Data slicer.

You'd think medical types would use a more professional-sounding term than "slice."

It's a little hard to know what your requirements are.

You might sim your thing in LT Spice, post the .ASC file here, and people could make suggestions.

LTspice got wiped out when W10 fell over - I never got around to re installing it.

At 3 components it could hardly be simpler - the signal is direct to one comparator input, the other input has an integrating capacitor fed via a resistor by the signal on the other pin.

As the capacitor integrates the input signal, it feeds the average level as Vref.

So you're finding the median voltage of the input, rather than the mean?

Inneresting idea. Still - what's the purpose, and why not just a low- pass filter?

I thought it was a LPF.

I'd like Vref to settle at the mean of the input signal, and the other pin to detect short term deviations exceeding the input differential voltage.

AFAICR: this circuit is among the NS application examples that have been doing the rounds for many years. That's just an example circuit with no details on calculating time constants.

If it's medical baseline wandering see Earpulse.pdf on the S.E.D/Schematics Page of my website.

Did that in the last century for one of those finger-thingies that shines light thru your fingertip and extracts your pulse rate. ...Jim Thompson

If I read this correctly, you're making a single-slope converter (the time of comparator switching is proportional to an input voltage). Also called time/amplitude converter.

It's a common enough technique; that's how IBM joystick adapters worked, and the reason so many MCU chips offer comparator inputs.

Well, it's certainly not a _linear_ LPF, but in some sense I suppose it's a low-pass filter.

Then use a linear low-pass filter, instead of a circuit that will settle to the median.

If you really want to determine when a signal exceeds the mean of the input by some amount, then low-pass filter the input and apply it to one pin, and offset the input by some amount and apply it to the other pin.

Or low-pass and offset on one, etc.

Are you assuming the CR is in feedback around the comparator?! In any case, a comparator can be linear if you roll it off with enough nfb.

In my circuit the comparator is used as a comparator - if there's any fb at all, it would be a tiny amount of pfb.

The CR simply integrates the input signal - the voltage developed on the capacitor is used as Vref for the comparator.

Vref tracks a slow moving DC offset on the input signal.

"Jim Thompson" wrote in message news: snipped-for-privacy@4ax.com...

UR off your meds again aren't you!!!

Time isn't a parameter.

The CR just integrates the mean level of the input - as that's used as Vref for the comparator; Vref tracks the input mean level.

The comparator responds to discrete pulses, but ignores slow fluctuations in the signal.

Ah, so you aren't going to integrate a small offset into a ramp, just low pass filter. Do you know the risetime of the pulse? Another way to detect pulses, is to take a derivative and sense dV/dt crossing a threshold. You can use a one-shot to detect the dV/dt lasting at-least-this-long.

Or I could just wire the flight deck of a Jumbo jet to it......................................................

The CR just integrates whatever happens on the input. Any sudden change on that input which goes directly to the other comparator pin, deviates from the assumed Vref and produces a response.

You might want to sell me a supercomputer to do a job that only needs 3 components - all I wanted to know is a search string that would enable Google to find info on the circuit I have in mind.

Ah. You confused me with your rank misuse of the word "integrator". An integrator integrates, and to the extent that it has a describable DC gain at all has a DC gain of infinity. An R-C filter is a low-pass filter, usually with a DC gain of 1.

Differentiation (practical type) takes two resistors, a capacitor, and an op amp; it's just a variation on your lowpass filter. Detection takes a comparator, and a one-shot is a timer chip. Real filtering for pulse detection can get elaborate, but this can be wired up in 20 minutes.

If you want something exotic, consider dual-delay-line amplifiers, used for pileup rejection of short-attack/long-delay pulse, in conjunction with Wilkinson peak-integrating A/D conversion. That takes less time, but more money; there's COTS boxes for it.

But it doesn't integrate. It filters. If it integrated then the reference would continually rise when the input is a constant positive voltage.

It wouldn't ignore slow changes. When slowly rising the comparator output=1, and when slowly falling, 0.

Wasn't there a big argument, over a latching relay, about how gain can't be infinite?

Someone else pointed out how this circuit does *not* track the MEAN level, but rather the MEDIAN level. The voltage on the cap will vary until the duty cycle on the comparator output reaches 50:50 which by definition is the MEDIAN, not the MEAN.

Do I misunderstand the circuit?

What the comparator does is to change the amplitude of the difference between the input and the reference so the output is two constant values for the two states removing amplitude information making the reference value a median rather than a mean which is what you get if you just LPF the input directly.

Ian,

your concept will work only if the duty factor of the input signal stays around 50% long term and short term.

If there is anytime the DF deviates from 50%, it may not work so well.

Mark