rubber. Plastics supply houses (Tap Plastics, out
here) have cool molding silicone. Make a prototype shapr out of
anything, cast the silicone around it, and pull it out when it cures.
Voila, a mold that *releases*. Doesn't even need draft angles, and you
can include fancy things like logos or bosses or textures.

Vacuum degass the epoxy for the ultimate bubble-free fill. Messy, too.

Potting loses its charm fast.

Not if it is required. Doh!

 

[DarkMatter]

On 3 Aug 2003 01:42:45 -0700, legg@magma.ca (R.Legg) Gave us:

A permanent shell
can protect it from shock that might cause fracture along flow lines,
due to violent handling. Well-formulated epoxy, on the other hand, can
be machined to a certain extent.

R

Very informative. The moral:

Always mix the proper amounts of the constituents, and always mix
very thoroughly. Evacuate with vacuum where required or desired for
maximum density, consistent grain structured potting matrices.

 

[Kevin Aylward]

dan williams wrote:

John Popelish wrote:

"Kevin Aylward" <kevin@anasoft.co.uk> wrote in message
news:<QDSWa.1846$LQ4.143820@newsfep2-gui.server.ntli.net>...
John Popelish wrote:

{snip}

Wow...you read all this, in detail....


My appologies for it being so long, it didn't get that way until I was
done, and I did cut it short.

Flyback transformers (or inductors) normally use cores with gaps in
the magnetic path to allow the core to store more energy without
saturation, Of course, this lowers their Al value, requiring more
turns, which requires more air gap to handle the higher amp turns
without saturating, etc. But energy goes as amp turns squared,
while Al drops as amp turns, so it settles down to a solution.
Basically, you can guess a gap, and check how many turns that gap
will require to get the inductance, and then check the peak amp
turns and see how high the core flux will go. For flyback
designs, I like to stay below about half of the saturation flux
rating.


Imo, I think this is probably a bit conservative. I don't see any
main reason to not go to just below its max flux density rating,
having duly taking into account the worst case operating
conditions. Its not like you get any extra life like you would by
derating a transistor. If your've got the space, you might save a
bit on core loss, but as a rule, you never have enough space

This rule of thumb is for a hand made one off, where I might change
my mind a bit about the worst case or have misjudged something else.
It saves having to go back an wind another transformer. A design for
mass production would squeeze things a bit more. Running closer to
saturation also increases the energy to be dumped into any snubber.

Is there a guideline for max power dissipated in a snubber? not that I
had a chance to apply some of these new techniques, but today I was
running some pretty cooking snubbers. Just one of the ways I'm
learning that its not designed right.

Its all common sense stuff. If the snuber loss is 10 times all other
losses, it wouldn't seem that optimum would it? Everything is an
engineering compromise.

It seems that this would be apropiate.

"Those who seek advice are generally seeking an accomplice".

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[Kevin Aylward]

dan williams wrote:

Fred Bloggs wrote:

Kevin Aylward wrote:
For flyback designs, I like to stay below
about half of the saturation flux rating.



Imo, I think this is probably a bit conservative. I don't see any
main reason to not go to just below its max flux density rating,
having duly taking into account the worst case operating
conditions. Its not like you get any extra life like you would by
derating a transistor. If your've got the space, you might save a
bit on core loss, but as a rule, you never have enough space


The final operating condition is a function of the material loss with
frequency more than anything else. If you are pushing the material in
frequency then you generally can't go near Bsat- unless you really
like burned up wiring and fuming magnetics. Even 1/2 Bsat may not be
good enough. The window area - core cross section product rule used
for initial sizing tells you nothing about ultimate core power
dissipation density- and you have to keep this under control to
limit temperature rise.

Here is another little tidbit I know of, but havn't found any
information on, what are the guidelines to do with core losses?
so far all my cores have been made of #77, #75, #63, #61. and I make
sure the volumes are atleast

V = ((Ipeak)^2 L u .4 Pi) / (B^2 * 10^-8)

thats almost the only factor I take in consideration when selecting a
core...

Care to elaborate on this?

