Propeller-mediated Sine Wave Frequency Synthesizer
Phil Pilgrim (PhiPi)
Posts: 23,514
In another thread I described how to clean the jitters from a Propeller clock signal, using an external PLL chip. To convert the cleaned square wave signal to a sine wave (e.g. for use in a transmitter VFO), you'd have to use a low-pass filter to eliminate the higher-order odd harmonics (3, 5, 7, ...) that are present. In this experiment, I wanted to see how easy it would be to generate sine waves externally and control their frequency with the Propeller. Here are my results.
The sine wave output is generated by a VCO (voltage-controlled oscillator) whose frequency is determined by a parallel LC tank circuit, consisting of a ferrite-core inductor and a varactor. A varactor is a diode whose junction capacitance is determined by the amount of reverse voltage impressed upon it. So it acts like a voltage-variable capacitor. The higher the voltage, the lower its capacitance, and the higher the VCO's output frequency. By setting the control voltage using a filtered DUTY mode counter output, the Propeller can control the output frequency. Here is a schematic of the VCO circuit I modified from one presented in The ARRL Handbook for Radio Communications (2010), Fig 9.39. It outputs a clean sine wave between 4100 and 4500 KHz, depending on the voltage on the input pin (0V - 3.3V):
The VCO's oscillation frequency, however, also depends on temperature, the proximity of other objects (like my hand) to the tuned circuit, and -- apparently -- the phase of the moon. So just setting the control voltage to a fixed value is not enough by itself to obtain good frequency control. Therefore we have to close the loop using feedback. The output from the VCO to the Prop is biased so the sinewave straddles Vdd/2. That way, the Prop will see a square wave of the VCO's frequency with about a 50% duty cycle. By generating a square wave of the desired frequency using one of the counter's NCO or PLL modes, we have something to compare the VCO's frequency to, in order to apply a correction. Is this beginning to sound like a phase-locked loop (PLL)? Well, that's what it is -- but in software.
In order to operate as a PLL, we need a phase detector. One common phase detector used in PLLs is the following:
As shown, when the VCO and target waveforms are exactly 90° out of phase, the pulse widths from the two outputs are equal. As the phase relationship changes, the pulse width ratio changes, and this information can be used to make corrections. When the VCO lags the Control, you increase the VCO frequency slightly, until it catches up; and vice-versa when the VCO leads the Control.
Once again, the Propeller's amazing counters can do the heavy lifting for us. These counters have a logic input mode that's tailor-made for implementing a combinatorial phase detector like this one. All we have to do is set it to count up for a period of time, as long as both inputs are low, then for the same period, obtain a count when one input is high and the other is low. By subtracting one count from the other, we get the relative amount of change required to correct the VCO control level.
One problem with this phase detector is that it does not have a very wide capture range. This is to say that if the two frequencies differ by more than a few kilohertz, it won't be possible to determine which way to make the correction, and the VCO will wander until the frequency gets close enough to capture and lock on. What's needed is a separate feedback mechanism that monitors frequency and makes larger corrections to the VCO control voltage until they are close enough for the phase detector to take over. This is done by counting edges for a period equal to clkfreq / 1024, or about 1ms. This count is then compared to the expected count from the chosen frequency, and the VCO frequency is adjusted until the difference is zero. Then the phase detector can take over.
In the attached program ctrb is used to control the DUTY mode output to the VCO. ctra is used, in turns, for the frequency counter, and the two phase sensors. The results turned out extremely well, with the locked-on VCO frequency never wandering by even 1Hz! Here is a scope trace of the VCO locked on to a reference clock from the Prop:
The red trace is an FFT plot of the VCO output. In contrast to those in my prior thread, there are virtually no harmonics present. One thing you will notice from the reference trace is the phase jitter from the Prop. But the phase jitter is completely absent from the VCO output. This is something that would be all too apparent while listening to the output on an AM receiver with a BFO (beat frequency oscillator) turned on, and none was noticeable when I checked it.
One might wonder: why not just use a DDS (direct digital synthesis) chip and be done with it? Well, DDS ICs are a little expensive, and they still require some external passive components for proper operation. Also, if you're going to use the Propeller in a radio applicaiton, why let it loaf when it could be doing a lot of the grunt work?
