FPGA Central

hello,

I woud like to design a 32,768 Hz stable input for
calibrated real-time operation of a FPGA-centric system,
as power-of-two frequencies make digital computations
very easy, using shifts instead of divides for motion equations.

* I "could" use a multiple of this frequency as my
main clock but I find no suitable canned oscillator
(a 3.3V 5x7mm one is preferred, cheaper is better)

examples :
- 18.423MHz / 32768 = 562.5 => almost there...
so I would have to divide it by 5*5*5*3*3
to get a power-of-two frequency, but only 16KHz.
- 24.576MHz (close to my target 25MHz) = 2*3*5*5*5 * 32768
so I get a nice 65536Hz after 4 small prescalers
but my other 25MHz circuits won't be happy by 1.7%.
And I have already bought the 25MHz oscillators :-/

* I "could" make a custom oscillator with a cheap&tiny "wristwatch"
cristal but the power consumption, stability and acuracy
would not be optimised.

* My target is the Actel A3P250 in TQFP100 package.
I "could" use his Fusion sibling (that has an integrated
and separately powered RTC oscillator) but the price
difference is ... not enchanting. And the package is different.
And I have the A3P250 already :-/

* I "could" use a common external I2C clock/calendar
but i'd prefer SPI and the cheap chips don't (rarely ?) provide
the 32KHz output (that I reuse to sequence stuff).

So... what's left ?

yg
--
http://ygdes.com / http://yasep.org

On Nov 26, 9:22*am, whygee <y...@yg.yg> wrote:
>
> So... what's left ?

It's not clear exactly what you are asking ?

Epson, EM Microelectronic et al, have 32Khz osc
modules, that include the crystal.

Maxim have one that is TCXO as well, and
if precision matters, then you will need txco.

NXP PCF2129 is another candidate, but less stocked.

MicroCrystal.com also have a range.

Some tcxo's are fixes in time, not frequency, so
watch the details

-jg

On Nov 25, 3:22*pm, whygee <y...@yg.yg> wrote:
> hello,
>
> I woud like to design a 32,768 Hz stable input for
> calibrated real-time operation of a FPGA-centric system,
> as power-of-two frequencies make digital computations
> very easy, using shifts instead of divides for motion equations.
>
> * * I "could" use a multiple of this frequency as my
> main clock but I find no suitable canned oscillator
> (a 3.3V 5x7mm one is preferred, cheaper is better)
>
> examples :
> * - 18.423MHz / 32768 = 562.5 => almost there...
> * so I would have to divide it by 5*5*5*3*3
> * *to get a power-of-two frequency, but only 16KHz.
> * - 24.576MHz (close to my target 25MHz) = 2*3*5*5*5 * 32768
> * *so I get a nice 65536Hz after 4 small prescalers
> * *but my other 25MHz circuits won't be happy by 1.7%.
> And I have already bought the 25MHz oscillators :-/
>
> * * I "could" make a custom oscillator with a cheap&tiny "wristwatch"
> cristal but the power consumption, stability and acuracy
> would not be optimised.
>
> * * My target is the Actel A3P250 in TQFP100 package.
> I "could" use his Fusion sibling (that has an integrated
> and separately powered RTC oscillator) but the price
> difference is ... not enchanting. And the package is different.
> And I have the A3P250 already :-/
>
> * * I "could" use a common external I2C clock/calendar
> but i'd prefer SPI and the cheap chips don't (rarely ?) provide
> the 32KHz output (that I reuse to sequence stuff).
>
> So... what's left ?
>
> yg
> --http://ygdes.com/http://yasep.org


I think you've covered all the bases given your apparently conflicting
requirements. Often engineering is a matter of determining an optimal
combination of all requirements, but sometimes it is a matter of
trading off one against the other so that neither is optimal, but both
are adequate. So if you *must* use 25 MHz, then you need a 25 MHz
oscillator, not 24.576. That means you have to provide another
clock. If using a 32.576 kHz oscillator is too power consuming, how
do you expect to get this second clock? I don't see how a clock chip
would be any better than a 32.576 kHz oscillator. The clock chip
still uses a crystal.

Rick

On Nov 26, 9:22*am, whygee <y...@yg.yg> wrote:
> hello,
>
> I would like to design a 32,768 Hz stable input
<snip>
> And I have already bought the 25MHz oscillators :-/
>
> So... what's left ?