The *key* formula for a transformer is

ET=NBA

To determine the minimum area of the core.

Then one chooses the primary inductance such that its magnetising
current in not too large a fraction of the load current, for obvious
reasons. i.e.

Imag =ET/L

This allows the minium length to be determined from Le=Al.N^2/L

The above will give the minium volume.

*After* this is done, one can have a look to see if this minium volume
generates too much loss.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

 

[onestone]

How about using 16:1 analog muxes for the read function and a cross
point for the square wave function. Even an 8:1 cross point would do.
The cross point allows you to supply a signal to up to n of n lines,
while the mux lets you read 1 of 16. a 16:1 cross point would simplify
addressing somewhat, and a parallel controlled device like the old
harris (now Intersil) parts would eb ideal. You could then do this with
just 14 devices. it depends on the signal levels you require. For high
voltages look at the MAX7506, else look at the DG506A/DG406. IIRC most
of these handle single or dual supply rails. Some up to 40V, except for
the MAX7506 they are all around US$3.50 in 1k lots.

Al


EROMLIGNOD wrote:

Hello and thanks John:

Yes I do have details, which I hope don't spur a deluge of questions.

I have invented a self-tuning piano (US pat. #6,559,369). With the help of QRS
Music Technologies I have built a prototype and it works great. I just need to
cut costs. You might have heard interviews with me on NPR or CBC (Canada), or
heard it mentioned on Paul Harvey. There was also an article about it in the
New York Times in January as well as articles in New Scientist, Der Spiegel,
Nikkei Marketing Journal, etc.

I have individual magnetic pickup/drive coils for each string in the piano.
Actually they are merely ferrite-core inductors soldered to a PCB. The signal
is only used for tuning, so you never hear it and very crude (obviously) coils
can be used. The coils are used both to pick up the vibration of the strings
(like a guitar pickup) and to sustain the strings. If you impose a wave onto a
coil and place it near a string, it will cause it to vibrate.

I have a universal signal that can make a coil vibrate any string at its
natural frequency and I apply it to all of the strings at once (it sounds
pretty wild!). I individually poll each pickup by simultaneously disconnecting
it from the driving signal and connecting it to a signal bus (thus the SPDT)
back to my evaluation circuit. The residual vibrating shows up as a signal and
I determine how far out-of-tune it is and go on to the next string. I measure
the period of the string, not the frequency, so it is done very quickly. I can
theoretically measure all 219 strings in about one second. I have been
switching individual strings with a separate, crude circuit up till now. I
need a cheap way to continuously poll the strings during the tuning process.

I tune the strings by passing electrical current through them, which warms them
up, expanding them and lowering their pitch. The piano is initially tuned at
the factory at an elevated temperature, so I also have the option of lowering
the current to cool them and raise the pitch. I keep polling the pickups and
adjusting the PWM to each string until they are all in tune (match a
factory-set tuning), then this duty cycle is maintained while you play. You
shut it off when not in use. It only goes through this tuning cycle when you
first turn it on.

Any more good ideas for polling those coils? I'm actually an ME, so use
little-bitty words!

Don

 

[James Meyer]

On Sun, 03 Aug 2003 00:25:12 -0700, dan williams <dan_williams@sunshine.net>
wroth:

James Meyer wrote:

To paraphrase Shakespear when he said, "Get thee to a nunnery.", I
suggest you "Get thee to Linear Technology's web site."

Jim

There be more? The linear app WAS one of the top two docs I found...
Have to go-a-browsing...


dan

There be *many* more than one or two. Lin Tech ought to offer degrees
based on reading their ap notes. Something like a Micro$oft certification.
After digesting several, you should be able to answer questions involving SMPS
rather than asking them.

Jim "Not affiliated, just grateful." Meyer

 

[John Popelish]

dan williams wrote:

Is there a guideline for max power dissipated in a snubber? not that I
had a chance to apply some of these new techniques, but today I was
running some pretty cooking snubbers. Just one of the ways I'm learning
that its not designed right.

