Clock Freq

g3cwi

I have an application that needs a reasonably jitter-free variable frequency. I have read the application note and thus realise that jitter depends on the pll settings etc and can be avoided (to some extent) with some ratios. My application requires very small changes in frequency (0.5Hz) at high frequencies (10MHz).

Ideas:

1) Could I fool the Prop by telling it that my crystal frequency (clkfreq) was something other that it actually is to make sure that the frequency I need is a "sweetspot"?
2) Can I dynamically re-define the Prop's "clock freq" in the code to move the sweetspot a little?
3) If all else fails I guess that I could use a hardware method to pull the actual crystal freq.

Thoughts from Propellor-land?

Regards

Richard

Toby Seckshund

From what i gather the sweet spot is very much restricted to powers of twos of the xtal, so a 5MHz xtal would lend itself to 5, 10, 20 .... MHz. I have told the Prop lies (or perhaps the truth) in the freq declaration to get the banding to be less objectionable on PAL. When I measured the Xtal freq it did show it to be about 400Hz low but some of that error could have been the act of measuring it.

Alan (ex g8lcu)

jmg

g3cwi wrote: »

I have an application that needs a reasonably jitter-free variable frequency. I have read the application note and thus realise that jitter depends on the pll settings etc and can be avoided (to some extent) with some ratios. My application requires very small changes in frequency (0.5Hz) at high frequencies (10MHz).

Flip this into the time domain : you are asking for 100ns, for the 10MHz, which so far sounds easy, but that 0.5Hz, is 0.05ppm,
or expressed in time, you want 99.9999950000002499999 ns - yes, that's 5 femto seconds steps you are asking for.
(google low jitter clocks, to get an idea of real jitter numbers, from the best sources. )

Something running at 12.5ns, is going to have shiploads of jitter, relative to 5fs.

How many of those 0.5Hz steps do you need ? A VCXO may be the best high purity solution ?

Mark_T

Given the standard tolerance on crystals of +/-30ppm or there abouts, an accuracy of 0.05ppm is going to be an issue. I'd consider a specialized clock generator like the Si5351 (although I have to say I can't figure out quite what it can do from the datasheet(!) - it does have a frequency-pulling input pin though.

Are you looking to modulate a signal or does the frequency have to be stable long-term to 0.5Hz accuracy?

g3cwi

@Mark T - you are possibly confusing two measures here. Tolerance is not quite the same as short term frequency accuracy. I agree with you about the use of a proper oscillator, this is more of a thought experiment.

@jmg People are already using the prop with 2Hz shifts at 10MHz for ham radio applications:

http://ko7m.blogspot.co.uk/2012/01/wspr-on-propeller.html

I was just thinking of ways to get the shifts closer to what is required by the modulation method (actually 1.5Hz - not 0.5Hz as I originally thought). You are right about the VCXO of course but I was just wonedring what could be done to improve the "prop standalone solution".

Thanks all.

Any more thoughts?

Regards

Richard

Phil Pilgrim (PhiPi)

Richard,

"No" to questions 1 and 2, because they don't change anything physically, so the oscillation with a given frqa will be the same regardless. And it's the value of frqa that determines how much jitter there is. As to question 3, yes you could add a cap to the crystal circuit and pull the frequency a bit. The Propeller itself would have no way of knowing how much it's being pulled, though, because it'd be pulling its own timebase. However, I'm sure this is something you could calibrate ahead of time.

I've done some work with a Propeller-mediated DPLL that you might find useful, here:

http://forums.parallax.com/showthread.php?123791

You can apply the same principle to the local oscillator in mixers such as the SA602 or '612.

I always tell people that you can't filter out oscillator jitter, once it's produced, since the vestigial sidebands it produces are so close to the fundamental. However, your case might be an exception, since your frequency is fixed, and your modulation range is so narrow. I wonder if a crystal lattice filter could be tuned sharply enough to provide a passband only a few Hertz wide.

-Phil, AD7YF

jmg

g3cwi wrote: »

@jmg People are already using the prop with 2Hz shifts at 10MHz for ham radio applications:

http://ko7m.blogspot.co.uk/2012/01/wspr-on-propeller.html

I was just thinking of ways to get the shifts closer to what is required by the modulation method (actually 1.5Hz - not 0.5Hz as I originally thought).

So you are trying to do the same thing ? send 683ms duration, of one of 4 frequencies, ~1.5Hz spaced ?
(I think 10.000MHz is in the very centre of the 4 ? so you never send exactly 10.000)

That link 'works' because they ignored the spectrum - see Phil's link.

You don't really want to be radiating in a broad splash, so a cheap VCXO would give the best results, and the Prop could lock that to a better OCXO or GPS reference pulse, if the 10.000MHz average is important.

PLL chips also have better spectrums, but are a very big hammer. The Si5351 we have played with, this has a sub GHZ Vco, and is relatively cheap, but the better ones have VCOs in the some-GHz region, and they have better spectrums.

Given you know the Centre F and the small dF is nicely within a VCXO, and they are cheap, why not do it properly ?

Edit: Checking into TXCO/VCXO, it seems 10MHz is giving way to 20MHz in the smaller models.

