Is there a way to extrapolate this D flip flop-based/XOR modified sine-wave generator to be a little less "steppy" using a couple more flops or XOR gates? or provide outputs of different phases?
I was thinking about a thing using the TinyLogic series flops/gates to clock it very fast.
Unfortunately my LTSpice sim of it with 74HC logic models shows glitching on some transitions that becomes worse with speed:
Don't know yet if shows up in the real circuit.
Those glitches look real to me. (it's the propogation time of the second flip-flop) but if you're going to low-pass the output they'll disappear. ther'll only be a slight saw-toothiness caused by that propogation delay.
If you want glitch-free build a 3 stage johnson counter
-- When I tried casting out nines I made a hash of it.
Probably not.
offers a more sensible approach. You cook up a long binary number where the mark-to-space varies along it's length, and cycle it past a low-pass filter.
Carefully chosen numbers can have zero low harmonic content, but they do seem to be quite long.
Direct digital synthesis chips are a lot more practical.
-- Bill Sloman, Sydney
Bill Sloman wrote in news: snipped-for-privacy@googlegroups.com:
Hehehe. Run it through one of the distortion pedals guitarists use. Definitely has harmonics. ;-)
I suspect the glitches are built-in.
A square wave can be seen as the sum of all the odd harmonics of a sine wav e of the same period (with the harmonic amplitude decreasing in proportion to harmonic number).
Bitrex is knocking out the third harmonic content of his basic square wave by subtracting out the first harmonic content of his triple frequency squar e wave, but this doesn't help with higher harmonics - the odd harmonics of three time the frequency aren't all the odd harmonics of the base frequency .
Not a great idea, and it doesn't get better if you try to elaborate it.
-- Bill Sloman, Sydney
The faster you try to clock it the worse things will get with propagation delays in the flip flops as drawn.
To take out most of the 5th harmonic component you will need a divide by
5 counter in the right phase. Each extra component you try to compensate seriously limits the maximum output frequency that it will work at. The fundamental will be lowered from f/3 to f/15 by adding this factor.The least bad fix might be to treat the output as a current source and integrate it so that you get a step wise linear approximation. (or aggressively low pass filter against the fifth harmonic and above).
What are you trying to do?
-- Regards, Martin Brown
Well, you can go for small improvements with the 'magic sine waves' approach, and get a factor of two or four, or the more-parts solution of switched capacitor filtering (clock at a hundred times the sinewave frequency, there's an MF-10 application note on this somewhere).
The issues, though, are that you only push the harmonic distortions higher in frequency, they don't go away unless you ALSO have a filter - and that will limit any adjustment range for the sinewave frequency.
For direct synthesis, any of the lossless audio schemes can put a high quality sinewave onto your iPod, or boxes with buttons on front can synthesize the waveforms offered in the menus; usually sine, triangle, square and sometimes others.
If you think that is bad, try looking at the clock input to the first FF.
The circuit is a divide by three to generate (1) and the inverted signal (2 ), the slowest input to the two bit DAC. That signal should not be glitchy . The clock signal is the fastest signal, again, not glitchy. When the cl ock signal changes it takes time for that signal to propagate through the X OR and the two FFs to be combined with the clock signal in the DAC so you g et the glitches as the clock changes first, then the signal (2).
To drive a glitchless DAC output with N values, you would need N-1 FFs driv en in a thermometer code. This is very practical and would be glitchless s ince the glitches come from race conditions and a thermometer code only cha nges one bit at a time. A Johnson ring counter would do the job just fine. So an 8 bit register chip and an inverter (a chip or a transistor) would provide a thermometer code that first fills all the bits, then empties them in the same direction.
000 100 110 111 011 001 000Use equal weight resistors on the outputs and you get a triangle wave. Tai lor the weights with larger resistors at the ends of the thermometer code t o shape it more like a sine wave. I'll let you do the math. You can casca de more than one octal register and get a very good sine wave.
-- Rick C. - Get 2,000 miles of free Supercharging - Tesla referral code - https://ts.la/richard11209
o
(2), the slowest input to the two bit DAC. That signal should not be glitc hy. The clock signal is the fastest signal, again, not glitchy. When the clock signal changes it takes time for that signal to propagate through the XOR and the two FFs to be combined with the clock signal in the DAC so you get the glitches as the clock changes first, then the signal (2).
iven in a thermometer code. This is very practical and would be glitchless since the glitches come from race conditions and a thermometer code only c hanges one bit at a time. A Johnson ring counter would do the job just fin e. So an 8 bit register chip and an inverter (a chip or a transistor) woul d provide a thermometer code that first fills all the bits, then empties th em in the same direction.
ailor the weights with larger resistors at the ends of the thermometer code to shape it more like a sine wave. I'll let you do the math. You can cas cade more than one octal register and get a very good sine wave.
If you use more and more bits in creating your sine wave both the frequency and amplitude of your harmonics are reduced. Filtering may or may not be needed, but will be much more effective by the smaller step sizes.
I use a sigma-delta DAC in a design I build which shows staircase steps in the output waveforms. The steps are small and well above the audio range, so not a problem. Use enough FFs and resistors and your harmonics will be out of the audible range so no filter will be required unless there is some sensitive equipment in the path which I highly doubt.