A snubber is an electronic heat sink for some switching device (among
other properties). So its losses should be reasonable losses for the
switching device. If you are designing a switching regulator to
improve on the losses of a linear, you wouldn't want snubber losses to
approach that of a linear regulator. A good switcher may be 80 to 90%
efficient, so if more than 10% of the output power is ending up in
your snubbers, you should start to wonder why your switch losses (that
are being transferred to the snubber) are getting so high.


--
John Popelish

 

[John Popelish]

dan williams wrote:

John Popelish wrote:
(snip)
Usually, the forward converter adds an extra winding to dump the core
magnetization energy back into the supply, but it has some nice
features, like lower peak and ripple currents. But at 15 watts,
flyback is not a bad choice, especially is you back off a bit from 100
kHz.

I have lots of extra room on the core I mentioned, another of the main
reasons for staying away from a forward is that I'm not entirly sure of
how to size the indictance (being that its not used for power storage,
all the stuff I worked out for inductor sizing is based on magnetic
storage) I would almost prefer to use a forward based on its outputs
being more capable of filtering.
to say more...
as I understand it, the output will act like a buck sully when the
switch is off. giving it a more continious output

Almost, A buck puts out a current (ramped up and down by the output
inductor. A forward converter puts out a low impedance (transformed
copy) of the input switched voltage waveform. So it needs an
additional inductor in the filter to have a buck characteristic.

The inductive current in the transformer is just an additional burden
the switch and copper have to handle. The more inductance the better,
just as with any normal AC transformer. And, similarly, size, and
cost limit how much this optimum is pushed. A buck transformer is
just an AC voltage transformer handling DC current and using only half
its BH loop.
(snip)

There is such a thing as too big a transistor. You have to charge all
that gate capacitance up and down every cycle (and fast).


at 20Khz I wouldn't think there much loss during switching transitions,
but I can see it for a point of optimization. and there arn't swithces
much bigger for me to use...

The losses per cycle tend to be just about constant (for a given
switch and input voltage and current). But at 100 kHz, there are 5
times as many of those cycles as at 20 kHz.

I think the 10 Vout spec is a bit overdone. Especially when you have
such a narrow range od input voltage. A 45 volt unit will have a
lower forward drop and almost certainly handle the reverse voltages it
will see.


I agree, I dont know why motorola uses 10.

The only advantage of the high multiplier is that it allows a very
sloppy (under snubbed) ringing waveform.

(snip)

(Ignoring the turns ration for the moment Since the output voltage
will be pretty constant, you won't need its time to vary much. You
input varies over a 2 to 3 ratio, so it will take this range of on
times to produce the same volt seconds area in your timing diagram.
If we allow the full duty cycle (low input voltage, highest output
current) case to reach 50% (because equal charge and discharge times
are pretty efficient and you want efficiency at maximum output). So
the worst case will be 50% duty cycle on the verge of continuous
current operation. 25 us to charge and 25 us to discharge. In order
to have 15 watts to be the average input power at 10 volts in, the
average current (neglecting losses) will be (draw that triangle
current wave and figure out how you get a peak current from the
average) from the 15/10*4=6 amps. Lets guess 20% more for losses. So
the switch must handle 7.2 amps, peak under this case.

now I'm interested, NONE of the docs I can find take this obvious
approach...

Well, there are a lot of simultaneous equations to solve, and some of
them are messy. Many approaches get over complicated, to reduce the
number of iterations you have to go through to solve all the equations
by including lots of minor factors in the first pass. This is fine
for computer solutions, but really muddy up the basic concepts.

The only sense I can make of the four multiplier is via the following:
Iave = Ipk/2
so
Ipk = 2 Iave
this needs to be multipled again by 2 because of the 50% duty cycle.
so
Ipk = 4 Iave

is that right?

