[sdiy] Idea - Triangle wave DCO core

[Magnus Danielson]

On 03/14/2012 11:56 AM, Neil Johnson wrote:
> Hi Tom,
>
>> It occurred to me today that you could fairly easily build a triangle core DCO. This wouldn't use a reset pulse, but would instead use a bipolar DAC.
>>
>> You'd simply output a charging CV from a DAC to an integrator. After half a cycle has been timed by the uP, it switches the DAC charging CV to the opposite polarity. And so on. Instant triangle wave at known frequency.
>>
>> Has anyone ever done/seen anything like this?
>>
>> Presumably once you'd got the basic version working, you could tweak the rise/fall CV calculation to get variable waveforms between ramp(ish) and triangle. Aside from that, I can't see any massive advantages beyond simplicity.
>>
>> What do you think?
>
> You'd need some sort of servo mechanism around the integrator to deal
> with any DC drift, otherwise you'll end up hitting one of the rails
> sooner or later as DC terms get integrated over time.

It suffice to assist the integrator cap leakage. Typically using a 
resistor across the cap. It just raises the cut-off frequency of the 
integration, which would degrade over time anyway due to accumulation of 
contamination on the board.

> A sawtooth is easy as you always reset to a known starting position.
> With the scheme you propose there is no known starting position -- all
> you're doing is changing direction at known points in time, but you
> have some control over how far each segment goes.  The integrator will
> integrate all the current flowing into it, which as well as the output
> of the DAC will also include any op-amp offsets, noise, etc.  You're
> also relying on the DAC to output EXACTLY the same magnitude of
> current for both signs.  Unlikely in reality.
>
> Analogue triangle cores don't suffer this as the timing and direction
> is handled by level-sensitive comparators.  The handle you turn to set
> frequency is solely the integrator current.  Since the amplitude is
> set by the comparators then any DC offsets simply disappear.

Indeed. Still, the "flip-time" can still cause a shift, but I suspect 
much lower, the expo converter is more likely to cause worse deviations.

Cheers,
Magnus