multi-phase oscillator

[zachary vex]

i've seen a multi-phase sine oscillator somewhere and i can't remember where.  it had 0, 90, 180, and 270 degree outputs.  any ideas?

[gez]

EPE magazine had a few sketches for CMOS 'loop' oscillators.  You basically run inverters in a loop (similar to a state variable filter) and use an auto gain circuit to control amplitude.  Inverters outputs are tapped to provide phase-shifted outputs...

[airhole]

Hi Mr Zvex,

I don't know what you are looking for, but can i dare suggest a quadrature oscillator with inverters?

i.e. quadrature oscillator outputs 2 waveforms 90 degrees from each other, buffer them, then invert them, hence obtain 4 waveforms?

cheers,
george

[puretube]

Bode/Moog;
J. Haible;
:wink:

[Paul Perry (Frostwave)]

http://www.ti.com/sc/docs/apps/msp/journal/aug2000/aug_07.pdf.

might help. Who can resist an oscillator called "Bubba" (unless I was hallucinating) :shock:

[zachary vex]

thanks, everybody.  excellent references.

[StephenGiles]

Hey zachary - what are you up to out there? Quadrature oscillators mean only one thing to me, possibly 2 - if I'm right, I look forward to seeing it!
Stephen

[Nasse]

There was some papers at Mark Hammer pages if I remember, more than one story

I had an idea some time ago about an electro-mechanical multi phase generator  :lol:  :shock:

[zachary vex]

Hey zachary - what are you up to out there? Quadrature oscillators mean only one thing to me, possibly 2 - if I'm right, I look forward to seeing it!
Stephen

8^)  not what you'd think.

[puretube]

I do hope, you`re not after the 3rd ("my") thing...

[Paul Perry (Frostwave)]

Quadrature oscillators??
let me see.. frequency shifters, feedback eliminators, barberpole illusion, leslie simulators, laser displays...
and the concertina time warp stretcher, but that is another story :wink:

[DiyFreaque]

Then there are tri-phase oscillators - 0, 120 and 240 degree outputs.  Used in the classic ensemble choruses.  Two tri-phase oscillators, three BBD's and you got yerself the makings for a Solina chorus.

[R.G.]

Save yourself some trouble. Use CMOS shift registers and shaping resistors to make ring counters. Any variable speed digital oscillator plus some ring counters makes for as many phases as you want to make the resistor summing networks. It's easy to get two, three, four, six phases.

The stairstepping is easily filtered out by a simple RC, although an active filter gives you primo results. Read up on it in Lancasters CMOS Cookbook.

The other way is with phase shift oscillators. You can get quadrature with two integrators and an inverter, but three integrators gives you three phase. Sounds simple, right?

Yeah, but like all analog oscillators, it has critical gains to be taken care of and you need to have some kind of gain bounding. Clipping works, but introduces distortion, usually worse distortion than with digital synthesis. By the time you get through with a good one, you've spent as many chips as with digital and a lot more time.

The other answer is - get a PIC. A PIC can output you multi-phase LFOs directly in PWM format, as well as reading the speed pot.

Better living through science.

[airhole]

Hey R.G,

You da man! Thanks for the good read!

cheers,
George

[puretube]

well, err, ehh, of course I know of a "different concept" quad-opamp
constant amplitude quadrature sinewave generator with very low parts count, which can go from once in 2 minutes to 1kHz without switching caps (something you wouldn`t even want to dial in with 1 knob  :roll: ),
but that probably will hit the market only after it visited the patent office...  :wink:

btw.: analogue, of course - and: the "different concept" is successfully tested with 12AX7`s, too...  :wink:

and: it does respect Barkhausen`s criteria...

[gez]

Any variable speed digital oscillator plus some ring counters makes for as many phases as you want to make the resistor summing networks. It's easy to get two, three, four, six phases

RG, if the outputs from each counter are all summed in parallel, how can you tap the phase relationships?  Presumably this is done using another chip?  Could you elaborate please?

[R.G.]

well, err, ehh, of course I know of a "different concept" quad-opamp
constant amplitude quadrature sinewave generator with very low parts count, which can go from once in 2 minutes to 1kHz without switching caps (something you wouldn`t even want to dial in with 1 knob  ),
but that probably will hit the market only after it visited the patent office...