-Phil
Post Edited (Phil Pilgrim (PhiPi)) : 7/8/2010 10:50:10 PM GMT
The sine wave output is generated by a VCO (voltage-controlled oscillator) whose frequency is determined by a parallel LC tank circuit, consisting of a ferrite-core inductor and a varactor. A varactor is a diode whose junction capacitance is determined by the amount of reverse voltage impressed upon it. So it acts like a voltage-variable capacitor. The higher the voltage, the lower its capacitance, and the higher the VCO's output frequency. By setting the control voltage using a filtered DUTY mode counter output, the Propeller can control the output frequency. Here is a schematic of the VCO circuit I modified from one presented in The ARRL Handbook for Radio Communications (2010), Fig 9.39. It outputs a clean sine wave between 4100 and 4500 KHz, depending on the voltage on the input pin (0V - 3.3V):
The VCO's oscillation frequency, however, also depends on temperature, the proximity of other objects (like my hand) to the tuned circuit, and -- apparently -- the phase of the moon. So just setting the control voltage to a fixed value is not enough by itself to obtain good frequency control. Therefore we have to close the loop using feedback. The output from the VCO to the Prop is biased so the sinewave straddles Vdd/2. That way, the Prop will see a square wave of the VCO's frequency with about a 50% duty cycle. By generating a square wave of the desired frequency using one of the counter's NCO or PLL modes, we have something to compare the VCO's frequency to, in order to apply a correction. Is this beginning to sound like a phase-locked loop (PLL)? Well, that's what it is -- but in software.
In order to operate as a PLL, we need a phase detector. One common phase detector used in PLLs is the following:
As shown, when the VCO and target waveforms are exactly 90° out of phase, the pulse widths from the two outputs are equal. As the phase relationship changes, the pulse width ratio changes, and this information can be used to make corrections. When the VCO lags the Control, you increase the VCO frequency slightly, until it catches up; and vice-versa when the VCO leads the Control.
Once again, the Propeller's amazing counters can do the heavy lifting for us. These counters have a logic input mode that's tailor-made for implementing a combinatorial phase detector like this one. All we have to do is set it to count up for a period of time, as long as both inputs are low, then for the same period, obtain a count when one input is high and the other is low. By subtracting one count from the other, we get the relative amount of change required to correct the VCO control level.
One problem with this phase detector is that it does not have a very wide capture range. This is to say that if the two frequencies differ by more than a few kilohertz, it won't be possible to determine which way to make the correction, and the VCO will wander until the frequency gets close enough to capture and lock on. What's needed is a separate feedback mechanism that monitors frequency and makes larger corrections to the VCO control voltage until they are close enough for the phase detector to take over. This is done by counting edges for a period equal to clkfreq / 1024, or about 1ms. This count is then compared to the expected count from the chosen frequency, and the VCO frequency is adjusted until the difference is zero. Then the phase detector can take over.
In the attached program ctrb is used to control the DUTY mode output to the VCO. ctra is used, in turns, for the frequency counter, and the two phase sensors. The results turned out extremely well, with the locked-on VCO frequency never wandering by even 1Hz! Here is a scope trace of the VCO locked on to a reference clock from the Prop:
The red trace is an FFT plot of the VCO output. In contrast to those in my prior thread, there are virtually no harmonics present. One thing you will notice from the reference trace is the phase jitter from the Prop. But the phase jitter is completely absent from the VCO output. This is something that would be all too apparent while listening to the output on an AM receiver with a BFO (beat frequency oscillator) turned on, and none was noticeable when I checked it.
One might wonder: why not just use a DDS (direct digital synthesis) chip and be done with it? Well, DDS ICs are a little expensive, and they still require some external passive components for proper operation. Also, if you're going to use the Propeller in a radio applicaiton, why let it loaf when it could be doing a lot of the grunt work?
-Phil
Post Edited (Phil Pilgrim (PhiPi)) : 7/8/2010 10:50:10 PM GMT
867 x 307 - 5K
723 x 127 - 6K
640 x 480 - 19K
Comments
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Leon Heller
Amateur radio callsign: G1HSM
Yes, I agree. The '4046's phase detector II is what I used in my prior thread, and it gave excellent results. The problem for the Prop, though, is that it's a stateful phase comparator and too complicated to implement with the counters. I suppose I could do it in software at low frequencies, but I was hoping eventually to implement this at 14MHz.