Yes.
(snip)

a clarification on leakage inductance, during "that spike", is or isn't
there a potential on the secondary of scaled magnitude? As I understood
leakage inductance thus far, the output rectifier simply hasn't kicked
in yet, (making the term leakage inductance quite inapproprite)

Leakage inductance is effective in series with the transformer effect
that keeps the primary and secondary voltages scaled to each other.
It represents magnetic flux that surrounds the primary coils that do
not also surround the secondary windings. The air gap in a flyback
transformer increase this uncoupled flux. Imaging the primary and
secondary windings as solenoids wound side by side on a long cylinder
of ferrite. Picture their flux lines passing through the ferrite
inside the solenoids, but fringing back around the outside of the
coils from the ends of the cylinder sticking the ends of the coils.
Now picture how less of the primary flux passes through the secondary
coil as you slide the two coils farther apart on the cylinder. At
some point, the two coils will be almost independent inductors, and
there will be almost no transformer effect between them. Your goal is
to merge the two coils as much as other limitations allow, to minimize
their independent inductance.

The snubber absorbs energy that was stored in flux that did not couple
to the secondary, but still must be removed from the primary before it
can be reenergized.
(snip)

--
John Popelish

 

[Owen Lawrence]

In your schematic the GND from IC2 is connected to PE by the net
"N$27". Your other PE's are on the net "N$4". Eagle seems to use the
net names for deciding what to connect. You can use the Show tool (the
eye) to see the net names and change them using (Menu)Edit->Name. I use
names like "GND" for ground so they are easier to follow when drawing
the layout.

Thank you. That fixed the error. Now I only have two warnings left, the
VDD vs. PE one. They both seem to be part of SUPPLY1. I think it makes no
difference to the board, but eventually I'll want to figure out the "right"
thing to do.

Unless you have a good reason for it, you might want move your
components a bit to avoid putting traces between the pads of a
component. For example, moving T8 to (3.20 0.55) and rotating it 90
degrees CW.

I do try to avoid close traces, but I'm only marginally successful. You
should have seen the first layout I had! I shuffled a few things around,
and am incorporating your suggestion. Thanks for that, too.

- Owen -




-----= Posted via Newsfeeds.Com, Uncensored Usenet News =-----
http://www.newsfeeds.com - The #1 Newsgroup Service in the World!
-----== Over 80,000 Newsgroups - 16 Different Servers! =-----

 

[dan williams]

I have a few documents on the suggested reading list.
I'v been writing software to do design assistance. I havn't tried the
software suggested by [memory lapse] but I dont use windows much, and I
suspect it is a windows program.
Is anyone interested in multiplatform smps design software? I'd be
willing to share what I come up with as soon as I get it working.
ofcourse if you already know how to build one...


dan

James Meyer wrote:

On Sun, 03 Aug 2003 00:25:12 -0700, dan williams <dan_williams@sunshine.net
wroth:

James Meyer wrote:

To paraphrase Shakespear when he said, "Get thee to a nunnery.", I
suggest you "Get thee to Linear Technology's web site."

Jim

There be more? The linear app WAS one of the top two docs I found...
Have to go-a-browsing...


dan

There be *many* more than one or two. Lin Tech ought to offer degrees
based on reading their ap notes. Something like a Micro$oft certification.
After digesting several, you should be able to answer questions involving SMPS
rather than asking them.

Jim "Not affiliated, just grateful." Meyer

--
Dan Williams, Owner
Electronic Device Services
(604) 741 8431
RR8 855 Oshea rd
Gibsons BC Canada
V0N 1V8

 

[dan williams]

Kevin Aylward wrote:

dan williams wrote:
Fred Bloggs wrote:

Kevin Aylward wrote:
For flyback designs, I like to stay below
about half of the saturation flux rating.



Imo, I think this is probably a bit conservative. I don't see any
main reason to not go to just below its max flux density rating,
having duly taking into account the worst case operating
conditions. Its not like you get any extra life like you would by
derating a transistor. If your've got the space, you might save a
bit on core loss, but as a rule, you never have enough space


The final operating condition is a function of the material loss with
frequency more than anything else. If you are pushing the material in
frequency then you generally can't go near Bsat- unless you really
like burned up wiring and fuming magnetics. Even 1/2 Bsat may not be
good enough. The window area - core cross section product rule used
for initial sizing tells you nothing about ultimate core power
dissipation density- and you have to keep this under control to
limit temperature rise.

