Tracking N-path filter

Before the question is answerable you will have to describe what you are actually trying to do and at what sweep rates and frequencies.

My instinct is that in the digital domain you would be better off doing the sweep by subsampling and/or interpolating into a largish DAC lookup table by the half cycle. Depends how much jitter you can live with.

There are surprisingly good DDS chips these days for not that much.

The VCO in a '4046 gives a square wave; you'd have to integrate that to get a triangle, and it'd have frequency-dependent amplitude. For controlled triangle VCO, it takes voltage-current conversion and current polarity switching (LM13700 and a comparator, for instance) and a capacitor.

All the digital tricks need a low-pass filter, and a tracking low-pass is a LOT of work. I'd consider, first, doing two 10 MHz-ish oscillators, mixing down, and THEN the low-frequency just has to stop 10MHz-ish, and 20 MHz-and-up. Something like a classical IF strip of LC's is a very nice, passive, filter for a known stopband that's not too wide, and there's SAW filters for a high end.

That is a very strange chip.

A cheap uP with an onboard DAC can do pretty good software DDS sine wave generation up to at least 10s of KHz.

A shift register with the right value resistor on each tap can make a tracking low pass filter. You sum the currents through each of the resistors into a virtual, and what comes out can be a pretty respectable sine wave. It's a finite impulse response digital filter with most of the computation handled by the analog output.

I did it once for a gadget that was designed to confuse echo-locating bats in a predictable way. The customer recently got a one page obituary in Acoustics Today, so it was a while ago.

Looks like it's two separate chips that then got interconnected in the same package, somehow. There's a "Matrix 0" and "Matrix 1" in the designer, each have their own parts, and can only talk to each other over a fixed number of lines. There's a lot of bits & bobs to make stuff out of in there, though.

Done that and it's fine but this is a push-the-limits experiment, hacking an analog filter into a part that's not really designed to do it.

The MF10 active filter IC was cool. The lead time at Mouser is May of

On a sunny day (Fri, 1 Apr 2022 15:54:17 -0700 (PDT)) it happened whit3rd snipped-for-privacy@gmail.com wrote in snipped-for-privacy@googlegroups.com:

No idea what he is actually doing, but indeed 4046 My old signal generator was a 4046 driving a binary counter that was connected to an EPROM programmed with a 256 steps sine wave... Nice variable frequency sine wave output in the audio range.

Yes, a switched cap low pass filter could be all you need. If LPF is high (like 5th) order then don't even need to premake a stepped sine input, a square wave is enough. AoE has this since at least Ed 2 (poss even Ed 1).

piglet

MF10s were awful. They were super noisy and aliased like crazy.

N-paths should alias too.

Yes, that has good possibilities; if you use a table for the sine, there's going to be steps (jaggies) at the table granularity, which means a 1000-point table of sines (one quadrant of a wave) gives at least .001 of the peak amplitude as 'steps'. DAC step size is also a granularity, assumed small. A FIR filter with analog components only needs a fast summing junction and a bevy of hold capacitors, and a tracking clock (easier to make a clock track than a passive filter element) and thus smooth the steps. Expect a bit of switching noise, still, but a CCD array of capacitors can make an on-the-chip integrated implementation.

Is there an off-the-shelf implementation available? If nothing else, a modulated-capacitance CCD array can be imagined, that would cheaply reproduce such a filter characteristic.

What is commonly done is linear or perhaps quadratic interpolation between table values.

Joe Gwinn

Y'all got a schematic of what this "FIR with analog components" looks like?

I was imagining a CCD that instead of being a simple chain was a branched chain, portioning off fractions of the DAC voltage/charge into a summing junction. That allows the FIR filter coefficients to be printed on-chip. The single DAC output would be delayed differently for each tap, and it tracks with the CCD clock (a multiple of the DAC clock) so it can effect its filtering up to that higher clock limit.

The device I put together didn't have any hold capacitors. It would have produced a stepped waveform if it had been perfect, but the steps were at 32-times the clock frequency, and small, so it didn't take much low-pass filtering to make them inperceptible.

Never heard of one.

After crunching through the sinc values to get the 32 resistances, and then having to do it again to throw in Hamming windowing to get rid of the Gibbs oscillations on the output, I'd be a bit surprised if any mass market application could throw up enough customers to pay for a mask set.