-Phil
The top (yellow) trace is the output of the VCO. The middle (cyan) trace is the reference frequency generated by the Propeller. The bottom (magenta) trace indicates whether the PLL software is correcting the frequency (high) or the phase (low).
The reference frequency is switching back and forth between 4.260MHz and 4.460MHz, a difference of 200KHz. As you can see, when the switch is made, the PLL software reverts momentarily to frequency mode to make the rapid corrections, then back to phase mode to acquire and maintain a phase lock. Also, the Prop's output jitter is plainly visible, but it does not seem to affect the stability of the VCO's output. (Yes, I realize I'm triggering on the VCO signal, which would natrually make it look more stable, but other tests have confirmed a lack of jitter.)
For some reason I do not yet understand, the software acquires a lock more quickly when transitioning lower in frequency than when transitioning higher. But both transition rates are acceptable for the app I have in mind. I just hope I can get it to work upwards of 14MHz. I got some different varactors today from Mouser, but they're in 0402 packages! I should've looked more closely at the datasheet before I ordered them. Now where did I put that loupe ...
-Phil
I wonder if there is a translation between the jitter from the Prop and amplitude of the output of the VCO.
Could you zoom in to the upper or lower peaks of the VCO output?
Which leads to another question, is the jitter biased so that the net amount of jitter is zero, or is there a measurable drift in one direction or the other?
...And how would the output of the PLL compare to the output of a tuned LC tank at say 4.36MHz?
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
The longer time constant in the software control loop, along with the DUTY mode filtering, is what keeps the jitter from appearing in the VCO's output. I zoomed in on the VCO peaks but didn't see any variation in amplitude -- at least none that could be detected at 100mV per division. (The VCO amplitude variation seen between the two frequencies is a function of the transistor oscillator, which has no built-in AGC capability. Nontheless, I'm surprised by the amount of change, given the small frequency deviation.)
The jitter from the Prop comes from the fact that the NCO's frqa register has a lot of one bits set. It's not biased and definitely doesn't drift, although the jitter's overall period will depend on the distance between the most- and least-significant one bits in frqa. I'd get slightly less jitter if I used the counter's PLL mode and probably will at higher frequencies.
'Not sure what you mean by the last question, since the VCO already uses a tuned LC tank. Can you rephrase, please?
-Phil
'Not sure what you mean by the last question, since the VCO already uses a tuned LC tank. Can you rephrase, please?
I'm just curious what a side by side would look like with the output you have compared to a tuned LC output fixed at 4.36MHz directly driven from the Propeller I/O at the two frequencies you tested with but now that I look at it 200kHz is a bit of a spread for a fixed LC setup.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
Thanks for the suggestion!
-Phil
This opens some new possibilities alright!
-Phil
"The LC tank alone seems to calm the Prop's jitters." - Bahhh humbug in July!!
I think though there is a small amount of translation, but I can't see it with my equip. ... basically any jitter should slightly affect the phase at which you are resonating your LC tank. If the bias of the jitter has a net of Zero, then there shouldn't be any drift over the long term. However because the jitter is either in phase or slightly out of phase with the resonating circuit this 'should' cause an amplitude translation in the output of the LC tank. ... however small, it still should be present.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
I assume the degree of the calming effect depends on the Q of the tank. IOW, a low-pass or band-pass filter would not have the same effect, right?
-Phil
In a way the LC tank is a bandpass filter .... errr a 'notch' filter rather. The lower the Q, the wider the bandwidth that's allowed to pass.
Since the associated jitter has for the most part a high frequency component, then a low-pass filter might work if constructed from LC components. An RC low-pass filter I wouldn't even consider.
Think of an LC tank as a swing at a playground ... the square wave that you put into the LC tank from the I/O is equivalent to when you would push (or pull, or both) the swing. If you are in 'sync' with pushing or pulling then you are in resonance or peak efficiency with the swing. If your rate changes and you are slightly faster or slower, then the efficiency goes down and you lose momentum or amplitude ... In theory, this is what jitter should accomplish. If the net jitter is not Zero, then I would expect to see a low frequency beat frequency superimposed within the signal of interest.
... that said, without the reciprocating (symbiotic-handshaking) associated with an LC, you don't get a typical sine wave action. Instead, with an RC filter, the edges of a square wave (rising/falling) are rounded because the high frequency components are removed from the low-pass filter, but in the body of the square wave (either rail) there is no filtering. So the illusion at high frequency is that an RC filter will produce a sine wave. With an LC filter, the momentum is carried through the 'body of the square wave' and a valid sine wave is produced.