Here is another little tidbit I know of, but havn't found any
information on, what are the guidelines to do with core losses?
so far all my cores have been made of #77, #75, #63, #61. and I make
sure the volumes are atleast

V = ((Ipeak)^2 L u .4 Pi) / (B^2 * 10^-8)

thats almost the only factor I take in consideration when selecting a
core...

Care to elaborate on this?

no. I pulled it from linear application note 19. inductor and
transformer basics.
like all equations in all the documents I have read so far, they dont
tell you how they derive the equations (which is great fun when your
trying to figure out why there all different (ultimitly the same, but
different))

I'm sure that if I spend another few hours at it, I can work out their
equation from what you have said.
What really gets me is when they have arbitrary safety margins in their
equations (BUT WHERE DOES THIS .2 COME FROM!)
It wasn't until writing this origional post that I realized the
significance of duty/Frequincy as an inductor charge time.
There are other goodies in designing these that aren't mentioned
anywhere in a document like linear app 19, like capping max duty at 50%,
their calcs lead me to a design that likes to run at about 77% (hrm,
well, that's what their "ratio optimizer" says...) open resistors,
shorted diodes, and burt fingers later.... (ah what I would give for a
thermal camera)

I'm going to read the suggested documents (going to take me a while),
redesign a 12v -> 5V 1A flyback converter (small explosions as opposed
to big ones) before carrying on with the target project. Before or after
testing it (I think its my tendency to dive into things that earned me
the nick "Dangerous Dan" It didn't take me long working on switching
supplies before I started adding fuses to input supplies) I'll post my
process and results.

I put in the deriviations for my equations as a diagnosis tool for "what
I did wrong" I may skip that in futher comments, if anyone is
interested, just say so.

The *key* formula for a transformer is

ET=NBA

To determine the minimum area of the core.

Then one chooses the primary inductance such that its magnetising
current in not too large a fraction of the load current, for obvious
reasons. i.e.

Imag =ET/L

This allows the minium length to be determined from Le=Al.N^2/L

The above will give the minium volume.

*After* this is done, one can have a look to see if this minium volume
generates too much loss.

Kevin Aylward
salesEXTRACT@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.

I'm sure I'v heard of anasoft before....

dan

--
Dan Williams, Owner
Electronic Device Services
(604) 741 8431
RR8 855 Oshea rd
Gibsons BC Canada
V0N 1V8

 

[dan williams]

John Popelish wrote:

dan williams wrote:

John Popelish wrote:
(snip)
Usually, the forward converter adds an extra winding to dump the core
magnetization energy back into the supply, but it has some nice
features, like lower peak and ripple currents. But at 15 watts,
flyback is not a bad choice, especially is you back off a bit from 100
kHz.

I have lots of extra room on the core I mentioned, another of the main
reasons for staying away from a forward is that I'm not entirly sure of
how to size the indictance (being that its not used for power storage,
all the stuff I worked out for inductor sizing is based on magnetic
storage) I would almost prefer to use a forward based on its outputs
being more capable of filtering.
to say more...
as I understand it, the output will act like a buck sully when the
switch is off. giving it a more continious output

Almost, A buck puts out a current (ramped up and down by the output
inductor. A forward converter puts out a low impedance (transformed
copy) of the input switched voltage waveform. So it needs an
additional inductor in the filter to have a buck characteristic.

I seem to have had a funny memorization of the output of a forward
converter in my head, corrected...

The inductive current in the transformer is just an additional burden
the switch and copper have to handle. The more inductance the better,
just as with any normal AC transformer. And, similarly, size, and
cost limit how much this optimum is pushed. A buck transformer is
just an AC voltage transformer handling DC current and using only half
its BH loop.

I concider the buck converter to be pretty elementrary, at worst you
treat it like a lowpass filter, duty cycle dictates output. I suppose
the primary concideration is to make sure the core can hold the power...