A different approach...

How stable ? - If you have 25Mhz Osc, then
divide by 762/763 in a ratio basis, gets you
to within 5.2ppm of 32.768 on average, and
gives just 40ns of period jitter. Tolerable?

1-((2*(762/25e6)+29*(763/25e6))/31)*32768
ans = 5.20258064516129032e-6

Is your 25MHz better than 5ppm

-jg

On Nov 26, 9:22*am, whygee <y...@yg.yg> wrote:
> hello,
>
> I woud like to design a 32,768 Hz stable input for
> calibrated real-time operation of a FPGA-centric system,
> as power-of-two frequencies make digital computations
> very easy, using shifts instead of divides for motion equations.

If this is for motion control, it's likely the 40ns jitter and 5ppm of
my earlier example is fine...

... but in case it is not, if you are going to x4 the 25MHz anyway in a
PLL, then this can apply :

1-((1*(3051/100e6)+3*(3052/100e6))/4)*32768
ans = 2.56e-6

ie from 100MHz, you get 2.56ppm, and 10ns jitter,
and any binary time under 32768/4, has no jitter
at all.

-jg

-jg wrote:
> There are other solutions:
>
> My original one, from 25Mhz, has a 31 modulo
> swallow : with /762 /763 counter
> 1-((2*(762/25e6)+29*(763/25e6))/31)*32768
> ans = 5.20258064516129032e-6
>
> and there is a solution at 50MHz
> 1-((1*(1525/50e6)+7*(1526/50e6))/8)*32768
> ans = 2.56e-6

is the barrell of aspirine also provided ? :-)

> - interesting this is also 2.56ppm, but with
> a modulo 8 swallow.
> Can you get a 11 bit preloadable counter, at 50Mhz ?
it should do, but that's looking overengineered :-/

> Or, this can be done as fixed /25; /61, and a single width CE pulse,
> to swallow one clock, to get /1526
a proper crystal frequency sounds better :-/

>> The only effect is on ego (or marketing) because 100MHz is a "sweet spot"
>> and 98 looks more like a miss, even though static timing analysis
>> reports better than that. But who said that marketing was compatible
>> with engineering ?
> Marketing would call a 98Mhz point a "100Mhz system" anyway
My CPU is memory-synchronous so it is limited by the external
asynch SRAM's speed anyway... The fastest ones I could find are
8ns so I'm not sure that a 10ns cycle is realistic.
I have synchronous RAMs that are faster but they introduce delays...

....

I just looked at the datasheets and the ENC28J60 provides a "clock out"
that is configured as 25/4=6.25MHz after power-on reset. This pin is also
available on the module's output \o/ I looked further and Microchip
was careful : the oscillator is not disabled when the chip sleeps,
only when the clock control register is set so. So no dirty business
with getting around stupid system interlocks during sleep :-)

I could switch the integrated CPU clock source to the 25MHz when booted.
However, the 10BasetT module is optional, and I have only one PLL
so frequency multiplication is possible on only one main clock...

The question of price is very important : I could get a bunch
of suitable crystals for a good price, but canned oscillators are
not as available, and not for the desired frequency. So I'm looking
at how I can safely drive the crystal without adding too many analog
parts. In fact, a cheap PIC12F could do the trick and the remaining
pins could replace the voltage supervisors. The space eaten by
the HC49S can is compensated by some removed SOT23s ...
The price of the crystal + cheap PIC is lower than a smaller oscillator :-/

> -jg
yg

--
http://ygdes.com / http://yasep.org

-jg wrote:
> A 9 bit, 16Mhz counter should be viable ?
did anyone tell you that you're obsessed ? :-D

> -jg
yg

--
http://ygdes.com / http://yasep.org

-jg wrote:
> I thought 'standard values' and price were important?
I go for 12.288MHz as main clock,
I can divide easily by simple integers to get RS232, 32.768KHz
and some audio frequencies. I can also PLL up to 98MHz wich
is close enough to 100MHz. I can reuse the 25MHz oscillators later.

I'm now looking around for better deals
than the usual/mainstream distributors.
I have found crystals but driving them
properly opens a new can of worms...
like price, surface, complexity, calibration...