-- Rick C. + Get 2,000 miles of free Supercharging + Tesla referral code - https://ts.la/richard11209
I have an I/Q demodulator/detector scheme that has to run pretty fast at very low voltage, the 0 degree phase path goes through some op amps with somewhat limited bandwidth and slew rate because they're low voltage types. it's fed by a reference clock.
the whole system would work fine with square waves if I had the budget for infinite GBW and slew rate parts but the op amps I can afford on the budget in the path don't really have enough GBW and slew rate to support a square wave at the speed I need to go so, I don't want to shove a raw square into them like a brutal person.
this would work fine to produce a modified sine and a quadrature square wave and is just what I need, aside from the glitches which causes problems down the line. using lower value resistors I can just drive the input of the analog section directly from the flop outputs, the amps are single-supply biased at the mid-point of the same supply and just drive it in, nice.
I want to maintain a tight phase relationship between the two signals so IDK if aggressive low passing is an option here. a couple flops and gates is within my price/power budget so I'm wishing there was a way to eliminate the glitch at the source without piling on parts, at that point I'd likely just resort to DDS. but I'd prefer to be able to provide a solution that works OK driven by a clock from whatever source in this case
Yep, the input to the XOR and the output of the XOR (in red) shows why the glitch happens:
Looking at the waveforms it looks like it might be possible to "lock out" the glitch by using another flop and gate or two to synchronize the timing of the inputs to the XOR. that would also be satisfactory
That violates all the usual "rules" of logic design. The clock to the first flop is gated, and the pair of flops is a ripple counter. Expect worse glitches if more flops are added in this style.
From the age of the drawing and the site it's on looks like it was designed by communists!!! they probably never ran it fast with their commmunist-logic so it was okay.
The glitch comes from this issue right here with the XOR, with the output in red and the other two colors inputs:
Without it, it does produce a quite nice modified sine with a minimum of parts. I think it might be possible to lock out the glitch by synchronizing the XOR input with a third flop, maybe.
If we are to build a successful gay space communist future in space, for everyone, then we must correct the failures of the past!
o
F.
l (2), the slowest input to the two bit DAC. That signal should not be gli tchy. The clock signal is the fastest signal, again, not glitchy. When th e clock signal changes it takes time for that signal to propagate through t he XOR and the two FFs to be combined with the clock signal in the DAC so y ou get the glitches as the clock changes first, then the signal (2).
Yep.
driven in a thermometer code. This is very practical and would be glitchle ss since the glitches come from race conditions and a thermometer code only changes one bit at a time. A Johnson ring counter would do the job just f ine. So an 8 bit register chip and an inverter (a chip or a transistor) wo uld provide a thermometer code that first fills all the bits, then empties them in the same direction.
Tailor the weights with larger resistors at the ends of the thermometer co de to shape it more like a sine wave. I'll let you do the math. You can c ascade more than one octal register and get a very good sine wave.
That way lies madness. You said you wanted a better sinewave. Why not use more FFs in a single chip? You can find lots of chips with 8 FFs and if y ou dig a bit I think you can find 10. Why is one 8 bit chip worse than one 2 bit chip? You will be amazed at the improvement in the quality of the s ine wave.
It will require a faster clock. Is that the issue?
-- Rick C. -- Get 2,000 miles of free Supercharging -- Tesla referral code - https://ts.la/richard11209
e
to
FF.
nal (2), the slowest input to the two bit DAC. That signal should not be g litchy. The clock signal is the fastest signal, again, not glitchy. When the clock signal changes it takes time for that signal to propagate through the XOR and the two FFs to be combined with the clock signal in the DAC so you get the glitches as the clock changes first, then the signal (2).
s driven in a thermometer code. This is very practical and would be glitch less since the glitches come from race conditions and a thermometer code on ly changes one bit at a time. A Johnson ring counter would do the job just fine. So an 8 bit register chip and an inverter (a chip or a transistor) would provide a thermometer code that first fills all the bits, then emptie s them in the same direction.
. Tailor the weights with larger resistors at the ends of the thermometer code to shape it more like a sine wave. I'll let you do the math. You can cascade more than one octal register and get a very good sine wave.
ese more FFs in a single chip? You can find lots of chips with 8 FFs and if you dig a bit I think you can find 10. Why is one 8 bit chip worse than o ne 2 bit chip? You will be amazed at the improvement in the quality of the sine wave.
And just to be clear, this design produces no glitches. The glitches are b ecause more than one signal is changing at a time and the timing will never be perfect. So work with a design where only one signal changes at a time . Johnson ring counter is the way. Use the force Luke.
-- Rick C. -+ Get 2,000 miles of free Supercharging -+ Tesla referral code - https://ts.la/richard11209
How about a 2 flip flop Johnson ring counter then? That can be configured to generate the quadrature square waves.
Integrate them to a triangle wave and use diode shaping. How wide a range of frequencies does this have to work at?
-- Regards, Martin Brown
Just one, around 200-300kHz. Here's the drive current into my load thru the op amp network at about 200k (slight visible phase shift is due to the AD8515 being one of the few low voltage op amps I can find in the LTSPice library, I'll likely be using the twice-as-fast OPA4332):
If not for the glitchin' it'd be near perfect, with only 3 parts. I'm on a pretty tight budget for per-unit cost...had to blow most of the budget on fast low voltage op amp....frustrating!
I don't think the glitches will affect the performance much...I may just try to live with it...compromise...:(
That range allows for some options; the 'integrate' is a filter that takes the harmonics down 3 dB/octave, but you can just as easily use a double-tuned IF transformer, and get 12 dB/octave. Some of 'em can be slug-tuned, too. It's power-stingy, and in quantity, those little wire twists are really inexpensive.
Standard off-the-shelf designs would take some external capacitors to pull to your frequency range (from 457 kHz), but a manufacturer can build them for any center frequency, and with your choice of input and output impedance.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.