(snip)
There is such a thing as too big a transistor. You have to charge all
that gate capacitance up and down every cycle (and fast).


at 20Khz I wouldn't think there much loss during switching transitions,
but I can see it for a point of optimization. and there arn't swithces
much bigger for me to use...

The losses per cycle tend to be just about constant (for a given
switch and input voltage and current). But at 100 kHz, there are 5
times as many of those cycles as at 20 kHz.

"save the high frequincy for things that HAVE to be small"?

I think the 10 Vout spec is a bit overdone. Especially when you have
such a narrow range od input voltage. A 45 volt unit will have a
lower forward drop and almost certainly handle the reverse voltages it
will see.


I agree, I dont know why motorola uses 10.

The only advantage of the high multiplier is that it allows a very
sloppy (under snubbed) ringing waveform.

(snip)
(Ignoring the turns ration for the moment Since the output voltage
will be pretty constant, you won't need its time to vary much. You
input varies over a 2 to 3 ratio, so it will take this range of on
times to produce the same volt seconds area in your timing diagram.
If we allow the full duty cycle (low input voltage, highest output
current) case to reach 50% (because equal charge and discharge times
are pretty efficient and you want efficiency at maximum output). So
the worst case will be 50% duty cycle on the verge of continuous
current operation. 25 us to charge and 25 us to discharge. In order
to have 15 watts to be the average input power at 10 volts in, the
average current (neglecting losses) will be (draw that triangle
current wave and figure out how you get a peak current from the
average) from the 15/10*4=6 amps. Lets guess 20% more for losses. So
the switch must handle 7.2 amps, peak under this case.

now I'm interested, NONE of the docs I can find take this obvious
approach...

Well, there are a lot of simultaneous equations to solve, and some of
them are messy. Many approaches get over complicated, to reduce the
number of iterations you have to go through to solve all the equations
by including lots of minor factors in the first pass. This is fine
for computer solutions, but really muddy up the basic concepts.

The only sense I can make of the four multiplier is via the following:
Iave = Ipk/2
so
Ipk = 2 Iave
this needs to be multipled again by 2 because of the 50% duty cycle.
so
Ipk = 4 Iave

is that right?

Yes.
(snip)

a clarification on leakage inductance, during "that spike", is or isn't
there a potential on the secondary of scaled magnitude? As I understood
leakage inductance thus far, the output rectifier simply hasn't kicked
in yet, (making the term leakage inductance quite inapproprite)

Leakage inductance is effective in series with the transformer effect
that keeps the primary and secondary voltages scaled to each other.
It represents magnetic flux that surrounds the primary coils that do
not also surround the secondary windings. The air gap in a flyback
transformer increase this uncoupled flux. Imaging the primary and
secondary windings as solenoids wound side by side on a long cylinder
of ferrite. Picture their flux lines passing through the ferrite
inside the solenoids, but fringing back around the outside of the
coils from the ends of the cylinder sticking the ends of the coils.
Now picture how less of the primary flux passes through the secondary
coil as you slide the two coils farther apart on the cylinder. At
some point, the two coils will be almost independent inductors, and
there will be almost no transformer effect between them. Your goal is
to merge the two coils as much as other limitations allow, to minimize
their independent inductance.

so, in a case where I dont need 1500V of isolation, for a 1:1
I could, in theroy take 12 strans of (for the sake of argument) #30 blue
wire, and 12 strans of #30 red wire, twist them all togethor, wrap up
the core, and split them out to blue = primary, red = secondary, lots of
capacitance, but almost no leakage.

I wouldn't see capacitance like this being a problem, as long as the
offset between the input and output voltages are static.
I am also disregarding the terrible things that may happen during
failure of such a device. I presume the losses may also go up a bit as
there would be minor differences in stran lengths in a single
"conductor".

dan

The snubber absorbs energy that was stored in flux that did not couple
to the secondary, but still must be removed from the primary before it
can be reenergized.
(snip)

the transforers I'v been making must have tonnes of leakage inductace...

dan

--
John Popelish

--
Dan Williams, Owner
Electronic Device Services
(604) 741 8431
RR8 855 Oshea rd
Gibsons BC Canada
V0N 1V8

 

[John Popelish]

dan williams wrote:

John Popelish wrote:
(snip)
The inductive current in the transformer is just an additional burden
the switch and copper have to handle. The more inductance the better,
just as with any normal AC transformer. And, similarly, size, and
cost limit how much this optimum is pushed. A buck transformer is
just an AC voltage transformer handling DC current and using only half
its BH loop.

