AD654-based VFC(VCO) drop-in

I don't think the AD654 does what you want here. It does 500kHz at full control voltage input, which is 5VDC (9V -4V) in this case, if I read the datasheet right.

The AD654 really shines in its linearity used as a VCO, but that's unimportant to you, I think.

Even if you used a CMOS 555, I think the dominant source of variation will be capacitor tolerance/tempco.

Those are really vague requirements... at the low end you have a UJT astable and at the high end you've got rubudium or cesium frequency standards. Tell us how many percent or how many ppm error you can tolerate and then you'll get a workable solution. Temperature and voltage variations will help too... but I suspect that since you said "resonator" that the right solution will involve a crystal and a PLL and/or divider chain.

Tim.

Referencing page 2 of the datasheet for the AD654, I'm seeing the AD654 has a frequency range of 0 to 500kHz. As such, it can act as a VCO by using a trimmer pot to adjust the input voltage. It's my understanding from the data sheet that if _Vs+_ is 5VDC and _Vin_ is 1VDC (_Vs+_ - 4 = _Vin_max) the ouput full scale square wave is 1V @ 500kHz. Now, with these voltages and outputs given, the obvious equation of [.5*_Vin_=_Fout_], shows me that 983mV on _Vin_ should give the required output of 491.5kHz. So, by creating a voltage drop of roughly

4V from the intended 9V supply or by using a 5V voltage regulator and using a trimmer pot on _Vin_, I can hit these voltages and, by extension, the requisite frequency.

With these things in mind, a 20pF cap across the _Ct_ pins "should" give me 491.5kHz on _Fout_ @ 983mV. The question is what value resistor network do I need for the whole thing.

I had considered using a 1.966MHz crystal and a divider chain however, crystals in that specific frequency are hard to find. Crystals in the original frequency (491.5kHz) are no longer available. What I've also found is the AD654 is capable of handling this task with fewer components. If there is a better solution, I'd love to hear it.

As far as ppm error, this really shouldn't matter a great deal as this signal is a reference used to create a QRS wave for ECG/EKG test equipment.

Thanks for the help in advance!

Wow, a 2 MHz crystal isn't good enough for you, you need 1.966 MHz, yet you have no problem using a RC oscillator, and then tell us this is used to calibrate medical equipment so tolerance doesn't matter at all.

All I can say is, once you do figure out your requirements, I'm sure it can be done.

Tim.

The 1.966MHz crystal would make it easier to use a simple "divide-by-four" circuit to govern the input signal. Granted, it seems odd that this particular signal's tolerance isn't critical, especially considering the application, but it is only a redesign of a currently functional device that will have to be dropped if a suitable substitute for the original crystal/resonator can't be found. I only need to figure out if this will work long enough to build a prototype of the drop-in replacement. From what I can see so far, calibration of the frequency will be challenging at the least. If it doesn't work, no big; I'll go the long route and outsource a full redesign, updating the full scope of the circuitry and everything else.

Please understand one more thing: This task was dropped in my lap because I have SOME electronics experience. I'm NOT by any means an engineer. I need assistance in this, not condescention and ridicule. I'm on a learning curve right now and need input in order to decide if chasing after this option is plausible not to mention feasible. I'm asking, with great respect, if anyone can help me design the necessary circuitry to build this prototype drop-in.

Again, THIS IS ONLY A PROTOTYPE. It will not be used for production runs unless and until it is proven consistent and safe for the application for which it is used. It must be tolerant to the normal hazards of running from a 9V dry-cell, such as voltage drop due to a dying battery. That is why I am seeking the aforementioned 1V input voltage while regulating the supply voltage @ 5VDC.

Tim, you probably do know much better than I how this could turn out. I really and truly need some input. You may ask questions that I don't know the answer to or even what it is you are asking. If you could rough in a possible answer, I can find out more and give answers that are more fitting to the questions you are asking.

Now, to answer your earlier questions about ppm error. The AD654 fits well within an acceptable range for ppm error. The original resonator had a + or - 1kHz stamped on the case. Unfortunately, there is no other information about that device available due to records being destroyed. This is what is making this even more difficult. I do know that the original resonator only created a baseline from which the other board components governed timing. This part functioned ONLY in that aspect. There is sufficient adjustment in the test device as a whole to compensate for drift and ppm error where the output QRS wave is concerned. (There is a complete manual on calibrating this calibration device.) The most critical outputs are heart rate (beats per minute) and the QRS wave itself. So long as the QRS is shaped correctly and sustains a 1mV peak, the overall baseline frequency error isn't critical. QRS wave shaping is handled by a completely different set of microdevices on the board. The wave will be shaped correctly so long as the input frequency is close to the 491.5kHz target.

So, considering the little extra light I can shed on this, would you please give me a hand in this? Thank you in advance!

