Great PWM filter CKT, but how does it work?

[Jack J]

Here is a circuit that low pass filters PWM signal. It does a great
job smoothing out the PWM pulses and yet, its response time is quick
compared to conventional Sallen-key type active filters. I know the
circuit works but I don't know how it works. Does anyone know?

The circuit diagram is at http://webpages.charter.net/unis.

Thanks,

 

[legg]

http://webpages.charter.net/unis.

Troll?

RL

 

[Yzordderex]

Nice try jack!

 

[Helmut Sennewald]

Here is a circuit that low pass filters PWM signal. It does a great
job smoothing out the PWM pulses and yet, its response time is quick
compared to conventional Sallen-key type active filters. I know the
circuit works but I don't know how it works. Does anyone know?

The circuit diagram is at http://webpages.charter.net/unis.

Hello Jack,
I have seen this circuit the first time.
I simulated it immediately with LTSPICE and the result is great.
It achieves sub millivolts of ripple voltage even with a low PWM-
frequency of 100Hz. Settling time is in the range of 1 to 2 seconds.
An equivalent RC-filter would have a settling time 20 times longer.

The filtered output voltage is proportional to the inverted PWM signal.
Please exchange the inverter outputs feeding the resistors for positive PWM.

Principle:
There is a remaining ripple voltage after the 200kOhm and 47nF capacitors.
The opamp tries to cancel this ripple voltage.
Hint: After the voltage is settled, the diodes are always off.
So the 47nF and the 10MEG resistor give a high gain at the pWM frequency.
This gain is the key to cancel the ripple voltage.

Best regards,
Helmut


PS: Send me an email if you want the LTSPICE circuit.
LTSPICE is one of the best SPICE programs and you can get it for free.
Download it from www.linear.com .

There is newsgroup dedicated to LTSPICE:
http://groups.yahoo.com/group/LTspice/

 

[Fred Bartoli]

Hello Jack,
I have seen this circuit the first time.
I simulated it immediately with LTSPICE and the result is great.
It achieves sub millivolts of ripple voltage even with a low PWM-
frequency of 100Hz. Settling time is in the range of 1 to 2 seconds.
An equivalent RC-filter would have a settling time 20 times longer.

The filtered output voltage is proportional to the inverted PWM signal.
Please exchange the inverter outputs feeding the resistors for positive PWM.

Principle:
There is a remaining ripple voltage after the 200kOhm and 47nF capacitors.
The opamp tries to cancel this ripple voltage.
Hint: After the voltage is settled, the diodes are always off.
So the 47nF and the 10MEG resistor give a high gain at the pWM frequency.
This gain is the key to cancel the ripple voltage.

Best regards,
Helmut

Helmut,
an easy way to think about this circuit is to change it to linear operation,
i.e. change the 7404 for just a gain of 1 linear inverter.
When doing an AC analysis you see this as a 3rd order LPF.

When comparing this circuit against a simple 3x 9Hz real poles RC filter,
you have almost no difference between them :
slightly better ondulation for the 3xRC filter and smaller delay at startup
for the "magic" filter.
Anyway, they have exactly the same 1% (and after) settling time on a 5V/50%
duty cycle signal.

Except for curiosity, this circuit is without interest. (unless I've
overlooked something, but it's late, so don't blame me

Oh, one funny thing I've just seen is that, when you remove the diodes, it
rings like a bitch.

 

[Jack J]

Having a great 90% response time is a big asset for this filter and
this filter gets to the 90% point much better than 3rd order
conventional filter.

I still can't figure out what the opamp ckt is doing.

Jack

 

[Helmut Sennewald]

"Fred Bartoli"

"Helmut Sennewald" <[email protected]> a écrit dans le message de

Helmut,
an easy way to think about this circuit is to change it to linear operation,
i.e. change the 7404 for just a gain of 1 linear inverter.
When doing an AC analysis you see this as a 3rd order LPF.

Hello Fred,
thanks for the idea of a .AC Spice analysis. I did it and it looks indeed
like a badly designed 3th order lowpass filter. It has a gain peak of 7.4dB.

When comparing this circuit against a simple 3x 9Hz real poles RC filter,
you have almost no difference between them :
slightly better ondulation for the 3xRC filter and smaller delay at startup
for the "magic" filter.
Anyway, they have exactly the same 1% (and after) settling time on a 5V/50%
duty cycle signal.

If it comes to mV precision(settling), then this filter is by factor of
2 to 5 times slower than a ordinary 3th oder lowpass filter for large steps.
I investigated it further and discovered that it behaves even more awfully
for smaller steps in duty cycle. As smaller the step as less damping
by the diodes.

Except for curiosity, this circuit is without interest. (unless I've
overlooked something, but it's late, so don't blame me

Summary:
This circuit isn't worth to be used.
I hope that somebody has patented it.
This will keep the rest of the world from a useless circuit


Best regards,
Helmut