I concider the buck converter to be pretty elementrary, at worst you
treat it like a lowpass filter, duty cycle dictates output. I suppose
the primary concideration is to make sure the core can hold the power...

To be clear on this, the forward converter does not intentionally
store any energy in the core field. The energy passes directly from
primary to secondary as in any good transformer. The core has to
stand the volt seconds of the excitation (and requires the same volt
seconds in reverse to reset the core flux back to near zero, each
cycle). Coil voltage requires flux rate of change. Voltage for time
requires some total change of flux. And a forward converter uses flux
in only one direction, so it cannot support only half as many volt
seconds as it could if the flux could swing from positive saturation
to negative saturation.

(snip)

The losses per cycle tend to be just about constant (for a given
switch and input voltage and current). But at 100 kHz, there are 5
times as many of those cycles as at 20 kHz.


"save the high frequincy for things that HAVE to be small"?

Yes.

(snip)

Leakage inductance is effective in series with the transformer effect
that keeps the primary and secondary voltages scaled to each other.
It represents magnetic flux that surrounds the primary coils that do
not also surround the secondary windings. The air gap in a flyback
transformer increase this uncoupled flux. Imaging the primary and
secondary windings as solenoids wound side by side on a long cylinder
of ferrite. Picture their flux lines passing through the ferrite
inside the solenoids, but fringing back around the outside of the
coils from the ends of the cylinder sticking the ends of the coils.
Now picture how less of the primary flux passes through the secondary
coil as you slide the two coils farther apart on the cylinder. At
some point, the two coils will be almost independent inductors, and
there will be almost no transformer effect between them. Your goal is
to merge the two coils as much as other limitations allow, to minimize
their independent inductance.


so, in a case where I dont need 1500V of isolation, for a 1:1
I could, in theroy take 12 strans of (for the sake of argument) #30 blue
wire, and 12 strans of #30 red wire, twist them all togethor, wrap up
the core, and split them out to blue = primary, red = secondary, lots of
capacitance, but almost no leakage.

Bingo. The only improvement on that is to wind coax, and use the
shield as one winding, and the center conductor as the other.

One other detail. If you add a thick spacer for core gap, it helps to
put a spacer on the bobbin, directly over the gap, so that no turns
fall in the highly fringed field, at least in the first layer or two
of windings. That spot is a leakage inductance concentration. The
fringe field also causes extra copper losses in that area.

I wouldn't see capacitance like this being a problem, as long as the
offset between the input and output voltages are static.
I am also disregarding the terrible things that may happen during
failure of such a device. I presume the losses may also go up a bit as
there would be minor differences in stran lengths in a single
"conductor".

The biggest problem with multi filar windings in a flyback converter
(aside from insulation failure) involves high common mode noise passed
from the switch to both sides of the output. A common mode choke
after the rectifier and first filter cap can help, and can be quite
small, since the load current produces no field in the core of this
choke, so it can have a high permeability, ungapped core.

The snubber absorbs energy that was stored in flux that did not couple
to the secondary, but still must be removed from the primary before it
can be reenergized.
(snip)


the transforers I'v been making must have tonnes of leakage inductace...

Building the perfect transformer is sort of a religious quest. I
meditate while I wind. Especially toroids.

--
John Popelish

 

[Bob Myers]

"Frank Pickens" <frankpickens@verizon.net> wrote in message
news:3F2AFD2A.3F9AAADF@verizon.net...

I have bought a couple of 1X16 LCD display modules (parallel driven) and

can't find from any of the
data sheets that I have located whether static is a factor with these
devices. My concern is soldering
onto the 16 pads on the circuit board. Is special handling needed with
these devices ??