So a reasonable guess at accuracy required is 1 part out of 491.5 kHz, which is 0.2%.

I'd be very reluctant to redesign a piece of equipment which had originally specced a 0.2% resonator oscillator with a RC oscillator. All the good things being said about the AD654, it is still a RC oscillator, and you will learn that no matter how good your AD654, you will be at the mercy of the capacitor tolerances.

The obvious solution to me is a crystal-controlled solution. You may note that 491.5 kHz is almost exactly 15 times 32.768 kHz. (It's so close that it may not be a coincidence!) 32.768 kHz is the most common crystal frequency of all. A crystal oscillator at 32.768kHz, with a CD4046 or 74HC4046 VCO/PLL running at 491.5kHz and locked to the

32.768kHz reference via a divide-by-15 made out of 74HC logic, seems very straightforward to me.

If you had to go with a more simplistic solution, a LC resonant oscillator can be done to 0.2% at the frequency range you're in, but it will take some care in design and feeding and alignment for every one produced.

Tim.

How about a 3.932160 Mhz crystal? This is 8x 491520 Hz.

Digi-Key shows this on their web site at $.40/ea, $.36 if you buy ten so it's probably a standard frequency.

X013-ND ECS-39-17-1 CRYSTAL 3.932160MHZ 17PF HC49/UA.

But wouldn't a multiply-by-15 circuit give me a pulse train with varying pulse widths?

I don't believe that 0.2% is the actual spec on that resonator. That may have been the spec on the one we were using but I'm not sure if that is an absolute spec.

As far as usng the CD4046 or 74HC4046, I like that possibility, however; we're now increasing the number of discreet components and, thus, cost. The internal schematic of the AD654 shows an oscillator inside. It seems to spec out close to what I need and keeps a handle on the number of discreet parts I'd need to make a drop-in replacement.

Are there any other possible solutions that keep the number of parts to a minimum and aren't going to be a major PITA to implement? And could you draft up something to show me this better way?

I wish a crystal-controlled solution in a single package were still available for this!!

That may well be an excellent solution to this dilemma as well. Again, if I can implement a way to lock the signal to 491.5kHz with a minimum of extra parts, I'd have this thing whipped!!!

Thanx, I'll look into this as well!!!

Nope, look up how PLL's work. Mike's suggestion for dividing down may be easier..

Seeing as how this is expressly for calibrating medical equipment my gut feeling is to run away and not offer any advice :-).

I've seen too many medical readings where the result "must" be accurate to 4 digits but nobody was ever able to explain to me why they needed more than one digit. It's like a classful of kids with 8-digit calculators.

The AD DDS chips come real close...

If 491.52 kHz is close enough, you could use a 4.9152 MHz crystal, and just divide it by ten. If you use a 74x90 (probably HC these days), you can divide by 5 first, then 2, to get a square wave. Mouser # ... Ah, crap, the catalog page is .pdf. (can't copy/paste an image of a part number )-; )

Good Luck! Rich

I was thinking that dividing down would not only be easier but would allow me to use a less accurate crystal. After all, when we divide down, the overall precision inproves, right? I mean, if I multiply up

15x's from 32.768kHz on a 0.1% tolerance crystal, doesn't that increase the error by 150%, giving me 1.5% tolerance? Or does this not amplify in a linear manner, i.e. 0.1% becomes 22.5% due to drift and other factors, etc. Dividing down, if my thoughts ARE correct, would allow me to use a 2% tolerance crystal and, per Rich's comment, dividing would render me an effective 0.2% tolerance, falling well within the spec you noted earlier.

Here we go again about the freq tolerance... ;-) Again, this is a baseline signal for timing only. It isn't actually used to generate the QRS. This signal only marks out the timing of the QRS pulses. Considering Mr. Grise's input very carefully, I find that a drop-in based on his idea would render a very stable solution from which the QRS _could_ be built.

Take a look at Rich's response. I think he may be on to something that not only will be simple to make but will also use the fewest discreet parts. What do you think?

The stability of these crystals seems really good. Great idea!

Now, how the heck do I do a divide-by-10 circuit?

(If you haven't figured it out already, where frequencies and other electronic stuff is concerned, I'm kind of at a loss.....)

I'll have to reveiw my response to see if I've misled you. Dividing a frequency doesn't change the _percentage_ tolerance. Lessee, .1% of

32768 is 32.768, right? So the range of absolute frequencies is 32735.232 through 32800.768 - divide them down, and the range is still +- .1% of nominal. I'll let you guys verify my math. :-)

Cheers! Rich

Here's the data sheet, you can get these at Mouser too -

On page 6, under Bi-Quinary. See Note B. :-)

I've also crossposted this to s.e.basics, which it's kind of sounding like it belongs. :-)

Good Luck! Rich

You're right, Rich. The overall tolerance doesn't shift. The range of error narrows but the tolerance percentage remains the same. My bad.