> -jg
yg

--
http://ygdes.com / http://yasep.org

On Nov 26, 6:25*pm, whygee <y...@yg.yg> wrote:
> > Can you get a 11 bit preloadable counter, at 50Mhz ?
>
> it should do, but that's looking overengineered :-/
>
> > Or, this can be done as fixed /25; /61, and a single width CE pulse,
> > to swallow one clock, to get /1526
>
> a proper crystal frequency sounds better :-/

?

but a lot more expensive, than < 20FF in a FPGA.

I thought 'standard values' and price were important?

-jg

On Nov 26, 8:51*am, whygee <y...@yg.yg> wrote:
> -jg wrote:
> > I thought 'standard values' and price were important?
>
> I go for 12.288MHz as main clock,
> I can divide easily by simple integers to get RS232, 32.768KHz
> and some audio frequencies. I can also PLL up to 98MHz wich
> is close enough to 100MHz. I can reuse the 25MHz oscillators later.
>
> I'm now looking around for better deals
> than the usual/mainstream distributors.
> I have found crystals but driving them
> properly opens a new can of worms...
> like price, surface, complexity, calibration...
>
> > -jg
>
> yg
>
> --http://ygdes.com/http://yasep.org

you cant get it all.
getting GOOD smd oscillators at 0.50$ at low qty
in A3P250 you do have the PLL for free but it has its lower frequency
so if you use 32khz you cant use the PLL and can not have any
frequency highere then 32KHz at all, unless you run own ring-
oscillator
with dynamic calibration

Antti

Hi Antti !
I can't reach you by email :-(

Antti wrote:
> you cant get it all.
> getting GOOD smd oscillators at 0.50$ at low qty
I believe you but I don't know where/how.

> in A3P250 you do have the PLL for free but it has its lower frequency
the PLL input is minimum 1.5MHz.

> so if you use 32khz you cant use the PLL and can not have any
> frequency highere then 32KHz at all, unless you run own ring-
> oscillator with dynamic calibration
I don't understand.
What kind of trick do you want to use ?

> Antti
yg
--
http://ygdes.com / http://yasep.org

Antti wrote:
> you cant get it all.
> getting GOOD smd oscillators at 0.50$ at low qty
While we're speaking about oscillators, I have a couple
of remarks :

- power consumption : I see that many osc draw from 6 to 32mA
(depending on some factors). That's a lot ! I know it is specified
for a 15pf load and a short trace to a FPGA pin won't draw as much,
but it's still an order of magnitude above what I imagined.
I can choose a lower frequency (like 3.5458MHz) to reduce the
toggle rate but are there other methods ? Can a discrete
oscillator consume less ? I doubt, but I could compare with
circuits based on the SN74LVC2G14 or LMV7219.

- output signal : my first proto used a 25MHz osc,
directly connected to the PLL input pin of the A3P250.
I checked that it worked with my 'scope and I saw a distorded
signal with almost 1V of undershoot and overshoot.
The trace is short (15mm ?) so no such effect should occur.
I added a series SMT resistor (68 Ohms that I had around)
and the trace looked better. I have not seen schematics
that add a series resistor (I analysed Actel's and ACME's
boards), and I have never heard about oscillator overshoot.
The power rails are abundantly decoupled and the PLL works
well with the serie resistor. Did I miss something ?


> Antti
yg
--
http://ygdes.com / http://yasep.org

On Nov 26, 6:41*pm, whygee <y...@yg.yg> wrote:
> Antti wrote:
> > you cant get it all.
> > getting GOOD smd oscillators at 0.50$ at low qty
>
> While we're speaking about oscillators, I have a couple
> of remarks :
>
> * - power consumption : I see that many osc draw from 6 to 32mA
> (depending on some factors). That's a lot ! I know it is specified
> for a 15pf load and a short trace to a FPGA pin won't draw as much,
> but it's still an order of magnitude above what I imagined.
> I can choose a lower frequency (like 3.5458MHz) to reduce the
> toggle rate but are there other methods ? Can a discrete
> oscillator consume less ? I doubt, but I could compare with
> circuits based on the SN74LVC2G14 or LMV7219.
>
> * - output signal : my first proto used a 25MHz osc,
> directly connected to the PLL input pin of the A3P250.
> I checked that it worked with my 'scope and I saw a distorded
> signal with almost 1V of undershoot and overshoot.
> The trace is short (15mm ?) so no such effect should occur.
> I added a series SMT resistor (68 Ohms that I had around)
> and the trace looked better. I have not seen schematics
> that add a series resistor (I analysed Actel's and ACME's
> boards), and I have never heard about oscillator overshoot.
> The power rails are abundantly decoupled and the PLL works
> well with the serie resistor. Did I miss something ?
>
> > Antti
>
> yg
> --http://ygdes.com/http://yasep.org

canned oscillator make extreme power supply noise, and their output
can easy overshoot as well