They're no more - or less - sensitive than any other modern
electronic device. Unprotected, you'll be running the risk
of damaging the inputs to the on-panel driver chips. The
LCD itself is relatively immune to ESD damage, but it's
not the only thing in the package.

Bob M.

 

[Ziguy]

I'm still here... It's just that a have hard time to follow/understand the
discussion... I think I will make tests/mesurements of my setup!

But, can I conclude that the case to ambient resistance will not increase as
long as the additional layers that I put are more conductive than the air?

Thanks again guys!

 

[eromlignod]

Mark Fergerson <mfergerson1@cox.net> wrote in message news:<3F2F7F36.70105@cox.net>...

Multiplexer as suggested by others looks simplest
parts-count wise (with sample-and-hold period/v converter to
PWM drivers?). I never heard of a 1-of-219 chip, so you may
have to learn PICs. Or, you could cascade 1/whatever's handy
and skip unused inputs.

Actually the multiplexing and PWM control isn't that bad. I use a
74F675A serial in, parallel/serial out shift register with
output-hold. Since it has a serial output, you can cascade several
together to basically get one giant shift register. With fourteen of
them in a row I can convert a serial signal to a big 224-bit word. I
connect each bit to a power transistor to control current to the
strings.

I refresh this word every few milliseconds and send a series of 100
words in a continuous cycle. So if I want to control string #48, I
just put a "1" in the 48th place of the word to turn it on and a "0"
to turn it off. The duty cycle is set by which word of the cycle of
100 words that I change the bit. For example, if I want string #48 to
have a 60% duty cycle, I set bit #48 in the word to "1" in the first
word I send, then set it back to "0" on the 60th word I send. Does
this make sense?

I wonder about your "theoretically" though, since you
need 1/20th second to get one period at the usual lower
limit of hearing (I'm relatively piano-illiterate; your
lowest-pitched string may go lower?). That gets better as
you go higher in pitch, but _that_ much better?

The lowest string in a piano is 27.5 Hz for a period of 1/27.5 = 36
ms. The highest note is 4180 Hz for a period of 239 us. The majority
of strings are higher fequency since they have triple strings. The
lower ones are single strings, so they don't add up to as much time.
It really does come out to nearly one second.

Also, how do you cope with the slight shifts in frequency
when the key is first struck and when the little felt pad
thingies damp the strings' vibration when a key is released,
by disabling polling a specific key while it is being played?

The tuning process only occurs when you first turn on the system. You
must depress the pedal (which lifts the dampers) and let the strings
sustain and warm until they are in tune (20 seconds). Then the PWM
duty cycle values are stored in memory and maintained while you play.
When you're done, you just turn it off. The next time you turn it on,
it tunes again. This way you get a custom tuning for the room
conditions at that moment. After the piano is tuned by hand at the
factory the sustainers sustain the strings at a specified level before
recording their value. They are again sustained at this level when
tuning in the field. In this way the tuning is identical to the hand
tuning.

This is indeed very slick!

Thanks!

More dribble: BTW, what are you using as a standard
during play, an ovened crystal oscillator or what? Also,
what is it referred against, the usual tuning fork? How does
the piano "remember" the "factory-set tuning" ratios, a set
of fixed dividers off the oscillator?

The pickups are so crude that they produce basically a fundamental
sine wave with very little filtering (to get overtones requires an
expensive pickup!). I Schmitt-trigger this signal to get a
fundamental square wave. Then I feed this signal to the clock input
of a one-shot counter and set the "count" number to "1". When I
trigger it, the output goes low at the first low-to-high trasition of
the input wave and then high again at the next low-to-high (to signify
that it has counted to "one"). So the output is a pulse equal to
exactly one vibration of the string.

I feed this pulse to the "gate" of another counter and a 10 MHz
oscillator to the "clock" input. Now it will count how many "ticks"
of the oscillator happen in one string vibration. I read this number
from the chip into my MPU. This is the number that is compared to the
correct one, etc.

Do you have a "failsafe" option for when power dies?

No, you're hosed if the power dies. But then, who wants to play in
the dark?

Don