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
The code I used to test it is attached. The procedure was this: output a square wave to drive the LC tank, then adjust the varactor capacitance to bring it to resonance with the square wave. Locating resonance turned out to be pretty easy. The reactance of an LC tank at resonance is purely resistive, which means that the drivng waveform and the output waveform will be in phase. Off resonance, the output waveform will be out of phase with the driving waveform in a direction dependent on the difference between the driving frequency and the LC's resonant frequency. So this is just another exercise in using the Prop as a PLL to phase lock the two signals. There is one wrinkle, however. With a stateless phase detector, which is the only kind possible using the Prop's counters, it's not possible to determine the directiion of phase shift between two signals in order to bring them into phase. Recall from the previous example that the phases were locked 90° apart. So, in order to keep the input and output signals in phase, another signal, in quadrature with the first one, has to be produced, to which the sine wave signal is locked 90° out of phase. This required another cog, but it worked quite well and lock was obtained quickly.
So I tuned in to the signal on a shortwave receiver to discern its purity. Frankly, it didn't sound as clean as the VCO-generated tone against the BFO, having a slight "fuzz" impressed upon it. But the real problem was one I discovered accidentally. When tuned 200KHz above the signal, I heard it again. When I checked 200KHz below the signal, there it was once more! Had I just adjusted the FFT in my scope to look for them, the extra signals would have been plain as day, as this trace shows:
These spurious sidebands are the direct consequence of phase jitter and are not harmonics of the fundamental frequency. They present a Catch 22 situation for this method of signal generation for the following reason: Suppose we could increase the selectivity of the LC tank by boosting its Q enough to filter out the "spurs", as they're called. The problem with this is that, off-resonance, there would not be any output signal to compare phases with to determine the direction of adjustment for the varactor. Could it be done with multiple stages of filtering? Perhaps. But each stage would require its own frequency adjustment, based on the output from the first stage. To pull that off would require highly-matched inductors and varactors. I don't see it as a viable solution.
Here's an article that demonstrates another method for eliminating jitter: www.intersil.com/data/tb/tb318.pdf But again, that's more complicated than phase locking an external VCO.
So it looks like it's back to either the VCO method or to filtering an external PLL like the '4046. One thing that's difficult about using varactors these days is that the varactor manufacturers have abandoned the HF market. The BB212 device I used for these tests is no longer manufactured, and only lower-capacitance devices are available. But those limit the available adjustment bandwidth to a few tens of kilohertz, entailing a need to switch additional capacitance or inductance in and out of the tank circuit. So, I'm not sure what's next. If Chip is successful in creating spectrally-pure sine waves with the Prop II, that could be the ultimate answer.
-Phil
Post Edited (Phil Pilgrim (PhiPi)) : 7/11/2010 7:47:17 PM GMT
rats!! ... thanks for giving it a go
▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Beau Schwabe
IC Layout Engineer
Parallax, Inc.
You mentioned briefly switching in capacitors to alter the resonant frequency of the tank - would this have a chance of working? What are the separations of the frequency channels you're chasing around the 14 Mhz area?
My ultimate goal is a ham radio transceiver that can work at frequencies up to the 20-meter band (14.000 - 14.350 MHz). For transmitting, the signal has to be squeaky clean, so that's why my obsession with removing the jitters.
The idea of switching caps in and out of the LC tank is just to overcome the limited capacitance range of currently-available varactors. They would act like band switches, but for selecting amongst narrow partitions within a single band. Basically, you just use a MOSFET to ground the ground pin of each cap to activate it in the LC tank. I'm sure it would work, along with one varactor for trimming purposes. But I'm disappointed at the prospect of using extra Prop pins to do it. The single DUTY output to a wide-range varactor seemed so clean.
I'm really interested in your XOR phase detector and PLL work. How did you implement it? What kind of lock range did you get with it? What was your app?
-Phil
I never did apply a PLL to it, which is why your setup is so interesting. I was investigating a simple phase meter to replace some aging equipment here, but there may be applications in touch sensing or other areas. Having 8 or perhaps even 16 detectors on a single chip opens up possibilities. Here's the code snippet. While A<>B makes sense for phase detection, I cannot explain why I was also measuring A==B, rather than something that would have determined phase lead/lag. I think I was chasing down some artefact whereby the sum total counts across both were always the same as the 500_000 clock interval. I hope to fire this up again later this week.