This nice one, is ~$3.50/100+ from Digikey (535-9702-1-ND)

Series	ASVTX-09
Type	VCTCXO
Frequency	20MHz
Function	-
Output	Clipped Sine Wave
Voltage - Supply	 (Vdd pin) 3V
Frequency Stability : Initial	±0.5ppm
Frequency Stability : (Over Temp  -30°C to 75°C) ±2.5ppm
Pullability ± 8ppm min
Control Voltage : (Vcc pin) 1.5V±1.0
Operating Temperature	-30°C ~ 75°C
Current - Supply (Max)	2mA
Mounting Type	Surface Mount
Size / Dimension	0.197" L x 0.126" W (5.00mm x 3.20mm)
Height	0.063" (1.60mm)
Package / Case	4-SMD, No Lead (DFN, LCC)

http://www.abracon.com/Oscillators/ASVTX-09.pdf

At 0.5ppm with a 8ppm pull range, this is good for many precise projects.

The clipped sine out may need a 1GU04 buffer, or maybe the Xtal buffer is enough ? (set to Max Speed?)
Or, the 74HC6323 can be used to both amplify, then divide to a square wave drive.
3V supply allows for a small RC filter from 3v3.

The pull range gain is only broadly defined, so a gain-calibrate step would be needed, min gain looks like 8ppm/V, so the (extremely) narrow band FM in the link above, of 146.48ppb/step, is a modulation control dV of ~18.31mV.

Abracon do not give a modulation bandwidth, but TAITIEN mentioned 10KHz on their vTCXO, which is no problems to support slow 683ms bit timings.

The Prop can manage Calibrate and DAC, if you want to lock within that 2.5ppm, the modulation is about 6% of the lock voltage (+/- 312mv about 50% of Vdd )

Phil Pilgrim (PhiPi)

Here's another way to go about it. If you need to generate a frequency-modulated 10 MHz, get a 4.9152 MHz crystal to clock the Prop with. They're pretty easy to get hold of, since they divide down nicely to common baud rates. With that, generate two frequencies using the Prop's counters in NCO mode: 4.9152 MHz x 2 = 9.8304 MHz and 10.0000 MHz - 9.8304 MHz = 169.6 KHz. The higher frequency will have no jitter, of course. The lower frequency will have some, but it's such a low frequency that it won't matter, and you can modulate it with relative impunity. Now, if you mix these two frequencies, you will get 9.8304 MHz ± 169.9 KHz. The sum will be 10.0000 MHz. There will be an absolute splattering of other mixer products in there, too, since the square wave outputs are rife with odd harmonics. You can either filter these out ahead of the mixer with low-pass filtering or post-mixing.

As a very crude experiment, I let the Propeller do the "mixing" by outputting both signals to the same pin, effectively ORing them:

CON

  _clkmode      = xtal1 + pll16x
  _xinfreq      = 4_915_200

  OUT_PIN       = 0

PUB  start | time

  ctra := %00100 << 26 | OUT_PIN
  ctrb := %00100 << 26 | OUT_PIN
  dira[OUT_PIN]~~
  frqa := frq_val(9_830_400)
  frqb := frq_val(169_600)
  repeat

PRI frq_val(frq) : r

'' Compute the required value of frqx to obtain the frequency frq.
 
  repeat 32
     frq <<= 1
     r <<= 1
     if frq => clkfreq
        frq -= clkfreq
        r++

This produced the waveform shown, with the expected mess of "mixer" products:



Here's what it looks like zoomed in on the fundamental frequency.



Next, I constructed a simple narrow-band filter with some parallel-resonant (20 pF) 10 MHz crystals I had:

(All caps are 22 pF.)

Here is what the output and broad spectrum look like:



Here's the zoomed-in spectrum:



You will probably obtain even better results doing the mixing externally in an SA602/612. But at least, the principle appears to be sound, with spurious frequencies more than 40dB down in my crude experiment (on a solderless breadboard, no less).

By this time next week (hopefully), I should have a better spectrum analyzer and will repeat the experiment to see if there are any hidden gotchas that don't show up here.

-Phil

jmg

Phil Pilgrim (PhiPi) wrote: »

As a very crude experiment, I let the Propeller do the "mixing" by outputting both signals to the same pin, effectively ORing them:
...
You will probably obtain even better results doing the mixing externally in an SA602/612.

Some nice footwork. Is that filter out voltage over 3V p-p ? Who needs a SA602 ?

Uses 5 generic crystals, but may still need a lock ? (varicap on the 4.9152MHz, or maths adder on the 169.6 KHz ? )

Another modulator choice is a XOR gate - be interested to compare the filter out voltage from OR and XOR choices.
Could a series RC as first stage, allow removal of a Crystal stage ?

I can see one fish-hook in this approach, and that is even at the slow modulation rate, the numbers may not quite stack up ?
The NCO granularity/frame rate, is of the same order as the modulation size and speed, so will the quantize noise, compromise demodulation ?

Demodulation of this is asking for resolve of just +1/+2/-1/-2 whole cycles, in the 682.687ms frame.

g3cwi

Nice work Phil. It does show that the prop is capable of some really great things with a bit of creative thought. You will perhaps be wanting to get your new WSPR system on the air to try it out. If so, this will help you to see how it is working:

http://wsprnet.org/drupal/wsprnet/map

regards

Richard