Yeah, so do I. Check prior art pretty thoroughly before spending the money on patent filings. If it's the one I know of, it has its own problems, though.

Shoot, maybe I ought to file some of these old circuits. The patent office is pretty lax on prior art searches these days.

RG, if the outputs from each counter are all summed in parallel, how can you tap the phase relationships? Presumably this is done using another chip? Could you elaborate please?

Well, first you start with a ring counter. This is not your ordinary binary counter, but an adaptation of a shift register that (usually) shifts a load 1/2 the ring's length of 1's and half of 0's around the ring like a snake chasing its tail. It *is* a counter, but you have to decode the bits to get a binary counting-up or -down word.

But binary counting is not all it's good for. The pattern of a block of 1's and 0's chasing themselves around the ring lends itself to decoding easily.  This is how the CD4017 works - it's a ring counter (also called a Johnson counter) inside with decoding so you get ten decoded outputs with much simpler internal logic and more importantly, no glitches, than you'd get with a binary and a decoder.

If, for instance, you wanted a sine wave out of this thing, you could set up an eight stage shift register with an inverter feeding back from the last Q output to the input. At power on, all the stages are set to 0, and this positions a 1 from the inverter to be shifted in. With each clock, a 1 is shifted into the input until the first 1 gets to the output. Then the inverter feedback feeds in a 0. So it starts with all 0's, feeds in a 1 from the input until all the stages are 1, then starts filling with 0's.

As you recognize, this is inherently a phase-preserving setup. If you connect up resistors from the shift register outputs in carefully chosen values, you can connect them all together and get a stepped approximation of a sine wave - or triangle wave, if you use different resistors - at the summing junction. In the case of the eight stage counter, you have to clock the thing sixteen times to get a full sine wave, so the sine wave comes out at 1/16th of the digital clock's frequency.

If you do a spectrum analysis on the waveform, you find that the first distortion harmonic (if you've done a good job on the resistor values!) is the seventh and the next is the ninth. You can make *good* sine waves this way. The filtering is easy, as the harminics are well separated from the fundamental.

And shift registers are cheap. It's almost as easy to make a 24 stage counter as it is an 8, so you can run the sine wvae distortion way down.

But back to the phase. If you think about it, the phase of the recovered sine wave just depends on which stages you hook the summer network to. If you take them off at two different places in the ring, the two summer networks have the same waveform, but different phases. Quadrature is easy with rings that have multiples of 8 stages.

There is one fluke that I discovered trying to make this practical. While we conceive of equal blocks of 1's and 0's chasing themselves around a ring, the real Johnson counter only implements half the ring. That is, when all the stages are filled with 1's, we conceive of there being an equal number of 0's on the "back" side of the ring being shifted around, which in fact there is not - there's only the feedback inverter generating those 'hidden' 0's for us.

So to make long-ring multiphase LFOs out of ring counters, you have to have a shift register extend on out from the point that the feedback is taken off to contain the extra phase bits that are 'hidden' in the back of the ring that doesn't exist. For instance, eight stage ring counter making a sine wave only needs 8 stages. If you want quadrature, you need to put four more stages on the end of the ring to have enough outputs to generate the second output. This end extension has no feedback, its bits just shift off into the ether. It's just a place to hold the extra bits to make the second phase.

This takes a lot longer to type out than to see operating on a simulator.

But then I guess most things do.

[gez]

So to make long-ring multiphase LFOs out of ring counters, you have to have a shift register extend on out from the point that the feedback is taken off to contain the extra phase bits that are 'hidden' in the back of the ring that doesn't exist

Ahh, everything is illuminated!!  

This takes a lot longer to type out than to see operating on a simulator

.

Thank you for taking the time, much appreciated.

[StephenGiles]

Seems perfectly straightforward to me!  :lol:  :lol:  :lol:  :lol:  :lol:

[Paul Perry (Frostwave)]

Can  I be the first to say "you can do this in a PIC?" :wink:

OK, to apologise for being a smatrarse, here is a useful link:
http://tinaja.com/magsn01.asp

As for making money by patenting ANY kind of analog oscillator
in 2004, "A day late and a dollar short"