I had a design(protoboard) where 100mhz canned oscillator did give
such noise that the Spartan 3E
did constantly return Virtex-4 JTAG ID, problems was noise on VCCINT
(from the osciallor, passing
via LDO or GND)

series resistor is not un-common, and local inductor-cap or res-cap on
osc supply is always a good idea

so small inverter with crystal may be a better option

Antti

Hi,

Antti wrote:
> canned oscillator make extreme power supply noise, and their output
> can easy overshoot as well
This is never explained in the datasheets.
but when a small thing like that draws 32mA,
it rings an alarm bell...

> I had a design(protoboard) where 100mhz canned oscillator did give
> such noise that the Spartan 3E did constantly return Virtex-4 JTAG ID,
> problems was noise on VCCINT (from the osciallor, passing via LDO or GND)
OK, I have an inductor on the Vpll but not for the supply of the osc.
It would not solve the overshoot problem, because this probably comes
from the crystal itself, it looks unbuffered.
Then if this is the case, the output is probably unsufficiently
capacitively loaded ?

> series resistor is not un-common, and local inductor-cap or res-cap on
> osc supply is always a good idea
I'll add an inductor next time, a resistor seems unsufficient.

> so small inverter with crystal may be a better option
now, the questions are
- how to correctly tune the circuit ?
by a strange coincidence I just got an old frequency meter
but this device certainly requires a good recalibration :-/
- how to be sure that the circuit consumes less ?
well... we'll have to measure. But hints are welcome...

> Antti
yg

--
http://ygdes.com / http://yasep.org

On Nov 27, 5:41*am, whygee <y...@yg.yg> wrote:
> Can a discrete oscillator consume less ?

Yes, but you need care.

Numbers from a 74LVC1GX04 on the test bench :
http://pdf.dzsc.net.cn/CAR/CARD_74LVC1GX04.pdf

Xtal: 25.8MHz
Vcc_Start = 1.3V
Vcc_stop = 0.9V
[email protected] = 1.93mA [6MHz ~ 500uA]
[email protected] = 4.04mA
[email protected] = 8.26mA

-jg

-jg wrote:
> On Nov 27, 5:41 am, whygee <y...@yg.yg> wrote:
>> Can a discrete oscillator consume less ?
> Yes, but you need care.
of course ...

> Numbers from a 74LVC1GX04 on the test bench :
> http://pdf.dzsc.net.cn/CAR/CARD_74LVC1GX04.pdf
I know this circuit already.
But it is as expensive as a cheap osc :-/
(at least from my resellers)

> Xtal: 25.8MHz
> Vcc_Start = 1.3V
> Vcc_stop = 0.9V
> [email protected] = 1.93mA [6MHz ~ 500uA]
> [email protected] = 4.04mA
> [email protected] = 8.26mA

Sweeeeet !
So if I use a 3.6468MHz crystal, it will draw even less.
I could use the 1.5V core voltage as well
but it will require some capacitive coupling
and some resistors for level translation to 3.3V-land.

What about frequency stability ?

> -jg
yg

--
http://ygdes.com / http://yasep.org

On Nov 27, 7:45*am, whygee <y...@yg.yg> wrote:
>
> What about frequency stability ?

You get what you pay for

If you need 2.5ppm, then you have to get a TCXO.
(eg the FOX924 series - 6mA Icc is good for a TCXO)

Simple Crystals are usually 50ppm-100ppm, and you ADD the tempco
curves on top of that.

You can pay more, and get down to 10ppm initial, but that still leaves
the tempco curves...

The advantage of using a OscModule, is you can usually choose what
Price/ppm balance your product needs, without a PCB change. (or, offer
a premium product model, with a TXCO)

Most important step: What ppm do you ACTUALLY NEED ?

-jg