The varicaps I have are ISV149.
regards
Lachlan
--
'**** set up counters
dira[noparse][[/noparse] 0 ]~ 'set as input
dira[noparse][[/noparse] 1 ]~
FRQA:=1
FRQB:=1
CTRA:=%11001_000 <<23 + 1 <<9 + 0 'A==B counter
CTRB:=%10110_000 <<23 + 1 <<9 + 0 'A<>B counter
PHSA:=0
PHSB:=0
' TempA:=Cognew (Ctr2(@X),@Stack1)
Time := cnt
repeat
Time +=500_000
waitcnt(Time)
'store reading then restart
TempA :=PHSA
TempB:=PHSB
PHSA:=0
PHSB:=0
debug.out(" ")
debug.dec(TempA)
debug.out(",")
debug.out(" ")
debug.dec(TempB)
debug.out(",")
debug.dec(Ctr1A)
I'm thinking of building the VCO shown in your schematic in first post of this thread, and then trying to modify it for use on the 30 meter ham band. Thought I'd check first to see if you have made any mod's to the circuit since this first post. (I see that the raw tank circuit was tested and did not work as well.)
Do you think it might work at 10Mhz with changes in the resonant circuit?
Thanks for your help.
Michael KD4SGN
I haven't done much with the circuit since presenting it here. I don't see why it couldn't work at 10 MHz, though. The difficulty will be finding varactors with a wide-enough tuning range to be useful. My future plans are to use the oscillator in an IC mixer (e.g. SA612) to convert inputs in a receiver down (or up) to 10 MHz, an IF that's easy to filter with a compact crystal lattice and to detect I/Q-style in the Propeller itself. Again, the varactor issue is a thorny one for HF, and I might have to resort to switching discrete caps in and out of the tank circuit to get the band coverage. (I wonder if there's some sort of C2C analog of the R2R matrix used for DACs.)
-Phil
AD7YF
For use on the digital modes of the 30 meter band, the VCO doesn't have to have as much bandwidth. I have some smaller capacitance varactors here that I plan to try out.
73,
Michael KD4SGN
Lawson
Thanks for the references!
The BB212 covers 500pF to 140pF over 0-3V. Some experimenters have used 1N4007s. The problem for design, though, is that diodes that are not designed to be sold as varactors are not screened for capacitance, so it's hard to know what you're getting from batch to batch.
Tuning a ferrite inductor with a DC bias sounds very intriguing. I had never heard of that technique before and would love to get my hands on more info. Can you recommend some reading for me?
Thanks again!
-Phil
As far as parts, I'd look for small RF "saturable reactors", these parts are designed to do exactly what I propose. Another good target would be small power transformers intended for "forward conver" DC-DC power supplies, as these would be optimized for low magnetization current and high flux. After that, just look for inductors that have much larger maximum current than saturation current so they can be saturated continuously, or for testing just about any inductor should work pulsed. (i.e. take an inductor with a 10mA saturation current and dump 100mA through it for a little bit)
Lawson
445-6150-1-ND might be a good target? Looks like it could be pushed well past 100mA (~25mA saturation current) if it was soldered between large thermal pads and operated at room temperature.
-Phil
Yes, two independent wraps would work well. Current requirements to saturate depend on the core material and configuration. It can be quite low if the core has high permeability and no air gaps. Most cores designed for inductors have relatively low permeability (reduces L by 'x') so more turns (increases L by 'x^2') can be put on a core before it saturates.
Was looking at parts for this a bit more, Ethernet isolation transformers are likely to work well for this.
The biggest potential problem I see is that the AC current circulating in the LC tank circuit will also effect the inductance. (I'd expect the vari-cap circuit to suffer from a similar effect) If the circuit drives too much AC current through the inductor the output will start to look "peaky".
Lawson
P.S. I'd expect the photo-diodes I linked earlier to have a tight distribution of capacitance. Photo-diode circuits REALLY care about the diode's capacitance, so suddenly getting 2x would be grounds for a vendor change.
BTW, I've got a nice collection of small toroids now (FT37-43, FT37-61, T37-2, T37-6), so there may be an experiment in the offing!
-Phil
Magnetics... Is this the same principle as the magnetic amplifier?
Lawson
I must admit you really crack me up at times.
Bruce